Split (1)

train · 1.47k rows

filename stringlengths 13 34	markdown stringlengths 0 359k
10.1371_journal.pone.0256297	RESEARCH ARTICLE The influence of childhood adversities on mid to late cognitive function: From the perspective of life course Jing Ma1☯, Yuanyuan Yang2☯, Yang Wan3, Chao Shen2, Peiyuan QiuID 3* 1 Office of Cancer Prevention and Treatment, Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, School of Medicine, University of Electronic Science and Technology of China, Chengdu, China, 2 The Brown School, Washington University in Saint Louis, Saint Louis, MO, United States of America, 3 West China School of Public Health and West China Fourth Hospital, Sichuan University, Chengdu, China ☯ These authors contributed equally to this work. * qiupeiyuan@scu.edu.cn Abstract Background a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Ma J, Yang Y, Wan Y, Shen C, Qiu P (2021) The influence of childhood adversities on mid to late cognitive function: From the perspective of life course. PLoS ONE 16(8): e0256297. https:// doi.org/10.1371/journal.pone.0256297 Editor: Sze Yan Liu, Montclair State University, UNITED STATES Received: March 4, 2021 Accepted: August 3, 2021 Published: August 16, 2021 Copyright: © 2021 Ma et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The data was derived from the second and third wave of the China Health and Retirement Longitudinal Study (CHARLS) (data and documentation are available at http:// charls.pku.edu.cn/). Funding: PQ received a grant from China Medical Board (grant number CMB-14-198). The funders initiated the idea, developed the plan for analysis, and revised the paper. Competing interests: The authors have declared that no competing interests exist. The effects of childhood adversities on cognitive function in later life are well reported. How- ever, few studies have examined the cumulative mechanism, especially in Chinese popula- tion. This study aims to explore this cumulative effects of childhood adversities on mid to late cognitive decline in China. Methods Data were drawn from the second and third wave of the China Health and Retirement Longi- tudinal Study (CHARLS). We included 9,942 respondents aged 45 and above and retro- spectively collected information on childhood adversities. Cognitive function was measured in three dimensions: orientation and calculation, immediate memory, and delayed memory. A structural equation model was employed for analysis. Results Age (β = -0.155, P<0.001) and mid to late depressive symptoms (β = -0.041, P<0.001) showed direct effects on cognitive decline. Low mid to late life socioeconomic status (SES) showed a direct effect on mid-late cognitive impairment (β = 0.603, P<0.001) and an indirect effect through depression (β = 0.007, P<0.001). Low childhood SES (β = 0.310, P<0.001), lack of friends (β = 0.208, P<0.001), parental mental health problems (β = 0.008, P<0.001), and poor relationship with parents (β = 0.001, P<0.001) had an indirect effect on cognitive impairment. Conclusions Childhood adversities had negative effects on cognitive function among middle aged and elderly population in China. The findings suggest that early counter measures on childhood adversities may lead to an effective reduction of cognitive impairment. PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 1 / 16 PLOS ONEThe influence of childhood adversities on mid to late cognitive function Introduction With the escalation of the aging population worldwide, cognitive impairment has emerged as a major public health concern [1,2]. Alzheimer’s Disease International (ADI) [3] estimated that by 2050 the number of people with Alzheimer’s disease (AD) globally would increase to 132 million from 47 million in 2015. China accounted for 25% of the world’s elderly patients with dementia in 2016, which has brought an immense socioeconomic burden [4]. It has been demonstrated that the annual cost of AD in China was over US $167 billion in 2015 and is pro- jected to reach US $1.89 trillion by 2050, emphasizing the importance of dementia as a public health priority [5]. A sizable body of research have emphasized the importance of identifying the biological, psychological, and social factors for maintaining or improving cognitive function [6]. Age, educational attainment, family history, chronic diseases such as diabetes, mental health factors like depression, and repeated stress were major risk factors for cognitive decline [7–9]. Addi- tionally, some studies indicated that adversities in early life such as an absent parent, bad early child–parent relationship quality, and inadequate social support were all negatively associated with cognitive capability in later life [10–14]. However, few studies have explored the mediat- ing effects of childhood adversities on cognitive function in later life. Life course theory was introduced to estimate the contribution of early life experiences to later life outcomes over the whole life process, which has provided a widely used framework for psychological and biological research [15]. Moreover, it provides significant insight into the study of aging and the accumulation of inequality. Therefore, this study aimed to examine the cumulative effects of childhood adversities on cognitive impairment among middle aged and elderly Chinese people using the structural equation model (SEM). Theoretical framework for constructing the SEM Based on the framework of life course, we propose the following theoretical SEM model (Fig 1), which hypothesizes that four childhood adversities and other potential risk factors might act on mid to late life cognitive function in China. Age, gender, and cognitive function. Age is regarded as the greatest risk factor for cogni- tive decline in mid to late life [16–19]. The percentage of people with Alzheimer’s increases dramatically with age: 3% for people aged 65–74, 17% for people aged 75–84, and 32% for peo- ple aged 85 or above [17]. In contrast, findings of a gender difference in cognitive deficit among middle aged and elderly populations remain controversial [20]. Some previous studies suggested gender differences in cognitive performance among Chinese older adults [21,22]. By contrast, other studies indicated non-significant direct association between gender and cogni- tive performance [23]. In our study, we intended to test the direct effects of age and gender on cognitive function in mid to late life. Health status and cognitive function. Health status in middle aged and elderly people, such as chronic diseases, can also affect cognitive abilities. For example, several systematic reviews have shown an increased risk of cognitive decline among individuals with diabetes [24,25]. So, in our study we hypothesized that poor health status could directly affect cognitive function in mid to late life among the Chinese population. Depression and cognitive function. Many studies have found that depression can adversely affect cognitive functioning in late life [26–30]. A recent longitudinal study showed that depressive symptoms increased cognitive decline in older adults [31], which is consistent with the findings of Chinese studies in which depression was generally associated with a PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 2 / 16 PLOS ONEThe influence of childhood adversities on mid to late cognitive function Fig 1. Initial structural equation model of the influence of childhood adversities on mid-late cognitive function. https://doi.org/10.1371/journal.pone.0256297.g001 certain degree of cognitive impairment [32,33]. Thus, we aimed to examine the direct effect of mid to late life depressive symptoms on cognitive impairment. Socioeconomic status and cognitive function. A large number of studies have docu- mented the impact of low SES in childhood on cognitive deficit in later life [34–36]. Lower SES has been found to be associated with a host of negative outcomes, including poorer general health, inequitable access to health services and increased risk for mental illness such as depression and anxiety [37–40]. Aartsen’s study further asserted that advantaged childhood SES was connected with higher cognitive functioning but stronger cognitive decline in older age [41]. A 5-year period cohort study also found that childhood SES is closely related to late- life baseline cognition [42]. Therefore, we intend to verify the direct and indirect association between low SES in mid to late life (associating it with low SES in childhood) and cognitive deficit mediated by depressive symptoms. Childhood relationships with parents and cognitive function. Several studies reported the effect of suboptimal parent–child relationships on cognitive decline in mid to late life [14,43,44]. Adverse experiences from early life had a significant impact on individual out- comes in late life [45]. People who experienced childhood abuse were more likely to have trauma-associated symptoms such as personality disorders, substance abuse, posttraumatic stress disorder, chronic physical conditions, depression, and suicidal ideations [46–48]. Addi- tionally, child neglect and domestic violence between parents have been shown to be associ- ated with depression and health impairments [47,49,50]. Therefore, it is essential to test the potential connection between poor parent–child relationships and cognitive function indi- rectly through depression in our study. Childhood parental mental health and cognitive function. Parental mental health prob- lems were reported as important risk factors for cognitive functioning issues in children [51– PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 3 / 16 PLOS ONEThe influence of childhood adversities on mid to late cognitive function 53]. Familial factors in childhood such as shared genetic factors and a shared living environ- ment had profound effects on children’s health outcomes [45]. For instance, Bennett’s study argued that the severity of mental health conditions for children aged 2–17 was positively related to parental mental illness [54]. Moreover, a 30-year-follow-up study found that chil- dren of parents with depressive symptoms were linked to a higher morbidity and mortality rate related to depression [55]. As such, we aimed to examine the indirect relationship between parental mental health issues and cognitive decline in their offspring through the mediation effect of late-life depressive symptoms. Childhood friendship and cognitive function. Previous studies asserted that friendship support was a positive predictor for cognitive development [56–58]. Friendship is closely related to social adaptability, subjective well-being, and mental health. People with fewer friends were at a higher risk of suicide ideation, which was largely explained by self-assessed depression [59]. Teo’s research confirmed that high-quality social relationships were protective against depression [60]. An 18-year follow-up study also demonstrated that individuals with no friends were approximately twice as likely to experience internalizing symptoms (e.g. depression, anxiety, psychosomatic complaints) compared to those who had at least one friend in childhood [61]. Therefore, it is reasonable to test the association between lack of friends in childhood and mid to late life cognitive decline through depression in this study. Materials and methods Ethical approvals The Ethics Review Committee of Peking University approved the study and informed consent was obtained from all participants. All methods were carried out in accordance with the rele- vant guidelines and regulations. Respondents The data were derived from the second and third wave of the China Health and Retirement Longitudinal Study (CHARLS) (data and documentation are available at http://charls.pku.edu. cn/), which is a nationally representative longitudinal survey. CHARLS employed multistage probability sampling to recruit 150 counties of 28 provinces of mainland China except Hainan, Ningxia and Tibet. At the household level, CHARLS conducted mapping and operations within each village-level unit to create the sample frame. Therefore, households with a member 39 years of age or older were included. Then, randomly sampling was employed to recruit one Individual aged 39 years and over in the household. Selected individuals aged 45 years or older and their spouses were interviewed in the first wave of 2011, and those who were between 39 and 45 years of age were not interviewed and designated for inclusion in a future refreshment sample. More detailed information of the study design and sampling procedure can be found in the cohort profile of CHARLS [62]. The interviewers were trained at Peking University by CHARLS staff members. Data was collected in respondents’ homes by well-trained clinicians in a face-to-face, computer-aided personal interview (CAPI) manner. A total of 17,708 individuals agreed to participate in the baseline survey. The second wave of CHARLS, conducted in 2013, was a regular follow-up sur- vey, which included demographic information, family structure, health status, income, and expenditures. In the second wave, 1,938 individuals were lost to follow-up, of which 431 were dead. Besides, 2,835 nonresponse sample and refresh sample in wave 1 were added for inter- view in wave 2. Totally, 18,605 individuals were surveyed in the second wave. The third wave survey, performed in 2014, was a special survey that retrospectively collected life history infor- mation of all longitudinal responsive samples. This wave included information regarding PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 4 / 16 PLOS ONEThe influence of childhood adversities on mid to late cognitive function Fig 2. Flow chart of participants in the study. https://doi.org/10.1371/journal.pone.0256297.g002 childhood SES, childhood history, health and health care history, and so forth. In the third wave, 2,134 individuals were lost to follow-up, of which 292 individuals were dead. And 4,072 refresh sample from wave 1 and non-response sample in wave 1 and wave 2 were included for interview in wave 3. Overall, 20,543 individuals were interviewed in the third wave. We matched the individuals from Waves 2 and 3 based on their unique IDs in order to trace childhood adversities. The wave 3 survey successfully re-interviewed 16,545 individuals among the respondents in wave 2. To be eligible for the study, respondents had to satisfy three inclusion criteria: 1) they must have been aged 45 or older; 2) they fully provided critical infor- mation on childhood adversities and other potential risk factors; and 3) they were interviewed in both Waves 2 and 3. In the current analyses, 340 individuals were younger than 45-year-old, 2,565 individuals did not complete depression measurements in wave 2, and 3,698 individuals did not provide critical information on childhood adversities in wave 3. They were excluded. Thus, there were 9,942 individuals included in the final sample (Fig 2). Measures Assessments of cognitive function. Cognitive function was measured in three dimen- sions (ten items for each dimension, 30 items in total), including orientation and calculation, immediate memory, and delayed memory (S1 Table). A 10-item questionnaire was adapted to assess orientation and calculation. It required respondents to tell exactly the current year, month, date, week, and season, and subtract 7 from 100 serially for five iterations. As for immediate memory assessment, participants were asked to immediately repeat in any order ten Chinese nouns just read to them. For assessment of delayed memory, respondents were required to recall the 10 words that had been read before by the investigators. A wrong answer for each item received a score of 0, and a correct answer got a score of 1. Adding up scores for each item generated a valid range from 0–30, with lower scores indicating a higher severity of cognitive impairment. PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 5 / 16 PLOS ONEThe influence of childhood adversities on mid to late cognitive function Assessments of depression. The 10-item Center for Epidemiological Studies Depression Scale (CESD-10) was deployed for the measurement of depressive symptoms [63,64]. It had 10 self-reported items, and responses to each item were rated on a 4-point Likert scale ranging from 0 (rare) to 3 (most or all of the time). As for the frequency of negative emotions, the answers were rated from 0 (rarely or none of the time) to 3 (most or all of the time). Regarding the frequency of positive emotions such as “I was happy”, the score was reversely rated from 0 (most or all of the time) to 3 (rarely or none of the time). The sum of these 10 items (range: 0 to 30) reflected individuals’ depression. Thus, higher scores prompted worse depressive symp- toms, and the cut-off point for depression was equal to or greater than 10 [64]. The Cronbach alpha coefficient of CESD-10 in this study was 0.797, consistently indicating comparable reli- ability with previous studies on depression among Chinese middle aged and older adults [65,66]. Assessments of socioeconomic status in mid to late life. SES in mid to late life was assessed by three indicators: residency, educational attainment, and household economic sta- tus. Residency was dichotomized into either an urban or a rural area. Educational attainment was categorized into six groups from illiteracy to bachelor’s degree or above. In most cases, household economic status refers to an income index such as family income. However, due to inaccurate income reporting in China, there are potential limitations when evaluating house- hold economic status using income indicators. Thus, researchers proposed using an asset- based method in which they constructed an asset index to assess the household economic sta- tus [67,68]. In this study, we divided all households into five levels of “household economic sta- tus” based on the asset index we generated using principal component analysis on a scale from 1 (very poor) to 5 (very good) (S1 Table). Assessments of poor health status in mid to late life. Health status in mid to late life was evaluated in four aspects: self-rated health status, the presence of hypertension, the presence of diabetes, and the presence of cardiovascular disease. The self-reported health status was assessed using a 5-point Likert scale from 1 (very good) to 5 (very poor). The remaining three items were dichotomous (S1 Table). People who self-reported having hypertension or diabetes, and who had objective measured values higher than the diagnostic standard were defined as having hypertension or diabetes. Cardiovascular diseases were self-rated by the participants. Assessments of childhood adversities. Low childhood socioeconomic status. We measured childhood SES using parents’ education, father’s occupation, and self-assessed household eco- nomic status. Education level was categorized into six groups from illiteracy to bachelor’s degree or above. Father’s occupation was divided into nonagricultural, farming, and unem- ployment. In addition, a 5-point scale from 1 (very poor) to 5 (very good) was applied to esti- mate household economic status (S1 Table). Lack of friends. Lack of friends was measured in three dimensions: the frequency of discom- fort, the frequency of unhappiness, and existence of good friendship. The first two indicators used a 4-point Likert scale ranging from 1 (never), 2 (not very often), 3 (sometimes), to 4 (often). The last item was dichotomous (S1 Table). Childhood parental mental health problems. We assessed parental mental health problems using the frequency of nervousness, the frequency of anxiety, the frequency of sadness, and panic for parents. All four indicators used a 4-point scale ranging from (most of the time) to 4 (a little of the time) (S1 Table). Poor parent–child relationships. Parent–child relationships were evaluated by three indica- tors: a self-assessment of the mother–child relationship, a self-assessment of the father–child relationship, and the relationship between parents. All three indicators followed a 5-point Likert scale ranging from 1 (poor) to 5 (excellent) (S1 Table). PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 6 / 16 PLOS ONEThe influence of childhood adversities on mid to late cognitive function Model construction and statistical method Based on the theoretical framework, we constructed SEM to analyze the effect of childhood adversities on cognitive function among middle-aged and elderly Chinese individuals. Latent variables and observed variables in the model are shown in S1 Table. The initial structural equation model is shown in Fig 1. Structural equation modelling analyses were performed with SPSS version 19.0 software (SPSS Inc., Chicago, IL, USA), using the weighted least squares means and variance adjusted estimation (WLSMV). The model was considered to have a good fit when root mean square error of approximation (RMSEA)<0.05 [69], comparative fit index (CFI)>0.90, the goodness of fit index (GFI)>0.90, and the normed fit index (NFI)>0.90 [70]. In the present study, the chi-square value was excluded when determining whether a structural equation model had a good fit or not as it is sensitive to sample size, and the chi-square value increased with a larger sample size. Adjusting or deleting the path between the two variables with lager modification index (MI) will be more conducive to the adjustment and optimization of the model. Accord- ing to the model results, we reconstructed the model by removing non-significant associations and re-assessing the model fitness. Standardized regression coefficients (equivalent to path coefficients) among endogenous and exogenous latent variables were shown in the final model. Results Demographic characteristics The description of respondents’ demographic characteristics is presented in Table 1. Of the 9,942 respondents, 52.9% were female. The average age was 59.93 (SD = 8.34), and only 1.86% of the respondents had an education level of some college or above. For location, 63.92% of participants lived in a rural area, and 81.78% reported fair or worse health status. Respondents with hypertension, diabetes, and cardiovascular diseases accounted for 38.74%, 8.32%, and 15.74% respectively. The score of depressive symptoms followed a skewed distribution: the median score was 6.00 (interquartile range, 4–11). The average cognitive function score was 13.53 (SD = 5.57). Other descriptive information about childhood adversities is shown in S2 Table. The differences of demographic characteristics between the excluded (n = 6,603) and included (n = 9,942) individuals were reported in S3 Table. Structural equation modeling The confirmatory factor analysis for the measuring model based on the theoretical framework indicated an appropriate factor structure with a good model fit: RMSEA = 0.056, GFI = 0.904, TLI = 0.781, CFI = 0.799. All factor loadings from observed to latent variables were significant. Two hypothesized paths failed to reach significance: 1) from health status in mid to late life to cognitive function in mid to late life (P = 0.418) and 2) from gender to cognitive function in mid to late life (P = 0.062). Therefore, we deleted the insignificant pathways and reassessed each model. Then, new results indicated that the modification index (MI) between "lack of friends" and "socioeconomic status in mid to late life" was large (MI = 1417.749, parameter change = 0.569). After adding the pathway from "lack of friends" to" socioeconomic status in mid to late life", the final model was ascertained (Fig 3). A more satisfactory model was attained with a good model fit: RMSEA = 0.041, GFI = 0.952, TLI = 0.908, CFI = 0.918. All path coefficients in the final model were standardized and significant (Table 2). Within the final model, lack of friends significantly indicated low SES in mid to late life (β = 0.338, SE = 0.014) and more severe depressive symptoms (β = -0.044, SE = 0.012). Low SES in mid to PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 7 / 16 PLOS ONEThe influence of childhood adversities on mid to late cognitive function Table 1. Demographic characteristics of respondents (weighted). Variable Age, Mean Age Gender Educational attainment Aged 45–60 Aged 60–75 Aged 75–90 Aged 90+ Male Female Illiterate Primary school Junior high school High school (secondary specialized school) Some college Bachelor degree or above Residency Household economic status Rural Urban Very poor Poor Fair Good Very good The presence of hypertension The presence of diabetes Yes No Yes No The presence of cardiovascular disease Self-rated health status Yes No Very poor Poor Fair Good Very good Cognitive function, Mean https://doi.org/10.1371/journal.pone.0256297.t001 Overall (N = 9,942) �x� ± SD/n(%) 59.93±8.34 5543 (55.75) 3800 (38.22) 593 (5.96) 6 (0.06) 4680 (47.07) 5262 (52.93) 2488 (25.03) 4058 (40.82) 2182 (21.95) 1029 (10.35) 133 (1.34) 52 (0.52) 6355 (63.92) 3587 (36.08) 1821 (18.32) 1959 (19.70) 2035 (20.47) 2051 (20.63) 2076 (20.88) 3852 (38.74) 6090 (61.26) 827 (8.32) 9115 (91.68) 1565 (15.74) 8377 (84.26) 1281 (12.88) 3571 (35.92) 3279 (32.98) 1178 (11.85) 633 (6.37) 13.53±5.57 late life was strongly associated with higher severity of cognitive impairment (β = 0.603, SE = 0.051) and significantly predicted more severe depressive symptoms (β = -0.174, SE = 0.017). In addition, depression in mid to late life was significantly associated with cogni- tive impairment (β = -0.041 SE = 0.031). Parental mental health problems during childhood PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 8 / 16 PLOS ONEThe influence of childhood adversities on mid to late cognitive function Fig 3. Final structural equation model of the influence of childhood adversities on mid to late cognitive function. https://doi.org/10.1371/journal.pone.0256297.g003 and bad parent–child relationships both had a significant influence on cognitive decline (β = -0.190, SE = 0.011; β = −0.033, SE = 0.012). We also found that increased age predicted a lower level of cognitive function (β = -0.155, SE = 0.028), and SES in childhood was strongly associ- ated with SES in mid to late life (β = 0.508, SE = 0.081). Standardized direct and indirect effects of childhood adversities on cognitive function in mid to late life are summarized in Table 3. It has been reported that SES in mid to late life has the largest effect on cognitive function (β = 0.610, P<0.001). Low SES in mid to late life had a positive effect on the cognitive impairment directly and indirectly, with the indirect effects Table 2. Path coefficients and standard errors for the final SEM model (N = 9,942). Independent variable Low childhood SES Lack of friends Lack of friends Childhood parental mental health problems Low SES in mid to late life Poor parent–child relationships Depression Age Low SES in mid to late life Dependent variable Low SES in mid to late life Low SES in mid to late life Depression Depression Depression Depression Cognitive function Cognitive function Cognitive function SES, socioeconomic status; β, path coefficient; SE, standard error; C.R., critical ratio. a, standardized parameter. https://doi.org/10.1371/journal.pone.0256297.t002 βa 0.508 0.338 -0.044 -0.19 -0.174 -0.033 -0.041 -0.155 0.603 SE 0.081 0.014 0.012 0.011 0.017 0.012 0.031 0.028 0.051 CR 16.477 18.799 -2.719 -15.765 -9.954 -2.789 -3.472 -15.107 29.918 P <0.001 <0.001 0.007 <0.001 <0.001 0.005 <0.001 <0.001 <0.001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 9 / 16 PLOS ONETable 3. The standardized direct effects, indirect and total effects of the childhood adversities on cognitive function in mid to late life (N = 9,942). The influence of childhood adversities on mid to late cognitive function Variable Age Low Childhood SES Lack of friends Childhood parental mental health problems Poor parent–child relationships Depression Low SES in mid to late life SES, socio-economic status. � � � P<0.001. https://doi.org/10.1371/journal.pone.0256297.t003 Standardized direct effects Standardized indirect effects Standardized total effects � � � -0.155 - - - - � � � -0.041 � � � 0.603 - � � � 0.310 � � � 0.208 � � � 0.008 � � � 0.001 - � � � 0.007 � � � -0.155 � � � 0.310 � � � 0.208 � � � 0.008 � � � 0.001 � � � -0.041 � � � 0.610 mediated by depression. SES in childhood had the second largest total effect on mid to late life cognitive decline (β = 0.310, P<0.001) before the total effect of lack of friends during child- hood (β = 0.208, P<0.001). Low SES in childhood had an indirect effect on poor cognitive per- formance mediated by low SES in mid to late life (β = 0.310, P<0.001). Lack of friends affected cognitive impairment through SES in mid to late life and depressive symptoms (β = 0.208, P<0.001). Parental mental health problems during childhood and poor parent–child relation- ships presented no direct effect on cognitive decline but contributed to depression (β = 0.008, P<0.001; β = 0.001, P<0.001). Additionally, age and depressive symptoms both indicated direct effects on cognitive function (β = -0.155, P<0.001; β = -0.041, P<0.001). The correla- tions of four latent variables of childhood adversities are demonstrated in Table 4. Discussion In our study, we intended to unpack the mechanism explanation between childhood adversi- ties and cognitive impairment among middle aged and elderly Chinese people using different mediators. Consistent with our hypotheses and previous studies, age, depressive symptoms, and SES in mid to late life were directly associated with cognitive deficit [28,71,72]. Low SES in mid to late life showed the largest total effect on cognitive function among all the direct rela- tionships, which is possibly due to reasons such as less access to health services, limited social support, fewer opportunities for success, and a higher probability of exposure to life adversities [40,73]. However, we did not find a significant direct association between gender and cognitive performance, which is consistent with the findings of a previous study in China [23]. It was shown that lack of friends in childhood indirectly affected cognitive decline in mid to late life mediated by depressive symptoms and low SES in mid to late life. Social support and Table 4. Correlations of four latent variables of childhood adversities (N = 9942). Latent variables Lack of friends Lack of friends Lack of friends Childhood parental mental health problems Poor parent–child relationships Low childhood SES Childhood parental mental health problems Poor parent–child relationships Childhood parental mental health problems Low childhood SES Poor parent–child relationships Low childhood SES SES, socio-economic status; r, correlation coefficient; SE, standard error; C.R., critical ratio. a, standardized parameter. https://doi.org/10.1371/journal.pone.0256297.t004 ra 0.133 0.197 0.263 0.035 0.141 0.048 S.E. 0.007 0.007 0.005 0.005 0.003 0.002 C.R. 10.452 14.845 12.24 3.075 8.558 3.399 P <0.001 <0.001 <0.001 0.002 <0.001 <0.001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 10 / 16 PLOS ONEThe influence of childhood adversities on mid to late cognitive function social networks were protective factors for people’s mental health outcomes including cogni- tive ability in later life [74,75]. We argued that satisfying social relationships with others could enhance children’s psychological well-being by providing strong emotional support and strengthening coping abilities when children encounter life adversities. Additionally, friend- ship is closely related to parental educational level and occupation, which indirectly reflects SES [13]. Therefore, our results indicate that lack of friends was associated with an increasing risk of depressive symptoms and low SES in mid to late life, which subsequently increased the likelihood of low cognitive ability in mid to late life. In accord with our hypothesis, it was shown that children whose parents had poor mental health were more likely to experience cognitive decline in later life, indirectly mediated through depressive symptoms in mid to late life. A recent study among older Chinese adults demonstrated that children who were exposed to parents with mental health problems were more likely to suffer depression in mid to late life [50]. Similarly, a significant association between depression and cognitive impairment in mid to late life was also found in our model, which is consistent with other findings [8,76,77]. A likely explanation could be that children living with parents who have poor mental health tend to receive inadequate emotional care and be influenced by substance abuse, anxiety, depression, suicide ideation, maltreatment, and violence from parents. These risk factors predict children’s mental health outcomes from childhood and have a cumulative impact on their depression and cognitive decline in mid to late life. Our study also found a significant positive indirect association between poor parent–child relationships in childhood and a deficiency in cognitive function in mid to late life, which was mediated by depressive symptoms. A systematic review illustrated that children with secure attachments to parents tend to develop more positive health and mental health outcomes including better cognitive functioning [43]. Korkeila also found that good parent–child inter- actions were associated with increasing optimism, which could serve as a partial buffer when confronting adversities [78]. The indirect effects can be explained through the long-term impact of poor parent–child relationships developing over time on depression in mid to late life, which also significantly predicts cognitive impairment. In addition, the results showed that experiencing low SES in childhood predicted a higher probability of cognitive decline, through the mediation effect of low SES in mid to late life as shown in previous studies [79,80]. Children growing up with low SES are less likely to gain adequate financial backups, higher educational attainment, and supportive social networks, which are crucial factors for future success. They have a higher risk of low SES in mid to late life, which in turn may increase their likelihood of experiencing poor nutrition and decrease their ability to afford health services such as prevention and treatments for depression. Although this study benefitted from the use of life course theory to explore cumulative effects of childhood adversities on cognitive deficit in later life, we recognize several limitations as well. First, the CHARLS does not possess data from three provinces in mainland China (Hainan, Ningxia and Tibet), hence it is not entirely representative of the nation. Second, childhood information was collected with a retrospective method, which may cause recall bias. Third, as all the data were collected by questionnaire and were self-reported, it may have reporting bias. Fourth, individuals who had adverse events earlier in life may have died before reaching 45 years old, which may led to survival bias. Fifth, although the CHARLS survey was collected following a well-administered process with a low lost-to-follow-up rate, attrition bias and non-response bias may exist. Last, adverse events and other unobserved confounding fac- tors were not available from the CHARLS, so we could not further test their effects on depres- sion and cognitive impairment. PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 11 / 16 PLOS ONEThe influence of childhood adversities on mid to late cognitive function Conclusions This study examined the potential paths from four aspects of childhood adversities including lack of friends, poor parental mental health status, poor parent–child relationships, and low SES to cognitive impairment in middle-aged and elderly Chinese populations. The results demonstrated that all four variables were associated with mid to late life cognitive decline through indirect processes. From the life course perspective, the negative impacts of adverse experiences in childhood on people’s mental health are not isolated. Instead, these experiences cumulatively influence cognitive deficit in mid to late life. These important findings suggest the urgent need to invest available resources to prevent childhood adversities, subsequently reducing the prevalence of cognitive decline. Supporting information S1 Table. Assignment of latent variables and observed variables. (PDF) S2 Table. Description of childhood adversities (N = 9,942). (PDF) S3 Table. Demographic characteristics between the excluded and included individuals. (PDF) Acknowledgments We would like to acknowledge the China Health and Retirement Longitudinal Study (CHARLS) team for providing high quality, nationally representative data. Author Contributions Conceptualization: Peiyuan Qiu. Data curation: Jing Ma. Formal analysis: Jing Ma. Funding acquisition: Peiyuan Qiu. Methodology: Peiyuan Qiu. Software: Jing Ma. Visualization: Chao Shen. Writing – original draft: Jing Ma, Yuanyuan Yang. Writing – review & editing: Jing Ma, Yuanyuan Yang, Yang Wan, Peiyuan Qiu. References 1. WHO A. Dementia: a public health priority. Geneva: World Health Organization. 2012. 2. Sepanlou SG, Parsaeian M, Krohn KJ, Afshin A, Farzadfar F, Roshandel G, et al. Disability-Adjusted Life-Years (DALYs) for 315 diseases and injuries and Healthy Life Expectancy (HALE) in Iran and its neighboring countries, 1990–2015: findings from Global Burden of Disease Study 2015. Arch. Iran. Med. 2017; 20(7):403–18. PMID: 28745902 3. Prince MJ. World Alzheimer Report 2015: the global impact of dementia: an analysis of prevalence, inci- dence, cost and trends: Alzheimer’s Disease International; 2015. 4. Nichols E, Szoeke CEI, Vollset SE, Abbasi N, FoadAbd-Allah, Abdela J, et al. Global, regional, and national burden of Alzheimer’s disease and other dementias, 1990–2016: a systematic analysis for the PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 12 / 16 PLOS ONEThe influence of childhood adversities on mid to late cognitive function Global Burden of Disease Study 2016. Lancet Neurol. 2019; 18(1):88–106. https://doi.org/10.1016/ S1474-4422(18)30403-4 PMID: 30497964 5. Jia J, Wei C, Chen S, Li F, Tang Y, Qin W, et al. The cost of Alzheimer’s disease in China and re-estima- tion of costs worldwide. Alzheimers Dement. 2018; 14(4):483–91. https://doi.org/10.1016/j.jalz.2017. 12.006 PMID: 29433981 6. Hendrie HC, Albert MS, Butters MA, Gao S, Knopman DS, Launer LJ, et al. The NIH cognitive and emo- tional health project: report of the critical evaluation study committee. Alzheimers Dement. 2006; 2:12– 32. https://doi.org/10.1016/j.jalz.2005.11.004 PMID: 19595852 7. Elias MF, Elias PK, Sullivan LM, Wolf PA, D’Agostino RB. Obesity, diabetes and cognitive deficit: the Framingham Heart Study. Neurobiol. Aging. 2005; 26(1):11–6. https://doi.org/10.1016/j.neurobiolaging. 2005.08.019 PMID: 16223549 8. Panza F, Frisardi V, Capurso C, D’Introno A, Colacicco AM, Imbimbo BP, et al. Late-life depression, mild cognitive impairment, and dementia: possible continuum? Am. J. Geriatr. Psychiatry. 2010; 18 (2):98–116. https://doi.org/10.1097/JGP.0b013e3181b0fa13 PMID: 20104067 9. Yuen EY, Wei J, Liu W, Zhong P, Li X, Yan Z. Repeated stress causes cognitive impairment by sup- pressing glutamate receptor expression and function in prefrontal cortex. Neuron. 2012; 73:962–77. https://doi.org/10.1016/j.neuron.2011.12.033 PMID: 22405206 10. Santos DN, Assis AMO, Bastos CSA, Santos LM, Santos CAS, Strina A, et al. Determinants of cognitive function in childhood: A cohort study in a middle income context. BMC Public Health. 2008; 8:202. https://doi.org/10.1186/1471-2458-8-202 PMID: 18534035 11. Majer M, Nater UM, Lin J-MS, Capuron L, Reeves WC. Association of childhood trauma with cognitive function in healthy adults: a pilot study. BMC Neurol. 2010; 10(1):61. https://doi.org/10.1186/1471- 2377-10-61 PMID: 20630071 12. Burri A, Maercker A, Krammer S, Simmen-Janevska K. Childhood Trauma and PTSD Symptoms Increase the Risk of Cognitive Impairment in a Sample of Former Indentured Child Laborers in Old Age. PLoS One. 2013; 8(2):e57826. https://doi.org/10.1371/journal.pone.0057826 PMID: 23469076 13. Lin Z, Chen X. Adverse Childhood Circumstances and Cognitive Function in Middle-aged and Older Chinese Adults: Lower Level or Faster Decline? SSM-Popul. Hlth. 2021; 14:100767. https://doi.org/10. 1016/j.ssmph.2021.100767 PMID: 33855158 14. Peng C, Burr JA, Yang D, Nan L. Early Child–Parent Relationship Quality and Cognitive Function in Older Rural Chinese Adults: The Mediating Role of Educational Attainment. J. Aging Health. 2021; 33 (2):089826432199656. https://doi.org/10.1177/0898264321996562 PMID: 33625258 15. Elder GH. The life course as developmental theory. Child Dev. 1998; 69:1–12. PMID: 9499552 16. Barnes JN. Exercise, cognitive function, and aging. Adv Physiol Educ. 2015; 39(2):55–62. https://doi. org/10.1152/advan.00101.2014 PMID: 26031719 17. Gaugler J, James B, Johnson T, Marin A, Weuve J. 2019 Alzheimer’s disease facts and figures. Alzhei- mers Dement. 2019; 15(3):321–87. 18. Gaugler JE, Duval S, Anderson KA, Kane RL. Predicting nursing home admission in the US: a meta- analysis. BMC Geriatr. 2007; 7(1):13. 19. Tabbarah M, Crimmins EM, Seeman TE. The relationship between cognitive and physical performance: MacArthur Studies of Successful Aging. J. Gerontol. 2002; 57(4):M228–35. https://doi.org/10.1093/ gerona/57.4.m228 PMID: 11909888 20. Han M, Huang X-F, Xiu MH, Hui L, Liu H, Kosten TR, et al. Gender differences in cognitive function of patients with chronic schizophrenia. Prog. Neuropsychopharmacol. Biol. Psychiatry. 2012; 39(2):358– 63. https://doi.org/10.1016/j.pnpbp.2012.07.010 PMID: 22820676 21. 22. Lei X, Smith JP, Sun X, Zhao Y. Gender differences in cognition in China and reasons for change over time: evidence from CHARLS. J. Econ. Ageing. 2014; 4:46–55. https://doi.org/10.1016/j.jeoa.2013.11. 001 PMID: 25530942 Lei X, Hu Y, McArdle JJ, Smith JP, Zhao Y. Gender Differences in Cognition among Older Adults in China. J. Hum. Resour. 2012; 47(4):951–71. https://doi.org/10.3368/jhr.47.4.951 PMID: 24347682 23. Kesavayuth D, Liang Y, Zikos V. An active lifestyle and cognitive function: Evidence from China. J. Econ. Aging. 2018; 12:183–91. 24. Gudala K, Bansal D, Schifano F, Bhansali A. Diabetes mellitus and risk of dementia: A meta-analysis of prospective observational studies. J. Diabetes Investig. 2013; 4(6):640–50. https://doi.org/10.1111/jdi. 12087 PMID: 24843720 25. Cheng G, Huang C, Deng H, Wang H. Diabetes as a risk factor for dementia and mild cognitive impairment: a meta-analysis of longitudinal studies. Intern. Med. J. 2012; 42(5):484–91. https://doi.org/ 10.1111/j.1445-5994.2012.02758.x PMID: 22372522 PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 13 / 16 PLOS ONEThe influence of childhood adversities on mid to late cognitive function 26. Biringer E, Mykletun A, Dahl AA, Smith AD, Engedal K, Nygaard HA, et al. The association between depression, anxiety, and cognitive function in the elderly general population—the Hordaland Health Study. Int. J. Geriatr. Psychiatry. 2005; 20:989–97. https://doi.org/10.1002/gps.1390 PMID: 16163751 27. Donovan NJ, Wu Q, Rentz DM, Sperling RA, Marshall GA, Glymour MM. Loneliness, Depression and Cognitive Function in Older U.S. Adults. Int. J. Geriatr. Psychiatry. 2017; 32:564–73. https://doi.org/10. 1002/gps.4495 PMID: 27162047 28. McDermott LM, Ebmeier KP. A meta-analysis of depression severity and cognitive function. J. Affect. Disord. 2009; 119:1–8. https://doi.org/10.1016/j.jad.2009.04.022 PMID: 19428120 29. Vinkers DJ, Gussekloo J, Stek ML, Westendorp RGJ, van der Mast RC. Temporal relation between depression and cognitive impairment in old age: prospective population based study. BMJ. 2004; 329 (7471):881. https://doi.org/10.1136/bmj.38216.604664.DE PMID: 15345599 30. Kuring JK, Mathias JL, Ward L. Risk of Dementia in persons who have previously experienced clinically- significant Depression, Anxiety, or PTSD: A Systematic Review and Meta-Analysis. J. Affect. Disord. 2020; 274(2020):247–61. https://doi.org/10.1016/j.jad.2020.05.020 PMID: 32469813 31. Holmquist S, Nordstro¨ m A, Nordstro¨m P. The association of depression with subsequent dementia diagnosis: A Swedish nationwide cohort study from 1964 to 2016. PLoS Med. 2020; 17(1):e1003016. https://doi.org/10.1371/journal.pmed.1003016 PMID: 31917808 32. Chi I, Chou KL. Depression predicts cognitive decline in Hong Kong Chinese older adults. Aging Ment Health. 2000; 4(2):148–57. 33. Giri M, Chen T, Yu W, Lu¨ Y. Prevalence and correlates of cognitive impairment and depression among elderly people in the world’s fastest growing city, Chongqing, People’s Republic of China. Clin. Interv. Aging. 2016; 11:1091–8. https://doi.org/10.2147/CIA.S113668 PMID: 27574409 34. Karlamangla AS, Miller-Martinez D, Aneshensel CS, Seeman TE, Wight RG, Chodosh J. Trajectories of cognitive function in late life in the United States: demographic and socioeconomic predictors. Am. J. Epidemiol. 2009; 170:331–42. https://doi.org/10.1093/aje/kwp154 PMID: 19605514 35. Lee Y, Back JH, Kim J, Byeon H. Multiple socioeconomic risks and cognitive impairment in older adults. Dement. Geriatr. Cogn. Disord. 2010; 29:523–9. https://doi.org/10.1159/000315507 PMID: 20606433 36. Hackman DA, Farah MJ. Socioeconomic status and the developing brain. Trends Cogn. Sci. 2009; 13 (2):65–73. https://doi.org/10.1016/j.tics.2008.11.003 PMID: 19135405 37. Goodman E, Slap GB, Huang B. The public health impact of socioeconomic status on adolescent depression and obesity. Am. J. Public Health. 2003; 93(11):1844–50. https://doi.org/10.2105/ajph.93. 11.1844 PMID: 14600051 38. Fryers T, Melzer D, Jenkins R. Social inequalities and the common mental disorders. Soc. Psychiatry Psychiatr. Epidemiol. 2003; 38(5):229–37. https://doi.org/10.1007/s00127-003-0627-2 PMID: 12719837 39. Barbeau EM, Krieger N, Soobader M-J. Working Class Matters: Socioeconomic Disadvantage, Race/ Ethnicity, Gender, and Smoking in NHIS 2000. Am. J. Public Health. 2004; 94(2):269–78. https://doi. org/10.2105/ajph.94.2.269 PMID: 14759942 40. Zhang C, Lei X, Strauss J, Zhao Y. Health Insurance and Health Care among the Mid-Aged and Older Chinese: Evidence from the National Baseline Survey of CHARLS. Health Econ. 2017; 26:431–49. https://doi.org/10.1002/hec.3322 PMID: 26856894 41. Aartsen MJ, Cheval B, Sieber S, Van der Linden BW, Gabriel R, Courvoisier DS, et al. Advantaged socioeconomic conditions in childhood are associated with higher cognitive functioning but stronger cognitive decline in older age. Proc. Natl. Acad. Sci. U. S. A. 2019; 116:5478–86. https://doi.org/10. 1073/pnas.1807679116 PMID: 30804194 42. Sha T, Yan Y, Cheng W. Associations of childhood socioeconomic status with mid-life and late-life cog- nition in Chinese middle-aged and older population based on a 5-year period cohort study. Int. J. Ger- iatr. Psych. 2018; 33(10):1335–45. https://doi.org/10.1002/gps.4930 PMID: 29971839 43. Ranson KE, Urichuk LJ. The effect of parent-child attachment relationships on child biopsychosocial outcomes: a review. Early Child Development and Care. 2008; 178(2):129–52. 44. 45. 46. Thompson RA. Early attachment and later development: Familiar questions, new answers. New York: NY: The Guilford Press; 2008. Ferraro KF, Shippee TP. Aging and cumulative inequality: How does inequality get under the skin? The Gerontologist. 2009; 49(3):333–43. https://doi.org/10.1093/geront/gnp034 PMID: 19377044 Levitan RD, Rector NA, Sheldon T, Goering P. Childhood adversities associated with major depression and/or anxiety disorders in a community sample of Ontario: Issues of co-morbidity and specificity. Depress. Anxiety. 2003; 17(1):34–42. https://doi.org/10.1002/da.10077 PMID: 12577276 PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 14 / 16 PLOS ONEThe influence of childhood adversities on mid to late cognitive function 47. Herrenkohl TI, Hong S, Klika JB, Herrenkohl RC, Russo MJ. Developmental impacts of child abuse and neglect related to adult mental health, substance use, and physical health. J. Fam. Violence. 2013; 28 (2):191–9. 48. Gonzalez A, Boyle MH, Kyu HH, Georgiades K, Duncan L, MacMillan HL. Childhood and family influ- ences on depression, chronic physical conditions, and their comorbidity: findings from the Ontario Child Health Study. J. Psychiatr. Res. 2012; 46(11):1475–82. https://doi.org/10.1016/j.jpsychires.2012.08. 004 PMID: 22959202 49. Hovens JG, Wiersma JE, Giltay EJ, Van Oppen P, Spinhoven P, Penninx BW, et al. Childhood life events and childhood trauma in adult patients with depressive, anxiety and comorbid disorders vs. con- trols. Acta Psychiatr. Scand. 2010; 122(1):66–74. https://doi.org/10.1111/j.1600-0447.2009.01491.x PMID: 19878136 50. Tian F, Meng SS, Qiu P. Childhood adversities and mid-late depressive symptoms over the life course: Evidence from the China health and retirement longitudinal study. J. Affect. Disord. 2019; 245:668–78. https://doi.org/10.1016/j.jad.2018.11.028 PMID: 30445392 51. McManus BM, Poehlmann J. Parent-child interaction, maternal depressive symptoms and preterm infant cognitive function. Infant Behav. Dev. 2012; 35(3):489–98. https://doi.org/10.1016/j.infbeh.2012. 04.005 PMID: 22721747 52. Slomian J, Honvo G, Emonts P, Reginster J-Y, Bruyère O. Consequences of maternal postpartum depression: A systematic review of maternal and infant outcomes. Womens Health. 2019; 15 (1780):174550651984404. https://doi.org/10.1177/1745506519844044 PMID: 31035856 53. Stein A, Pearson RM, Goodman SH, Rapa E, Rahman A, McCallum M, et al. Effects of perinatal mental disorders on the fetus and child. The Lancet. 2014; 384(9956):1800–19. https://doi.org/10.1016/S0140- 6736(14)61277-0 PMID: 25455250 54. Bennett AC, Brewer KC, Rankin KM. The association of child mental health conditions and parent men- tal health status among US children, 2007. Matern. Child Hlth. J. 2012; 16(6):1266–75. 55. Weissman MM, Berry OO, Warner V, Gameroff MJ, Skipper J, Talati A, et al. A 30-year study of 3 gen- erations at high risk and low risk for depression. JAMA Psychiatry. 2016; 73(9):970–7. https://doi.org/ 10.1001/jamapsychiatry.2016.1586 PMID: 27532344 56. Mazurek MO, Kanne SM. Friendship and internalizing symptoms among children and adolescents with ASD. J. Autism Dev. Disord. 2010; 40:1512–20. https://doi.org/10.1007/s10803-010-1014-y PMID: 20405193 57. Mikami AY. The importance of friendship for youth with attention-deficit/hyperactivity disorder. Clin. Child Fam. Psychol. Rev. 2010; 13:181–98. https://doi.org/10.1007/s10567-010-0067-y PMID: 20490677 58. Van Harmelen AL, Kievit RA, Ioannidis K, Neufeld S, Jones PB, Bullmore E, et al. Adolescent friend- ships predict later resilient functioning across psychosocial domains in a healthy community cohort. Psychol. Med. 2017; 47:2312–22. https://doi.org/10.1017/S0033291717000836 PMID: 28397612 59. Marver JE, Galfalvy HC, Burke AK, Sublette ME, Oquendo MA, Mann JJ, et al. Friendship, Depression, and Suicide Attempts in Adults: Exploratory Analysis of a Longitudinal Follow-Up Study. Suicide Life- Threat. 2017; 47(6):660–71. 60. Teo AR, Choi H, Valenstein M. Social relationships and depression: ten-year follow-up from a nationally representative study. PLoS One. 2013; 8(4):e62396. https://doi.org/10.1371/journal.pone.0062396 PMID: 23646128 61. Sakyi KS, Surkan PJ, Fombonne E, Chollet A, Melchior M. Childhood friendships and psychological dif- ficulties in young adulthood: an 18-year follow-up study. Eur. Child Adolesc. Psychiatry. 2015; 24 (7):815–26. https://doi.org/10.1007/s00787-014-0626-8 PMID: 25316094 62. Zhao Y, Hu Y, Smith JP, Strauss J, Yang G. Cohort profile: the China health and retirement longitudinal study (CHARLS). Int. J. Epidemiol. 2014; 43:61–8. https://doi.org/10.1093/ije/dys203 PMID: 23243115 63. Wang X, Wang X, Ma H. Psychological assessment scale manual. Beijing, P.R.China: Chinese Mental Health Journal press. 1999. 64. Andresen EM, Malmgren JA, Carter WB, Patrick DL. Screening for depression in well older adults: eval- uation of a short form of the CES-D (Center for Epidemiologic Studies Depression Scale). Am. J. Prev. Med. 1994; 10(2):77–84. PMID: 8037935 65. Huang Q, Wang X, Chen G. Reliability and validity of 10-item CES-D among middle aged and older adults in China. China Journal of Health Psychology. 2015(7):24. 66. Yang L, Jia C-X, Qin P. Reliability and validity of the Center for Epidemiologic Studies Depression Scale (CES-D) among suicide attempters and comparison residents in rural China. BMC Psychiatry. 2015; 15 (1):76. PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 15 / 16 PLOS ONEThe influence of childhood adversities on mid to late cognitive function 67. Filmer D, Pritchett LH. Estimating wealth effects without expenditure data—or tears: an application to educational enrollments in states of India. Demography. 2001; 38(1):115–32. https://doi.org/10.1353/ dem.2001.0003 PMID: 11227840 68. Morris SS, Carletto C, Hoddinott J, Christiaensen LJJJoE, Health C. Validity of rapid estimates of household wealth and income for health surveys in rural Africa. J. Epidemiol. Community Health. 2000; 54(5):381–7. https://doi.org/10.1136/jech.54.5.381 PMID: 10814660 69. Browne MW, Cudeck R. Alternative ways of assessing model fit. Sociol. Method Res. 1992; 21(2):230– 58. 70. Bentler PM. Comparative fit indexes in structural models. Psychol. Bull. 1990; 107(2):238. https://doi. org/10.1037/0033-2909.107.2.238 PMID: 2320703 71. Deary IJ, Corley J, Gow AJ, Harris SE, Houlihan LM, Marioni RE, et al. Age-associated cognitive decline. Br. Med. Bull. 2009; 92(1):135–52. https://doi.org/10.1093/bmb/ldp033 PMID: 19776035 72. Lyu J, Burr JA. Socioeconomic status across the life course and cognitive function among older adults: An examination of the latency, pathways, and accumulation hypotheses. J. Aging Health. 2016; 28 (1):40–67. https://doi.org/10.1177/0898264315585504 PMID: 26006338 73. Nurius PS, Fleming CM, Brindle E. Life Course Pathways From Adverse Childhood Experiences to Adult Physical Health: A Structural Equation Model. J. Aging Health. 2019; 31:211–30. https://doi.org/ 10.1177/0898264317726448 PMID: 28845729 74. Fratiglioni L, Wang H-X, Ericsson K, Maytan M, Winblad B. Influence of social network on occurrence of dementia: a community-based longitudinal study. The Lancet. 2000; 355(9212):1315–9. https://doi.org/ 10.1016/S0140-6736(00)02113-9 PMID: 10776744 75. Gow AJ, Pattie A, Whiteman MC, Whalley LJ, Deary IJ. Social support and successful aging: Investigat- ing the relationships between lifetime cognitive change and life satisfaction. J. Indiv. Differ. 2007; 28 (3):103–15. 76. Richard E, Reitz C, Honig LH, Schupf N, Tang MX, Manly JJ, et al. Late-life depression, mild cognitive impairment, and dementia. JAMA Neurol. 2013; 70(3):374–82. 77. Steffens DC, Otey E, Alexopoulos GS, Butters MA, Cuthbert B, Ganguli M, et al. Perspectives on depression, mild cognitive impairment, and cognitive decline. Arch. Gen. Psychiatry. 2006; 63(2):130– 8. https://doi.org/10.1001/archpsyc.63.2.130 PMID: 16461855 78. Korkeila K, Kivela¨ S-L, Suominen S, Vahtera J, Kivima¨ ki M, Sundell J, et al. Childhood adversities, par- ent-child relationships and dispositional optimism in adulthood. Soc. Psychiatry Psychiatr. Epidemiol. 2004; 39(4):286–92. https://doi.org/10.1007/s00127-004-0740-x PMID: 15085330 79. Everson-Rose SA, Mendes de Leon CF, Bienias JL, Wilson RS, Evans DA. Early life conditions and cognitive functioning in later life. Am. J. Epidemiol. 2003; 158(11):1083–9. https://doi.org/10.1093/aje/ kwg263 PMID: 14630604 80. Kaplan GA, Turrell G, Lynch JW, Everson SA, Helkala E-L, Salonen JT. Childhood socioeconomic posi- tion and cognitive function in adulthood. Int. J. Epidemiol. 2001; 30(2):256–63. https://doi.org/10.1093/ ije/30.2.256 PMID: 11369724 PLOS ONE \| https://doi.org/10.1371/journal.pone.0256297 August 16, 2021 16 / 16 PLOS ONE
10.1371_journal.pone.0265395	RESEARCH ARTICLE Chronic kidney disease in Ecuador: An epidemiological and health system analysis of an emerging public health crisis 1, Rachel Sippy2,3, Kevin Louis Bardosh4, Ramya Bhargava2, Martı´n Lotto- Irene TorresID Batista2,5, Abigail E. Bideaux2, Ramon Garcia-Trabanino6,7, Amelia Goldsmith8, Sriram S. Narsipur2, Anna M. Stewart-Ibarra2,9 1 Fundacion Octaedro, Quito, Ecuador, 2 SUNY Upstate Medical University, Syracuse, New York, United States of America, 3 University of Florida, Gainesville, Florida, United States of America, 4 Center for One Health Research, School of Public Health, University of Washington, Seattle, Washington, United States of America, 5 Department of Epidemiology, Helmholtz Centre for Infection Research, Brunswick, Germany, 6 Centro de Hemodia´ lisis, San Salvador, El Salvador, 7 Emergency Social Fund for Health, Tierra Blanca, El Salvador, 8 Johns Hopkins University Bloomberg School of Public Health, Baltimore, Maryland, United States of America, 9 Inter-American Institute for Global Change Research (IAI), Montevideo, Uruguay irene.torres@octaedro.edu.ec Abstract The absence of a chronic kidney disease (CKD) registry in Ecuador makes it difficult to assess the burden of disease, but there is an anticipated increase in the incidence of CKD along with increasing diabetes, hypertension and population age. From 2012, augmented funding for renal replacement therapy expanded dialysis clinics and patient coverage. We conducted 73 in-depth sociological interviews with healthcare providers in eight provinces and collected quantitative epidemiological data on patients with CKD diagnoses from six national-level databases between 2015 and 2018. Datasets show a total of 17,484 dialysis patients in 2018, or 567 patients per million population (pmp), with an annual cost exceeding 11% of Ecuador’s public health budget. Each year, there were 139–162 pmp new dialysis patients, while doctors reported waiting lists. The number of patients on peritoneal dialysis was static; those on hemodialysis increased over time. Only 13 of 24 provinces were found to have dialysis services, and nephrologists were clustered in major cities, which limits access, delays medical attention, and adds a travel burden on patients. Prevention and screening programs are scarce, while hospitalization is an important reality for CKD patients. CKD is an emerging public health crisis that has increased dramatically over the last decade in Ecuador and is expected to continue, making coverage for all patients impossible and the current structure, unsustainable. A patient registry would help health policymakers and administra- tors estimate the demand and progression of patients with consideration for comorbidities, disease stage, requirements and costs, mortality and follow-up. This should be used to help identify where to focus prevention and improved treatment efforts. Organized monitoring of CKD patients would benefit from improvements in patient referral. Community-based educa- tion and prevention programs, the strengthening of primary healthcare capacity (including basic routine tests) and improved nephrology services are also urgently needed. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Torres I, Sippy R, Bardosh KL, Bhargava R, Lotto-Batista M, Bideaux AE, et al. (2022) Chronic kidney disease in Ecuador: An epidemiological and health system analysis of an emerging public health crisis. PLoS ONE 17(3): e0265395. https://doi.org/10.1371/journal. pone.0265395 Editor: Pasqual Barretti, Universidade Estadual Paulista Julio de Mesquita Filho, BRAZIL Received: April 8, 2021 Accepted: March 1, 2022 Published: March 16, 2022 Copyright: © 2022 Torres et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Data underlying this study cannot be shared publicly due to Ecuador government regulations that they are requested directly. Direct requests to the Ministry of Public Health can be made by contacting ventanillaunica. msp@msp.gob.ec and to the Ecuadorian Institute of Social Security at https://www.iess.gob.ec/ contactos. PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 1 / 21 PLOS ONEFunding: IT, AMSI, ML, SN, KB and RS were supported by a grant from Dialysis Clinic, Inc (DCI), award number 84232. Competing interests: The authors have declared that no competing interests exist. Chronic kidney disease in Ecuador: An epidemiological and health system analysis Introduction Chronic kidney disease (CKD), defined as “abnormalities of kidney structure or function, present for >3 months, with implications for health”, is considered a globally important non- communicable disease (NCD). In 2017, over 9% of the global population, approximately 700 million people, were estimated to suffer from CKD, making it the twelfth leading cause of death [1]; another study estimated worldwide prevalence at 11–13% in 2016, making CKD more common than diabetes [2]. In Andean Latin America, CKD was the fifth leading cause of death in 2017 [3]; in Ecuador, it was estimated as the fourth leading cause of death and fifth leading cause of premature mortality [4]. CKD is a huge burden to health systems. Although patients with CKD stage 5, or end-stage renal disease (ESRD), comprise only 0.1–0.2% of developed countries’ population, they require 2–3% of healthcare spending [5]. This cost is attributed to the need for renal replacement ther- apy (RRT)—regular dialysis or transplantation—to survive. In low- and middle-income coun- tries, it is difficult to assess the exact burden of CKD, but there is an anticipated increase in the incidence of CKD along with diabetes, hypertension and population age [6, 7]. The estimated prevalence of ESRD in Latin American is 660 pmp [8]. Also, shortage of nephrologists is not uncommon [9]. Most preventable deaths from ESRD patients lacking access to RRT occur in low and middle-income countries [6], affecting population health, public health budgets and livelihoods. Ecuador had a population of 16.8 million in 2017, and there were 5739 estimated CKD deaths and 1.2 million estimated prevalent CKD cases [3]. In 2008, Ecuador’s government recognized CKD as a ‘catastrophic illness’ and from 2012 committed to guarantee medical attention to CKD patients, including dialysis during ESKD [4]. Increased patient coverage resulted in better reporting and capture of CKD but the lack of a systematic registry in accordance with international standards [8] makes it impossible to assess true patterns of CKD epidemiology. Assessing the true incidence and prevalence would help determine resource allocation for CKD management, mitigate disease burden, and reduce lost productivity. According to the Ministry of Public Health (MSP), an estimated 33,000 peo- ple in 2015 were at CKD stage 5 in Ecuador and 45% of all CKD patients in stages 4 and 5 (roughly 30,000 people) were expected to die due to unavailability of RRT [4]. Research in neighboring Peru reported a similar effect on CKD mortality from insufficient access to dialy- sis in parts of the country [10]. The structure, policies and functioning of national health systems impact CKD outcomes. Ecuador’s health sector is highly fragmented [11]. 41.6% of the population (approximately 7 million) had public health insurance coverage in 2019 [12] through the Ecuadorian Institute of Social Security (IESS in Spanish) and Farmer’s Social Security (SSC in Spanish). An estimated 41.4% of total health spending in Ecuador is out-of-pocket [13], with primary-level care being largely provided by newly-graduated medical professionals as part of the MSP’s one-year com- pulsory service program. Access to medical care in Ecuador continues to be a challenge for many, especially in rural areas. Although Ecuador is classified as an “upper middle-income country” by the World Bank [14], 25% of the population live below the poverty line [15], which is defined as a monthly household income of US$84.82 per person. The extensive economic costs of CKD can push entire families into poverty [16], while poverty negatively impacts CKD via poor diet, hazard- ous occupational conditions, psychosocial stress and sub-optimal access to healthcare [17]. Lack of adherence to therapy in CKD negatively impacts quality of care and life [18, 19]. Poor access to RRT increases mortality in advanced CKD. Limited screening and surveillance, delayed medical care, inadequate training of medical practitioners, and the lack of a national registry limits opportunities for prevention, early detection and management [20]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 2 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis Public health implications of CKD have been explored for limitations and challenges to the nephrology workforce [9], physician perspectives on the burden and possible causes of CKD [21] or on care planning for patients [22], and the need for improving awareness and screening for prevention [23]. Addressing risk factors such as overweight and obesity through integrated or community-based interventions has also gained attention in the past years [24–27], as has the need to prevent the rising prevalence of chronic kidney disease of unknown origin (CKDu, or Mesoamerican nephropathy) in Latin America [28]. Studies on CKD patients in Ecuador are limited; the most updated data are from 2014, and are unofficial, relying on practitioners’ estimates collected through a survey [29], which is insufficient to determine health system needs. Reports on outstanding government debt with dialysis providers in recent times point towards financial pressure of ESRD likely escalating in the country [30–32]. We undertook an exploratory, integrated sociological and epidemiological study of CKD in Ecuador with the following aims: (1) Estimate the burden of CKD by analyzing databases and other publicly available epidemiologic information; (2) Understand mechanisms of healthcare delivery and its effect on CKD; and (3) Explore the perceptions and experiences of CKD among nephrologists and other physicians. Methods Our study combined qualitative data from healthcare provider’s interviews and quantitative national-level CKD patient data with a review of grey and scientific literature published on CKD and health systems in Ecuador from the MSP and other sources. In this manuscript, we present results in an integrated narrative, organized by the most salient thematic categories that emerged in the data analysis. Ethics Ethics committee approval (CEU-084) was obtained from the Universidad Internacional del Ecuador; the Institutional Review Board (IRB) of the State University of New York-Upstate Medical University deemed the study protocol exempt. All interviewees were adults (18 years of age or older) and no personal identifying information was recorded. Verbal informed con- sent was obtained by the study investigator (IT), and recorded in audio recordings of the inter- views prior to conducting the interview. Due to the conversational approach used with the interviewees and, potentially, cultural acceptance of this approach, verbal consent was most appropriate and written consent was not deemed necessary by the IRB committee. The epide- miological data used in this analysis were anonymized with a unique identifier to allow for duplicate entries to be removed. Epidemiological data & analysis Several national-level datasets were collected for analysis in this paper. There were five datasets from the employee public insurance systems in Ecuador (IESS, SSC, Armed Forces Institute of Social Security, ISSFA, and the Police Force Institute of Social Security, ISSPOL), including retired and dependent persons in these systems, for 2015–2018. These data included all hemo- dialysis visits at IESS facilities (IESS Hemodialysis), all dialysis procedures (from any facility) billed to the insurance system (IESS Procedures), all visits to an IESS hospital with an 10th International Classification of Disease (ICD-10) code N18 diagnosis (IESS Hospital), all diag- noses of ICD-10 code N18 billed to the insurance system (IESS Diagnosis), and all external (non-IESS facility) consultation or emergency room visits billed to the IESS system with an ICD-10 code N18 diagnosis (IESS Externals). Another dataset from the public health care PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 3 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis system in Ecuador included all patients with an N18 diagnosis who used MSP services from 2014–2018. All data were anonymized. Information on dialysis patients in the IESS system was compiled from the IESS Hemodial- ysis and IESS Procedures datasets. Individual identification numbers, which were previously anonymized, were compared to eliminate duplicates between datasets and from repeated visits. Summary characteristics (sum and proportion by type of visit, patient sex and patient age cate- gory, and mean number of annual visits per patient) were calculated by year at the national level. New patients in the dialysis system were calculated by matching patient identification numbers across years then counting unique patient numbers each year. To project the number of dialysis patients in Ecuador in 2023, IESS and MSP data from 2018, and IESS data from 2015–2017 were used to estimate total (IESS and MSP) dialysis patients for 2015–2018, then total patients in 2023 and a 95% confidence interval for this total were estimated using damp- ened Holt’s method [33]. Total IESS patients with a CKD diagnosis were estimated by combining the IESS Hospital, IESS Diagnosis, and IESS Externals datasets. Individual identification numbers were compared to eliminate duplicates between datasets and from repeated visits. Demographic characteristics (sum and proportion by patient sex and age category) were calculated by year. This list was cross-checked with the list of dialysis patients in the IESS system to determine the number of individuals with CKD on dialysis in the IESS system. Total MSP patients with a CKD diagnosis were obtained from the MSP dataset. Demographic characteristics (sum and proportion by patient sex and age category) were calculated for each year. The IESS Hospitalized dataset was used to assess the number and types of comorbidities among patients with a CKD diagnosis. These include patients with CKD who were hospitalized for any reason. Summary characteris- tics (total hospitalizations, mean duration of hospitalization, range of hospitalization duration, total patients hospitalized, mean number of hospitalizations per patient, and range in number of hospitalizations per patient) were calculated by year. The five most common comorbidity ICD-10 codes were determined by looking at total comorbidities. Dialysis type was only available for IESS dialysis patients. To estimate the number of patients on hemodialysis and peritoneal dialysis in the MSP system, the proportion of hemodi- alysis and peritoneal patients in the IESS system were applied to the MSP dialysis patients. For the IESS system, dialysis patient travel burden was determined using the IESS Diagnosis and IESS Procedures datasets. The location of the clinic where the patient received their initial referral for dialysis was assumed to be within their home province. For each dialysis visit in 2015–2018, the location of the dialysis clinic was compared to the home province and was cate- gorized as the same province, a neighboring (border-sharing) province, or from further away. For the MSP system, patient residence data was available; the location of residence was com- pared to the location of dialysis clinics during all dialysis visits in 2014–2018, to create the same categories. Information on health clinics and dialysis services were obtained from the 2008 and 2017 Health Resources and Activities Survey (Recursos y Actividades de Salud, in Spanish) con- ducted by INEC (https://www.ecuadorencifras.gob.ec/actividades-y-recursos-de-salud/). All data preparation, analyses, and visualization were performed in R version 4.0.3 [34], in RStudio version 1.3.1093 [35], using packages forecast [36], ggplot2 [37], ggpubr [38], haven [39], maps [40], maptools [41], raster [42], RColorBrewer [43] and rgdal [44]. Qualitative interviews A convenience sample of 73 healthcare providers was selected. First, through discussions with the Nephrology Society of Ecuador, IESS, and MSP from the 286 nephrologists registered to PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 4 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis practice, to include as many geographic locations as possible, in both public and private set- tings. Second, we added physicians and health administrators in areas where there are no nephrologists based on locations recommended by participants. We conducted in-depth interviews with the healthcare providers in eight provinces. This was divided into two phases. The first, spanning 2 weeks in June 2019, included 34 nephrolo- gists and health administrators in the cities Guayaquil (n = 19) and Quito (n = 15), which are the largest cities and have the highest number of nephrology clinics and services. The second phase spanned 4 weeks between July and August 2019 and included 39 nephrologists, health administrators and physicians in the provinces of Manabı´, Guayas, Los Rı´os, Cañar, Esmeraldas, Santo Domingo de Los Tsa´chilas, Pichincha and Imbabura. Participant details are in S5 Table. Phase 1 interviews were conducted in person by two authors (IT and AMS) in 20 hospitals or clinics in Guayaquil and Quito. Phase 2 interviews were conducted in 24 hospitals or clinics, in person (n = 28) or by telephone (n = 11) by one author (IT). During interviews, we intro- duced study objectives and obtained verbal consent to record the interview. Our interview template included 14 open-ended questions and probes for this study (See S1 Text) including: professional background, common causes of kidney disease, knowledge and experience with CKD, characteristics of kidney disease patients, standard diagnosis and treatment strategies, and barriers to early detection, treatment and care. All interviews were conducted in Spanish, audio-recorded, transcribed and translated into English. IT and KB analyzed them. After transcription, all respondents were identified through a code number to ensure anonymity. Transcripts were subsequently coded using a modified grounded theory approach to develop our thematic categories. Analysis involved open coding in the margins of the transcripts. Once all transcripts had been coded and a list of codes devel- oped, the transcripts were revisited using the selective list of codes. Grouping of codes into concepts and then categories relied on memos and sorting. The large sample size supports a reasonable level of saturation. Patient and public involvement This paper does not report a clinical study, so no patients were involved. While the public was not involved in the development and design of the study, we will draft a report of results in Spanish for the Ministry of Public Health of Ecuador and study participants, which we will also disseminate among specialized and other media, and public health professional and advo- cacy organizations. Results Results are divided into four main sections: (1) Increasing incidence and prevalence of CKD and ESRD, where we also analyze available data on comorbidities; (2) Public health, hospitali- zation and dialysis service expenditures; (3) Nephrologists, access to treatment and the prob- lem with dialysis, where we also analyze the determinants and guidelines of dialysis treatment, and available data on transplants; (4) Early diagnosis and prevention, where we also analyze the involvement of primary care physicians in detection. In each section, we present the quan- titative epidemiological data and the qualitative sociological data concurrently and in an inte- grated manner. Increasing incidence and prevalence of CKD and ESRD Increasing incident/prevalent dialysis patients. According to Ecuador’s Health Activi- ties and Resources Database [45], there were 59,183 dialysis treatments in 2008, which is about 400 chronic patients, assuming 3 treatments per week. Dialysis was unaffordable to most PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 5 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis patients and ESRD resulted in premature death before the policy change in 2012 that expanded care: “All the patients who died [before] without diagnosis in their home or any other hospital, became supported by the ministry, and so we had a huge flow of kidney patients that before died without treatment. . .all those repressed patients suddenly entered the system.” (Quito, Participant 13) IESS and MSP datasets show a total of 17 484 dialysis patients in 2018 (9641 in IESS and 7843 in MSP), or 567 patients per million population (pmp). Summary statistics for IESS patients are in S1 Table; detailed data on MSP dialysis patients were not available. Number of patients receiving dialysis annually within the IESS system from 2015–2018 are shown in Fig 1; during this period, the total number of patients increased by more than 2600. Each year, there were 2311–2569 new dialysis patients (i.e. patients who had not been receiving dialysis the pre- vious year), or 139–162 pmp, likely indicating a high mortality rate among this patient group. Fig 1. Number of IESS patients receiving dialysis, 2015–2018. Plot of patients in IESS system receiving dialysis each year from 2015–2018, by type: peritoneal (blue) and hemodialysis (red). The number of new patients annually is represented by the black line. https://doi.org/10.1371/journal.pone.0265395.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 6 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis The number of patients on peritoneal dialysis was static (from 350 to 357 patients), while those on hemodialysis increased over the 4 years (from 6663 to 9284 patients). If this annual rate of increase continues, the number of CKD patients demanding dialysis by 2023 could reach at least 21 365 (95% confidence interval: 18 255–24 474) or 1170 pmp. Increased dialysis access has not solved the “silent epidemic” of CKD; most interviewees viewed kidney disease as an emerging public health crisis that increased dramatically over the last decade and is expected to continue: “the [number of patients] is growing abysmally, just to give you an example. In 2009, there were only 4 dialysis centers in Guayaquil. Now. . .I counted 16 last year” (Guayaquil, Participant 3). The number of health facilities offering hemodialysis has grown by 420%, from 19 in 2008 to 76 in 2017 [45, 46] (S4 Fig), as has the number of hemodialysis machines. Whereas dialysis centers previously had few (3–5) machines, inter- viewees reported that now many have upwards of 30 each, and most shifts were full. The mean number of machines per facility is 16 [46], and the medical directors who were interviewed reported patient waiting lists. MSP and IESS databases show increasing CKD diagnoses (Figs 2 and 3). Among patients in the IESS system, the number with CKD increased by 99.3% between 2015–2018, from 14 747 Fig 2. IESS patients with a CKD diagnosis. For each year, the number of individual patients with an N18 diagnosis who received care (emergency, inpatient, or outpatient) billed to the IESS system is given. Colors denote dialysis status. https://doi.org/10.1371/journal.pone.0265395.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 7 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis Fig 3. MSP patients with CKD diagnosis. For each year, the number of individual patients who received care for CKD (emergency, inpatient, or outpatient) at an MSP-associated facility is given. https://doi.org/10.1371/journal.pone.0265395.g003 to 29 418. Summary statistics for these patients are in S2 Table. In the MSP system, the number with CKD increased by 107.3% between 2014–2018, from 14 525 in 2014 to 30 117 in 2018. Summary statistics for these patients are in S3 Table. CKD and comorbidities. Interviewees reported that the rise of CKD was linked to the parallel rise in diabetes and hypertension; interviewees estimated that 60% to 90% of all CKD cases in Ecuador were caused by these two NCDs. Causes of CKD mentioned by interviewed nephrologists centered heavily on a syndemic relationship between kidney disease, diabetes, hypertension, obesity, socio-economic status and environmental factors, such as diet (espe- cially salt intake), water quality (brackish, with sediment), chemical exposure and genetic fac- tors. Primarily, the rise in CKD was seen as a result of the rise in other NCDs: “Go to my [dialysis] ward at some point, in one shift, 100% are diabetics” (Guayaquil, Participant 5). Many complications are triggered by kidney damage, and nephrologists often saw diabetic and hypertensive patients who were referred to them after evaluation by cardiologists, endocrinol- ogists, internal medicine specialists or emergency physicians. PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 8 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis “There are prevention programs in the ministry, early care of diabetes, hypertension, obesi- ty. . .but not chronic kidney disease. Why? Because within the five or seven causes [of death], in the society [of nephrologists], and PAHO and WHO, chronic kidney disease is not the main cause. Rather it is presented as a consequence [of other conditions like diabetes and hyperten- sion]. [But] now we know that it is not only a consequence. . .” (Guayaquil, Participant 17). Comorbidities were common among hospitalized IESS CKD patients. The number of CKD patients with comorbidities are in S1 Fig. In 2018, 22.1% had no comorbidities, 15.0% had one comorbidity, and 62.9% had two or more comorbidities. The most common comorbidities among patients in 2018 were primary hypertension (7.1%), type-II or type-I diabetes (3.4% each), other urinary tract disorders (2.7%), heart failure (1.7%), and bacterial pneumonia (1.0%) (Fig 4). Public health, hospitalization and dialysis service expenditures With the increasing number of dialysis patients, interviewees lamented the current financial burden of dialysis, and many had dire predictions of the fiscal pressure of these costs in the immediate present and future: “[dialysis] is a black [fiscal] hole, you put money in it, the money goes, the money goes, the money goes” (Guayaquil, Participant 5). This was reflected in a domi- nant belief that the exclusive focus on dialysis has done little to address the risk factors and causes of late-stage CKD. These costs are largely borne by social security schemes that operate under the IESS and MSP, and to a lesser extent, private insurance companies. According to our analysis (S4 Table), at an annual cost of US$17,472/patient for hemodialysis and US $14,940/patient for peritoneal dialysis, 11.04% of Ecuador’s US$2.665 billion public health budget [47] was spent on dialysis alone in 2018. Because dialysis machines are imported, some nephrologists interviewed believed that spending limits needed to be better defined and younger patients prioritized for dialysis and transplants. Diabetic patients were also viewed as a transplant priority, because of their high treatment cost; CKD expenses include frequent and long-term hospitalizations for patients on hemodialysis. Many clinicians felt that if unaddressed, CKD poses a major fiscal challenge to the health system: “. . .we do not have the money to treat all these people, and the budget that one disease eats up, they are going to take away from another. . . the Ecuadorian state cannot afford not to treat a patient with chronic kidney disease. . . the [total health] budget is going to become so reduced [due to this], you have no idea” (Quito, Participant 7). Hospitalization is an important reality for CKD patients. Among patients in the IESS sys- tem, we found that 43.2–48.6% of those with CKD were hospitalized for at least one day annu- ally. In 2018, we found that 3,427 CKD patients were hospitalized, for an average duration of 8.4 days (Table 1). Nephrologists, access to treatment and the problem with dialysis Access to nephrologists and dialysis services. The crisis of CKD means that the primary, secondary and tertiary care networks are overwhelmed with patients seeking dialysis, as reported by our interviewees. Government support for dialysis encouraged proliferation of PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 9 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis Fig 4. Comorbidities among CKD patients in IESS, 2015–2018. Annual number of hospitalized CKD patients in IESS system with comorbidities, 2015–2018. Comorbidities were determined by individual ICD-10 codes. Many patients had more than one comorbidity. https://doi.org/10.1371/journal.pone.0265395.g004 dialysis centers and access to treatment has increased in major cities (Guayaquil, Cuenca, Quito) as well as in many provincial capitals. While this addressed some access barriers, inter- viewees explained that dialysis centers remain concentrated in urban areas, with long delays to see nephrologists. In 2019, there were 286 registered nephrologists in Ecuador for a population of 17.2 million people that year, or 16.6 nephrologists pmp. Nephrology is a relatively new spe- ciality in Ecuador and most interviewees reported having studied abroad (S3 Table). Table 1. Hospitalization of IESS-associated patients with CKD. Total Hospitalizations Total Persons Hospitalized Mean Number of Annual Hospitalizations per Patient Range in Number of Annual Hospitalizations per Patient Mean Duration of Hospitalization Range of Hospitalization Duration https://doi.org/10.1371/journal.pone.0265395.t001 2015 3,529 2,682 1.3 1—8 2016 3,417 2,707 1.3 1—10 2017 3,973 3,094 1.3 1—12 2018 4,641 3,627 1.3 1—12 9.0 days 1—311 days 8.9 days 1—150 days 8.4 days 1—168 days 8.4 days 1—120 days PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 10 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis Fig 5. Map of facilities offering dialysis, 2017. As of 2017, 76 health facilities offered dialysis. These facilities are concentrated in urban areas, in 21 municipalities (cantons) across 13 provinces, as indicated in the map. Guayaquil Canton has 25 such facilities, Quito Canton has 17, and Cuenca Canton has 6. https://doi.org/10.1371/journal.pone.0265395.g005 Quevedo and Esmeraldas, cities of approximately 210,000 people, did not have a nephrolo- gist in the public system during our study period. Some nephrologists we interviewed empha- sized the need for nephrologists to be based in “the provinces” to ensure that medium-sized cities are sufficiently covered—otherwise the financial burden of travel for patients can be overwhelming: “In the rural areas, there is not a single dialysis unit.” (Quito, Participant 13). As of 2017, there were dialysis centers in only 13 of 24 provinces [46]. A map showing the national distribution of facilities is shown in Fig 5. The inter-provincial travel necessary for IESS and MSP dialysis patients to access dialysis services is in S2 and S3 Figs, respectively. Interviewees explained that healthcare for ESRD patients in rural, peripheral and small cit- ies may not begin at primary care level (S4 Fig). Patients commonly presented in ESRD directly to emergency services at local hospitals that lacked a nephrologist or dialysis services, and may die in emergency or intensive care if there are no openings in a dialysis clinic. In our interviews, nephrologists insisted on the need for individualized care and not a stan- dardized approach: “human beings cannot be standardized. . .we think differently, here dialysis is according to what we see is the risk profile of the patient” (Quito, Participant 13). The cost of dialysis includes treatment and a support package. In private clinics, patients also have access PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 11 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis to monthly nutritionist, psychologists, social workers and other medical specialists such as vas- cular surgeons. By contrast, such care appears to be infrequently available at many MSP and IESS facilities. Most nephrologists, in our interviews, reported low capacity in both private and public systems, with patients not being followed up as indicated in the new CKD guidelines for non-dialysis patients [4]. Life expectancy and the social determinants of dialysis treatment. Mortality in CKD can be improved with appropriate management early in the disease course. In our interviews, the low life expectancy of many patients was blamed on the asymptomatic presentation of early-stage CKD, making it difficult for patients or primary doctors to detect it: “so many patients arrive. . .when 90% of kidney is damaged” (Quito, Participant 3). Interviewees reported that the lack of symptoms also contributes to a series of problems in patient referral and health screening behavior: people do not accept their diagnosis, especially in the early stages, seek alternative therapies, and as much as 20% may be lost to follow-up until the very late stages. Some nephrologists stressed that patients have a great deal of fear around initiating dialysis: “They arrive with a lot of fear, a lot of disinformation. . .the patient goes through limbo for months sometimes. We just have a patient who was admitted, who left [X] hospital and has spent two months without dialysis. . .Why? He is terrified of dialysis because he had a bad hos- pital experience [in the past]” (Quito, Participant 13). Interviewees described the strong psychological and mental health dimension to CKD [17, 18]. They found that their patients frequently viewed dialysis as a “death sentence” and treat- ment strained the whole family emotionally and financially. Tragically, some patients are aban- doned by their families and are depressed; as one clinician stated: “those who do not accept the disease are the fastest to go.” Interviewees identified barriers to health care for CKD patients including the lack of access to health insurance, three times weekly travel to dialysis centers (most providers do not cover travel), specific bureaucratic and administrative issues (e.g. referral to a nephrologist is only valid for 2 months and it can take longer for patients to book an appointment within this win- dow) and the overarching challenge of poverty and low socio-economic status: “people are poor here, they do not have resources. . .you see the emergency room is full the first days of the month and at end of the month because they have money for the [bus] ticket. . . sometimes we do not have 100% stock of the medicine required. . .” (Guayaquil, Participant 17). Dialysis treatment guidelines. To address the need for standards of care in CKD, in 2018 the MSP published a set of clinical practice guidelines of CKD, which are described in the pub- lication as “general recommendations” based on international guidelines and scientific evi- dence by public and private specialists. Although the new guidelines do not explicitly discuss this, nephrologists interviewed emphasized widespread preference for hemodialysis (calcu- lated at 90% of total) rather than peritoneal dialysis. Regionally, the average ratio is 6.6 hemo- dialysis per peritoneal dialysis patient; in 2018 in Ecuador, we estimated that the ratio was 26.0 (derived from IESS system data). The focus on hemodialysis means that other modalities of RRT such as peritoneal dialysis and transplants are not encouraged. Many clinicians recommended encouraging peritoneal PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 12 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis dialysis (which is often used for children), including using it as a front-line treatment for all patients. However, it is not standard practice for patients to be informed of this option: “the majority of patients who come to hemo[dialysis] are almost not aware of peri[toneal], especially acute emergency cases when they begin [treatment only] after arriving to emergency” (Guayaquil, Participant 3). In our interviews, many nephrologists stressed that, compared to hemodialysis, peritoneal is less aggressive, can be performed at home and can prolong life expectancy, espe- cially for older patients, those with comorbidities and young children. Hemodialysis requires patients to visit dialysis centers three times weekly for roughly four hours each visit. Monthly exams are also required, as is a regimen of medications for complica- tions and other chronic conditions. Peritoneal dialysis is performed at home and requires daily dialysis of approximately 6–8 hours overnight or 4–6 exchanges during the daytime. It requires aseptic measures and training, making family involvement crucial (for hemodialysis, family members must arrange transportation and accompany the patient, since patients are often ill post-dialysis). Other factors limiting the uptake of peritoneal dialysis, according to our interview participants, included: medical team preferences, limited numbers of automated machines, and potential risks, including catheter infections or dysfunction, peritonitis, and anemia. Our inter- views described reluctance towards peritoneal dialysis, arguing that patients who are not careful can get infections or because they do not have the necessary hygiene conditions at home. Con- currently, nephrologists emphasized that managing peritoneal treatment is not difficult for fam- ily members to learn and should be expanded: “If you know how to handle an Android phone you know how to operate a peritoneal dialysis machine, and everyone has an Android telephone nowadays, at least that access is universalized, so it is possible” (Quito, Participant 1). Transplants. Nephrologists reported underuse of transplants and suggested that restruc- turing the kidney treatment cascade to include transplants could contribute significantly to life expectancy. According to the National Institute of Organ Donation and Transplants (INDOT, Spanish acronym, 2019), in 2018 only one public hospital and one paediatric public hospital were accredited to perform kidney transplants; we found that there were only 235 kidney transplants in 2018 [48], of which 22 were from living donors, and 226 in 2019, of which 4 were from living donors [49]. We estimated an annual transplant rate of 2.4% of dialysis patients in Ecuador for 2018. Nephrologists emphasized the need to promote a “culture of donation” in Ecuador, since the number of donors has not increased significantly. According to one study [50], kidney transplants went from 58 in 2007 to 249 in 2018. INDOT reported in 2018 there were 1623 patients in the kidney waiting list, with 420 new patients (206 more than the previous year); 44% of those on the waiting list are people between 30 to 50 years of age. Given the limited organ supply, there is a need to develop stronger policy guidelines for trans- plant recipients. Many nephrologists believed that diabetic patients and children should be pri- oritized due to health expense of both patients and positive health outcomes: “the diabetic patient costs 5 times more than any other patient because he has many comorbi- dities. . .and the patient who dies the most in dialysis. 70% die within 5 years, due to cardio- vascular problems, so that patient is the one who would benefit most with a transplant. . .” (Quito, Participant 11) Missing pieces: Early diagnosis and prevention An unintended consequence of the focus on dialysis has been the creation of a “reactive nephrology” and “expensive medicine” that has: “transform[ed] us into hemodialysis doctors. That is the problem. That is a major problem” (Guayaquil, Participant 3). Despite discussions PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 13 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis about an integrated healthcare model and agreement about the critical need for preventative interventions, the predominant public health focus has been on terminal treatment. One clini- cian, reflecting the opinion of others, associated the new Ecuadorian clinical guide for primary care to the US “invasive” terminal approach (which costs more), rather than the European emphasis on more diagnostic testing to identify early CKD cases (Guayaquil, Participant 17). Basic routine testing. Our interviews revealed that simple, low-cost urine tests are not widely used to detect the early stages of kidney damage. Some viewed the urine test as a “liquid biopsy” that provided the “first alert” to emerging kidney damage and should be a routine part of medical checkups; others emphasized the need to integrate proteinuria, creatinine and microalbuminuria (urinary sediment) tests in primary care. There was a widely shared senti- ment that some general practitioners do not request any kidney tests, despite routinely testing blood glucose, cholesterol, and triglycerides. Others emphasized that using urine tests would require educating the physicians to look for protein in the urine, which is a simple, low-cost and non-invasive test. Nephrologists also highlighted the need to address physician interpretation of kidney func- tion tests. They considered an over-emphasis on “normal values” in tests as problematic and the need to calculate kidney function as glomerular filtration rate by validated formulae. MSP’s kidney disease guide for evaluating creatinine levels of early kidney disease lacks speci- ficity in some areas, such as how to evaluate early stages and include multiple factors. Depend- ing on age and gender, and comorbidities such as hypertension and diabetes, values need to be interpreted differently. Ideally, tests should be performed annually to establish baseline kidney function and monitor changes over time. Involvement of primary care physicians. Nephrologists frequently mentioned inade- quate diagnoses and clinical histories from primary care physicians, contributing to the inabil- ity of the health system to address kidney disease at primary and secondary levels of care which has “overwhelmed nephrologists” (Guayaquil, Participant 3). The triage system was seen as “broken”, and nephrologists working with rural patients in cities with 76,000–320,000 people mentioned that not even occupational health physicians at private companies would detect CKD. According to several interviewees, in their experience, up to 100% of patients could present to nephrologists with ESRD, having never received nephrology services before, which is reflected in the few staged CKD individuals in statistics of the National Institute of Statistics and Census (INEC) and the lower number of reported CKD patients compared to MSP and IESS data. Strengthening early detection and prevention would alleviate the strain on nephrologists and address the problem of losing patients to follow-up. Little attention is paid to delaying progression, including the need for dialysis, for example by addressing CKD in ear- lier stages. Some viewed this as a direct result of the “lobbying by companies, the hemodialysis companies” (Guayaquil, Participant 5). Whatever the cause, there was a ubiquitous agreement that prevention is nonexistent and that this reflected a systemic problem in the organization of primary healthcare and health policy: “We have completely deformed the health system. . . Here everything is atomized [put into ver- tical programs] and that is what makes us have nothing, in the end. . .we cannot prevent [our patients] from advancing to stage 5 [since there is no focus on prevention].” (Quito, Participant 13) To many of the nephrologists we interviewed, the main barrier to early detection was lack of clinical knowledge at lower healthcare levels due to gaps in medical education. Kidney func- tion testing by means of a simple blood test is sparse at the level of primary and secondary care PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 14 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis despite the presence of significant co-morbidities such as hypertension and anemia. According to our interviewees, there is a need to re-design nephrology training for medical students, interns and general practitioners, so they have the capacity to recognize early CKD. Community-based prevention. Our interviews highlighted the need for community- based prevention programs. Although individual nephrologists mentioned some local preven- tion programs, these were rare, not systematic and many lamented the lack of attention and funding to this critical area. Nephrologists trained abroad emphasized the importance of mass media communication; those working with indigenous populations highlighted the need to work in collaboration with local leaders because traditional medicine is highly valued among them. Some recommended that awareness campaigns become integrated with diabetes and hypertension prevention; others, that these efforts are not reduced to specific events but take place throughout the year. While some dialysis companies have considered prevention, nephrologists expressed concern that the private sector could eclipse national programs to reduce the need for dialysis in the first place. This included community-level screening, urine tests, and an annual kidney profile for high-risk groups, especially those with diabetes and hypertension and those working with agrochemicals. This could be based on existing models for maternal and child health: “. . .95% of kidney diseases [can be identified] in the urine test, which costs 50 cents. Why is it not done? We could just do a screening in the schools, in the community. . .we have a screen- ing that costs 50 cents and that should be used every year at all levels. . .to screen, screen at an extremely low price. And if we [combine urine tests with] blood pressure tests. . . almost 100% of kidney diagnoses would be made.” (Quito, Participant 13) Nephrologists also reported a lack of accessible, community-level information about kidney disease, contributing to low knowledge among patients and the wider community. Since CKD is only symptomatic in later stages, many patients are “surprised” and “skeptical” when nephrologists talk about the seriousness of the disease: “[Some] patients on dialysis do not know about their illness or their treatment, not even in a general way. . . they are totally unaware. So, I think that. . .prevention is failing, it is failing a lot” (Quito, Participant 5). Another nephrologist reflected: “In Spain [where I trained], it is impressive because people would come and say ’I have this, I have this, I have to take this [medication] and this other’ and they would almost come with the diagnosis. Here people arrive, ’I take a blue pill but I do not know anything about what it is’. . .it is important to know the basics of their disease, know how to take care of themselves, what to take, what not to take. . .” (Quito, Participant 3). Discussions about prevention also stressed the need for population-level lifestyle changes and education for changing risk behaviors, specifically for diet (reducing salt), drinking more fluids, and less self-medication with NSAID drugs. Some nephrologists associated specific hot- spots of CKD with regional diets in high-salt consumption and junk food such as soda; as one stated: “Look how obesity has increased. . .it was not like that in our country [before]. There are so many obese children, who eat on the street, fast foods, and that creates a high risk of kidney failure, hypertension, [and being] overweight. . .” (Guayaquil, Participant 14). PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 15 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis Discussion This paper set out to examine the three interrelated aspects of chronic kidney disease (CKD) in Ecuador: burden of disease, healthcare access and delivery mechanisms, and the perceptions and experiences of nephrologists and other physicians. This research has uncovered critical CKD policy and practice gaps in Ecuador and provided insights into how to best address them. We observed that the number of patients diagnosed with CKD and end stage renal disease (ESRD) has increased over time in Ecuador, with many arriving directly to renal replacement therapy during emergency care before detection in primary care. The recognition of CKD as a catastrophic illness deserving free-of-cost coverage involves redirecting a significant portion of public health expenditure to dialysis services, which may soon become unsustainable. Hospi- talization costs additionally contribute to these expenditures. Repeated reports of government nonpayment to dialysis providers show that pressure on the public health sector is already at a breaking point [30–32]. Our research has revealed a relative neglect of basic routine tests, primary healthcare physi- cians and community-based prevention efforts to address the CKD problem in Ecuador. The focus appears to remain on a curative, hospital-based approach. Instead, community-based education and prevention programs that are adequately funded could systematically provide community-level information or, better yet, design and implement strategies in collaboration with local leaders. Furthermore, if the primary level of care were strengthened through better training and resources, media-based campaigns to create awareness could also motivate people to seek medical attention, where they would be screened periodically for comorbidities that put them at risk of CKD. Once patients are diagnosed with CKD, similar to other countries in Latin America [10], healthcare coverage is not available for all of them in Ecuador, where it is estimated that only 36% of ESRD patients are in renal replacement therapy and 45% of CKD patients in stages 4 and 5 may die before receiving renal replacement therapy. Not all provinces have nephrologists while some only have them available in the private sector. Depending on their place of resi- dence, patients may have to travel long distances to access dialysis services. Care restricted to urban areas is a major barrier to access and implies costly and lengthy transportation as well as delays in medical attention. Other countries in the region such as Argentina and Uruguay have 34.54 and 46.98 nephrologists (per million people) pmp, respectively [9], which is more than double Ecuador’s 16.6 nephrologists pmp. Because chronic kidney disease and demand for renal replacement therapy are prevalent in all provinces of Ecuador, but there are limited opportunities to specialize in nephrology locally, the government should consider supporting in-country specialization programs as well as redistributing nephrologists or defining another strategy to secure nephrology consultation through the public sector, at least in main cities. In addition, peritoneal dialysis is underutilized in Ecuador, as also found in other Latin American countries [29], and represents an opportunity to improve quality of life and life expectancy, and reduce patient financial burden. CKD cannot be managed without a national patient registry including morbidity and mor- tality data. Kidney disease registries permit the estimation of demand, levels of comorbidities, the calculation of costs and health care services, and the identification of geographic hotspots. The health system would benefit if policy clearly stated CKD definitions and required patient data, including disease stage and necessary follow-up statistics. This is an urgent, short-term feasible policy step to improve CKD policy and management in Ecuador. Together with improvements in referral and counter-referral with considerations for data confidentiality and security, this would facilitate more organized monitoring of patients and provide evidence and data for health administrators and policymakers. PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 16 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis Our research has identified several urgent policy interventions that can help address this looming health crisis. We summarize these here. First, there is a need, to improve awareness and knowledge of CKD and related NCDs in the general population. Second, changes in medi- cal education curricula and primary care guidelines, together with professional training, are also needed to improve detection. Third, the government should invest in primary health care diagnosis and treatment to improve coverage and patient referral at the community level, both prevent and impede rapid progression of CKD, and help build a comprehensive patient regis- try for adequate monitoring and follow up. Fourth, conditions related to CKD such as over- weight and obesity should be addressed through public health measures, (e.g. food labelling, restrictions on unhealthy food advertising and sale, and media campaigns to change consumer behavior) [27], and the integration of interventions in development programs [24] and com- munity-based health promotion [25, 26] should be seriously considered. Finally, since chronic kidney disease of unknown origin (CKDu), or Mesoamerican nephropathy, has emerged as a major public health concern in populations with similar labor conditions as those experienced in coastal Ecuador, future research is needed to determine whether it is present in Ecuador, as has been documented in Central America and other countries. This study has several limitations. Our estimate of 567 dialysis patients pmp in Ecuador does not include all possible ESRD patients in the country including undiagnosed or unre- ported cases, and patients possibly utilizing private healthcare systems We also lack data on the number of patients who need dialysis but are on waiting lists and patients relying solely on the private sector. As in other Latin American countries [29], the lack of data on CKD staging and RRT waiting lists limits our assessment of CKD and ESRD. The datasets did not have the detailed level of medical records, meaning we had an incomplete picture of each patient’s health. Data on comorbidities was restricted to those in IESS system, and could also be under- reported, therefore their burden among CKD patients cannot be quantified more precisely. However, the datasets analyzed here represent the most complete national data available, com- piling for the first time data from both the MSP and IESS systems. Supporting information S1 Table. IESS patients on dialysis, 2015–2018. Data are provided as numbers or percentages as indicated. (DOCX) S2 Table. IESS patients with CKD, 2015–2018. Data are provided as numbers or percentages as indicated. (DOCX) S3 Table. MSP patients with CKD, 2014–2018. Data are provided as numbers or percentages as indicated. (DOCX) S4 Table. Estimated cost of dialysis to the Ecuadorian public health system. (DOCX) S5 Table. Characteristics of interview participants. (DOCX) S1 Fig. Comorbidities among IESS patients with CKD, 2015–2018. Comorbidity data were available for IESS patients with CKD who had been hospitalized in each year. The total num- ber of comorbidities for each patient are in the plot. (DOCX) PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 17 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis S2 Fig. Patient travel to IESS dialysis clinics, 2015–2018. Data include all dialysis service vis- its from 2015–2018 for patients in the IESS system, which covers 17 provinces out of 24. The location of their initial referral was compared to visits for dialysis services and whether dialysis was provided in the same province as the referral (green), a neighboring province (blue), or from farther away (red). Patients from Los Rı´os, Chimborazo, and Loja had to travel most fre- quently to another province for dialysis. (DOCX) S3 Fig. Patient travel for MSP services, 2014–2018. Data include all CKD-related visits from 2014–2018 for patients in the MSP system, which covers 23 provinces out of 24 (Gala´pagos province lacks dialysis services). The location of patient residence was compared to visits for CKD-related visits and whether these visits were the same province as the patient’s residence (green), a neighboring province (blue), or from farther away (red). Patients from Bolı´var and Carchi in the Andean highlands, and Zamora Chinchipe and Orellana in the Amazon region, had to travel most frequently to another province for CKD-related service. (DOCX) S4 Fig. Ecuadorian health services network from the perspective of an end-stage CKD patient. SOURCES: Statistical Registry of Health Resources and Activities [46]; Government of Ecuador Official Register Nº 428, Supplement, January 30, 2015; Ministry of Health Techni- cal Guidelines 2014 [Norma técnica Subsistema de referencia, derivación, contrareferencia, referencia inversa y transferencia del Sistema Nacional de Salud]; Licensed kidney health spe- cialized centers 2019 [Centros especializados en salud renal con licencias emitidas]; Interviews. (DOCX) S1 Text. Key informant interview questions in English and Spanish. (DOCX) Acknowledgments We thank Fundacion Octaedro and the Ministry of Public Health for procuring and providing de-identified data, and thank all of the people who participated in the study. Author Contributions Conceptualization: Kevin Louis Bardosh, Ramya Bhargava, Ramon Garcia-Trabanino, Ame- lia Goldsmith, Anna M. Stewart-Ibarra. Data curation: Irene Torres, Rachel Sippy. Formal analysis: Irene Torres, Rachel Sippy, Kevin Louis Bardosh, Martı´n Lotto-Batista. Funding acquisition: Sriram S. Narsipur, Anna M. Stewart-Ibarra. Investigation: Irene Torres, Abigail E. Bideaux. Methodology: Kevin Louis Bardosh, Amelia Goldsmith, Anna M. Stewart-Ibarra. Supervision: Sriram S. Narsipur. Validation: Ramya Bhargava, Ramon Garcia-Trabanino. Visualization: Rachel Sippy. Writing – original draft: Irene Torres, Rachel Sippy, Kevin Louis Bardosh. PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 18 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis Writing – review & editing: Irene Torres, Rachel Sippy, Kevin Louis Bardosh, Ramya Bhar- gava, Ramon Garcia-Trabanino, Anna M. Stewart-Ibarra. References 1. Bikbov B, Purcell CA, Levey AS, Smith M, Abdoli A, Abebe M, et al. Global, regional, and national bur- den of chronic kidney disease, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. The Lancet. 2020 Feb 29; 395(10225):709–33. 2. Hill NR, Fatoba ST, Oke JL, Hirst JA, O’Callaghan CA, Lasserson DS, et al. Global Prevalence of Chronic Kidney Disease—A Systematic Review and Meta-Analysis. PloS One. 2016; 11(7):e0158765. https://doi.org/10.1371/journal.pone.0158765 PMID: 27383068 3. Global, regional, and national burden of chronic kidney disease, 1990–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet Lond Engl. 2020 Feb 29; 395(10225):709–33. 4. Ministry of Health of Ecuador. Prevencio´ n, diagno´ stico y tratamiento de la enfermedad renal cro´ nica: Guı´a pra´ctica clı´nica [Prevention, diagnosis and treatment of cronic kidney disease: Clinical practice guide]. Quito: Direccio´n Nacional de Normatizacio´ n MSP; 2018. 5. 6. Jha V, Garcia-Garcia G, Iseki K, Li Z, Naicker S, Plattner B, et al. Chronic kidney disease: global dimen- sion and perspectives. The Lancet. 2013 Jul 20; 382(9888):260–72. https://doi.org/10.1016/S0140- 6736(13)60687-X PMID: 23727169 Liyanage T, Ninomiya T, Jha V, Neal B, Patrice HM, Okpechi I, et al. Worldwide access to treatment for end-stage kidney disease: a systematic review. The Lancet. 2015 May 16; 385(9981):1975–82. https:// doi.org/10.1016/S0140-6736(14)61601-9 PMID: 25777665 7. Yang C, Wang H, Zhao X, Matsushita K, Coresh J, Zhang L, et al. CKD in China: Evolving Spectrum and Public Health Implications. American Journal of Kidney Diseases. 2020; 76(2):258–64. https://doi. org/10.1053/j.ajkd.2019.05.032 PMID: 31492486 8. Luxardo R, Kramer A, Gonza´lez-Bedat MC, Massy ZA, Jager KJ, Rosa-Diez G, et al. The epidemiology of renal replacement therapy in two different parts of the world: the Latin American Dialysis and Trans- plant Registry versus the European Renal Association-European Dialysis and Transplant Association Registry. Rev Panam Salud Publica Pan Am J Public Health. 2018; 42:e87. https://doi.org/10.26633/ RPSP.2018.87 PMID: 31093115 9. Sharif MU, Elsayed ME, Stack AG. The global nephrology workforce: emerging threats and potential solutions! Clin Kidney J. 2015 Nov 14; 9(1):11–22. https://doi.org/10.1093/ckj/sfv111 PMID: 26798456 10. Herrera-Añazco P, Atamari-Anahui N, Flores-Benites V. Nu´ mero de nefro´logos, servicios de hemodia´ li- sis y tendencia de la prevalencia de enfermedad renal cro´ nica en el Ministerio de Salud de Peru´ . Rev Peru Med Exp Salud Publica. 2019; 36(1):62–7. https://doi.org/10.17843/rpmesp.2019.361.4253 PMID: 31116340 11. Torres I, Lo´pez-Cevallos DF. Institutional challenges to achieving health equity in Ecuador. Lancet Glob Health. 2018 Aug 1; 6(8):e832–3. https://doi.org/10.1016/S2214-109X(18)30245-6 PMID: 30012260 12. Ecuadorian Institute of Social Security. Estimacio´n de poblacio´ n beneficiaria del Seguro General de Salud—julio 2019. Subdireccio´n Nacional de Vigilancia y Gestio´n de la Informacio´n del Seguro de SaludI, ESS; 2019. 13. Chang AY, Cowling K, Micah AE, Chapin A, Chen CS, Ikilezi G, et al. Past, present, and future of global health financing: a review of development assistance, government, out-of-pocket, and other private spending on health for 195 countries, 1995–2050. The Lancet. 2019 Jun 1; 393(10187):2233–60. 14. Data for Ecuador [Internet]. The World Bank IBRD-IDA Data. [cited 2020 Mar 5]. https://data. worldbank.org/?locations=EC-XT 15. National Institute of Statistics and Census. Encuesta Nacional de Empleo, Desempleo y Subempleo (ENEMDU)—December 2019 [Internet]. Quito: INEC; 2020. https://www.ecuadorencifras.gob.ec/ documentos/web-inec/POBREZA/2019/Diciembre-2019/201912_PobrezayDesigualdad.pdf 16. Morton RL, Schlackow I, Gray A, Emberson J, Herrington W, Staplin N, et al. Impact of CKD on House- hold Income. Kidney Int Rep. 2018 May 1; 3(3):610–8. https://doi.org/10.1016/j.ekir.2017.12.008 PMID: 29854968 17. Crews DC, Novick TK. Social Determinants of CKD Hotspots. Semin Nephrol. 2019 May; 39(3):256– 62. https://doi.org/10.1016/j.semnephrol.2019.02.003 PMID: 31054624 18. Cukor D, Halen NV, Kimmel PL. Chapter 27—Psychosocial Issues in Chronic Kidney Disease Patients. In: Kimmel PL, Rosenberg ME, editors. Chronic Renal Disease ( Second Edition) [Internet]. Academic Press; 2020. p. 413–23. http://www.sciencedirect.com/science/article/pii/B9780128158760000279 PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 19 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis 19. Perales-Montilla CM, Garcı´a-Leo´n A, Reyes-del Paso GA. Predictores psicosociales de la calidad de vida en pacientes con insuficiencia renal cro´ nica en tratamiento de hemodia´ lisis. Nefrol Madr. 2012; 32:622–30. 20. Nugent RA, Fathima SF, Feigl AB, Chyung D. The Burden of Chronic Kidney Disease on Developing Nations: A 21st Century Challenge in Global Health. Nephron Clin Pract. 2011; 118(3):c269–77. https:// doi.org/10.1159/000321382 PMID: 21212690 21. Ramirez-Rubio O, Brooks DR, Amador JJ, Kaufman JS, Weiner DE, Scammell MK. Chronic kidney dis- ease in Nicaragua: a qualitative analysis of semi-structured interviews with physicians and pharmacists. BMC Public Health. 2013 Apr 16; 13:350. https://doi.org/10.1186/1471-2458-13-350 PMID: 23590528 22. O’Hare AM, Szarka J, McFarland LV, Taylor JS, Sudore RL, Trivedi R, et al. Provider Perspectives on Advance Care Planning for Patients with Kidney Disease: Whose Job Is It Anyway? Clin J Am Soc Nephrol. 2016 May 6; 11(5):855. https://doi.org/10.2215/CJN.11351015 PMID: 27084877 23. Chen N, Hsu C-C, YAMAGATA K, Langham R. Challenging chronic kidney disease: Experience from chronic kidney disease prevention programs in Shanghai, Japan, Taiwan and Australia. Nephrology. 2010 Jun 1; 15(s2):31–6. 24. Pullar J, Allen L, Townsend N, Williams J, Foster C, Roberts N, et al. The impact of poverty reduction and development interventions on non-communicable diseases and their behavioural risk factors in low and lower-middle income countries: A systematic review. PLOS ONE. 2018; 13(2):e0193378. https:// doi.org/10.1371/journal.pone.0193378 PMID: 29474454 25. Philip P, Kannan S, Parambil N. Community-based interventions for health promotion and disease pre- vention in noncommunicable diseases: A narrative review. J Educ Health Promot. 2018;(7):141. PMID: 30596113 26. Jeet G, Thakur J, Prinja S, Singh M. Community health workers for non-communicable diseases pre- vention and control in developing countries: Evidence and implications. PLOS ONE. 2017; 12(7): e0180640. https://doi.org/10.1371/journal.pone.0180640 PMID: 28704405 27. OECD. Obesity Update 2017 [Internet]. Organisation for Economic Co-operation and Development; 2017. https://www.oecd.org/health/obesity-update.htm 28. Garcı´a-Trabanino R, Cerdas M, Madero M, Jakobsson K, Barnoya J, Crowe J, et al. Nefropatı´a mesoa- mericana: revisio´ n breve basada en el segundo taller del Consorcio para el estudio de la Epidemia de Nefropatı´a en Centroame´ rica y Me´ xico (CENCAM). Nefrol Latinoam. 2017 Jan 1; 14(1):39–45. 29. Gonza´ lez-Bedat M, Rosa-Diez G, Ferreiro-Fuentes A, Douthat W, Cueto-Manzano A, Ferna´ ndez-Cean J. Burden of disease: Closing the gaps in the burden of end-stage kidney disease in Latin America. Clin Nephrol. 2020; 93(1):55–9. https://doi.org/10.5414/CNP92S109 PMID: 31793873 30. El Comercio. Dialysis centers’ patients demand government pay debt [Pacientes de centros de dia´lisis reclaman el pago de deuda al Gobierno]. 2020 Nov 26; https://www.elcomercio.com/actualidad/ pacientes-centros-dialisis-pago-deuda.html 31. El Comercio. 19 hemodialysis clinics demand Ministry of Health pays outstanding debts of USD 10 mil- lion [19 clı´nicas de hemodia´ lisis reclaman al MSP el pago pendiente de USD 10 millones]. 2019 Nov 6; https://www.elcomercio.com/actualidad/clinicas-hemodialisis-msp-deuda-reclamo.html 32. Edicio´ n Me´dica. Dialysis centers self-declare in emergency due to Ministry of Health debts [Centros de dia´ lisis se declaran en emergencia por deudas del Ministerio de Salud]. 2020 Jan 13; https://www. edicionmedica.ec/secciones/empresas/centros-de-dialisis-se-declaran-en-emergencia-por-deudas- del-ministerio-de-salud–95172 33. Makridakis S, Hibon M. The M3-Competition: results, conclusions and implications. M3- Compet. 2000 Oct 1; 16(4):451–76. 34. R Core Team. R: A language and environment for statistical computing [Internet]. Vienna, Austria: R Foundation for Statistical Computing; 2020. https://www.R-project.org 35. RStudio Team. RStudio: Integrated Development for R [Internet]. Boston, MA: RStudio, PBC; 2020. http://www.rstudio.com 36. Hyndman R, Khandakar Y. Automatic time series forecasting: the forecast package for R. J Stat Softw. 2008; 26(3):1–22. 37. Wickham H. ggplot2: Elegant Graphics for Data Analysis. New York: Springer-Verlag; 2016. 38. Kassambara A. ggpubr: “ggplot2” Based Publication Ready Plots [Internet]. 2020. (R package). https:// CRAN.R-project.org/package=ggpubr 39. Wickham H, Miller E. haven: Import and Export “SPSS”, “Stata” and “SAS” Files. R package version 2.3.1. 2020; https://CRAN.R-project.org/package=haven 40. Becker R, Wilks A, Brownrigg R, Minka T, Deckmyn A. maps: Draw Geographical Maps. R package ver- sion 3.3.0. 2018; https://CRAN.R-project.org/package=maps PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 20 / 21 PLOS ONEChronic kidney disease in Ecuador: An epidemiological and health system analysis 41. Bivand R, Lewin-Koh N. maptools: Tools for Handling Spatial Objects. R package version 1.0–2. 2020; https://CRAN.R-project.org/package=maptools 42. Hijmans R. raster: Geographic Data Analysis and Modeling. R package version 3.4–10. 2021; https:// CRAN.R-project.org/package=raster 43. Neuwirth E. RColorBrewer: ColorBrewer Palettes. R package version 1.1–2. 2014; https://CRAN.R- project.org/package=RColorBrewer 44. Bivand R, Keitt T, Rowlingson B. rgdal: Bindings for the “Geospatial” Data Abstraction Library. R pack- age version 1.5–23. https://CRAN.R-project.org/package=rgdal 45. 46. INEC. Statistical Registry of Health Resources and Activities of Ecuador [Registro Estadı´stico de Recur- sos y Actividades de Salud de Ecuador] [Internet]. Quito: Instituto Ecuatoriano de Estadı´sticas y Cen- sos; 2008. https://www.ecuadorencifras.gob.ec/actividades-y-recursos-de-salud/ INEC. Statistical Registry of Health Resources and Activities of Ecuador [Registro Estadı´stico de Recur- sos y Actividades de Salud de Ecuador] [Internet]. Quito: Instituto Ecuatoriano de Estadı´sticas y Cen- sos; 2017. https://www.ecuadorencifras.gob.ec/actividades-y-recursos-de-salud/ 47. Ecuadorian Ministry of Finance. Accountability Report Year 2018 [Informe de Rendicio´ n de Cuentas Año 2018] [Internet]. 2018. https://www.finanzas.gob.ec/wp-content/uploads/downloads/2019/03/ Informe-de-Rdc-2018-final.pdf 48. Ecuadorian Institute of Donation and Transplant of Organs, Tissues and Cells. Executive Report. Accountability 2018 [Informe ejecutivo. Rendicio´ n de cuentas 2018] [Internet]. INDOT; 2019. http:// www.donaciontrasplante.gob.ec/indot/wp-content/uploads/2019/02/Informe_Rendicion_de_Cuentas_ 2018.pdf 49. INDOT. Informe estadı´stico te´cnico-Trasplantes de Enero a Diciembre 2019 [Internet]. Quito: Instituto Nacional de Donacio´n y Trasplante de O´ rganos, Tejidos y Ce´lulas; 2019. http://www.donaciontrasplante. gob.ec/indot/wp-content/uploads/2020/01/TRASPLANTES_DICIEMBRE_2019.pdf 50. Huertas J, Garrido D, Serpa F. Organ Transplantation in Ecuador. Transplantation. 2020; 104(3):445– 7. https://doi.org/10.1097/TP.0000000000002974 PMID: 32106198 PLOS ONE \| https://doi.org/10.1371/journal.pone.0265395 March 16, 2022 21 / 21 PLOS ONE
10.1371_journal.pone.0270562	RESEARCH ARTICLE ‘I don‘t need an eye check-up’. A qualitative study using a behavioural model to understand treatment-seeking behaviour of patients with sight threatening diabetic retinopathy (STDR) in India 1☯, Rani Mohanraj1☯, Rajiv Raman2☯, Geetha Kumar2‡, Sanjay Luvies3‡, Shuba KumarID Shivani Sunil Machhi4‡, Subhratanu Chakrabarty5‡, Janani Surya2‡, Radha Ramakrishnan6‡, Dolores Conroy6‡, Sobha Sivaprasad6☯ 1 Social Science Department, Samarth, Chennai, Tamil Nadu, India, 2 Department of Ophthalmology, Sankara Nethralaya, Chennai, Tamil Nadu, India, 3 Department of Ophthalmology, Giridhar Eye Institute, Cochin, Kerala, India, 4 Department of Ophthalmology, Aditya Jyot Foundation for Twinkling Little Eyes, Mumbai, Maharashtra, India, 5 Department of Ophthalmology, VMA Netra Niramay Niketan, Haldia, West Bengal, India, 6 Department of Ophthalmology-NIHR Biomedical Research Centre, Moorfields Eye Hospital NHS Foundation Trust, London, United Kingdom ☯ These authors contributed equally to this work. ‡ GK, SL, SSM, SC, JS, RR, and DC also contributed equally to this work. shubakumar@samarthngo.org Abstract Diabetic Retinopathy (DR) affects about 27% of patients with diabetes globally. According to the World Health Organization (WHO), DR is responsible for37 million cases of blindness worldwide. The SMART India study (October 2020-August 2021) documented the preva- lence of diabetes, and DR in people40 years and above across ten Indian states and one Union Territory by conducting community screening. About 90% of people with sight threat- ening diabetic retinopathy (STDR) were referred from this screening study to eye hospitals for management, but failed to attend. This qualitative study, a component of the SMART India study, explored perceptions of referred patients regarding their susceptibility to eye related problems in diabetes and the benefits/barriers to seeking care. Perceived barriers from the viewpoint of ophthalmologists were also explored. Guided by the Health Beliefs Model (HBM), 20 semi structured interviews were carried out with consenting patients diag- nosed with STDR. They included nine patients who had sought care recruited from eight eye hospitals across different states in India and eleven patients who did not seek care. Eleven ophthalmologists also participated. Four themes of analysis based on the HBM were, understanding of DR and its treatment, perceptions about susceptibility and severity, perceived barriers, perceived benefits and cues to action. Findings revealed poor under- standing of the effects of diabetes on the eye contributing to low risk perception. Prohibitive costs of treatment, difficulties in accessing care services and poor social support were major barriers to seeking care. Ophthalmologists acknowledged that the absence of symptoms and the slow progressive nature of the disease deluded patients into thinking that they were a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Kumar S, Mohanraj R, Raman R, Kumar G, Luvies S, Machhi SS, et al. (2023) ‘I don‘t need an eye check-up’. A qualitative study using a behavioural model to understand treatment- seeking behaviour of patients with sight threatening diabetic retinopathy (STDR) in India. PLoS ONE 18(6): e0270562. https://doi.org/ 10.1371/journal.pone.0270562 Editor: Kanchan Thapa, Province Health Service Directorate Surkeht, Technical Support Unit (TSU), NEPAL Received: June 11, 2022 Accepted: May 30, 2023 Published: June 15, 2023 Copyright: © 2023 Kumar et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Data will be made available on request. The request can be sent to Ms. KV. Sripriya, administrative officer, Samarth. She can be contacted at admin@samarthngo.org. Funding: The grant was received by Dr. Sobha Sivaprasad This work was supported by the Global Challenges Research Fund and UK Research and Innovation through the Medical Research Council PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 1 / 23 PLOS ONE(grant Number MR/P027881/1) This funding source had no role to play in the study design, data collection and analysis, decision to publish or preparation of the manuscript. Competing interests: The authors have declared that they have no competing interests. Treatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India fine. The study attests to the need for greater health literacy around diabetes, DR and STDR; for making treatment more affordable and accessible and for the development of effective patient education and communication strategies towards increasing compliance. Introduction Diabetic retinopathy (DR) is a common microvascular complication of diabetes mellitus (DM) that can result in irreversible visual impairment if the sight threatening complications are not identified early and treated [1]. Clinically, DR can be graded as non-sight-threatening diabetic retinopathy (NSTDR) and sight-threatening diabetic retinopathy (STDR) [2, 3]. As STDR can be asymptomatic in the initial stages and timely treatment can prevent visual loss, retinal screening for DR is recommended for all people with diabetes [4]. Once DR is identified through the process of screening, the person is referred to an ophthalmologist equipped to provide treatment, if required. These treatments include retinal laser, intra-vitreal injections of drugs and/or vitreo-retinal surgery depending on the complexity of the retinopathy. Globally, about 27% of individuals with diabetes have DR [5]. The prevalence of DR is esti- mated to be 31.6% in Africa [6]. According to the World Health Organization (WHO), DR is responsible for an estimated 37 million cases of blindness worldwide [7]. It is anticipated that the overall number of people with DR will rise due to the increasing prevalence of diabetes and the ageing and expanding global population. DR-related blindness is on the decline in high-income countries (USA,UK) as a result of improved therapies, concerted public health efforts and increased public awareness about DR screening [8–10]. On the other hand, low and middle income countries (India, China) bear the brunt of the diabetes epidemic and its complications including DR. This is partly attribut- able to inadequate forward planning and limited access to high-quality affordable eye care [11]. Owing to its potential to cause vision loss and a commensurate decline in quality of life, its early detection and effective management will be the key to improving overall health out- comes in people with diabetes [12]. DR in the Indian context Despite the high prevalence of DM in India [13, 14] there exists wide variations in awareness about the disease in the general population. A report [15] suggested that approximately 50% of participants were not even aware of diabetes. Awareness about DR and other ocular related problems elicited through community based research studies carried out across different regions in India also revealed wide variations in awareness of DR ranging from 16.1% to 71.3% [16–19]. Of significance is the fact that among those diagnosed with DM, awareness about DR ranged from 17.01% to 93.2% [20–22]. Ramasamy et al. reported that although study partici- pants were cognizant of DM, understanding of DR and its implications was poor [23]. The absence of symptoms; difficulties in doctor–patient interactions; burden of hospital visits and high costs were major deterrents to seeking care [24]. An earlier study of ours [25] that explored barriers to screening for DR, in addition to the above, also explored perceptions of health care providers wherein difficulties in communicating information about DR to less lit- erate patients, heavy work pressure and silent progression of the disease were reported as major barriers to care delivery. Another major challenge is the non-adherence to follow-up and treatment [26] and conse- quent adverse outcomes including irreversible visual loss [27]. Although this is partly explained by the lack of collaboration between screening and treatment services, there may be PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 2 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India other reasons that could contribute to their non-attendance. All these findings highlight the need to improve strategies to enhance compliance. The SMART India study The SMART India study (Multi centre statistical and economic modelling of risk based strati- fied and personalised screening of diabetes and its complication in India) was a cross-sectional screening study conducted between August 2018-June 2020 on people aged 40 years or above across 10 Indian states and 1 Union Territory (UT) [28]. Twenty tertiary level eye hospitals located in the above states and UT constituted the participating centres. The aims of the study were to assess the prevalence of diabetes and eye related complications, namely, DR and STDR towards developing risk models to identify and treat these conditions. Both urban and rural areas were included. The study found that out of the 42,416 people screened, a total of 7910 (18.8%)were identified to have diabetes. These included 5689 persons with known diabetes and 2221 with undiagnosed diabetes. The overall prevalence of DR was estimated at 12.5% and that of STDR was 4% [29]. People identified with DR in the study were referred to designated retinal departments in their respective sites where free treatment was offered. A total of 324 individuals were identi- fied to have STDR but only 10% (n = 32) attended treatment despite repeated reminders given about the possible threat of losing their vision. To better understand this phenomenon of poor care seeking we carried out a qualitative study to explore the perceptions of patients with STDR regarding their understanding of and susceptibility to eye related problems and the ben- efits/barriers to seeking care. From the health care providers (HCPs) we explored what they believed were barriers to care seeking among STDR patients and also sought their insights into what could contribute to improved care seeking in such patients. Conceptual framework In undertaking this qualitative study we used the Health Belief Model (HBM) which is a social psychological health behaviour change) model developed to explain and predict health-related behaviours, particularly with regard to the use of health services [30, 31]. Employing a theoreti- cal model such as the HBM we believed, was particularly well suited to our study because this model focuses on intra-personal factors, including risk-related beliefs which influence individ- uals’ health-related decision making [32]. Given that our study explored perceptions of disease risk, barriers to undergoing regular eye tests and receiving appropriate treatment among per- sons with STDR, the HBM provided us with the necessary structure and insights on strategies to improve decision making by families. Including the perceptions of HCPs contributed to a holistic understanding of the phenomenon of treatment seeking and served to triangulate our findings. The five components of the HBM model that predict an individual’s readiness to bring about behaviour change are described in Fig 1. Materials and methods This qualitative component of the SMART India study (grant number MR/P027881/1) carried out using semi structured interviews (SSIs) complied with the declaration of Helsinki. It was approved by the Institutional Ethics Committees of all the participating institutions namely, Sankara Nethralaya Vision Research Foundation Institutional Review Board, Chennai, Tamil Nadu, Giridhar Eye Institute Ethics Committee Cochin, Kerala, LV Prasad Eye Institute Ethics Committee, Hyderabad, Telengana, Aditya Jyot Eye Hospital Ethics Committee, Mumbai, Maharashtra, LV Prasad Eye Institute Ethics Committee, Bhubaneshwar, Odisha, Asram Netra Niramay Niketan Institutional Review Board Haldia, West Bengal, Shri Aurobindo PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 3 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India Fig 1. The Health Belief Model (HBM). https://doi.org/10.1371/journal.pone.0270562.g001 Medical Research Centre Institutional Review Board Raipur, Chhattisgarh and the Sri Sankar- adeva Nethralaya Institutional Ethics Committee, Guwahati, Assam. Study details were explained to participants in the local vernacular following which written informed consent was obtained. The study period was from October 2020-August 2021. Sites Of the 20 sites that had participated in the SMART India quantitative survey, we purposively selected 8 sites to provide some sense of geographical representation. The selected sites were: Chennai-state of Tamil Nadu, Cochin- state of Kerala and Hyderabad- state of Andhra Pra- desh as three cities from southern India. Mumbai- state of Maharashtra from the west, Bhuva- neshwar- state of Orissa and Haldia- state of West Bengal from the east, Guwahati- state of Assam from the north east and Raipur- state of Chhattisgarh from central India. Each site was led by a locally based eye hospital with a trained ophthalmologist who was responsible for the study at that site. Sampling Our participants sample included STDR patients who were referred to visit a health facility fol- lowing screening. Those who attended the eye clinics were categorised as ‘sought care’ while those who did not attend despite being advised to do so were categorised as ‘not sought care’. From the SMART India survey, the research team prepared a list of those who had sought care and those who had not. Each patient on the list was contacted to determine their willingness to participate. Given our constraints of time and money, the mandate by our ethics committees to provide a sample size for their approval, our relatively narrow objectives (namely risk per- ceptions and barriers to seeking care) and the fact that an a-priori theoretical model (HBM) PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 4 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India guided our study, we decided on a sample size of 32 patients who had not sought care (4 from each site) and 16 patients who had sought care (2 from each site). We were however, unable to achieve this number despite repeated calls and requests to the patients to participate in the study and were able to recruit only 11 patients from the not sought care group and nine patients from the sought care group. Inclusive of all 8 sites a total of 20 SSIs were carried out with persons with STDR. A combination of the COVID pandemic and the resultant lock- downs imposed by the central and state governments, poor phone and internet connectivity in some regions coupled with a reluctance to participate by some patients affected our ability to achieve the sample size that was originally planned. The HCPs were ophthalmologists who treated persons with DR and were working in the eye clinics. The original plan was to do 16 interviews- two HCPs per site- but only 11 HCPs participated in the SSIs. This was because five HCPs (two each from the participating centres in Assam and West Bengal and one from Chhattisgarh) had relocated and were therefore not available to participate in the interview. All interviews with the HCPs were done at the respec- tive facilities where they worked after obtaining written informed consent. Separate interview guides (S1 Text) were developed for each of the stakeholder groups. The core elements were common for both patients and HCPs (i.e perceptions on susceptibility, severity, barriers to and benefits of care seeking). However, from the HCPs we also examined the roles they played in encouraging patients to seek regular care and sought their opinions on possible strategies that would aid better compliance. Data collection The research teams who carried out the SSIs comprised optometrists who had a minimum of three years work experience and were employed as staff in the participating eye clinics in each of the study sites. In India, optometrists usually carry out all the initial assessments of the eye, interact closely with patients and serve as a link between the patient and the ophthalmologist. Prior to the recruitment of study participants an online training programme on qualitative research methods was conducted by SK and RM (both trained in qualitative research methods) over zoom sessions held for two hours every day over five days. The training sessions focused on orienting the field teams to the study methodology, the interview guides and to equipping them with the skills needed to carry out the SSIs. A combination of didactic lectures, demon- strations, mock interviews and hands on practical exercises were used towards enhancing the learning experiences of the research teams. Interviews with the patients were done face to face with those who were willing to come to the hospital. Others who were unwilling to come on account of the risks posed by the COVID -19 pandemic were provided the option of an online/phone interview. Interviews with the HCPs were carried out at their respective eye clinics. Data management and analysis All the audio recorded SSIs were transcribed verbatim and then translated into English (wher- ever necessary). The English transcripts were loaded into NVIVO. Data was analysed using a hybrid approach of qualitative thematic analysis [33] which incorporated i) a deductive a priori template of codes and themes derived from our research questions, the conceptual framework and the interview guides and ii)a data-driven inductive approach that was carried out follow- ing data collection. Given that the HBM constituted the basic theoretical framework that guided our study we put together a codebook (S2 Text) comprising categories/codes that reflected the HBM, namely, perceived susceptibility, perceived severity, perceived benefits, perceived barriers and cues to action. Following data collection, we used an inductive PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 5 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India approach in data coding and applied the principles of thematic analysis as described by Clark and Braun [34], which involved six stages. The first was the data familiarisation stage which required us to become familiar with the data through repeated readings of the interview tran- scripts. We also began to note down our initial observations during this stage. Next was the process of coding. Three transcripts were initially coded independently by two coders (SK and RM). After independently coding 3 transcripts, we expanded the existing code book and added codes and categories inductively derived from the interviews. Any differences in coding were discussed and resolved. The remaining transcripts were coded using this code book, new issues identified in these interviews relevant to our study were given a new code and inserted into the code book and were organized according to the HBM categories. In the third stage we examined our coded data in the context of the HBM, looking to see how they fitted and whether it addressed our research questions. In the fourth stage we began to formalise our themes of analysis and to reflect on whether these themes actually related well to our data, were convincing and credible. In the fifth stage we labelled and defined each theme, describing in detail what it signified and in the 6th stage of the analytic process we brought forth a coher- ent explanation of our study findings. Sifting through the data we then sorted and selected quotes and placed them under appropriate themes. Results Socio-demographic characteristics of participants Of the nine patients from the sought care group, three were men and six were women while the 11 participants from the not sought care group comprised two men and nine women. Those from the sought care group were a little older with a mean age of 59.8 years (Std ±9.19) while participants from the not sought care group had a mean age of 53 years (Std ±6.60). Five participants from the sought care group had completed five years of schooling (primary school), three were non-literate and one was a graduate. In the not sought care group, six par- ticipants had completed five years of schooling; three had completed nine years of schooling (secondary school) and two were non–literate. All participants were married; the women were all homemakers while the occupations of the men ranged from casual labourers, tailors to run- ning a petty business. The 11 HCPs included five men and six women, all were consultant oph- thalmologists. Seven of them were retinal specialists. Themes of analysis We mapped the care seeking journey of patients with STDR guided by the HBM and present four themes that we believe best explained our data i) Understanding of STDR and its treat- ment ii) Perceptions about susceptibility and severity (Perceived threat) iii) Perceived barriers iv) Perceived benefits and cues to action. i) Understanding of STDR and its treatment. Perceptions about susceptibility to a par- ticular disease/illness are considerably influenced by the patient’s understanding and aware- ness of the disease, its symptoms and whether they believe it could seriously impair their health. We explored understanding of STDR from the point of view of those patients who had sought care and those who had not and triangulated this with the opinions of the HCPs. Sought care group. Patients who had sought care indicated some awareness about STDR when they reported having been told that the disease could affect their vision perhaps even resulting in loss of sight. As described by this participant from Bhuvaneshwar, “When I was ill I have visited the government hospital, after check-up I came to know that I have diabetes. Doctor told me that it is not that only my health will be affected but my eyes PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 6 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India can also be affected. I have consulted an eye doctor and the doctor informed that in many cases one may even lose vision”, (Bhuvaneshwar-sought care-02, Female, 65 years). Information about STDR had been communicated to them in some cases, by the doctor who had been treating them for their diabetes and in other cases by the eye specialists to whom they had been referred. One patient described STDR as, “bleeding in the retina”, while another spoke of the risk of a person dying when she said, “When the sugar reaches into the brain then person may die or eye also may get affected. These are not new things, such general information must be known by all of us”, (Chennai-sought care-01, Female, 35 years). Thus, patients who had sought care professed to not having heard about DR in the begin- ning but became cognizant of it over the years when their doctors told them about it and advised them to visit an eye specialist. They described being told about the importance of con- trolling their blood sugars and realised that if this control was not adequate it could affect their eyes leading to possible loss of vision. One participant from Guwahati said, “The doctor informed me that it is the beginning of Diabetic Retinopathy. I did not know earlier but, after the doctor explained I know a little better now. I have been suffering from diabetes for the last 10 years but I never knew it can also affect the eyes. . . .The doctor also informed me that this is a sight threatening disease and can lead to complete blindness at any time, without my knowing. I am well aware now and I have also begun the treatment for it”, (Guwahati-sought care-01, Male, 47 years). They were also aware that treatment for STDR involved injections and/or laser treatment. Undergoing regular eye check-up, either once in 6 months or once every year, was therefore considered very important. The HCPs believed that STDR was a serious problem and it was critical that it be detected early so appropriate treatment could be started. Failing this, they felt sceptical of the patient recovering vision despite best efforts. Patients were believed to have poor understanding of the importance of consistent control of blood sugars as highlighted by this HCP from Chennai, who said, “. . .they tend to think that if one day they have good control [of food intake] then it should reflect on them the very next day or improve their vision. . ... So we have to explain to them and make them understand that diabetic retinopathy when it reaches the stage of severity we can probably only stabilize the disease, we cannot improve too much or reverse the complica- tions of the disease completely”, (Chennai-HCP-02, Male, 38 years). Doctors from Haldia and Raipur said that overall awareness about diabetes and its complications was very low among rural and tribal populations. The HCP from Raipur said, PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 7 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India “The problem is in the villages, patients of advanced diabetic retinopathy come from there because there is no one to investigate and due to lack of information about the disease, there is more problem”, (Raipur-HCP-01, Male, 40 years). Further, he added that in rural areas testing facilities for diabetes were not well developed and hence patients lost out on early detection. The doctor from Cochin said that sometimes diabetes was either not diagnosed or people had not started treatment for diabetes with oral hypoglycaemic agents. He also said that some individuals took alternative medicines for diabe- tes that did not help in glycaemic control. Such patients would invariably come at an advanced stage of the disease. Not sought care group. A mixed picture emerged with respect to those who had not sought care or who had not returned to the hospital for a follow up, with some being quite unaware and others being familiar with STDR. Those who had not heard of STDR or of its possible effects on the eyes had been seeking treatment for management of their diabetes but had never gone for an eye check-up. They also said that their health care providers had not advised them to do so. One participant even questioned the need to go for an eye check-up and said, “I don’t need to go for an eye check-up, I am able to see and do my daily routine and all is ok.. After diabetes only changes made are in my food, nothing else and I am seeing everything as usual. . . you said something about diabetic retinopathy... I don’t know about this and have never checked up for this. I have gone to the village health camp, I check up there and they didn’t say anything about this.. . . no one informed me for first time I am hearing from you”, (Bhuvaneshwar-not sought care 01-Female, 61 years). Some HCPs too, blamed the physician/diabetologist of these patients as they had failed to educate them adequately. As reported by one of them from Haldia, ““. . . ., most of them (patients)say that they were not aware that diabetes can cause vision loss, nor the treating physician has educated them about it and asked them to get their eyes checked. I think the treating physician should educate every patient about diabetic retinopa- thy and should ask them to get their eyesight checked with an Ophthalmologist”, (Haldia-HCP-01, Male, 37 years). Other HCPs said that patients, despite being advised by their doctors did not go for regular eye tests “. . .some patients despite telling them multiple times they don’t lend the ear. They feel that the eye treatment is different and the diabetic control is different. So they have to understand that the glycaemic control, the blood sugar control is very important along with the treatment for diabetic retinopathy. . . the awareness is lacking”, (Chennai-HCP- 01, Female, 38 years). All the HCPs felt that treating physicians played an important part in educating patients about DR and STDR. The group who were aware of STDR, yet had not sought care, appeared to be cognizant of the possible harms to their health in terms of vision loss brought on by their diabetic PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 8 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India condition. Their doctors had explained to them in fair detail about the importance of regular visits to the eye hospital and of keeping their blood sugars under control. However, as they were not experiencing any symptoms they did not believe anything was wrong, “I only have vision problem in eye since 6 to 7 years that too after I was diagnosed with diabe- tes. I am not taking any medicines for my eyes till now. . ... I do not see any changes. Every- thing seems normal to me. . .only thing is it looks dull and is not bright”, (Chennai- not sought care-01, Female, 56 years). Another participant despite repeated reminders did not perceive any urgency to seek care and said, “I was advised to do an eye check when I came to the hospital last year. . .but then corona came and I came to the village during this period. I used to get calls from the hospital asking me to come for a check-up. I told them that I will come when I come to Mumbai”, (Mumbai-not sought care-01, Male, 55 years) The HCPs agreed that the slow progression of STDR combined with the virtual absence of symptoms deluded patients into thinking that they were fine. Patients were unaware that dia- betes could in fact affect their vision. In this context, one HCP from Haldia declared that many patients would perhaps seek care only when symptoms become debilitating. “The main problem with diabetic retinopathy is that the patients’ eyesight is normal in the earliest stage which causes delay in seeking treatment. Only when they develop macular oedema or vitreous haemorrhage they seek treatment which by then has become sight threatening”, (Haldia-HCP-01,Male, 37 years). Endorsing this lack of understanding of the disease, another HCP said, “I think that it is really serious problem because most of our population don’t know about how seriously sight threatening diabetic retinopathy can be. . . they don’t understand that dia- betic retinopathy or diabetes itself is a very morbid disease. They feel that if they take medica- tion, it can get controlled and so there is nothing to be worried about”, (Mumbai-HCP-02, Female, 28 years). ii) Perceptions about susceptibility and severity (Perceived threat). Given the apparent lack of threat of a possible loss of vision consequent to a diagnosis of STDR- evident from the reports of many patients and validated by the HCPs- we explored threat perceptions as per- ceived by both the ‘sought care’ and ‘not sought care groups’. Important to note is the fact that both groups had their share of dissenting voices. Here too we have triangulated these percep- tions with the reports of the HCPs. Sought care group. Typically, participants who had sought care were perceptive to the threat of possible vision deterioration/loss consequent to the diagnosis of STDR and to that extent were sensitive to the need to be consistent in the care of their eyes, PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 9 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India “Yes I know about diabetic retinopathy, When I visited the government run PHC (Primary health Centre) the doctor explained to me about diabetic eye disease and informed that in many cases one may lose one’s vision. . .; I am taking medicines for it, doing exercise and regu- lar eye check-up because diabetes can affect my eyes, many people ignore them, we should check our eyes regularly and take care”, (Bhuvaneshwar-sought care-01, Female, 67 years). Such patients tended to take necessary precautions to safeguard their eyes. However, even among the sought care group there were a few who appeared unconvinced–the dissenting voice as it were—about the potential risk of vision loss, as this patient who said, “I don’t know if sugar patients (diabetes is colloquially referred to as ‘sugar disease’) might lose vision without any pain or symptoms. I am not having any pain. My distance vision only is bad, near vision is good. I do not know if sugar patients should visit an eye doctor once in a year. I can see well. I feel this is not going to affect my eyes because my blood sugar is normal”, (Hyderabad-sought care-02,Female, 74years). The absence of symptoms was thus an overarching feature that lulled patients into a false sense of security. The fact that patients did not perceive the seriousness of their illness in the absence of any symptoms was also highlighted by the HCPs. One of them from Mumbai said, “. . . for them (referring to patients) this is not important as it has not affected them yet. It is only when the disease starts to affect them that they become compliant or that they become responsive. . . . They don’t understand the seriousness of the situation,” (Mumbai-HCP-02, Female, 28years). She went on to say that patients may continue to ignore the retinopathy when it is in its early stages little realising the rapidity with which deterioration could set in. “even if they develop a mild form of diabetic retinopathy, it will progress rapidly because they don’t know that they need to get checked up. . . they don’t have the knowledge. So, I think the slum population, urban. . . you know below poverty line kind of population, uneducated peo- ple are the ones I feel who are most at risk”, (Mumbai-HCP-02, Female, 28 years). Not sought care group. As regards the ‘not sought care group’ while poor risk perceptions attributable to some extent to inadequate understanding of the disease had a bearing on their failure to seek care, there were a few who were different. They were different in that a sense of fear of the possibility of losing vision had begun to permeate their consciousness. One such participant voiced his apprehensions and said, “I’m worried whether my problem will be cured. Will I lose my eyesight as informed by your health worker? I never knew such an eye disease even existed. Having high sugar is such a problem. . .- Initially I was not serious at all, as I did not know much about it. But when I was informed by the health worker that it might lead to sudden complete blindness I got very PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 10 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India afraid. The scariest part is that such a serious issue was happening in my eye and I wasn’t even aware about it.—I did not know earlier that diabetes can lead to loss of eyesight”, (Guwahati-not sought care-01, Female, 40years). Further throwing light on patients’ care seeking behaviours, the HCPs said that patients tended to attribute the deterioration of their vision to cataract which was not considered a threatening condition and one which could be rectified with surgery. This contributed to reduced threat perceptions and to delays in seeking care. The situation was compounded by the fact that many patients either remained unaware about their diabetic status or else were not consistent in maintaining their blood sugars. Highlighting this, an HCP from Bhuvanesh- war said, “Longer the years with diabetes greater are the chances of developing DR. Poor glycemic con- trol, association of hypertension and dyslipidemia may raise the chances of vision loss. Renal disease, high body mass index, smoking, alcohol and poor health seeking behaviour may lead to early progression of the disease”, (Bhuvaneshwar-HCP-01, Male, 48 years). iii) Perceived barriers. We next explored specific barriers that impeded care seeking among patients with STDR. Our analysis revealed that similar barriers were reported by patients from both groups and therefore we present our findings under four sub-themes, i) Cost of care, access to care and financial constraints ii) Covid and the lockdown iii) Support iv) Gender related issues. The last sub-theme on gender related issues reflects the opinions of the HCPs alone. Cost of care, access to care and financial constraints. The predominant barrier reported by patients across both groups was the high cost of treatment for STDR. Patients advised monthly intra-vitreal injections found it difficult to bear the high cost of these injections (there are many types of such injections ranging in cost from around Rs.10,000 to Rs. 100,000)which severely compromised their ability to seek care. Furthermore, the progress of treatment was slow and required repeated visits which not only added to their costs but was deeply frustrat- ing. As reported by this patient, “Economic reason is the main reason for not coming early. . .. As the treatment of my disease is very costly and takes time it is unaffordable by people like us. . .I am in a lot of tension as my right eye has a lot of problems and I need urgent treatment. Eleven months ago the doctor told me to come. . . it is already late and now I am almost going blind”, (Haldia -not sought care-02, Male, 51 years). For patients from rural and remote areas, accessing care facilities posed a major challenge as these were usually located in cities and towns, far away from their homes, “I am worried that I won’t be able to do the treatment because this is the only eye hospital in the district and my place is around hundred kilometres from here. I have also heard that the injections that I need is also very costly and I might not able to afford the treatment at all”, (Haldia-not sought care-03, Female, 49 years). PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 11 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India Difficulty in taking time off from work was another barrier. A visit to the hospital meant a whole day off from work which poor patients could ill afford, “Yes, actually I won’t get time. I work in a bungalow [does household work], so doing work there itself occupies my time. I was told to spend a day to consult in an eye hospital, but I didn’t get time, I left just like that”, (Chennai-not sought care-02, Female, 56 years). The HCPs admitted that the management of the disease was expensive, complex and required regular monitoring. These were major deterrents to seeking care for patients. Those without the benefit of any health insurance cover, mainly the daily wage earners were particu- larly hard hit. Neither could they afford the treatment costs, nor could they afford to take a day off from work to visit the hospital as this would mean loss of wages. Such patients they said tended to drop out early. For other patients the fact that they needed to continuously monitor their health, deal with other diabetic related complications coupled with the realization that this would be a life-long feature was very exhausting. As one doctor from Mumbai explained, “Patients sometimes do lose faith. . . they tend to drop out either because they are financially unable to afford it or because they think it is not being cured or they think it should happen faster. Sometimes the other systemic diseases are so severe that they are not able to come for treatment of their eyes”, (Mumbai-HCP-01,Female, 34 years). Covid and the lockdown. The lockdown imposed due to the COVID-19 pandemic was another reason cited by a few patients that came in the way of their care and treatment. One patient from Mumbai who did not seek care said that her family members forbade her to seek treatment for her eyes because of the lockdown and the added fear of her contracting the COVID-19 infection. A patient from Haldia regretted having delayed seeking care and with the lockdown in force, feared the effect this additional delay would have on her eye sight. She said, “I couldn’t come due to lockdown and monetary problems. I wish I had listened to my doctor and not waited for so long to get my eyes checked. Maybe then my problem would not have been so bad”, (Haldia-not sought care-03, Female, 49 years). A patient from Cochin describing the difficulties she faced in accessing the eye hospital said, “I am living in a rural area the mode of transportation is by boat and bus. Due to the pan- demic the bus has stopped the service and the boats are limited which makes it really difficult for us to travel to the eye hospital, especially because of its timing”, (Cochin-not sought care-01, Female, 62 years). Patients living in rural areas reported that the government usually organized free eye camps periodically which gave them an opportunity to get their eyes tested. The camps did not hap- pen due to the lockdown which was a major blow to the village folk. PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 12 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India All the HCPs were unanimous in their opinion that the Covid pandemic and the resultant lockdowns were a major setback to patients as many were unable to obtain the care they needed. Support. The need for assistance, be it from a family member or a friend to accompany them to the hospital for a consultation or a follow–up visit was expressed by several patients. The problem was exacerbated for those who had to travel long distances to access care. Cou- pled with their impaired vision, negotiating the challenges of boarding buses and trains on their own, severely affected their ability and desire to seek care, as reflected in this quote from a participant from Cochin, “Actually I am having blurred vision since 1 year and I consulted in a hospital at Ernakulum and they told me to consult at a higher center. I have no one to accompany me and also con- sidering my age I was unable to go”, (Cochin-not sought care-01, Female,61 years). The HCPs reiterated the need for good family support. They stated that a large number of such patients were older in age and some also suffered from low vision. One HCP from Cochin added that diabetes and STDR required lifelong care for the patient. This necessitated that patients have the support of their families without which their care and recovery would be compromised, “. . .one important barrier is it requires life-long care which is not easy because patient will be having multiple diseases they have to go to the diabetologist, and if they have nephropathy, they have to go to nephrologist and the ophthalmologist is one among them. . . it is not that they come once, they get treatment and it is over, they have to keep coming again andagain.. they need family support. Most of the patients are generally middle aged or elderly and with poor vision they won’t be able to come alone”, (Cochin-HCP-02, Female, 51 years). Gender related issues. With respect to gender, a few HCPs were of the opinion that men sought care early when compared to women. Explaining this, the HCP from Chennai said that often women tended to come when symptoms became very debilitating, an indication that the disease had progressed to an advanced stage, “The females are probably more delayed because even after a significant drop in vision, they do not reach because they are mainly at the household, so they don’t reach us. Males, reach a little early because they are working population and it affects their work”, (Chennai-HCP-01, Female, 38 years). Others spoke of these differences in care seeking as more evident in rural areas and said, “The patriarchal nature of society here appears to be a barrier to the females. Also most females do not want to be a burden on the family. So somewhere health becomes less of a pri- ority for the females”, (Haldia-HCP- 02, Female, 34 years). Another doctor, also from Haldia believed that there were many cases of undiagnosed dia- betes among women, primarily because these women were uneducated, totally dependent on PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 13 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India their husbands/families and did not feel comfortable speaking of their diabetes or eye prob- lems to their in-laws. Care seeking among these women was therefore poor. “Females are mostly dependent on the caregivers, that is the husband and offspring with regards to their health check-up. . .many of them are housewives and not very educated. Even younger females with DR present late and mostly the reason I see is they are not very comfort- able speaking of their poor vision and about their diabetes to their family members/in laws. There are lot of undiagnosed diabetes in females with DR”, (Haldia-HCP-02, Female, 34 years). iv) Benefits and cues to action. Our final theme sought to understand from patients whether they perceived any benefits in seeking care and in bringing about lifestyle changes and what cues would help sustain these behaviours. We also explored the concept of “cues to action” from the perspectives of the HCPs. We believed that by virtue of their years of experi- ence treating such patients they would be in a position to provide insights into what could help improve care seeking in these patients. Benefits. Fear of poor health because of diabetes and of the potential of losing vision, was a strong motivating factor that pushed patients into seeking care and to following the doctor’s recommendations. If this meant going for regular eye check-up and keeping blood sugars under control through proper medication and lifestyle changes, then these patients felt this was worth it. As stated by a patient, “As much as possible I take medicine prescribed by doctor. I feel better after medication and also my sugar level is in control.. I have diabetes, regular eye check-up is necessary for me so I can protect my eyes”, (Bhuvaneshwar-sought care-02, Female, 65 years). Another patient feared becoming helpless and dependant on others and said, “I will not be able to take a glass of water. . . . our children do not stay with us.. No one stays nearby.. We, husband and wife are alone.. my husband will go out for work and I will be alone in the house.. If I will not be able to see properly then who will look after me.. How will I manage without my eyes.. So we came to the hospital for treatment”, (Mumbai- sought care-01, Female, 58 years). Even those who had not sought care, in hind sight, regretted their actions and understood the damage this had done to them, “I fear that I might lose my sight completely, I cannot cook food for myself. I have to depend on my family members a lot. I used to do all the work myself before, now it is not possible. I wish I had come earlier to the hospital”, (Haldia-not sought care-01, Female, 59 years). Thus, many patients appreciated the benefits of regular eye check-up and care of the eyes but had been hindered from accessing care on account of the barriers described earlier. There- fore, the cues to action recommended by both patients and HCPs are primarily aimed at addressing these barriers. PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 14 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India Cues. Several cues to action were reported by both patients and HCPs which we categorized as improving awareness, improving access to and reducing costs of care and need for support. Improving awareness. The need for making the public and in particular, persons with diabe- tes more aware about the disease was considered important to promoting better care seeking by many patients, “I think it is extremely important to create awareness among diabetic people regarding DR. All diabetes treatment centres should hold awareness camps on DR from time to time. In your hospital you can display videos on DR informing diabetic patients to do a proper retina check- up at least once a year. You can also appoint health workers to increase awareness in the communities”, (Guwahati-sought care-01,Male, 47 years). Those who had not sought care highlighted that it would be helpful if the treating physi- cians advised and provided proper guidance, “my doctor said to me “your eyes are normal only”. He said there is no problem in my eyes due to diabetes. . .If there is problem in the eye then he would have told me right? He said it’s normal so I left it like that”, (Chennai-not sought care-01, Female, 56 yrs). Sending regular reminders to patients to come for an eye check-up and warning them of possible repercussions of not seeking treatment for STDR was also seen as a way to both edu- cate and get patients seek care. The HCPs believed that poor awareness about STDR constituted a major challenge to care seeking among patients and saw this as a critical first step to encouraging patients to seek proper and timely treatment for STDR. They believed that the poor and less literate patients were particularly hard hit and often failed to appreciate what they were up against, “Poor patients and those with low literacy may fail to understand the importance of regular follow up. Well informed patients however participate in the treatment better. . .”, (Bhuvaneshwar-HCP-01, Male, 48 years). According to them this was where the treating physicians—generally the first point of con- tact for most patients- could play a significant role. In this context one HCP said, “The physician should tell the patient that because you are diabetic you need an eye check-up, you have to go to ophthalmologist, you should take treatment on time. . . that guidance if it comes from their family physician that will have a good influence on patients and they will tend to obey. . .I think these physicians should play a more important role”, (Cochin-HCP-02, Female, 51 years). Another HCP spoke about the value of involving the community health workers in spread- ing awareness about DR and STDR to people in the communities in which they lived. Display- ing messages on the print and social media as a way of publicly disseminating information on STDR was also reported. In addition to educating patients on STDR, HCPs also stressed the need for counselling and motivating patients to cope positively and more effectively with the PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 15 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India disease. An HCP from Chennai described using words of encouragement and appreciation with his patients and said, “So every time I see patients, I tell them, “you are doing good and your blood sugar is under control. I have to keep seeing you regularly then only you will be maintaining the same good vision. So, positive reinforcement does work”, (Chennai-HCP- 02,Male,38 years). Improving access to care and reducing costs. For patients from rural and other remote areas having hospitals located closer to their villages and towns was believed to both motivate them to seek care and enable easy access. In this context one woman participant from Haldia suggested, “My home is very far away. It takes 3–4 hours to reach here and also costs a lot of money to come here. So it would not be possible to come here for check-up regularly. It will be helpful if we could go for follow-up to any nearby eye hospital as it is not possible to come here all the way for regular follow-up”, (Haldia-not sought care-03, Female, 49 years). Conducting free eye camps in their villages and in nearby towns were also suggested as helpful strategies that would make care accessible and affordable The HCPs admitted that treatment for STDR was expensive and involved injections or laser or surgery or a combination of them. Describing this, an HCP said, “Treatment mainly depends on stage of disease in which the patient presents to you. Early stages of DR can be managed with good glycaemic control and control of associated risk fac- tors. But in later stages of disease progression intra-vitreal injections and lasers are the main- stay of treatment”, (Bhuvaneshwar-HCP-02, Male, 33years). Therefore, seeking early and regular care was strongly suggested. Another suggestion to improve both access and reduce cost for patients was for the government to step in and equip government hospitals (where care is provided free) with the capability to deliver such care, “The government should take active role in training all their primary healthcare providers on the importance of eye check-up in treatment of diabetes”, (Haldia-HCP-02, Female, 34 years). Reducing the cost of the intra-vitreal injections and making them available in the govern- ment hospitals were recommended. The HCPs also highlighted the existence of various gov- ernment health insurance schemes which patients need to be made aware of and which could help to subsidise the cost of care. Support. Support in terms of having someone to accompany them to the hospital was seen as an important factor that would aid better compliance with hospital visits. In many cases both patients and their primary caregivers, usually the spouse were elderly, making travel to the hospital a challenge for both. Younger family members like their children or other relatives who were called upon to take the patient to the hospital were often constrained by their own PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 16 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India work and home responsibilities. Therefore, one suggestion was to have the hospital organize a vehicle to pick up patients from their homes for a check up on specified days. As reported by this patient, “The hospital can arrange van for patients to travel. Prior to travel they can send a reminder and inform patients that we have arranged van for travelling. . .So trip can be organized to pick up and drop the patients. Like these if you inform, people will show interest to come, we poor people will expect these things only. . .so if hospital provide such services it will be helpful to us”, (Chennai-sought care-01, Female, 35 years) The HCPs highlighted that STDR was a slow progressive condition, and the treatment too was long drawn and expensive. In view of this, they felt that the support of family members to motivate and encourage patients to maintain a healthy lifestyle, assist with financial costs and accompany patients to the hospital whenever required were critical cues to the patient’s well- being and to improved care seeking as well. In this context an HCP said, “It is the family who keeps on motivating when the patient loses hope as treatment for DR is a prolonged one. Just treatment of the eye alone is not enough. It’s the holistic improvement in the patient’s blood sugar control other systemic factors and most important the stress levels. This is where family support plays an important role. Most diabetic patients are aged and need the help of their family. . .”, (Haldia-HCP-02, Female, 34 years). Discussion The findings from our study revealed poor appreciation of the need for early and sustained care among most patients with STDR. Added to this, the absence of symptoms and the slow progression of the disease contributed to low risk perceptions which further compromised seeking early care. Barriers in terms of prohibitive costs of treatment, difficulties in accessing care services and inadequate social support complicated the scenario by increasing health disparities. Low awareness contributed considerably to poor care seeking and perceptions of risk, a fact that is well acknowledged in low and middle income countries such as India [35–40]. There are studies showing that those with less education who usually belong to the poorer sections of society tend to have poor knowledge about DR and its complications [41]. By virtue of this fact they tend not to perceive the seriousness of the situation [25] and therefore avoid or delay care seeking. In our study, although most patients who had sought care were aware that diabetes could affect their vision there were also those who did not appreciate this fact. They nurtured the belief that as they were asymptomatic and had no vision related difficulties, they were fine. To them having diabetes essentially meant controlling diet, exercising and taking the required medication. An earlier study carried out in rural Tamil Nadu [38] found that while 90% of the study participants were aware of the importance of regular eye examinations, around one- third believed that if blood sugars were controlled there was no requirement to see an ophthal- mologist. The study further highlighted that despite having knowledge about DR, only half the study participants, knew about the effects of good control of diabetes on DR. The HCPs in our study also emphasized that significant numbers of patients did not recognize the importance PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 17 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India of consistent control of blood sugars and added that in many cases the treating physicians of these patients did not advise them to go for an eye test. Previous studies have shown that dura- tion of diabetes, poor glycaemic and blood pressure control and dyslipidemia [42, 43] acceler- ate the development and progression of DR suggesting the need for more intensive management of the condition which can be overwhelming to patients. Given this, enhancing understanding and awareness about STDR would perhaps be a key priority to address to encourage better compliance and help prevent vision loss. What was particularly alarming in our study was that despite all reminders to attend the eye hospital and the offer of free screening, patients did not seek care. In an earlier study [25], we had recommended giving due credence to a patients’ understanding capacity when providing health care awareness as a means to enhance better learning and thereby to improving treat- ment compliance. As stated by Piyasena et al. [40]“an individual’s better understanding of their susceptibility to vision loss may increase motivation to attend a screening examination”. Perhaps, involving their treating physicians/diabetologists- with whom patients share a relationship of trust—in this effort could be a step in this direction. In this context, Sanghamitra et al. [44] have spoken about the need to strengthen the capacity of physicians in the private sector to counsel patients about the importance of seeking appropriate care, thereby enabling patient recovery. Equally important will be the need to create more patient friendly environments making the entire testing and treatment process less frightening and intimidating to patients. Further, the insidious progression of STDR contributes to a false sense of security increas- ing the perceptions that regular eye tests are unnecessary [45, 46]. In this context, Xiong et al. [36] reported that presence of symptoms can serve as triggers which could alert patients to the need to seek care. Understandably, this does not apply to DR as visual impairment occurs late in the disease and is usually associated with sight threatening complications such as diabetic macular oedema and complications of proliferative diabetic retinopathy. Devenney and O Neill [47] described the sense of loss of independence and a variety of social losses faced by DR patients with fading vision which could serve as possible cues to action. However, by then the disease would have progressed significantly with potential for irreversible visual loss. There- fore, educating patients with diabetes about DR would be most essential for informed decision making and appropriate care seeking. In our study we found that where women were concerned, not only was awareness poor but women refrained from reporting their problems to their family members as they felt awk- ward or else did not want to pose a burden. Studies have shown [40] that women-many of whom are dependants and have no income of their own -are reluctant to seek eye care and unwilling to draw on the limited family resources for their treatment. Das et al. [48] described that women participants in their study felt that men tended not to consider women’s health problems as serious and often believed that they were faking it. Greenwood et al. [49] explored similarities and differences in values and attitudes towards women in the work place and at home across several countries in Asia. They reported that, “culturally ingrained gender role expectations keep women from full equality at work and at home”. Given this climate, women may lack both the voice and the economic capability to seek care for themselves. The need to sensitise families and men in particular to be attentive to and supportive of the health needs of women members in their family would be extremely important. The prohibitive cost of treatment was another impediment to seeking/continuing treat- ment. Financial barriers to DR treatment is a universal problem in both developed and devel- oping countries [39, 50, 51] attesting to the urgent need to address this issue. Our study reported that patients who were poor or who lived in rural/remote areas were particularly affected. The HCPs confirmed that it was challenging for patients on monthly intra-vitreal injections to bear the high cost of these injections in the absence of any insurance cover, often PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 18 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India leading to non-adherence to treatment regimens. Suggestions towards making these injections available in government facilities at reduced cost and informing patients about the existence of various government health insurance schemes were made by both patients and HCPs in our study. In this context the government of India under the NPCDCS (National Programme for Prevention and Control of Cancer, Diabetes, Cardiovascular disease and Stroke-2013-17) [52] has focused on strengthening infrastructure, human resources, early diagnosis, management and referral for non–communicable diseases such as diabetes. Provision has also been made for free diagnostic facilities and drugs for patients attending these clinics. Incorporating screening and treatment for DR and STDR under this programme will facilitate a larger pro- portion of patients to avail and access these services at more reasonable costs. Poor support in terms of not having any one to accompany them to the hospital was another major challenge reported by participants in our study. Sanghamitra et al. [44] under- scored the importance of family and friends who were seen as enablers in terms of providing encouragement and motivation to patients to manage their diabetes. Studies have shown that presence of such support improves diabetes control, knowledge, and psychosocial functioning [53]. We could perhaps also draw upon the strengths of the community health workers usually attached to the government run primary health centres (PHCs) located in villages and urban communities who by virtue of the outreach work they do, are in close contact with the com- munities they serve. They could be trained to both educate and support men and women to more easily access eye care services. Strengths and limitations This study used the HBM with its focus on intra-personal factors, including risk-related beliefs which influence decision making as a means to understand the phenomena of care seeking for DR. Using a hybrid approach of qualitative thematic analysis involving a combination of deductive and inductive strategies is the strength of this study. At the outset, the use of the HBM provided logic and structure to the study and also allowed for a data-driven inductive approach that was carried out following data collection. We were however, unable to achieve the desired sample size which could be seen as a limita- tion. We compensated for this lack by taking care to explore each domain of the HBM as thor- oughly as possible with each respondent during the interviews. At the stage of coding we kept our minds open to the generation of new codes and categories beyond the ones developed a priori thereby helping to further saturate each domain of the HBM. It is important to highlight here that our patients, albeit from different cities across India, were all persons with diabetes, aged between 40 to 60 years and predominantly belonging to low and middle class socioeco- nomic backgrounds. Further, across different states in India, issues concerning awareness of, access to and compliance with care for DR were found to be similar in this study. While we did not undertake any comparison of data between states on account of the very small sample sizes in each state, the fact remains that patients and health care providers across all the states raised similar issues thereby enhancing the transferability of these findings to poor and socioeconom- ically deprived populations in our country. Conclusion The findings from this study add to the growing body of literature from different countries attesting to the urgent need for greater health literacy around diabetes, DR and STDR. Undoubtedly, a real understanding of the disease was lacking among patients in our study evi- denced by the fact that many did not appreciate the effects of good control of diabetes on DR. While recommendations to enhance health literacy around DR have been repeatedly made PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 19 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India [38, 54], it is unfortunate that this continues to be a major issue severely compromising care. Given its irreversible nature eventually leading to vision loss if left untreated, a redoubling of efforts towards improving patient awareness and compliance will be vital to ensuring a better quality of life for DR patients. Supporting information S1 Text. Study guides. (DOCX) S2 Text. Study code book. (DOCX) Author Contributions Conceptualization: Shuba Kumar, Rani Mohanraj, Rajiv Raman, Sobha Sivaprasad. Data curation: Rani Mohanraj, Geetha Kumar, Sanjay Luvies, Shivani Sunil Machhi, Subhra- tanu Chakrabarty, Janani Surya, Radha Ramakrishnan. Formal analysis: Shuba Kumar, Rani Mohanraj. Funding acquisition: Rajiv Raman, Sobha Sivaprasad. Methodology: Shuba Kumar, Rani Mohanraj, Dolores Conroy, Sobha Sivaprasad. Project administration: Rajiv Raman, Geetha Kumar, Sanjay Luvies, Shivani Sunil Machhi, Subhratanu Chakrabarty, Janani Surya, Radha Ramakrishnan, Dolores Conroy. Resources: Rajiv Raman, Radha Ramakrishnan, Dolores Conroy. Supervision: Shuba Kumar, Geetha Kumar, Sanjay Luvies, Shivani Sunil Machhi, Subhratanu Chakrabarty, Janani Surya, Radha Ramakrishnan, Dolores Conroy. Validation: Shuba Kumar, Rajiv Raman, Sobha Sivaprasad. Writing – original draft: Shuba Kumar. Writing – review & editing: Shuba Kumar, Rani Mohanraj, Rajiv Raman, Geetha Kumar, San- jay Luvies, Shivani Sunil Machhi, Subhratanu Chakrabarty, Janani Surya, Radha Rama- krishnan, Dolores Conroy, Sobha Sivaprasad. References 1. Wild S, Roglic G, Green A, Sicree R, King H. Global prevalence of diabetes: estimates for the year 2000 and projections for 2030. Diabetes Care. 2004; 27: 1047–1053. https://doi.org/10.2337/diacare.27.5. 1047 PMID: 15111519 2. Alemu Mersha G, Tsegaw Woredekal A, Tilahun Tesfaw M. Sight-threatening diabetic retinopathy and associated risk factors among adult diabetes patients at Debre Tabor General Hospital, Northwest Ethi- opia. Clin Ophthalmol. 2020; 14: 4561–4569. https://doi.org/10.2147/OPTH.S285606 PMID: 33408458 3. Forbes JM, Cooper ME. Mechanisms of diabetic complications. Physiol Rev. 2013; 93: 137–188. https://doi.org/10.1152/physrev.00045.2011 PMID: 23303908 4. Scanlon PH. Update on screening for sight-threatening diabetic retinopathy. Ophthalmic Res. 2019; 62: 218–224. https://doi.org/10.1159/000499539 PMID: 31132764 5. Thomas RL, Halim S, Gurudas S, Sivaprasad S, Owens DR. IDF Diabetes Atlas: A review of studies uti- lising retinal photography on the global prevalence of diabetes related retinopathy between 2015 and 2018. Diabetes Res Clin Pract. 2019; 157:107840. https://doi.org/10.1016/j.diabres.2019.107840 PMID: 31733978 PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 20 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India 6. Burgess PI, MacCormick IJ, Harding SP, Bastawrous A, Beare NA, Garner P. Epidemiology of diabetic retinopathy and maculopathy in Africa: a systematic review. Diabet Med; 30(4):399–412. https://doi.org/ 10.1111/j.1464-5491.2012.03756.x PMID: 22817387 7. Ting DS, Cheung GC, Wong TY. Diabetic retinopathy: global prevalence, major risk factors, screening practices and public health challenges: a review. Clin Exp Ophthalmol. 2016; 44(4):260–77. https://doi. org/10.1111/ceo.12696 PMID: 26716602 8. Wong TY, Mwamburi M, Klein R, Larsen M, Flynn H, Hernandez-Medina M, et al. Rates of progression in diabetic retinopathy during different time periods: a systematic review and meta-analysis. Diabetes care. 2009; 32(12):2307–13. https://doi.org/10.2337/dc09-0615 PMID: 19940227 9. Liew G, Michaelides M, Bunce C. A comparison of the causes of blindness certifications in England and Wales in working age adults (16–64 years), 1999–2000 with 2009–2010. BMJ open. 2014; 4(2): e004015. https://doi.org/10.1136/bmjopen-2013-004015 PMID: 24525390 10. Claessen H, Kvitkina T, Narres M, Trautner C, Zo¨llner I, Bertram B, et al. Markedly decreasing incidence of blindness in people with and without diabetes in southern Germany. Diabetes care. 2018; 41(3):478– 84. https://doi.org/10.2337/dc17-2031 PMID: 29317450 11. Wong TY, Sabanayagam C. Strategies to tackle the global burden of diabetic retinopathy: from epidemi- ology to artificial intelligence. Ophthalmologica. 2020; 243(1):9–20. https://doi.org/10.1159/000502387 PMID: 31408872 12. Unwin N. Definition and diagnosis of diabetes mellitus and intermediate hyperglycemia. Report of a WHO/IDF Consultation. 13. Tandon N, Anjana RM, Mohan V, Kaur T, Afshin A, Ong K, et al. The increasing burden of diabetes and variations among the states of India: the Global Burden of Disease Study 1990–2016. The Lancet Glob Health. 2018; 6: e1352–1362. https://doi.org/10.1016/S2214-109X(18)30387-5 PMID: 30219315 14. Federation ID. IDF Diabetes Atlas, tenth. International Diabetes. 2021. 15. Deepa M, Bhansali A, Anjana RM, Pradeepa R, Joshi SR, Joshi PP, et al. Knowledge and awareness of diabetes in urban and rural India: the Indian Council of Medical Research India diabetes study (phase I): Indian Council of Medical Research India diabetes 4. Indian J Endocrinol Metab. 2014; 18: 379–385. https://doi.org/10.4103/2230-8210.131191 PMID: 24944935 16. Deepa M, Deepa R, Shanthirani CS, Manjula D, Unwin NC, Kapur A, et al. Awareness and knowledge of diabetes in Chennai—the Chennai urban rural epidemiology study [CURES-9]. J Assoc Physicians India. 2005; 53: 283–287. PMID: 15987011. 17. Dandona R, Dandona L, John RK, McCarty CA, Rao GN. Awareness of eye diseases in an urban popu- lation in southern India. Bull World Health Organ. 2001; 79: 96–102. PMID: 11242828 18. Namperumalsamy P, Kim R, Kaliaperumal K, Sekar A, Karthika A, Nirmalan PK. A pilot study on aware- ness of diabetic retinopathy among non-medical persons in South India. The challenge for eye care pro- grammes in the region. Indian J Ophthalmol. 2004; 52: 247–251. PMID: 15510470 19. Hussain R, Rajesh B, Giridhar A, Gopalakrishnan M, Sadasivan S, James J, et al. Knowledge and awareness about diabetes mellitus and diabetic retinopathy in suburban population of a South Indian state and its practice among the patients with diabetes mellitus: A population-based study. Indian J Ophthalmol. 2016; 64: 272–276. https://doi.org/10.4103/0301-4738.182937 PMID: 27221678 20. Saikumar SJ, Giridhar A, Mahesh G, Elias A, Bhat S. Awareness about eye diseases among diabetic patients: a survey in South India. Community Eye Health. 2007; 20: 16–17. PMID: 17515978 21. Koshy J, Varghese DL, Mathew T, Kaur G, Thomas S, Bhatti SM. Study on KAP of ocular complications due to diabetes among type II diabetics visiting a tertiary teaching hospital. Indian J Community Health. 2012; 24: 27–31. 22. Srinivasan NK, John D, Rebekah G, Kujur ES, Paul P, John SS. Diabetes and Diabetic Retinopathy: Knowledge, Attitude, Practice (KAP) among Diabetic Patients in A Tertiary Eye Care Centre. J Clin Diagn Res. 2017; 11: NC01–NC07. https://doi.org/10.7860/JCDR/2017/27027.10174 PMID: 28892947 23. Ramasamy K, Raman R, Tandon M. Current state of care for diabetic retinopathy in India. Curr Diab Rep. 2013; 13: 460–468. https://doi.org/10.1007/s11892-013-0388-6 PMID: 23657764 24. Shukla R, Gudlavalleti MV, Bandyopadhyay S, Anchala R, Gudlavalleti AS, Jotheeswaran AT, et al. Perception of care and barriers to treatment in individuals with diabetic retinopathy in India: 11-city 9- state study. Indian J Endocrinol Metab. 2016; 20: S33–41. https://doi.org/10.4103/2230-8210.179772 PMID: 27144135 25. Kumar S, Kumar G, Velu S, Pardhan S, Sivaprasad S, Ruamviboonsuk P, et al. Patient and provider perspectives on barriers to screening for diabetic retinopathy: an exploratory study from southern India. BMJ Open. 2020; 10: e037277. https://doi.org/10.1136/bmjopen-2020-037277 PMID: 33303431 PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 21 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India 26. Kumar G, Velu S, Prakash VJ, Kumar S, Sivaprasad S, Sharma A, et al. Attendance Rate in Patients with Diabetic Macular Edema Receiving Short Messages. Ophthalmol Retina. 2021; 5: 1054–1056. https://doi.org/10.1016/j.oret.2021.03.005 PMID: 33691182 27. Kashim RM, Newton P, Ojo O. Diabetic retinopathy screening: A systematic review on patients’ non- attendance. Int J Environ Res Public Health. 2018; 15: 157. https://doi.org/10.3390/ijerph15010157 PMID: 29351207 28. Sivaprasad S, Raman R, Rajalakshmi R, Mohan V, Deepa M, Das T, et al. Protocol on a multi centre statistical and economic modelling study of risk-based stratified and personalised screening for diabetes and its complications in India (SMART India). BMJ Open. 2020; 10: e039657. https://doi.org/10.1136/ bmjopen-2020-039657 PMID: 33310798 29. Raman R, Vasconcelos JC, Rajalakshmi R, Prevost AT, Ramasamy K, Mohan V, et al. Prevalence of diabetic retinopathy in India stratified by known and undiagnosed diabetes, urban–rural locations, and socioeconomic indices: results from the SMART India population-based cross-sectional screening study. Lancet Glob Health. 2022; 10: e1764–e1773. https://doi.org/10.1016/S2214-109X(22)00411-9 PMID: 36327997 30. Rosenstock IM. Historical origins of the health belief model. Health Educ Monogr. 1974; 2: 328–335. 31. Carpenter CJ. A meta-analysis of the effectiveness of health belief model variables in predicting behavior. Health Commun. 2010; 25: 661–669. https://doi.org/10.1080/10410236.2010.521906 PMID: 21153982 32. Green, E. C., & Murphy, E. Health belief model. In W. C. Cockerham, R. Dingwall, & S. R. Quah (Eds.), The Wiley 8 G. ZAIDLIN ET AL. Blackwell encyclopedia of health, illness, behavior, and society. 2014 (pp. pp. 766–769). Wiley Blackwell. 33. Swain J. A hybrid approach to thematic analysis in qualitative research: Using a practical example. SAGE Publications Ltd; 2018. 34. Clarke V, Braun V. Teaching thematic analysis: Overcoming challenges and developing strategies for effective learning. Psychologist. 2013; 26: 120–123. 35. Agarwal S, Mahajan S, Rani PK, Raman R, Paul PG, Kumaramanickavel G, et al. How high is the non- response rate of patients referred for eye examination from diabetic screening camps?. Ophthalmic Epi- demiol. 2005; 12: 393–394. https://doi.org/10.1080/09286580500278418 PMID: 16283991 36. Xiong Y, Liu L, Chen Y, Zhao J. Survey on the awareness of diabetic retinopathy among people with dia- betes in the Songnan community of Shanghai. Int Eye Sci. 2015; 15: 1117–1121. 37. Adriono G, Wang D, Octavianus C, Congdon N. Use of eye care services among diabetic patients in urban Indonesia. Arch Ophthalmol. 2011; 129: 930–935. https://doi.org/10.1001/archophthalmol.2011. 147 PMID: 21746983 38. Rani PK, Raman R, Subramani S, Perumal G, Kumaramanickavel G, Sharma T. Knowledge of diabetes and diabetic retinopathy among rural populations in India, and the influence of knowledge of diabetic ret- inopathy on attitude and practice. Rural Remote Health. 2008; 8: 1–9. PMID: 18656993 39. Thapa R, Paudyal G, Maharjan N, Bernstein PS. Demographics and awareness of diabetic retinopathy among diabetic patients attending the vitreo-retinal service at a tertiary eye care center in Nepal. Nepal J Ophthalmol. 2012; 4: 10–16. https://doi.org/10.3126/nepjoph.v4i1.5844 PMID: 22343990 40. Piyasena MM, Murthy GV, Yip JL, Gilbert C, Peto T, Premarathna M, et al. A qualitative study on barri- ers and enablers to uptake of diabetic retinopathy screening by people with diabetes in the Western Province of Sri Lanka. Trop Med Health. 2019; 47: 1–0. 41. Alharbi AMD, Alhazmi AMS. Prevalence, Risk Factors, and Patient Awareness of Diabetic Retinopathy in Saudi Arabia: A Review of the Literature. Cureus. 2020; 12: e11991. https://doi.org/10.7759/cureus. 11991 PMID: 33437546 42. Klein R, Lee KE, Gangnon RE, Klein BE. The 25-year incidence of visual impairment in type 1 diabetes mellitus: the Wisconsin Epidemiologic Study of Diabetic Retinopathy. Ophthalmol. 2010; 117: 63–70. https://doi.org/10.1016/j.ophtha.2009.06.051 PMID: 19880184 43. Solomon SD, Chew E, Duh EJ, Sobrin L, Sun JK, VanderBeek BL, et al. Diabetic retinopathy: a position statement by the American Diabetes Association. Diabetes Care. 2017; 40: 412–418. https://doi.org/10. 2337/dc16-2641 PMID: 28223445 44. Sanghamitra P, Eunice L, Sandipana P, Shayma D, Pranab M. Type 2 diabetes and physical activity: barriers and enablers to diabetes control in Eastern India. Prim Health Care Res Dev. 2019; 20: 44. 45. Green J, Siddall H, Murdoch I. Learning to live with glaucoma: a qualitative study of diagnosis and the impact of sight loss. Soc Sci Med. 2002; 55: 257–267. https://doi.org/10.1016/s0277-9536(01)00169-1 PMID: 12144140 46. Crabb DP, Smith ND, Glen FC, Burton R, Garway-Heath DF. How does glaucoma look?: patient per- ception of visual field loss. Ophthalmol. 2013; 120: 1120–1126. https://doi.org/10.1016/j.ophtha.2012. 11.043 PMID: 23415421 PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 22 / 23 PLOS ONETreatment-seeking behaviour of patients with sight threatening diabetic retinopathy in India 47. Devenney R, O’Neill S. The experience of diabetic retinopathy: A qualitative study. Br J health Psychol. 2011; 16: 707–721. https://doi.org/10.1111/j.2044-8287.2010.02008.x PMID: 21988061 48. Das M, Angeli F, Krumeich AJ, van Schayck OC. The gendered experience with respect to health-seek- ing behaviour in an urban slum of Kolkata, India. Int J Equity Health. 2018; 17: 1–4. 49. Greenwood RA, George R, Murphy Jr EE, Teahan J, Madero S, Monserrat S, et al. An exploration of cross-cultural and gender differences in East Asia, South Asia and Southeast Asia. 2016 50. Sheppler CR, Lambert WE, Gardiner SK, Becker TM, Mansberger SL. Predicting adherence to diabetic eye examinations: development of the compliance with Annual Diabetic Eye Exams Survey. Ophthal- mol. 2014; 121: 1212–1219. 51. Yahya T, Nazzal Z, Abdul-Hadi AR, Belkebir S, Hamarshih M, Fuqaha A, et al. Diabetic retinopathy screening barriers among Palestinian primary health care patients: a qualitative study. J Diabetes Metab Disord. 2020; 19: 875–881. https://doi.org/10.1007/s40200-020-00575-4 PMID: 33520809 52. Ministry of Health & Family welfare, India. National programme for prevention and control of cancer, dia- betes, cardiovascular disease & stroke (NPCDCS revised operational guideline) [Internet]. New Delhi, 2013: 1–78. https://dghs.gov.in/WriteReadData/userfiles/file/Publication/ OperationalGuidelinesofNPCDCS(Revised—2013-17)_1.pdf 53. Van Dam HA, Van Der Horst FG, Knoops L, Ryckman RM, Crebolder HF, Van Den Borne BH. Social support in diabetes: a systematic review of controlled intervention studies. Patient Educ Couns. 2005; 59: 1–2. https://doi.org/10.1016/j.pec.2004.11.001 PMID: 16198213 54. Durai V, Samya V, Akila GV, Shriraam V, Jasmine A, Muthuthandavan AR, et al. Self-care practices and factors influencing self-care among type 2 diabetes mellitus patients in a rural health center in South India. J. Educ. Health Promot. 2021; 10. PLOS ONE \| https://doi.org/10.1371/journal.pone.0270562 June 15, 2023 23 / 23 PLOS ONE
10.1371_journal.pone.0263447	RESEARCH ARTICLE Stakeholder perceptions of bird-window collisions Georgia J. RiggsID, Omkar Joshi, Scott R. Loss Department of Natural Resource Ecology and Management, Oklahoma State University, Stillwater, Oklahoma, United States of America georgia.riggs@okstate.edu Abstract Bird-window collisions are a major source of human-caused avian mortality for which many mitigation and prevention options are available. However, because very little research has characterized human perspectives related to this issue, there is limited understanding about the most effective ways to engage the public in collision reduction efforts. To address this research need, we: (1) evaluated how two stakeholder groups, homeowners and conserva- tion practitioners, prioritize potential benefits and obstacles related to bird-window collision management, (2) compared priorities between these groups, and (3) evaluated potential conflicts and collective strength of opinions within groups. We addressed these objectives by merging the strengths, weaknesses, opportunities, and threats (SWOT) and analytic hierarchy process (AHP) survey approaches. Specifically, survey respondents made pair- wise comparisons between strengths and weaknesses (respectively, direct outcomes and barriers related to management, such as fewer collisions and increased costs) and opportu- nities and threats (indirect outcomes and barriers, such as increased bird populations and fewer resources for other building-related expenses). Both homeowners and conservation practitioners ranked strengths and opportunities higher than weaknesses and threats, indi- cating they have an overall positive perception toward reducing bird-window collisions. How- ever, key obstacles that were identified included costs of management and a lack of policy and guidelines to require or guide management. These results suggest that substantial advances can be made to reduce bird-window collisions because both homeowners and conservation practitioners had positive views, suggesting their receptivity toward collision management measures. However, because of more neutral views and conflicting responses within the homeowner group, results also highlight the importance of targeting homeowners with education materials that provide information about bird-window collisions and solutions that reduce them. Because bird-window collisions are a human-caused phe- nomenon, such information about human perspectives and priorities will be crucial to addressing this threat and thus benefitting bird populations. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Riggs GJ, Joshi O, Loss SR (2022) Stakeholder perceptions of bird-window collisions. PLoS ONE 17(2): e0263447. https://doi.org/ 10.1371/journal.pone.0263447 Editor: Christopher A. Lepczyk, Auburn University, UNITED STATES Received: May 31, 2021 Accepted: January 19, 2022 Published: February 10, 2022 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0263447 Copyright: © 2022 Riggs et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. Funding: This research was funded by Oklahoma State University Department of Natural Resource Ecology and Management (https://go.okstate.edu/ PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 1 / 20 PLOS ONEundergraduate-academics/majors/natural- resource-ecology-and-management.html) and Hatch Grant funding (grant numbers: OKL02915, OKL03150) from the USDA National Institute of Food and Agriculture (https://nifa.usda.gov/). Funding was obtained by SRL. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Stakeholder perceptions of bird-window collisions Introduction As earth’s human population continues to grow [1], human actions and ways of life increas- ingly affect wildlife and their habitats, and the many sources of unintended, direct wildlife mortality are a major component of these human impacts [2–4]. Among direct sources of avian mortality, collisions of birds with buildings and their windows are a top global threat. Window collisions cause between 365 and 988 million bird deaths annually in the United States alone [5] and are also a top threat to birds in other countries (e.g., Canada, Mexico, Bra- zil, Spain, Singapore, South Korea) [6–11]. Birds collide with glass because they are unable to perceive it as a barrier due to its reflective and transparent qualities [12], and because artificial light at night confuses and draws migrating birds near buildings, elevating collision risk [13, 14]. Bird collisions occur at a wide variety of building types; tall buildings such as skyscrapers have higher per-building collision rates, but smaller and far more abundant residential build- ings account for higher cumulative mortality despite lower per building collision rates [5, 7]. Many studies have identified factors that lead to spatiotemporal variation in bird-building collisions. Temporal factors include weather, seasonality, migration phenology, and fluctua- tions in bird abundance [15–17]. Spatial factors include building-related features like amount of glass, building shape, and nearby vegetation [18–20], as well as broader landscape features like surrounding greenspace and urbanization intensity [21]. Research into correlates of bird- window collisions has led to development of recommendations and management approaches that can be used to reduce collisions. Technologies and commercially available products that reduce glass reflection and transparency have been developed, tested, and marketed, and guidelines to make newly constructed buildings bird-friendly (e.g., by reducing amount of glass or using opaque, fritted, or colored glass) have also been summarized [18, 22, 23]. Munic- ipal, state, and federal policy guidelines and regulations to implement such bird-friendly approaches have also been adopted or are under consideration. These include, for example, Standards for Bird-Safe Buildings in San Francisco, California, U.S.A [24], Buildings, Bench- marks, and Beyond in Minnesota, U.S.A. [25], Best Practices for Bird-friendly Glass and Best Practices for Effective Lighting in Toronto, Canada [26], and the Bird Safe Buildings Act of 2021 currently under consideration by the U.S. federal government [27]. Bird-window collisions occur in areas with human infrastructure, and humans regularly encounter the bird carcasses that result. However, although significant resources have gone into designing and testing mitigation approaches to reduce bird-window collisions, and into developing and implementing bird-friendly policies and guidelines, only two studies have eval- uated human perceptions and priorities related to these practices. In fact, there is a general lack of human dimensions research for nearly all sources of direct, human-caused bird mortal- ity, including other kinds of bird collisions (e.g., with wind turbines, communication towers, and vehicles; but see studies of wildlife predation by domestic cats) [28, 29]. One of the studies that evaluated human perspectives related to bird-window collisions examined the Canadian public’s willingness to pay (WTP) to reduce collisions at their homes [30] and found that WTP was positively associated with homeowner age, income, and interest in birds, among other fac- tors. The other study investigated public perceptions and knowledge about this issue in Costa Rica and concluded that participants were aware of bird-window collisions but not of the large magnitude of the problem [31]. Clarifying how people perceive bird-window collisions, and how much they support mitigation and prevention techniques, is crucial for bird conservation because implementing effective practices generally entails adoption of new products and tech- nologies on buildings, and therefore, requires buy-in from multiple stakeholder groups (e.g., residential homeowners, owners/managers of commercial buildings, building architects, policymakers). PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 2 / 20 PLOS ONEStakeholder perceptions of bird-window collisions We began to address this major research gap by exploring and quantifying perceptions and priorities related to bird-window collisions among a diverse pool of respondents in North America. Our objectives were to: (1) evaluate how two important stakeholder groups (owners of individual residences, i.e., “homeowners,” and conservation practitioners in state, federal, and non-government conservation organizations) perceive and prioritize potential benefits and obstacles related to bird-window collision management, (2) compare priority rankings for benefits and obstacles to management between homeowners and conservation practitioners, and (3) evaluate potential conflicts in priorities within each stakeholder group, as well as the collective strength of group opinions. To address objectives 1 and 2, we merged the strengths, weaknesses, opportunities, and threats (SWOT) and analytical hierarchy process (AHP) analy- ses; the approach of merging these two analyses is frequently used to quantitatively assess and rank perceived benefits and obstacles related to management actions and decisions [32–35]. To address objective 3, we used Manfredo et al.’s [36] potential for conflict index (PCI) to visu- alize within-group conflicts and strength of group opinions, information that can lend addi- tional insight into factors potentially limiting progress in managing bird-window collisions. Methods Study design This study, the survey distribution strategy, and the survey contents were approved by and comply with the Oklahoma State University Institutional Review Board’s (IRB) standards and regulations (approved IRB protocol # IRB-20-202). All survey participants gave consent for participation upon completion of surveys, and data were also analyzed anonymously. To address objectives 1 and 2, we used a combined SWOT-AHP perception analysis approach (i.e., a strengths, weaknesses, opportunities, and threats analysis linked with an analytic hierar- chy process analysis). This merged approach is often used to quantify and rank perceptions about major benefits and obstacles related to issues, actions, and decisions of interest, and to compare benefit and obstacle rankings among diverse stakeholder groups, including for issues in conservation and natural resource management like renewable energy, ecotourism, and land management and policy [32–35, 37, 38]. In the SWOT framework [39], there are 4 catego- ries of factors related to the issue, action, or decision under consideration: strengths, weak- nesses, opportunities, and threats. Strengths and weaknesses are considered internal to an issue, action, or decision. In our case, strengths are direct, immediate outcomes of implement- ing bird-window collision management (e.g., fewer bird collisions) and weaknesses are direct barriers or obstacles to implementing management (e.g., the financial cost of management). Opportunities and threats are considered external to an issue, action, or decision. In our case, opportunities are non-immediate and/or secondary outcomes that indirectly result from implementing management (e.g., increased bird populations due to fewer collisions), and threats are barriers that are not directly related to management but that could arise as manage- ment is carried out (e.g., with collision management expenses, reduced financial resources for other building management-related costs). We used the SWOT approach to ask surveyed stakeholders to prioritize strengths, weaknesses, opportunities, and threats related to bird-win- dow collision mitigation and prevention (the specific factors used for each of these 4 SWOT categories are under “Survey Questionnaire Details”). The ultimate goal of a SWOT analysis is to determine perceptions of stakeholders to help develop a strategy that optimizes the tradeoff between strengths and weaknesses of various options, while considering both internal and external factors. When used alone, SWOT does not allow quantitative ranking of factors within or across different categories, making it difficult to draw conclusions about perceptions. The AHP, however, is a generalized method to rank decision problems that assumes independence PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 3 / 20 PLOS ONEStakeholder perceptions of bird-window collisions among options; when combined with SWOT, AHP allows quantitative comparisons of differ- ent SWOT factors, which helps determine the relative importance of a decision [39]. As a multi-criteria decision-making tool, AHP assigns relative weights to factors of interest based on 2-way comparisons between factors [40]; this allows objective evaluation of the degree of agreement (or disagreement) between factors. Stakeholder groups and strategy to distribute survey questionnaire Initially, we sought to investigate priorities of four stakeholder groups: architects, home- owners, and conservation practitioners in both government agencies and non-governmental organizations (NGOs). Each of these groups can play a key role in managing bird-window col- lisions. Architects can help reduce collisions by working from the top down to incorporate mitigation and prevention measures, within policy parameters, into design and construction of new buildings [41, 42]. Homeowners act from the bottom-up as consumers by expressing their values and desires, buying and living in houses, and deciding whether to manage their properties in ways that benefit birds (e.g., feeding birds or applying films/decals to windows to reduce collisions) [42, 43]. Government and NGO conservation practitioners are both knowl- edgeable about and advocate for wildlife, but these two groups may enact change in different ways. Government (federal, state/provincial, and tribal) practitioners help inform policy devel- opment with research and management, and while NGOs can also help inform policy, they typically engage members of the public through activities such as education campaigns, volun- teering, and public funding [41, 42]. To recruit respondents from all stakeholder groups (architects, homeowners, government conservation practitioners, and NGO conservation practitioners) and from as broad of a geo- graphic area as possible, we used snowball sampling, a nonprobability sampling method that uses gateway contacts who can take the survey themselves and are asked to forward the survey invitation to relevant contacts in their stakeholder group [44]. For this study, gateway contacts were the authors’ personal or professional contacts in each stakeholder group, including 17 architects, 66 homeowners, 36 government practitioners, and 20 NGO conservation practi- tioners. Most of these contacts lived and worked in the United States (18 U.S. states repre- sented), but Canada was also represented. Recruitment emails were tailored to each stakeholder group and sent from the authors to gateway contacts; these emails contained a brief overview of the project, a request for participation, a link to sign up to take the survey, a link to a recruitment video on YouTube, and a request that respondents share recruitment materials with colleagues [45]. The recruitment video contained a brief overview about the issue of bird-window collisions and the objectives of this research project, as well as a request for participation and to forward the recruitment materials. In addition to using gateway con- tacts, we also actively recruited respondents using social media platforms, including Facebook and Twitter [46, 47]. Recruitment via Facebook and Twitter included brief posts on the authors’ profile pages, which are followed by numerous professional contacts with formal posi- tions in conservation science and management (including government and NGO conservation practitioners), and by nonprofessional contacts that include numerous homeowners. These Facebook and Twitter posts contained information about the project, the recruitment video, a call for participation, a link to sign up to take the survey, and a request to share recruitment materials. Of note, mixed data collection methods involving focus group meetings, web sur- veys, and email contacts have been commonly adopted in SWOT-AHP based studies [34, 37, 48]. Accordingly, to broaden participation and increase replication of responses from mem- bers of the homeowner stakeholder group, we reached out to multiple neighborhood home- owner’s associations (HOA) in Stillwater, Oklahoma, USA, the location of the authors’ home PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 4 / 20 PLOS ONEStakeholder perceptions of bird-window collisions institution (Oklahoma State University). We used this approach because we expected that snowball sampling would result in recruitment of relatively few homeowners. Recruitment materials were sent to publicly available email addresses of HOA board member contacts; again, we requested participation in the survey and dissemination of recruitment materials to other HOA board members and neighborhood residents. Survey questionnaire details Using the merged SWOT-AHP approach first entails development of a survey that contains a list of top strengths, weaknesses, opportunities, and threats regarding the issue at hand. These SWOT lists are often developed from a longer list of candidate factors with assistance of sub- ject-matter experts [37]. We created a list of candidate SWOT factors related to bird-window collision management based on our own subject matter expertise, which includes familiarity with the scientific and gray literature on this topic, and years of interactions and collaborations with key stakeholders in federal/state agencies and NGOs. After drafting the initial list of can- didate SWOT factors, we asked three external bird-window collision experts to rank them by importance. Expert responses for each candidate factor were counted and weighted based on ranking to create a final SWOT list containing the four top-ranked factors in each category (Table 1). Following methodology used by similar SWOT studies, we next solicited stakeholder opin- ions in two rounds of surveys, with each containing multiple pairwise comparisons between SWOT factors using a scale of one to nine [32, 37, 49]. Specifically, a value of 1 indicated an opinion that one factor was “extremely important,” a value of 9 indicated an opinion that the other factor was extremely important, and a value of 5 indicated an opinion that the two fac- tors were “equally important” (see Fig 1 for visual representation of scale). For Survey 1, all possible pairwise comparisons were made between factors within (but not between) SWOT categories. For example, all possible 2-way comparisons were made among strengths (e.g., Fewer collisions compared to Fewer bird carcasses to clean up), but in this survey, strengths were not compared to weaknesses, opportunities, or threats (see example comparison in Fig 1 and S1 File for full Survey 1 contents). We created Survey 2 based on top-ranked factors Table 1. List of all SWOT factors. Strengths Weaknesses S1: Fewer collisions W1: No economic incentives building for bird-friendly buildings S2: Fewer carcasses to clean up W2: Lack of architect experience in bird-friendly design S3: Fewer people witnessing collisions W3: Lack of availability of expert consultation for bird- friendly design S4: Fewer stunned birds that die of other causes while recovering from colliding W4: Financial burden of treating glass or including bird- friendly design in building process Opportunities O1: Recovering bird populations O2: Public exposure to bird-friendly options Threats T1: Unknown social acceptance of bird-friendly treatments and design T2: Lack of understanding of federal/state policy on bird- window collisions O3: Consideration of birds in building design becoming a norm/standard T3: Reduced resources available to spend on other facilities maintenance/improvements O4: Greater energy efficiency of buildings T4: No federal/state policy in many areas Finalized list of strengths, weaknesses, opportunities, and threats (SWOT) containing the top four factors for each category that were used to evaluate stakeholder perceptions regarding bird-window collisions. https://doi.org/10.1371/journal.pone.0263447.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 5 / 20 PLOS ONEStakeholder perceptions of bird-window collisions Fig 1. SWOT survey example. Examples of pairwise comparisons within the strengths category of the strengths, weaknesses, opportunities, and threats (SWOT) analysis; this example illustrates the format of Survey 1 distributed to stakeholder groups to evaluate their perceptions and priorities regarding bird-window collision management. https://doi.org/10.1371/journal.pone.0263447.g001 calculated from Survey 1 for each SWOT category (see details of these calculations under “Data Analysis”). These calculations were made separately for each stakeholder group, which allowed us to tailor Survey 2 to each group, a standard practice for SWOT studies. In Survey 2, respondents were asked to make pairwise comparisons of all top-ranking factors between SWOT categories. For example, within the homeowner group, the factor Fewer collisions was identified as the top strength in Survey 1, and No federal/state policy in many areas was the top threat. Thus, respondents were asked to compare these two factors (see S2 and S3 Files for full Survey 2 contents for each stakeholder group). All surveys were administered using the online platform Qualtrics [50], and both surveys had the same general format. Both surveys contained an introductory page displaying informa- tion about the study, including the study’s purpose, what to expect, risks associated with par- ticipating, and a confidentiality statement. Next, the survey asked respondents to indicate which stakeholder group they belonged to. The following section contained a brief introduc- tion to the issue of bird-window collisions (to give respondents introductory background or to reorient them to the issue), as well as a table containing all of the SWOT factors. To minimize the collection of personally identifiable information and to retain survey anonymity, we only collected contact information (names and emails) of potential respondents during the initial recruitment period (i.e., the period during which we asked stakeholders to sign up to take the survey, but before the survey was distributed). During survey periods, surveys were completed anonymously; therefore, we could not monitor which people who signed up (including gate- way contacts and other people reached through snowball and purposive sampling) actually PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 6 / 20 PLOS ONEStakeholder perceptions of bird-window collisions completed the surveys. For Survey 2, all individuals who signed up to take Survey 1 were again contacted, but we requested that only those that completed Survey 1 complete Survey 2. Survey 1 was administered from 1 June 2020 to 30 June 2020, and Survey 2 was administered from 13 July 2020 to 12 August 2020. For all stakeholder groups and sampling approaches, we waited two weeks before sending one reminder to complete the survey to allow adequate time for par- ticipants to respond to the original request [51]. Data analysis Analyses of survey response data followed methods of other SWOT-AHP studies (e.g., Starr et al. 2019 and Joshi et al. 2020) [37, 52] that adapted their analyses from Saaty [40]. The same general procedures were used to analyze results from Survey 1 (comparisons within SWOT cate- gories) to determine factor priorities for Survey 2, and to analyze results from Survey 2 (compari- sons of top-ranked factors between SWOT categories). First, to calculate the weighted geometric mean for each factor in each SWOT category, and also separately for each stakeholder group, we collated response data for each pairwise comparison into counts according to the selection scale of one to nine (See S1 Dataset for calculated geometric means). Counts were then weighted reciprocally, multiplied, and taken to the power of one over the total number of counts [53]. Each weighted geometric mean was entered into a standard reciprocal matrix, and values were then normalized and placed into a weighted reciprocal pairwise matrix. The weighted reciprocal pairwise matrix was used to calculate factor priority values for each factor in each SWOT cate- gory and stakeholder group; these values were used to evaluate relative importance of factors within each SWOT category (all factor priority values within a category sum to one). The stan- dard reciprocal matrix and factor priority values for each category were also used to calculate a consistency index, which when used with a predetermined random index (based on the number of SWOT factors within a category) determines the consistency ratio, a metric indicating the consistency of responses among respondents within a stakeholder group [39, 52]. Pairwise com- parisons within each SWOT category were determined to be internally consistent if the consis- tency ratio (calculated for each SWOT category within each stakeholder group and for both surveys) was less than 10%; however, consistency ratios up to 20% are considered acceptable [34, 40, 49, 52]. When we conducted preliminary analyses of Survey 1 responses, we calculated unac- ceptably high consistency ratios within the architect and NGO practitioner groups that were most likely attributable to small sample sizes of recruited respondents (n = 12 for each group). We therefore excluded data for architects, and due to similarities between the groups and to pre- vent data loss, we combined government practitioners (n = 26) and NGO practitioners into a single group (conservation practitioners, n = 38). Thus, our final analysis of Survey 1 (and subse- quently, Survey 2) included two stakeholder groups, homeowners (Survey 1: n = 52; Survey 2: n = 33) and conservation practitioners (Survey 1: n = 38; Survey 2: n = 41). Our receipt of more conservation practitioner responses for Survey 2 than Survey 1 was unexpected because we only asked recruits to complete the second survey if they had already completed the first survey. This result likely arose because we had to exclude a small number of Survey 1 responses that were incomplete or contained response errors (7 surveys excluded for homeowners; 4 for conserva- tion practitioners). Regardless of the explanation, we have no reason to believe that receiving slightly more Survey 2 results biased our results. The last steps in the SWOT-AHP analysis were to calculate global and group priority values. Global priority values rank individual SWOT factors among all categories for each stakeholder group; these values allow for comparison among stakeholders’ perceptions and priorities, as well as evaluation of SWOT factor priority rankings against each other [32, 37, 49]. Global pri- ority values within each SWOT category were then added together to create group priority PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 7 / 20 PLOS ONEStakeholder perceptions of bird-window collisions values that represent the priority of each SWOT category as a whole. We also followed previ- ous literature (e.g., Dwivedi & Alavalapati 2009 and Joshi et al. 2018) [32, 34] to generate per- ception maps, which illustrate differences in global priority values and allow direct comparisons among all SWOT factors and between stakeholder groups. To address objective 3, we applied Manfredo et al.’s [36] potential for conflict index (PCI) to the Survey 2 responses (see S1 Dataset for PCI calculations); the PCI allows visualization of potential conflicts in perceptions within stakeholder groups, and of the collective strength (vs. neutrality) of group opinions [54], information that can lend additional insight into factors potentially limiting progress in addressing bird-window collisions. We used the PCI2, an extension of PCI that is used for response data from a scalar survey to visually display degree of conflict (i.e., opposite of agreement) in responses among respondents in a stakeholder group, as well as neutrality of responses [36, 54]. In this case, the scalar survey questions were pairwise comparisons that respondents completed in Survey 2. With regard to neutrality, pair- wise comparisons that are near five for a stakeholder group indicate factors perceived as Equally important (indicated as bubbles close to the x-axis on PCI graphs). Comparisons that are lower (near one) or higher (near nine) toward either of the factors being compared repre- sent an average group perception that one factor is Extremely important relative to the other (bubbles farther from the x-axis). Regarding degree of conflict, this value ranges between 0 and 1, with values close to 0 indicating little conflict (strong agreement on a pairwise comparison among respondents in a group, indicated as small bubbles), and values close to 1 indicating complete conflict (i.e., responses on a pairwise comparison equally divided between the two extreme values on the response scale, indicated as large bubbles) [36, 55]. Results Stakeholder priorities for different SWOT categories Our survey likely had a nationwide or even broader scope, as our gateway contacts represented at least 18 U.S. states and Canada. However, the exact geographic distribution of survey respondents is unknown because surveys were completed anonymously to minimize collection of personally identifiable information, and because the snowball sampling method we used entailed recruitment of additional respondents beyond our gateway contacts. For all SWOT categories except two in the conservation practitioner group for Survey 1, consistency ratios were <10%, indicating consistent responses within stakeholder groups. For conservation prac- titioners, the weaknesses and opportunities categories had consistency ratios of 19% and 18%, respectively, indicating some inconsistency. Nonetheless, consistency ratios <20% are consid- ered acceptable for drawing inferences [34, 49]. A summary of SWOT factor, group, and global priorities for homeowners and conservation practitioners is in Table 2. Group priorities for homeowners for strengths, weaknesses, oppor- tunities, and threats were 24%, 15%, 40%, and 21%, respectively, and group priorities for con- servation practitioners were 24%, 15%, 52%, and 9%, respectively. For homeowners and conservation practitioners, perceptions about potential outcomes of bird-window collision mitigation and prevention were generally positive, as evidenced by summed percentages of group priorities for strengths and opportunities (64% and 76% for homeowners and conserva- tion practitioners, respectively). As indicated by group priority values for threats, homeowners gave greater priority (21%) to threats than did conservation practitioners (9%). Stakeholder priorities for different factors within SWOT categories As evident from the above-presented group priority values, homeowners prioritized opportu- nities overwhelmingly over strengths, weaknesses, and threats. Among opportunities, PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 8 / 20 PLOS ONEStakeholder perceptions of bird-window collisions Table 2. Factor, global, and group priorities for all SWOT factors for each stakeholder group. SWOT Factors Factor Priority Global Priority Homeowner Conservation Practitioner Homeowner Conservation Practitioner S1: Fewer collisions S2: Fewer carcasses to clean up S3: Fewer people witnessing collisions S4: Fewer stunned birds that die of other causes while recovering from colliding Group Priorities for Strengths W1: No economic incentives for building for bird-friendly buildings W2: Lack of architect experience in bird-friendly design W3: Lack of availability of expert consultation for bird-friendly design W4: Financial burden of treating glass or including bird-friendly design in building process Group Priorities for Weaknesses O1: Recovering bird populations O2: Public exposure to bird-friendly options O3: Consideration of birds in building design becoming a norm/standard O4: Greater energy efficiency of buildings Group Priorities for Opportunities T1: Unknown social acceptance of bird-friendly treatments and design T2: Lack of understanding of federal/state policy on bird-window collisions T3: Reduced resources available to spend on other facilities maintenance/ improvements T4: No federal/state policy in many areas Group Priorities for Threats 0.46 0.11 0.09 0.34 0.23 0.18 0.31 0.28 0.34 0.18 0.25 0.23 0.19 0.25 0.25 0.31 0.60 0.06 0.07 0.27 0.36 0.13 0.26 0.25 0.45 0.15 0.20 0.21 0.14 0.16 0.36 0.35 0.11 0.03 0.02 0.08 0.24 0.03 0.03 0.05 0.04 0.15 0.14 0.07 0.10 0.09 0.40 0.04 0.05 0.05 0.07 0.21 0.15 0.02 0.02 0.07 0.24 0.05 0.02 0.04 0.04 0.15 0.23 0.08 0.10 0.11 0.52 0.01 0.01 0.03 0.03 0.09 Summary of factors used in strengths, weaknesses, opportunities, and threats (SWOT) analyses related to perceptions and potential outcomes of bird-window collision mitigation and prevention. Factor priority values indicate the relative importance of a single factor within a SWOT category among other factors in the same category (boldfaced factor priority values are the highest prioritized factor for each SWOT category). Global priority values rank individual SWOT factors among all factors and can be compared across SWOT categories. Group priority values (the boldfaced values in “Global Priority” columns) are the sum of global priority values within each SWOT category and are used to compare categories against each other. https://doi.org/10.1371/journal.pone.0263447.t002 Recovering bird populations was the top factor priority (34%), followed by Consideration of birds in building design becoming a norm/standard (25%) and Greater energy efficiency of build- ings (23%). Homeowners prioritized strengths next; highest priority strengths were Fewer colli- sions (46%) and Fewer stunned birds that die of other causes while recovering from colliding (34%). The anthropocentric strengths received lower priority, including: Fewer carcasses to clean up (11%) and Fewer people witnessing collisions (9%). For threats, which homeowners prioritized only slightly behind strengths, the top factor was No federal/state policy in many areas (31%), followed by two equally ranked (25%) priorities: Lack of understanding of federal/ state policy on bird-window collisions and Reduced resources available to spend on other facilities maintenance/improvements. Homeowners prioritized weaknesses lowest, with Lack of avail- ability of expert consultation for bird-friendly design being the top priority (31%) within this category (Table 2). Based on group priority values, conservation practitioners also prioritized opportunities as most important; among opportunities, Recovering bird populations was the top-priority factor (45%). Strengths was the second-highest prioritized category, and top factors in this category were Fewer collisions (60%) and Fewer stunned birds that die of other causes while recovering from colliding (27%). Conservation practitioners gave weaknesses and threats lowest priority. PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 9 / 20 PLOS ONEStakeholder perceptions of bird-window collisions The most highly prioritized weakness was No economic incentives for building bird-friendly buildings (36%); the two top threats were Reduced resources available to spend on other facilities maintenance/improvements (36%) and No federal/state policy in many areas (35%) (Table 2). Stakeholder priorities for different factors across SWOT categories Perception maps (Fig 2A and 2B) illustrate differences in global priorities and allow direct comparisons among all SWOT factors and between stakeholder groups. For homeowners, the opportunity Recovering bird populations (O1) received the highest global priority among all SWOT factors, closely followed by the strength Fewer collisions (S1). Although homeowner priorities for weaknesses and threats were lower than for strengths and opportunities, all threats and some weaknesses still received higher global priorities than the strengths Fewer people witnessing collisions (S2) and Fewer carcasses to clean up (S3). The opportunity Recover- ing bird populations (O1) followed by the strength Fewer collisions (S1) also received the two highest global priorities for conservation practitioners. Additionally, this group prioritized weaknesses over threats while homeowners ranked these categories in the opposite order. Although the two groups had similar broad priorities, such as valuing strengths and oppor- tunities over weaknesses and threats, conservation practitioners gave higher priority to the top factor in some categories, suggesting stronger perceptions toward these factors. Specifically, although Recovering bird populations (O1) was the highest global priority among all SWOT factors for both stakeholder groups, it received a greater global priority value for conservation practitioners (0.23) than homeowners (0.14). Similarly, the top strength (and second highest global priority among all SWOT factors) for both stakeholder groups (Fewer collisions; S1) received a greater global priority value for conservation practitioners (0.15) than for home- owners (0.11) (Table 2). Global priorities also illustrated that both homeowners and conserva- tion practitioners gave low priority to Fewer people witnessing collisions (S2) and Fewer carcasses to clean up (S3) relative to other strengths and many other weakness and threats. Potential for conflict and strength of opinions within stakeholder groups Regarding potential for conflict indices (PCI2) for Survey 2, comparison of the bubbles for homeowners (Fig 3A) and conservation practitioners (Fig 3B) for each pairwise comparison illustrates there was more conflict among responses for homeowners than conservation practi- tioners for 4 of 6 comparisons. Additionally, relative locations of bubbles on the y-axis (which indicates the difference in preference for each priority in a pairwise comparison) illustrate that homeowners were more neutral than conservation practitioners for all 6 pairwise comparisons. Discussion Our results suggest that both homeowners and conservation practitioners have an overall posi- tive perception toward potential benefits related to bird-window collision mitigation and pre- vention measures. This indicates stakeholders may believe that benefits of implementing management to reduce bird-window collisions outweigh any obstacles that may impede such measures. Although generally similar in their positive views, the two stakeholder groups dis- played some differences in their specific priorities regarding strengths, weaknesses, opportuni- ties, and threats surrounding this issue. Specifically, homeowners gave greater priority than conservation practitioners to threats, indicating more concern among homeowners about external obstacles (financial and policy related) that may impede bird-window collision man- agement efforts. PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 10 / 20 PLOS ONEStakeholder perceptions of bird-window collisions Fig 2. Perception maps of SWOT global priorities for each stakeholder group. Perception maps illustrating homeowner (a) and conservation practitioner (b) strength, weakness, opportunity, and threat (SWOT) global priorities for a study evaluating perceptions about potential outcomes of bird-window collision mitigation and prevention. Factors with the highest global priority are farthest from the origin. S1: Fewer collisions; S2: Fewer carcasses to clean up; S3: Fewer people witnessing collisions; S4: Fewer stunned birds that die of other causes while recovering from colliding. W1: No economic incentives for building for bird-friendly buildings; W2: Lack of architect experience in bird-friendly design; W3: Lack of availability of expert consultation for bird-friendly design; W4: Financial burden of treating glass or including bird-friendly design in building process. O1: Recovering bird populations; O2: Public exposure to bird-friendly options; O3: Consideration of birds in building design becoming a norm/standard; O4: Greater energy efficiency of buildings. T1: Unknown social acceptance of bird-friendly treatments and design; T2: Lack of understanding of federal/state policy on bird-window collisions; T3: Reduced resources available to spend on other facilities maintenance/improvements; T4: No federal/state policy in many areas. https://doi.org/10.1371/journal.pone.0263447.g002 Stakeholder perceptions about bird-window collision management Results indicate that the homeowner and conservation practitioner groups, while in general agreement on their positive perceptions about managing bird-window collisions, each have unique aspects of their perceptions that are important to consider in order to make headway PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 11 / 20 PLOS ONEStakeholder perceptions of bird-window collisions Fig 3. Potential for conflict indices from survey 2 for each stakeholder group. Illustration of the potential for conflict index (PCI2) based on homeowner (a) and conservation practitioner (b) responses to Survey 2 in a study evaluating perceptions about potential outcomes of bird-window collision mitigation and prevention. Bubble size and values correspond and indicate the dispersion (conflict) among respondent answers (larger bubbles/numbers indicate greater conflict). The location of the bubble indicates the scale mean or the direction respondents lean in their answers to pairwise comparisons (e.g., 5 indicates completely neutral; values lower and higher than 5 indicate more non- neutral perceptions). Each bubble is an individual pairwise comparison indicated by the labels. Pairwise comparisons correspond visually to the y-axis scale (e.g., for S1-W3, 1 corresponds to S1 and 9 corresponds to W3). For a description of all strengths (S), weaknesses (W), opportunities (O), and threats (T), see Table 1. https://doi.org/10.1371/journal.pone.0263447.g003 in addressing this conservation issue. As evidenced by the PCI analysis, homeowners had more conflict in their responses to pairwise comparisons than conservation practitioners, indi- cating differing opinions within the group. PCI analysis also indicated that homeowners were more neutral than conservation practitioners in their responses, demonstrating differing or a potential lack of strong opinions within the group. Although we provided contextual informa- tion about this project in the survey’s introductory materials, a lack of prior knowledge about PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 12 / 20 PLOS ONEStakeholder perceptions of bird-window collisions the issue—which was anecdotally revealed from comments made by gateway contacts in the homeowner group—could have contributed to their relatively neutral perceptions and con- flicting responses. The less-conflicting responses within the conservation practitioner group could be due to greater knowledge about the issue or more cohesion within the group due to a shared field of profession and its associated sources of information. Specifically, those in the field of wildlife conservation likely have greater, and perhaps more consistent, exposure to major bird conservation issues through training opportunities, professional conferences, social media networks, newsletters, and scientific publications. It is important to note that the home- owner group included gateway contacts from a wide variety of professional backgrounds, which could explain the lesser degree of agreement within the group. As evidenced by high group priority values for the strength and opportunity categories, as well as high global priority values for individual strengths and opportunities, our results indi- cate that both stakeholder groups have positive views about bird-window collision mitigation and prevention measures. Members of these groups may therefore be willing to participate in or support implementation of measures to reduce bird collisions. Because the top ranked strengths and opportunities capture outcomes related to bird conservation and welfare (e.g., recovering bird populations), not anthropocentric benefits (e.g., no longer having to clean up or observe collisions), our results suggest that stakeholders value mitigating and preventing collisions for the sake of the birds themselves. This result demonstrates that stakeholders may have a general sense of caring and responsibility for birds—and/or that they view birds as aes- thetically, culturally, or economically valuable [56, 57]—which lends additional support to the potential acceptability and implementation of management measures. Due to a greater degree of neutrality and lack of strong opinions within the homeowner group (as illustrated by the PCI), and because some homeowners in our study were not previously aware of bird-window collisions and underlying challenges, our findings suggest a strong need for public education on this issue. Advantageously, the positive perceptions about reducing bird-window collisions, and the apparently bird-centric reasons behind these positive perceptions, suggest that members of the public may be receptive to further education about this issue. Menacho-Odio [31] also investi- gated public perception and knowledge of bird-window collisions in Monteverde, Costa Rica, and concluded that while participants had general knowledge of the issue, few were aware of the magnitude of the problem. This previous study recommended targeted education that informs people about the large number of bird-window collisions that occur, as well as effec- tive methods for preventing collisions. There are multiple publicly available resources from which individuals can learn about bird-window collisions and ways to reduce them. For exam- ple, the American Bird Conservancy (ABC) has published a website geared toward the public [58], a Bird-Friendly Building Design booklet targeting all types of building owners and man- agers, as well as architects [22], interactive web resources and educational materials for home- owners and architects, and a framework to help policy makers develop ordinances and legislation to reduce collisions. Similar and complementary resources to improve stakeholder knowledge about bird-window collisions have also been developed by other conservation orga- nizations and agencies (e.g., USFWS 2021; National Audubon Society 2021; FLAP Canada 2019) [59–61]. While many resources are available, active education on this topic would also be beneficial. Specifically, increased funding and staffing to expand the delivery and interpreta- tion of such resources to stakeholders, along with research to improve understanding of how best to develop and distribute these resources to ensure they are used, are needed to make fur- ther headway in reducing bird-window collisions. As evident from the factor and global priority values for threats, homeowners highly priori- tized policy-related obstacles to bird-window collision mitigation and prevention. However, PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 13 / 20 PLOS ONEStakeholder perceptions of bird-window collisions importantly, multiple states, cities, and municipalities across North America have already enacted policies designed to reduce bird-window collisions, including San Francisco, Califor- nia, U.S.A. [24] and Minnesota, U.S.A. [25]. The U.S. House of Representatives also approved legislation (Bird Safe Buildings Act of 2021) that would require bird-friendly measures at many new and renovated U.S. federal buildings; however, this act has not yet passed the U.S. Senate [27]. Thus, while there is concern among homeowners about potential policy-related obstacles, many may not know that relevant policies already exist. This points again to the importance of education, as increasing awareness of existing and proposed policies could increase support for them among the public, and therefore, among policymakers. Beyond educating homeowners about existing and planned policies related to bird-window collisions, homeowners could also be informed that implementing bird-friendly measures at homes might be their responsibility even with policies in existence. To date, no legislation and policies have focused on residential structures, and the proposed U.S. federal bill only focuses on public buildings. Thus, there are no formal mechanisms to ensure that collisions are reduced at residences, even though residences collectively cause a large proportion of total bird collisions [5, 7]. Although public education may encourage some homeowners to expend their own resources on measures to reduce bird-window collisions, formal programs to encourage these actions may be necessary to ensure that a large proportion of homes become bird-friendly in the future, especially for lower income residents that lack expendable resources to pay for such measures. Examples of such potential programs include conservation grants/subsidies that help pay for materials that make existing windows more bird-friendly, and revisions to existing sustainability or wildlife-friendly certification programs to specifically incorporate considerations related to reducing bird-window collisions. Our analysis identified other potential barriers to widespread bird-window collision man- agement. For example, homeowners identified a lack of availability of expert consultation as another top threat. Although the above-mentioned education campaigns could help empower homeowners to reduce collisions themselves, this result suggests that widespread adoption of collision management practices at homes may require increased training of consultants and outreach professionals that convey information about collision management. Conservation practitioners identified a lack of resources available to spend on other facilities/maintenance improvements as a top threat arising from the costs of collision management. In addition to emphasizing the need for low-cost management options, this result suggests that approaches that reduce collisions while meeting other facilities-related needs may be especially likely to be adopted. Notably, some approaches that are highly effective in reducing bird-window colli- sions, including reducing nighttime lighting [14] and some of the films, coatings, and decals adhered to windows to make them more visible [22], also may contribute to reducing build- ing-related energy costs. Communicating the dual benefits of such approaches may lead to greater adoption of bird-friendly building management techniques. Limitations and future research While this research provides valuable information to advance efforts to manage bird-window collisions, there were some limitations and potential biases related to our analyses. We were, for example, unable to analyze perspectives of architects as an independent stakeholder group due to limited recruitment for participation in our surveys. Architects are a crucial stakeholder in the issue of bird-window collisions, and further research should seek to thoroughly evaluate their perceptions about this topic. The low number of respondents for architects leads to the question of how best to reach and engage with this stakeholder group. Potential routes to engage architects include having bird-window collision researchers present at architectural PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 14 / 20 PLOS ONEStakeholder perceptions of bird-window collisions society conferences, creating publication materials geared toward architects, or reaching out directly to architectural societies or firms about bird-window collisions. Another limitation concerns the representativeness of our sample of survey respondents, which relates both to the limited sample size of respondents and mixed-data collection approach that used gateway contacts and recruitment through social media platforms. Nota- bly, the AHP approach does not require large sample sizes to result in statistically robust results that are useful for understanding stakeholder perceptions and informing management decisions [62]. Instead, reliability of results from this approach is interpreted using consistency ratios, which indicate the degree of consistency of responses within stakeholder groups. Con- sistency ratios for groups used in our analyses were considered acceptable [63], suggesting our results are reliable. However, because many of the gateway contacts we recruited for the home- owner group were our personal and professional contacts, our sample of homeowners could have been biased toward bird enthusiasts rather than providing full representation of the diver- sity within this group. Nonetheless, our homeowner sample contained many respondents beyond the gateway contacts that we did not know personally, indicating that there may have been variation in levels of interest or support for bird-window collision management and wild- life conservation more broadly. Although our approach does not require large sample sizes, we caution against making broad generalizations from our results, especially for the homeowner group, due to these potential issues regarding sample representativeness. Our results lay a foundation for future research into stakeholder perceptions, priorities, and potential disputes and conflicts related to bird-window collision management. Conducting research to better understand motivations and barriers to behavioral change will be crucial for designing collision management programs that garner broad support and participation from the public. In this study, we examined stakeholder perceptions and priorities, but other impor- tant factors that influence behavioral changes (e.g., social and cultural norms, institutional and economic factors) should also be evaluated [64]. Further, research that identifies social-psy- chological barriers that may lead to conflicts among groups (e.g., conservation organizations recommending collision management approaches vs. building management entities resistant to recommendations) could facilitate more-rapid adoption of bird-friendly building design, and similar research related to the green building movement may be instructive for this issue [65]. We did not collect demographic information from respondents, nor did we know the geographic representation of our sample other than for gateway contacts. Because the factors that influence behaviors, perceptions, and conflicts can vary regionally and among demo- graphic groups (e.g., among different age groups), future research could evaluate how percep- tions about bird-window collisions vary regionally and in relation to various demographic factors. Another essential area of future research is to evaluate stakeholders’ willingness to pay (WTP) for measures to reduce collisions. Our study shows that the stakeholder groups we eval- uated are receptive to bird-window collision management, but that does not necessarily trans- late into a willingness to pay for these measures, especially if doing so at private residences is the responsibility of homeowners. Past research evaluating WTP for conservation practices indicates that the public is often receptive to wildlife conservation and willing to pay for it [66– 69]. The public’s WTP for conservation practices can be heavily influenced by sense of place, or the value and meaning that individuals attach to a physical location [70, 71]. This suggests that informational materials that tie the issue of bird-window collisions to an individual’s loca- tion or experience may be a particularly effective way to increase WTP. For example, educa- tional materials could highlight the likely number of collisions that occur in areas where residents live and how collisions may be affecting locally important bird species. Another study found that while members of the public were willing to pay for bird conservation, they PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 15 / 20 PLOS ONEStakeholder perceptions of bird-window collisions believed the government should also play a role [68], a finding that lends additional support to grant, subsidy, and/or certification programs specifically geared toward reducing bird-window collisions. Although homeowners are a critical group to examine with regard to WTP to reduce bird-window collisions, other stakeholders such as business owners and agencies oper- ating in larger buildings are also important stakeholders to study. Birds face multiple human-related threats, including climate change, habitat loss, and other direct mortality sources (e.g., cat predation, other types of collisions) [3]. While it is important to investigate bird-window collisions specifically, understanding human perceptions of other threats is also necessary because this may lead to insights about which conservation actions are most and least likely to be supported and implemented by the public. Understanding percep- tions of different threats, as well as willingness to pay and/or willingness to change behaviors in ways that mitigate these threats, could also lead to more effective conservation strategies that optimize the tradeoff between addressing the most substantial threats and addressing the threats for which substantial management inroads are possible. Conclusions This study provides novel insight about how important stakeholder groups view and prioritize benefits and obstacles related to bird-window collision mitigation and prevention. Our research suggests that substantial advances can be made to reduce bird-window collisions because both homeowners and conservation practitioners had positive views, suggesting their receptivity toward and acceptability of collision management measures. However, because of the more neutral views and more conflicting responses within the homeowner group, our results also highlight the importance of targeting these stakeholders with education materials that provide information about bird-window collisions and policies and publicly available solutions that reduce them. Homeowners are a critical stakeholder group because a large pro- portion of collisions occur at residential buildings; having their support and participation in bird-window collision mitigation and prevention could help significantly reduce collisions. Future research needs related to human dimensions of bird-window collisions and other avian mortality sources include evaluating perceptions of other stakeholder groups (e.g., architects and policymakers), studying social-psychological barriers to reducing collisions, determining willingness to pay for collision mitigation and prevention, and clarifying relative perceptions about impacts and management of human-related threats other than bird-window collisions. Because bird-window collisions are a human-caused phenomenon, understanding human per- spectives and priorities about this issue will be crucial to addressing this threat and thus benefitting bird populations. Supporting information S1 File. SWOT Survey 1. Strengths, weaknesses, opportunities, and threats (SWOT) survey distributed to all respondents (i.e., Survey 1 described in main text) consisting of all pairwise comparisons between factors in each SWOT category using a scale of one to nine. For this sur- vey, all possible pairwise comparisons were made between factors within (but not between) each SWOT category (e.g., all strengths compared to each other, but strengths not compared to weaknesses, opportunities, and threats). Analysis of responses to this survey revealed top- ranked SWOT factors in each category, which were unique to each stakeholder group and used to generate comparisons in Survey 2. (PDF) PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 16 / 20 PLOS ONEStakeholder perceptions of bird-window collisions S2 File. SWOT Survey 2 for homeowners. Strengths, weaknesses, opportunities, and threats (SWOT) survey distributed to respondents in the homeowner stakeholder group (i.e., Survey 2 for homeowners described in main text) based on their responses to Survey 1. For this survey, all possible pairwise comparisons were made between the top-ranking factors from each SWOT category for homeowners (e.g., top homeowner strength compared to top weakness, opportunity, and threat). (PDF) S3 File. SWOT Survey 2 for conservation practitioners. Strengths, weaknesses, opportuni- ties, and threats (SWOT) survey distributed to respondents in the conservation practitioner stakeholder group (i.e., Survey 2 for conservation practitioners described in main text) based on their responses to Survey 1. For this survey, all possible pairwise comparisons made between the top-ranking factors from each SWOT category for conservation practitioners (e.g., top conservation practitioner strength compared to top weakness, opportunity, and threat). (PDF) S1 Dataset. SWOT and PCI data analysis. This file contains all response data generated from strengths, weaknesses, opportunities, and threats (SWOT) Surveys 1 and 2 (see main text and S1–S3 Files for details about these surveys) along with data and analysis for the potential for conflict index (PCI). (XLSX) Acknowledgments We thank Christine Sheppard, Daniel Klem, and Stephen Hager for providing preliminary rankings of candidate SWOT factors, Samantha Cady, Jared Elmore, and Timothy O’Connell for insightful feedback on methods and an earlier version of the manuscript, and all survey respondents for their participation. We also thank the handling editor and two anonymous reviewers for their constructive feedback and suggestions that greatly improved the manuscript. Author Contributions Conceptualization: Georgia J. Riggs, Omkar Joshi, Scott R. Loss. Data curation: Georgia J. Riggs. Formal analysis: Georgia J. Riggs. Funding acquisition: Scott R. Loss. Investigation: Georgia J. Riggs, Scott R. Loss. Methodology: Georgia J. Riggs, Omkar Joshi. Supervision: Omkar Joshi, Scott R. Loss. Validation: Omkar Joshi. Writing – original draft: Georgia J. Riggs. Writing – review & editing: Omkar Joshi, Scott R. Loss. PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 17 / 20 PLOS ONEStakeholder perceptions of bird-window collisions References 1. United Nations, Department of Economic and Social Affairs, Population Division. World Population Prospects 2019: Ten Key Findings; 2019. 2. Calvert A, Bishop C, Elliot R, Krebs E, Kydd T, Machtans C, et al. A synthesis of human-related avian mortality in Canada. Avian Conservation and Ecology. 2013; 8(2). 3. Loss SR, Will T, Marra PP. Direct mortality of birds from anthropogenic causes. Annual Review of Ecol- ogy, Evolution, and Systematics. 2015; 46. 4. Nyhus PJ. Human–wildlife conflict and coexistence. Annual Review of Environment and Resources. 2016; 41:143–71. 5. Loss SR, Will T, Loss SS, Marra PP. Bird–building collisions in the United States: estimates of annual mortality and species vulnerability. The Condor. 2014; 116(1):8–23. 6. Kim E-M, Choi C-Y, Kang C-W. Causes of injury and mortality of Fairy Pitta Pitta nympha on Jeju Island, Republic of Korea. Forktail. 2013; 29:145–8. 7. Machtans C, Wedeles C, Bayne E. A first estimate for Canada of the number of birds killed by colliding with building windows. Avian Conservation and Ecology. 2013;8(2). 8. Aymı´ R, Gonza´lez Y, Lo´ pez T, Gordo O. Bird-window collisions in a city on the Iberian Mediterranean coast during autumn migration. Catalan Journal of Ornithology. 2017; 33:17–28. 9. David J, Yong DL, Low BW, Owyong A, Alfred C. Anthropogenic Sources of Non-Migratory Avian Mor- talities In Singapore. The International Journal of Tropical Veterinary and Biomedical Research. 2017; 2 (2):17–27. 10. Santos LPS, de Abreu VF, de Vasconcelos MF. Bird mortality due to collisions in glass panes on an Important Bird Area of southeastern Brazil. Revista Brasileira de Ornitologia. 2017; 25(2):90–101. 11. Go´mez-Martı´nez MA, Klem D, Rojas-Soto O, Gonza´ lez-Garcı´a F, MacGregor-Fors I. Window strikes: bird collisions in a Neotropical green city. Urban Ecosystems. 2019; 22(4):699–708. 12. Klem D. Bird: window collisions. The Wilson Bulletin. 1989:606–20. 13. Winger BM, Weeks BC, Farnsworth A, Jones AW, Hennen M, Willard DE. Nocturnal flight-calling behaviour predicts vulnerability to artificial light in migratory birds. Proceedings of the Royal Society B. 2019; 286(1900):20190364. https://doi.org/10.1098/rspb.2019.0364 PMID: 30940055 14. Lao S, Robertson BA, Anderson AW, Blair RB, Eckles JW, Turner RJ, et al. The influence of artificial night at night and polarized light on bird-building collisions. Biological Conservation. 2020; 241:108358. 15. Kahle LQ, Flannery ME, Dumbacher JP. Bird-window collisions at a west-coast urban park museum: analyses of bird biology and window attributes from Golden Gate Park, San Francisco. PLoS One. 2016; 11(1):e0144600. https://doi.org/10.1371/journal.pone.0144600 PMID: 26731417 16. Nichols KS, Homayoun T, Eckles J, Blair RB. Bird-building collision risk: an assessment of the collision risk of birds with buildings by phylogeny and behavior using two citizen-science datasets. PloS One. 2018; 13(8):e0201558. https://doi.org/10.1371/journal.pone.0201558 PMID: 30092094 17. Loss S, Lao S, Eckles J, Anderson A, Blair R, Turner R. Factors influencing bird-building collisions in the downtown area of a major North American city. PloS One. 2019; 14(11):e0224164–e. https://doi. org/10.1371/journal.pone.0224164 PMID: 31693699 18. Klem D. Preventing bird–window collisions. The Wilson Journal of Ornithology. 2009; 121(2):314–22. 19. Hager SB, Cosentino BJ, McKay KJ, Monson C, Zuurdeeg W, Blevins B. Window area and develop- ment drive spatial variation in bird-window collisions in an urban landscape. PloS One. 2013; 8(1): e53371. https://doi.org/10.1371/journal.pone.0053371 PMID: 23326420 20. Riding CS, O’Connell TJ, Loss SR. Building fac¸ade-level correlates of bird–window collisions in a small urban area. The Condor. 2020; 122(1):duz065. 21. Hager SB, Cosentino BJ, Aguilar-Go´mez MA, Anderson ML, Bakermans M, Boves TJ, et al. Continent- wide analysis of how urbanization affects bird-window collision mortality in North America. Biological Conservation. 2017; 212:209–15. 22. Sheppard C, Fry M, Parr M, Law A. Bird-friendly building design: American Bird Conservancy (ABC); 2011. 23. Sheppard CD. Evaluating the relative effectiveness of patterns on glass as deterrents of bird collisions with glass. Global Ecology and Conservation. 2019; 20:e00795. 24. San Francisco Planning. Standards for Bird-Safe Buildings. 2019; Available from: https://sfplanning. org/standards-bird-safe-buildings. 25. Regents of the University of Minnesota. B3 –Building Benchmarks and Beyond. 2017; Available from: https://www.b3mn.org/. PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 18 / 20 PLOS ONEStakeholder perceptions of bird-window collisions 26. City of Toronto. Bird-friendly Guidelines. 2021; Available from: https://www.toronto.ca/city-government/ planning-development/official-plan-guidelines/design-guidelines/bird-friendly-guidelines/. 27. United States Congress. H.R.1986 –Bird Safe Buildings Act of 2021. 2021; Available from: https://www. congress.gov/bill/117th-congress/house-bill/1986?s=1&r=3. 28. Wald DM, Jacobson SK, Levy JK. Outdoor cats: Identifying differences between stakeholder beliefs, perceived impacts, risk and management. Biological Conservation. 2013; 167:414–24. 29. Gramza A, Teel T, VandeWoude S, Crooks K. Understanding public perceptions of risk regarding out- door pet cats to inform conservation action. Conservation Biology. 2016; 30(2):276–86. https://doi.org/ 10.1111/cobi.12631 PMID: 26379227 30. Warren EJ. A clear threat to conservation: using public policy to reduce bird collisions with windows in homes. 2013. 31. Menacho-Odio RM. Local perceptions, attitudes, beliefs, and practices toward bird-window collisions in Monteverde, Costa Rica. UNED Research Journal. 2018; 10. 32. Dwivedi P, Alavalapati JR. Stakeholders’ perceptions on forest biomass-based bioenergy development in the southern US. Energy Policy. 2009; 37(5):1999–2007. 33. Darshini D, Dwivedi P, Glenk K. Capturing stakeholders views on oil palm-based biofuel and biomass utilisation in Malaysia. Energy Policy. 2013; 62:1128–37. 34. Joshi O, Parajuli R, Kharel G, Poudyal NC, Taylor E. Stakeholder opinions on scientific forest manage- ment policy implementation in Nepal. PloS One. 2018; 13(9):e0203106. https://doi.org/10.1371/journal. pone.0203106 PMID: 30183725 35. Kurttila M, Pesonen M, Kangas J, Kajanus M. Utilizing the analytic hierarchy process (AHP) in SWOT analysis—a hybrid method and its application to a forest-certification case. Forest Policy and Econom- ics. 2000; 1(1):41–52. 36. Manfredo M, Vaske J, Teel T. The potential for conflict index: A graphic approach to practical signifi- cance of human dimensions research. Human Dimensions of Wildlife. 2003; 8(3):219–28. 37. Starr M, Joshi O, Will RE, Zou CB. Perceptions regarding active management of the Cross-timbers for- est resources of Oklahoma, Texas, and Kansas: a SWOT-ANP analysis. Land Use Policy. 2019; 81:523–30. 38. Arsić S, Nikolić D, Zˇ ivković Zˇ . Hybrid SWOT-ANP-FANP model for prioritization strategies of sustain- able development of ecotourism in National Park Djerdap, Serbia. Forest Policy and Economics. 2017; 80:11–26. 39. Saaty TL. Decision making—the analytic hierarchy and network processes (AHP/ANP). Journal of Sys- tems Science and Systems Engineering. 2004; 13(1):1–35. 40. Saaty TL. A scaling method for priorities in hierarchical structures. Journal of Mathematical Psychology. 1977; 15(3):234–81. 41. Klem D. Bird–window collisions: a critical animal welfare and conservation issue. Journal of Applied Ani- mal Welfare Science. 2015; 18(sup1):S11–S7. https://doi.org/10.1080/10888705.2015.1075832 PMID: 26440494 42. Snep RP, Kooijmans JL, Kwak RG, Foppen RP, Parsons H, Awasthy M, et al. Urban bird conservation: presenting stakeholder-specific arguments for the development of bird-friendly cities. Urban Ecosys- tems. 2016; 19(4):1535–50. 43. Kummer J, Bayne E. Bird feeders and their effects on bird-window collisions at residential houses. Avian Conservation and Ecology. 2015; 10(2). 44. Bernard HR. Research methods in anthropology: Qualitative and quantitative approaches: Rowman & Littlefield; 2017. 45. YouTube. San Bruno, CA. 2022; Available from: https://www.youtube.com/. 46. 47. 48. Facebook. Menlo Park, CA. 2022; Available from: https://www.facebook.com/. Twitter. San Francisco, CA. 2022; Avalable from: https://twitter.com/. Tumpach C., Dwivedi P., Izlar R., & Cook C. Understanding perceptions of stakeholder groups about Forestry Best Management Practices in Georgia. Journal of Environmental Management. 2018; 213:374–381. https://doi.org/10.1016/j.jenvman.2018.02.045 PMID: 29502022 49. Catron J, Stainback GA, Dwivedi P, Lhotka JM. Bioenergy development in Kentucky: a SWOT-ANP analysis. Forest Policy and Economics. 2013; 28:38–43. 50. Qualtrics. Provo, UT. 2022; Available from: https://www.qualtrics.com/. 51. Dillman DA, Smyth JD, Christian LM. Internet, phone, mail, and mixed-mode surveys: the tailored design method: John Wiley & Sons; 2014. PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 19 / 20 PLOS ONEStakeholder perceptions of bird-window collisions 52. Joshi O, Chapagain B, Kharel G, Poudyal NC, Murray BD, Mehmood SR. Benefits and challenges of online instruction in agriculture and natural resource education. Interactive Learning Environments. 2020:1–12. 53. Krejčı´ J, Stoklasa J. Aggregation in the analytic hierarchy process: why weighted geometric mean should be used instead of weighted arithmetic mean. Expert Systems with Applications. 2018; 114:97– 106. 54. Vaske JJ, Beaman J, Barreto H, Shelby LB. An extension and further validation of the potential for con- flict index. Leisure Sciences. 2010; 32(3):240–54. 55. Lute ML, Carter NH, Lo´pez-Bao JV, Linnell JD. Conservation professionals agree on challenges to coexisting with large carnivores but not on solutions. Biological Conservation. 2018; 218:223–32. 56. Sekercioglu C¸ H, Wenny DG, Whelan CJ. Why birds matter: avian ecological function and ecosystem services: University of Chicago Press; 2016. 57. Hedblom M, Knez I, Gunnarsson B. Bird diversity improves the well-being of city residents. Ecology and Conservation of Birds in Urban Environments: Springer; 2017. p. 287–306. 58. American Bird Conservancy. Glass collisions: preventing bird window strikes. 2022; Available from: https://abcbirds.org/glass-collisions/. 59. U.S. Fish and Wildlife Service. Buildings & glass. 2021; Available from: https://www.fws.gov/birds/bird- enthusiasts/threats-to-birds/collisions/buildings-and-glass.php. 60. National Audubon Society. Reducing collisions with glass. 2021; Available from: https://www.audubon. org/news/reducing-collisions-glass. 61. Fatal Light Awareness Program Canada. FLAP Canada. 2019; Available from: https://flap.org/. 62. Darko A, Chan APC, Ameyaw EE, Owusu EK, Pa¨ rn E, Edwards DJ. Reivew of application of analytic hierarchy process (AHP) in construction. International Journal of Construction Management. 2019; 19 (5):436–452. 63. Margles SW, Masozera M, Rugyerinyange L, Kaplin BA. Participatory planning: using SWOT-AHP analysis in buffer zone management planning. Journal of Sustainable Forestry. 2010; 29(6–8):613– 637. 64. Kollmuss A, Agyeman J. Mind the gap: why do people act environmentally and what are the barriers to pro-environmental behavior?. Environmental Education Research. 2002; 8(3):239–60. 65. Hoffman AJ, Henn R. Overcoming the social and psychological barriers to green building. Organization & Environment. 2008; 21(4):390–419. 66. 67. Tsi E, Ajaga N, Wiegleb G, Mu¨ hlenberg M. The willingness to pay (WTP) for the conservation of wild animals: Case of the Derby Eland (Taurotragus derbianus gigas) and the African wild dog (Lycaon pic- tus) in North Cameroon. African Journal of Environmental Science and Technology. 2008; 2(3):51–8. Zander KK, Ainsworth GB, Meyerhoff J, Garnett ST. Threatened bird valuation in Australia. PloS One. 2014; 9(6):e100411. https://doi.org/10.1371/journal.pone.0100411 PMID: 24955957 68. Steven R, Morrison C, Castley JG. Exploring attitudes and understanding of global conservation prac- tice among birders and avitourists for enhanced conservation of birds. Bird Conservation International. 2017; 27(2):224–36. 69. Getzner M, Meyerhoff J, Schla¨pfer F. Willingness to pay for nature conservation policies in state-owned forests: An Austrian case study. Forests. 2018; 9(9):537. 70. Kudryavtsev A, Stedman RC, Krasny ME. Sense of place in environmental education. Environmental Education Research. 2012; 18(2):229–50. 71. Nielsen-Pincus M, Sussman P, Bennett DE, Gosnell H, Parker R. The influence of place on the willing- ness to pay for ecosystem services. Society & Natural Resources. 2017; 30(12):1423–41. PLOS ONE \| https://doi.org/10.1371/journal.pone.0263447 February 10, 2022 20 / 20 PLOS ONE
10.1371_journal.pone.0266237	RESEARCH ARTICLE Long-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test Ehlimana Osmanović OmerdićID Kulenović1, orđe Medarević4, Branka Ivković5, Dragana Vasiljević4 1, Larisa Alagić-Dzˇambić2, Marko Krstić3, Maja Pasˇ ić- 1 Development and Registration Department, Bosnalijek d.d., Sarajevo, Bosnia and Herzegovina, 2 Quality Assurance and Quality Control Department, Bosnalijek d.d., Sarajevo, Bosnia and Herzegovina, 3 Department of Analytical Chemistry, University of Belgrade—Faculty of Pharmacy, Belgrade, Serbia, 4 Department of Pharmaceutical Technology and Cosmetology, University of Belgrade—Faculty of Pharmacy, Belgrade, Serbia, 5 Department of Pharmaceutical Chemistry, University of Belgrade—Faculty of Pharmacy, Belgrade, Serbia ehlimana.osmanovic.omerdic@bosnalijek.ba Abstract a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Osmanović Omerdić E, Alagić-Dzˇambić L, Krstić M, Pasˇić-Kulenović M, Medarević , Ivković B, et al. (2022) Long-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test. PLoS ONE 17(4): e0266237. https://doi.org/ 10.1371/journal.pone.0266237 Editor: Mª A´ngeles Peña, University of Alcala´, SPAIN Received: January 11, 2022 Accepted: March 16, 2022 Published: April 4, 2022 Copyright: © 2022 Osmanović Omerdić et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Funding: This research was funded by the Ministry of Education, Science and Technological Development, Republic of Serbia through Grant Agreement with University of Belgrade - Faculty of Pharmacy No: 451-03-9/2021-14/200161. The funders had no role in study design, data collection Formulation of solid dispersions (SDs), in which the drug substance is dissolved or dis- persed inside a polymer matrix, is one of the modern approaches to increase the solubility and dissolution rate of poorly soluble active pharmaceutical ingredients (APIs), such as clo- pidogrel. In the form of a free base, clopidogrel is unstable under increased both high mois- ture and temperature, so it is most often used as its salt form, clopidogrel hydrogen sulfate (CHS).The aim of this study was the formulation, characterization, and long-term stability investigation of CHS solid dispersions, prepared with four different hydrophilic polymers (poloxamer 407, macrogol 6000, povidone, copovidone) in five API/polymer ratios (1:1, 1:2, 1:3, 1:5, 1:9). SDs were prepared by the solvent evaporation method, employing ethanol (96% v/v) as a solvent. Initial results of the in vitro dissolution test showed an increase in the amount of dissolved CHS from all prepared SD samples compared to pure CHS, corre- sponding physical mixtures (PMs), and commercial tablets. SDs, prepared with poloxamer 407, macrogol 6000, and copovidone, at CHS/polymer ratios 1:5 and 1:9, notably increased the amount of dissolved CHS (> 80%, after 60 min), thus they were selected for further char- acterization. To assess the SDs long-term stability, in vitro dissolution studies, clopidogrel content determination, differential scanning calorimetry (DSC), and Fourier transform infra- red spectroscopy (FT-IR) were performed initially and after 12 months of long-term stability studies under controlled conditions (25˚C, 60% RH) meeting the ICH guideline Q1A (R2) requirements. The clopidogrel content in the selected samples was very similar at the begin- ning (96.13% to 99.93%) and at the end (95.98% to 99.86%) of the conducted test. DSC curves and FT-IR spectra of all SD samples after 12 months of stability study, showed the absence of CHS crystallization, which is an indication of good stability. However, the in vitro dissolution test showed a considerable reduction in CHS released from SDs with macrogol 6000. The amount of dissolved CHS from SDs with macrogol 6000 was initially 94.02% and 92.01%, and after 12 months of stability study, only 65.13% and 49.62%. In contrast, the amount of dissolved CHS from SDs prepared with poloxamer 407 and copovidone was very PLOS ONE \| https://doi.org/10.1371/journal.pone.0266237 April 4, 2022 1 / 14 PLOS ONELong-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. similar after 12 months of the stability study compared to the initial values. Results obtained indicated the great importance of the in vitro dissolution test in determining the long-term stability and quality of SDs. 1. Introduction A number of formulation strategies have focused on increasing solubility and dissolution rate with the aim to improve the oral bioavailability of poorly soluble drugs like clopidogrel. One of the most promising is solid dispersions (SDs) formulation. SDs can be defined as a dispersion of one or more active pharmaceutical ingredients (APIs) in an inert carrier or matrix in the solid-state prepared by melting method, solvent evaporation method, or combination of these two methods [1]. SDs can be prepared using cost-effective manufacturing technologies with well-known, widely used, safe excipients [2]. Mechanisms of enhancement solubility and dis- solution rate of API in SDs include particle size reduction, formation of solid solutions of drug in an inert polymer matrix, enhanced porosity of SDs, improved wettability, solubilization of API, or conversion of drug from crystalline into the amorphous state [3, 4]. It was recently reported that between 2007 and 2017, the US Food and Drug Administration approved 19 commercial SDs products, mainly in tablets and capsules dosage forms [5]. The increase in the number of approved and commercially available products gives additional importance to more substantial research of SDs. Due to its complexity, there are numerous challenges regard- ing of the physicochemical properties of used API and polymers, and therefore influencing on the stability of SDs and formulation of the final dosage form. It has been reported that stability, which is one of the main issue of SDs, can be successfully resolved by the use of appropriate polymer and preparation method [2, 6, 7]. Clopidogrel, methyl (+) -(S)-α-(2-chlorophenyl)-6,7-dihydrothieno[3,2-c] pyridine-5(4H)- acetate, is an inhibitor of platelet activation and aggregation through the irreversible binding of its active metabolite to the P2Y12 class of ADP receptors on platelets. The inhibiting activity makes clopidogrel an effective API for reducing the induce of ischemic strokes, myocardial infarction, and vascular disease [8, 9]. In a free base form, clopidogrel is an oily substance with high viscosity, unstable under increased moisture and temperature, so salt forms are com- monly used. The most commonly used salt is clopidogrel hydrogen sulfate (CHS) (syn. clopi- dogrel bisulfate) [10]. CHS belongs to BCS class II APIs with low solubility that limits its oral bioavailability and therapeutic effectiveness. CHS has been found to exist in six polymorphic forms and an amorphous form. Still, only polymorphic forms I and II are used in the formula- tion of dosage forms due to better stability and bioavailability compared to other forms [10– 12]. Form II has a melting point range between 176–178˚C, is thermodynamically more stable, and has better compactibility than form I, and therefore is more acceptable for the manufacturing process [12–14]. The aim of this study was the formulation, characterization, and long-term stability evalua- tion of CHS solid dispersions, prepared with four different hydrophilic polymers (poloxamer 407, macrogol 6000, povidone, copovidone) in five API/polymer ratios (1:1, 1:2, 1:3, 1:5, 1:9). 2. Materials and methods 2.1. Materials In this study, clopidogrel hydrogensulfate (Ph. Eur. 10.0) was used as a model API. Four differ- ent hydrophilic polymers: poloxamer 407 (Kolliphor1 P407, BASF, Germany), macrogol 6000 PLOS ONE \| https://doi.org/10.1371/journal.pone.0266237 April 4, 2022 2 / 14 PLOS ONELong-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test (Polyglykol 6000 S, Clariant Export AG, Switzerland), povidone (Kollidon1 30, BASF, Ger- many), and copovidone (Kollidon1 VA 64, BASF, Germany), were used as carriers. Ethanol 96% (v/v) was used as a solvent for SDs preparation by solvent evaporation method. All excipi- ents were of pharmaceutical grade. Other reagents and chemicals used for the characterization of SDs were of analytical grade. API, excipients, and other reagents and chemicals were kindly provided as a gift from Bosnali- jek d.d. (Sarajevo, Bosnia and Herzegovina). 2.2. Methods 2.2.1. Preparation of solid dispersions and physical mixtures. CHS solid dispersions were prepared by the standard solvent evaporation method, employing ethanol (96% v/v) as a solvent. The composition of prepared SDs is presented in Table 1. The required quantities of CHS and polymer were dissolved in the corresponding amount of solvent to get a clear solu- tion. The solvent was then removed by evaporation under reduced pressure (vacuum) on a rotary vacuum evaporator (Rotavapor1 R-205, Bu¨chi, Switzerland) at a temperature of 55˚C with constant mixing at 95 rpm. The resulting solid mass was gentle ground in a mortar with a pestle to obtain powder form and shifted through a sieve 500 μm (Ph.Eur.). Prepared SDs were filled and stored in a sealed amber glass bottle with a polyethylene screw. The physical mixtures (PMs) of CHS and polymers (Table 1) were prepared by mixing the API and polymer using mortar and pestle. Obtained mixtures were shifted through sieve 500 (Ph. Eur), filled, and stored in a sealed amber glass bottle with a polyethylene screw. For in vitro dissolution studies, SDs and PMs were filled into hard gelatin capsules which contained a therapeutic dose of clopidogrel (75 mg) and stored in a sealed amber glass bottle with a polyethylene screw until analyses were performed. Table 1. Composition of CHS solid dispersions and corresponding physical mixtures. No. of formulation Formulation 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 P1 P2 P3 P5 P9 M1 M2 M3 M5 M9 K1 K2 K3 K5 K9 C1 C2 C3 C5 C9 https://doi.org/10.1371/journal.pone.0266237.t001 Polymer Poloxamer 407 Poloxamer 407 Poloxamer 407 Poloxamer 407 Poloxamer 407 Macrogol 6000 Macrogol 6000 Macrogol 6000 Macrogol 6000 Macrogol 6000 Povidone Povidone Povidone Povidone Povidone Copovidone Copovidone Copovidone Copovidone Copovidone CHS/ polymer ratio Physical mixure 1:1 1:2 1:3 1:5 1:9 1:1 1:2 1:3 1:5 1:9 1:1 1:2 1:3 1:5 1:9 1:1 1:2 1:3 1:5 1:9 PPM1 PPM2 PPM3 PPM5 PPM9 MPM1 MPM2 MPM3 MPM5 MPM9 KPM1 KPM2 KPM3 KPM5 KPM9 CPM1 CPM2 CPM3 CPM5 CPM9 PLOS ONE \| https://doi.org/10.1371/journal.pone.0266237 April 4, 2022 3 / 14 PLOS ONELong-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test 2.2.2. In vitro dissolution studies. In vitro dissolution studies were performed using USP Apparatus 1 (basket), rotation speed of 75 rpm, in 900 mL of phosphate buffer (pH 6.8) at 37˚C ± 0.5˚C. Tests were conducted with a Vankel VK 7010 dissolution bath and a Varian fraction sample collection module (Agilent, USA). Aliquots of 1.5 mL were removed at predetermined time points (15, 30, 45, and 60 min) and replaced with an equal volume of fresh dissolution medium. Samples were filtered through a 35 μm filter (Micron Full Flow Filter, Agilent, USA) and collected directly into an HPLC vial. The CHS concentration was determined by the RP-HPLC method, as described in the previ- ously published paper [15]. In vitro dissolution studies were performed in triplicate, and the data are expressed as the mean value ± standard deviation. The obtained dissolution profiles were fitted into different empirical models, and the corresponding correlation coefficients (R2) were calculated. 2.2.3. Determination of clopidogrel content. Accurately weighed quantity of selected SDs, equivalent to 75 mg of clopidogrel, was diluted with the mobile phase to a volume of 100 ml. 1 ml of this filtered solution was further diluted with 10 ml mobile phase. The content of clopidogrel was determined by the mentioned RP-HPLC method [15]. 2.2.4. Differential scanning calorimetry (DSC). DSC analyses of API, polymers, and selected SDs were performed using a DSC1 instrument (Mettler Toledo, Giessen, Germany). A precisely measured amount of sample (1–5 mg) was placed into pierced 40 μl aluminium pans, and subjected to heating from 20 to 250˚C with a heating rate of 10˚C/min using nitrogen as purge gas at a flow rate of 50 ml/min. 2.2.5. Fourier transform infrared spectroscopy (FT-IR). FT-IR spectra of API, poly- mers, and selected SDs were acquired using Nicolet iS10 (Thermo Scientific, Waltham, USA) FT-IR spectrometer, equipped with a single reflection ATR system (Smart iTR, Thermo Scien- tific, Waltham, USA) with diamond plate and Zn-Se lens. The samples were directly kept on the sample holder, and spectra were recorded as an average of 16 scans in the frequency range from 4000 to 650 cm-1, with a resolution of 4 cm-1. Sixteen recordings were made for each sample and then averaged to obtain a spectrum. 2.2.6. Long-term stability studies. Selected SDs, filled into sealed amber glass bottles with a polyethylene screw, were stored for 12 months in chambers for stability testing (Pharma 2000, Weiss Technik, Germany) under the conditions of 25˚C ± 2˚C and 60% RH ± 5% RH, according to ICH guideline Q1A (R2) [16]. To assess the long-term stability of SDs, in vitro dissolution studies, clopidogrel content, DSC, and FT-IR were performed initially and after 12 months of stability studies under mentioned conditions. 3. Results and discussions In this study, 20 samples of clopidogrel hydrogensulfate SDs were prepared, with different hydrophilic polymers (poloxamer 407, macrogol 6000, povidone, and copovidone) and differ- ent API/polymer ratios (1:1, 1:2, 1:3, 1:5, and 1:9). Initially, an in vitro dissolution test was per- formed. This test is an easy, fast, inexpensive method, which during research and development may show differences in the dissolution rate of API from the tested samples. Only samples from which at least 80% CHS was dissolved after 60 min were selected for further characterization. 3.1. In vitro dissolution studies Firstly, an in vitro dissolution test was performed with 20 prepared SDs, corresponding PMs, pure CHS, and commercial immediate release film coated tablet. PLOS ONE \| https://doi.org/10.1371/journal.pone.0266237 April 4, 2022 4 / 14 PLOS ONELong-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test Fig 1. Comparative in vitro dissolution profiles of pure CHS, SDs and commercial tablet. a) SDs with poloxamer 407 b) SDs with macrogol 6000 c) SDs with povidone d) SDs with copovidone. https://doi.org/10.1371/journal.pone.0266237.g001 Dissolution profiles of pure CHS, SDs, and commercial tablet are presented in (Fig 1A– 1D). After 60 min, the amount of dissolved CHS was between 30.74% and 96.38% for SDs (Table 2), 32.20% for pure CHS, and 37.62% for commercial tablet. Except in the case of sam- ple P1, the amount of dissolved CHS from SDs was notably higher than the corresponding PMs (Table 2), pure CHS, and commercial tablet. Sample P1 contained the least amount of polymer (CHS/poloxamer 407 ratio 1:1), so it can be assumed that in this case, the amount of poloxamer 407 was not sufficient to increase the solubility and dissolution rate of CHS. In other samples, it could be assumed that SDs increased the amount of dissolved CHS due to the transition of the crystalline form of API to amorphous, reducing particle size, enhanced wetta- bility, and polymers’ ability to solubilize API, as mentioned earlier in literature [17–20]. In the case of SDs prepared with poloxamer 407, povidone, and copovidone, with the increase in the content of the polymer in the SDs, there was an increase in the amount of dis- solved CHS. However, in macrogol SDs, the larger amount of CHS (94.03%) was dissolved from the sample M5, which had a CHS/polymer ratio of 1:5. The dissolved CHS amount from PLOS ONE \| https://doi.org/10.1371/journal.pone.0266237 April 4, 2022 5 / 14 PLOS ONELong-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test Table 2. The amount of dissolved CHS after 60 min from SDs and PMs. Solid dispersion Dissolved CHS, % (S.D.) Physical mixture Dissolved CHS, % (S.D.) P1 P2 P3 P5 P9 M1 M2 M3 M5 M9 K1 K2 K3 K5 K9 C1 C2 C3 C5 C9 30.74 (0.68) 47.38 (1.00) 60.34 (0.98) 89.21 (1.86) 96.38 (1.68) 43.72 (1.50) 57.22 (1.01) 63.01 (1.51) 94.03 (0.85) 92.01 (1.60) 40.98 (0.32) 54.75 (0.71) 57.34 (2.34) 70.20 (0.55) 77.22 (2.58) 54.05 (0.61) 59.28 (1.06) 62.56 (0.35) 81.01 (0.93) 90.44 (1.33) PPM1 PPM2 PPM3 PPM5 PPM9 MPM1 MPM2 MPM3 MPM5 MPM9 KPM1 KPM2 KPM3 KPM5 KPM9 CPM1 CPM2 CPM3 CPM5 CPM9 29.60 (0.58) 32.60 (2.04) 34.94 (0.51) 42.71 (1.04) 42.71 (1.21) 27.34 (0.87) 30.83 (0.53) 36.06 (0.96) 45.20 (1.40) 50.45 (2.14) 22.31 (0.93) 23.07 (2.13) 25.30 (1.98) 27.04 (0.10) 29.91 (3.06) 30.93 (1.28) 31.86 (0.54) 33.22 (1.10) 28.77 (3.41) 44.35 (0.53) S.D. standard deviation. https://doi.org/10.1371/journal.pone.0266237.t002 the sample with a CHS/macrogol 6000 ratio of 1:9 (M9) was 92.01%. This could be explained by the assumption that in the sample M9 around the CHS particles, a viscous polymer layer was formed, which prevent further dissolution improvement [21]. Based on the obtained results, it could be concluded that SDs, prepared with three different polymers (poloxamer 407, macrogol 6000, and copovidone), at CHS/polymer ratios 1:5 and 1:9, notably increase the amount of dissolved CHS. SDs in which the amount of dissolved CHS after 60 min was greater than 80% (P5, P9, M5, M9, C5, and C9) were selected for further char- acterization. In the case of SDs prepared with povidone (K1 –K9), the amount of dissolved CHS after 60 min was from 40.98% to 77.22%. These samples did not meet the stated require- ment and were excluded from further study. 3.2. Determination of clopidogrel content Initially, clopidogrel content in the selected SDs was in the range of 96.13% to 99.93% (Table 3). These results were in line with the requirements of the guideline “Specifications and Control Tests on the Finished Product-3AQ11a”, where maximum acceptable deviations in the APIs content should not exceed ± 5% [22]. 3.3. Differential scanning calorimetry (DSC) DSC has been used to measure the melting point of API and polymer, API-polymer miscibil- ity, polymorphic form transformation, a crystallization tendency of an API, and molecular mobility [23]. In the SDs, if the drug is in amorphous form, no endothermic peak will be pres- ent on the DSC curve. Figs 2–4 illustrate the DSC curves of pure CHS, polymers, and selected SDs (initially and after 12 months of stability studies). A small number of thermal events were PLOS ONE \| https://doi.org/10.1371/journal.pone.0266237 April 4, 2022 6 / 14 PLOS ONELong-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test Table 3. Clopidogrel content in selected SDs (initially, and after 12 months of stability studies). SDs P5 P9 M5 M9 C5 C9 Clopidogrel content (%) Clopidogrel content (%) 0 months 12 months 99.93 99.91 98.99 96.13 96.77 96.35 99.86 99.38 98.51 97.41 95.82 95.98 https://doi.org/10.1371/journal.pone.0266237.t003 observed in all the DSC curves. DSC curve of pure CHS showed a single endothermic peak at a temperature of 178˚C, corresponding to the melting point of the CHS. This indicated that CHS was in polymorphic form II, which is in accordance with the literary data of melting point range between 176–178˚C [10, 12, 13]. Polymorphic form I of CHS has a melting point at slightly higher temperatures, 198–200˚C [10, 12, 13]. Polymorphic form II of CHS is the most stable form at ambient conditions, and it is also used more often in commercial dosage forms (tablets and film-coated tablets) [14]. DSC curves of polymers showed an endothermic peak for poloxamer 407 at 57˚C (Fig 5) and macrogol 6000 at 63˚C (Fig 6), corresponding to the melting points of these polymers [11, 21, 24]. Broad endotherms of copovidone were observed at 63˚C and 100–107˚C, respectively (Fig 7) [19, 25], which can result from combina- tion of several thermal events, such as glass transition, evaporation of absorbed water and poly- mer relaxation. Characteristic peaks for the melting point of crystalline CHS in all SDs curves wholly disappeared, which was indicated that most of the API was existed in the amorphous state. Peaks that correspond to poloxamer 407 and macrogol 6000 in SDs curves were remained in close positions as in pure polymers. Copovidone endotherms were further stretched or almost completely disappeared in SDs curves. Small endothermic peaks seen on Fig 2. DSC curves of CHS, poloxamer 407, and SDs with poloxamer 407 (P5 and P9) initially, and after 12 months of stability studies. https://doi.org/10.1371/journal.pone.0266237.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0266237 April 4, 2022 7 / 14 PLOS ONELong-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test Fig 3. DSC curves of CHS, macrogol 6000, and SDs with macrogol 6000 (M5 and M9) initially, and after 12 months of stability studies. https://doi.org/10.1371/journal.pone.0266237.g003 DSC curves of API and SDs after the melting temperature of CHS were most likely caused by the thermally induced degradation of CHS [12]. 3.4. Fourier transform infrared spectroscopy (FT-IR) FT-IR spectroscopy was used to examine interactions between the CHS and the polymers in solid dispersions. Intermolecular interactions between API and polymer are essential in Fig 4. DSC curves of CHS, copovidone, and SDs with copovidone (C5 and C9) initially, and after 12 months of stability studies. https://doi.org/10.1371/journal.pone.0266237.g004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0266237 April 4, 2022 8 / 14 PLOS ONELong-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test Fig 5. FT-IR spectra of CHS, poloxamer 407, and SDs with poloxamer 407 (P5 and P9) initially, and after 12 months of stability studies. https://doi.org/10.1371/journal.pone.0266237.g005 stabilizing API amorphous form in the SDs matrix. These interactions also play the most cru- cial role in preventing the crystallization of API [26, 27]. The FT-IR spectra of pure CHS, polymers, and selected SDs (initially and after 12 months of stability studies) are presented in Figs 5–7. The FT-IR spectrum of CHS shows characteristic Fig 6. FT-IR spectra of CHS, macrogol 6000, and SDs with macrogol 6000 (M5 and M9) initially, and after 12 months of stability studies. https://doi.org/10.1371/journal.pone.0266237.g006 PLOS ONE \| https://doi.org/10.1371/journal.pone.0266237 April 4, 2022 9 / 14 PLOS ONELong-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test Fig 7. FT-IR spectra of CHS, copovidone, and SDs with copovidone (C5 and C9) initially, and after 12 months of stability studies. https://doi.org/10.1371/journal.pone.0266237.g007 bands at 3121 cm–1 (aromatic CH vibrations), due to the presence of chlorophenyl ring, 1750 cm-1 with higher intensity, due to the presence of the ester functional group (C = O stretching vibrations), 1437 cm-1 (N-H deformations), 1186 cm-1 (in-plane motion of the C-H) and 1150 cm-1 (rocking of CH3) and 1028 cm-1 (pyridine-methylene rock) which is in line with the spec- tra previously reported in the literature [12, 13, 28]. The SDs spectra looked very similar to the spectra of individual components, CHS and polymers, with minimal differences. For SDs with poloxamer 407 (P5 and P9) and macrogol 6000 (M5 and M9) the differences between observed spectra for most characteristic band is about 1–5 cm-1. Shifting of the peak from 1150 to 1147 cm-1 were observed on these spectra. These subtle changes probably resulted in hydrogen bonds between CHS and polymers. For SDs with copovidone (C5 and C9) absorption band at 1750 cm-1 were observed at 1730–1731 cm-1 and CHS band positioned at 1150 cm-1 shifted to 1166 cm-1. For C5 solid dispersion CHS band at 1186 and 1028 cm-1 shifted to 1181 and 1034 cm-1, respectively, and for C9 same bands shifted to 1184 and 1042 cm-1, respectively. These observed changes in FTIR spectra of SDs could indicate the existence of intermolecular interactions between CHS and polymers. Physical interactions between molecules can be held responsible for reducing the level of CHS crystallinity and improving CHS dissolution [24, 29]. However, no additional peaks were observed in the FT-IR spectra of SDs, indicating the absence of any chemical interaction. 3.5. Long-term stability studies Physical stability is the main issue for the development SDs in a final dosage form. Long-term stability of SDs is necessary to ensure an acceptable shelf-life of the final dosage form. The PLOS ONE \| https://doi.org/10.1371/journal.pone.0266237 April 4, 2022 10 / 14 PLOS ONELong-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test optimal type of polymer and API loading, good miscibility between API and polymer, and enhancing drug-polymer interaction have proven to be a good strategy for improving the sta- bility of SDs [30]. After 12 months of stability studies, selected SDs (P5, P9, C5, C9, M5, and M9) were analyzed using in vitro dissolution tests, clopidogrel content, DSC, and FT-IR, and were compared to initial results. The amount of dissolved CHS from SDs with poloxamer 407 and copovidone after 12 months of stability studies were very similar to initial results (Figs 1A–1C, and 8). In the case of SDs with macrogol 6000 (M5 and M9), after 12 months of stabil- ity studies, a decrease in the amount of dissolved CHS was observed (65.13% and 49.62%, respectively) compared to the initial values (94.02% and 92.01%, respectively) (Figs 1C and 8). This could be explained by changing the physical properties of the polymer during storage. Damian et al. [31] showed that in SD with an antiviral drug UC-781 and hydrophilic polymer macrogol 6000, an increase in enthalpy of fusion was observed during stability studies, espe- cially expressed at a temperature of 25˚C. The absence of crystallinity of the UC-781 in the SDs with macrogol 6000 initially and after 12 months of stability studies was confirmed by DSC and X-ray powder diffraction. It was assumed that reason for decreasing in UC-781 dissolution was the reorganization in the crystalline structure of the polymer that occurred during storage, which led to a decrease in the dissolution rate of API from the SDs. A similar decrease in the dissolution of API from SDs with macrogol 6000 after stability studies were also reported by Ford and Rubinstein [32, 33]. When compared SDs with macrogol 6000, for M9 was noticed a higher decrease in the amount of dissolved API, probably because of a higher amount of poly- mer in the formulation. To describe CHS release mechanism from SDs (initially and after 12 months of stability studies), mathematical models such as zero-order, first-order, Korsmeyer-Peppas, Higuchi, and Hixson-Crowell models were used. Initially, all selected SDs had the greatest R2 in the Korsmeyer-Peppas model (Table 4). According to this model, the drug was transported via Fickian diffusion when the parameter n was below or equal to 0.45. In contrast, the n value between 0.45 and 0.89 indicated anomalous transport (non-Fickian diffusion). Case I transport is defined as a zero-order release model when n value was equal to 0.89, while n values were above 0.89, suggesting a super case II transport [34, 35]. In the case of selected SDs, the n value was more than 0.89, indicating that CHS was released by matrix polymer diffusion and relaxa- tion [36]. The Korsmeyer-Peppas remained the best kinetic model for SDs with poloxamer 407 and copovidone after 12 months of stability studies. Small changes in the release kinetics have occurred for SDs with macrogol 6000, where the R2 values were almost equal for the Higuchi and Korsmeyer-Peppas model (Table 4). The Higuchi model describes the release mechanism of APIs based on Fick’s diffusion and is commonly used to describe release kinet- ics for the modified-release dosage form [37]. Fig 8. Comparative dissolution profiles of selected SDs after 12 months of stability studies. https://doi.org/10.1371/journal.pone.0266237.g008 PLOS ONE \| https://doi.org/10.1371/journal.pone.0266237 April 4, 2022 11 / 14 PLOS ONELong-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test Table 4. Release kinetic of dissolution data (initially and after 12 months of stability studies). SD P5 P9 M5 M9 C5 C9 0-order R2 I order R2 Korsmeyer-Peppas R2 N Higuchi R2 Hixon-Crowell R2 0m 0.6430 0.5450 0.6402 0.6573 0.6464 0.6338 12m 0.6419 0.7168 0.8102 0.7097 0.6602 0.6291 0m 0.8287 0.7126 0.8805 0.8869 0.7858 0.8342 12m 0.8240 0.9714 0.9094 0.7720 0.7858 0.8111 0m 0.9576 0.8433 0.9673 0.9970 0.9287 0.9863 12m 0.9628 0.9895 0.9713 0.9163 0.9840 0.9594 0m 1.8725 1.9614 1.8928 1.8761 1.8268 1.8867 12m 1.7098 1.6071 1.3559 1.2724 1.6257 1.7410 0m 0.8840 0.8113 0.8794 0.8915 0.8851 0.8751 12m 0.8830 0.9300 0.9714 0.9193 0.8952 0.8737 0m 0.7609 0.6397 0.7956 0.8057 0.7348 0.7619 12m 0.7575 0.8988 0.8787 0.7507 0.7434 0.7447 https://doi.org/10.1371/journal.pone.0266237.t004 Clopidogrel content in the chosen SDs was in range of 95.98% to 99.86% after 12 months of stability studies (Table 3). Presented results for clopidogrel content were very similar com- pared to initial results and followed ICH Q1A (R2) guideline requirements [16]. DSC curves for all selected SDs after 12 months of stability studies (Figs 2–4) showed mini- mal changes in appearance and position of endothermic peaks, which confirmed that drug physical state was not changed during stability studies. The FT-IR spectra of selected SDs after 12 months of stability studies (Figs 5–7) showed similar spectra with minimal or no differences in characteristic band positions and the absence of CHS crystallization in SDs samples which indicated that intermolecular interactions could stabilize the amorphous form of CHS in solid dispersions. 4. Conclusions Initial results of the in vitro dissolution test showed that SDs increased the amount of dissolved CHS from all prepared samples compared to pure CHS, corresponding physical mixtures, and commercial tablet. The results obtained during the long-term stability study of selected clopi- dogrel hydrogen sulfate SDs (clopidogrel content, DSC, and FT-IR) indicated good stability of SDs prepared with poloxamer 407, macrogol 6000, and copovidone, at CHS/polymer ratios 1:5 and 1:9. However, the in vitro dissolution test showed a considerable reduction in CHS released from SDs with macrogol 6000. Results obtained indicated the great importance of the in vitro dissolution test in determining the long-term stability and quality of SDs. Further stud- ies are needed to explain better the changes in SDs with macrogol 6000, which occur over time, leading to a considerable reduction in the CHS released amount from these samples. Acknowledgments The authors gratefully acknowledge Bosnalijek d.d. for their support. Author Contributions Conceptualization: Ehlimana Osmanović Omerdić, Dragana Vasiljević. Data curation: Ehlimana Osmanović Omerdić, Marko Krstić. Formal analysis: Larisa Alagić-Dzˇambić, Marko Krstić, Ðorđe Medarević, Branka Ivković. Funding acquisition: Maja Pasˇić-Kulenović. Investigation: Ehlimana Osmanović Omerdić, Larisa Alagić-Dzˇambić. Methodology: Ehlimana Osmanović Omerdić, Larisa Alagić-Dzˇambić, Ðorđe Medarević, Branka Ivković, Dragana Vasiljević. PLOS ONE \| https://doi.org/10.1371/journal.pone.0266237 April 4, 2022 12 / 14 PLOS ONELong-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test Project administration: Maja Pasˇić-Kulenović. Resources: Maja Pasˇić-Kulenović, Dragana Vasiljević. Supervision: Dragana Vasiljević. Visualization: Ehlimana Osmanović Omerdić. Writing – original draft: Ehlimana Osmanović Omerdić. Writing – review & editing: Ehlimana Osmanović Omerdić, Marko Krstić, Ðorđe Medarević, Branka Ivković, Dragana Vasiljević. References 1. Paudwal G, Rawat N, Gupta R, Baldi A, Singh G, Gupta PN. Recent advances in solid dispersion tech- nology for efficient delivery of poorly water-soluble drugs. Curr Pharm Des. 2019; 25: 1524–1535. https://doi.org/10.2174/1381612825666190618121553 PMID: 31258070 2. Van Doung T, Van den Mooter G. The role of the carrier in the formulation of pharmaceutical solid dis- persions. Part II: amorphous carriers. Expert Opin Drug Deliv. 2016; 13: 1681–1694. https://doi.org/10. 1080/17425247.2016.1198769 PMID: 27267583 3. Van den Mooter G. The use of amorphous solid dispersion: a formulation strategy to overcome poor sol- ubility and dissolution rate. Drug Discov Today Technol. 2012; 9: 79–85. https://doi.org/10.1016/j.ddtec. 2011.10.002 PMID: 24064267 4. Vasconcelos TF, Marques S, das Neves J, Sarmento B. Amorphous solid dispersions: Rational selec- tion of a manufacturing process. Adv Drug Deliv Rev. 2016; 100: 85–101. https://doi.org/10.1016/j.addr. 2016.01.012 PMID: 26826438 5. 6. 7. 8. Jermain SV, Brough C, Williams RO III. Amorphous solid dispersions and nanocrystal technologies for poorly water-soluble drug delivery–An update. Int J Pharm. 2018; 535: 379–392. https://doi.org/10. 1016/j.ijpharm.2017.10.051 PMID: 29128423 Li B, Harich K, Wegiel L, Taylor SL, Edgar JK. Stability and Solubility enhancement of ellagic acid in cel- lulose ester solid dispersions. Carbohydr Polym. 2013; 92: 1443–1450. https://doi.org/10.1016/j. carbpol.2012.10.051 PMID: 23399175 Jang WS, Kang JM. Improved oral absorption and chemical stability of everolimus via preparation of solid dispersion using solvent wetting technique. Int J Pharm. 2014; 473: 187–193. https://doi.org/10. 1016/j.ijpharm.2014.06.006 PMID: 25003829 Liron LR, Ishai KE, Wizel S, Maydan AS, Hadas LR. Polymorphs of clopidogrel hydrogensulfate. US patent 7 074 928 B2, 2006. https://patentimages.storage.googleapis.com/05/62/19/527f20830e72bc/ US7074928.pdf (accessed on 22 October 2021). 9. Center for drug evaluation and research, FDA, Full prescribing information for: Plavix, (clopidogrel bisul- fate). Malvern, USA, 2018. https://www.accessdata.fda.gov/drugsatfda_docs/label/2018/ 020839s070lbl.pdf (accessed on 22 October 2021). 10. 11. Zupančić V, Smrkolj M, Benkič P, Simonič I, Plevnik M, Ritlop G et al. Preformulation Investigation of Some Clopidogrel Addition Salts. Acta Chim Slov. 2010; 57: 376–385. PMID: 24061734 Jassim ZE, Hussein AA. Formulation and evaluation of Clopidogrel tablet incorporating during nanopar- ticles. Int J Pharm Pharm Sci. 2014; 6: 838–851. 12. Koradia V, Chawla G, Bansal AK. Qualitative and quantitative analysis of clopidogrel bisulfate poly- morphs. Acta Pharm. 2004; 54: 193–204. PMID: 15610616 13. Lestari MLAD, Suciati, Indrayanto G, Brittain GH. Clopidogrel bisulfate. In: Brittain GH, editor. Profiles of drug substances, excipients, and related methology. New York: Elsevier Inc; 2010. pp. 71–115. https://doi.org/10.1016/S1871-5125(10)35002-3 14. Raijada KD, Singh S, Bansal KA. Influence of microenviromental pH, humidity, and temperature on the stability of polymorphyc and amorphous forms of clopidogrel bisulfate. Pharm Sci Tech. 2010; 11 (1): 197–203. https://doi.org/10.1208/s12249-010-9376-1 PMID: 20112136 15. Osmanović OE, Dzˇambić AL, Krstić M, Kulenović PM, Odović D, Vasiljević D. In Vitro Dissolution Study of Acetylsalicylic Acid and Clopidogrel Bisulfate Solid Dispersions: Validation of the RP-HPLC Method for Simultaneous Analysis. Appl Sci. 2020; 10(14):4792. https://doi.org/10.3390/app10144792 16. European Medicines Agency. ICH Topic Q1A (R2)—Stability Testing of new Drug Substances and Products. 2003. https://www.ema.europa.eu/en/documents/scientific-guideline/ich-q-1-r2-stability- PLOS ONE \| https://doi.org/10.1371/journal.pone.0266237 April 4, 2022 13 / 14 PLOS ONELong-term stability of clopidogrel solid dispersions—Importance of in vitro dissolution test testing-new-drug-substances-products-step-5_en.pdf (accessed on 12 November 2021). PMID: 14631936 17. Homayouni A, Sadeghi F, Nokhodchi A, Varshosaz J, Garekani AH. Preparation and characterization of celecoxib solid dispersions; comparison of poloxamer-188 and PVP-K30 as carriers. Iran J Basic Med Sci. 2014; 17: 322–331. PMID: 24967060 18. Sapkal S, Babhulkar M, Rathi A, Mehetre G, Narkhede BM. An overwiev on the mechanisms of solubil- ity and dissolution rate enhancement in solid dispersion. Int J Pharm Tech Res. 2013; 5: 31–39. 19. Mehatha KA, Suryadevara V, Lankapalli RS, Deshmukh MA, Sambath PL. Formulation and Optimiza- tion of Ezetimibe Containing Solid Dispersions Using Kollidon VA64. Turk J Pharm Sci. 2014; 11 (2): 113–126. 20. Fouad SA, Malaak FA, El-Nabarawi MA, Abu Zeid K, Ghoneim AM (2021) Preparation of solid disper- sion systems for enhanced dissolution of poorly water soluble diacerein: In-vitro evaluation, optimization and physiologically based pharmacokinetic modeling. PLoS ONE 16(1): e0245482. https://doi.org/10. 1371/journal.pone.0245482 PMID: 33471832 21. Bolourchian N, Mahboobian MM, Dadashzadeh S. The Effect of PEG Molecular Weights on Dissolution Behavior of Simvastatin in Solid Dispersions. Iran J Pharm Res. 2013; 12: 11–20. PMID: 24250667 22. European Medicines Agency. Guideline 3AQ11a - Specifications and Control Tests on the Finished Product. 1992. https://www.ema.europa.eu/en/documents/scientific-guideline/specifications-control- tests-finished-product_en.pdf (accessed 15 November 2021). 23. Liu X, Feng X, Williams RO III, Zhang F. Characterization of amorphous solid dispersions. J Pharm Inv. 2018; 48: 19–41. https://doi.org/10.1007/s40005-017-0361-5 24. Kolasˇinac N, Kachrimanis K, Homsˇek I, Grujić B, urić Z, Ibrić S. Solubility enhancement of desloratadine by solid dispersion in poloxamers. Int J Pharm. 2012; 436: 161–170. https://doi.org/10.1016/j.ijpharm. 2012.06.060 PMID: 22772487 25. Song Y, Wang L, Yang P, Wenslow RM Jr, Tan B, Zhang H, et al. Physicochemical Characterization of Felodipine-Kollidon VA64 Amorphous Solid Dispersions Prepared by Hot-Melt Extrusion. J Pharm Sci. 2013; 102 (6): 1915–1923. https://doi.org/10.1002/jps.23538 PMID: 23580396 26. Paudel A, Worku ZA, Meeus J, Guns S, Van den Mooter G. Manufacturing of solid dispersions of poorly water soluble drugs by spray drying: formulation and process considerations. Int J Pharm. 2013; 453(1): 253–284. https://doi.org/10.1016/j.ijpharm.2012.07.015 PMID: 22820134 27. Patel BB, Patel JK, Chakraborty S, Shukla D. Revealing facts behind spray dried solid dispersions tech- nology used for solubility enhancement. Saudi Pharm J. 2015; 23: 352–365. https://doi.org/10.1016/j. jsps.2013.12.013 PMID: 27134535 28. Srivastava A, Mishra S, Tandon P, Patel S, Ayala AP, Bansal AK, et al. Molecular structure and vibra- tional spectroscopic analysis of an antiplatelet drug; clopidogrel hydrogen bisulfate (form 2)- A com- bined experimental and quantum chemical aproach. J Mol Struc. 2010; 964: 88–96. https://doi.org/10. 1016/j.saa.2012.11.093 PMID: 23277180 29. Alladi S, Nalini S. Preparation, physico chemical characterization of solid dispersion of tenoxicam with poloxamer. J Pharm Sci Technol. 2010; 20: 308–301. 30. Meng F, Gala U, Chauhan H. Classification of solid dispersions: correlation to (i) stability and solubility (ii) preparation and characterization techniques. Drug Dev Ind Pharm. 2015; 41(9): 1–15. https://doi. org/10.3109/03639045.2015.1018274 PMID: 25853292 31. Damian F, Blaton N, Van den Mooter G. Physical stability of solid dispersions of the antiviral agent UC- 781 with PEG 6000, Gelucire® 44/14 and PVP K30. Int J Pharm. 2002; 244: 87–98. https://doi.org/10. 1016/s0378-5173(02)00316-2 PMID: 12204568 32. 33. Ford JL, Rubinstein MH. Ageing of indomethacin-polyethylene glycol 6000 solid dispersions. Pharm Acta Helv. 1979; 54: 353–358. PMID: 538042 Ford JL, Rubinstein MH. Formulation and ageing of tablets prepared from indomethacin-polyethylene glycol 6000 solid dispersions. Pharm Acta Helv. 1980; 55: 1–7. PMID: 7367448 34. Costa P, Sousa Lobo JM. Modeling and comparison of dissolution profiles. Eur J Pharm. Sci. 2001; 13: 123–133. https://doi.org/10.1016/s0928-0987(01)00095-1 PMID: 11297896 35. Savu NS, Silvestro L, Mircioiu C, Anuta V. Development of in vitro in vivo correlation models for clopido- grel tablets to describe administration under fasting and fed conditions. Farmacia. 2016; 64: 302–312. 36. Karthikeyan M, Deepa MK, Bassim E, Rahna CS, Raj KRS. Investigation of kinetic drug release charac- teristics and in vitro evaluation of sustained-release matrix tablets of a selective COX-2 inhibitor for rheumatic diseases. J Pharm Innov. 2021; 16: 551–557. https://doi.org/10.1007/s12247-020-09459-9 37. Higuchi T. Mechanism of sustained-action medication: theoretical analysis of rate of release of solid drugs dispersed in solid matrices. J Pharm Sci. 1963; 52: 1145–1148. https://doi.org/10.1002/jps. 2600521210 PMID: 14088963 PLOS ONE \| https://doi.org/10.1371/journal.pone.0266237 April 4, 2022 14 / 14 PLOS ONE
10.1371_journal.pone.0264137	RESEARCH ARTICLE Association between living in municipalities with high crowding conditions and poverty and mortality from COVID-19 in Mexico Viridiana Rı´os1☯, Edgar Denova-Gutie´ rrezID 2☯, Simo´ n Barquera2* 1 Independent Researcher, Mexico City, Mexico, 2 Centro de Investigacio´ n en Nutricio´ n y Salud, Instituto Nacional de Salud Pu´ blica, Cuernavaca, Morelos, Me´xico ☯ These authors contributed equally to this work. * sbarquera@insp.mx a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 Abstract Background OPEN ACCESS Citation: Rı´os V, Denova-Gutie´rrez E, Barquera S (2022) Association between living in municipalities with high crowding conditions and poverty and mortality from COVID-19 in Mexico. PLoS ONE 17(2): e0264137. https://doi.org/10.1371/journal. pone.0264137 Editor: Omar Yaxmehen Bello-Chavolla, Instituto Nacional de Geriatria, MEXICO Received: August 30, 2021 Accepted: February 3, 2022 Published: February 22, 2022 Copyright: © 2022 Rı´os et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The data was stored in Zenodo. The digital object identifier is:10.5281/ zenodo.5999622. Funding: VR received funding form Ha´bitat para la Humanidad, Me´xico. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. The World Health Organization stated a pandemic by severe acute respiratory syndrome coronavirus SARS-Cov2 (COVID-19) on March, 2020 with devastating implications for pop- ulations, healthcare systems, and economies globally. Objective The present study explores the association between patients living in municipalities with crowding conditions and poverty and mortality from COVID-19 in Mexico; specifically evalu- ating the socioeconomic characteristics of the municipality in which the patients reside and some individual characteristics. Methods In the present study, we examined public information collected from the National Epidemio- logical Surveillance System informing all persons tested for SARS-CoV-2 and published by the Ministry of Health. The present analysis was restricted to those with the date of registra- tion to October 12, 2021. The association between the main exposures (overcrowded condi- tions and poverty) and the outcomes of interest (death by COVID-19) was explored using Cox proportional hazard regression models, including frailty penalties to accommodate mul- tilevel data and random effects for the municipality of case occurrence. Results A total of 9619917 subjects were included in the Epidemiological Surveillance System for viral respiratory disease platform. Of those for which results were available, 6141403 were negative for COVID-19 and 3478514 were positive for COVID-19; with a total of 273216 deaths in those who tested positive. Among those positive to COVID-19 mean age was 46.9. Patients living in municipalities with high rates of crowding conditions increased the risk of dying from COVID-19 by 8% (95% CI: 1.03, 1.14). Individuals living in municipalities PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 1 / 15 PLOS ONECrowding conditions and the risk of mortality from COVID-19 Abbreviations: SARS-Cov2 (COVID-19), severe acute respiratory syndrome coronavirus; SISVER, Epidemiological Surveillance System for viral respiratory disease; USMER, sentinel surveillance health units (Spanish acronym for Unidades de Salud Monitoras de Enfermedad Respiratoria); CONAPO, Mexican Population Council (Spanish acronym for Consejo Nacional de Poblacio´n); CONEVAL, Mexican National Council on Evaluation of Social Policy (Spanish acronym for Consejo Nacional de Evaluacio´n de la Polı´tica de Desarrollo Social); CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; SD, standard deviations; HR, Hazard ratio; 95% CI, 95% confidence intervals. with indigenous background was associated with an increased risk of dying from COVID-19 (HR = 1.10; 95% CI: 1.04, 1.17). Individuals living in municipalities with illiteracy (HR = 1.09; 95% CI: 1.03, 1.11), poverty (HR = 1.17; 95% CI: 1.14, 1.19), food insecurity (HR = 1.094; 95% CI 1.02, 1.06), limited access to social security (HR = 1.10; 95% CI: 1.08, 1.13) and health services (HR = 1.06; 95% CI: 1.04, 1.08) had a higher risk of mortality from COVID- 19. Conclusion Our data suggest that patients living in municipalities with higher rates of crowding condi- tions and higher rates of poverty had elevated risk of mortality from COVID-19. In Mexico, the COVID-19 pandemic is a systemic crisis linked to human development since we have seen that it affects less developed and more vulnerable municipalities. Policies to reduce vulnerabilities and develop strategies to deal with health crises like the current one needs to be considered. Introduction The World Health Organization stated a pandemic by severe acute respiratory syndrome coro- navirus SARS-Cov2 (COVID-19) in March 2020 [1] with devastating implications for popula- tions, healthcare systems, and economies globally. Understanding the predictors of mortality by COVID-19 is essential for targeting preventing efforts around the world. Extensive litera- ture has proven that age, obesity [2], cardiovascular disease [3, 4], hypertension [5], and type 2 diabetes [6] are established predictors of adverse COVID-19 but we have yet to understand whether socioeconomic determinants play a role on mortality risk. Based in the “Social Determinants of Health” approach and the ecosocial epidemiology, the cumulative and dynamic interplay processes of exposure, susceptibility, and resistance, which influences health at the singular and general levels, could have been an important factor to understand processes which shape population health—as well as the COVID-19 pandemic [7– 11]. Some studies have already suggested that socioeconomic deprivation [12–14], such as crowding conditions [15], living in poorly ventilated spaces [16, 17], and social inequality have a major impact on the transmission of the virus [18]. We also know that these effects may be higher for pregnant women [19]. There is evidence that there is an over-representation of black, Asian and minority ethnic in the COVID-19 positive groups [20–23]. Other communicable and non-communicable disease pandemics show similar socioeco- nomic patterns [24]. People living in neighborhoods with a high concentration of poverty are more likely to develop diabetes [25, 26], chronic obstructive pulmonary disease [27], and air obstruction [28]. Yet, we do not know what the mechanisms behind such relationship may be. We studied the role of socioeconomic conditions on the effects of COVID-19 by dividing its potential impacts on two mechanisms: those related with living in crowding conditions, and those related directly with income poverty. Closed environments may facilitate secondary transmis- sion of coronavirus disease [17]. The association between COVID-19 and some social determinants in Mexico has been studied [7, 29–32], however, some important questions remain with this topic. Thus, the PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 2 / 15 PLOS ONECrowding conditions and the risk of mortality from COVID-19 present study explores the association of living in municipalities with higher rates of crowding conditions and poverty and the risk of mortality from COVID-19 in Mexico. In this sense, we hypothesize that a) the risk of mortality will be higher for individuals living in municipalities with higher rates of crowding conditions and poverty, and b) the risk of mortality will be higher for individuals with obesity, diabetes, and hypertension and living in municipalities with higher rates of crowding conditions. Methods Study design and data source In the present study, we examined public information collected from the National Epidemio- logical Surveillance System (In Spanish, Sistema Nacional de Vigilancia Epidemiolo´gica [SINAVE]) informing all persons tested for COVID-19 and published by the Ministry of Health [33]. This data is updated every day and included: demographic characteristics, certain comorbidities (e.g. Obesity, type 2 diabetes, hypertension, among others), test result (negative or positive to COVID-19), place of residence and whether the patient died. Due to the inci- dence of severe disease related to COVID-19 and the prevalence of non-communicable dis- eases is very low in the population under 20 years of age, we only included patients aged 20 years or older [34]. The database contains information of patients who had a COVID-19 test in Mexico at pub- lic or private laboratory services and were registered by the Epidemiological Surveillance Sys- tem for Viral Respiratory Disease (In Spanish, Sistema de Vigilancia Epidemiolo´gica de Enfermedad Respiratoria Viral [SISVER]). Additionally, the patients included could be both hospitalized and ambulatory. Furthermore, as we reported previously [34], the information included patients from the SISVER in which 475 sentinel surveillance health units (In Spanish, Unidades de Salud Moni- toras de Enfermedades Respiratorias [USMER]), across the country test at least 10% of all cases with mild respiratory disease and 100% of severe respiratory disease. Also, patients from hospital base surveillance (no USMER) with severe symptoms of respiratory disease were included (S1 Fig). Assessment of mortality in patients with a positive diagnosis of COVID-19 For the initial analysis, we only included those who were negative or positive for COVID-19 and who had complete information (S1 Table). In the final analysis, only those who were posi- tive for COVID-19 (laboratory confirmed and diagnosed by using the real-time reverse-tran- scription polymerase chain reaction method or by rapid Antigen testing) and were recorded as deaths from COVID-19 were included (S2 Fig). Assessment of overcrowded conditions To assess social indicators across the study we used the estimations of the Mexican Population Council (In Spanish, Consejo Nacional de Poblacio´n [CONAPO]) [35] and the Mexican National Council on Evaluation of Social Policy (In Spanish, Consejo Nacional de Evaluacio´n de la Polı´tica de Desarrollo Social [CONEVAL]) [36]. Given that at the individual level it is not possible to know directly whether someone lives in crowding conditions, the municipal crowding rate was considered as a probability that the individual lives in high crowding condi- tions. In this sense, the definition of CONAPO and CONEVAL were followed. For this, two possible measurements were considered at the municipal level. First, the percentage of house- holds that have one bedroom with three or more occupants, or with two bedrooms with five or PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 3 / 15 PLOS ONECrowding conditions and the risk of mortality from COVID-19 more occupants. Second, the percentage of households with 2.5 inhabitants or more per room, considering the kitchen but excluding hallways and bathrooms, as measured by the CONEVAL. Assessment of municipal level sociodemographic characteristics These characteristics were divided into three categories: 1) Those related to poverty and income distribution, 2) Those that identify deficiencies and vulnerabilities, and 3) Other spe- cific deficiencies. These characteristics were selected for its possible theoretical relationship with the objective of the study. Poverty and income distribution. Constituted of six different variables: 1) Gini index, which measures the economic inequality of a society, by exploring the level of concentration that exists in the distribution of income among the population. 2) The income below the welfare line, which is equivalent to the total value of the national basic food basket and the non-food basket per person per month. 3) The minimum welfare line, which is equivalent to the value of the food basket per person per month. 4) The percentage of the population of the municipality that lives in a condition of multidisciplinary poverty, defined as the per- centage of people that has at least one social deprivation (in the indicators of educational backwardness, access to health services, access to social security, quality and spaces of the house, basic services in the house and access to food) and if their income is insufficient to purchase the goods and services they require to meet their food and non-food needs. 5) The extreme multidisciplinary poverty, defined as the percentage of people who have three or more social deprivations out of six possible and, in addition, their total income is less than the minimum welfare line. The population in this situation has such a low income that even if it were entirely devoted to purchasing food, it would not be able to access those who make up the food basket. And 6) The income ratio which measures the ratio of the average income of the extreme poor population and the average income of the non-poor and non-vulnera- ble population [35, 36]. Deficiencies and vulnerabilities. In this section we included six variables: 1) Percentage of the municipality that has limited access to food, defined as the percentage of people who have moderate to severe food insecurity. 2) Lack of access to social security, defined as the per- centage of people who have deficiencies in access to social security according to their age and employment status. 3) Lack of access to basic housing services, defined as the percentage of people who do not meet any of the following four criteria: piped water inside the home or out- side the home but on the property; drainage connected to the public water service or to a septic tank; electricity obtained from the public service, from a solar panel or from another source, and that the fuel for cooking is LP gas or natural gas, electricity, and if it is firewood or charcoal that the kitchen has a fireplace. 4) Lack of access to health services, percentage of people who do not have an affiliation or the right to receive health services from any public health institu- tion. 5) Educational backwardness, percentage of people who are of school age and who do not attend school, or if according to their age they have not completed primary or secondary school, according to the criteria indicated above. And 6) Vulnerability due to income, when he/she does not have sufficient income to buy the aggregation of the basic food basket with the basic non-food basket but has no social deprivations. Other specific indicators of vulnerability. Seven additional variables describing specific deficiencies were identified: the percentage of people with illiteracy by municipality, with dirt floors in their homes, who live in localities with less than 5,000 inhabitants, who have incom- plete primary school, housing without electricity, housing without a toilet or drainage, and who speak an indigenous language. PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 4 / 15 PLOS ONECrowding conditions and the risk of mortality from COVID-19 Assessment of individual level characteristics For this case, we included sociodemographic characteristics (e.g., age, sex, state of birth, state and municipality of residence, nationality, migration status, and, for imported cases, country of origin and native language); we also incorporated event management variables related to COVID-19 like ambulatory or hospitalized management, date of hospitalization (for those who were hospitalized: diagnosis of pneumonia, intubation, and treatment in the intensive care unit), date of symptoms onset, and date of death. Finally, variables related to identification of the case like date of the report, type of testing facility (USMER or not), and healthcare pro- vider were included. Assessment of non-communicable diseases These variables were obtained from SISVER, a public dataset published by the Ministry of Health [33]. Eight variables describe pre-existing non-communicable diseases obtained by self-report like: obesity, hypertension, diabetes, cardiovascular disease, chronic kidney disease (CKD), chronic obstructive pulmonary disease (COPD), asthma, and immunosuppression. Statistical analysis Descriptive analyses of the main characteristics of interest were performed for all patients tested and stratifying by test result (positive or negative to COVID-19). Categorical variables were described as percentages, and continuous variables were defined as means and standard deviations (SDs). In this primary analysis, we compared those who tested positive and negative to COVID-19. In a second step, we only evaluated patients who tested positive for COVID-19. To analyze this group by tertiles (low, medium and high) of municipalities with information on crowding conditions. To model COVID-19 mortality risk we computed Cox proportional hazard regression models, including frailty penalties to accommodate multilevel data and random effects for the municipality of case occurrence, approximating iterations using the Newton–Raphson algo- rithm. Random effect estimates were exponentiated to calculate baseline mortality hazards across municipalities to represent geographical heterogeneity [37]. We tested the joint associa- tion of living in municipalities with higher rates of crowding conditions and access to health services with mortality from COVID-19. To assess a possible effect modification, we explored stratified analyses by access to health services and chronic conditions (obesity, type 2 diabetes and hypertension) and living in municipalities with higher rates of crowding conditions. The statistical analyses were conducted using the Stata Software statistical software package, version 13.0 (StataCorp, College Station, Texas). All P values presented are two-sided; P < 0.05 was considered statistically significant. Results Characteristics of individuals tested for COVID-19 A total of 9619917 individuals who were tested for COVID-19 and were included in the SIS- VER platform, their characteristics are listed in S1 Table. Of those for which results were avail- able, 6141403 were negative for COVID-19 and 3478514 were positive for COVID-19; with a total of 273216 deaths in those who tested positive for COVID-19. The mean age of study participants was 43.6 years, among those positive to COVID-19, mean age was 46.9 and 40.8 among those negative to COVID-19. Approximately 47.6% of our study population were women. A higher proportion of individuals who tested positive to PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 5 / 15 PLOS ONECrowding conditions and the risk of mortality from COVID-19 COVID-19 live in communities with higher indigenous background (7.3 vs 5.0%). Gini index was slightly higher among negative to COVID-19 (41.0 vs 39.0%). Individuals who tested posi- tive to COVID-19 live in communities with higher rates of lack of access to health services (10.7 vs 9.1%) and vulnerability due to income (8.8 vs 8.6%). Prevalence of chronic conditions were higher among those who tested positive to COVID-19. Characteristics of individuals tested positive COVID-19 Sociodemographic characteristics of the 3478514 individuals who tested positive for COVID- 19 according to levels of crowding conditions are listed in Table 1. Compared to those living in municipalities with lower rates of crowding conditions a higher proportion of municipali- ties with higher rates of indigenous population, population with incomplete elementary school and illiterate was observed among individuals living in municipalities with higher rates of crowding conditions. In terms of poverty and income distribution, Gini index was higher among those who live in municipalities with higher rates of crowding conditions, we also observed that these individuals live in municipalities with higher rates of multidisciplinary poverty and extreme multidisciplinary poverty when compared to the individuals living in municipalities with lower rates of crowding conditions. Gaps and vulnerabilities such as food insecurity, social security and health services were more present in subjects living in munici- palities with higher rates of crowding conditions. People living in municipalities with worse Table 1. Sociodemographic characteristics and chronic conditions of patients with COVID-19 according to the municipality level of crowding conditions. Variable Age Women, (%) Indigenous population, (%) Incomplete Primary school, (%) Illiteracy, (%) Gini index, (%) Lower income welfare line, (%) Lower income minimum welfare line, (%) Multidisciplinary poverty, (%) Extreme multidisciplinary poverty, (%) Educational backwardness, (%) Vulnerability due to income, (%) Limited access to food, (%) Limited access to social security, (%) Lack of access to basic housing services, (%) Limited access to health services, (%) Chronic conditions Obesity, (%) Diabetes, (%) Hypertension, (%) Cardiovascular disease, (%) Chronic obstructive pulmonary disease, (%) Chronic kidney disease, (%) Inmunosupresio´n, (%) Asthma, (%) https://doi.org/10.1371/journal.pone.0264137.t001 Low n = 1085954 Medium n = 1167697 High n = 1224863 Municipality level of crowding conditions 48.3 49.6 4.8 18.1 2.2 37.9 33.5 7.2 24.2 1.2 9.5 6.8 12.2 39.6 2.4 16.3 11.9 10.9 14.0 1.4 1.1 1.4 0.8 2.4 47.3 51.1 4.9 22.3 4.6 39.3 43.1 11.0 33.1 2.6 12.3 9.3 17.7 46.8 5.7 17.6 12.3 11.2 14.7 1.4 1.1 1.4 0.9 2.0 49.5 49.5 5.3 29.7 8.0 40.5 52.1 17.4 45.2 7.3 17.7 9.9 24.2 59.2 22.9 17.7 13.8 12.5 15.8 1.6 1.2 1.5 1.0 2.4 PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 6 / 15 PLOS ONECrowding conditions and the risk of mortality from COVID-19 crowding conditions were more likely to have a diagnosis of obesity, type 2 diabetes and hyper- tension compared to those living in municipalities with lower rates of crowding conditions. Role of overcrowding conditions and poverty on mortality Crude and adjusted hazard ratios and 95% CIs of mortality amongst those who tested positive for COVID-19 are presented in Table 2. Living in a municipality with higher rate of crowding conditions increased the risk of dying from COVID-19 by 8% (95% CI: 1.03, 1.14). Living in a municipality with higher rates of indigenous background was associated with an increased risk of dying from COVID-19 (HR = 1.10; 95% CI: 1.04, 1.17). Individuals living in a municipality with higher rates of illiteracy (HR = 1.09; 95% CI: 1.03, 1.11), poverty (HR = 1.17; 95% CI: 1.14, 1.19), food insecurity (HR = 1.04; 95% CI 1.02, 1.06), limited access to social security (HR = 1.10; 95% CI: 1.08, 1.13) and limited health services (HR = 1.06; 95% CI: 1.04, 1.08) had a higher risk of dying from COVID-19. Table 2. Hazard ratios (95% confidence intervals) for mortality amongst those tested positive for COVID-19. Variables Crude Adjusted Municipality rates of crowding conditions HR (95% CI) Low Medium High P for trend 1.00 1.01 1.06 Municipality rates idigenous population No Yes P value Municipality rates of illiteracy Low Medium High P for trend Municipality rates of poverty Low Medium High P for trend 1.00 1.16 1.00 1.01 1.05 1.00 1.05 1.12 Municipality rates of food insecurity Low Medium High P for trend 1.00 1.01 1.05 Municipality rates of access to social security High Medium Low P for trend 1.00 1.01 1.08 – 1.00, 1.02 1.03 1.08 <0.001 – 1.09, 1.24 <0.001 – 1.00, 1.02 1.04, 1.07 <0.001 0.001 – 1.04, 1.06 1.11, 1.13 – 0.98, 1.03 1.04, 1.08 <0.001 – 1.00, 1.02 1.07, 1.10 <0.001 Municipality rates of access to health services High access 1.00 – PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 HR 1.00 1.02 1.08 1.00 1.10 1.00 1.01 1.09 1.00 1.08 1.17 1.00 1.00 1.04 1.00 1.02 1.10 1.00 (95% CI) – 1.01, 1.05 1.03, 1.14 <0.001 – 1.04, 1.17 <0.001 – 0.99, 1.03 1.03, 1.11 <0.001 – 1.07, 1.10 1.14, 1.19 <0.001 – 0.98, 1.02 1.02, 1.06 – 1.01, 1.04 1.08, 1.13 <0.001 – (Continued ) 7 / 15 PLOS ONECrowding conditions and the risk of mortality from COVID-19 Table 2. (Continued) Variables Middle access Low access P for trend Age � 20 to < 40 years � 40 to < 60 years � 60 years P for trend Sex Women Men P value Crude Adjusted HR 0.99 1.04 (95% CI) 0.98, 1.00 1.03, 1.05 <0.001 1.00 5.41 15.90 1.00 1.45 <0.001 – 4.30, 7.92 12.11, 19.90 – 1.30, 1.59 HR 1.00 1.06 1.00 5.13 14.76 1.00 1.40 (95% CI) 0.99, 1.02 1.04, 1.08 <0.001 – 4.11, 7.35 11.20, 19.41 – 1.28, 1.54 <0.001 <0.001 Municipalities with localities with less than 5,000 inhabitants Yes P value 1.18 1.10, 1.24 1.14 1.04, 1.25 <0.001 <0.001 Adjusted for: Age, sex, indigenous population, illiteracy, poverty rate, lack of access to food, lack of social security, lack of health services, Gini index, localities <5 thousand inhabitants, reduced mobility, obesity, diabetes, hypertension, COPD, immunosuppression, asthma, cardiovascular disease, chronic kidney disease. Adjusted for: Age, sex, indigenous population, illiteracy, poverty rate, lack of access to food, lack of social security, lack of health services, Gini index, localities <5 thousand inhabitants, reduced mobility, smoking, obesity, diabetes, hypertension, COPD, immunosuppression, asthma, cardiovascular disease, chronic kidney disease, time from symptom onset up to death. https://doi.org/10.1371/journal.pone.0264137.t002 Interaction between living in municipalities with overcrowding conditions, access to health services and chronic conditions We tested the joint association between living in municipalities with higher rates of crowding conditions and access to health services and chronic conditions with mortality from COVID- 19 (Figs 1 and 2). In Fig 1, the reference group for comparisons was composed of subjects liv- ing in municipalities with lower rates of crowding conditions and subjects living in municipal- ities with higher rates of access to health services. Relative to the reference group, the HR of the individuals living in a municipality with higher rates of crowding conditions and individu- als living in municipalities with lower rates of access to health services was 1.53 (95% CI: 1.40, 1.67; P for interaction <0.001). Additionally, we observed that relative to the reference group, the HR of the group living in a municipality with higher rates of crowding conditions and liv- ing with type 2 diabetes was 1.20 (95% CI: 1.14, 1.27; P for interaction <0.001). Discussion The aim of the present study was to explore the association between crowding conditions and poverty and the mortality risk from COVID-19 in Mexico, with additional emphasis in other social determinants like some deficiencies (e.g. food insecurity) and vulnerabilities (e.g. indige- nous population, illiteracy), as well as the interaction with some comorbidities. From a total of 3478514 cases who tested positive for COVID-19, our data suggests that municipalities with higher rates of crowding conditions, and higher rates of poverty (measured by the income PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 8 / 15 PLOS ONECrowding conditions and the risk of mortality from COVID-19 Fig 1. Interaction between crowding conditions and access to health services and mortality on patients with COVID-19. https://doi.org/10.1371/journal.pone.0264137.g001 below the welfare line) are associated with higher risk of mortality from COVID-19 in Mexi- can adult population. In addition, we also found, that in municipalities with higher rates of conditions like being indigenous, illiterate, food insecure, lack of access to social security, health services, from places with <5,000 inhabitants increase the risk of dying from COVID- 19. Also, there is multiple evidence suggesting that noncommunicable diseases and COVID-19 may be associated at different levels, increasing the probability of severe disease and death [34, 38]. Furthermore, given the high burden of disease attributable to non-communicable diseases in Mexico, it is necessary to generate information on their contribution to the risk of mortality from COVID-19. Our findings suggest that individuals living in a municipality with higher Fig 2. Interaction between crowding conditions and a) Obesity, b) Type 2 diabetes and c) Hypertension and mortality on patients with COVID-19. https://doi.org/10.1371/journal.pone.0264137.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 9 / 15 PLOS ONECrowding conditions and the risk of mortality from COVID-19 rates of crowding conditions and living with obesity, hypertension, and/or diabetes, were asso- ciated to a higher risk of mortality from COVID-19. Our study suggests that social inequalities, like higher rates of crowding conditions and poverty have a relevant contribution and role in the COVID-19 pandemic in Mexico. Our analysis found that patients living in municipalities with higher rates of crowding conditions had 8% (95% CI: 1.03, 1.14; P trend <0.001) higher risk to die from COVID-19, compared to those living in a municipality with lower rates of crowding conditions. A recent study con- ducted by Krieger et al. [39] found a similar association; households in the highest crowding group had greater risk ratios (RR = 1.7; 95% CI: 1.0, 2.9) of mortality from COVID19 com- pared to those in lowest crowding group. In this sense, previous studies have suggested that higher rates of crowding conditions has been related to certain factors such as poor ventilation, air pollution and poor sanitation within homes which has been related with infectious diseases, particularly respiratory diseases [40]. Additionally, other study has proposed that higher rates of crowding conditions could increase anxiety, stress and depressive symptoms which has been related with obesity and other non-communicable diseases via low grade inflammation and this could also explain the relation with COVID-19 [41]. Initially in Mexico, the spread of COVID-19 began in subjects of high socioeconomic status who live in more developed municipalities of the country; however, our data suggest that more vulnerable populations living in small areas, which are exposed to a persistent and historical context of social deprivation, have been more affected by COVID-19. Thus, we observed that patients living in municipalities with higher rates of poverty had 17% (95% CI: 1.14, 1.19; P trend <0.001) higher risk of mortality from COVID-19, compared to those living in munici- palities with lower rates of poverty. Consistent with our findings, previous studies have sug- gested that poverty and other vulnerabilities are associated with an increased risk of mortality from COVID-19 [39, 42–44]. These findings are of concern specially since these populations lack the resources to imple- ment or adopt preventive measures. For example, given their socio-economic conditions, hav- ing no financial reserves, and depending on emergency government assistance, they will scarcely be able to adhere to special recommendations such as social isolation, wearing masks, and hand hygiene. Also, the lack of access to minimum resources, such as potable clean water and basic sanitation (e.g. drainage) can increase the risk of illness due to COVID-19, as observed with other respiratory diseases [45]. To our knowledge, there are no previous studies evaluating the joint effect of crowding con- ditions and lack of access to health services with the risk of mortality from COVID-19. Our results suggest that patients living in municipalities with higher rates of crowding conditions and with lower rates of access to health services had 53% (95% CI: 1.34, 1.72; P for interaction <0.001) higher risk to die from COVID-19, compared with those patients living in municipali- ties with lower rates of crowding conditions and living in municipalities with higher rates of access to health services. We also found a joint effect of living in municipalities with higher rates crowding condi- tions and living with certain comorbidities like obesity, diabetes, and/or hypertension, and the risk of mortality from COVID-19. For example: our data proposed that compared with patients living in municipalities with lower rates of crowding conditions and without obesity, patients living with obesity and living in municipalities with higher rates crowding conditions had 1.15 times higher risk of dying from COVID-19. As these communities face crowding conditions, poverty, lack of access to health care, food insecurity, lack of access to basic sanitation services, among other deficiencies, there is a need to focus more on the underlying factors that put them at higher risk. For that reason, in these vulnerable groups the governments should place more emphasis on their social distancing PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 10 / 15 PLOS ONECrowding conditions and the risk of mortality from COVID-19 behaviors and assess their needs for basic resources such as food, medicine, personal protective equipment and other essential supplies necessary to socially distance themselves during the current pandemic. COVID-19 has the potential to substantially exacerbate socioeconomic and ethnic inequali- ties in health, unless measures are taken to mitigate these inequalities [46]. Thus, policy response is essential to guarantee that the health system is reactive to the requirements of these vulnerable groups. The present study has some important limitations. First, collider bias is an important problem and could result in several ways, including differential healthcare requesting, dif- ferential examination and differential diagnosis. Despite this, we did not find evidence to suggest that differential medical care seeking, or testing would explain the pattern of results observed in our study. Second, as we noted in a previous study [34], Although the informa- tion is from subjects from all over Mexico, patients who were asymptomatic or who were treated at home are not part of our data, so our study represents only those most severe cases of COVID-19, and the results observed cannot be extrapolated to non-severe COVID-19 cases. Moreover, since sentinel units were not randomly designated, our results are not likely to be representative of the entire Mexican population. Furthermore, at the national level, by protocol, tests were only performed on those patients with severe respira- tory conditions, and to a lesser extent on subjects with mild respiratory conditions; there- fore, our results may underestimate how living in municipalities with higher rates of crowding and higher rates of poverty are associated with the COVID-19 prognosis, as it is highly likely that people living in municipalities with higher rates of crowding and higher rates of poverty have less access to health services. In addition, in relation of the association with non-communicable diseases, the protocol mentioned above may also underestimate the relationships, as those with mild respiratory conditions will have a better prognosis. Finally, our indicators of crowding, poverty, vulnerabilities, and other conditions, share these indicators at the municipality level; thus, it does not reflect the condition of the indi- vidual but that of the municipality where people live. Although the limitations, the present study has some strengths. Comparative studies between crowding conditions and poverty as well as other vulnerabilities associated with the risk of mortality from COVID-19 are scarce; therefore, our findings may help identify these relations. Although it is probable that our results are not representative of the entire popula- tion, our study contains nationwide data. Lastly, as previously reported [34], multivariate anal- ysis, due to our large sample size, reduce de possibility of confounding factors. Conclusion This study adds information that emphasize that vulnerable communities have a greater risk of mortality from COVID-19. Our data suggest that patients living in municipalities with higher rates of crowding conditions, higher rates of poverty, and higher rates of other condi- tions; indigenous population, illiterate, food insecurity, lack of access to social security, lack of access to health services had higher risk of mortality from COVID-19. Also, we found a joint effect of living in municipalities with higher rates of crowding conditions and living with cer- tain comorbidities like obesity, diabetes, and/or hypertension, and the risk of mortality from COVID-19 in Mexican population. In Mexico, the COVID-19 pandemic is a systemic crisis linked to human development since we have seen that it affects less developed and more vulnerable localities. Policies to reduce vulnerabilities and develop strategies need to consider the following points: 1) adoption of effective evidence-based prevention mechanisms that consider collective risk and the social PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 11 / 15 PLOS ONECrowding conditions and the risk of mortality from COVID-19 context; 2) expand and prepare the health system at all levels; 3) guarantee social protection; and 4) develop strategies that promote vaccination of vulnerable populations. Supporting information S1 Fig. Flowchart of the Mexican National Epidemiological Surveillance System for viral respiratory disease. (TIF) S2 Fig. Standardized guideline for Epidemiological and Laboratory Surveillance of viral respiratory disease. (TIF) S1 Table. Sociodemographic characteristics and chronic conditions of the study popula- tion. (DOCX) Acknowledgments We thank Mariel White for her support in proofreading. Author Contributions Conceptualization: Viridiana Rı´os, Simo´n Barquera. Data curation: Viridiana Rı´os. Formal analysis: Viridiana Rı´os, Edgar Denova-Gutie´rrez. Funding acquisition: Viridiana Rı´os. Investigation: Edgar Denova-Gutie´rrez, Simo´n Barquera. Methodology: Edgar Denova-Gutie´rrez. Supervision: Simo´n Barquera. Writing – original draft: Edgar Denova-Gutie´rrez. Writing – review & editing: Viridiana Rı´os, Edgar Denova-Gutie´rrez, Simo´n Barquera. References 1. World Health Organization. WHO Director-General’s opening remarks at the media briefing on COVID- 19. 2020 Mar. World Health Organization 2020. Available from: https://www.who.int/dg/speeches/ detail/who-director-general-s-opening-remarks-at-the-mediabriefing-on-covid-19-11-march-2020 2. Sattar N, McInnes IB, McMurray J J. Obesity a risk factor for severe COVID-19 infection: multiple poten- tial mechanisms. Circulation 2020; 142(1):4–6. https://doi.org/10.1161/CIRCULATIONAHA.120. 047659 PMID: 32320270 3. Zheng YY, Ma YT, Zhang JY, Xie X. COVID-19 and the cardiovascular system. Nature Reviews Cardi- ology 2020, 17(5), 259–260. https://doi.org/10.1038/s41569-020-0360-5 PMID: 32139904 4. Clerkin KJ, Fried JA, Raikhelkar J, Sayer G, Griffin JM, Masoumi A, et al. COVID-19 and cardiovascular disease. Circulation 2020, 141(20), 1648–1655. https://doi.org/10.1161/CIRCULATIONAHA.120. 046941 PMID: 32200663 5. 6. Zhou F, Yu T, Du R, Fan G, Liu Y, Liu Z, et al. Clinical course and risk factors for mortality of adult inpa- tients with COVID-19 in Wuhan, China: a retrospective cohort study. Lancet 2020; 395(10229):1054– 1062. https://doi.org/10.1016/S0140-6736(20)30566-3 PMID: 32171076 Fang L, Karakiulakis G, Roth M. Are patients with hypertension and diabetes mellitus at increased risk for COVID-19 infection? Lancet Respiratory Medicine 2020; 8(4):e21. https://doi.org/10.1016/S2213- 2600(20)30116-8 PMID: 32171062 PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 12 / 15 PLOS ONECrowding conditions and the risk of mortality from COVID-19 7. Antonio-Villa NE, Fernandez-Chirino L, Pisanty-Alatorre J, Mancilla-Galindo J, Kammar-Garcı´a A, Var- gas-Va´ zquez A, et al. Comprehensive evaluation of the impact of sociodemographic inequalities on adverse outcomes and excess mortality during the COVID-19 pandemic in Mexico City. Clin Infect Dis 2021: ciab577. https://doi.org/10.1093/cid/ciab577 PMID: 34159351 8. Borde E, Herna´ ndez M. Revisiting the social determinants of health agenda from the global South. Glob Public Health 2019; 14:847–862. https://doi.org/10.1080/17441692.2018.1551913 PMID: 30500313 9. Spiegel JM, Breilh J, Yassi A. Why language matters: insights and challenges in applying a social deter- mination of health approach in a North-South collaborative research program. Global Health 2015; 11:9. https://doi.org/10.1186/s12992-015-0091-2 PMID: 25880442 10. Waitzkin H, Iriart C, Estrada A, Lamadrid S. Social medicine then and now: lessons from Latin America. Am J Public Health 2001; 91:1592–601. https://doi.org/10.2105/ajph.91.10.1592 PMID: 11574316 11. Krieger N. Theories for social epidemiology in the 21st century: an ecosocial perspective. Int J Epide- miol 2001; 30:668–677. https://doi.org/10.1093/ije/30.4.668 PMID: 11511581 12. Williams, Michelle A. COVID-19 and the Social Determinants of Health. Center for Primary Health. Har- vard Medical School. Available from: http://info.primarycare.hms.harvard.edu/blog/covid-social- determinants-health. (Accessed July 18, 2020). 13. Mulholland RH, Sinha IP. Ethnicity and COVID-19 infection: are the pieces of the puzzle falling into place? BMC Med 2020; 18(1):206. https://doi.org/10.1186/s12916-020-01669-9 PMID: 32605617 14. Soltan M, Crowley LE, Melville CR, Varney J, Cassidy S, Mahida R, et al. To what extent are social determinants of health, including household overcrowding, air pollution and housing quality deprivation, modulators of presentation, ITU admission and outcomes among patients with SARS-COV-2 infection in an urban catchment area in Birmingham, United Kingdom?. BMC Public Health, 2020. 15. Niedzwiedz CL, O’Donnell CA, Jani BD, Demou E, Ho FK, Celis-Morales C, et al. Ethnic and socioeco- nomic differences in SARS-CoV-2 infection: prospective cohort study using UK Biobank. BMC Med 2020; 18(1):160. https://doi.org/10.1186/s12916-020-01640-8 PMID: 32466757 16. Cardoso MR, Cousens SN, de Go´ es Siqueira LF, Alves FM, D’Angelo LA. Crowding: risk factor or pro- tective factor for lower respiratory disease in young children? BMC Public Health 2004; 4:19. https://doi. org/10.1186/1471-2458-4-19 PMID: 15176983 17. Nishiura H, Oshitani H, Kobayashi T, Saito T, Sunagawa T. Closed environments facilitate secondary transmission of coronavirus disease 2019 (COVID-19). MedRxiv 2020; https://www.medrxiv.org/ content/10.1101/2020.02.28.20029272v1. 18. Emeruwa UN, Ona S, Shaman JL, Turitz A, Wright JD, Gyamfi-Bannerman C, et al. Associations Between Built Environment, Neighborhood Socioeconomic Status, and SARS-CoV-2 Infection Among Pregnant Women in New York City. JAMA 2020; 324(4):390–392. https://doi.org/10.1001/jama.2020. 11370 PMID: 32556085 19. Greene SK, Levin-Rector A, Hadler JL, Fine AD. Disparities in Reportable Communicable Disease Inci- dence by Census Tract-Level Poverty, New York City, 2006–2013. Am J Public Health 2015; 105(9): e27–34. https://doi.org/10.2105/AJPH.2015.302741 PMID: 26180961 20. Raisi-Estabragh Z, McCracken C, Bethell MS, Cooper J, Cooper C, Caulfield MJ, et al. Greater risk of severe COVID-19 in Black, Asian and Minority Ethnic populations is not explained by cardiometabolic, socioeconomic or behavioural factors, or by 25(OH)-vitamin D status: study of 1326 cases from the UK Biobank. J Public Healt 2020; 42(3):451–460. https://doi.org/10.1093/pubmed/fdaa095 PMID: 32556213 21. de Lusignan S, Joy M, Oke J, McGagh D, Nicholson B, Sheppard J, et al. Disparities in the excess risk of mortality in the first wave of COVID-19: Cross sectional study of the English sentinel network. J Infect 2020:4817. https://doi.org/10.1016/j.jinf.2020.08.037 PMID: 32858068 22. Garg S, Kim L, Whitaker M, O’Halloran A, Cummings C, Holstein R, et al. A. Hospitalization Rates and Characteristics of Patients Hospitalized with Laboratory-Confirmed Coronavirus Disease 2019— COVID-NET, 14 States, March 1–30, 2020. MMWR Morb Mortal Wkly Rep 2020; 69(15):458–464. https://doi.org/10.15585/mmwr.mm6915e3 PMID: 32298251 23. Brown CS, Ravallion M. Inequality and the Coronavirus: Socioeconomic Covariates of Behavioral Responses and Viral Outcomes Across US Counties. NBER Working Papers 27549, National Bureau of Economic Research, Inc, 2020. 24. Mendenhall E, Kohrt BA, Norris SA, Ndetei D, Prabhakaran D. Non-communicable disease syndemics: poverty, depression, and diabetes among low-income populations. Lancet 2017; 389:951–963. https:// doi.org/10.1016/S0140-6736(17)30402-6 PMID: 28271846 25. Hubbard D, Colantonio LD, Tanner RM, Carson AP, Sakhuja S, Jaeger BC, et al. Prediabetes and Risk for Cardiovascular Disease by Hypertension Status in Black Adults: The Jackson Heart Study. Diabetes Care 2019; 42(12):2322–2329. https://doi.org/10.2337/dc19-1074 PMID: 31591089 PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 13 / 15 PLOS ONECrowding conditions and the risk of mortality from COVID-19 26. Raju S, Keet CA, Paulin LM, Matsui EC, Peng RD, Hansel NN, et al. Rural Residence and Poverty Are Independent Risk Factors for Chronic Obstructive Pulmonary Disease in the United States. Am J Respir Crit Care Med 2019; 199(8):961–969. https://doi.org/10.1164/rccm.201807-1374OC PMID: 30384774 27. Townend J, Minelli C, Mortimer K, Obaseki DO, Al Ghobain M, Cherkaski H, et al. The association between chronic airflow obstruction and poverty in 12 sites of the multinational BOLD study. Eur Respir J 2017; 49(6):1601880. https://doi.org/10.1183/13993003.01880-2016 PMID: 28572124 28. Patel JA, Nielsen FBH, Badiani AA, Assi S, Unadkat VA, Patel B, et al Poverty, inequality and COVID- 19: the forgotten vulnerable. Public Health 2020; 183:110–111. https://doi.org/10.1016/j.puhe.2020.05. 006 PMID: 32502699 29. Bello-Chavolla OY, Gonza´lez-Dı´az A, Antonio-Villa NE, Fermı´n-Martı´nez CA, Ma´ rquez-Salinas A, Var- gas-Va´ zquez A, et al. Unequal Impact of Structural Health Determinants and Comorbidity on COVID-19 Severity and Lethality in Older Mexican Adults: Considerations Beyond Chronological Aging. J Gerontol A Biol Sci Med Sci 2021; 76(3):e52–e59. https://doi.org/10.1093/gerona/glaa163 PMID: 32598450 30. Sosa-Rubı´ SG, Seiglie JA, Chivardi C, Manne-Goehler J, Meigs JB, Wexler DJ, et al. Incremental Risk of Developing Severe COVID-19 Among Mexican Patients With Diabetes Attributed to Social and Health Care Access Disadvantages. Diabetes Care 2021; 44(2):373–380. https://doi.org/10.2337/ dc20-2192 PMID: 33208487 31. Ibarra-Nava I, Flores-Rodriguez KG, Ruiz-Herrera V, Ochoa-Bayona HC, Salinas-Zertuche A, Padilla- Orozco M, et al. Ethnic disparities in COVID-19 mortality in Mexico: A cross-sectional study based on national data. PLoS One 2021; 16(3):e0239168. https://doi.org/10.1371/journal.pone.0239168 PMID: 33690607 32. Gutierrez JP, Bertozzi SM. Non-communicable diseases and inequalities increase risk of death among COVID-19 patients in Mexico. PLoS One 2020; 15(10):e0240394. https://doi.org/10.1371/journal.pone. 0240394 PMID: 33031467 33. Direccio´ n General de Epidemiologı´a. Datos abiertos 2021. (Accessed on October 12, 2021) Availabre from: https://www.gob.mx/salud/documentos/datosabiertos. 34. Denova-Gutie´ rrez E, Lopez-Gatell H, Alomia-Zegarra JL, Lo´pez-Ridaura R, Zaragoza-Jimenez CA, Dyer-Leal DD, et al. The Association of Obesity, Type 2 Diabetes, and Hypertension with Severe Coro- navirus Disease 2019 on Admission Among Mexican Patients. Obesity 2020; 28 (10):1826–1832. https://doi.org/10.1002/oby.22946 PMID: 32610364 35. Emilio Gutie´ rrez, Rubli A, Tavares T. Delays in death reports and their implications for tracking the evo- lution of COVID-19. COVID Economics 2020; 34:116–144. 36. Consejo Nacional de Poblacio´ n, CONAPO (2020). I´ndice de marginacio´ n (carencias poblacionales) por localidad, municipio y entidad [I´ndice de marginacio´ n por municipio]. Available from: https://datos.gob. mx/busca/dataset/indice-de-marginacion-carencias-poblacionales-por-localidad-municipio-y-entidad. 37. Consejo Nacional de Evaluacio´n de la Polı´tica de Desarrollo Social, CONEVAL (2020). Medicio´n de la pobreza. Available from: https://www.coneval.org.mx/coordinacion/entidades/Paginas/Informes_ Pobreza_Evaluacion_2020.aspx (accessed October 10, 2021). 38. Austin PC. A Tutorial on Multilevel Survival Analysis: Methods, Models and Applications. Int Stat Rev 2017; 85(2):185–203. https://doi.org/10.1111/insr.12214 PMID: 29307954 39. Bello-Chavolla OY, Bahena-Lo´pez JP, Antonio-Villa NE, Vargas-Va´zquez A, Gonza´ lez-Dı´az A, Ma´r- quez-Salinas A, et al. Predicting mortality due to SARS-CoV- 2: A mechanistic score related obesity and diabetes to COVID-19 outcomes in Mexico. J Clin Endocrinol Metab 2020; 105:dgaa346. https:// doi.org/10.1210/clinem/dgaa346 PMID: 32474598 40. Krieger N, Waterman PD, Chen JT. COVID-19 and Overall Mortality Inequities in the Surge in Death Rates by Zip Code Characteristics: Massachusetts, January 1 to May 19, 2020. Am J Public Health 2020:e1–e3. https://doi.org/10.2105/AJPH.2020.305913 PMID: 33058698 41. Singleton R, Salkoski AJ, Bulkow L, Fish C, Dobson J, Albertson L, et al. Housing characteristics and indoor air quality in households of Alaska Native children with chronic lung conditions. Indoor Air 2017; 27:478–486. https://doi.org/10.1111/ina.12315 PMID: 27317363 42. 43. 44. Lin EJ, Sun M, Choi EY, Magee D, Stets CW, During MJ. Social overcrowding as a chronic stress model that increases adiposity in mice. Psyc honeuroendocrinology 2015; 51:318–30. https://doi.org/ 10.1016/j.psyneuen.2014.10.007 PMID: 25462904 Fielding-Miller RK, Sundaram ME, Brouwer K. Social determinants of COVID-19 mortality at the county level. PLoS One 2020; 15(10):e0240151. https://doi.org/10.1371/journal.pone.0240151 PMID: 33052932 de Souza CDF, Machado MF, do Carmo RF. Human development, social vulnerability and COVID-19 in Brazil: a study of the social determinants of health. Infect Dis Poverty 2020; 9(1):124. https://doi.org/ 10.1186/s40249-020-00743-x PMID: 32867851 PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 14 / 15 PLOS ONECrowding conditions and the risk of mortality from COVID-19 45. Adhikari S, Pantaleo NP, Feldman JM, Ogedegbe O, Thorpe L, Troxel AB. Assessment of Community- Level Disparities in Coronavirus Disease 2019 (COVID-19) Infections and Deaths in Large US Metro- politan Areas. JAMA Netw Open 2020; 3(7):e2016938. https://doi.org/10.1001/jamanetworkopen.2020. 16938 PMID: 32721027 46. Douglas M, Katikireddi SV, Taulbut M, McKee M, McCartney G. Mitigating the wider health effects of covid-19 pandemic response. BMJ 2020; 369:m1557. https://doi.org/10.1136/bmj.m1557 PMID: 32341002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0264137 February 22, 2022 15 / 15 PLOS ONE
10.1371_journal.pone.0262773	RESEARCH ARTICLE Comparison of direct and inverse methods for 2.5D traction force microscopy Johannes W. Blumberg, Ulrich S. SchwarzID* Heidelberg University, Institute for Theoretical Physics and Bioquant, Heidelberg, Germany * schwarz@thphys.uni-heidelberg.de Abstract Essential cellular processes such as cell adhesion, migration and division strongly depend on mechanical forces. The standard method to measure cell forces is traction force micros- copy (TFM) on soft elastic substrates with embedded marker beads. While in 2D TFM one only reconstructs tangential forces, in 2.5D TFM one also considers normal forces. Here we present a systematic comparison between two fundamentally different approaches to 2.5D TFM, which in particular require different methods to deal with noise in the displacement data. In the direct method, one calculates strain and stress tensors directly from the dis- placement data, which in principle requires a divergence correction. In the inverse method, one minimizes the difference between estimated and measured displacements, which requires some kind of regularization. By calculating the required Green’s functions in Fourier space from Boussinesq-Cerruti potential functions, we first derive a new variant of 2.5D Fou- rier Transform Traction Cytometry (FTTC). To simulate realistic traction patterns, we make use of an analytical solution for Hertz-like adhesion patches. We find that FTTC works best if only tangential forces are reconstructed, that 2.5D FTTC is more precise for small noise, but that the performance of the direct method approaches the one of 2.5D FTTC for larger noise, before both fail for very large noise. Moreover we find that a divergence correction is not really needed for the direct method and that it profits more from increased resolution than the inverse method. Introduction Mechanical forces are important for a wide range of essential cellular processes such as cell adhesion, migration and division [1]. The new field of mechanobiology has evolved around the role of forces in biological systems [2]. However, in general forces cannot be observed directly, but have to be measured by their effect to create deformation and flow. The key idea to measure cell forces therefore is to use a calibrated material as a measurement devise. The simplest approach is to replace the glass or plastic substrates commonly used for cell culture by soft elastic substrates and to measure their deformations by tracking embedded marker beads. This method is known as traction force microscopy (TFM) and is widely used in cell biology and biophysics [3–8]. Depending on whether one aims at reconstructing only tangential or also normal forces, one speaks of 2D TFM or 2.5D TFM, respectively [8]. If one goes beyond a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Blumberg JW, Schwarz US (2022) Comparison of direct and inverse methods for 2.5D traction force microscopy. PLoS ONE 17(1): e0262773. https://doi.org/10.1371/journal. pone.0262773 Editor: Jose Manuel Garcia Aznar, University of Zaragoza, SPAIN Received: June 23, 2021 Accepted: January 4, 2022 Published: January 20, 2022 Copyright: © 2022 Blumberg, Schwarz. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The traction force microscopy code developed for this project is available at https://github.com/usschwarz/ DirectMethod. Funding: This research has been funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy via the Excellence Cluster 3D Matter Made to Order (EXC-2082/1 – 390761711). JWB thanks the Carl Zeiss Foundation for financial support. The funders had no role in study design, data collection PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 1 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. planar substrates and reconstructs cell forces applied to a fibrous marix surrounding the cell, then one speaks of 3D TFM [7–10]. Recently the concept of measuring cell forces on the sur- face of calibrated material has been extended beyond planar substrates by using elastic beads deformed by cell traction [11–13]. Recalling the concepts of continuum mechanics, in principle it is straight-forward to calcu- late cell traction from displacement data. Given the three-dimensional displacement field, one can first calculate the strain tensor, which is its derivative. Using the constitutive law of the material, it can be converted into a stress tensor. For linear isotropic material, this is simply a linear transformation. Finally the surface traction follows as contraction of the stress tensor with the surface normal. This workflow is known as the direct (or forward) method (DM). It has been pioneered for 2.5D TFM [14–16], but also has been applied to 3D TFM [17, 18] and to elastic beads with embedded markers that are deformed by the traction forces in cell aggre- gates [11]. The DM is especially suited to deal with large deformations and non-linear material laws, but it also can be used in the linear regime, which is typically used for 2D or 2.5D TFM on soft elastic substrates. However, its use is not very common in this field, for two main rea- sons. First the DM requires a 3D image volume in order to be able to reconstruct the full stress tensor, which is challenging given the anisotropic point spread functions of standard micro- scopes and the time required to acquire image stacks. Planar substrates are more compatible with imaging in 2D focal planes rather than 3D image volumes. Secondly, it is a numerical challenge to calculate the required derivatives in the presence of the unavoidable experimental noise, especially when microscope resolution is low. Rather than using the direct method, on soft elastic substrates it is common practice to use the inverse method that was conceived by the pioneers of 2D TFM [19, 20]. In order to avoid derivatives, one does not explicitly calculate strain or stress tensors, but stays on the level of displacement fields. By minimizing the difference between the measured displacement field and a displacement field calculated from an estimated traction field, one arrives at the best esti- mate given the experimental data. Because elasticity theory leads to an ill-posed inverse prob- lem due to its long-ranged deformation fields, one usually deals with the noise problem by invoking some regularization procedure, e.g. zero-order Tikhonov regularization [4, 8, 20, 21]. In order to choose the correct value of the regularization parameter, different schemes have been suggested, including the L-curve criterion or generalized cross-validation [22, 23]. The need for explicit regularization can be avoided by using TFM-schemes that effectively filter the deformation data, like image smoothing [24, 25] or the Finite Element Method (FEM) [17, 18, 26]. While the inverse method avoids the calculation of derivatives, it requires to repeatedly cal- culate deformation from estimated traction (direct problem). This can be implemented with different methods, but the two most common ones are Green’s functions (GFs) and FEM. GFs can be used only for linear elasticity, but offer several advantages. First, the GF for thick elastic substrates is well known (Boussinesq solution) [27] and also the GF for a finite thickness sub- strate has been derived [28]. Second, for flat cells on planar substrates it is sufficient to know the displacements in a two-dimensional plane close to the gel surface. This reduces the Green’s function from a 3x3 to a 2x2 matrix. And third, one can use fast Fourier methods to convert the convolution of GF and traction into a product. Due to its speed, Fourier Transform Trac- tion Cytometry (FTTC) [24] therefore has become the method of choice for high resolution measurements. Moreover regularization can be formulated in Fourier space and therefore many schemes can be applied to deal with the noise issue [23, 29, 30]. Together, these advances make the inverse method very attractive for measuring cell traction on soft elastic substrates. If cell forces in the third dimension are also of interest (2.5D TFM), one often switches to imple- mentations that use FEM to solve the direct problem [31]. Alternatively, an analytical scheme PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 2 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy has been proposed for 2.5D FTTC before, which however is rather complicated because it has been derived for substrates with finite thickness [28]. In summary, TFM has become a very diverse field with a mix of different approaches, each of which have their advantages and disadvantages in a certain context. Very rarely however are these different methods directly compared against each other. A notable exception is a recent work that compares FEM-based implementations of the direct and inverse methods for 3D TFM [17, 18]. Here we aim at a similar comparison, but for 2.5D TFM and with GF-based methods. Rather than simulating experimental setups, we use simple test cases and simulations with displacement noise to provide a comprehensive comparison of the mathematical proper- ties of direct versus inverse methods for 2.5D TFM. In addition to this interest in fundamental questions of TFM, our work is also motivated by different recent experimental developments. First, the DM seems to be an attractive choice for non-planar geometries like elastic beads, for which it is very challenging to calculate appropri- ate GFs [11]. We expect that this line of research will become more important in the future with the promise of additive manufacturing to print 3D elastic material that is compatible with cell culture and deforms under cell traction [32–35]. Such systems might by approached best with FEM-approaches, but in some cases (like elastic beads) GF-based inverse methods are possible [12, 36]. Here we use this recent development as a motivation to compare direct and inverse methods in the traditional setup of 2.5D TFM for planar substrates. Second, new microscopy methods have been used to achieve better image resolution for 2.5D TFM, including Stimulated Emission Depletion (STED) microscopy [37], Structured Illumination Microscopy (SIM) [26, 38], astigmatic SIM [39] and fluctuation-based TFM [40]. The displacement data resulting from these experimental advances is often analyzed with one of the different TFM-methods, but a systematic comparison between different methods is usu- ally not performed. In this context, it is interesting to know how direct and inverse methods compare in regard to improvements in sampling density. Here we use this recent development as a motivation to also study the effect of varying sampling distance. Our work is structured as follows. We first provide an in-depth introduction to direct and inverse methods for 2.5D TFM. We use Boussinesq-Cerruti potential functions to derive a novel and very efficient version of the inverse method 2.5D FTTC. For the 2.5D DM, we dis- cuss different schemes for numerically calculating the required derivatives and introduce a divergence correction motivated by similar schemes from hydrodynamics. We define several simple test cases for force reconstruction based on analytical solutions for Hertz-like adhesion patches. To test the robustness of our different methods, we simulate different levels of dis- placement noise, which is a common method to lump noise that might originate from optical microscopy or gel preparation into one parameter. In addition, we investigate the effect of varying sampling distance. Our results show that 2.5D FTTC usually performs better than the 2.5D DM, but that the performance of the DM approaches the one of FTTC for larger noise, before both fail at very large noise. We also find that the divergence correction for the DM is not needed and that DM profits more from increased sampling density. Theory and methods Inverse versus direct methods The standard workflow of a TFM analysis for planar elastic substrates is outlined in Fig 1A. Deformations are usually determined by comparing the situation in which the cell has fully adhered to the substrate with a reference situation in which the substrate is relaxed and the cells exert no forces, e.g. by removing or relaxing the cell. The state of the substrate can also be observed at multiple time steps to observe the temporal evolution of the cell-to-substrate PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 3 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy Fig 1. General overview of traction force microscopy. (A) By determining with image processing the movement of marker beads in the substrate due to cell forces, a deformation field is obtained. Traction force microscopy (TFM) estimates the cellular traction field from the displacement data. (B) Here we discuss two fundamentally different methods for TFM. The inverse method estimates a force distribution that results in an optimal match for the global displacement field. In Fourier Transform Traction Cytometry (FTTC) one makes use of fast Fourier transforms. A regularization scheme is introduced to address the fact that the inverse problem is ill-posed. In the direct method, point-wise computational methods are used to determine the stress tensor locally. A divergence correction can be applied to ensure that the physical force balance @i σij = 0 is satisfied. https://doi.org/10.1371/journal.pone.0262773.g001 adhesion [41]. Deformations in the substrate are commonly monitored with the help of fluo- rescent marker beads, whose change in position can be observed in the microscope. The bead displacement is obtained using common tracking and correlation techniques [3, 4, 42]. Next, traction force routines are employed to estimate the traction force vector field from the dis- placement vector field. Here we compare two methods as shown in Fig 1B. The traditional approach introduced by the pioneers of 2D TFM is the inverse method (top), which minimizes the difference between the experimental and estimated displacement fields. The direct method (bottom) cannot be used in a purely 2D setup, but requires 3D image data. In this method, traction is calculated directly from the displacement field by differentiation and (linear) trans- formation between strain and stress. For both methods, elements of elasticity theory are required. Elasticity theory Traction forces are forces acting on the boundary of elastic solids. They are quantified by a three-component vector field τ, which is defined for all points at a surface and describes the area density of the force. Its component normal to the surface describes the pressure balance between the solid and its surroundings. In contrast, the tangential components describe shear forces. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 4 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy The stress tensor σ describes the force per area acting on any (real or fictitious) surface of the system. It is defined by the relation: τS ¼ σnS : ð1Þ Here, the left hand side represents the surface force density for a given surface S. The vector nS is the unit normal vector of S. In the case of a solid which is delimited by a planar surface at the z = 0 plane and whose outwards normal is defined in negative z-direction, the traction stress reads: τ ¼ (cid:0) ðs13; s23; s33Þjz¼0 : ð2Þ This means that by determining the stress tensor in a region close to the surface, the surface traction is known. The different entries to the stress tensor are not independent physical quantities, because they have to obey force and torque balances. The stresses lead to material movement that is described by the deformation gradient tensor F, which is the Jacobian of the coordinate trans- formation between deformed and undeformed configurations. Alternatively the movement can be described by the deformation field u, which is related to the deformation gradient ten- sor by Fij ¼ dij þ ∂ui ∂xj : ð3Þ Note that the derivatives must be taken with respect to the coordinates of the reference con- figuration (Lagrange frame). Stresses within the material and changes in its configuration are related by a constitutive equation σ = σ(F), which is specific to the material in question (material law). For relatively stiff homogeneous and isotropic materials, a linear approximation can be used that is given by � sij ¼ E 2ð1 þ nÞ Fij þ Fji þ 2nðtrðFÞ (cid:0) 1Þ (cid:0) 2 1 (cid:0) 2n dij � or in terms of the displacement field sij ¼ E 2ð1 þ nÞ ∂ui ∂xj þ ∂ui ∂xj þ 2n 1 (cid:0) 2n ðruÞdij : ! ð4Þ ð5Þ The material constants E and ν are known as the Young’s modulus and the Poisson’s ratio, respectively, and describe the stiffness and compressibility of the material. Note that the differ- ence between the coordinates of the reference configuration (Lagrange frame) and the coordi- nates describing positions in the Euclidean space (Euler frame) are neglected in this linear approximation. Whether a substrate is considered stiff or soft is determined by the relation between the typ- ical traction force amplitude and the Young’s modulus. The dimensionless deformation gradi- ent will typically be in the same magnitude as the ratio σij/E. We must consider the substrate to be soft if we expect the magnitude of the traction forces to be much larger than the substrates Young’s modulus. Then the linear approximation as given here does not necessarily apply any- more and might has to be replaced by a more complicated (non-linear) mapping. Typical sub- strate materials used for TFM are polyacrylamide (PAA) or polydimethlysiloxane (PDMS), that can be created for a wide range of stiffnesses. For TFM-experiments, a typical value is 10 kpa, which according to the above considerations is a rather stiff substrates assuming typical PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 5 / 25 PLOS ONE Comparison of direct and inverse methods for 2.5D traction force microscopy traction stresses in the range of 1 kpa. Therefore the linear relation is typically satisfied. Both types of substrates are usually considered to be close to incompressible, with a Poisson ratio close to 1/2. It should be noted that incompressibility implies ru = 0 in the linear case assumed in Eq (5). This counteracts the apparent divergence for ν ! 1/2 that would otherwise occur in the last term of Eq (5). Assuming small deformations and a linear material law, the traction forces can be related to deformations directly using a convolution relation: uðx; y; zÞ ¼ Z S Gðx (cid:0) x0; y (cid:0) y0; zÞ � τðx0; y0Þdx0 dy0; ð6Þ where the coordinate system is chosen in such a way that the traction stresses are exerted on the z = 0 plane and the substrate is confined to the z > 0 halfspace. The GF G is known analyti- cally for an elastic halfspace (Boussinesq solution) [27]: Gðx; y; zÞ ¼ 2ð1 (cid:0) nÞr þ z rðr þ zÞ þ ð2rðnr þ zÞ þ z2Þx2 r3ðr þ zÞ2 ð2rðnr þ zÞ þ z2Þxy r3ðr þ zÞ2 ð2rðnr þ zÞ þ z2Þxy r3ðr þ zÞ2 2ð1 (cid:0) nÞr þ z rðr þ zÞ þ ð2rðnr þ zÞ þ z2Þy2 r3ðr þ zÞ2 ð1 (cid:0) 2nÞx rðr þ zÞ þ xz r3 p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi x2 þ y2 þ z2 ð1 (cid:0) 2nÞy rðr þ zÞ þ yz r3 . Cases where the substrate is not sufficiently thick and inter- where we used r ¼ actions at the bottom surface of a finite-thickness substrate can not be neglected have also been studied [28]. In these cases the convolution relation is still valid, but a different GF has to be used. Note that the 1/r-dependence of this GF is the reason why the inverse problem is ill- posed. ; ð7Þ 1 C C C C C C C C C C C C A xz r3 yz r3 (cid:0) (cid:0) ð1 (cid:0) 2nÞx rðr þ zÞ ð1 (cid:0) 2nÞy rðr þ zÞ 2ð1 (cid:0) nÞ r þ z2 r3 2pE 1 þ n 0 B B B B B B B B B B B B @ 2.5D Fourier Transform Traction Cytometry (FTTC) FTTC is the most widely used inverse method and uses the fact that the convolution integral from Eq (6) becomes a product in Fourier space and that fast Fourier transforms allow one to quickly switch between real and Fourier space [24, 29, 37, 43]. To use it, the displacement field u(x) must first be interpolated onto a regular, rectangular grid covering the whole image, either on the surface or in a fixed depth z = h within the substrate. We will designate the values at the sample points xij by uij in the following. Then the traction force τ(x, y) is described by a set of plane waves ^f mnðx; yÞ τðx; yÞ ¼ XNx XNy m¼0 n¼0 ^τ mn ^f mnðx; yÞ ; ^f mnðxÞ ¼ 1 NxNy eikmn�x : ð8Þ The wave vectors kmn are chosen in accordance with the sampling grid. This choice ensures that the expansion coefficients ^τ mn are in fact the 2D discrete Fourier transform of the traction sampled at the xij. Due to the Fourier convolution theorem, Eq (6) reduces to PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 ^umn ¼ ~GððkmnÞx; ðkmnÞy; hÞ � ^τ mn; ð9Þ 6 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy where ^umn and uij are also related by a 2D discrete Fourier transform. The term ~Gðkx; ky; zÞ describes the function obtained from applying a continuous Fourier transform in the two tan- gential directions on the real space GF. For 2D FTTC, one uses only the planar part at z = 0 [4, 21, 23, 24]: 0 ð1 (cid:0) nÞk2 x þ nk2 y (cid:0) nkxky ~G2Dðkx; kyÞ ¼ @ 1 mk3 (cid:0) nkxky ð1 (cid:0) nÞk2 y þ nk2 x 1 A: ð10Þ Here we used k ¼ q ffiffiffiffiffiffiffiffiffiffiffiffiffiffi x þ k2 k2 y and μ = E/2(1 + ν). Note that this result differs in an important minus sign in the off-diagonal elements from the one given in the original publication on FTTC [24]. 2.5D FTTC was first introduced by [28] for thin substrates using a rather lengthy expression to derive a stress-strain relation. Here, we assume a thick substrate and present a more straight forward calculation to derive a closed form solution for the GF in Fourier spaces, that is consis- tent with the 2D formula presented in (10) and effectively gives the same results as the proce- dure given earlier in the context of a finite thickness halfspace [28]. The general problem of finding the deformation field in an infinite halfspace with surface traction as boundary condition and no internal forces follows from Eq 5 as @ @xj @ui @xj þ @uj @xi þ 2n 1 (cid:0) 2n ! ruð Þdij ¼ 0; (cid:0) sizjz¼0 ¼ ti: ð11Þ The GF in real space is known as described by Eq 6 with the Boussinesq equation [27]. By making use of the convolution theorem, Eq 6 can be reduced into the simple product expres- sion ~uiðkx; ky; zÞ ¼ X3 j¼1 ~G ijðkx; ky; zÞ~qjðkx; kyÞ: ð12Þ Here a tilde above the quantities represents the Fourier transform along the x and y axis. Unfortunately, finding an analytical expression for ~G ij directly from transforming the real space Boussinesq solution would require us to solve a complex integral expression. Instead of doing so, we will derive the relation between ~τ and ~u directly from the boundary problem stated in Eq 11. We can then extract the analytical expression for ~G ij by comparing this relation to Eq 12. In general, our procedure follows the same steps as described in Ref [27] to derive the real space GF in the first place. We start by introducing a set of harmonic potential functions that solve the boundary value problem stated by: r2Pi ¼ 0 ; @3 z Pijz¼0 ¼ (cid:0) ti : ð13Þ These Boussinesq-Cerruti potential functions have been described e.g. in Refs. [44, 45]. A solution of Eq (11) is now given by PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 ui ¼ X3 uðjÞ i ; j¼1 ð14Þ 7 / 25 PLOS ONE Comparison of direct and inverse methods for 2.5D traction force microscopy together with the x-tangential contributions 2muðxÞ x ¼ 2n@2 xPx þ 2@2 z Px (cid:0) z@2 x@zPx 2muðxÞ y ¼ 2n@x@yPx (cid:0) z@x@y@zPx 2muðxÞ z ¼ ð1 (cid:0) 2nÞ@x@zPx (cid:0) z@x@2 z Px; the y-tangential contributions 2muðyÞ x ¼ 2n@x@yPy (cid:0) z@x@y@zPy 2muðyÞ y ¼ 2n@2 yPy þ 2@2 z Py (cid:0) z@2 y@zPy 2muðyÞ z ¼ ð1 (cid:0) 2nÞ@y@zPy (cid:0) z@y@2 z Py and the normal (z) contributions 2muðzÞ x ¼ (cid:0) ð1 (cid:0) 2nÞ@x@zPz (cid:0) z@x@2 z Pz 2muðzÞ y ¼ (cid:0) ð1 (cid:0) 2nÞ@y@zPz (cid:0) z@y@2 zPz 2muðzÞ z ¼ 2ð1 (cid:0) nÞ@2 z Pz (cid:0) z@3 z Pz: ð15Þ ð16Þ ð17Þ By applying the mentioned two-dimensional Fourier transform to Eq (13), the different modes decouple and we obtain the following initial value problem: @2 z ~P iðkx; ky; zÞ (cid:0) ðk2 x þ k2 yÞ~Piðkx; ky; zÞ ¼ 0 ; @3 z ~P ijz¼0 ¼ (cid:0) ~ti : ð18Þ Because the differential equation is linear, it can easily be solved and we find for the Fourier transform of the potential functions: ~Piðkx; ky; zÞ ¼ q p 1 ffiffiffiffiffiffiffiffiffiffiffiffiffiffi x þ k2 k2 y 3 e(cid:0) ffiffiffiffiffiffiffiffi x þk2 k2 y z~tiðkx; kyÞ: In the next step, we apply the Fourier transform onto Eqs (14) to (17) and obtain: ~ui ¼ X3 ~uðjÞ i ; j¼1 together with the x-tangential contributions 2m~uðxÞ x ¼ (cid:0) 2nk2 xPx þ 2@2 z Px þ k2 xz@zPx 2m~uðxÞ y ¼ (cid:0) 2nkxkyPx þ kxkyz@zPx 2m~uðxÞ z ¼ ikxð1 (cid:0) 2nÞ@zPx (cid:0) ikxz@2 z Px; the y-tangential contributions 2m~uðyÞ x ¼ (cid:0) 2nkxkyPy (cid:0) z@zPy 2m~uðyÞ y ¼ (cid:0) 2nk2 yPy þ 2@2 z Py þ k2 yz@zPy 2m~uðyÞ z ¼ ikxð1 (cid:0) 2nÞ@zPy (cid:0) ikxz@2 z Py PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 ð19Þ ð20Þ ð21Þ ð22Þ 8 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy and the normal (z) contributions 2m~uðzÞ x ¼ (cid:0) ð1 (cid:0) 2nÞikx@zPz (cid:0) ikxz@2 z Pz 2m~uðzÞ y ¼ (cid:0) ð1 (cid:0) 2nÞiky@zPz (cid:0) ikyz@2 z Pz ð23Þ 2m~uðzÞ z ¼ 2ð1 (cid:0) nÞ@2 z Pz (cid:0) z@3 zPz: We can now insert Eq (19) into these equations. This results in a simple product expression relating ~τ to ~u. By comparing the result to Eq (12) we can extract the following Greens func- tion in Fourier space: ~Gðkx; ky; zÞ ¼ 0 e(cid:0) kz 2mk3 B B B B B @ 2k2 (cid:0) ð2n þ kzÞk2 x (cid:0) ð2n þ kzÞkxky ð1 (cid:0) 2n (cid:0) kzÞikkx (cid:0) ð2n þ kzÞkxky 2k2 (cid:0) ð2n þ kzÞk2 y ð1 (cid:0) 2n (cid:0) kzÞikky (cid:0) ð1 (cid:0) 2n þ kzÞikkx (cid:0) ð1 (cid:0) 2n þ kzÞikky 2ð1 (cid:0) nÞk2 þ k3z 1 C C C C C A : ð24Þ The analytical form of the GF given here allows us to perform 2.5D FTTC in a very efficient manner. We also checked both by analytically taking the limit of an infinitely thick substrate and by numerical comparison that our method agrees with the one described earlier in the context of a finite thickness elastic halfspace [28]. Inverse problem and regularization In principle, a suitable traction field can be found simply by calculating the inverse of ~GððkmnÞx; ðkmnÞy; hÞ and applying the resulting matrix to ^umn to get the traction field ^τ mn. How- ever, such a procedure ignores the important fact that the inverse problem of elasticity is ill- posed because elastic effects are long-ranged, meaning that local changes in traction will have non-negligible effects on the displacement even over large distances. This is a problem because the input data for the displacement field will be always subject to experimental noise due to limitations in resolution or inhomogeneities in the medium [4]. A naive inversion will try to reproduce all the fine details of the input field by changing global properties of the force field; this ill-posed nature of the inverse problem will be reflected by a large condition number of the inverse matrix. This problem can be addressed either by filtering the displacement data, e.g. by image smoothing [24, 25], or by introducing a regularization scheme [21]. Because here we focus on the mathematical aspects of TFM, which should be kept sepa- rately from issues of image processing, we refrain from smoothing procedures and use a 0th order Tikhonov regularization, which has been found earlier to be very appropriate for the kind of TFM-procedures discussed here [29]. With our regularization approach, Eq (6) leads to τðx; yÞ ¼ argmin s � � � � Z S Gðx (cid:0) x0; y (cid:0) y0; zÞ � sðx0; y0Þdx0dy0 (cid:0) uðx; yÞ � 2 � � � 2 þl2 sðx; yÞ k k2 2 : ð25Þ Eq (9) now leads to � ^τ mn ¼ ð ~Gy ~G þ l2IÞ(cid:0) 1 ~Gy � ^umn≕ ^G# l;mn^umn: PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 ð26Þ 9 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy Here the superscript † designates the Hermitian conjugate. In addition, ~GððkmnÞx; ðkmnÞy; hÞ has been denoted simply by ~G for visual clarity. The regularization parameter λ must be chosen with care [23, 30]. If the value is chosen too large, this will result in a loss of accuracy and resolution in the resulting force map. If it is cho- sen too small, the result will be dominated by noise [22]. Here we use a generalized cross vali- dation technique (GCV) [46] for finding a regularization parameter efficiently. The estimation function is defined by GðlÞ ¼ k ~G^τ l (cid:0) ^uk2 ðtrð1 (cid:0) ~G ~G#ÞÞ2 2 ð27Þ and always has a minimum for a strictly positive value of λ, at which an optimal regularization can be found. It can easily be calculated for a large number of values using the singular value decomposition of ~G which is known. Using the determined value for λ, an estimate for the deformation field can be calculated. Direct method and divergence correction The direct method for TFM uses the constitutive equation to calculate the local stress tensor of the material from the displacement derivatives or the deformation gradient tensor (compare Fig 1B). The surface traction can be determined from the stress tensor using Eqs (1) and (2). Here we apply this method to the same setup as commonly used with the inverse method. How- ever, the direct method is fundamentally a 3D method and the deformation must be known not only in a single plane, but in a volume below the surface, in order to obtain the full stress tensor. In contrast to the inverse method based on GFs, the direct method can more easily be extended to non-planar surfaces (going beyond Eq (2)) and non-linear material (going beyond (4)). The strains @ui/@xj and the components Fij of the deformation gradient tensor must be obtained numerically. To do so, the displacement field u is first sampled on a regularly spaced 3D grid. Then both quantities can be easily obtained using a finite difference scheme. Frank et al [15, 47] determine the derivative by fitting a 1st order polynomial uðxÞ ¼ ax þ by þ cz þ d ð28Þ to a local region in the resulting traction profile. The components of a, b and c contain the strains @ui/@xj directly. We refer to this technique as the 3x3x3 patch method as a 3 by 3 by 3 data point support is used to estimate the deformation field gradient. We investigate how well this approach performs in comparison to other approximation techniques. The most simple alternative is a simple two-point finite difference method (only the equation for z-derivatives is stated): � � � � @ui @z klm ui;kl mþ1 ð � Þ (cid:0) ui;kl m(cid:0) 1 2Dx ð Þ : ð29Þ Increasing the number of sampling points contributing to the derivatives as done in Eq (28) should in theory decrease the uncertainty of the result, but will decrease resolution. Another alternative to this constitutes a four-point scheme: � � � � @ui @z klm � (cid:0) ui;klðmþ2Þ þ 8ui;klðmþ1Þ (cid:0) 8ui;klðm(cid:0) 1Þ þ ui;klðm(cid:0) 2Þ 12Dx : PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 ð30Þ 10 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy In all three cases special non-symmetric expressions are used close to the boundary that make use of the same number of support points, but avoid contributions outside the observed area. All methods for numerical derivatives are dependent on the sampling distance defined by the sampling lattice. A smaller sampling distance should reduce the systematic error, but on the other hand the statistical error on the deformation gradient is proportional to the ratio between the statistical error of the displacement field and the spacing and therefore will increase for smaller sampling distance. While the inverse method always gives a valid displacement field because it is calculated from a force distribution as a direct problem, the direct method calculates a displacement field that might not be valid as it violates force and torque balances. In a static system, the force bal- ance leads to the Cauchy momentum equation [45] (here stated in the form used in linear elas- ticity): ∂sij ∂xj ¼ 0 : ð31Þ In other words the stress tensor must be divergence-free (solenoid). This property is not restricted to a specific geometry of the system. It describes a fundamental property resulting from the fact that the stress tensor represents local force densities that must be in balance for a static situation free of external forces. The coordinate of the initial deformation fields where the corresponding stress tensor field satisfies Eq (31) are called compatible [45]. The impor- tance of this condition has been pointed out before in the context of monolayer stress micros- copy [48] and 3D TFM [17]. While the physical deformation field u(T) is always compatible, the measured deformation field u(C) is not, as it contains errors attributed to noise and thus the obtained stress tensor may not satisfy Eq (31). If we describe σij(u) to be the stress tensor obtained from a deforma- tion field u, we easily see by @sijðuðCÞÞ @xj ¼ @ðsijðuðCÞÞ (cid:0) sijðuðTÞÞÞ @xj þ Þ @sij u Tð Þ ð @xj \|{z} ¼0 ð32Þ that only the noise part of the deformation field will contribute to the divergence of the obtained stress tensor σij(u(C)). This means that we can use this divergence to determine what ratio of the input data is attributed to noise. The divergence of the calculated stress tensor σij(u(C)) can be calculated on a local basis by using a symmetric two-point form. Instead of the tensor based formulation Eq (31), the compatibility equation can also be rep- resented as a condition for each column of the stress tensor: r � a ¼ 0; r � b ¼ 0; r � c ¼ 0; ð33Þ where we defined a ¼ ^e1σ, b ¼ ^e2σ and c ¼ ^e3σ. This observation immediately generates a relation to hydrodynamics, because a similar situation can be found for hydrodynamic descriptions of incompressible fluids, where the mass conservation equation reduces to r � v ¼ 0 ð34Þ for a flow field v. This property has been exploited by various studies and using a variety of techniques to remove potential noise from v [49–51]. Because of the relation between the stress tensor and the surface traction Eq (1), removing the noise contributions from σij for the full substrate might also yield a better result for the surface traction, similar to regularization for PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 11 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy the inverse method. An efficient method for performing this kind of divergence correction on a flow profile has been described by Wang et al. [52] and this method can be generalized to any divergence-free vector field. A short description of their technique is given in the S1 Appendix. This technique makes use of the particular structure of the problem to result in an algorithm that applies matrix operations only on one coordinate at the same time, which strongly reduces computational complexity as well as avoids complex operations like matrix inversion. The noise removal technique now consists of an iterative process. We start with the initial condition σ(0) ≔ σexp and i ≔ 0 and iterate the following steps: 1. Split the stress tensor into column vectors: (aexp, bexp, cexp) = σ(i−1) 2. Find ac, bc and cc that satisfy Eq (33) using the mentioned procedure for vector fields using aexp, bexp and cexp as input. 3. Reassemble the vector fields into a new tensor w = w(i) ≔ (aexp, bexp, cexp). 4. Obtain a symmetric approximation for the stress tensor σ(i) = 1/2(w + wT). In each iteration, the divergence of the resulting stress tensor σ(i) is reduced. We tested this reduction for a number of different inputs, including a white noise input, and found that a fixed iteration of 20 cycles was sufficient to remove the noise of the input (compare S1 Appendix). Results Design of simulated traction patterns Because here we aim at testing different traction force reconstruction methods, we design trac- tion patterns that are useful for the task at hand and for which we can calculate the deforma- tions analytically. We then add displacement noise to these solutions and finally reconstruct the traction and compare with the original pattern. This process is illustrated in Fig 2 for an example that includes the two most important features known from adherent cells, namely tangential traction at focal adhesions and the normal push by the nucleus (which due to Fig 2. Workflow for reconstruction for cell-like traction pattern. (A) As an introductory example, we include the two most important features of adherent cells, namely focal adhesions with mainly tangential traction and the normal push of the nucleus into the substrate. The plots in the upper row show the tangential components, while the lower row contains the normal component. (B) From this given traction field one can then calculate the analytical solution for the deformation field. (C) Noise is added to simulate experimental data (here σN/< kuk > = 0.2). (D) For this low noise level, 2.5D FTTC works very well. (E) The direct method gives similar results for the tangential tractions, but performs less well for normal traction. Details on the simulation parameters for this profile can be found in the S1 Appendix. https://doi.org/10.1371/journal.pone.0262773.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 12 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy momentum conservation has to be balanced by counteracting normal forces at the focal adhe- sions). We discuss the properties of the different reconstructions in the subsequent section. For the analytically tractable patterns, we choose linear combinations of Hertz-like patches: τPatch ¼ q 3 2pa3 Q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi a2 (cid:0) ðx (cid:0) x0Þ2 (cid:0) ðy (cid:0) y0Þ2 Yða2 (cid:0) ðx (cid:0) x0Þ2 (cid:0) ðy (cid:0) y0Þ2Þ; ð35Þ where the force transmitted by the patches is represented by Q, the adhesion size a and the positions x0 and y0 are parameters that can be chosen freely, and Θ is the Heaviside function. We call these traction profiles Hertz-like because they match the profiles in the Hertz and Cer- rutti contact problems. The traction fields for this problem have been extensively studied for materials that obey linear elasticity [53]. Based on an approach suggested by [54], we derive the full 3D solution for Eq (35) in the S1 Appendix. The analytical solution is then sampled on an lattice grid with cuboid unit cells, where the distance between sampling points in x and y direction is equal. The distance in z-direction is usually chosen larger to reflect the anisotropy of the point spread function of traditional optical microscopes. In the next step, Gaussian noise is added to the displacements. For each data point in the grid sampled deformation field a random number is added. This number is drawn from a Gaussian distribution with mean μ = 0 and standard deviation σ being fixed for all sam- pling points and chosen with respect to the amplitude of deformation averaged over the whole field, which we designate as < kuk >. In the following, we give the magnitude of the noise in σ/< kuk >. It has been show before that the noise distribution does indeed have a Gaussian shape [55]. The perturbed fields now form the input for the actual TFM-analysis. The deter- mined traction profile τrecon can then be compared to the initial analytical profile τ. Studies that describe new methods to improve the image processing part of TFM often include a simulation of the bead distribution, sometimes also assuming a specific point spread function [18, 56, 57]. However, noise can also arise from different sources, e.g. inhomogeneities in the gel or the bead distributions. Because here we do not aim at simulating experimental setups, but focus on the mathematical properties of different TFM-procedures, we simply simulate displacement noise [21, 23, 29]. Every TFM-method will eventually fail at very high noise lev- els, but here we ask if direct or inverse methods perform better for low or high noise levels. For each dataset used, we list the full set of parameters, including the sample point spacing, the material parameters and the parameters of the Hertz-like patches superposition to generate the profile in the S1 Appendix. Performance of direct method We first optimize the direct method by assessing the performances of the numerical derivatives and the importance of a divergence correction. Towards this aim, we choose traction profiles that emphasize the normal component. This is not the standard case in TFM, but in this way, we can best test the performance of the different approximations for derivatives that are required for the direct method. Moreover, this choice demonstrates our ability to deal with three dimensions. Fig 3 shows the two simple Hertzian traction patterns investigated below together with the cell-like profile presented in Fig 2. The first profile corresponds to a simple Hertz contact, which means that it is a force monopole pushing into the substrate. This setup could be recreated experimentally using a spherical indenter that presses into the substrate. Note that in contrast to FTTC, the direct method works in real space and also can reconstruct force monopoles. The second profile mimics a situation in which pushing and pulling tractions are exerted in different regions in a ring-like pattern. For example, this resembles the way can- cer cells invade tissue with invadopodia or funghi invade plants. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 13 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy Fig 3. Two different Hertzian traction profiles used for normal analysis with the direct method. (A) Indenter monopole with Fz = 10Aˆ μN. (B) Ring dipole with Fz = 0 μN. The upper row present a heat map of the normal traction τz at the substrate surface. There is no tangential traction in these cases. The lower row present the change of the normal traction along a lower-left to upper-right diagonal section. The results of the analysis corresponding to these profiles are presented in Fig 4. The other simulation parameters for these profiles can be found in the S1 Appendix. https://doi.org/10.1371/journal.pone.0262773.g003 To quantitatively compare the different methods of determining the deformation gradient as well as the effects of our divergence correction scheme, we calculate different quantities of interest. First we calculate the force monopole components Fx, Fy, Fz defined by Z Fi ¼ tiðx; yÞdx dy S ð36Þ which describe the total force transmitted between the sample and the substrate in our field of view. Numerically, the integration is performed using the Simpson formula [58]. These quanti- ties serve as indicators on numerical inaccuracies that add up. If they differ from their pre- dicted value, they will therefore indicate asymmetric and systematic errors in the analysis. The amplitude of the tangential components Fx and Fy should be zero for all our normal indenta- tion profiles, as no tangential force is transmitted. Due to symmetry, we expect the Fy compo- nent to show the same behavior as the Fx component. The Fz component should vanish for the dipolar patterns. If Fz was different from the expected value, this may indicate a systematic error due to the non-symmetric way of taking the derivative at the surface. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 14 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy Next we calculate the total L2-difference between the TFM results and the reference defined as dL2 ¼ kτrecon (cid:0) τtruek2 ¼ r ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Z : j τreconðx; yÞ (cid:0) τtrueðx; yÞj2dx dy S ð37Þ This quantity measures how well the reconstructed field matches the analytical solution. Again, the integration is performed using the Simpson formula. In contrast to the monopole, this value will not only capture systematic aberration offset, but also general noise and gives a measure for the uncertainty of the results. A high dL2 indicates that the reconstruction contains a high amount of noise artifacts. Fig 4 shows the performance of the different variants of the direct method as assessed through these metrics. First we observe that as expected, the variance in the monopoles and the distance metrics increase with increasing noise levels such that σN/< kuk > = 1 has to be considered to be large noise. At much higher noise levels, all methods will fail. For the DM investigated here, we find that using a four-point (4P) form instead of a simple two-point (2P) form does not improve the result, but causes a significantly higher level of overall noise, likely due to overfitting. In contrast, using the 3x3x3 patch fit significantly improves the noise sup- pression, both in the full field as well as the background. However, it will result in an underesti- mate for the force monopole z-component Fz in case of the indenter profile (A). This is likely due to the fact that uz/z will be underestimated due to the non-symmetric derivative. Applying the divergence correction algorithm does improve the result in this case and also offers a Fig 4. Quantitative comparison between different variants of the direct method. The plots use the Hertzian traction profiles introduced in Figs 2 and 3. A normalization factor < kuk > used to compare noise levels for different amplitudes is calculated by taking the mean of the amplitudes of the deformation field. In the upper line, we plot the variation in the predicted total force in x direction as a function of the standard deviation of Gaussian noise added to the input data. The line indicates the mean and the colored area indicates the standard deviation. In the mid row the same is done for the total traction in z-direction. In the lower row, we plot the total difference in norm, where the variation between the different samples is shown to be negligible. The different colors designate the different ways of calculating the deformation gradients as well as whether the divergence correction scheme (DCS) is used. Fy is not shown due to its similarity to Fx. https://doi.org/10.1371/journal.pone.0262773.g004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 15 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy slightly better noise reconstruction. However, an opposite effect is found for Fz for profiles (B) and (C), where the noise removal introduces a systematic offset in the normal traction compo- nent, as well as for FN, where it also increases the aberration in the force monopole normal components. Notably the difference in norm describes a straight line with negligible variance between the samples. This comes from the fact that for all methods, the deformation gradient and therefore also the resulting stress have a linear relation to the input deformation field. This implies a linear relation in the variance due to noise. This proportionality is then shared by the dL2 parameter. Although the divergence correction ensures that formally force and torque bal- ance are satisfied, it does not improve the performance of the DM. Comparison of direct method and FTTC Now that we have optimized the direct method, we next compare it to the inverse method, namely with FTTC-calculations both in 2.5D (2.5D FTTC) as well as with calculations in which contributions in the normal dimension are not considered, as described by Eq (10) (2D FTTC). Because we now compare with FTTC, we work with the standard choice for TFM on planar substrates, namely a collection of circular adhesion sites with mainly tangential trac- tions, as commonly observed for contractile cell types which adhere to flat substrates through focal adhesions. While Fig 2 showed traction reconstructions with different methods for the cell-like pattern for small noise (σN/< kuk > = 0.2), in Fig 5 we now show such reconstructions for variable Fig 5. Reconstruction for cell-like traction pattern for different noise levels and reconstruction methods. The images show the reconstruction for no (upper row), large (σN/< kuk > = 1, mid row) and very large (σN/< kuk > = 2, lower row) Gaussian noise added to the analytical solution for the displacement before reconstruction with the different methods as indicated. The simulation parameters are identical to the ones used in Fig 2 and details can be found in the S1 Appendix. https://doi.org/10.1371/journal.pone.0262773.g005 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 16 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy noise levels (σN/< kuk > = 0, 1 and 2 from top to bottom). For simplicity, here we only show the components in the xy-plane. A visual inspection suggests that FTTC reconstructs the shape of the deformation profile most faithfully for low noise levels. For high noise levels, it predicts a too high level of force. For very high noise levels, the adhesion sites can be more easily identi- fied by the DM compared to FTTC. The DM and in particular the DM using divergence cor- rection seem to predict a traction strength amplitude that is largely independent of noise. On the other hand, the divergence correction scheme produces some artifacts, independent of the noise level. This suggests that the divergence removal algorithm converts local noise into a more distributed signal that is not directly related to the physical force generators. In order to make these qualitative assessments more objective, we next evaluated a series of established metrics for the three different traction profiles shown in the upper two rows of Fig 6. As before, the analytical solutions are known and it is easy to add Gaussian noise to the resulting deformation fields. The analysis results in a traction field described by a discrete sam- ple vector τrecon for each site sampling point j, that can be compared to its theoretical equivalent τtrue j structions using the following five metrics [23, 29]: predicted from the analytical solution. We estimate the accuracy of the different recon- j • The deviation of traction magnitude at adhesions (DTMA) is given by: DTMA ¼ 1 NP i � X meanjðiÞ (cid:0) � � � � � � �τrecon � jðiÞ � � 2 � � �τtrue jðiÞ � � � � � 2 : � � � �τtrue jðiÞ � � � � � 2 ð38Þ meanjðiÞ Here Np is the number of adhesion patches and the index i iterates over the individual patches. For each patch, the mean is taken over all sampling points j(i) belonging to the given patch. The DTMA determines how well the average magnitude of the patches is pre- dicted. A good reconstruction would yield an DTMA close to zero, while a positive or nega- tive value would indicate an over- or underestimation, respectively. • The deviation of traction magnitude at adhesions restricted to the two tangential dimensions (tDTMA) is given by: tDTMA ¼ 1 NP X i � meanjðiÞ n2ðτrecon jðiÞ Þ (cid:0) n2ðτtrue jðiÞ Þ � jðiÞ Þ meanjðiÞ n2ðτtrue � � : ð39Þ Here Np is the number of adhesion patches and the index i iterates over the individual patches and n2ðτÞ ¼ . For each patch, the mean is taken over all sampling points j ffiffiffiffiffiffiffiffiffiffiffiffiffiffi x þ t2 t2 q y (i) belonging to the given patch. The tDTMA determines how well the average magnitude of the patches is predicted. In contrast to DTMA, we do only take into account the two tangential components. This focuses on situations in which the effect on the normal component behaves differently from the tangential one. A good reconstruction would yield an tDTMA close to zero, while a positive or negative value would indicate an over- or underestimation, respectively. • The signal to noise ratio (SNR) is defined by: X i 1 Np � meanjðiÞ kτrecon j k2 � stdkðkτrecon k k2Þ : SNR ¼ ð40Þ 17 / 25 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy Fig 6. Quantitative comparison of direct and inverse methods. Performance of different force reconstruction techniques as a function of noise for different force profiles. The point profiles tested are shown in the top two row: In the first row, the heat map indicates the amplitude of the the tangential traction, while the white arrows indicate their direction. In the second row, the heat map indicates the normal traction. Below are different metrics shown, describing the quality of the reconstruction at different locations. The plots in one column all correspond to the same force profile. Details on the simulation parameters for all three profiles can be found in the S1 Appendix. https://doi.org/10.1371/journal.pone.0262773.g006 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 18 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy Here i, j(i) and k are defined as above. The signal to noise ratio describes how well the adhe- sion sites are realized in comparison to background noise. The value should be significantly larger them 1 to indicate a good separation between traction sites and noise. • The deviation of traction magnitude in the background (DTMB) is given by: DTMB ¼ 1 Np k2Þ meankðkτrecon � X meanjðiÞ kτtrue k j k2 i � : ð41Þ Here k runs over all sampling points not belonging to any patch. i and j(i) again iterate over the patches and their sampling points respectively. The DTMB describes the level of back- ground noise in the reconstruction. Ideally it takes a value close to zero indicating a low level of artifacts in the background. • The deviation of traction maximum at adhesions (DMA) is defined by: DMA ¼ 1 NP X maxjðiÞ i � � � � �τrecon jðiÞ � � � � 2 � � maxjðiÞ � (cid:0) maxjðiÞ � � � � �τrecon jðiÞ � � � � 2 � � � �τtrue jðiÞ � � � � � 2 : ð42Þ Again i and j(i) are defined as above. The DMA is similar to the DTMA, but rather them using the average traction over the whole adhesion site, the peak traction is taken into account. This emphasizes the reconstruction of the correct amplitude in core area over the correct profile close to the boundary. Like with the DTMA a good reconstruction would yield a DMA close to 0, while a positive or negative value would indicate an over- or under- estimation, respectively. Fig 6 shows the performance of the three different methods as a function of increasing noise level and for different traction patterns as shown in the top row. For all metrics except the SNR, the optimal value is shown as black horizontal line. In general, we see that all meth- ods fail at very high noise levels. We find that FTTC is better than the DM in predicting the correct strength of the adhesions at low noise levels as seen in the DTMA, tDTMA and DMA results. However, the situation is reversed for higher noise, as the regularization is not sufficent to prevent overfitting above a certain noise level. While the application of the diver- gence correction scheme does reduce the SNR result, it improves the results for DTMB and DMA, by bringing them closer to zero, particular for high noise levels. A beneficial effect of the divergence removal is to reduce the difference between the DTMA and tDTMA scores, meaning that the algorithm does predict the orientation of the force vector more correctly. In contrast, the FTTC algorithm shows significantly better values for the DTMB metric, which means that it is more effective in preventing artifacts in the outside of adhesion sites. Surprisingly, the signal to noise ratio is lower in the case of FTTC compared to the direct method. This can be attributed to the fact that FTTC works in Fourier space and errors in the reconstruction effect the whole field of view, not only the area close to the adhesion sites. We conclude that both methods perform similarly well, that the divergence correction as used here is in fact a disadvantage, that FTTC works best for small noise and that for larger noise the direct method becomes comparable and gives a clearer visualisation due to the higher SNR. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 19 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy Fig 7. Effects of variation in sample density. Plots A to E show how the different metrics are affected when using a different distance of the sampling points when setting up the input data without adding any noise. The profile first shown in Fig 2 is used. The default sample point spacing in x and y for this profile is 0.4 μm and here is varied up and down. The z-component is scaled proportionally. Since there is no noise, in general FTTC performs much better. The direct method shows a consistent improvement for decreasing sampling point distance. Details on the simulation parameters can be found in the S1 Appendix. https://doi.org/10.1371/journal.pone.0262773.g007 Effect of sampling density We finally study the effect of variation in sampling density, which experimentally is related to marker bead density and the resolution of the optical microscope. Because in practise displace- ment noise is expected to change with sampling distance and because we now focus on the effect of sampling density, in Fig 7 we show the results for vanishing displacement noise. As expected, overall all metrics become worse with increasing sampling distance. Interestingly, however, the performance of FTTC seems to be more robust, while the DM quickly decreases in performance. This implies that a decrease in sampling distance (that is an increase in sam- pling density) will be much more beneficial for the DM. We note that for FTTC, the SNR and (for 2.5D FTTC) also DTMB can even slightly improve with decreasing sampling density. This surprising (but weak) effect might be related to the fact that for the Fourier method, increasing sampling distance amounts to stronger filtering of the data, thus focusing on the overall adhe- sion pattern. We also checked that the same trends persist for variation of sampling distance at finite noise (S1 Appendix). We noticed that for FTTC the SNR now significantly improves for a decrease in sampling density. This confirms that noise and sample point density are in fact correlated, because an increase in sampling density also increases the number of nodes that contribute noise towards the calculation. FTTC and DM are affected differently by an increase in sampling density in the presents of noise. For the DM the increase of the numbers of nodes improves the accuracy of the numerical gradients. For FTTC, the quality of reconstruction at the sampling points does not significantly improve when increasing the resolution, but the increase of the number of nodes increases the negative effect of noise on the result. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 20 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy Discussion Motivated by the observation that different TFM-methods are often advanced in specific con- texts, but rarely compared to each other, here we have conducted an in-depth comparison of inverse and direct methods in the framework of 2.5D TFM. Recently such a comparison has been conducted using FEM-approaches for 3D TFM [17, 18], but here we focus on 2.5D TFM as the traditional setup for high-resolution experiments. For the inverse method, we have used the commonly used method, namely the Fourier method FTTC, which is very fast and reliable when combined with a regularization scheme, for which here we have chosen zero-order Tikhonov regularization with generalized cross-validation for identification of the regulariza- tion parameter λ. Although we here work with the setup of a flat elastic substrate, which usu- ally is treated with 2D TFM, in order to account for the 3D-nature of the direct method, we included the third dimension by considering also normal forces (2.5D TFM). For this purpose, we have implemented a version of 2.5D FTTC based on newly derived GFs with normal com- ponents. With these advances for 2.5D FTTC, the two methods can be directly compared to each other. For each performance test, we have first designed traction patterns that are representative for experiments, suitable for the task at hand and analytically tractable. Motivated by the obser- vation that experimental noise is Gaussian-distributed in experiments [55], we have added Gaussian noise to the displacements and then performed the reconstructions with different methods. In the future, this procedure could be complemented by a stronger focus on the actual image generation occurring in TFM-experiments, in particular by using specific image processing algorithms and point spread functions [18, 56, 57]. We have evaluated our reconstructions using commonly used metrics [23, 29]. This meth- odology then was used in three ways. We first optimized the direct method, then compared it with FTTC, and finally studied the effect of sampling distance. For the direct method, we found that the 3x3x3 patch calculation of the derivatives is indeed the best solution and that the standard divergence correction from hydrodynamics works, but does not necessarily improve our solutions and in fact worsens the visual appearance of the traction pattern. Assuming a perfect elastic material, divergence is generated only by noise, which is known to be essentially Gaussian in experimental TFM data. This means that it is uncorrelated between different dimensions. Divergence removal however couples the different dimensions and therefore does not counteract the process that generated the divergence. We conclude that although required from the viewpoint of elasticity theory, divergence removal is not really needed for the TFM procedures used here. Our main result is the demonstration that the direct method for TFM can be used to reli- ably predict the traction field at the surface of a flat elastic substrates and in fact performs com- paratively well to the best inverse method, that is FTTC, at least for large noise, when FTTC worsens more quickly. The direct method also offers an interesting alternative to costly FEM- simulations, in situations where FTTC cannot be applied. The 3x3x3 patch method that has proven to be a reliable method for deformation gradient calculation can in fact be easily adapted for curved surfaces. Although FTTC is expected to remain the standard method for 2D TFM, we believe that the direct method is a valuable alternative even in this case, e.g. because it can also reconstruct monopolar traction patterns. For 2.5D TFM we believe that the choice between the direct and inverse methods should depend on context, but that in princi- ple, both might work well. In the future, the direct method might become more important due to several experimental developments. Recent advances in microfabrication and additive manufactoring will lead to completely new geometries and setups [32–35], for which the analytical solutions required for PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 21 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy methods like FTTC will not be possible anymore, except for simple geometries like elastic beads [12, 36]. Inverse methods implemented in FEM-environments are the method of choice in this case, but because they are computational costly, the DM is a attractive alternative. Moreover image quality for TFM quickly improves, e.g. due to super-resolution microscopy [26, 37–40], and this might play in favor of the direct method, which without displacement noise becomes better with smaller sampling distance. Supporting information S1 Appendix. Supplementary calculations related to the Hertz-like force profile, details of the divergence correction algorithm and parameters used for the simulated profiles. (PDF) Acknowledgments We thank Peter Bastian for helpful discussions on the divergence correction. Author Contributions Conceptualization: Johannes W. Blumberg, Ulrich S. Schwarz. Data curation: Johannes W. Blumberg. Formal analysis: Johannes W. Blumberg. Funding acquisition: Ulrich S. Schwarz. Investigation: Johannes W. Blumberg. Methodology: Johannes W. Blumberg. Project administration: Ulrich S. Schwarz. Resources: Ulrich S. Schwarz. Software: Johannes W. Blumberg. Supervision: Ulrich S. Schwarz. Validation: Johannes W. Blumberg. Visualization: Johannes W. Blumberg. Writing – original draft: Johannes W. Blumberg, Ulrich S. Schwarz. Writing – review & editing: Johannes W. Blumberg, Ulrich S. Schwarz. References 1. Yamada KM, Sixt M. Mechanisms of 3D cell migration. Nature Reviews Molecular Cell Biology. 2019; 20:738–752. https://doi.org/10.1038/s41580-019-0172-9 PMID: 31582855 2. Iskratsch T, Wolfenson H, Sheetz MP. Appreciating force and shape—the rise of mechanotransduction in cell biology. Nature Reviews Molecular Cell Biology. 2014; 15:825–833. https://doi.org/10.1038/ nrm3903 PMID: 25355507 3. Style RW, Boltyanskiy R, German GK, Hyland C, MacMinn CW, Mertz AF, et al. Traction force micros- copy in physics and biology. Soft Matter. 2014; 10:4047. https://doi.org/10.1039/c4sm00264d PMID: 24740485 4. Schwarz US, Soine´ JRD. Traction force microscopy on soft elastic substrates: A guide to recent compu- tational advances. Biochimica et Biophysica Acta (BBA)—Molecular Cell Research. 2015; 1853:3095– 3104. https://doi.org/10.1016/j.bbamcr.2015.05.028 PMID: 26026889 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 22 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy 5. Polacheck WJ, Chen CS. Measuring cell-generated forces: a guide to the available tools. Nature Meth- ods. 2016; 13:415–423. https://doi.org/10.1038/nmeth.3834 PMID: 27123817 6. Roca-Cusachs P, Conte V, Trepat X. Quantifying forces in cell biology. Nature Cell Biology. 2017; 19:742–751. https://doi.org/10.1038/ncb3564 PMID: 28628082 7. Ando T, Bhamidimarri SP, Brending N, Colin-York H, Collinson L, De Jonge N, et al. The 2018 correla- tive microscopy techniques roadmap. Journal of physics D: Applied Physics. 2018; 51:443001. https:// doi.org/10.1088/1361-6463/aad055 PMID: 30799880 8. Mulligan JA, Bordeleau F, Reinhart-King CA, Adie SG. Traction Force Microscopy for Noninvasive Imaging of Cell Forces. In: Dong C, Zahir N, Konstantopoulos K, editors. Biomechanics in Oncology. Advances in Experimental Medicine and Biology. Cham: Springer International Publishing; 2018. p. 319–349. 9. Legant WR, Miller JS, Blakely BL, Cohen DM, Genin GM, Chen CS. Measurement of mechanical trac- tions exerted by cells in three-dimensional matrices. Nature Methods. 2010; 7:969. https://doi.org/10. 1038/nmeth.1531 PMID: 21076420 10. Steinwachs J, Metzner C, Skodzek K, Lang N, Thievessen I, Mark C, et al. Three-dimensional force microscopy of cells in biopolymer networks. Nature Methods. 2016; 13:171–176. https://doi.org/10. 1038/nmeth.3685 PMID: 26641311 11. Mohagheghian E, Luo J, Chen J, Chaudhary G, Chen J, Sun J, et al. Quantifying compressive forces between living cell layers and within tissues using elastic round microgels. Nature Communications. 2018; 9:1878. https://doi.org/10.1038/s41467-018-04245-1 PMID: 29760452 12. Vorselen D, Wang Y, de Jesus MM, Shah PK, Footer MJ, Huse M, et al. Superresolved microparticle traction force microscopy reveals subcellular force patterns in immune cell-target interactions. Nature Communications. 2020; 11:20. https://doi.org/10.1038/s41467-019-13804-z PMID: 31911639 13. Tra¨ ber N, Uhlmann K, Girardo S, Kesavan G, Wagner K, Friedrichs J, et al. Polyacrylamide Bead Sen- sors for in vivo Quantification of Cell-Scale Stress in Zebrafish Development. Scientific Reports. 2019; 9. https://doi.org/10.1038/s41598-019-53425-6 PMID: 31745109 14. Maskarinec SA, Franck C, Tirrell DA, Ravichandran G. Quantifying cellular traction forces in three dimensions. Proceedings of the National Academy of Sciences. 2009; 106:22108–22113. https://doi. org/10.1073/pnas.0904565106 PMID: 20018765 15. 16. Franck C, Maskarinec SA, Tirrell DA, Ravichandran G. Three-Dimensional Traction Force Microscopy: A New Tool for Quantifying Cell-Matrix Interactions. PLoS ONE. 2011; 6:e17833. https://doi.org/10. 1371/journal.pone.0017833 PMID: 21468318 Toyjanova J, Bar-Kochba E, Lo´pez-Fagundo C, Reichner J, Hoffman-Kim D, Franck C. High Resolu- tion, Large Deformation 3D Traction Force Microscopy. PLoS ONE. 2014; 9:e90976. https://doi.org/10. 1371/journal.pone.0090976 PMID: 24740435 17. Sanz-Herrera JA, Barrasa-Fano J, Co´ ndor M, Oosterwyck HV. Inverse method based on 3D nonlinear physically constrained minimisation in the framework of traction force microscopy. Soft Matter. 2021. 18. Barrasa-Fano J, Shapeti A, de Jong J, Ranga A, Sanz-Herrera JA, Oosterwyck HV. Advanced in silico validation framework for three-dimensional traction force microscopy and application to an in vitro model of sprouting angiogenesis. Acta Biomaterialia. 2021; 126:326–338. https://doi.org/10.1016/j. actbio.2021.03.014 PMID: 33737201 19. Dembo M, Oliver T, Ishihara A, Jacobson K. Imaging the traction stresses exerted by locomoting cells with the elastic substratum method. Biophysical Journal. 1996; 70:2008–2022. https://doi.org/10.1016/ S0006-3495(96)79767-9 PMID: 8785360 20. Dembo M, Wang YL. Stresses at the Cell-to-Substrate Interface during Locomotion of Fibroblasts. Bio- physical Journal. 1999; 76:2307–2316. https://doi.org/10.1016/S0006-3495(99)77386-8 PMID: 10096925 21. Schwarz US, Balaban NQ, Riveline D, Bershadsky A, Geiger B, Safran SA. Calculation of Forces at Focal Adhesions from Elastic Substrate Data: The Effect of Localized Force and the Need for Regulari- zation. Biophysical Journal. 2002; 83:1380–1394. https://doi.org/10.1016/S0006-3495(02)73909-X PMID: 12202364 22. Soine´ JRD, Brand CA, Stricker J, Oakes PW, Gardel ML, Schwarz US. Model-based Traction Force Microscopy Reveals Differential Tension in Cellular Actin Bundles. PLoS Computational Biology. 2015; 11:e1004076. https://doi.org/10.1371/journal.pcbi.1004076 PMID: 25748431 23. Huang Y, Schell C, Huber TB, Şimşek AN, Hersch N, Merkel R, et al. Traction force microscopy with optimized regularization and automated Bayesian parameter selection for comparing cells. Scientific Reports. 2019; 9. https://doi.org/10.1038/s41598-018-36896-x PMID: 30679578 24. Butler JP, Tolić-Nørrelykke IM, Fabry B, Fredberg JJ. Traction fields, moments, and strain energy that cells exert on their surroundings. American Journal of Physiology-Cell Physiology. 2002; 282:C595– C605. https://doi.org/10.1152/ajpcell.00270.2001 PMID: 11832345 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 23 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy 25. Bauer A, Prechova´ M, Gregor M, Fabry B. pyTFM: A tool for Traction Force and Monolayer Stress Microscopy. PLoS Computational Biology. 2021; 17:e1008364. https://doi.org/10.1371/journal.pcbi. 1008364 PMID: 34153027 26. Colin-York H, Javanmardi Y, Barbieri L, Li D, Korobchevskaya K, Guo Y, et al. Spatiotemporally Super- Resolved Volumetric Traction Force Microscopy. Nano Letters. 2019; 19:4427–4434. https://doi.org/10. 1021/acs.nanolett.9b01196 PMID: 31199151 27. 28. Landau LD, Lifshitz EM. Theory of elasticity. 2nd ed. Oxford etc.: Pergamon Press etc.; 1963. del A´ lamo JC, Meili R, A´ lvarez-Gonza´ lez B, Alonso-Latorre B, Bastounis E, Firtel R, et al. Three-Dimen- sional Quantification of Cellular Traction Forces and Mechanosensing of Thin Substrata by Fourier Traction Force Microscopy. PLoS ONE. 2013; 8:e69850. https://doi.org/10.1371/journal.pone.0069850 PMID: 24023712 29. Sabass B, Gardel ML, Waterman CM, Schwarz US. High Resolution Traction Force Microscopy Based on Experimental and Computational Advances. Biophysical Journal. 2008; 94:207–220. https://doi.org/ 10.1529/biophysj.107.113670 PMID: 17827246 30. Huang Y, Gompper G, Sabass B. A Bayesian traction force microscopy method with automated denois- ing in a user-friendly software package. Computer Physics Communications. 2020; 256:107313. https://doi.org/10.1016/j.cpc.2020.107313 31. Hur SS, Zhao Y, Li YS, Botvinick E, Chien S. Live cells exert 3-dimensional traction forces on their sub- strata. Cellular and Molecular Bioengineering. 2009; 2:425–436. https://doi.org/10.1007/s12195-009- 0082-6 PMID: 19779633 32. Klein F, Richter B, Striebel T, Franz CM, Freymann Gv, Wegener M, et al. Two-component polymer scaffolds for controlled three-dimensional cell culture. Advanced Materials. 2011; 23:1341–1345. https://doi.org/10.1002/adma.201004060 PMID: 21400593 33. Brand CA, Linke M, Weißenbruch K, Richter B, Bastmeyer M, Schwarz US. Tension and elasticity con- tribute to fibroblast cell shape in three dimensions. Biophysical Journal. 2017; 113:770–774. https://doi. org/10.1016/j.bpj.2017.06.058 PMID: 28755755 34. Hippler M, Lemma ED, Bertels S, Blasco E, Barner-Kowollik C, Wegener M, et al. 3D Scaffolds to Study Basic Cell Biology. Advanced Materials. 2019; p. 1808110. https://doi.org/10.1002/adma.201808110 PMID: 30793374 35. Hippler M, Weißenbruch K, Richler K, Lemma ED, Nakahata M, Richter B, et al. Mechanical stimulation of single cells by reversible host-guest interactions in 3D microscaffolds. Science Advances. 2020; 6: eabc2648. https://doi.org/10.1126/sciadv.abc2648 PMID: 32967835 36. Wang Y, Zhang X, Cai W. Spherical harmonics method for computing the image stress due to a spheri- cal void. Journal of the Mechanics and Physics of Solids. 2019; 126:151–167. https://doi.org/10.1016/j. jmps.2019.01.020 37. Colin-York H, Shrestha D, Felce JH, Waithe D, Moeendarbary E, Davis SJ, et al. Super-Resolved Trac- tion Force Microscopy (STFM). Nano Letters. 2016. https://doi.org/10.1021/acs.nanolett.6b00273 PMID: 26923775 38. Aramesh M, Mergenthal S, Issler M, Plochberger B, Weber F, Qin XH, et al. Functionalized Bead Assay to Measure Three-dimensional Traction Forces during T-cell Activation. Nano Letters. 2021; 21:507– 514. https://doi.org/10.1021/acs.nanolett.0c03964 PMID: 33305952 39. Li D, Colin-York H, Barbieri L, Javanmardi Y, Guo Y, Korobchevskaya K, et al. Astigmatic traction force microscopy (aTFM). Nature Communications. 2021; 12:2168. https://doi.org/10.1038/s41467-021- 22376-w PMID: 33846322 40. Stubb A, Laine RF, Miihkinen M, Hamidi H, Guzma´ n C, Henriques R, et al. Fluctuation-Based Super- Resolution Traction Force Microscopy. Nano Letters. 2020; 20:2230–2245. https://doi.org/10.1021/acs. nanolett.9b04083 PMID: 32142297 41. Hanke J, Probst D, Zemel A, Schwarz US, Ko¨ ster S. Dynamics of force generation by spreading plate- lets. Soft Matter. 2018; 14:6571–6581. https://doi.org/10.1039/C8SM00895G PMID: 30052252 42. Holenstein CN, Silvan U, Snedeker JG. High-resolution traction force microscopy on small focal adhe- sions - improved accuracy through optimal marker distribution and optical flow tracking. Scientific Reports. 2017; 7. https://doi.org/10.1038/srep41633 PMID: 28164999 43. Han SJ, Oak Y, Groisman A, Danuser G. Traction microscopy to identify force modulation in subresolu- tion adhesions. Nature Methods. 2015; 12:653–656. https://doi.org/10.1038/nmeth.3430 PMID: 26030446 44. Love AEH. A Treatise on the mathematical theory of elasticity. Repr. 4. ed. ed. Dover books on engi- neering and engineering physics. New York: Dover Publ.; 1972. 45. Howell P, Kozyreff G, Ockendon JR. Applied solid mechanics. Cambridge texts in applied mathematics. Cambridge etc.: Cambridge Univ. Press; 2009. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 24 / 25 PLOS ONEComparison of direct and inverse methods for 2.5D traction force microscopy 46. Golub GH, Heath M, Wahba G. Generalized Cross-Validation as a Method for Choosing a Good Ridge Parameter. Technometrics. 1979; 21:215–223. https://doi.org/10.1080/00401706.1979.10489751 47. 48. Franck C, Hong S, Maskarinec SA, Tirrell DA, Ravichandran G. Three-dimensional Full-field Measure- ments of Large Deformations in Soft Materials Using Confocal Microscopy and Digital Volume Correla- tion. Experimental Mechanics. 2007; 47:427–438. https://doi.org/10.1007/s11340-007-9037-9 Tambe DT, Croutelle U, Trepat X, Park CY, Kim JH, Millet E, et al. Monolayer stress microscopy: limita- tions, artifacts, and accuracy of recovered intercellular stresses. PLoS ONE. 2013; 8:e55172. https:// doi.org/10.1371/journal.pone.0055172 PMID: 23468843 49. Song SM, Napel S, Glover GH, Pelc NJ. Noise reduction in three-dimensional phase-contrast MR veloc- ity measurementsl. Journal of Magnetic Resonance Imaging. 1993; 3:587–596. https://doi.org/10.1002/ jmri.1880030407 PMID: 8347951 50. Vennell R, Beatson R. A divergence-free spatial interpolator for large sparse velocity data sets. Journal of geophysical research. 2009; 114. https://doi.org/10.1029/2008JC004973 51. de Silva CM, Philip J, Marusic I. Minimization of divergence error in volumetric velocity measurements and implications for turbulence statistics. Experiments in Fluids. 2013; 54:1557. https://doi.org/10.1007/ s00348-013-1557-8 52. Wang C, Gao Q, Wei R, Li T, Wang J. Weighted divergence correction scheme and its fast implementa- tion. Experiments in Fluids. 2017; 58. https://doi.org/10.1007/s00348-017-2307-0 53. Johnson KL. One Hundred Years of Hertz Contact. Proceedings of the Institution of Mechanical Engi- neers. 1982; 196:363–378. 54. Hamilton GM, Goodman LE. The Stress Field Created by a Circular Sliding Contact. Journal of Applied Mechanics. 1966; 33:371–376. https://doi.org/10.1115/1.3625051 55. Plotnikov SV, Sabass B, Schwarz US, Waterman CM. High-Resolution Traction Force Microscopy. In: Methods in Cell Biology. Elsevier; 2014. p. 367–394. 56. Jorge-Peñas A, Izquierdo-Alvarez A, Aguilar-Cuenca R, Vicente-Manzanares M, Garcia-Aznar JM, Oosterwyck HV, et al. Free Form Deformation–Based Image Registration Improves Accuracy of Trac- tion Force Microscopy. PLoS ONE. 2015; 10:e0144184. https://doi.org/10.1371/journal.pone.0144184 PMID: 26641883 57. Bar-Kochba E, Toyjanova J, Andrews E, Kim KS, Franck C. A Fast Iterative Digital Volume Correlation Algorithm for Large Deformations. Experimental Mechanics. 2014; 55:261–274. https://doi.org/10. 1007/s11340-014-9874-2 58. Abramowitz M, Stegun IA. Handbook of Mathematical Functions with Formulas, Graphs, and Mathe- matical Tables. 10th ed. U.S. Goverment Printing Office; 1964. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262773 January 20, 2022 25 / 25 PLOS ONE
10.1371_journal.pone.0251901	RESEARCH ARTICLE Effect of disinfection agents and quantification of potentially viable Leptospira in fresh water samples using a highly sensitive integrity-qPCR assay Elise Richard1,2, Pascale Bourhy2, Mathieu PicardeauID Se´ bastien Wurtzer1 2, Laurent MoulinID 1, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Richard E, Bourhy P, Picardeau M, Moulin L, Wurtzer S (2021) Effect of disinfection agents and quantification of potentially viable Leptospira in fresh water samples using a highly sensitive integrity-qPCR assay. PLoS ONE 16(5): e0251901. https://doi.org/10.1371/journal.pone.0251901 Editor: Odir Antonio Dellagostin, UFPL, BRAZIL Received: March 25, 2021 Accepted: May 5, 2021 Published: May 26, 2021 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0251901 Copyright: © 2021 Richard et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Funding: This work is part of the PhD thesis of E. R., who received financial support from the "DIM 1 Eau de Paris, DRDQE, Ivry-Sur-Seine, France, 2 Institut Pasteur, Unite´ Biologie des Spirochètes, Paris, France * mathieu.picardeau@pasteur.fr (MP); laurent.moulin@eaudeparis.fr (LM) Abstract Leptospirosis is an emerging worldwide zoonotic disease, but the general biology of the causative agents is still poorly understood. Humans are an occasional host. The main risk factors are water-associated exposure during professional or recreational activities or during outbreaks in endemic areas. Detecting the presence of pathogenic bacteria in aquatic envi- ronments and their capacity to resist various inactivation processes are research fields that need to be further developed. In addition, the methods used for detecting and enumerating Leptospira still need to be improved. We aimed to describe a new quantitative polymerase chain reaction coupled to propidium monoazide treatment (PMAqPCR) that targets not only total Leptospira but also discriminates pathogenic from non-pathogenic Leptospira while also addressing PCR inhibitors, a frequently encountered problem when studying environ- mental water. In a second step, the killing efficiency of Leptospira to different treatments was tested and PMAqPCR compared to culture-based enumeration. This provided informa- tion about the effects of temperature, as well as ultraviolet and chlorine disinfection, that are both related to water treatment processes, in particular for the production of drinking water, on the persistence of both saprophytic and pathogenic Leptospira. Finally, PMAqPCR was used for the detection of Leptospira in freshwater samples for a proof-of-concept. In conclu- sion, our method could be used for routine freshwater monitoring and allows better evalua- tion of the presence of Leptospira, allowing evaluation of the bacterial dynamics in a designated area or assessment of the efficacy of water disinfection processes. Introduction Pathogenic Leptospira are responsible for a global zoonosis, leptospirosis, in which humans are found to be occasional hosts in a cycle involving wild and domestic animals. One million severe cases are reported every year worldwide [1]. Leptospirosis can take various forms, from a flu-like syndrome (fever, myalgia, or headache) to a multisystem disorder, with icteric and hemorrhagic syndrome accounting for 20% of cases, causing at least 60,000 deaths a year. PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 1 / 17 PLOS ONE1Health 2017 from Ile-de-France region". The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Effect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay This disease occurs worldwide but the incidence is the highest in tropical regions [1–3]. However, developed countries, including those in Europe [4, 5], have also observed an increase in the number of reported cases. Animal reservoirs, mainly rodents, excrete Leptospira through urine and contaminate the environment and, potentially, water resources. The dissemination of these bacteria into the environment can allow other animals or humans to be newly infected. Humans can be infected through direct or indirect contact with urine or water contaminated by Leptospira. In this con- text, leptospirosis can be both an occupational disease (affecting veterinarians, farmers, sewer workers, etc.) and a recreational disease associated with water-related activities (bathing, kay- aking, canyoning, etc.) [6, 7]. Leptospirosis is considered to be an emerging zoonotic disease partially due to global warming [8] and more frequent and severe flooding events [9]. Water exposure appears to be the major risk factor [10]. Floods increase the risk by exposing humans and animals to Leptospira that are flushed out of their environment. In France, a study con- ducted from 1995 to 2005 showed that 42% of patients became infected after practicing water sports, 19% after contact with backwater (ponds, swamps, wells, water holes), and 19% during professional activities [11]. Recreational water activities are becoming increasingly popular and, for example, a recent study described a cluster of 14 kayakers that exhibited leptospirosis symptoms after contact with water in Britanny, France [12]. Leptospirosis cases have also been reported after the consumption of drinking water. Contamination may be caused by failure during water treatment processes or the lack of any water treatment [13, 14]. To date, sixty-four species and more than three hundred serovars of Leptospira have been described and classified into four phylogenetic subclades: pathogenic P1 and P2 and sapro- phytic S1 and S2 [15]. Saprophytes are non-infectious species that can multiply in the environ- ment, whereas pathogens are mostly isolated from both humans and animals and, occasionally the environment, in which they can survive for a few weeks [16]. Leptospira are slow-growing bacteria (generation time of 5 and 20 hours for saprophytes and pathogens, respectively), requiring a specific and rich medium [17, 18]. They are fastidious to isolate because of possible contamination with fast-growing interfering flora [19–21]. The recent development of a cock- tail of antibacterial and antifungal antibiotics to which Leptospira are resistant (sulfamethoxa- zole, trimethoprim, Fosfomycin, 5-fluorouracil, amphotericin B,), thus limiting the development of interfering microorganisms [22], should facilitate their culture. Little information is available on Leptospira contamination of surface water and its seasonal evolution. Several authors have highlighted the importance of assessing the bacterial concen- tration in water resources [19]. Such evaluation is essential for monitoring population expo- sure to Leptospira to ensure public health. Currently, scarce data are available about the persistence of Leptospira in the environment and their resistance to physico-chemical parame- ters or disinfection. According to Huang et al. [23], Leptospira genomes were still detected in tap water after treatment, whereas most other pathogens disappeared. The effects of temperature and pH on Leptospira were studied by Chang et al. [24]. In this study, the authors showed that optimal survival conditions for Leptospira are neutral to slightly alkaline pH and a temperature of approximately 25 to 27˚C. Other studies [25, 26] showed that pathogenic strains of Leptospira could survive for > 20 months, despite deleterious storage conditions (cold, nutrient-poor acidic water, etc.). To date, no study has described the antimi- crobial effect of ultraviolet (UV) radiation exposure, a process commonly used in tap water production, on the viability of Leptospira. Molecular diagnostic methods appear to be more sensitive, specific, and rapid than culture [27, 28]. qPCR appears to be applicable for Leptospira detection, although there is no consen- sus concerning molecular methods. In most recent studies, real-time PCR assays have been based on Taqman technology [29, 30] instead of the SYBR Green approach [27, 31] due to its PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 2 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay higher specificity [29]. Molecular methods for Leptospira detection have been mainly based on the detection of housekeeping genes, such as rrs 16S [32], gyrB [33], or secY [27]. Pathogen- specific Leptospira can be detected using genes such as LipL32 [29], ligA, or ligB [34]. However, most analyses solely allow quantification of the Leptospira genome, irrespective of the viability of the bacteria. These methods tend to overestimate the true risk, which is solely linked to via- ble bacteria. The use of qPCR coupled with a DNA intercalating agent pretreatment was already described to evaluate the integrity of Leptospira [35].This approach avoids the amplifi- cation of permeable bacteria (i.e. damaged or dead bacteria) and, thus, partially limits their overestimation while maintaining the advantages of molecular methods [36]. Our objective was to create a rapid and sensitive method for the quantification of poten- tially viable Leptospira in water samples, discriminating pathogens from saprophytes. This PMAqPCR detection method was also used to evaluate the resistance of Leptospira to different treatments such as heat inactivation, chlorine treatment, and UV exposure. Materials and methods Sample collection and culture media Twenty-five Leptospira DNA samples, including that of eight saprophytes (four for S1 and four for S2) and seventeen pathogens (twelve for P1 and five for P2) (S1 Table) were used to test the perfor- mance of the PCR assays. The pathogen L. interrogans serovar Manilae strain L495 and the sapro- phyte L. biflexa serovar Patoc strain Patoc 1 were used for inactivation tests. Strains were incubated at 30˚C in liquid Ellinghausen, McCullough, Johnson, and Harris (EMJH) medium [37, 38]. For colony numeration, EMJH medium supplemented with 1.2% agar was used. Bacterial numbers were determined using a Petroff-Hausser chamber and dark-field microscopy. Lep- tospira DNA and strains were provided by the National Reference Center (NRC) for Leptospi- rosis (Institut Pasteur, Paris, France). Other bacterial strains used to determine the specificity of the pan-Leptospira PCR assay were provided by the Eau De Paris Laboratory. Extraction For inactivation tests, DNA was extracted using the Q400 protocol with a QIAsymphony1 DSP Virus/Pathogen Midi Kit (Qiagen) and the MagNA Pure Compact System (Roche1) for environmental monitoring. Nucleic acids were extracted from 200 μL of sample by elution into 50 μL, according to the manufacturer’s protocols. PCR assays The multiplex qPCR was tested using Leptospira strains isolated from patients and environ- mental strains. The specificity of the qPCR methodology was established by the analysis of Lep- tospira strains and other bacterial species (S1 Table). The Pan-Leptospira PCR was based on the rrs (16S) gene sequence alignment of 113 Leptos- pira species, including saprophytes and pathogens. The primers 16S-F267 (5’-GGCCACAA TGGAACTGAG-3’) and 16S-R336 (5’-CCCATTGAGCAAGATTCTTAAC-3’), associated with the probe 16S-P286 (5’- FAM-CACGGTCCATACTCCT-NFQ-MGB-3’), achieve the amplification of a 70-bp fragment. The pathogenic-Leptospira PCR was based on the LipL32 gene sequence alignment of 30 pathogenic Leptospira species. The primers LipL32-F164 (5’- CTGTGATCAACTATTACGG-3’) and LipL32-R298 (5’-GGGAAATCATACGAACTC-3’), associated with the probe LipL32-P188 (5’-HEX-TAAAGCCAGGACAAGCGCCG-BHQ1-3’), achieve the amplification of a 135-bp fragment. PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 3 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay Sequences from the Genbank database of the National Center for Biotechnology Informa- tion (NCBI) were used to design the primers (S2 and S3 Tables). The primers and probes for both PCR assays were designed using AlleleID1 software version 7 (http://premierbiosoft. com/bacterial-identification/index.html). Two plasmids (16S and LipL32) were generated as positive PCR controls using pCR2.1 (Topo TA-cloning, Life Technologies, Carlsbad, CA). An internal positive competitive amplifi- cation control (IPC) was used [39] to evaluate the presence of PCR inhibitors. The IPC is com- posed of a partial sequence of the human β-actin gene and was cloned into the pCR™2.1-TOPO1 vector (Life Technologies, Carlsbad, CA) and flanked by the LipL32 prim- ers using an approach similar to that described by Wurtzer et al. [40]. The primers were LipL32_BACT-F1146 (5’-CTGTGATCAACTATTACGGttGCAGGAGTATGACGAGT-3’) and LipL32_BACT-R1215 (5’-GGGAAATCATACGAACTCttCAAGAAAGGGTGTAACGCA ACTAA-3’). The probe was BACT_P1172 (5’ -CCCCTCCATCGTCCACCGCAAATG-3’). Each PCR reaction was performed using the TaqMan ™ Fast Virus 1-Step Master Mix (#4444434). Unlike other polymerases, this Taq polymerase was less affected by inhibition due to the environmental sample matrix. The RT step was removed to achieve rapid diagnosis. Never- theless, residual activity of the RT step remained, thus conferring better sensitivity than other kits. For the pan-Leptospira PCR (16S), the F267 primer was used at 500 nM and the R336 primer and P286 probe at 100 nM. Simultaneously, oligonucleotides targeting pathogenic-Leptospira PCR (LipL32) were used at 600 nM for primer F164 and primer R298, and 200 nM for probe P188. IPC was added to the reaction mixture at 104 copies and detected using the IPC probe at 100 nM. The PCR reaction was performed in a 20-μL reaction volume using a ViiA 7 real-time PCR system (Life Technologies, Carlsbad, CA) in 96-well plates. The thermal profile consisted of an initial denaturation step at 94˚C for 20 s, followed by 45 cycles at 94˚C for 5 s and 60˚C for 30 s. FAM Yakima Yellow and Tamra fluorescence were detected at the end of the elongation step. The three plasmids, positive controls, and IPC were quantified using an ultra-sensitive fluo- rescent nucleic acid stain for quantitating double-stranded DNA (Quant-iT™ PicoGreen1 dsDNA reagent), according to the manufacturer’s protocol. Propidium monoazide treatment Before DNA extraction, samples were incubated with propidium monoazide (PMAxx) to ensure bacterial integrity [41–43]. The PMAxx solution was diluted in molecular grade water to obtain a final concentration of 10 mM and aliquots were stored at -20˚C. Based on pilot studies (data not shown), the PMAxx dye was used at a final concentration of 0.1 mM to pre-treat the samples. After mixing, samples were incubated on ice, in the dark, for 30 min. Photo-activation was per- formed for 15 min using the PhaST Blue system (IUL, Barcelona, Spain). In this study, PMAxx- qPCR was also called integrity qPCR and is sometimes wrongly called viability PCR. Persistence tests Persistence tests were simultaneously performed in duplicate on two laboratory strains: Leptos- pira biflexa serovar Patoc (saprophyte) and Leptospira interrogans serovar Manilae (pathogen). For each test, bacterial inactivation was modeled using GraphPad Prism version 8 software (GraphPad, La Jolla, CA). After testing several models, the Leptospira reduction data were adjusted using a sigmoidal dose−response model based on the equation: Y ¼ Ct þ ðC0 (cid:0) CtÞ (cid:0) Þa 1 þ IC50 X PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 4 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay In this equation, X was the studied parameter (temperature, CT value, time, UV dose), the variable “Ct” the Leptospira concentration at a given time during the assay, “C0” the Leptospira concentration at T0, and IC50 the value of X at which the response was halfway between C0 and C0. Finally, α described the slope of the curve. As mentioned in every test referenced below, two detection methods were used in this study: integrity qPCR (described above) and culture on specific EMJH medium. Heat exposure. Bacterial cultures were adjusted to a final concentration of approximately 104 Leptospira/mL in 1X PBS (pH 7.4). Each sample was aliquoted in duplicate and incubated for 1 h at various temperatures in a thermal cycler (Mastercycler1 nexus, Eppendorf). Sam- ples tested at 4˚C were stored on ice. Half of each sample was used for plating after resuspension in EMJH medium and half for PMA-qPCR analysis after resuspension in 1X PBS and the addition of PMAxx (100 μM final concentration). Nucleic acids were extracted using a QIAsymphony instrument (QIAGEN). Ultraviolet (UV) exposure. Exponential phase cultures of L. interrogans and L. biflexa were centrifuged at 8,000 x g for 10 min and the pellet resuspended in 1X PBS to a final con- centration of 109 Leptospira/mL. The bacterial suspension was split into microtubes (450 μL/ tube). For each strain, eight UV conditions were tested in duplicate, from 0 mJ/cm2 to 40 mJ/ cm2 (the last is the UV dose applied in drinking-water treatment plants) using a 254 nm UV lamp (Phillips, Amsterdam, Netherlands) at room temperature. A digital UVC radiometer (IL Metronic Sensortechnik GmbH, Germany) was used to monitor UV irradiation. Samples were subjected to various exposure times according to the tested dose and half used for plating on EMJH agar and half for PMA-qPCR analysis. Chlorine treatment. Before starting the experiments, laboratory glassware was prepared by soaking it in a sodium hypochlorite solution containing 40 mg/L free active chlorine. Glassware was then intensively cleaned with chlorine demand-free (CDF) water. CDF water was used to prepare all experimental solutions and was prepared using a Milli-Q1 Purification system with a Biopak1 Polisher (Merck, Darmstadt, Germany). A stock solution of sodium hypochlorite (81 g/L) was used to prepare an intermediate solution of 0.5 mg/L with CDF water (pH 6). As described previously, L. interrogans and L. biflexa were incubated at 30˚C in liquid EMJH medium until reaching a concentration of approximately 108 Leptospira/mL. Leptospira cultures were centrifuged at 5,000 x g for 15 min, the supernatant discarded, the pellet washed in 0.9% NaCl, and finally resuspended in 0.9% NaCl. Leptospira suspensions were prepared by adding bacteria (final concentration of 104 Leptospira/mL) to a chlorine solution (0.5 mg/L). The free chlorine concentration was measured before and after addition of the Leptospira sus- pension using a Pocket colorimeter II (Hach Lange, Dusseldorf, Germany) after activation of the DPD reagent (N,N-diethyl-p-phenylene-diamine). Unreacted free chlorine was quenched by the addition of sodium thiosulfate (100 mg/L) to stop the activation reaction. After chlorine treatment, each sample was analyzed by culture, qPCR, and integrity qPCR. Environmental sample collection Thirty-four surface-water samples were collected in Paris, including an area which includes a controlled bathing section during the summer period (48.885441128096, 2.37411186180 21346). The surface water had not undergone any sanitation treatment. Samples were analyzed within 24 h of collection. Surface-water samples (500 mL) were con- centrated by successive centrifugation steps down to 400 μL. Half of the sample was treated with PMAxx (as described in the section on propidium monoazide treatments) and half remained untreated. All samples (with and without PMAxx) were then extracted and analyzed by qPCR. PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 5 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay Statistical analysis with GraphPad software In addition to its use in the persistence test to model bacterial inactivation, GraphPad Prism software version 8 (GraphPad, La Jolla, CA) was also used for statistical analysis. Normality of the distribution was assessed using the Shapiro-Wilk test. Paired groups were tested using the nonparametric Wilcoxon matched pairs signed rank test. P-values < 0.05 were considered significant. Results Leptospira multiplex qPCR First designed to operate separately, the three qPCR assays were adapted to be performed in a multiplex assay to obtain a single reaction (effectiveness demonstrated below in the section on the analytical sensitivity of the method). This was made possible through the use of different fluorophores. Total Leptospira, pathogenic Leptospira, and IPC were targeted by different dyes: FAM, Yakima Yellow, and TAMRA, respectively. Several enzymatic master mixes were tested to optimize the qPCR reaction (Fig 1). The Fast Virus 1-Step Master mix resulted in better detection limits than amplification using two other enzyme mixes, with a difference of approximately 10 CT between qPCRs. In addition, the effi- ciency and coefficient of determination (R2) showed the Fast Virus 1-step master mix to out- perform the others in this study (Table 1). As indicated previously, the two qPCR assays for detecting Leptospira and the IPC were first designed to operate independently. We compared the performance of the qPCR assays separately (simplex mode) or together (multiplex mode). The simplex and multiplex modes provided the same results for the 16S target (efficiency of approximately 100% and R2 of approximately 0.98) (Fig 2) and these two parameters were slightly higher in the multiplex mode (89% of efficiency of 89% and R2 of 0.987) than the simplex mode (efficiency of and 84% R2 of 0.984) for the LipL32 target (Table 2). Fig 1. qPCR tests on range of concentrations of the Leptospira 16S plasmid with various enzymatic mixes (Taqman1 Fast Virus 1-Step Master Mix, QuantiNova1 Multiplex Master Mix, Taqman1 Fast Advanced Master Mix). The colored dotted lines correspond to the number of cycles required to obtain the smallest detectable quantity of genome, shown by the black dotted line. https://doi.org/10.1371/journal.pone.0251901.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 6 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay Table 1. Comparison of R2 and efficiency between various enzymatic mixes. TaqMan1 Fast Virus 1-Step Master Mix QuantiNova1 Taqman1 Fast Advanced Master Mix R2 Efficacity (%) 0.984 95.8 0.996 81.9 0.994 71.3 The efficiency and R2 were compared between various enzymatic mixes from two suppliers: TaqMan1 (Applied Biosystems) and QuantiNova1 (Qiagen). https://doi.org/10.1371/journal.pone.0251901.t001 Ten-fold serial dilutions (ranging from 106 copies/μL to 103 copies/μL), followed by two- fold serial dilutions (from 103 copies/μL to 1 copy/μL), of L. interrogans serovar Manilae were used to determine the true analytical sensitivity of the detection method. The limit of detection (LoD) was determined as the quantity of plasmid that could be detected in 95% of the repli- cates. The limit of quantification (LoQ) was the lowest concentration of plasmid that could be properly quantified in a standard range. The LoD, LoQ, and amplification range were sepa- rately determined for each target (16S, LipL32, and IPC) to determine their own parameters. The LoD and LoQ of 16S were both at a CT value of 33.09, corresponding to 1 bacterium/well. For LipL32, the LoQ was at a CT value of 37.97, corresponding to 125 bacteria/well (Fig 3), and the LoD at 38.66, corresponding to 86 bacteria/well. We tested the specificity of the pan-Leptospira PCR on bacterial strains other than Leptos- pira. The assay was found to be specific for Leptospira spp., as none of the six other pathogenic organisms were amplified. Moreover, we tested the specificity of pathogenic-Leptospira PCR. None of the nonpathogenic Leptospira were detected by the assay. The sensitivity was also measured, and results were mentioned in S1 Table. Evaluation of the efficacy of disinfection treatments of Leptospira Addition of the integrity assay to the detection by qPCR enabled the specific detection of unal- tered bacteria and viable but non-cultivable (VBNC) bacteria in the samples [44]. An interca- lating agent was added and photoactivated to avoid the amplification of “free” or not protected DNA by the qPCR. This protocol was tested on Leptospira subjected to disinfection treatments (temperature, UV radiation, chlorine). The bacterial concentration was evaluated by three dif- ferent methods to validate the use of PMAqPCR: microscopic enumeration, colony-forming units by plating, and PMAqPCR (Table 3). Fig 2. Comparison between simplex and multiplex qPCR for the two targets: 16S gene and LipL32 gene; using range of plasmids. https://doi.org/10.1371/journal.pone.0251901.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 7 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay Table 2. Comparison of R2 and efficiency between the simplex and multiplex qPCR assays developed in this study. R2 Efficacity (%) rrs (16S) simplex 0.9827 104 LipL32 simplex 0.9846 84 rrs (16S) multiplex 0.9820 104 LipL32 multiplex 0.9873 89 The efficiency and R2 were compared between the simplex and multiplex modes for two targets: rrs (16S) gene for all Leptospira and LipL32 gene for pathogenic strains. https://doi.org/10.1371/journal.pone.0251901.t002 The titrations obtained by microscopy and plating were similar, giving approximately 107 Leptospira/mL. Quantification of the two genes indicated concentrations 25 to 60 times higher. Various experiments were independently carried out with fresh Leptospira suspensions. Thus, the initial concentration of bacteria could have differed. Nonetheless, this parameter did not interfere with the interpretation of results because the analysis was based on the log of inactivation. Heat inactivation was assessed by culture analysis and PMA-qPCR analysis. The pathogenic strain appeared to be more resistant than the saprophytic strain. Indeed, the curve of the slope from the culture analysis was lower for L. interrogans (-6.945) than L. biflexa (-8.211). How- ever, the temperature which induced a reduction of the concentration of the bacteria by half was broadly similar for both strains (32.12˚C for L. interrogans and 32.91˚C for L. biflexa) (Fig 4). Leptospira appeared to not be cultivable on EMJH solid plates beyond 37˚C and not detect- able by PMA-qPCR beyond 55˚C. The time of exposure to free chlorine was calculated from the kinetics of free chlorine con- sumption and adjusted to reach a CT value equal to 10 mg.min/L (Fig 5). This CT value corre- sponded to the concentration of this powerful oxidant (mg/L) multiplied by the time (min) of exposure. In this case, we selected the time at which the area under the free chlorine CT curves Fig 3. Determination by PMAqPCR of LoD for the LipL32 gene using serial dilution (eight replicates for each dilution). https://doi.org/10.1371/journal.pone.0251901.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 8 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay Table 3. Comparison of Leptospira detection between microscopy, plating and molecular analysis. Microscopic enumeration Plating enumeration (EMJH) PMAqPCR (PBS) PMAqPCR (EMJH) L. biflexa Patoc L. interrogans Manilae n = 9 1.00E+07 1.00E+07 n = 9 1.66E+07 1.63E+07 (16S) (16S) (LipL32) n = 6 4.25E+08 1.11E+09 7.45E+08 n = 6 4.21E+08 7.10E+08 4.65E+08 Microscopic enumeration was assessed by dark field microscopy using a Petroff-Hausser chamber. Plating enumeration was performed using EMJH semi-solid medium after seven days of incubation for L. biflexa and more than two weeks of incubation for L. interrogans. Detection by molecular analysis was performed in PBS and EMJH liquid medium. https://doi.org/10.1371/journal.pone.0251901.t003 was 10 mg.min/L. The time of exposure ranged from 0 to 27 min. Chlorine-dependent inacti- vation was assessed by culture and molecular methods. We observed a slight decrease in the concentration of L. interrogans in the presence of the hypochlorous acid during the assay by PMAqPCR. The effect was stronger for L. biflexa, with 1.5 log removal at CT = 4mg.min/L and up to 2 logs of inactivation at the end of the experiment (CT = 10mg.min/L). Both strains showed strong sensitivity to chlorine treatment by culture assay (Fig 6). No L. interrogans grew on the EMJH culture media after exposure to the lowest chlorine dose (CT = 0.001 mg.min/L), whereas total inactivation of L. biflexa occurred at CT = 0.1 mg.min/L. UV254 light inactivates microorganisms by targeting their nucleic acids, resulting in the inhibition of DNA replication and thus their growth in culture. Culture analysis showed the same kinetics for the action of UV radiation for both strains (Fig 4). Three logs of removal were achieved at 10 mJ/cm2, whereas 40 J/cm2 is currently used in drinking water treatment plants. UV light kills cells by damaging their DNA and does not usually result in cell lysis; the Fig 4. Heat and UV inactivation for Leptospira biflexa serovar Patoc (A&B, respectively) and for Leptospira interrogans serovar Manilae (C&D, respectively). The X-axis indicates the temperature to which Leptospira were exposed for 1 h (A, C) or the UV dose of exposure (B, D). The left Y-axis represents the median bacterial culture results (in colony-forming units) after several days of incubation in EMJH semi-solid medium at 30˚C (circle). The right Y- axis represents the median PMAxxqPCR results (in genome units) after extraction and molecular biology assay based on the 16S gene (triangle). https://doi.org/10.1371/journal.pone.0251901.g004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 9 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay Fig 5. Free chlorine consumption kinetics for L. biflexa serovar Patoc and L. interrogans serovar Manilae. The time of exposure to free chlorine was adjusted to maintain a CT-value equal to 10 mg.min.L- 1. https://doi.org/10.1371/journal.pone.0251901.g005 lack of decrease for PMAqPCR signal is therefore consistent with the preservation of mem- brane integrity. Proof-of-concept of application for environmental monitoring. Our PMAxx-qPCR method was tested on 32 surface water samples, with and without PMAxx. The results are summarized in Fig 7. A Wilcoxon signed-rank test was performed to show whether the data with PMAxx were significantly different from those without. The two sets of data were signifi- cantly different (p-value < 0.0001). The genome concentration using PMAxx-qPCR was con- sistently lower than that estimated using conventional qPCR in 100% of cases, signifying that a significant proportion of target DNA was “free” or inside permeable bacteria. Adding PMAxx reduced overestimation by amplifying only the genomes of non-permeable bacteria (poten- tially viable bacteria). Fig 6. Chlorine inactivation for L. biflexa serovar Patoc and L. interrogans serovar Manilae by culture and PMAqPCR. Results were gathered for the two strains L. biflexa serovar Patoc (blue, red) and L. interrogans serovar Manilae (green, orange). The X-axis indicates the CT value: powerful oxidant concentration (mg/L) � time (min) of exposure. The left Y-axis represents the median bacterial culture results (in colony-forming units) after several days of incubation in EMJH semi-solid medium at 30˚C (triangle). The right Y-axis represents the median PMAxxqPCR results (in genome units) after extraction and molecular biology assay based on the 16S gene (circle). https://doi.org/10.1371/journal.pone.0251901.g006 PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 10 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay Fig 7. Distribution and median of Leptospira genome (16S gene) with (red) and without PMAxx (blue) on environmental samples. This analyze was based on 34 environmental freshwater samples collected in the Ourcq canal (Paris) in June-September 2018. https://doi.org/10.1371/journal.pone.0251901.g007 Discussion Leptospirosis is an emerging waterborne zoonosis of global importance for both humans and animals. However, there is also an urgent need for a robust and easy-to-use Leptospira detection method for environmental samples. Traditional culture methods are fastidious and Leptospira are slow-growing bacteria and cultures can be contaminated by other microorganisms [28]. Alternative detection methods, such as specific qPCR, have thus been proposed but no gold- standard has been implemented for environmental monitoring. In general, molecular methods allow a rapid measurement of the pathogen concentration coupled with high sensitivity and specificity. In the present study, we implemented a triplex qPCR based on Taqman technology for the detection and quantification of both pathogenic and saprophytic Leptospira spp. using the rrs gene and specifically pathogenic strains based on LipL32 gene amplification. An internal competitive amplification control was added to the analysis to evaluate the inhibition of amplifi- cation resulting from the samples. In addition, we coupled this method with a bacterial integrity assay. A triplex PCR (two genes and an internal control) was designed and used to discriminate Leptospira from subclades P1 and P2 in animal samples [45]. This multiplex qPCR was set up, according the MIQE guidelines [46], to detect all Leptospira and selectively discriminate between pathogenic and saprophytic strains. An internal control was also added as a supplementary mon- itor to check for inhibition of amplification in environmental samples. Genome amplification by qPCR also has certain limitations. Amplification of any targeted DNA present in the sample makes it impossible to distinguish between live and dead cells, resulting in potential overestimation of the bacterial concentration and the risk of infection. It was also shown that the use of propidium iodide to distinguish between living bacteria from dead bacteria was not relevant under certain experimental conditions and in particular to esti- mate the effect of chlorination [47]. Here, we used propidium monoazide (PMAxx) to avoid PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 11 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay the amplification of DNA from degraded cells, permeable to the dye, and solely amplify DNA from potentially viable bacteria [35] to reduce potential overestimation. Validation of the qPCR method was performed by comparing the molecular results (using PMAxx coupled with the RT qPCR measurement) with culture-based methods. A higher num- ber of Leptospira was counted by the qPCR. This difference can be explained by various fac- tors, such as colonies not issued from one bacteria (i.e. aggregated bacteria), genome multiplication before cells separation, count of viable but non cultivable bacteria and multiple copies of the 16S gene in Leptospira [48]. We compared the survival of pathogenic and saprophytic strains of Leptospira in several environments. Heat inactivation experiments revealed different kinetics between pathogenic and saprophytic strains, the pathogenic strain being more tolerant to heat. This relative toler- ance may explain their higher prevalence in tropical areas as well as the role of these strains in human infections. Other inactivation processes did not show any significant differences in inactivation kinetics between the two strains. Our results based on bacterial culture methods suggest that Leptospira is rapidly inactivated by free chlorine. However, the use of PMAqPCR shows that leptospiral membranes would not be directly damaged by free chlorine. Several studies have shown that free chlorine inactivates E. coli without damaging its cell membranes [49–52] further indicating that the use of PMAqPCR is not appropriate to determine the capacity of chlorine to kill bacteria. In absence of impact of disinfection treatment, it was essential to also consider that the absence of cultivability does not necessarily indicate cell lysis as bacteria could remain as a “viable but non-cultivable” state [53]. Contrary to classical PCR, the integrity PCR approach allowed the use of more adaptable and faster molecular methods to assess inactivation efficiencies. Due to interfering flora or organic matter, the culture is too complicated to implement, especially on complex samples whose matrix negatively impacts the re-cultivation of Leptospira. Certain physicochemical parameters (for example, salty water) can alter the survival of Lep- tospira [24], whereas other parameters can provide a protective effect. For example, the pres- ence of organic matter or biofilms could increase the survival of Leptospira survival in aquatic environments [16, 54]. The PMAxx approach showed certain limitations concerning its use to evaluate the effi- ciency of disinfection treatments of Leptospira. Although useful results were obtained after heat or low-level free chlorine exposure, the use of PMAxx was not informative for the analysis of UV treatment. It is possible that treatments or conditions that affect bacterial integrity (tem- perature, chlorine) allow PMAxx to access the genome, contrary to UV radiation. However, when only DNA was targeted for the inactivation of the microorganism, PMAxx had no effect in improving the determination of Leptospira sensitivity (Fig 7). Such an observation has already been reported for a virus assay [55]. To date, Leptospira are not considered when investigating microorganisms in water. Our methods allowed us to obtain information on the presence and integrity of such bacte- ria in Parisian surface-water samples (Fig 7). This proof of concept should be applied to answer other questions, such as the influence of seasonal variations or the impact of rodent control campaigns. The development of a sensitive qPCR method using a rapid reverse- transcription step targeting the rrs (16S) or LipL32 genes improved the sensitivity of detec- tion [56]. Within the sampling area, the median concentration was approximately 103 eq. bacteria/L. Further studies are under way to determine whether it would be relevant and useful to routinely use this method to monitor Leptospira in surface water. These results could be considered to establish threshold alerts, leading to restrictions of access, after events that favor bacterial contamination (heavy rain, flooding, etc.). The impact of PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 12 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay various parameters, such as seasons, climatic conditions, and human activities, on Leptos- pira dynamics needs to be evaluated. Analysis of the presence of viable Leptospira in the environment and the measurement of the effectiveness of treatments are essential. The advantages of combining PMAxx addition and RT qPCR methods to detect low levels of non-permeable pathogens in water is now accepted [57]. The use of PMAxx combined with qPCR is therefore one possible solution for Leptospira measurement; it is time saving and avoids overestimation. By detecting non-perme- able and VBNC bacteria, this method pinpointed and prevented a Leptospira outbreak origi- nating from environmental water [58]. These data could be useful for quantitative microbiological risk assessment (QMRA) approaches. Although it is now well known that these bacteria can have a high impact on public health in endemic areas [59] or during specific seasons [60, 61], little is known about the environmental concentration in urbanized areas in Europe affected by the proliferation of rodents, which are the main animal reservoir of Leptospira. With ongoing social changes, water-related activities have increased in these regions, with the installation of urban beaches, bathing areas, and aquatic activities in areas with non-treated surface water. Our PMAqPCR method will be further used to better evaluate the presence of pathogenic Lep- tospira in bathing areas in Paris, France. The monitoring of microbial contamination is a requirement for establishing microbial risk assessment guidelines. Moreover, regular monitoring of Leptospira could help to provide a better description of infection events. Despite an increase in reported cases of leptospirosis and evidence that most cases are due to exposure to contaminated water, regulations are still based on fecal indicators and do not yet include pathogenic Leptospira, probably due to the absence of reliable measure- ment methods. Indeed, only Escherichia coli and intestinal enterococci are analyzed in France (French Public Health Code-D. 1332-15D1332-15). The only existing recommen- dations related to the risk of Leptospira exposure are to avoid bathing with skin lesions or in uncontrolled bathing areas. Leptospira monitoring could be implemented in fresh- water swimming facilities to improve awareness of Leptospira exposure. This could consist of identifying sources of pollution prone to affect water quality and the health of bathers. By detecting Leptospira in the environmental water during flooding, this assay can also contribute to early warning of potential outbreaks of leptospirosis. Supporting information S1 Table. Microorganism strains used for specificity tests and results from the TaqMan real-time multiplex (LipL32 and 16S) PCR assays. (DOCX) S2 Table. Leptospira sequences and their NCBI accession numbers required to design the primers and probe for the 16S qPCR. (DOCX) S3 Table. Leptospira sequences and their NCBI accession numbers required to design primers and probe for the LipL32 qPCR. (DOCX) Acknowledgments We are very grateful to the sampling department of Eau de Paris for providing samples to the laboratory. We thank the staff of the National Reference Center for Leptospirosis for support and the processing of some of the samples. PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 13 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay Author Contributions Conceptualization: Mathieu Picardeau, Laurent Moulin. Formal analysis: Elise Richard. Funding acquisition: Mathieu Picardeau, Laurent Moulin. Investigation: Elise Richard, Se´bastien Wurtzer. Methodology: Elise Richard, Pascale Bourhy, Se´bastien Wurtzer. Supervision: Mathieu Picardeau. Validation: Laurent Moulin, Se´bastien Wurtzer. Writing – original draft: Elise Richard. Writing – review & editing: Mathieu Picardeau, Laurent Moulin, Se´bastien Wurtzer. References 1. Costa F, Hagan JE, Calcagno J, Kane M, Torgerson P, Martinez-Silveira MS, et al. Global Morbidity and Mortality of Leptospirosis: A Systematic Review. Small PLC, editor. PLoS Negl Trop Dis. 2015 Sep 17; 9(9):e0003898. https://doi.org/10.1371/journal.pntd.0003898 PMID: 26379143 2. Tique V, Mattar S, Miranda J, Oviedo M, Noda A, Montes E, et al. Clinical and Epidemiological Status of Leptospirosis in a Tropical Caribbean Area of Colombia. BioMed Res Int. 2018 May 29; 2018:1–8. 3. Biscornet L, de Comarmond J, Bibi J, Mavingui P, Dellagi K, Tortosa P. An Observational Study of Human Leptospirosis in Seychelles.: 10. 4. Pijnacker R, Goris MGA, te Wierik MJM, Broens EM, van der Giessen JWB, de Rosa M, et al. Marked increase in leptospirosis infections in humans and dogs in the Netherlands, 2014. Eurosurveillance [Internet]. 2016 Apr 28 [cited 2021 Apr 7]; 21(17). Available from: https://www.eurosurveillance.org/ content/10.2807/1560-7917.ES.2016.21.17.30211. https://doi.org/10.2807/1560-7917.ES.2016.21.17. 30211 PMID: 27168584 5. Bourhy P. DIAGNOSTIC, SURVEILLANCE ET EPIDEMIOLOGIE DE LA LEPTOSPIROSE EN FRANCE / DIAGNOSIS, SURVEILLANCE, AND EPIDEMIOLOGY OF LEPTOSPIROSIS IN FRANCE.: 7. 6. Bharti AR, Nally JE, Ricaldi JN, Matthias MA, Diaz MM, Lovett MA, et al. Leptospirosis: a zoonotic dis- ease of global importance. Lancet Infect Dis. 2003 Dec; 3(12):757–71. https://doi.org/10.1016/s1473- 3099(03)00830-2 PMID: 14652202 7. Morgan J, Bornstein SL, Karpati AM, Bruce M, Bolin CA, Austin CC, et al. Outbreak of Leptospirosis among Triathlon Participants and Community Residents in Springfield, Illinois, 1998. Clin Infect Dis. 2002 Jun 15; 34(12):1593–9. https://doi.org/10.1086/340615 PMID: 12032894 8. Lau CL, Smythe LD, Craig SB, Weinstein P. Climate change, flooding, urbanisation and leptospirosis: fuelling the fire? Trans R Soc Trop Med Hyg. 2010 Oct; 104(10):631–8. https://doi.org/10.1016/j.trstmh. 2010.07.002 PMID: 20813388 9. Munoz-Zanzi C, Groene E, Morawski BM, Bonner K, Costa F, Bertherat E, et al. A systematic literature review of leptospirosis outbreaks worldwide, 1970–2012. Rev Panam Salud Pu´ blica. 2020 Jul 15; 44:1. https://doi.org/10.26633/RPSP.2020.78 PMID: 32684917 10. Mwachui MA, Crump L, Hartskeerl R, Zinsstag J, Hattendorf J. Environmental and Behavioural Deter- minants of Leptospirosis Transmission: A Systematic Review. Small PLC, editor. PLoS Negl Trop Dis. 2015 Sep 17; 9(9):e0003843. https://doi.org/10.1371/journal.pntd.0003843 PMID: 26379035 11. Abgueguen P, Delbos V, Blanvillain J, Chennebault JM, Cottin J, Fanello S, et al. Clinical aspects and prognostic factors of leptospirosis in adults. Retrospective study in France. J Infect. 2008 Sep; 57 (3):171–8. https://doi.org/10.1016/j.jinf.2008.06.010 PMID: 18656263 12. Guillois Y, Bourhy P, Ayral F, Pivette M, Decors A, Aranda Grau JH, et al. An outbreak of leptospirosis among kayakers in Brittany, North-West France, 2016. Eurosurveillance [Internet]. 2018 Nov 29 [cited 2020 Aug 10]; 23(48). Available from: https://www.eurosurveillance.org/content/10.2807/1560-7917. ES.2018.23.48.1700848. 13. Cacciapuoti B, Ciceroni L, Maffei C, Stanislao FD, Strusi P, Calegari L, et al. A WATERBORNE OUT- BREAK OF LEPTOSPIROSIS. Am J Epidemiol. 1987 Sep; 126(3):535–45. https://doi.org/10.1093/ oxfordjournals.aje.a114686 PMID: 3618584 PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 14 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay 14. Ramakrishnan R, Patel MS, Gupte MD, Manickam P, Venkataraghavan S. An institutional outbreak of leptospirosis in Chennai, South India. J Commun Dis. 2003 Mar; 35(1):1–8. PMID: 15239298 15. Vincent AT, Schiettekatte O, Goarant C, Neela VK, Bernet E, Thibeaux R, et al. Revisiting the taxonomy and evolution of pathogenicity of the genus Leptospira through the prism of genomics. Martins EAL, edi- tor. PLoS Negl Trop Dis. 2019 May 23; 13(5):e0007270. https://doi.org/10.1371/journal.pntd.0007270 PMID: 31120895 16. Thibeaux R, Geroult S, Benezech C, Chabaud S, Soupe´-Gilbert M-E, Girault D, et al. Seeking the envi- ronmental source of Leptospirosis reveals durable bacterial viability in river soils. Munoz-Zanzi C, editor. PLoS Negl Trop Dis. 2017 Feb 27; 11(2):e0005414. https://doi.org/10.1371/journal.pntd.0005414 PMID: 28241042 17. Faine S, Adler B, Bolin C, Perolat P. “Leptospira” and leptospirosis [Internet]. MediSci; 1999 [cited 2020 Oct 13]. Available from: https://research.monash.edu/en/publications/leptospira-and-leptospirosis. 18. Adler B, editor. Leptospira and Leptospirosis [Internet]. Berlin, Heidelberg: Springer Berlin Heidelberg; 2015 [cited 2020 Oct 13]. (Current Topics in Microbiology and Immunology; vol. 387). Available from: http://link.springer.com/10.1007/978-3-662-45059-8. 19. Goarant C. Leptospirosis: risk factors and management challenges in developing countries. Res Rep Trop Med. 2016 Sep;Volume 7:49–62. https://doi.org/10.2147/RRTM.S102543 PMID: 30050339 20. Schaechter M. Escherichia coli and Salmonella 2000: the View From Here. Microbiol Mol Biol Rev. 2001 Mar; 65(1):119–30. https://doi.org/10.1128/MMBR.65.1.119-130.2001 PMID: 11238988 21. Loureiro AP, Martins G, Pinto P, Narduche L, Teixeira RC, Lilenbaum W. Usage of a selective media (EMJH-STAFF) in primary culturing of pathogenic leptospires from bovine clinical samples. Lett Appl Microbiol. 2015 Dec; 61(6):603–6. https://doi.org/10.1111/lam.12501 PMID: 26408270 22. Chakraborty A, Miyahara S, Villanueva SYAM, Saito M, Gloriani NG, Yoshida S. A novel combination of selective agents for isolation of Leptospira species: Selective medium for Leptospira. Microbiol Immu- nol. 2011 Jul; 55(7):494–501. https://doi.org/10.1111/j.1348-0421.2011.00347.x PMID: 21545510 23. Huang K. A comprehensive insight into bacterial virulence in drinking water using 454 pyrosequencing and Illumina high-throughput sequencing. Ecotoxicol Environ Saf. 2014; 8. https://doi.org/10.1016/j. ecoenv.2014.07.029 PMID: 25129220 24. Chang SL, Buckingham M, Taylor MP. Studies on Leptospira Icterohaemorrhagiae: IV. Survival in Water and Sewage: Destruction in Water by Halogen Compounds, Synthetic Detergents, and Heat. J Infect Dis. 1948 May 1; 82(3):256–66. https://doi.org/10.1093/infdis/82.3.256 PMID: 18864110 25. Andre-Fontaine G, Aviat F, Thorin C. Waterborne Leptospirosis: Survival and Preservation of the Viru- lence of Pathogenic Leptospira spp. in Fresh Water. Curr Microbiol. 2015 Jul; 71(1):136–42. https://doi. org/10.1007/s00284-015-0836-4 PMID: 26003629 26. Bierque E, Soupe´ -Gilbert M-E, Thibeaux R, Girault D, Guentas L, Goarant C. Leptospira interrogans Retains Direct Virulence After Long Starvation in Water. Curr Microbiol. 2020 Oct; 77(10):3035–43. https://doi.org/10.1007/s00284-020-02128-7 PMID: 32683468 27. Ahmed A, Engelberts MFM, Boer KR, Ahmed N, Hartskeerl RA. Development and Validation of a Real- Time PCR for Detection of Pathogenic Leptospira Species in Clinical Materials. Bereswill S, editor. PLoS ONE. 2009 Sep 18; 4(9):e7093. https://doi.org/10.1371/journal.pone.0007093 PMID: 19763264 28. Podgorsˇek D, Ruzˇić-Sabljić E, Logar M, Pavlović A, Remec T, Baklan Z, et al. Evaluation of real-time PCR targeting the lipL32 gene for diagnosis of Leptospira infection. BMC Microbiol. 2020 Dec; 20(1):59. https://doi.org/10.1186/s12866-020-01744-4 PMID: 32160864 29. Stoddard RA, Gee JE, Wilkins PP, McCaustland K, Hoffmaster AR. Detection of pathogenic Leptospira spp. through TaqMan polymerase chain reaction targeting the LipL32 gene. Diagn Microbiol Infect Dis. 2009; 9. https://doi.org/10.1016/j.diagmicrobio.2009.03.014 PMID: 19395218 30. Mohd Ali MR, Mohd Safee AW, Ismail NH, Abu Sapian R, Mat Hussin H, Ismail N, et al. Development and validation of pan- Leptospira Taqman qPCR for the detection of Leptospira spp. in clinical speci- mens. Mol Cell Probes. 2018 Apr; 38:1–6. https://doi.org/10.1016/j.mcp.2018.03.001 PMID: 29524642 31. Levett PN. Detection of pathogenic leptospires by real-time quantitative PCR. J Med Microbiol. 2005 Jan 1; 54(1):45–9. https://doi.org/10.1099/jmm.0.45860-0 PMID: 15591254 32. Go¨kmen TG, Soyal A, Kalayci Y, O¨ nlen C, Ko¨ ksal F. COMPARISON OF 16S rRNA-PCR-RFLP, LipL32-PCR AND OmpL1-PCR METHODS IN THE DIAGNOSIS OF LEPTOSPIROSIS. Rev Inst Med Trop São Paulo [Internet]. 2016 [cited 2020 Jan 2]; 58(0). Available from: http://www.scielo.br/scielo. php?script=sci_arttext&pid=S0036-46652016005000248&lng=en&tlng=en. 33. Slack AT, Symonds ML, Dohnt MF, Smythe LD. [No title found]. BMC Microbiol. 2006; 6(1):95. 34. Palaniappan RUM, Chang Y-F, Chang C-F, Pan MJ, Yang CW, Harpending P, et al. Evaluation of lig- based conventional and real time PCR for the detection of pathogenic leptospires. Mol Cell Probes. 2005 Apr; 19(2):111–7. https://doi.org/10.1016/j.mcp.2004.10.002 PMID: 15680212 PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 15 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay 35. Casanovas-Massana A, Pedra GG, Wunder EA, Diggle PJ, Begon M, Ko AI. Quantification of Leptos- pira interrogans Survival in Soil and Water Microcosms. Appl Environ Microbiol [Internet]. 2018 Jun 18 [cited 2020 Oct 13]; 84(13). Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6007094/. https://doi.org/10.1128/AEM.00507-18 PMID: 29703737 36. Roussel C, Galia W, Leriche F, Chalancon S, Denis S, Van de Wiele T, et al. Comparison of conven- tional plating, PMA-qPCR, and flow cytometry for the determination of viable enterotoxigenic Escheri- chia coli along a gastrointestinal in vitro model. Appl Microbiol Biotechnol. 2018 Nov; 102(22):9793– 802. https://doi.org/10.1007/s00253-018-9380-z PMID: 30238141 37. Ellinghausen HC, Mccullough WG. NUTRITION OF LEPTOSPIRA POMONA AND GROWTH OF 13 OTHER SEROTYPES: FRACTIONATION OF OLEIC ALBUMIN COMPLEX AND A MEDIUM OF BOVINE ALBUMIN AND POLYSORBATE 80. Am J Vet Res. 1965 Jan; 26:45–51. PMID: 14266934 38. Johnson RC, Harris VG. Differentiation of Pathogenic and Saprophytic Leptospires I. Growth at Low Temperatures. J Bacteriol. 1967; 94(1):27–31. https://doi.org/10.1128/JB.94.1.27-31.1967 PMID: 6027998 39. Abdulmawjood A, Roth S, Bu¨ lte M. Two methods for construction of internal amplification controls for the detection of Escherichia coli O157 by polymerase chain reaction. Mol Cell Probes. 2002 Oct; 16 (5):335–9. https://doi.org/10.1006/mcpr.2002.0431 PMID: 12477437 40. Wurtzer S, Prevost B, Lucas F, Moulin L. Detection of enterovirus in environmental waters: A new opti- mized method compared to commercial real-time RT-qPCR kits. J Virol Methods. 2014 Sep 4; 209. https://doi.org/10.1016/j.jviromet.2014.08.016 PMID: 25196451 41. Karim MR, Fout GS, Johnson CH, White KM, Parshionikar SU. Propidium monoazide reverse transcrip- tase PCR and RT-qPCR for detecting infectious enterovirus and norovirus. J Virol Methods. 2015 Jul; 219:51–61. https://doi.org/10.1016/j.jviromet.2015.02.020 PMID: 25796356 42. Coudray-Meunier C, Fraisse A, Martin-Latil S, Guillier L, Perelle S. Discrimination of infectious hepatitis A virus and rotavirus by combining dyes and surfactants with RT-qPCR. BMC Microbiol. 2013; 13 (1):216. https://doi.org/10.1186/1471-2180-13-216 PMID: 24083486 43. Prevost B, Goulet M, Lucas FS, Joyeux M, Moulin L, Wurtzer S. Viral persistence in surface and drink- ing water: Suitability of PCR pre-treatment with intercalating dyes. Water Res. 2016 Mar; 91:68–76. https://doi.org/10.1016/j.watres.2015.12.049 PMID: 26773484 44. Nocker A, Sossa-Fernandez P, Burr MD, Camper AK. Use of Propidium Monoazide for Live/Dead Dis- tinction in Microbial Ecology. Appl Environ Microbiol. 2007 Aug 15; 73(16):5111–7. https://doi.org/10. 1128/AEM.02987-06 PMID: 17586667 45. Pe´ rez LJ, Lanka S, DeShambo VJ, Fredrickson RL, Maddox CW. A Validated Multiplex Real-Time PCR Assay for the Diagnosis of Infectious Leptospira spp.: A Novel Assay for the Detection and Differentia- tion of Strains From Both Pathogenic Groups I and II. Front Microbiol. 2020 Mar 20; 11:457. https://doi. org/10.3389/fmicb.2020.00457 PMID: 32265882 46. Bustin SA, Benes V, Garson JA, Hellemans J, Huggett J, Kubista M, et al. The MIQE Guidelines: Mini- mum Information for Publication of Quantitative Real-Time PCR Experiments. Clin Chem. 2009 Apr 1; 55(4):611–22. https://doi.org/10.1373/clinchem.2008.112797 PMID: 19246619 47. Virto R, Mañas P, A´ lvarez I, Condon S, Raso J. Membrane Damage and Microbial Inactivation by Chlo- rine in the Absence and Presence of a Chlorine-Demanding Substrate. Appl Environ Microbiol. 2005 Sep; 71(9):5022–8. https://doi.org/10.1128/AEM.71.9.5022-5028.2005 PMID: 16151082 48. Fukunaga M, Mifuchi I. Unique organization of Leptospira interrogans rRNA genes. J Bacteriol. 1989; 171(11):5763–7. https://doi.org/10.1128/jb.171.11.5763-5767.1989 PMID: 2808297 49. Mizozoe Otaki, Aikawa. The Mechanism of Chlorine Damage Using Enhanced Green Fluorescent Pro- tein-Expressing Escherichia coli. Water. 2019 Oct 16; 11(10):2156. 50. Bridges DF, Lacombe A, Wu VCH. Integrity of the Escherichia coli O157:H7 Cell Wall and Membranes After Chlorine Dioxide Treatment. Front Microbiol. 2020 May 15; 11:888. https://doi.org/10.3389/fmicb. 2020.00888 PMID: 32499765 51. Cho M, Kim J, Kim JY, Yoon J, Kim J-H. Mechanisms of Escherichia coli inactivation by several disinfec- tants. Water Res. 2010 Jun; 44(11):3410–8. https://doi.org/10.1016/j.watres.2010.03.017 PMID: 20427068 52. Kazama S, Otaki M. Quantitative Analysis of the Inactivation Mechanisms of Escherichia coli by a Newly Developed Method Using Propidium Monoazide. J Water Environ Technol. 2013; 11(6):507–17. 53. Xu L, Zhang C, Xu P, Wang XC. Mechanisms of ultraviolet disinfection and chlorination of Escherichia coli: Culturability, membrane permeability, metabolism, and genetic damage. J Environ Sci. 2018 Mar; 65:356–66. PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 16 / 17 PLOS ONEEffect of disinfection agents on Leptospira in water using a high sensitivity integrity-qPCR assay 54. Barragan VA, Mejia ME, Tra´vez A, Zapata S, Hartskeerl RA, Haake DA, et al. Interactions of Leptospira with Environmental Bacteria from Surface Water. Curr Microbiol. 2011 Jun; 62(6):1802–6. https://doi. org/10.1007/s00284-011-9931-3 PMID: 21479795 55. Prevost B, Lucas FS, Goncalves A, Richard F, Moulin L, Wurtzer S. Large scale survey of enteric viruses in river and waste water underlines the health status of the local population. Environ Int. 2015 Jun; 79:42–50. https://doi.org/10.1016/j.envint.2015.03.004 PMID: 25795193 56. Waggoner JJ, Balassiano I, Abeynayake J, Sahoo MK, Mohamed-Hadley A, Liu Y, et al. Sensitive Real-Time PCR Detection of Pathogenic Leptospira spp. and a Comparison of Nucleic Acid Amplifica- tion Methods for the Diagnosis of Leptospirosis. Lin B, editor. PLoS ONE. 2014 Nov 7; 9(11):e112356. https://doi.org/10.1371/journal.pone.0112356 PMID: 25379890 57. Gensberger ET, Polt M, Konrad-Ko¨szler M, Kinner P, Sessitsch A, Kostić T. Evaluation of quantitative PCR combined with PMA treatment for molecular assessment of microbial water quality. Water Res. 2014 Dec; 67:367–76. https://doi.org/10.1016/j.watres.2014.09.022 PMID: 25459225 58. Golpayegani A, Douraghi M, Rezaei F, Alimohammadi M, Nodehi RN. Propidium monoazide–quantita- tive polymerase chain reaction (PMA-qPCR) assay for rapid detection of viable and viable but non-cul- turable (VBNC) Pseudomonas aeruginosa in swimming pools. J Environ Health Sci Eng. 2019 Jun; 17 (1):407–16. https://doi.org/10.1007/s40201-019-00359-w PMID: 31297217 59. Pappas G, Papadimitriou P, Siozopoulou V, Christou L, Akritidis N. The globalization of leptospirosis: worldwide incidence trends. Int J Infect Dis. 2008 Jul; 12(4):351–7. https://doi.org/10.1016/j.ijid.2007. 09.011 PMID: 18055245 60. Bouscaren N, Benoit de Coignac C, Lastère S, Musso D, Teissier Y, Formont J, et al. Leptospirosis in French Polynesia: 11 years of surveillance data, 2007–2017. New Microbes New Infect. 2019 May; 29:100518. https://doi.org/10.1016/j.nmni.2019.100518 PMID: 30899522 61. Joshi YP, Kim E-H, Cheong H-K. The influence of climatic factors on the development of hemorrhagic fever with renal syndrome and leptospirosis during the peak season in Korea: an ecologic study. BMC Infect Dis. 2017 Dec; 17(1):406. https://doi.org/10.1186/s12879-017-2506-6 PMID: 28592316 PLOS ONE \| https://doi.org/10.1371/journal.pone.0251901 May 26, 2021 17 / 17 PLOS ONE
10.1371_journal.pone.0262786	RESEARCH ARTICLE Factors influencing nurses’ intention to care for patients with COVID-19: Focusing on positive psychological capital and nursing professionalism Sun-a Jeong1, Jinhee KimID 2* 1 Department of Nursing, Chosun University Hospital, Gwangju, South Korea, 2 Department of Nursing, College of Medicine, Chosun University, Gwangju, South Korea a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 * jinheeara@chosun.ac.kr Abstract Purpose OPEN ACCESS Citation: Jeong S-a, Kim J (2022) Factors influencing nurses’ intention to care for patients with COVID-19: Focusing on positive psychological capital and nursing professionalism. PLoS ONE 17(1): e0262786. https://doi.org/10.1371/journal. pone.0262786 Editor: Ce´sar Leal-Costa, Murcia University, Spain, SPAIN Received: August 25, 2021 Accepted: January 4, 2022 Published: January 19, 2022 Copyright: © 2022 Jeong, Kim. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting information files. Funding: This study was supported by research fund from Chosun University (K206904004, 2021). The funder did not play any role in the conduct or publication of the study. Competing interests: The authors have declared that no competing interests exist. It is necessary to identify factors that influence nurses’ intention to care for coronavirus dis- ease 2019 (COVID-19) patients to improve the quality of care during the pandemic. This study identifies factors that influence nurses’ intention to care for COVID-19 patients, focus- ing on positive psychological capital and nursing professionalism. Methods This study adopted a descriptive correlational design. Data were collected between August 16 and August 30, 2020, through self-administered questionnaires from 148 bedside nurses caring for COVID-19 patients, from four hospitals designated for COVID-19 treatment. Modi- fied versions of the Nursing Intention Questionnaire for SARS Patient Care, Psychological Capital Questionnaire, and Hall’s Professional Inventory were used. The collected data were analyzed using stepwise multiple regression. Results In total, 165 questionnaires were distributed, and 148 questionnaires (89.7%) were included in the final analysis. Factors influencing nurses’ intention to care were: age (30<: β = .18, p = .026; �50: β = .23, p = .005), department (ICU: β = -.26, p = .001), sufficient clinical experience and skills to care for COVID-19 patients (sufficient: β = .18, p = .019), and posi- tive psychological capital (β = .22, p = .044). The model’s explanatory power (R2) was 48%. Conclusions Strategies to increase nurses’ positive psychological capital are necessary to improve nurs- ing care quality by increasing intention to care when facing novel infectious diseases such as COVID-19. Furthermore, adequate education and training on managing novel infectious diseases should be implemented to provide nurses with relevant experience and skills regarding caring for patients infected with these diseases. Through various studies, PLOS ONE \| https://doi.org/10.1371/journal.pone.0262786 January 19, 2022 1 / 13 PLOS ONEFactors influencing nurses’ intention to care for patients with COVID-19 strategies for improving nurses’ positive psychological capital need to be suggested to improve the quality of care by increasing the nurses’ intention to care during the emergence of a novel infectious disease, such as COVID-19. Additionally, adequate education and training on managing the novel infectious diseases, sufficient for the nurses to believe they have the experience and skills for caring for the infected patients, will be needed. Introduction Following the first case of COVID-19 in Wuhan, Hubei Province, China in December 2019, the first confirmed case in South Korea was reported in January 2020 [1]. COVID-19 spread worldwide rapidly, and in March 2020, the World Health Organization (WHO) officially declared it a pandemic [2]. Currently, approximately 176,532,000 confirmed cases and 3,819,000 deaths have been reported worldwide [3], while in Korea, approximately 150,000 confirmed cases (cumulative) and 2,000 deaths have been reported [4]. Although most coun- tries have implemented strict social distancing measures such as school closures and restricting gatherings, the dissemination rate of the virus has not significantly decreased [3, 4]. COVID-19 patients require care from clinicians including nurses. Among clinicians, nurses are particularly affected by COVID-19; they have to provide direct bedside care including medication administration, sample collection, administration of intravenous injections, venti- lator management, and monitoring patient status while wearing personal protective equip- ment [5]. Most nurses were unprepared for the duties assigned to them with respect to nursing care of COVID-19 patients; as a result, they have been suffering from work burnout due to the burden of work as well as anxiety, depression, and fear [5, 6]. Such negative work experiences have decreased nurses’ intention to care, such as their refusal to care for patients and turnover intention [5, 7]. Thus, it is necessary to identify factors that positively influence nurses’ inten- tion to care and develop strategies that strengthen them. Intention to care refers to nurses’ willingness to care for patients voluntarily [8]. In the cur- rent crisis situation, identifying the factors influencing nurses’ intention to care is essential not only to reduce the levels of fear, anxiety, stress, and turnover intention among nurses, but also to improve the quality of nursing care [9]. Additionally, in the current COVID-19 pandemic, nurses are experiencing fear, anxiety, depression, helplessness, and exhaustion stemming from the fear of getting infected themselves or spreading the virus to their families [5, 6, 10]; all of these have a negative impact on individ- uals’ sense of optimism [11]. Reduced optimism can increase fear of COVID-19, leading to a vicious cycle [11]. As such, nurses should learn to utilize self-coping mechanisms that involve psychological and lifestyle adjustments [10]. Although nurses who care for COVID-19 patients may initially experience negative emotions, over time, their psychological defense mechanisms enable positive emotions to co-exist [5, 10]. Therefore, strategies to increase nurses’ optimism need to be developed. Positive psychological capital is a concept derived from an organizational application of positive psychology—organizational behavior research and evolving topics—and refers to the complex positive psychological state of members in an organization. It is a second-order factor with a multidimensional construct, and is composed of the sub-domains of self-efficacy, hope, optimism, and resilience [12]. Although previous studies reported a negative correlation between positive psychological capital and exhaustion and turnover intention [13], studies in context of nursing are still lacking. Furthermore, research on positive psychological capital in PLOS ONE \| https://doi.org/10.1371/journal.pone.0262786 January 19, 2022 2 / 13 PLOS ONEFactors influencing nurses’ intention to care for patients with COVID-19 the context of crises such as the COVID-19 pandemic are yet to be conducted. Thus, the posi- tive psychological capital of nurses should be explored to better understand their intention to care during the COVID-19 pandemic. The role of a nursing professional with a positive professional outlook is critical for an effec- tive response to unprecedented disaster situations such as the spread of a novel infectious dis- ease [14]. High-quality care and efficient performance is possible when nurses have a positive and clear sense of professionalism [15]. Certainly, nurses’ intention to care was low during the SARS and Ebola epidemics [7, 16]. Although nursing professionalism has been suggested as a factor that positively influences nurses’ intention to care for patients in the pandemic situation [17], verification is needed as some studies have reported no relevance in this regard [9, 18]. As crises caused by a novel infectious disease outbreak can recur in the future, improving the quality of nursing by identifying factors that influence nurses’ intention to care during these situ- ations is imperative. This study aims to examine nurses’ intention to care for COVID-19 patients in the prevailing pandemic situation, and to identify factors that influence their intention to care. The current study focuses on positive psychological capital and nursing professionalism to pro- vide preliminary data to support efforts to positively influence nurses’ intention to care. Methods Study design This study used a descriptive correlational investigation to examine the level of intention to care for patients among bedside nurses caring for COVID-19 patients. The study also aimed to identify factors that influence their intention to care, based on the concepts of positive psycho- logical capital and nursing professionalism. Participants and data collection The participants of this study were nurses working at one of the two university hospitals with nationally designated negative pressure isolation rooms, or two national hospitals designated as treatment facilities for patients with COVID-19. Participants had to satisfy the following conditions: first, they had to be nurses who directly cared for COVID-19 patients; and second, they had to provide informed consent and voluntarily agree to participate therein. The sample size for this study was calculated using the G�power 3.1.9.4 program following previous research [18] on factors influencing nurses’ intention to care. The minimum required number of participants was 147, considering a two-tailed test for multiple regression, signifi- cance level α = .05, statistical power (1-β) = .90, moderate effect size (f2) = .15, and 10 predic- tors included in the regression analysis. Ultimately, 165 nurses were included in the study to account for attrition. Data were collected between August 16 and 30, 2020, from two university hospitals with nationally designated negative pressure isolation rooms and two national hospitals designated as COVID-19 treatment centers. Prior to data collection, the nursing department at each hos- pital was effectively conveyed the purpose of the study and requested to cooperate, via tele- phone. Approval to conduct the study was also obtained prior to data collection. Although the investigator hoped to collect data through in-person visits to the three hospitals, with the exception of the investigator’s own workplace, the hospitals informed that in-person visits were not possible during the COVID-19 pandemic. Therefore, the investigator provided the description, consent form, questionnaires, and return envelopes to the nurse-in-charge of the COVID-19 patients at each hospital, for distribution among the participants. Study partici- pants then filled out the consent forms and the questionnaires and returned them to the nurse-in-charge in a sealed return envelope. The nurse-in-charge collected and boxed these PLOS ONE \| https://doi.org/10.1371/journal.pone.0262786 January 19, 2022 3 / 13 PLOS ONEFactors influencing nurses’ intention to care for patients with COVID-19 documents and mailed them to the investigator. At the hospital where the investigator is employed, the investigator delivered the aforementioned documents to the nurse-in-charge for data collection, and the latter distributed these documents to the participants. Signed con- sent forms and completed questionnaires were returned to the nurse-in-charge in the return envelope, which were then collected by the investigator. In total, 165 questionnaires were distributed, of which 156 were collected, indicating a recovery rate of 94.5%. Of these, 8 questionnaires were deemed unsuitable for statistical analy- sis due to inappropriate responses; these were excluded. Finally, 148 questionnaires were used in the final analysis, indicating a response rate of 89.7%. Questionnaires and measurements Intention to care. Intention to care was measured using a tool developed by Yoo et al. [19]; it assessed nurses’ intention to care for severe acute respiratory syndrome (SARS) patients. The instrument was then modified and revised by Lee and Kang [9] to ensure its validity for use on nurses caring for patients with novel infectious diseases. Permission to use the revised version was obtained from Lee. The tool consisted of three items measured on a seven-point Likert scale ranging from “strongly disagree” (-3 points) to “strongly agree” (3 points). The final score was calculated by obtaining the average of the scores of the three items. A higher average score indicated greater intention to care. The reliability of the tool according to Lee and Kang [9] was Cronbach’s α = .88, and in this study, Cronbach’s α = .97. Positive psychological capital. Positive psychological capital was measured using the Psy- chological Capital Questionnaire (PCQ) developed by Luthans et al. [12], translated to Korean and tested for validity by Lee and Choi [20]; it was then revised by Lee [21] to fit the nursing context. Prior to using the tool, approval was obtained from the Mind Garden, Inc. (www. mindgarden.com)—which owns the PCQ copyright—and the Korean translators and editors. The tool comprises 24 items in 4 subdomains: self-efficacy (6 items), optimism (6 items), hope (6 items), and resilience (6 items) measured on a six-point Likert scale ranging from “strongly disagree” (1 point) to “strongly agree” (6 points). The score was obtained by calculating the average of the scores on the 24 items. Higher scores indicated higher levels of positive psycho- logical capital. The reliability of this tool was reported as Cronbach’s α = .88–.89 in a study by Luthans et al. [12], while Lee and Choi [20] reported Cronbach’s α = .93. Similarly, Lee [21] reported a reliability of Cronbach’s α = .95. In this study, Cronbach’s α = .94. Nursing professionalism. Nursing professionalism was measured using Hall’s Profes- sional Inventory, an assessment tool developed by Hall [22], revised by Snizek [23], and trans- lated into Korean and validated by Baek and Kim [24]. Prior approval for use was obtained from Baek. The tool comprises 25 items in 5 subdomains: standard of professional organiza- tion (5 items), belief in public service (5 items), autonomy (5 items), belief in self-regulation (5 items), and vocational consciousness (5 items). The scores are measured on a five-point Likert scale ranging from “strongly disagree” (1 point) to “strongly agree” (5 points). The score was obtained by calculating the average of the scores on the 25 items. Higher scores indicated higher levels of nursing professionalism. When the tool was developed, its reliability was Cron- bach’s α = .86 [22], while Snizek [23] reported a reliability of Cronbach’s α = .78 and Baek and Kim [24] reported Cronbach’s α = .82. In this study, Cronbach’s α = .63. Ethical consideration Approval from the Institutional Review Board (IRB) was obtained (IRB File No. CHOSUN 2020-07-007) in addition to permission obtained from the nursing director of the institutions included in the study and the cooperation of pertinent individuals. An explanation regarding PLOS ONE \| https://doi.org/10.1371/journal.pone.0262786 January 19, 2022 4 / 13 PLOS ONEFactors influencing nurses’ intention to care for patients with COVID-19 the purpose of the study, voluntary participation, freedom to withdraw from or discontinue the study without penalty, anonymity, and the storage and destruction of data were provided to participants, and their written consent was obtained prior to completing the questionnaire. The questionnaire took approximately 15 minutes to complete, and participants received a small incentive (approximately $5) as a token of appreciation for their participation. The col- lected written consent forms were stored in a locked cabinet and the data were encrypted and computerized with a password to ensure the confidentiality of personal information. All docu- ments and data collected in this study will be stored for three years following the completion of the study, and then discarded. Data analysis Data analysis was performed using the SPSS/WIN 26.0 program according to the procedure described in this section. First, the differences between participants’ level of intention to care based on general and clinical experience characteristics were analyzed using an independent t- test or one-way ANOVA, along with the Scheffe´ test for the post-hoc analysis. Second, Pear- son’s correlation coefficient was calculated to examine the relationship among positive psycho- logical capital, nursing professionalism, and intention to care. Finally, a stepwise multiple regression analysis was performed using the stepwise selection method to examine the influ- ence of positive psychological capital and nursing professionalism on nurses’ intention to care for COVID-19 patients. Results Participants’ general characteristics and characteristics of their clinical experience The response rate for the questionnaire was 89.7%. Of the participants, 94.6% were women. Most participants were aged less than 30 years (51.4%), followed by individuals in their 30s (25.0%) and 40s (15.5%). The average age of participants was 32.67±8.60 years. Among the par- ticipants, 65% were not religious and 65.5% were not married. Most participants lived together with family (88.6%) and did not have children (69.6%). Most participants had a bachelor’s degree (63.5%). The most common duration of job experience as a nurse was more than 7 years (45.3%), followed by 1–3 years (31.8%). Only 6.1% had worked as a nurse for less than a year. The most prevalent job position was a general bedside nurse (86.5%), and the most common type of work was a three 8-hour shift pattern (91.2%). The most common subjective health sta- tus was moderate (45.9%) followed by healthy (45.3%), and unhealthy (8.8%) (Table 1). Based on the characteristics of participants’ clinical experience, 89.9% had received COVID-19-related training. Furthermore, 37.2% of the participants had previous experience of caring for patients infected with novel infectious diseases such as severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS), and 25.7% had experience caring for patients with severe respiratory symptoms. In addition, 30.4% had experience work- ing in an intensive care unit (ICU), and 20.9% had experience working in an emergency room (ER). Finally, 54.1% of the participants indicated that their clinical experience and skills were sufficient to care for COVID-19 patients (Table 2). Level of positive psychological capital, nursing professionalism, and intention to care The level of positive psychological capital of the study participants was 3.98±0.57. The level of nursing professionalism was 3.20±0.28, and level of intention to care was 1.04±1.58 (Table 3). PLOS ONE \| https://doi.org/10.1371/journal.pone.0262786 January 19, 2022 5 / 13 PLOS ONETable 1. Participants’ characteristics and differences in intention to care by their characteristics (N = 148). Variables Categories N (%) Intention to care Factors influencing nurses’ intention to care for patients with COVID-19 Gender Age (in years) Religion Marital status Living together with family members Children Education level Duration of job experience as a nurse (years) Job position Department Type of work Health status Female Male <30a 30-39b 40-49c �50d Follow Do not follow Partnered Single Yes No Have Do not have 3yr college a Bachelors b Master’s or higher c <1 1–3 4–6 �7 General-beside nurse a Supervising nurse b Head nurse c Ward a Intensive care unit b Others† c Three 8-hour shift pattern Fixed day or evening Day-time job Healthy Moderate Unhealthy Abbreviation: M = mean; SD = standard deviation. �Scheffe´ test; †Others included operating room, out-patients department, etc. https://doi.org/10.1371/journal.pone.0262786.t001 140(94.6) 8(5.4) 76(51.4) 37(25.0) 23(15.5) 12(8.1) 51(34.5) 97(65.5) 51(34.5) 97(65.5) 131(88.6) 17(11.4) 45(30.4) 103(69.6) 47(31.8) 94(63.5) 7(5.3) 9(6.1) 47(31.8) 25(16.9) 67(45.3) 135(91.2) 6(4.1) 7(4.7) 127(86.5) 14(9.5) 6(4.0) 135(91.2) 5(3.4) 8(5.4) 67(45.3) 68(45.9) 13(8.8) M±SD 1.03±1.60 1.21±1.17 1.06±1.54 0.41±1.50 1.29±1.60 2.42±1.09 1.29±1.59 0.91±1.57 1.18±1.59 0.97±1.58 1.11±1.54 0.53±1.82 1.4±1.60 0.89±1.55 0.60±1.68 1.15±1.49 2.57±0.74 1.59±1.22 1.18±1.51 0.73±1.59 0.99±1.66 0.93±1.53 1.83±2.40 2.52±0.60 1.14±1.52 -0.25±1.66 1.89±1.39 0.97±1.55 1.27±2.49 2.21±1.05 1.27±1.62 0.34±1.47 0.42±1.82 t/F 0.30 5.66 -1.40 0.75 1.43 -1.83 5.73 p .762 .001 (a,b<d)� .165 .457 .115 .069 .004 (a,b<c)� 0.84 .476 4.36 6.23 .015 (a<c)� .003 (a,c>b)� 2.44 .091 1.85 .160 Differences in intention to care by participants’ characteristics Statistically significant differences were observed in nurses’ intention to care based on the fol- lowing general characteristics of participants: age (F = 5.66, p = .001), education level (F = 5.73, p = .004), job position (F = 4.36, p = .015), and department (F = 6.23, p = .003). Spe- cifically, participants aged more than 50 years (2.42±1.09) and those with an education level higher than a master’s degree (2.57±0.74) demonstrated the highest level of intention to care. In terms of job position, head nurses (2.52±0.60) demonstrated the highest level of intention PLOS ONE \| https://doi.org/10.1371/journal.pone.0262786 January 19, 2022 6 / 13 PLOS ONETable 2. Characteristics of participants’ clinical experience and differences in intention to care based on their clinical experience (N = 148). Variables Categories N (%) Intention to care Factors influencing nurses’ intention to care for patients with COVID-19 COVID-19-related training Experience caring for patients infected with NID Experience caring for patients with SRS Experience working in an ICU Experience working in an ER Yes No Experienced Not experienced Experienced Not experienced Experienced Not experienced Experienced Not experienced Belief that their clinical experience and skills were sufficient to care for COVID-19 patients Sufficient Not sufficient 133(89.9) 15(10.1) 55(37.2) 93(62.8) 38(25.7) 110(74.3) 45(30.4) 103(69.6) 31(20.9) 117(79.1) 80(54.1) 68(45.9) M±SD 1.10±1.58 0.51±1.56 1.28±1.37 0.90±1.68 1.00±1.66 1.06±1.56 1.00±1.58 1.06±1.59 1.41±1.31 0.95±1.64 1.33±1.33 0.71±1.79 t/F 1.38 p .170 1.42 .158 -0.19 .847 -0.20 .843 1.46 .148 2.36 .020 Abbreviations: M = mean; SD = standard deviation; NID = novel infectious diseases; SRS = severe respiratory symptoms; ICU = intensive care unit; ER = emergency room. https://doi.org/10.1371/journal.pone.0262786.t002 to care, while in terms of department, those working in the ICU (-0.25±1.66) demonstrated the lowest level of intention to care (Table 1). Difference in intention to care by characteristics of participants’ clinical experience The level of intention to care was significantly higher in participants who believed their clinical experience and skills were sufficient to care for COVID-19 patients (1.33±1.33) compared to when they did not (0.71±1.79) (t = 2.36, p = .020) (Table 2). Correlation between positive psychological capital, nursing professionalism, and intention to care A statistically significant correlation was found between intention to care and both positive psy- chological capital (r = 0.30, p < .001) and nursing professionalism (r = 0.17, p = .041) (Table 4). Factors influencing intention to care To identify the factors influencing nurses’ intention to care, a stepwise multiple regression analysis using the stepwise selection method was performed on factors that demonstrated a difference in intention to care—age, education level, job position, department, and belief that Table 3. Level of positive psychological capital, nursing professionalism, and intention to care (N = 148). Variables PPC NP Intention to care Range 1–6 1–5 -3–3 M±SD 3.98±0.57 3.20±0.28 1.04±1.58 Min 1.71 2.52 -3 Max 5.54 4.12 3 Abbreviations: PPC = positive psychological capital; NP = nursing professionalism; M = mean; SD = standard deviation. https://doi.org/10.1371/journal.pone.0262786.t003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262786 January 19, 2022 7 / 13 PLOS ONEFactors influencing nurses’ intention to care for patients with COVID-19 Table 4. Correlation among positive psychological capital, nursing professionalism, and intention to care (N = 148). Variables PPC NP Intention to care PPC r(p) 1 .48(.001) .30 (< .001) NP r(p) 1 .17(.041) Intention to care r(p) 1 Abbreviations: PPC = positive psychological capital; NP = nursing professionalism. https://doi.org/10.1371/journal.pone.0262786.t004 their clinical experience and skills were sufficient to care for COVID-19 patients—along with positive psychological capital and nursing professionalism, which correlated with intention to care. Nurses’ age, education level, job position, department, and belief that their clinical experi- ence and skills were sufficient to care for COVID-19 patients were converted into dummy var- iables and included in the regression equation. To determine the presence of multicollinearity among independent variables, the basic assumption for regression analyses, the Durbin-Wat- son statistic and tolerance, and variance inflation factor (VIF) were calculated. The Durbin- Watson statistic was 2.185, close enough to 2 to confirm the absence of autocorrelation. Fur- ther, tolerance was 0.78–0.97, more than 0.1, and VIF was 1.072–2.352, less than 10, indicating that the issue of multicollinearity was not relevant to any of the variables. Moreover, the linear- ity and equal variance of the model was observed through residual analysis and the residuals were standardized to assume the normal distribution of the error terms, which indicated there was no value greater than the absolute value of 3 [25]. Factors influencing intention to care were identified as: nurses’ age (30<: β = .18, p = .026; �50: β = .23, p = .005), department (ICU: β = -.26, p = .001), belief that their clinical experience and skills were sufficient to care for COVID-19 patients (sufficient: β = .18, p = .019), and posi- tive psychological capital (β = .22, p = .044). The explanatory power of the model (R2) was 48.0% (Table 5). Discussion Based on the results of this study, participating nurses’ intention to care for COVID-19 patients was 1.04 (±1.58). The intention to care observed in the participants of this study was higher than that observed in previous studies that used the same tools to report nurses’ Table 5. Factors influencing intention to care (N = 148). Variables (Constant) Age Department <30 �50 ICU Belief that their clinical experience and skills were sufficient to care for COVID-19 patients Sufficient PPC B -1.94 0.57 1.32 -1.39 0.56 0.61 SE 0.96 0.25 0.47 0.40 0.24 0.23 β .18 .23 -.26 .18 .22 t -2.03 2.24 2.83 -3.46 2.37 2.61 p .044 .026 .005 .001 .019 .044 R2 = .480, Adj. R2 = .231, F = 8.51. p < .001, Durbin-Watson = 2.185. Abbreviations: SE = Standard error; ICU = Intensive care unit; PPC = positive psychological capital. Note: Age (reference: 30–39 years), educational level (reference: 3yrs college), job position (reference: general-duty nurse), department (reference: ward), subjective judgment that clinical experience and skills are sufficient to care for COVID-19 patients (reference: not sufficient), nursing professionalism, and PPC were included. https://doi.org/10.1371/journal.pone.0262786.t005 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262786 January 19, 2022 8 / 13 PLOS ONEFactors influencing nurses’ intention to care for patients with COVID-19 intention to care for patients with novel infectious diseases, regardless of prior experience [9, 18]. These findings reveal that nurses’ intention to care for patients infected with a novel infec- tious disease in a situation where a new infectious disease is prevalent, is higher than in normal situations that do not involve an outbreak. This may be attributed to the fact that during the COVID-19 pandemic, the severity of the situation was recognized at the national and global levels and shared widely through various platforms such as the media and SNS. In addition, support acknowledging the efforts of clinicians to overcome the situation was also extended. Furthermore, given that the data collection period of this study was prior to November 2020 when the number of cases surged in Korea—and the region in which data was collected had a lower of number cases than other regions, meant that nurses’ workload and stress levels were not as high; these may have influenced the study findings. A follow-up study to evaluate the appropriate number of patients with a novel infectious disease that should be assigned to each nurse considering their workloads, may be worthwhile. Positive psychological capital is a concept derived from positive psychology and refers to individuals’ positive psychological state [26]. This positive psychological state is an emerging paradigm for human resource development that is characteristically manageable with the potential for development and improvement [12, 26]. According to the findings of this study, the level of positive psychological capital of the nurses caring for COVID-19 patients was higher than that of nurses with a relatively shorter clinical experience ranging from 13 to 36 months [27]. Similarly, the positive psychological cap- ital of participants was higher than or similar to that of nurses in general hospitals and small- to medium-sized hospitals [28, 29]. These findings indicate that the level of positive psychological capital of nurses experiencing fear, anxiety, depression, and exhaustion in the COVID-19 pan- demic are similar to normal or higher than normal levels. Positive psychological capital is a pos- itive psychological state in which an individual pursues personal development, and that enables optimism and increased coping ability, while experiencing decreasing job-related burdens in challenging situations [30, 31]. Furthermore, positive psychological capital enhances life satis- faction through positive changes in individuals’ attitudes and behaviors toward their jobs and improves organization performance by inducing organizational change [30]. Amid the medical crisis of the COVID-19 pandemic, the utilization of nurses’ positive psychological capital should be actively sought as a personal and organizational coping strategy. This study identified positive psychological capital as a factor that positively affects the intention to care. A direct comparison of the findings was not possible because of the lack of previous research on the influence of positive psychological capital on nurses’ intention to care in a crisis such as COVID-19. However, a systematic review examining the factors related to positive psychological capital among clinical nurses by Lee et al. [13] reported a negative corre- lation between positive psychological capital and turnover intention, a finding similar to that in this study. Positive psychological capital not only has a positive influence on the relationship between nurses and patients in the clinical setting [32], but also helps to maintain nurses’ phys- ical, mental, and social well-being [33], while improving the quality of nursing care and orga- nizational performance [30, 32]. The use of positive psychological capital to help nurses overcome fear, anxiety, job stress, depression, social isolation, and exhaustion in the COVID- 19 pandemic should be explored. Previous research has framed positive psychological capital as a capacity that can be developed and improved through retrospective training and learning [34]. A follow-up study should be conducted to develop and test a program designed to increase nurses’ positive psychological capital, and an implementation strategy for its use dur- ing a medical crisis should be explored. According to the findings of this study, subjective judgment that clinical experience and skills are sufficient to care for COVID-19 patients was identified as a factor influencing nurses’ PLOS ONE \| https://doi.org/10.1371/journal.pone.0262786 January 19, 2022 9 / 13 PLOS ONEFactors influencing nurses’ intention to care for patients with COVID-19 intention to care. A direct comparison of results was not feasible as no study has yet explored the influence of this subjective judgment on nurses’ intention to care in a crisis such as the COVID-19 pandemic. Results for this study can be interpreted in the same ways as the find- ings from Ko et al. [35], which suggested the need for continuing education to induce inten- tion to care in SRAS patients, have previous emergency and disaster experience [36], findings from previous studies that report nurses’ increased intention to respond if they have previous experience in caring for patients with infectious diseases [37], and previous studies that report relationship of confidence in personal skills with willingness to work [38]. This indicates that providing an opportunity for nurses to gain sufficient knowledge and experience regarding novel infectious diseases can promote their participation in a pandemic or related crisis. Thus, it will be essential to provide information and training on novel infectious diseases in prepara- tion for future pandemics and for such efforts to be made at the government level as well as in the medical and nursing fields. The level of nursing professionalism among nurses caring for COVID-19 patients in this study was lower than that reported in previous research on nurses caring for patients with a novel infectious disease in a non-pandemic situation [9, 18], as well as nurses working in gen- eral hospitals under normal circumstances [39]. Such results indicate that the level of nursing professionalism decreases in medical crises such as the COVID-19 pandemic. From a funda- mental perspective of nursing as a profession, caring for patients is considered a nurse’s duty [9]. However, during an infectious disease outbreak, clinicians may face conflict between their safety as individuals and their duty to care as clinicians [40]. Nursing professionalism may be utilized as a strategy to overcome this ethical conflict. In this study, nursing professionalism was not identified as a factor influencing nurses’ intention to care. This supports the findings of previous studies examining the factors influ- encing nurses’ intention to care for patients with a novel infectious disease and Ebola patients [9, 16, 18]. Nevertheless, findings from another study that presented factors that influence nurses’ intention to care for MERS patients suggested otherwise [17]. Further research is nec- essary to explore the correlation between nurses’ intention to care and nursing professionalism in a medical crisis. This study is valuable in that it reviewed the factors influencing intention to care among bedside nurses providing direct care for COVID-19 patients during a pandemic. However, the study has a few limitations that should be considered when interpreting its findings. First, this study did not assess factors, such as fear, anxiety, job stress, depression, and burn out, in nurses who provided care for COVID-19 patients. Therefore, we could not interpret the influence these factors on the nurses’ intention to care for COVID-19 patients. Future studies that include and explore these factors should be conducted. Second, the participants of the study were nurses caring for COVID-19 patients randomly sampled from hospitals in regions with a relatively lower number of COVID-19 cases, which may present a challenge in generalizing the findings. Third, this study examined nurses’ intention to care for COVID-19 patients in a period during which COVID-19 response had gained the attention of the Korean nation and media. Thus, the potential for bias from social desirability in participants’ responses cannot be excluded and should be considered while interpreting the findings. Fourth, the data for the study were collected prior to November 2020 and thus did not reflect the situation after November 2020 when the number of cases surged. This should also be considered when inter- preting the findings of this study. A future study examining nurses’ intention to care organized by time periods according to the number of cases may be beneficial. Finally, the reliability of the nursing professionalism questionnaire used in this study was low: Cronbach’s α = .63. This should also be considered when interpreting the findings. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262786 January 19, 2022 10 / 13 PLOS ONEFactors influencing nurses’ intention to care for patients with COVID-19 Conclusion The results identified positive psychological capital as a factor influencing nurses’ intention to care for COVID-19 patients. Thus, the development of various programs to improve nurses’ positive psychological capital and follow-up studies analyzing their effects are needed. Further- more, the subjective judgment that clinical experience and skills are sufficient to care for COVID-19 patients was also identified as a factor influencing intention to care. Specifically, it will be necessary to provide information and training regarding novel infectious diseases; this will enable nurses to prepare for future novel infectious disease pandemics through efforts at the government level and in the medical and nursing fields. Nursing professionalism was not identified as a factor influencing nurses’ intention to care for COVID-19 patients. We hope that future studies will be conducted to assess the association between the nurses’ intention to care and their professionalism. Supporting information S1 Appendix. Questionnaire (Korean version). (PDF) S2 Appendix. Questionnaire data. (XLSX) Author Contributions Conceptualization: Sun-a Jeong, Jinhee Kim. Data curation: Sun-a Jeong. Formal analysis: Sun-a Jeong. Funding acquisition: Jinhee Kim. Investigation: Sun-a Jeong, Jinhee Kim. Methodology: Sun-a Jeong, Jinhee Kim. Project administration: Jinhee Kim. Resources: Jinhee Kim. Software: Sun-a Jeong. Supervision: Jinhee Kim. Validation: Jinhee Kim. Visualization: Jinhee Kim. Writing – original draft: Sun-a Jeong, Jinhee Kim. Writing – review & editing: Sun-a Jeong, Jinhee Kim. References 1. Kim JY, Choe PG, Oh Y, Oh KJ, Kim J, Park SJ, et al. The first case of 2019 novel coronavirus pneumo- nia imported into Korea from Wuhan, China: implication for infection prevention and control measures. Journal of Korean medical science. 2020; 35(5). https://doi.org/10.3346/jkms.2020.35.e61 PMID: 32030925 2. Cucinotta D, Vanelli M. WHO declares COVID-19 a pandemic. Acta Bio Medica: Atenei Parmensis. 2020; 91(1):157. https://doi.org/10.23750/abm.v91i1.9397 PMID: 32191675 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262786 January 19, 2022 11 / 13 PLOS ONEFactors influencing nurses’ intention to care for patients with COVID-19 3. COVID-19 Dashboard by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University (JHU) USA: Johns Hopkins University (JHU); 2021 [cited 2021 June 16]. https://www.arcgis. com/apps/opsdashboard/index.html#/bda7594740fd40299423467b48e9ecf6 4. Coronavirus Disease 2019 (COVID-19) Seoul: Korea Centers for Disease; 2021 [cited 2021 June 16]. http://ncov.mohw.go.kr/bdBoardList_Real.do?brdId=1&brdGubun=11&ncvContSeq=&contSeq= &board_id=&gubun= 5. Lee N, Lee H-J. South Korean Nurses’ Experiences with Patient Care at a COVID-19-Designated Hos- pital: Growth after the Frontline Battle against an Infectious Disease Pandemic. International Journal of Environmental Research and Public Health. 2020; 17(23):9015. https://doi.org/https%3A//doi.org/10. 3390/ijerph17239015 PMID: 33287343 6. Hu D, Kong Y, Li W, Han Q, Zhang X, Zhu LX, et al. Frontline nurses’ burnout, anxiety, depression, and fear statuses and their associated factors during the COVID-19 outbreak in Wuhan, China: A large- scale cross-sectional study. EClinicalMedicine. 2020; 24:100424. https://doi.org/10.1016/j.eclinm.2020. 100424 PMID: 32766539 7. Shiao JS-C, Koh D, Lo L-H, Lim M-K, Guo YL. Factors predicting nurses’ consideration of leaving their job during the SARS outbreak. Nursing Ethics. 2007; 14(1):5–17. https://doi.org/10.1177/ 0969733007071350 PMID: 17334166 8. 9. Lee JK. A study for investigating predictors of AIDS and patients care intention among nursing students. Journal of Korean Academy of Nursing. 2001; 31(2):292–303. Lee J, Kang SJ. Factors influencing nurses’ intention to care for patients with emerging infectious dis- eases: Application of the theory of planned behavior. Nursing & Health Sciences. 2020; 22(1):82–90. https://doi.org/10.1111/nhs.12652 PMID: 31670456 10. Sun N, Wei L, Shi S, Jiao D, Song R, Ma L, et al. A qualitative study on the psychological experience of caregivers of COVID-19 patients. American journal of infection control. 2020; 48(6):592–8. https://doi. org/10.1016/j.ajic.2020.03.018 PMID: 32334904 11. Bakioğlu F, Korkmaz O, Ercan H. Fear of COVID-19 and positivity: Mediating role of intolerance of uncertainty, depression, anxiety, and stress. International Journal of Mental Health and Addiction. 2020:1–14. https://doi.org/https%3A//doi.org/10.1007/s11469-020-00331-y PMID: 32837421 12. 13. Luthans F, Avolio BJ, Avey JB, Norman SM. Positive psychological capital: Measurement and relation- ship with performance and satisfaction. Personnel psychology. 2007; 60(3):541–72. https://doi.org/ https%3A//doi.org/10.1111/j.1744-6570.2007.00083.x Lee BY, Jung HM. Factors related to positive psychological capital among Korean clinical nurses: a sys- tematic review and meta-analysis. Journal of Korean Clinical Nursing Research. 2019; 25(3):221–36. https://doi.org/https%3A//doi.org/10.22650/JKCNR.2019.25.3.221 14. Oh N. Stress of nurses in local medical centers who experienced the MERS situation and nursing inten- tion for patients with new infectious diseases. Daegu: Kyungpook National University; 2016. 15. Hallin K, Danielson E. Registered nurses’ perceptions of their work and professional development. Jour- nal of Advanced Nursing. 2008; 61(1):62–70. https://doi.org/10.1111/j.1365-2648.2007.04466.x PMID: 18034817 16. Kim JS, Choi JS. Factors predicting clinical nurses’ willingness to care for Ebola virus disease-infected patients: A cross-sectional, descriptive survey. Nursing & health sciences. 2016; 18(3):299–305. https://doi.org/10.1111/nhs.12269 PMID: 26661457 17. Oh N, Hong N, Ryu DH, Bae SG, Kam S, Kim K-Y. Exploring nursing intention, stress, and professional- ism in response to infectious disease emergencies: the experience of local public hospital nurses during the 2015 MERS outbreak in South Korea. Asian nursing research. 2017; 11(3):230–6. https://doi.org/ 10.1016/j.anr.2017.08.005 PMID: 28991605 18. Hea-Jin M, Young PJ. Factors Influencing Intentions to Care For Emerging Infectious Disease Patients among National and Public Hospitals Nurses. Journal of Korean Academy of Fundamentals of Nursing. 2021; 28(1). 19. Yoo HR, Kwon BE, Jang YS, Youn HK. Validity and reliability of an instrument for predictive nursing intention for SARS patient care. Journal of Korean Academy of Nursing. 2005; 35(6):1063–71. https:// doi.org/10.4040/jkan.2005.35.6.1063 PMID: 16288149 20. 21. Lee DS, Choi YD. A study on antecedents and consequences of positive psychological capital in organi- zations. Korean Business Review. 2010; 39(1):1–28. Lee BS. The relationship of shift work nurse0s positive psychological capital and the quality of life. Seoul: Yonsei University; 2016. 22. Hall RH. Professionalization and bureaucratization. American sociological review. 1968:92–104. 23. Snizek WE. Hall’s professionalism scale: An empirical reassessment. American sociological review. 1972:109–14. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262786 January 19, 2022 12 / 13 PLOS ONEFactors influencing nurses’ intention to care for patients with COVID-19 24. Baek H-C, Kim Y-S. Translation and validation of Korean version of Hall’s professionalism inventory. Journal of Korean Academy of Nursing Administration. 2007; 13(4):509–15. 25. 26. Fox J. Applied regression analysis, linear models, and related methods: Sage Publications, Inc; 1997. Luthans F, Youssef CM. Emerging positive organizational behavior. Journal of management. 2007; 33(3):321–49. https://doi.org/https%3A//doi.org/10.1177/0149206307300814 27. Kwon HS, Ha Y. Effects of Positive Psychological Capital, Social Support and Head Nurses’ Authentic Leadership on Organizational Commitment of Nurses at the Advanced Beginner Stage. Journal of Korean Academy of Nursing Administration. 2020; 26(3):284–93. https://doi.org/http%3A//doi.org/10. 11111/jkana.2020.26.3.284 28. Yang H-M. Effects of the positive psychological capital, professional self-concept and self-leadership on nursing work outcome: Focused on the nurses in small and medium hospitals. The Journal of the Korea Contents Association. 2018; 18(11):87–97. https://doi.org/https%3A//doi.org/10.5392/JKCA.2018.18. 11.087 29. Kwon SN, Park HJ. Effects of Nurses’ Positive Psychological Capital, Self-Leadership, and Relational Bonds on Organizational Commitment. Journal of Korean Academy of Nursing Administration. 2020; 26(3):241–50. https://doi.org/http%3A//doi.org/10.11111/jkana.2020.26.3.241 30. 31. Lee SN, Kim JA. Concept analysis of positive psychological capital. Journal of Korean Academy of Nursing Administration. 2017; 23(2):181–90. https://doi.org/https%3A//doi.org/10.11111/jkana.2017. 23.2.181 Lee S-h, Park J-y, Lee K-m, Shin E-h, Park J-m, Nam H-, et al. The convergence effects of commitment of organization and positive psychological capital in nurse on burnout. Journal of Digital Convergence. 2016; 14(6):335–45. https://doi.org/https%3A//doi.org/10.14400/JDC.2016.14.6.335 32. Kim IS, Seo RB, Kim BN, Min AR. The effects of positive psychological capital, organizational commit- ment, customer orientation in clinical nurses. Journal of Korean Academy of Nursing Administration. 2015; 21(1):10–9. https://doi.org/https%3A//doi.org/10.11111/jkana.2015.21.1.10 33. 34. Laschinger HKS, Fida R. New nurses burnout and workplace wellbeing: The influence of authentic lead- ership and psychological capital. Burnout research. 2014; 1(1):19–28. https://doi.org/https%3A//doi. org/10.1016/j.burn.2014.03.002 Luthans F, Avey JB, Patera JL. Experimental analysis of a web-based training intervention to develop positive psychological capital. Academy of Management Learning & Education. 2008; 7(2):209–21. https://doi.org/https%3A//doi.org/10.5465/amle.2008.32712618 35. Ko N-Y, Feng M-C, Chiu D-Y, Wu M-H, Feng J-Y, Pan S-M. Applying theory of planned behavior to pre- dict nurses’ intention and volunteering to care for SARS patients in southern Taiwan. The Kaohsiung journal of medical sciences. 2004; 20(8):389–98. https://doi.org/10.1016/S1607-551X(09)70175-5 PMID: 15473650 36. Grimes DE, Mendias EP. Nurses’ intentions to respond to bioterrorism and other infectious disease emergencies. Nursing outlook. 2010; 58(1):10–6. https://doi.org/10.1016/j.outlook.2009.07.002 PMID: 20113750 37. Minuye B, Alebachew W, Melese Kebede SA, Belay DM. Intention to Care for COVID-19 Patients Among Nurses Working at Health Care Institutions of Debre Tabor Town, North Central Ethiopia. Risk Management and Healthcare Policy. 2021; 14:2475. https://doi.org/10.2147/RMHP.S311830 PMID: 34163266 38. Aoyagi Y, Beck CR, Dingwall R, Nguyen-Van-Tam JS. Healthcare workers’ willingness to work during an influenza pandemic: A systematic review and meta-analysis. Influenza and other respiratory viruses. 2015; 9(3):120–30. https://doi.org/10.1111/irv.12310 PMID: 25807865 39. Lee MA, Kang S, Hyun HS. Relationship among nursing professionalism, nursing work environment, and patient safety nursing activities in general hospital nurses. Journal of Korean Academy of Nursing Administration. 2019; 25(4):317–28. https://doi.org/http%3A//doi.org/10.11111/jkana.2019.25.4.317 40. Barr H, Macfarlane J, Macgregor O, Foxwell R, Buswell V, Lim W. Ethical planning for an influenza pan- demic. Clinical medicine. 2008; 8(1):49. https://doi.org/10.7861/clinmedicine.8-1-49 PMID: 18335669 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262786 January 19, 2022 13 / 13 PLOS ONE
10.1371_journal.pone.0284692	RESEARCH ARTICLE Chairman’s Communist Party of China member status and targeted poverty alleviation: Evidence from China Jian Xie1, Ruirui Gu2, Tianyi LeiID 3, Sen Yang4, Ruian Yu5 1 Department of International Accounting, School of International Education, Guangxi University of Finance and Economics, Nanning, China, 2 School of Management, Guangxi Minzu University, Nanning, China, 3 Department of Financial Engineering, School of Finance, Zhongnan University of Economics and Law, Wuhan, China, 4 Department of Business, Huanggang Normal University, Huanggang, China, 5 Department of Finance, School of Economics and Management, Hubei University of Technology, Wuhan, China tyray1992@163.com Abstract Based on the data of Chinese listed private companies from 2016 to 2020, this paper investi- gates the influence of the Chairman’s member status of Communist Party of China (CPC) on targeted poverty alleviation. The research results demonstrate that the Chairman’s CPC member status of private companies significantly increases the companies’ willingness and the amounts of investment in poverty alleviation. The construction of the CPC organization can strengthen the role of the chairman’s Communist Party of China member status in pro- moting targeted poverty alleviation. The conclusions are still valid through robustness tests, such as substituting dependent variables, adjusting the sample range, and PSM-paired samples. In addition, the Impact Threshold for a Confounding Variable is used to deal with endogenous problems. 1. Introduction Based on the upper echelon theory (Hambrick & Mason, 1984) [1], existing studies have found when the demographic background variable is taken as a proxy variable to describe the cogni- tive structure and values of managers, the certain personal characteristics of managers will affect the performance of corporate social responsibility. For example, demographic back- ground characteristics like gender, age, education [2], and acquired experience like overseas experience [3], poverty experience [4], and so on. The culture which is a deep-seated factor for managers to make differentiated and diversi- fied social responsibility decisions affects managers’ cognition, interaction and strategy choices [5].Cultural innovation is emerging in the field of finance [6], in which the impact of manag- ers’ cultural background characteristics on corporate social responsibility has also emerged as a hot research topic. China not only has traditional cultures such as Confucianism and Taoism, but also has an ideal of Communism that is gradually developing and radiating strong vitality. It enriches the research of ‘culture and corporate behavior’. Above mentioned two cultures a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Xie J, Gu R, Lei T, Yang S, Yu R (2023) Chairman’s Communist Party of China member status and targeted poverty alleviation: Evidence from China. PLoS ONE 18(6): e0284692. https:// doi.org/10.1371/journal.pone.0284692 Editor: Vu Quang Trinh, Newcastle University Business School, UNITED KINGDOM Received: August 19, 2022 Accepted: April 5, 2023 Published: June 15, 2023 Copyright: © 2023 Xie et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The third party financial data files are available from the China Economic and Financial Research Database (CSMAR) (https://www.gtarsc.com) and Wind database (https://www.wind.com.cn). The authors confirm that others would be able to access or request these data in the same manner as the authors. The authors also confirm that they did not have any special access or request privileges that others would not have. Funding: The funding named Hubei University of Technology Doctoral Research Start-up Fund PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 1 / 20 PLOS ONEProject (No.XJ2022008301), General Project of the National Social Science Foundation of China (No. 19BGL085), Research Project of Guangxi Philosophy and Social Science Planning in 2022 (No.22FGL037) support our this research for collecting data and it will support the fee for publication. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Chairman’s CPC member status and targeted poverty alleviation both make companies assume more social responsibilities. Communist Party of China (CPC) membership of the managers is an important carrier for participating in political learning and inheriting the red culture of communism. Studies have shown that the party membership of managers will increase corporate charitable donations [7] and promote corporate social responsibility [8]. Poverty alleviation is a special social responsibility given to private companies in this era. It is similar to other social responsibilities such as environmental protection, charitable dona- tions, stakeholder responsibility, and is also endowed with its special attributes. It aims to elim- inate poverty and achieve common prosperity. Besides, it is also a necessary process for the realization of Communism and an inevitable choice for the Socialist system. The data shows among the A-share private listed companies from 2016 to 2020, up to 31.1% of the chairmen were members of the CPC. Besides, both the willingness to participate in targeted poverty alle- viation and the amount of poverty alleviation investment in private companies with the status of the CPC member were significantly higher than the private companies to which the chair- man of the non-CPC member belongs. The research data demonstrates that there should be some important internal connection between the CPC member status of the chairman of a pri- vate company and targeted poverty alleviation. With the support of Chinese government, the private companies have become an important force for poverty alleviation. For one thing, when the private entrepreneurs are absorbed to become the CPC members, they need to learn systematic political knowledge such as Commu- nist values and pass strict political review. For another, the establishment of grassroots organi- zations of CPC in the companies has easy access to the ideology of Communist values. Besides, the CPC is the leadership core for the cause of socialism with Chinese characteristics. The setup of grassroots organizations of CPC contributes to the closer relationship between the government and companies. Therefore, we reasonably expect that the chairman of the CPC member has more identify with Communist values and also prefers to cooperate with the government. In this paper, we select Chinese listed private companies from 2016 to 2020 as samples to investigate the influence of chairman’s CPC membership of private companies on targeted poverty alleviation. The results find out that compared with private companies without CPC chairman status, the private companies with the chairman status of the CPC membership are significantly more willing to participate in targeted poverty alleviation and invest much more in poverty alleviation. Besides, the role of the chairman’s CPC member status in promoting targeted poverty alleviation depends on the construction of the Party organization. In other words, the chairman’s CPC member status has a more significant impact on targeted poverty alleviation in those private companies with grassroots Party organizations. The main contributions of this paper are as follows. First, it has enriched the research on the influencing factors of poverty alleviation in Chinese listed companies. Existing research mainly studies the influencing factors of targeted poverty alleviation from other aspects of companies, while this paper finds that private company executive politics identity is also an important factor in targeted poverty alleviation, and it is a useful supplement to this field. Sec- ond, it expands the research on the upper echelon theory in China. Based on the perspective of Communist culture, this paper uses the managerial CPC membership to characterize its cul- tural background characteristics, investigates its impact on targeted poverty alleviation, and expands the relevant research on the contextualization of the upper echelon theory in China. Third, this paper investigates the impact of senior management political membership on the targeted poverty alleviation of private companies, and has important enlightenment for how to effectively connect poverty alleviation and rural revitalization, and how to guide private com- panies to participate in poverty alleviation during the rural revitalization stage. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 2 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation 2. Institutional background and hypothesis development 2.1. Institutional background 2.1.1. Private listed companies’ participation in targeted poverty alleviation. The his- tory of human civilization is a history of the struggle against poverty. Since its inception, the People’s Republic of China has been committed to poverty governance and has made many remarkable achievements. China’s new generation of leaders has placed particular emphasis on this work. On November 3, 2013, President Xi Jinping first proposed ‘targeted poverty alle- viation’ in a visit to the Eighteenth Cave Village in Huayuan Town, Hunan Province. He stated that poverty alleviation should be practical and tailored to local conditions. In November 2015, President Xi Jinping made an important speech and pledged to completely eliminate poverty by 2020, which was called ‘Commitment 2020’. In order to complete this achievement, with the guidance of the CPC, the community has responded extensively, gradually forming a major pattern of poverty alleviation with the participation of many parties. Private companies are essential parts of the Socialist market economy with Chinese charac- teristics and play as important living forces in combating poverty. On 17 October 2015, the All-China Federation of Industry and Commerce, the State Council Poverty Alleviation Office and the China Society for Promotion of the Guangcai Program launched the ‘Ten Thousand Enterprises Helping Ten Thousand Villages’ program. The program took private companies as the major sponsors, with the financially disadvantaged villages and households established as the objects of financial aid. It focused on bridging village and companies, and mobilizing more than 10,000 private companies nationwide to help more than 10,000 poor villages to accelerate the process of poverty eradication. It aimed to promote all-round healthy growth of non-pub- lic economy and build a moderately prosperous society on all fronts within 3 to 5 years. Furthermore, aiming to guide and encourage capitals to flow in targeted poverty alleviation, in September 2016, China Securities Regulatory Commission encouraged listed companies to fulfill their social responsibility of targeted poverty alleviation. This served the national poverty alleviation for the interest of the national overall strategy. Subsequently, Chinese Stock Exchange unequivocally requires listed companies to reveal information on targeted poverty alleviation, including four aspects of targeted poverty alleviation planning, an annual summary of targeted poverty alleviation, the effectiveness of targeted poverty alleviation and follow-up precise poverty alleviation plans in the social responsibility in their annual reports. 2.1.2. Party-building system in private companies. With the reform and opening up pol- icy, the private economy has achieved historical development from scratch, from small to large, from weak to strong, and grown into an important part of the socialist market economy. Moreover, the private companies have undergone an evolutionary process from a ‘dissident force’ in socialism to an ‘important basis’ for China’s economic development, and private entrepreneurs are ‘their own people’. Private companies have also undergone an evolutionary process of deepening, improving and maturing from a socialist ‘dissident force’ to an ‘impor- tant foundation’ for China’s development. On one hand the strengthening of the Party-build- ing work of private companies could guide and regulate the development of the private economy. On the other hand, it is also an inherent requirement for the overall strengthening of the Party leadership. The work of the Party-building in private companies mainly includes two aspects of politi- cal absorption and the Party organization building. The Party organization of private compa- nies is a unique product of the socialist system, and by embedding grassroots Party organizations in private companies. It achieves the penetration of the CPC’s influence within private companies, thus ensuring its leadership of the private economy. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 3 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation 2.2. The upper echelon theory and corporate social responsibility The upper echelon theory was proposed by Hambrick and Mason [9] in 1984, which regarded managers as heterogeneous economic people with bounded rationality. It uses demographic characteristics such as gender, age, and educational background to portray the differences in the cognitive structure and values of managers, and believes that these differences will affect the behavior of companies. The proposal of this theory pushes the research on the influence of managers’ characteristics on corporate financial behavior to a hot topic. Existing research mainly explores the impact of the following three types of managers’ characteristics on social responsibility. First one is demographic background, which mainly focuses on examining the impact of personal characteristics on social responsibility [10] such as age, gender, education, tenure, social responsibility and so on [10]. Second is psychological characteristics, such as managers’ overconfidence [11], managers’ narcissism [12] and so on. Third is the acquired experience, such as overseas experience [3], and poverty experience [4]. All these individual characteristics are used to portray the differentiation and diversification of the cognitive structure and value concept of the heterogeneous and bounded rational man- agers, and culture is a deep-seated factor in the formation of managers’ cognitive structure and values. China provides an excellent research context for the research on this topic. Confucian culture is the mainstream culture of the Chinese nation with five-thousand-year civilization, and its thinking has always been the most basic mainstream value of the Chinese nation. The values advocating ‘propriety, righteousness, integrity, shame, benevolence, love, loyalty, and filial piety’ are still the norms for most Chinese people to act. The Confucian culture has also profoundly affected the behavior of Chinese companies. Studies have shown that Confucian culture reduces the level of risk-taking [13], improves the quality of internal control [14] and the level of social responsibility information disclosure [15]. In addition, the Communist cul- ture, which has grown rapidly and has a strong vitality in the past century, has also attracted the attention of the academic circles. Existing research shows that private companies will assume more social responsibilities such as investing more resources in pollution control and environmental protection [16,17], or more charitable donations [18] by building grassroots Party organizations or encouraging senior executives to join the CPC accepting the influence of Communist culture. Targeted poverty alleviation is a special social responsibility given to Chinese companies in this era. Only a few scholars have paid attention to the impact of individual characteristics of managers on targeted poverty alleviation. The particularity of targeted poverty alleviation is that for state-owned companies, it is a political task, unlike private company, and it is more of a voluntary act. Whether the identity of a chairman’s CPC membership of a private company will enable him to recognize the values of Communism and take the initiative to assume the responsibility of targeted poverty alleviation is a question that needs to be tested empirically. Therefore, based on the perspective of Communist culture, this paper studies the influence of private executives’ identity of the CPC membership on targeted poverty alleviation, to provide a useful supplement to the research in this field. 2.2.1. The influence of the chairman’s identity of the CPC membership of a private company on targeted poverty alleviation. The political beliefs of business managers or founders will have an important impact on business decision-making [19]. The empirical evi- dence from the U.S. market shows that representatives of different parties have different atti- tudes towards risk, there are significant differences in business strategies [20], tax avoidance behaviors [21,22], mergers and acquisitions [23], and other aspects of companies managed or created by supporters of the Democratic and Republican parties. Different from the capitalist system of Western countries, China is a socialist country under the leadership of the CPC. The PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 4 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation dominance of the public economy is the main feature of the socialist economy with Chinese characteristics, and the private economy is an important part. State-owned companies repre- senting the public economy not only have a sound governance mechanism for the Party orga- nizations, but most of their executives are held by the CPC members with strong political qualities; while for private companies representing the private economy, the construction of grassroots Party organizations is voluntary. There is also no mandatory requirement for Party membership for senior management positions. Given this, the research of political culture and corporate behavior based on the Chinese background also discusses state-owned companies and private companies separately. For state-owned companies, existing research mainly mea- sures the degree of overlap (two-way entry, cross-service) between primary Party organizations and corporate governance organizations (board of directors, board of supervisors, and man- agement) to measure Party organizations’ participation in corporate governance. The research found that it will have an important impact on executive compensation contracts [24] and executive corruption [25]. For private companies, the existing research mainly observes the political participation of private companies indirectly from political organization building or executives’ identity of the CPC membership, and studies have found that it has an important impact on charitable donations [8], financial violations [26], social responsibility [8] and other aspects. Communist culture is the macro-political and cultural environment of the socialist econ- omy with Chinese characteristics. This also determines that no matter how China’s economy is reformed and developed, the principle that the public economy dominates will never change. Besides, the CPC members are the carrier of this cultural inheritance, and the identity of the CPC members of the executives will subtly influence their cognitions, interactions and deci- sion-making [4], and have an important impact on their business decision-making. This paper reveals that private companies managed by the CPC members will more actively assume the responsibility of targeted poverty alleviation and invest more funds for it. The reasons are as follows. Firstly, the identity of the CPC membership makes the chairman agree with the values of Communism more. From an individual’s application for CPC member, a strict and lengthy process of review and training is required. All party members and comrades are people with lofty ideals who support the leadership of the CPC and have lofty Communist beliefs. In addi- tion, the CPC members should participate in the Party’s organizational life and receive educa- tion from the organization. This is an important system guarantee for Party members to maintain their advanced nature. Existing research shows that in China, entrepreneurs with a background as the CPC member possess Communist values [27,28], and Communist culture itself has the spirit of altruism and dedication to ‘serve the people wholeheartedly’, and this value concept will be internalized In the thinking mode of Party members and entrepreneurs, which positively affects their business decisions. Existing studies have also found that it is the entrepreneurs who are members of CPC to undertake more social responsibilities. Targeted poverty alleviation is a national strategy put forward in response to China’s poverty alleviation process at the critical stage, and it directly points to eliminating polarization and achieving common prosperity, which is the essential requirement of Socialism. Therefore, the identity of a chairman’s CPC membership of private company will enable them to endorse this strategy more willingly, and actively participate in targeted poverty alleviation, give play to its own advantages, and assume this special social responsibility. Secondly, the chairman who is a CPC member will pay more attention to the reputation of himself and the company. Communist culture also inherently requires CPC members to have an awareness of excellent ‘vanguard model’ and higher moral standards [7], play a leading role in all aspects, and reflect the advanced nature of Party members. Specific to the targeted PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 5 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation poverty alleviation that this article focuses on, to ensure a full victory in the targeted poverty alleviation in 2020, under the active call of the government, a large-scale poverty alleviation pattern with the active participation of all sectors of society has gradually formed. Chinese gov- ernment put forward a series of policies to disclose the information of targeted poverty allevia- tion in the social responsibility section of their annual reports. It helps guarantee listed companies’ indispensable role in fulfilling their duties for targeted poverty alleviation. Under such background, the targeted poverty alleviation has become the focus of attention from all walks of life. The chairman of the CPC members of private companies who have always been a ‘pioneer model’ also bears more social attention and higher expectations. Therefore, regardless of individual or corporate reputation, the chairman of the CPC membership of private compa- nies should be more actively involved in targeted poverty alleviation in order to maintain a consistently good social image. Based on the above analysis, this paper proposes the following research hypotheses: Hypothesis 1: Compared with other private companies, private companies in which chairman is a CPC member will be more actively involved in targeted poverty alleviation and invest more in poverty alleviation funds. 2.2.2. Chairman’s CPC membership, Party organization building, and targeted poverty alleviation by private companies. The Party organization is an important channel through which the chairman’s CPC membership of a private company plays a role in promoting tar- geted poverty alleviation. Firstly, the construction of Party organizations in private companies is a concrete manifestation of the mutual embedding of political organizations and economic organizations [29], by setting up grassroots party organizations to strengthen the construction of corporate political culture and further reinforce the role of the chairman’s CPC membership of private companies in promoting targeted poverty alleviation. Secondly, the establishment of political organizations facilitates the establishment of ties between private companies and the government, and provides opportunities for private companies to engage in the discussion of politics. In the meantime, the Party organization of private company is also a window for pri- vate companies to connect with the government’s national policies and guidelines. It can better convey the policies and guidelines of the Party and the government, thus private companies can accept and actively respond to the relevant government departments’ initiatives on tar- geted poverty alleviation, which forms a synergy with the chairman’s CPC membership to bet- ter fulfill the responsibility of targeted poverty alleviation. We previously conducted a special survey on targeted poverty alleviation of more than 20 listed companies and found a very interesting phenomenon. For private companies that have set up grassroots Party organiza- tions, the grassroots Party branches are responsible for their targeted poverty alleviation work, while the general trade unions are responsible for private companies that have not set up grass- roots Party organizations. Third, targeted poverty alleviation is not only a special social responsibility, but also a political task that must be completed for Party and government cadres at all levels. The construction of grassroots Party organizations is also convenient for the chair- man of private companies to accept and complete this political task. Based on the above analy- sis, this article proposes the following research hypotheses: Hypothesis 2: Compared with private companies that have not set up grassroots Party organi- zations, in private companies that have set up grassroots Party organizations, the chair- man’s CPC member status will promote targeted poverty alleviation even more significantly. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 6 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation 3. Empirical research 3.1. Sample selection and data source This paper selects the A-share private listed companies in Shanghai and Shenzhen Stock Exchanges from 2016 to 2020 as the initial sample. In September 2016, China’s Securities Reg- ulatory Commission encouraged the capital market and listed companies to participate in pre- cise poverty alleviation. Aiming to help the poor precisely, in December 2016, the Shanghai and Shenzhen stock markets in China simultaneously issued a notice requiring listed compa- nies to disclose information on precise poverty alleviation in the social responsibility section of their annual reports. Moreover, a comprehensive victory in the fight against poverty and the complete elimination of absolute poverty has made in 2020. Therefore, we choose the period of sample from 2016 to 2020. We exclude the companies of financial and insurance industries, ST and ST, and compa- nies that lack financial information. As a result, a total of 9,889 valid samples collected in 5 years were obtained. The identity data of the CPC members of the board of directors of private companies required for this study is non-mandatory disclosure information. We mainly obtain it manually through channels such as the WIND database, corporate website, Baidu Encyclopedia, Sina Finance, and Juchao Information. Targeted poverty alleviation data comes from the CNRDS database; other variables are all from the CSMAR database. Moreover, all continuous variables are winsorized at the 1% and 99% quantiles to alleviate the influence of extreme values on the results. 3.2. The variables 3.2.1. The dependent variables. 1. PA_D Whether to participate in the precision poverty alleviation dummy variable (PA_D), if it participates in the targeted poverty alleviation in the current year, the value is 1, otherwise it is 0. 2. PA_A The investment amount of targeted poverty alleviation (PA_A), equals to the natural logarithm of the total amount invested in the current year. 3.2.2. The independent variables. 1. CCPC Taking the leaders of the decision-making level, the chairman of the board as the object of investigation, constructing the dummy variable CCPC as chairman’s CPC member status. If the chairman is a member of the CPC, the value is 1; otherwise, is taken the value as 0. 2. Party Party organization building (Party) reflects the information whether a private listed company has established a grassroots CPC organization. If a grassroots CPC organization has been set up in the year, the value is 1. Otherwise, its value is 0. 3.2.3. Controlled variable. The control variables mainly include the following three cate- gories. (1) The control variables related to the individual characteristics of the chairman of the board [30], including: age (Age), gender (Gender), education (Edu), shareholding ratio (Share), and two-job integration (Dual). (2) Firm-level control variables [31,32], including firm size (Size), asset-liability ratio (Lev), profitability (RoA), operating capability (OC), corporate growth (Growth), holdings of the largest shareholder Share ratio (First), independent director PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 7 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation ratio (Inde) and listing age (FirmAge). (3) Annual and industry control variables. The main reason of poverty lies in the huge gap between rich and poor caused by uneven regional eco- nomic development should be considered. In addition, from the perspective of listed compa- nies participation in targeted poverty alleviation, there is serious regional mismatch between poverty alleviation resources and poverty alleviation needs, poverty alleviation may vary greatly between different regions. Therefore, regional control variables are also added to the model. Specific variable definitions are shown in Table 1. 3.3. Model In order to test the above-mentioned two research hypotheses, this paper constructs the fol- lowing multiple regression model: PA ¼ a0 þ a1CCPC þ X aiControls þ ε ð1Þ Controls stand for control variables. The dependent variables include two of PA_D and PA_A. PA_D is a dummy variable, the Logit regression model is used. PA_A gathers a large number of samples at 0 and Tobit regression model is used. When verifying research Table 1. The definition of variables. Type of variable symbol Variable name and metric Dependent variable PA_D Whether to participate in targeted poverty alleviation, private companies participating in targeted poverty alleviation is 1, otherwise it is 0 Independent variable PA_A CCPC Party The amount of targeted poverty alleviation investment is the natural logarithm of the total amount of private companies targeted poverty alleviation investment (Ten thousand RMB) Chairman Communist Party of China membership, if chairman of private companies is a CPC member, the value is 1, otherwise it is 0 Party organization construction, private companies in that year set up grassroots CPC organizations will take the value of 1, otherwise it is 0 Controlled variable Age, the natural log of the age of the chairman Age Gender Gender, with a value of 1 when the chairman was male and 0 for female Edu Share Dual Size Lev Roa OC Growth First Inde Education, 1 is for technical secondary school and below, 2 is for junior college, 3 is for bachelor’s degree, 4 is graduate student Holding proportion, the number of shares held by the chairman divided by the total number of shares The two positions are in one, and the chairman and CEO positions are worth 1, otherwise it is 0 Firm size, the natural log of the total assets of the company Financial leverage, liabilities divided by total assets Profitability, net profit divided by total assets Asset operating capacity, operating revenue divided by the average total assets Firm growth, the growth rate of operating revenue The shareholding proportion of the largest shareholder, the shareholding percentage of the largest shareholder The proportion of independent directors, the proportion of independent directors of the board of directors FirmAge The age of listing, the year since the listing time of private firms Year Ind Area Annual variable, taking a value of 1 when the observation is a given year, otherwise 0 Industry variable, value 1 when the observation belongs to a particular industry, or otherwise 0 Regional variable, taking a value of 1 when the observation value is registered to a particular province and city, or 0 otherwise https://doi.org/10.1371/journal.pone.0284692.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 8 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation hypothesis 1, the coefficient α1 of CCPC is expected to be significantly positive. To verify research hypothesis 2, it is expected that the coefficient of CCPC is more significant in private companies that have set up grassroots organizations of CPC, according to the Party organiza- tion construction (Party) variable group regression. 4. Empirical results 4.1. Descriptive statistics Table 2 shows the descriptive statistics of the variables including 9,889 observations. The mean value of PA_D is 0.183, suggesting that 18.3% of the observed private companies participate in targeted poverty alleviation and there is still a lot of room to explore. Moreover, it also shows that private companies are important force for targeted poverty alleviation. The standard devi- ation of PA_A is 5.342, indicating that the average investment in poverty alleviation funds of private companies participating in targeted poverty alleviation with large differences. The mean of CCPC of private companies is 31.1%, reflecting that although there is no requirement on whether the chairman of a private company is a member of the CPC, more than a quarter of the private corporate chairmen are still the CPC members. Part of this is the formation of the original government officials doing business and the reform of state-owned corporates, and some of them are private entrepreneurs who joined the CPC in the later period. Other control variables are consistent with previous studies. Table 2. Descriptive statistical results. Panel A: Descriptive statistical results of the whole sample Variables PA_D PA_A CCPC Party Age Gender Edu Share Dual Size Lev Roa OC Growth First Inde FirmAge N 9889 9889 9889 9889 9889 9889 9889 9889 9889 9889 9889 9889 9889 9889 9889 9889 9889 Mean 0.183 2.491 0.311 0.693 3.986 0.935 3.415 0.129 0.378 22.005 0.399 0.03 0.608 0.189 0.306 0.379 2.891 Std 0.387 5.342 0.463 0.461 0.156 0.247 0.844 0.152 0.485 1.092 0.199 0.097 0.389 0.478 0.131 0.052 0.32 Min Quantile 25 Median Quantile 75 0 0 0 0 0 0 0 0 3.219 3.912 0 0 0 0 19.735 0.06 -0.532 0.058 -0.682 0.081 0.333 0.693 1 3 0 0 21.225 0.239 0.015 0.355 -0.023 0.205 0.333 2.708 0 0 0 1 4.007 1 4 0.06 0 21.905 0.385 0.04 0.529 0.116 0.291 0.364 2.944 0 0 1 1 4.078 1 4 0.23 1 22.661 0.53 0.07 0.76 0.29 0.388 0.429 3.135 Max 1 17.623 1 1 4.443 1 4 0.56 1 25.342 0.962 0.211 2.354 3.099 0.671 0.571 4.143 Variable PA_D PA_A Sample number Sample of the chairman of the CPC members Sample of non-CPC member chairman Panel B: The univariate test results 0.240 3.306 3075 0.157 2.123 6814 T-test (T-Value) 0.082 (9.832) 1.182* (10.252) — Note: , , * * indicate the significance levels of 10%, 5%, and 1%, respectively. https://doi.org/10.1371/journal.pone.0284692.t002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 9 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation Panel B shows the test result of the univariate of targeted poverty alleviation between the sample of a CPC member and non-CPC chairman. Compared with the non-CPC chairman, if the chairman of the board is a member of the CPC, the willingness to participate and the amount of investment in targeted poverty alleviation of private companies are significantly greater, which implies that the chairman’s CPC membership of private companies plays a role in promoting targeted poverty alleviation. 4.2. Correlation coefficient analysis As the data shown in Table 3, the correlation coefficient between CCPC and PA_D is 0.098, which is significant at the 1% level. It implies that the chairman’s CPC membership is signifi- cantly positively correlated with the willingness of private companies to participate in targeted poverty alleviation. Moreover, the correlation coefficient between CCPC and PA_A is 0.103, which is significant at the 1% level. It implies that the chairman’s CPC membership is signifi- cantly positively correlated with the investment in targeted poverty alleviation in private com- panies. The above results support hypothesis 1. In addition, the correlation coefficients between all the variables are not above 0.5 except between PA_D and PA_A, suggesting that the subsequent regression conclusions are less affected by multicollinearity. 4.3. Regression results Table 4 illustrates the regression results for Model (1). Column (2) introduces control variables based on column (1), while Column (4) introduces control variables based on column (3) to make model more accurate. For regression of the four CCPC, the coefficients are all signifi- cant. Moreover, in column (2), the coefficient of CCPC is 0.3432, and it is significant at the 1% level, which shows that the chairman’s CPC membership can significantly increase the willing- ness of private companies to participate in targeted poverty alleviation. In column (4), the coef- ficient of CCPC is 0.7206, which is also significant at the 1% level. This shows that the chairman’s CPC members can also increase the amount of investment in targeted poverty alle- viation in private companies. The above conclusions support research hypothesis 1. From control variables of individual characteristics of the chairman, it takes column (2) as an example. The results show that female chairmen with higher education can promote the tar- geted poverty alleviation of private companies, while the chairman’s shareholding ratio inhib- its private companies fulfill their responsibilities for targeted poverty alleviation, similar with column (4). From the perspective of firm-level control variables, private companies with larger scale, stronger profitability, better growth and firm ages will more actively participate in tar- geted poverty alleviation with greater poverty alleviation efforts. 5. Mechanism: Party organizations Table 5 shows the regression results of the influence of chairman’s CPC membership on tar- geted poverty alleviation in private companies whether having been set up Party organizations. First, based on the method of group regression, the results are shown in column (2), (3), (5) and (6). Second, to make the conclusion more valid, we put the dummy variables for the Party branch setup in the model and moderate the impact of the CPC status. The results are pre- sented in column (1) and (4). According to Table 5, in column (2), the coefficient of CCPC is 0.3259, which is significant at the 1% level. In column (3), the coefficient of CCPC is not signif- icant. This indicates that in private companies with Party organizations, the status of chair- man’s CPC can significantly increase the willingness of companies to participate in targeted poverty alleviation. In private companies without a Party organization, the chairman’s CPC membership reduces the willingness of the company to participate in targeted poverty PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 10 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation e g A m r i F e d n I t s r i F h t w o r G C O a o R v e L e z i S l a u D e r a h S u d E r e d n e G e g A y t r a P C P C C A _ A P D _ A P e l b a i r a V . s i s y l a n a t n e i c i f f e o c n o i t a l e r r o c e h t f o s t l u s e R . 3 e l b a T 1 * * * 4 4 0 0 - . * * * 7 5 0 0 - . * * * 2 5 0 0 - . * * * 5 3 0 0 - . * * * 9 6 0 0 - . * * * 7 2 1 0 . * * * 9 1 1 0 . * * * 8 8 0 0 - . * * * 1 8 1 0 - . * * * 5 8 0 0 . * * * 8 3 0 0 - . * * * 8 4 0 0 . * * * 3 6 0 0 . * * * 8 0 1 0 . * * * 1 8 0 0 . * * * 3 8 0 0 . 1 * * * 9 3 0 0 . 1 * 7 1 0 0 . 4 0 0 0 - . * * * 7 1 1 0 . * * * 7 6 1 0 . 8 0 0 0 - . * * * 1 3 0 0 - . 6 1 0 0 - . 2 1 0 0 - . * * * 3 4 0 0 . * * * 8 6 0 0 . * * * 5 9 1 0 . * * * 5 7 0 0 - . * * * 7 4 0 0 - . * * * 4 3 0 0 . * * * 1 3 0 0 . * * * 8 2 0 0 . * * * 7 2 0 0 . * * 1 2 0 0 . * * * 9 8 0 0 - . * * * 6 3 1 0 . * * * 3 0 1 0 . * * 5 3 0 0 - . * * * 1 4 0 0 - . * * * 7 4 0 0 - . * * * 4 6 0 0 - . * * * 4 6 0 0 - . * * 1 3 0 0 - . * * * 9 2 0 0 - . t s r i F e d n I e g A m r i F . y l e v i t c e p s e r , % 1 d n a , % 5 , % 0 1 f o s l e v e l e c n a c i f i n g i s e h t e t a c i d n i * * * , * * , * : e t o N 3 0 0 t . 2 9 6 4 8 2 0 . e n o p . l a n r u o j / 1 7 3 1 . 0 1 / g r o . i o d / / : s p t t h 1 * * * 7 9 1 0 . * * * 6 3 2 0 . * * 6 2 0 0 . * * * 5 7 0 0 . 1 * * * 0 1 2 0 . * * * 8 9 0 0 . * * * 4 6 0 0 . 3 0 0 0 - . 5 1 0 0 . * * 3 2 0 0 . * * * 6 3 0 0 . 2 0 0 0 - . 3 2 0 0 . 3 1 0 0 . 0 1 0 0 . * * * 8 4 0 0 - . - * 7 1 0 0 - . 4 0 0 0 - . * * * 2 4 0 0 . 6 0 0 0 . 4 1 0 0 - . * * * 8 3 0 0 . * * * 6 3 0 0 . 2 1 0 0 - . 3 1 0 0 - . h t w o r G 1 * * * 0 5 3 0 - . * * * 8 5 0 0 . * * 1 2 0 0 . * * * 6 3 1 0 . * * * 4 5 0 0 - . 0 1 0 0 - . * * * 6 8 0 0 . * * * 6 4 0 0 . * * 3 2 0 0 . * * * 8 7 0 0 . * * * 4 7 0 0 . 1 * * * 2 3 4 0 . * * * 6 7 0 0 - . * * * 5 8 1 0 - . * * * 5 9 0 0 . * * 1 2 0 0 . * * * 6 9 0 0 - . * * * 6 6 0 0 . * * * 2 5 0 0 . * * * 6 6 0 0 . * * * 3 5 0 0 . 1 * * * 1 1 1 0 - . * * * 8 2 2 0 - . * * * 2 8 0 0 . 1 * * * 0 0 2 0 . * * * 0 3 0 0 . 4 1 0 0 . 3 1 0 0 . * * * 8 1 1 0 - . * * * 1 5 0 0 - . * * * 7 9 0 0 - . * * * 2 3 0 0 - . * * * 0 3 0 0 - . * * * 4 3 0 0 . * * * 5 4 1 0 . * * * 0 2 1 0 . * * * 2 2 2 0 . * * * 8 8 1 0 . 1 * * * 3 1 1 0 - . * * * 1 4 0 0 . * * * 4 9 0 0 . * * * 0 6 0 0 - . * * * 2 0 1 0 - . * * * 6 7 0 0 - . * * * 0 7 0 0 - . 1 . 5 0 0 0 0 - . * * * 8 8 1 0 - . 2 0 0 0 - . 4 1 0 0 - . * * * 0 4 0 0 . * * * 7 3 0 0 . 1 * * * 7 7 0 0 . * * * 4 4 0 0 . * * 4 2 0 0 . * * 1 2 0 0 - . * * 5 2 0 0 - . 1 * * * 3 4 0 0 . * * * 0 2 1 0 . * * * 2 3 0 0 . * * * 0 3 0 0 . * * * 4 3 2 0 . * * * 4 0 1 0 . * * * 8 9 0 0 . 1 * * * 3 0 1 0 . * * * 8 9 0 0 . 1 * * * 5 8 9 0 . 1 D _ A P A _ A P C P C C y t r a P e g A r e d n e G u d E e r a h S l a u D e z i S v e L a o R C O PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 11 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation Table 4. The regression results of the influence of chairman’s CPC membership on targeted poverty alleviation in private companies. Variable PA_D PA_A CCPC Age Gender Edu Share Dual Size Lev Roa OC Growth First Inde FirmAge Constant Year Industry Area Observations Pseudo R2 Wald Chi2/LR Chi2 (1) 0.4624* (8.3588) -1.7315* (-6.9009) control control control 9889 0.0419 381.43 (2) 0.3432* (5.9375) 0.2299 (1.1926) -0.2718* (-2.5769) 0.0753 (2.1626) -0.4805 (-2.3266) 0.0581 (0.9611) 0.3927* (13.1954) -0.0038 (-0.0206) 2.7199* (5.6214) 0.0799 (0.9952) -0.1657* (-2.6176) -0.0223 (-0.1011) -0.3691 (-0.6774) 0.2936* (2.9868) -12.1239* (-11.4323) control control control 9889 0.0807 644.46 (3) 1.0162* (8.8452) 2.7049* (5.0981) control control control 9889 0.0073 446.33 (4) 0.7206* (6.3086) 0.3727 (1.0694) -0.4972 (-2.3673) 0.1482 (2.3399) -0.9336 (-2.4822) 0.1033 (0.9315) 0.9834* (17.4814) -0.1204 (-0.3551) 3.6742* (5.7876) 0.2109 (1.3850) -0.3267* (-2.8299) 0.3251 (0.7768) -0.4791 (-0.4760) 0.4793* (2.7967) -21.8640* (-11.1287) control control control 9889 0.0304 949.15 Note: , , * * indicate the significance levels of 10%, 5%, and 1%, respectively. When the explained variable is PA_D, using the Logit model, the Z value of the coefficient is reported under the coefficient. When the explained variable is PA_A, the Tobit model is used, and the T value is reported in the square bracket under the coefficient (the same below). https://doi.org/10.1371/journal.pone.0284692.t004 alleviation. In column (5), the coefficient of CCPC is 0.7239 and is significant at the 1% level while in column (6), the coefficient of CCPC is not significant. It shows that in private compa- nies with the Party organizations, the chairman’s CPC members will have a significant positive effect on the company’s investment in targeted poverty alleviation. Otherwise, it will have a PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 12 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation Table 5. The channel role of CPC organization building. Variable CCPC CCPCParty Party Age Gender Edu Share Dual Size Lev Roa OC Growth First Inde FirmAge _cons Year Industry Area Observations Pseudo R2 Wald Chi2/LR Chi2 Full sample (1) 0.0518 (0.3497) 0.2686 (1.6809) 0.2793* (3.6949) 0.2385 (1.2382) -0.2921* (-2.7411) 0.0753 (2.1584) -0.4738 (-2.2848) 0.0587 (0.9686) 0.3830* (12.8350) -0.0159 (-0.0868) 2.6349* (5.4644) 0.0722 (0.8980) -0.1610 (-2.5284) -0.0353 PA_D Party = 1 (2) 0.3259* (5.0571) 0.0304 (0.1363) -0.1860 (-1.4285) 0.1138*** (2.7986) -0.4412* (-1.8309) 0.1516 (2.2057) 0.4040* (11.6525) -0.2074 (-0.9691) 2.1139* (4.1712) 0.0700 (0.7725) -0.1499 (-2.0170) -0.0831 (-0.1601) (-0.3391) -0.2668 (-0.4887) 0.2745* -0.3288 (-0.5088) 0.4237* Party = 0 Full sample (3) -0.0653 (-0.4167) 0.8094* (1.9465) -0.5326* (-2.7837) -0.0244 (-0.3513) -0.5757 (-1.3611) -0.2884 (-2.1555) 0.3441* (5.6038) 0.6102 (1.6425) 4.2301* (3.3270) -0.0278 (-0.1558) -0.1961 (4) -0.0306 (-0.1166) 0.7819* (2.6925) 0.4060* (3.1455) 0.3886 (1.1167) -0.5442* (-2.5926) 0.1486 (2.3492) -0.9163 (-2.4391) 0.1030 (0.9306) 0.9615* (17.0485) -0.1495 (-0.4414) 3.5222* (5.5514) 0.1958 (1.2877) -0.3208* (-1.5902) (-2.7814) 0.4050 (0.7876) 0.0478 (0.0441) -0.3837* 0.3216 (0.7692) -0.3142 (-0.3124) 0.4508* PA_A Party = 1 (5) 0.7239* (5.2521) Party = 0 (6) -0.0995 (-0.4433) 0.0194 (0.0439) -0.4521 (-1.5991) 0.2260*** (2.8280) -0.8958* (-1.8431) 0.3109 (2.1984) 1.0989* (15.3551) -0.4716 (-1.0669) 3.5612* (4.1686) 0.2251 (1.1630) -0.3419 (-2.1995) 0.1735 (0.3341) -0.2941 (-0.2247) 0.7408* 1.0619 (2.0074) -0.6621** (-2.3142) -0.0015 (-0.0151) -1.0001* (-1.8047) -0.3503 (-2.0991) 0.6532* (7.5833) 0.7261 (1.4744) 3.7200* (4.3101) -0.0178 (-0.0768) -0.3190 (-2.0392) 0.7983 (1.1777) -0.0900 (-0.0619) -0.5646 (2.8157) -12.0756* (3.6360) -12.0395* (-1.9335) -11.1544* (2.6327) -21.6257* (3.4416) -23.8865* (-2.0998) -14.1712*** (-11.4215) (-9.8760) (-4.9634) (-11.0212) (-9.6785) (-4.6098) Control Control Control 9889 0.0839 695.67 Control Control Control 6855 0.0803 511.31 Control Control Control 3034 0.0954 192.87 Control Control Control 9889 0.0160 979.58 Control Control Control 6855 0.0162 701.89 Control Control Control 3034 0.0157 276.29 https://doi.org/10.1371/journal.pone.0284692.t005 significant negative effect. Furthermore, by investigating the coefficient of interactive term between CCPC and the above dummy, it finds out that in column (1) the coefficient of CCPCParty is 0.2686, significant at the 10% level, and in column (4) the coefficient of CCPCParty is 0.7819, significant at the 1% level, which support the above conclusions. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 13 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation Combining the results of column (1) to (6), it implies that the grassroots Party branch is an important channel to play the role of the chairman’s CPC membership to promote targeted poverty alleviation in private companies. For private companies, having established grassroots Party organizations in these companies, the individual recognition of the Communist cultural values and the importance of the self-reputation, will be reflected in the targeted poverty allevi- ation responsibility. At this moment, the chairman’s CPC identity can speak out through Party organizations, and the Party organization is specifically responsible for the implementation of targeted poverty alleviation decisions. Moreover, it is also possible that after the establishment of grassroots Party branches, the governance mechanism of private companies is closer to that of state-owned companies, through the cross-employment of the Party organization and the corporate governance organization to strengthen the influence of the chairman’s CPC mem- bership on the performance of targeted poverty alleviation responsibilities in private compa- nies. On the contrary, the chairman’s CPC membership has even a negative effect on targeted poverty alleviation. Therefore, research hypothesis 2 has been supported. 6. Robustness test and endogenous problems 6.1. Replace dependent variables Table 6 reveals the robustness test results of adjusting the sample range. In previous research, we used the natural logarithm of the amount of investment in targeted poverty alleviation by private companies to describe the level of investment in targeted poverty alleviation to describe its relevant level. It is an absolute quantitative indicator, with insufficient comparability between companies of different sizes, which may have a certain impact on the previous research conclusions. Through the regression results of the previous article, the regression results of the firm-scale control variables also show that the firm-scale is an important factor that affects the degree of investment in targeted poverty alleviation by private companies. Therefore, in order to verify the robustness of the research conclusions of this paper, the previ- ous research conclusions were re-examined after standardizing the investment amount of pre- cision poverty alleviation with total assets. Thus, it can be found that after replacing the dependent variable, all the research conclusions remain robust. 6.2. Adjusting the sample range Table 7 represents the robustness test results of adjusting the sample range. The duration from 2016 to 2020 includes the outbreak of Covid-19 since the end of 2019. Covid-19 caused a big impact on companies all over the world, and many of them were in business difficulties includ- ing Chinese companies. In order to fight against the epidemic, companies were likely to abnor- mally reduce target poverty alleviation in 2020, thereby creating a braking effect. The braking effect might make the results unreliable. So, we deleted the data of year 2020 from the samples and conducted empirical regression based on data from 2016 to 2019. The results are shown in Table 7. The coefficients of CCPC in column (1) and (2) are positive, significant at the 1% level. The coefficients of CCPCParty and party in column (3) and (6) are positive and signifi- cant. The column (4) and (5), (7) and (8) are consistent with the former regression model in groups. Therefore, the conclusions are still valid. 6.3. PSM-paired samples Table 8 shows the results of the PSM robustness test. In order to alleviate the impact of the endogenous problems caused by self-selection on the conclusions of this research, we take the sample of the private company chairman who is a CPC member as the processing group, and PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 14 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation Table 6. The robustness test results of replacing independent variables. Variable CCPC CCPCParty Party Age Gender Edu Share Dual Size Lev Roa OC Growth First Inde FirmAge _cons Year Industry Area Observations R-squared F-Value Full sample Full sample PA_A (1) 0.0023 (2.1557) 0.0065 (2.0899) -0.0009 (-0.4286) 0.0017* (3.3277) -0.0017 (-0.4935) 0.0003 (0.2848) 0.0029* (4.7528) 0.0025 (0.8648) 0.0206* (5.2460) 0.0040 (2.5130) -0.0031* (-4.5921) 0.0016 (0.3949) -0.0066 (-0.8003) -0.0008 (-0.4958) -0.0619* (-3.0977) Control Control Control 9889 0.038 8.97 (2) -0.0032 (-2.2212) 0.0056* (2.9365) 0.0034* (3.2994) 0.0067 (2.1292) -0.0012 (-0.6154) 0.0017* (3.3393) -0.0015 (-0.4477) 0.0003 (0.2892) 0.0027* (4.4412) 0.0022 (0.7837) 0.0194* (4.9551) 0.0039 (2.4440) -0.0030* (-4.5164) 0.0016 (0.3806) -0.0053 (-0.6387) -0.0011 (-0.6268) -0.0600* (-3.0004) Control Control Control 9889 0.040 8.44 Party = 1 (3) 0.0023* (1.8244) Party = 0 (4) -0.0040 (-2.5451) 0.0036 (0.8809) -0.0002 (-0.0605) 0.0020* (2.9391) -0.0005 (-0.1078) 0.0015 (1.1661) 0.0031* (3.7888) 0.0009 (0.2192) 0.0243* (4.4126) 0.0063* (2.9747) -0.0042* (-4.2689) -0.0003 (-0.0618) -0.0028 (-0.2393) -0.0002 (-0.0876) -0.0522 (-1.9850) Control Control Control 6855 0.047 7.58 0.0136* (3.3464) -0.0025 (-0.9981) 0.0010* (1.7438) -0.0047 (-1.0770) -0.0028 (-2.2458) 0.0021* (2.9527) 0.0071 (2.2799) 0.0119 (2.5198) -0.0027 (-1.5550) -0.0011 (-1.6117) 0.0072 (1.0784) 0.0022 (0.2548) -0.0040 (-2.1183) -0.0812* (-3.4136) Control Control Control 3034 0.017 2.57 https://doi.org/10.1371/journal.pone.0284692.t006 use the PSM method to select the control group in the non-CPC chairman private company, and re-examine the previous research conclusions. The specific processes are as follows. (1) A sample of 3,620 companies whose chairman is a CPC member from 2016 to 2020 is used as the processing group. (2) Whether the chairman is a CPC member is taken as the indicator PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 15 / 20 PLOS ONETable 7. The robustness test results of adjusting the sample range. Chairman’s CPC member status and targeted poverty alleviation Party Age Gender Edu Share Dual Size Lev Roa OC Growth First Inde FirmAge _cons Year Industry Area Observations Pseudo R2 Wald Chi2/LR Chi2 Variable CCPC CCPCParty PA_D Full sample PA_A Full sample (1) 0.3274 (4.9313) (2) 0.6511* (5.1078) Full sample (3) -0.0051 (-0.0299) 0.3098* (1.6858) 0.2877*** (3.3282) 0.3871* (1.7101) -0.3265*** (-2.6659) 0.0690* (1.7733) -0.5406 (-2.2763) 0.0695 (0.9936) 0.3649* (10.3628) -0.0580 (-0.2706) 2.4387* (4.2397) 0.0739 (0.8082) -0.1868 (-2.4823) 0.0037 (0.0147) -0.2725 (-0.4267) 0.2530 PA_D Party = 1 (4) 0.3100* (4.1656) 0.2267 (0.8565) -0.2569* (-1.7213) 0.1099** (2.4130) -0.3867 (-1.4011) 0.1527* (1.9087) 0.3913* (9.5391) -0.1938 (-0.7665) 2.2906* (3.6359) 0.0583 (0.5641) -0.1801 (-2.0508) -0.0231 (-0.0815) -0.5012 (-0.6573) 0.3842* Party = 0 Full sample (5) -0.1264 (-0.7099) 0.7295 (1.5630) -0.4753 (-2.1481) -0.0494 (-0.6396) -1.0023 (-2.0517) -0.2340 (-1.5347) 0.3051* (4.2630) 0.4811 (1.1480) 3.0212 (2.4379) -0.0074 (-0.0368) -0.2151 (-1.5165) 0.3543 (0.6055) 0.5455 (0.4449) -0.3634 (6) -0.1210 (-0.4191) 0.7988 (2.4903) 0.4303* (2.9913) 0.6603* (1.6793) -0.6136* (-2.5850) 0.1357 (1.9725) -0.9755 (-2.3367) 0.1009 (0.8122) 0.8856* (13.8399) -0.1748 (-0.4572) 3.2919* (4.5972) 0.1809 (1.0721) -0.3591* (-2.8644) 0.3915 (0.8390) -0.3248 (-0.2886) 0.3993 PA_A Party = 1 (7) 0.6454* (4.1609) 0.4072 (0.8044) -0.6236* (-1.9400) 0.2155** (2.4715) -0.7183 (-1.3234) 0.2891* (1.8085) 1.0325* (12.5895) -0.3703 (-0.7357) 3.7991* (3.8999) 0.1975 (0.9152) -0.3868 (-2.2710) 0.2534 (0.4347) -0.4665 (-0.3143) 0.6468* Party = 0 (8) -0.1826 (-0.7399) 1.0216* (1.7559) -0.5407* (-1.6807) -0.0322 (-0.3046) -1.4254 (-2.3269) -0.2858 (-1.5420) 0.5584* (5.8240) 0.6352 (1.1558) 3.0481* (3.1657) -0.0279 (-0.1092) -0.3522 (-2.0879) 0.7928 (1.0608) 0.1962 (0.1230) -0.5171* 0.3868* (1.7095) -0.3078** (-2.5400) 0.0693* (1.7827) -0.5491 (-2.3193) 0.0694 (0.9916) 0.3745* (10.6509) -0.0515 (-0.2398) 2.5189* (4.3607) 0.0837 (0.9169) -0.1899 (-2.5386) 0.0227 (0.0891) -0.3809 (-0.5983) 0.2708** 0.6567* (1.6673) -0.5658 (-2.3816) 0.1351 (1.9604) -0.9962 (-2.3826) 0.1010 (0.8118) 0.9062* (14.1846) -0.1514 (-0.3952) 3.4445* (4.8062) 0.2002 (1.1852) -0.3628* (-2.8902) 0.4058 (0.8688) -0.5029 (-0.4463) 0.4231 (2.4231) -12.2482* (2.2449) -21.1447* (2.2846) -12.1816* (2.8970) -12.4085* (-1.6073) -10.1071* (2.1213) -20.9201* (2.7168) -23.7400* (-1.7682) -12.1085*** (-9.7627) (-9.4876) (-9.7379) (-8.5334) (-3.9042) (-9.4010) (-8.3927) (-3.5466) Control Control Control 7680 0.0799 516.42 Control Control Control 7680 0.0149 705.84 Control Control Control 7680 0.0834 535.76 Control Control Control 5255 0.0822 400.36 Control Control Control 2425 0.0900 147.45 Control Control Control 7680 0.0155 732.97 Control Control Control 5255 0.0162 533.75 Control Control Control 2425 0.0145 202.67 https://doi.org/10.1371/journal.pone.0284692.t007 variable, corporation size (Size), financial leverage (Lev), profitability (Roa), Asset Operations (OC), company growth (Growth) and age of market (FirmAge) are used as matching variables, using non-returning proximity matching to obtain 1,810 company annual samples as Control PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 16 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation Table 8. Results of the PSM robustness test. Variable CCPC CCPCParty Party Age Gender Edu Share Dual Size Lev Roa OC PA_D Full sample PA_A Full sample (1) 0.3225 (2) 1.0869* Full sample (3) 0.0359 PA_D Party = 1 Party = 0 (4) 0.2917* (5) -0.0909 0.1116 Full sample (6) PA_A Party = 1 Party = 0 (7) 0.9675* (8) -0.3657 (4.3546) (4.4660) (0.1967) (3.4700) (-0.4591) (0.1840) (3.5277) (-0.6085) 0.2470 (1.2449) 0.3513* (3.7108) 0.8323 (1.2649) 1.2146* (3.9931) 0.2960 1.0894 0.3153 0.1973 0.6442 1.1473 0.6696 (1.2889) -0.3751* (1.4320) -1.0373 (1.3735) -0.4047* (0.7480) -0.3351 (1.2475) -0.5967 (1.5137) -1.1288 (0.7688) -1.0393* 2.7347* (1.8040) -1.3667* (-2.7304) (-2.2893) (-2.9204) (-1.9964) (-2.5296) (-2.4985) (-1.8927) (-1.7754) 0.0479 0.1777 0.0431 0.0575 -0.0122 0.1603 0.2011 0.0104 (1.0908) -0.6285 (1.2375) -1.7476 (0.9808) -0.6045** (1.1186) (-0.1351) -0.4591 -0.8287 (-2.3355) (-1.9894) (-2.2379) -0.0201 -0.0683 -0.0142 (-1.4168) 0.1552* (-1.5387) -0.5710* (-0.2672) -0.0354 (-0.2757) 0.3281* (-0.1887) (1.7556) (-3.5241) -0.0474 -0.0276 -0.0405 (1.1203) -1.6448* (-1.8790) -0.0512 (-0.2074) 0.2870** (1.1942) (0.0393) -1.0924 -2.4709 (-1.0409) 0.4924* (-1.5699) -1.6969* (1.6958) 0.3724* (-3.6918) 0.2511 (-0.9492) (2.6873) (-1.2655) (-0.6307) (-0.5174) (2.3538) (2.6049) (1.0963) 0.1523 0.6483 0.1556 0.0674 0.4820 0.6687 0.3990 1.6455 (0.6041) (0.7863) (0.6153) (0.2229) (0.9263) (0.8142) (0.4081) 0.2429 1.6269 0.1612 -0.6991 1.9876 1.3462 -0.9375 (0.4328) (0.8957) (0.2871) (-0.9908) -0.1470 -0.4123 -0.1432 -0.1149 (1.6303) -0.3932* (0.7437) (-0.4159) -0.3934 -0.1983 (1.0899) 5.7766* (1.9220) -1.3552 (-1.5149) (-1.3025) (-1.4695) (-0.9992) (-1.8687) (-1.2476) (-0.5269) (-2.2879) Growth 0.1136 0.2404 0.1236 0.1721 0.0439 0.2686 0.4248 -0.0450 First Inde FirmAge (1.2973) (0.8523) (1.4004) (1.5807) (0.2749) (0.9551) (1.2117) (-0.0986) 0.0537 0.5039 0.0060 -0.0583 0.3871 0.3316 0.0395 1.8916 (0.1994) (0.5708) (0.0220) (-0.1885) (0.6500) (0.3768) (0.0389) (1.0800) -0.0653 -0.3723 0.1083 0.4763 -0.6836 0.2280 1.8551 -2.0477 (-0.0954) -0.2950 (-0.1641) -1.1322* (0.1574) -0.3261* (0.5811) -0.2055 (-0.4735) -0.9474* (0.1007) -1.2376* (0.6912) -0.8765* (-0.4929) -2.8707* (-2.4988) (-2.9140) (-2.7602) (-1.4818) (-3.6863) (-3.1924) (-1.9217) (-3.9198) _cons 0.6336 -0.8812 0.6556 0.3316 2.1859 -0.7891 -1.7212 1.1950 (0.4864) (-0.2046) (0.5014) (0.2214) (0.7458) (-0.1839) (-0.3488) (0.1376) Control Control Control 3620 0.0399 183.83 Control Control Control 3620 0.0097 237.02 Control Control Control 3620 0.0449 203.38 Control Control Control Control Control Control 2695 0.0453 154.71 925 0.0742 88.53 Control Control Control 3620 0.0109 265.05 Control Control Control Control Control Control 2695 0.0107 195.01 925 0.0182 111.23 Year Industry Area Observations Pseudo R2 Wald Chi2/LR Chi2 https://doi.org/10.1371/journal.pone.0284692.t008 group. (3) Re-examine the previous research conclusions with annual samples of 1,810 compa- nies in the processing group and the control group. To sum up, the research conclusions obtained after using the PSM matching samples are consistent with previous conclusions. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 17 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation 6.4. Endogenous problems There is also a potential endogeneity that the likelihood of firms having a chairman with or without the CPC status may be explained by firm characteristics. In order to deal with this problem, referring Larcker and Rusticus (2010) [33], Impact Threshold for a Confounding Variable (denoted as ITCV) is used to measure whether the endogeneity problem changes the regression results. If the endogeneity problem is not severe enough to affect the direction and significance of the OLS regression results, the problem can be ignored and the regression results are considered robust. ITCV is measured by the bias correlation between the dependent variable and unobservable variables multiplied by the bias correlation between the indepen- dent variable and unobservable variables, and is also the minimum value at which the signifi- cance of the results is altered. Exceeding the ITCV indicates that the endogeneity problem is severe enough to alter the regression results. Higher ITCV values indicate that the OLS results are more robust. As shown in Table 9, column (2) and (6) represents the regression results. The ITCV value in column (2) is 0.045, implying that once the biased correlation between PA_D and unobserv- able variables reach 0. 21(0. 0.045^1/2), the significance of the OLS estimates changes. In addi- tion, according to the coefficients in column (3) and (4), none of them is more than 0.21. Moreover, the ITCV value in column (6) is 0.0496, implying that once the biased correlation between PA_A and unobservable variables reach 0.22(0.0496^1/2), the significance of the OLS estimates changes. In addition, according to the coefficients in column (7) and (8), none of them is more than 0.22. Therefore, it can be concluded that the findings of this paper are robust. 7. Conclusions and implications This paper investigates the influence of the chairman’s CPC membership on targeted poverty alleviation based on the hand-collected data of chairman’s CPC membership and the data of A-share listed private companies from 2016 to 2020. The research finds out the chairman of a board as a CPC member not only strengthens the willingness of private companies to Table 9. The impact of unobservable confounding variables. (1) PA_D (2) (3) (4) (5) Coefficient(P-Value) 0.3616(0.000) ITCV Impact 0.0450 Impact raw Coefficient(P-Value) CCPC Age Gender Edu Share Dual Size Lev Roa OC Growth First Inde FirmAge 0.2678(0.157) -0.2597(0.011) 0.794(0.020) -0.4258(0.036) 0.0275(0.642) 0.3876(0.000) -0.1295(0.447) 2.8575(0.000) 0.1309(0.062) -0.2467*(0.000) 0.0107(0.960) -0.2933(0.581) 0.4497(0.000) 0.0021 0.0035 -0.0005 -0.0006 -0.0003 -0.0005 0.0021 -0.0001 0.0096 -0.0001 0.0014 -0.0001 0.0005 0.0004 0.0002 0.0044 0.0072 0.0029 0.0226 0.0027 0.0017 0.0002 0.0002 0.0006 0.0018 0.0089 0.7836(0.000) 0.4777(0.174) -0.4841(0.023) 0.1564(0.014) -0.8907(0.018) 0.0593(0.596) 0.9939*(0.000) -0.3274(0.318) 3.8749(0.000) 0.2976(0.035) -0.4811(0.000) 0.4247(0.307) -0.2603(0.798) 0.7929*(0.000) PA_A (6) (7) (8) ITCV Impact 0.0496 Impact raw 0.0022 0.0038 -0.0005 -0.0005 -0.0001 -0.0006 0.0021 -0.0001 0.0115 -0.0001 0.0014 -0.0001 0.0006 0.0005 0.0002 0.0040 0.0077 0.0031 0.0267 0.0034 0.0018 0.0002 0.0002 0.0008 0.0020 0.0087 https://doi.org/10.1371/journal.pone.0284692.t009 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 18 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation participate in targeted poverty alleviation, but also increases the amount of investment in tar- geted poverty alleviation by private companies. Moreover, the establishment of the CPC branch is an important channel for the chairman’s CPC member status to promote private companies to fulfill the responsibility of targeted poverty alleviation, and the construction of the CPC organization of private companies can strengthen the role of the chairman’s CPC member status in promoting targeted poverty alleviation. In addition, the above research con- clusions are still valid under replacing the explained and explanatory variable, and PSM paired sample alleviating endogeneity. The great success of targeted poverty alleviation is not the end of poverty alleviation, but a new beginning. For a long time in the future, poverty alleviation will become one of the impor- tant social responsibilities of listed companies. Therefore, this paper is of significance to con- solidate and expand the achievements of poverty alleviation and rural revitalization. And for the government, private companies have developed rapidly and become an important part of the Socialist economy with Chinese characteristics. It is necessary to attract more outstanding private entrepreneurs to join the CPC. Moreover, the construction of the CPC organization of private companies will be greatly accelerated to unleash maximum advantage of socialist mar- ket economy. Finally, for private companies, it is necessary to strengthen the construction of Party organizations, give play to the advanced role of CPC members, strive to fulfill social responsibilities, and continue to contribute to the cause of poverty alleviation. Author Contributions Data curation: Ruirui Gu. Funding acquisition: Jian Xie, Ruian Yu. Methodology: Jian Xie. Software: Ruirui Gu, Sen Yang. Supervision: Tianyi Lei, Ruian Yu. Writing – original draft: Jian Xie, Tianyi Lei, Sen Yang. Writing – review & editing: Jian Xie, Tianyi Lei. References 1. Hambrick D. & Mason P. (1984) Upper Echelons: The Organization as a Reflection of Its Top Managers. Academy of Management Review, 9: 193–206. 2. Aguinis H. & Glavas A. (2012) What We Know and Don’t Know About Corporate Social Responsibility: A Review and Research Agenda. Journal of Management, 38(4):932–968. 3. Xu Z. & Hou J. (2021) Effects of CEO overseas experience on corporate socia1 responsibility evidence from Chinese manufacturing listed companies. Sustainability, 13: 1–24. 4. Zhang Qiang, Wang Bo, Liao ChengZhong. (2020) An Investigation of the Relationship Between Corpo- ration Social Responsibility and Senior Managers Influence & Poverty Experience. Journal of Finance and Accounting, 8(1):1–8. 5. Dimaggio P. J. (1997) Culture and Cognition. Metaphilosophy, 23(1):263–287. 6. Zingales L. (2015) The ‘Cultural Revolution’ in Finance. Journal of Financial Economics, 117(1):1–4. 7. Wang S., et al. (2015) Opening the Black Box of CSR Decision Making: A Policy-Capturing Study of Charitable Donation Decisions in China. Journal of Business Ethics, 128(3):665–683. 8. Harvey C, et al. (2021) The Ethics of Entrepreneurial Philanthropy. Journal of Business Ethics, 171 (1):33–49. 9. Velte P. & Stawinoga M. (2017) Empirical research on corporate social responsibility assurance (CSRA): A literature review. Journal of Business Economics, 87(8):1–50. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 19 / 20 PLOS ONEChairman’s CPC member status and targeted poverty alleviation 10. Mcclelland P. L., et al. (2012) CEO Career Horizon and Tenure: Future Performance Implications under Different Contingencies. Journal of Business Research, 65(9):1387–1393. 11. Oh W. Y., Chang Y. K. & Cheng Z. (2016) When CEO Career Horizon Problems Matter for Corporate Social Responsibility: The Moderating Roles of Industry-Level Discretion and Blockholder Ownership. Journal of Business Ethics, 133(2):279–291. 12. Chen J., Zhang Z. & Jia M. (2019) How CEO narcissism affects corporate social responsibility choice? Asia Pacific Journal of Management, 38(4):1–28. 13. Safi A., et al. (2021) CEO overconfidence, firm-specific factors, and systemic risk: evidence from China. Risk Management, 23(1): 30–47. 14. Griffin D. W., et al. (2012) How Does Culture Influence Corporate Risk-Taking? Journal of Corporate Finance, 23(4):1–22. 15. Wu T. (2019). A review of the impact of traditional culture on auditing. American Journal of Industrial and Business Management, 9(3):780–788. 16. Michaels A. & Gru¨ning M. (2018) The impact of corporate identity on corporate social responsibility dis- closure. International Journal of Corporate Social Responsibility, 3(1), 1–13. 17. 18. 19. Lin H., et al. (2014) Can political capital drive corporate green innovation? lessons from china. Journal of Cleaner Production, 64(feb.1), 63–72. Zhang H., et al. (2014) Political connections, government subsidies and firm financial performance: evi- dence from renewable energy manufacturing in china. Renewable Energy, 63, 330–336. TW Li & Shi L. (2021) Donation, political connection and credit access: the case of Chinese small and medium companies. Finance Research Letters, 7, 102269. 20. Giuli A. D. & Kostovetsky L. (2014) Are Red or Blue Companies More Likely to Go Green? Politics and Corporate Social Responsibility. Journal of Financial Economics, 111(1):158–180. 21. Hutton I., Jiang D. & Kumar A. (2014) Corporate Policies of Republican Managers. Journal of Financial and Quantitative Analysis, 49(5–6):1279–1310. 22. Francis B. B., et al. (2016) CEO Political Preference and Corporate Tax Sheltering. Journal of Corporate Finance, 38(C):37–53. 23. Christensen D. M., et al. (2015) Top Management Conservatism and Corporate Risk Strategies: Evi- dence from Managers’ Personal Political Orientation and Corporate Tax Avoidance. Strategic Manage- ment Journal, 36(12):1918–1938. 24. Elnahas A. M. & Kim D. (2017) CEO Political Ideology and Mergers and Acquisitions Decisions. Journal of Corporate Finance, 45:162–175. 25. Lin H. C., Chou T. K. & Wang W. G. (2012) Capital structure and executive compensation contract design: a theoretical and empirical analysis. Journal of Banking & Finance, 36(1), 209–224. 26. Nain A., & Yao T. (2013) Mutual fund skill and the performance of corporate acquirers. Journal of Finan- cial Economics, 110(2): 437–456. 27. Billon S. & Gillanders R. (2016) State ownership and corruption. International Tax and Public Finance, 23:1074–1092. 28. 29. Fu P. P, & Tsui A. S. (2003) Utilizing Printed Media to Understand Desired Leadership Attributes in the People’s Republic of China. Asia Pacific Journal of Management, 20(4):423–446. Tietz M. A. & Parker S. C. (2014) Charitable donations by the self-employed. Small Business Econom- ics, 43(4), 899–916. 30. Bass, B. M. & Stogdill, R. M. Handbook of leadership: theory, research, and managerial applications. New York, NY: Free Press, 59–77. 31. Huang H., Shang R., Wang L. & Gong Y. (2022). Corporate social responsibility and firm value: evi- dence from Chinese targeted poverty alleviation. Management Decision, (ahead-of-print). 32. 33. Zhang H., Huang J., Wu K., Wang S., Nygaard C. & Qiu Y. (2022). Do political connections affect corpo- rate poverty alleviation decisions? Evidence from China. China Economic Review, 73, 101785. Larcker D. F. & Rusticus T. O. (2010) On the use of instrumental variables in accounting research. Jour- nal of accounting and economics, 49(3), 186–205. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284692 June 15, 2023 20 / 20 PLOS ONE
10.1371_journal.pone.0284699	RESEARCH ARTICLE COVID-19 pandemic: Impact on the cardiac implantable electronic devices’ implantation rates in Croatia Ivan Zeljkovic1‡, Jerko Ferri2‡, Pavao MiocID Zrinka Jurisic5, Janko Szavits Nossan6, Lana MaricicID Vjekoslav Radeljic1, Nikola Pavlovic9, Sime Manola9 1, Sandro BrusichID 3, Richard Matasic4, 7, Katica Cvitkusic Lukenda8, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Zeljkovic I, Ferri J, Mioc P, Brusich S, Matasic R, Jurisic Z, et al. (2023) COVID-19 pandemic: Impact on the cardiac implantable electronic devices’ implantation rates in Croatia. PLoS ONE 18(4): e0284699. https://doi.org/ 10.1371/journal.pone.0284699 Editor: Simone Savastano, Fondazione IRCCS Policlinico San Matteo, ITALY Received: January 30, 2022 Accepted: April 6, 2023 Published: April 26, 2023 Copyright: © 2023 Zeljkovic et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Funding: The authors received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. 1 Department of Cardiology, Sestre Milosrdnice University Hospital, Zagreb, Croatia, 2 Department of Cardiology, Dubrovnik General Hospital, Dubrovnik, Croatia, 3 Department of Cardiology, Rijeka University Hospital, Rijeka, Croatia, 4 Department of Cardiology, Zagreb University Hospital, Zagreb, Croatia, 5 Department of Cardiology, Split University Hospital, Split, Croatia, 6 Department of Cardiology, Magdalena Clinic, Krapinske Toplice, Croatia, 7 Department of Cardiology, Osijek University Hospital, Osijek, Croatia, 8 Department of Cardiology, General Hospital “Dr. Josip Bencevic”, Slavonski Brod, Croatia, 9 Department of Cardiology, Dubrava University Hospital, Zagreb, Croatia ‡ IZ and JF are share first authorship on this work. pavomioc@hotmail.com Abstract Introduction Coronavirus disease 2019 (COVID-19) pandemic has influenced health-care organization worldwide, including management of non-communicable diseases. The aim of this study was to determine the impact of COVID-19 pandemic on cardiac implantable electronic devices’ (CIEDs) implantation rates in Croatia. Methods A retrospective, observational, national study was conducted. The data on CIEDs’ implanta- tion rates from 20 Croatian implantation centres, between January 2018 and June 2021, were extracted from the national Health Insurance Fund registry. Implantation rates before and after COVID-19 pandemic started, were compared. Results The overall numbers of CIED implantations in Croatia during COVID-19 pandemic were not different in comparison to 2 years pre-COVID-19 time (2618 vs. 2807, p = .081). The pace- maker implantation rates decreased significantly (by 45%) during April (122 vs. 223, p < .001) and May 2020 (135 vs. 244, p = .001), as well as during November 2020 (177 vs. 264, p = .003), but significantly increased during summer months 2020 comparing to 2018 and 2019 (737 vs. 497, p<0.001). The ICD implantation rates decreased significantly by 59% in April 2020 (26 vs. 64, p = .048). PLOS ONE \| https://doi.org/10.1371/journal.pone.0284699 April 26, 2023 1 / 8 PLOS ONECOVID-19 impact on CIED implantation Conclusion To the authors best knowledge this is a first study including complete national data on CIED implantation rates and COVID-19 pandemic impact. A significant reduction in number of both pacemaker and ICD implants during specific months of the COVID-19 pandemic was determined. However, afterwards compensation in implants resulted in similar total number when the complete year was evaluated. Introduction Coronavirus disease 2019 (COVID-19), caused by the novel SARS-CoV2 virus, started to spread all over the world in December 2019, and has soon confirmed its pandemic potential. The first cases reported in Europe were on 24 January 2020, and by the end of February, more than 1500 daily new cases were reported [1]. The pandemic was officially declared by the World Health Organization on 11 March 2020 [2]. Since the COVID-19 pandemic has started, health care systems around the world began to change in order to provide the optimal care for COVID-19 patients. However, the necessary reorganization of health care systems has largely influenced the management of non-COVID-19 patients, who were underprioritized. Patients with cardiovascular diseases were not an exception to this general trend. Previous publications demonstrated the overall decrease of interventional cardiology procedures during COVID-19 [3–6]. Moreover, a substantial decline in the number of cardiac implantable electronic devices (CIEDs) has also been reported [7–10]. To date, the impact of COVID-19 pandemic on CIEDs implantation rates in Croatia hasn’t been determined, thus, the aim of this study was to explore whether the COVID-19 pandemic has altered the pacemaker and ICD implantation rates in Croatia. Methods This retrospective, observational study considered the total number of CIED implants and its generator replacements in 20 Croatian implantation centres between January 2018 and June 2021. The data were extracted from the Croatian Health Insurance Fund registry, in which the rates of implants and replacements from all centres in Croatia are recorded. Croatian Health Insurance Fund is one obligatory health insurance in Croatia and all the centres are required to report the procedures and the underlying diagnoses for all patients. These data are public and could be accessed on web [11]. The data are published in excel tables that include a detailed information on procedures based on diagnoses. The data presented in this study were completely extracted from these datasets. Out of 20 implanting centres, 12 are implanting both ICDs and pacemakers and 8 pacemakers only (S1 Table). One centre located in Zagreb (capital city) became dedicated COVID-19 hospital, whereas four other city’s hospitals hospitalized only non-COVID-19 patients, and the rest of the hospitals around the country treated both groups of patients. For the purposes of this study, time has been divided into two periods: "before COVID-19 period" (BCP) and "COVID-19 period" (CovP). In Croatia the first COVID-19 case was diag- nosed on February 26th, and the COVID-19 epidemic was officially declared on March 16th, 2020. Hence, data from March 2020 were excluded from the statistical analysis to avoid over- or under-estimation bias. The Ethics Committee of the coordinating centre (Sestre milosrdnice University Hospital, Zagreb) was consulted waiving the need for approval given the nature of the data. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284699 April 26, 2023 2 / 8 PLOS ONECOVID-19 impact on CIED implantation Statistical analysis Continuous variables were presented as means with standard deviations or median with inter- quartile range in case of skewed distribution. Continuous variables were compared with non- parametric Mann-Whitney U test. Categorical variables were presented as absolute values and/or percentages and were compared using Pearson χ2 test. Descriptive statistics were calcu- lated for the available cases. Two-sided P-value of < 0.05 was considered significant. All statis- tical analyses were performed using SPSS software for Windows, version 22.0 (IBM SPSS Statistics, Armonk, NY, USA). Results The total number of pacemaker implantations during 2020 was lower when compared to an average number of implants in two earlier years, however it did not reach statistical signifi- cance (BCP 2807 vs. CovP 2618, p = .081, Table 2). An average of 234 pacemakers per month were implanted in BCP in comparison to 206 pacemakers per month in CovP (p = NS). When the implant numbers were compared for specific months, there was a significant decrease in pacemaker implants during April (by 45%) and May 2020 (by 45%) when compared to both months in 2018/2019 (April 223 vs. 122, p < .001; May 244 vs. 135, p = .001). Also, significant decrease (by 33%) was noted in November 2020 (264 vs. 177, p = .003). In contrast, during summer months (July-August) 2020, there was a significant increase in number of pacemaker implants when compared to 2018/19 (497 vs. 737, p = .001). There was no significant differ- ence between the number of pacemaker implants between the BCP and CovP when the num- ber were compared monthly, bimonthly, quarterly, half-yearly or in other combination (Table 1 and 2, S2 Fig). When we compared first 6 months of 2021 with appropriate months of 2018/2019, there were no significant differences no matter if we compared monthly, bimonthly, quarterly or 6-months data (Tables 1 and 2). When the number of ICD implants was compared between BCP and CovP, there was a sig- nificant decrease of implants during April 2020 by 59% (64 vs. 26, p = 0.048). There was no other statistically significant difference between the BCP and CovP regarding ICD implants, including monthly, bimonthly, quarterly number or any other combination (Tables 3 and 4 Table 1. Impact of COVID-19 pandemic on pacemaker implantation rates in Croatia, single month analysis. Timeline January February March April May June July August September October November December Pacemaker implantation (n) p value 2018/2019 vs 2020 p value 2020 vs 2021 2018/2019** 2020 2021 186 227 261 223 244 324 251 117 231 257 264 239 210 218 249 122 135 228 301 208 286 284 177 193 211 223 281 234 264 265 262 216 242 374 119 259 .891 .845 .441 < .001 .001 .082 .659 .222 .165 .262 .003 .151 .930 .757 .456 . 611 .444 .676 .752 .890 .285 .041 .215 .752 * April-May 2018/2019 compared to April-May 2020/2021 mean value https://doi.org/10.1371/journal.pone.0284699.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284699 April 26, 2023 3 / 8 PLOS ONETable 2. Impact of COVID-19 pandemic on pacemaker implantation rates in Croatia, yearly, bimonthly, quarterly and 6-months analysis. Pacemaker implantation (n) p value 2018/2019 vs 2020 p value 2020 vs 2021 COVID-19 impact on CIED implantation Timeline January-December January-February April-May June-August September-December November-December April-December January-June 2018/2019 447 2807 467 467 497 958 503 1922 1374 377 2020 2618 257 737 858 370 1852 1454 2021 2926 434 498 478 752 378 2207 1395 * April-May 2018/2019 compared to April-May 2020/2021 ** mean value https://doi.org/10.1371/journal.pone.0284699.t002 .081 .687 .001 .020 .001 .085 .013 .431 .910 .752 .473 < .001 .065 .921 .252 .931 and S3 Fig). However, during April and May 2020, 4 out of 12 centres which implant ICD, reported no ICD implants. In addition, there was no significant difference in number of ICD implants during 2021 when compared to 2018/2019 (Table 4). Discussion The main findings of this multi-centre, national, registry-based study, which included data from all 20 CIED implanting centres in Croatia, are the following: 1) the overall numbers of CIED implants during COVID-19 pandemic were not different in comparison to 2 years pre- COVID time; 2) the pacemaker implantation rates decreased significantly during two months after the official declaration of COVID-19 epidemics (April and May 2020), as well as during November 2020; 3) the pacemaker implantation rates significantly increased during summer months 2020 in comparison to 2018/2019; 4) the ICD implantation rates decreased signifi- cantly during April 2020. Table 3. Impact of COVID-19 pandemic on ICD implantation rates in Croatia, single month analysis. Timeline January February March April May June July August September October November December ICD implantation (n) p value 2018/2019 vs 2020 p value 2020 vs 2021 2018/2019* 2020 2021 43 46 55 64 53 46 52 37 31 32 57 60 48 64 36 26 55 65 63 65 68 64 57 61 38 67 64 51 64 49 65 44 62 59 78 68 .992 .325 .412 .048 .890 .562 .756 .216 .121 .085 .995 .937 .878 .925 .212 .218 .188 .596 .989 .653 .898 .752 .662 .802 ICD–implantable cardioverter defibrillator mean value https://doi.org/10.1371/journal.pone.0284699.t003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284699 April 26, 2023 4 / 8 PLOS ONECOVID-19 impact on CIED implantation Table 4. Impact of COVID-19 pandemic on ICD implantation rates in Croatia yearly, bimonthly, quarterly and 6-months analysis. Timeline January-December January-February April-May June-August September-December November-December April-December January-June ICD implantation (n) p value 2018/2019 vs 2020 p value 2020 vs 2021 2018/2019 2020 2021 625 89 118 142 226 117 486 306 673 112 81 156 250 118 487 295 725 124 115 109 205 146 536 361 .552 .452 .137. .091 .484 .945 .992 .912 .483 .821 .240 .432 .212 .356 .742 .201 ICD–implantable cardioverter defibrillator **mean value https://doi.org/10.1371/journal.pone.0284699.t004 We analysed the overall national CIED implantation activity in Croatia before and during the COVID-19 pandemic. A significant reduction in a pacemaker implantation rates first two months after the official declaration of COVID-19 epidemic was determined. That was the period of the strictest anti-COVID-19 measures in Croatia. During these periods there was a significant decrease in number of visits in emergency departments that were not related to COVID-19 which we detected as a main reason of the observed decrease in implantation rate. The same trend was observed in November 2020 occurring during the peak of the second “COVID-19 wave” in Croatia which led to the second “lock down” in our country [12]. Monthly data are similar to the results of studies conducted in Italy, Spain, Peru and Portugal [7–10, 13]. Interestingly, in our study we also showed a significant increase in pacemaker implantation rates during summer 2020. This could possibly be explained by the fact that the numbers of COVID-19 cases in Croatia were relatively low, and the hospitals were gradually re-opening for non-urgent visits [12]. Patients who were not electively admitted to the hospital during lockdown, alongside with those who were previously scheduled for implantation, could have contributed to the overall increase in number of reported procedures during this period. Thus, in Croatia the total number of pacemaker implants per year did not differ in the BCP and CovP. There are no previous studies considering complete national data and including impact of different COVID-19 waves on CIED implantation rates [7–10]. Interestingly, there was no difference in implantation rates during 2021 despite having “lockdown” measures and third “COVID-19 wave”. It is important to emphasize that the indications for pacemaker implantations were the same in the pre-COVID period as well as during pandemic. Similarly, the ICD implantation rates declined continuously, but the statistically significant decrease was noted only in April 2020, whereas this was not noted in the rest of the following months of COVID-19 pandemic, including first 6 months of 2021. However, in one third of Croatian implantation centres in April and May 2020, no ICD implants were reported. Regarding ICD implantation, our data are also in line with previous results collected in Italy, Spain and Peru [7–10]. However, there was no compensation noted during summer period. This could be explained by the healthcare organisation in COVID-19 pandemic, and the fact that one implanting centre in the capital of Croatia (Zagreb) became dedicated COVID-19 centre, and the remaining 4 centres managed only non-COVID-19 patients. Determination of casual connections is beyond the scope of this paper, but the results provide a ground for a dis- cussion on potential causes that have led to the observed changes. Arbelo et al. [8] and Boriani et al. [7] in their papers speculated on different moments that are likely to give rise to the PLOS ONE \| https://doi.org/10.1371/journal.pone.0284699 April 26, 2023 5 / 8 PLOS ONECOVID-19 impact on CIED implantation reduction in implantation rates in Spain and Peru, respectively. Government directives on cancelling the elective procedures have formally made the treatment of non-urgent patients impossible. What is more, fear of contagion, associated with the individual perception of the risk present in the hospital areas could be a second important cause. Finally, during COVID- 19 “lockdown” the collective awareness of non-COVID-19 pathologies was reduced. Taking into consideration that the SARS-CoV2 has become almost ubiquitous and the fight against the pandemic is similar worldwide, we think that the underlying causes of the reductions in implant rates are similar, particularly comparing the countries within the European Union. Lockdown during COVID-19 pandemic caused reduction in emergency departments visits not only in Croatia, but also in other European countries [14], but on the other hand SARS- CoV2 infection and its complications, alongside with medication used in treatment COVID- 19 patients contributed to the relatively constant CIEDs’ implantation rates during pandemic. Transient decrease in implantation rates was due to lockdown, however in the following months the numbers were increasing. We see it as a “compensation” period. During summer 2020 there was not statistically significant increase in number of CIEDs implantation, but in July 2020 there was the greatest number of implanted pacemakers in last two years (301; Table 1, S2 Fig). We argue that the reason for the observed change in trends could be the fact that only highly symptomatic patients visited emergency departments during lockdown, as discussed by Colivicchi et al. and Baldi et al. [14, 15], and the others were just postponing their visits to the period after the lockdown. This multicentre retrospective study has several limitations. First, due to the limitation in registry’s data, especially during the COVID-19 pandemic, we have not been able to extract detailed data on relevant variables such as patients’ socio-demographic status, indication for CIED implantation, device type or clinical presentation. Second, the observational nature of the study did not allow any conclusion to be drawn in relation to the underlying causes for the significant variations in CIED implantation rates during the COVID-19 pandemic. Conclusion A significant reduction in total number of pacemaker implants during April, May and Novem- ber 2020 was determined in Croatia during the COVID-19 pandemic, whereas an increase of pacemaker implants was observed in the period between two “COVID-19 waves”. Regarding the ICD implants, significant decrease was noted only in April 2020, with no significant, but gradual compensation afterwards. Further studies are needed to elucidate the underlying causes of these trends. Supporting information S1 Table. Cardiac implantable electronic device implantation centres in Croatia. (PDF) S1 Fig. Total numbers of CIEDs implantations. (DOCX) S2 Fig. Pacemaker implantations monthly. (DOCX) S3 Fig. ICD implantations monthly. (DOCX) PLOS ONE \| https://doi.org/10.1371/journal.pone.0284699 April 26, 2023 6 / 8 PLOS ONECOVID-19 impact on CIED implantation Author Contributions Conceptualization: Ivan Zeljkovic, Jerko Ferri. Data curation: Ivan Zeljkovic, Sandro Brusich. Formal analysis: Ivan Zeljkovic, Pavao Mioc, Sandro Brusich, Richard Matasic, Zrinka Jurisic, Janko Szavits Nossan. Investigation: Pavao Mioc. Methodology: Ivan Zeljkovic, Pavao Mioc, Sandro Brusich, Richard Matasic, Zrinka Jurisic, Katica Cvitkusic Lukenda. Supervision: Ivan Zeljkovic, Sandro Brusich, Richard Matasic, Zrinka Jurisic, Janko Szavits Nossan, Lana Maricic, Katica Cvitkusic Lukenda, Vjekoslav Radeljic, Nikola Pavlovic, Sime Manola. Validation: Richard Matasic, Janko Szavits Nossan, Lana Maricic, Katica Cvitkusic Lukenda, Vjekoslav Radeljic, Nikola Pavlovic, Sime Manola. Writing – original draft: Ivan Zeljkovic, Jerko Ferri, Pavao Mioc. Writing – review & editing: Ivan Zeljkovic, Jerko Ferri, Sime Manola. References 1. Europe: COVID-19 cases by day. Available from: https://www.statista.com/statistics/1102209/ coronavirus-cases-development-europe/ 2. WHO Director-General’s opening remarks at the media. Available from: https://scholar.google.com/ scholar?q=WHO+Director-General%E2%80%99s+opening+remarks+at+the+media+briefing+on +COVID19+-March+2020+ 3. Garcia S, Albaghdadi MS, Meraj PM, Schmidt C, Garberich R, Jaffer FA, et al. Reduction in ST-Seg- ment Elevation Cardiac Catheterization Laboratory Activations in the United States During COVID-19 Pandemic. J Am Coll Cardiol. 2020; 75(22):2871. https://doi.org/10.1016/j.jacc.2020.04.011 PMID: 32283124 4. Rodrı´guez-Leor O, Cid-A´ lvarez B, Ojeda S, Martı´n-Moreiras J, Rumoroso JR, Lo´pez-Palop R, et al. Impact of the COVID-19 pandemic on interventional cardiology activity in Spain. REC Interv Cardiol. 2020; 2(2):82–9. 5. Pessoa-Amorim G, Camm CF, Gajendragadkar P, Maria GL de, Arsac C, Laroche C, et al. Admission of patients with STEMI since the outbreak of the COVID-19 pandemic: a survey by the European Soci- ety of Cardiology. Eur Heart J Qual Care Clin Outcomes. 2020; 6(3):210–6. https://doi.org/10.1093/ ehjqcco/qcaa046 PMID: 32467968 6. Rosa S de, Spaccarotella C, Basso C, Calabrò MP, Curcio A, Filardi PP, et al. Reduction of hospitaliza- tions for myocardial infarction in Italy in the COVID-19 era. Eur Heart J. 2020; 41(22):2083–8. https:// doi.org/10.1093/eurheartj/ehaa409 PMID: 32412631 7. Boriani G, Palmisano P, Guerra F, Bertini M, Zanotto G, Lavalle C, et al. Impact of COVID-19 pandemic on the clinical activities related to arrhythmias and electrophysiology in Italy: results of a survey pro- moted by AIAC (Italian Association of Arrhythmology and Cardiac Pacing). Intern Emerg Med. 2020; 15 (8):1445–56 https://doi.org/10.1007/s11739-020-02487-w PMID: 32889687 8. Arbelo E, Angera I, Trucco E, Rivas-Ga´ndara N, Guerra JM, Bisbal F, et al. Reduction in new cardiac electronic device implantations in Catalonia during COVID-19. Europace. 2021; 23(3):456–63. https:// doi.org/10.1093/europace/euab011 PMID: 33595062 9. Gonzales-Luna AC, Torres-Valencia JO, Alarco´ n-Santos JE, Segura-Saldaña PA. Impact of COVID-19 on pacemaker implant. J Arrhythm. 2020; 36(5):845–8. https://doi.org/10.1002/joa3.12411 PMID: 32837668 10. Migliore F, Zorzi A, Gregori D, del Monte A, Falzone PV, Verlato R, et al. Urgent pacemaker implanta- tion rates in the veneto region of Italy after the covid-19 outbreak. Circ Arrhythm Electrophysiol. 2020; 13:576–8. https://doi.org/10.1161/CIRCEP.120.008722 PMID: 32434373 11. CEZIH. Available from: http://www.cezih.hr/index.html PLOS ONE \| https://doi.org/10.1371/journal.pone.0284699 April 26, 2023 7 / 8 PLOS ONECOVID-19 impact on CIED implantation 12. COVID-19 tjedno izvjesˇće. Available from: https://www.koronavirus.hr/covid-19-tjedno-izvjesce-17- kolovoza/765 13. de Almeida Fernandes D, Cadete R, Anto´nio N, Ventura M, Cristo´vão J, Elvas L et al. Impact of the COVID-19 lockdown in urgent pacemaker implantations: A cross-sectional study. J Arrhythm. 2021; 38 (1):137–44. https://doi.org/10.1002/joa3.12658 PMID: 35222760 14. Colivicchi F, Di Fusco SA, Magnanti M, Cipriani M, Imperoli G. The Impact of the Coronavirus Disease- 2019 Pandemic and Italian Lockdown Measures on Clinical Presentation and Management of Acute Heart Failure. J Card Fail. 2020, 26(6):464–5 https://doi.org/10.1016/j.cardfail.2020.05.007 PMID: 32417376 15. Baldi E, Auricchio A, Klersy C, Burkart R, Benvenuti C, Vanetta C et al. Out-of-hospital cardiac arrests and mortality in Swiss Cantons with high and low COVID-19 incidence: A nationwide analysis. Resusc Plus. 2021; 6:100105. https://doi.org/10.1016/j.resplu.2021.100105 PMID: 34223367 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284699 April 26, 2023 8 / 8 PLOS ONE
10.1371_journal.pone.0267316	RESEARCH ARTICLE Transcriptomic analysis of chloride tolerance in Leptospirillum ferriphilum DSM 14647 adapted to NaCl Javier Rivera-Araya1, Thomas Heine2, Renato Cha´ vez1, Michael Schlo¨ mann2, Gloria Levica´ nID 1* 1 Biology Department, Faculty of Chemistry and Biology, University of Santiago of Chile (USACH), Santiago, Chile, 2 Environmental Microbiology, Institute of Biosciences, TU Bergakademie Freiberg, Freiberg, Germany * gloria.levican@usach.cl Abstract Chloride ions are toxic for most acidophilic microorganisms. In this study, the chloride toler- ance mechanisms in the acidophilic iron-oxidizing bacterium Leptospirillum ferriphilum DSM 14647 adapted to 180 mM NaCl were investigated by a transcriptomic approach. Results showed that 99 genes were differentially expressed in the adapted versus the non-adapted cultures, of which 69 and 30 were significantly up-regulated or down-regulated, respectively. Genes that were up-regulated include carbonic anhydrase, cytochrome c oxidase (ccoN) and sulfide:quinone reductase (sqr), likely involved in intracellular pH regulation. Towards the same end, the cation/proton antiporter CzcA (czcA) was down-regulated. Adapted cells showed a higher oxygen consumption rate (2.2 x 10−9 ppm O2 s-1cell-1) than non-adapted cells (1.2 x 10−9 ppm O2 s-1cell-1). Genes coding for the antioxidants flavohemoprotein and cytochrome c peroxidase were also up-regulated. Measurements of the intracellular reactive oxygen species (ROS) level revealed that adapted cells had a lower level than non-adapted cells, suggesting that detoxification of ROS could be an important strategy to withstand NaCl. In addition, data analysis revealed the up-regulation of genes for Fe-S cluster biosyn- thesis (iscR), metal reduction (merA) and activation of a cellular response mediated by dif- fusible signal factors (DSFs) and the second messenger c-di-GMP. Several genes related to the synthesis of lipopolysaccharide and peptidoglycan were consistently down-regulated. Unexpectedly, the genes ectB, ectC and ectD involved in the biosynthesis of the compatible solutes (hydroxy)ectoine were also down-regulated. In line with these findings, although hydroxyectoine reached 20 nmol mg-1 of wet biomass in non-adapted cells, it was not detected in L. ferriphilum adapted to NaCl, suggesting that this canonical osmotic stress response was dispensable for salt adaptation. Differentially expressed transcripts and experimental validations suggest that adaptation to chloride in acidophilic microorganisms involves a multifactorial response that is different from the response in other bacteria studied. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Rivera-Araya J, Heine T, Cha´vez R, Schlo¨mann M, Levica´n G (2022) Transcriptomic analysis of chloride tolerance in Leptospirillum ferriphilum DSM 14647 adapted to NaCl. PLoS ONE 17(4): e0267316. https://doi.org/10.1371/ journal.pone.0267316 Editor: Benjamin J. Koestler, Western Michigan University, UNITED STATES Received: May 21, 2021 Accepted: April 6, 2022 Published: April 29, 2022 Copyright: © 2022 Rivera-Araya et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Data were submitted to European Bioinformatic Institute, EBI (https:// www.ebi.ac.uk/). The accession was assigned as ArrayExpress accession E-MTAB-11136. Additionally, this information is available as a footnote in Tables 2 and 3 of the manuscript. Funding: This work was funded by Fondo Nacional de Desarrollo Cientı´fico y Tecnolo´gico, FONDECYT, from the government of Chile (Grant N˚ 1170799/ 1211386). JR-A and RC received support from DICYT_USACH (Proyecto POSTDOC_DICYT, COD PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 1 / 18 PLOS ONE022043CR_POSTDOC). MS thanks Dr. Erich Kru¨ger foundation for generous support as part of the Biohydrometallurgical Center Freiberg. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Transcriptomic of Na-Cl adapted Leptospirillum ferriphilum Introduction Leptospirillum ferriphilum is a Gram-negative, and obligately aerobic iron-oxidizing chemoau- totroph able to thrive in a pH range from 1.3 to 2.0 [1]. This bacterium belongs to the bioleach- ing microbial communities involved in solubilization of metals from sulfide ores [2,3]. L. ferriphilum, like most acidophilic microorganisms, shows extreme sensitivity to chloride and other anions (with the notable exception of sulfate) [4]. Acidophiles possess positive membrane potentials which facilitates the influx of permeable anions into cells [5]. The mass entry of chloride and other anions favor the influx of protons causing the collapse of positive internal potential, and therefore the disruption of the proton motive force, as well as acidification of the cytoplasm and a general detrimental effect in the cell [4,6,7].Nevertheless, the molecular basis of the response to chloride in L. ferriphilum and other acidophilic microorganisms are poorly understood. The mechanisms operating in acidophiles in response to chloride have been investigated just during recent years. In L. ferriphilum, the osmotic stress induced by sodium chloride leads to the up-regulation of genes encoding the potassium transporter Kdp, and for the biosynthesis or uptake of compatible solutes such as (hydroxy)ectoine and trehalose [8–11]. In members of the genus Acidithiobacillus, like At. ferrooxidans and At. caldus, the use of proline and betaine as osmoprotec- tants has been reported [12,13], whilst the moderately halotolerant Acidihalobacter prosperus has a response based on the synthesis and uptake of ectoine [14]. In addition, A. prosperus also seems to have developed a more specific adaptive response that involves changes in the amino acid compo- sition of rusticyanin to protect the copper ion present in the active site of this protein [14]. To respond to cytoplasm acidification induced by chloride exposure, acidophiles synthesize a greater number and diversity of cation/H+ antiporters, proteins that modify the cell mem- brane, and proteins of the electron-transport chain. These changes result in the presumed export of protons, at the expense of increasing the respiratory rate [1,4,14,15]. Recently, Rivera-Araya et al. [4] described that exposure of L. ferriphilum to chloride led to a significant increase in intracellular reactive oxygen species (ROS). It is believed that ROS enhancement is produced by the increment in respiratory activity and by disruption of metallic centers of pro- teins due to osmotic imbalance. In addition, Fe2+ and other cations can trigger Fenton chemis- try and induce the generation of hydroxyl radicals [16]. In agreement with these observations, the activation of antioxidant mechanisms seems to play an important complementary role in the response to chloride. The exposure of L. ferriphilum to 50–150 mM NaCl has been shown to up-regulate the activity of thioredoxin and cytochrome c peroxidase [4]. Similarly, in other microorganisms, like At. caldus and Acidimicrobium ferrooxidans, the up-regulation of antiox- idative proteins in response to NaCl has also been reported [7,13]. Therefore, based on the evidence from the individual studies described above, it is possible to state that in L. ferriphilum and other acidophilic microorganisms, the exposure to chloride trig- gers a response that involves the participation of different mechanisms to withstand osmotic, acid and oxidative stress. However, it is envisioned that a chloride challenge activates a global and complex physiological response that has yet to be well deciphered. In the present study, we report on transcriptomic analyses conducted in L. ferriphilum DSM 14647 adapted and exposed to 180 mM NaCl. This study also included the measurements of specific parameters such as oxygen con- sumption rate, intracellular pH, and ROS and (hydroxy)ectoine content. Materials and methods Bacterial strains and growth conditions L. ferriphilum DSM 14647 [17] used in this study was provided by Leibniz Institute DSMZ. The bacterial cells were cultured in DSMZ 882 medium (pH 1.8) supplemented with 72 mM PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 2 / 18 PLOS ONETranscriptomic of Na-Cl adapted Leptospirillum ferriphilum ferrous sulfate (FeSO4�7H2O). Bacterial growth was carried out in Erlenmeyer flasks at 180 rpm and 37˚C. Adaptation of L. ferriphilum DSM 14647 to 180 mM NaCl The adaptation of L. ferriphilum DSM 14647 was performed in growth medium (see above) with increasing NaCl concentrations (50-100-120-150-180 mM) and supplementation with 1 mM ectoine as a compatible solute (Sigma-Aldrich). The adaptation was performed sequen- tially and with 3 passages per salt concentration. Cultures were maintained until the late expo- nential phase and used to inoculate fresh NaCl and ectoine-containing medium (10% v/v) and generate a new culture. After the 180 mM NaCl-adapted culture had been obtained, the com- patible solute was gradually (1–0.5–0 mM) removed from the medium. Adapted cells were constantly grown in the presence of 180 mM NaCl. Growth curve determination The experiment was carried out in 250 mL Erlenmeyer flasks. Each flask contained 100 mL DSMZ 882 medium with 0 or 180 mM NaCl for non-adapted and adapted cells, respectively. Samples were taken periodically for determination of cell growth, which was measured by direct microscopic counting using a modified Neubauer chamber. The initial cell density was 1 x 106 cells mL-1. Measurement of minimum inhibitory concentrations (MIC) of NaCl This assay was carried out on planktonic cells according to Rivera-Araya et al. [11] with some modifications. Briefly, to determine the MIC of NaCl, non-adapted and adapted cells of L. fer- riphilum were cultured in DSMZ 882 medium at pH 1.4, 1.8, 2.4 or 3.0 in the presence of dif- ferent NaCl concentrations, ranging from 0 to 600 mM. The experiments were performed in triplicate in 6-well plates, each well containing 5 mL of the medium. Bacteria were inoculated to a concentration of 1 x 106 cells mL-1 and later incubated at 37˚C for 72–86 h, until the con- trol sample (without salt) reached the stationary phase. The MIC value corresponds to the minimal NaCl concentration where no bacterial growth was observed. mRNA isolation and transcriptomic analysis mRNA isolation. Cells from control (non-adapted, non-exposed to NaCl), and adapted in 180 mM NaCl conditions were grown until the late exponential phase. Cells were harvested by centrifugation at 8,000 x g for 15 min (at 4˚C) and washed once with cold 10 mM H2SO4 and twice with 10 mM sodium citrate pH 7.0. Total RNA was isolated using the RNeasy Mini Kit (Qiagen). DNA was removed by DNase I treatment (New England, Biolabs) according to the manufacturer’s instructions. cDNA library preparation and Illumina sequencing. The quality and integrity of the total RNA were evaluated using an Agilent Bioanalyzer 2100 and an RNA 600 Nano Kit (Agi- lent Technologies). Three RNA preparations of high quality (RNA integrity number above 7) were pooled together and submitted for transcriptome analysis as previously described [18,19]. Before library preparation, ribosomal RNA was depleted using the MICROBExpress kit (Thermo Fisher). Then, a TruSeq stranded mRNA library prep kit (Illumina) was used to generate cDNA libraries for whole transcriptome analysis. The resulting libraries were sequenced on an Illumina MiSeq system with v3 chemistry and 2 x 75-nucleotide read lengths (paired end). PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 3 / 18 PLOS ONETranscriptomic of Na-Cl adapted Leptospirillum ferriphilum Differential expression analysis. Raw reads from RNA sequencing of non-adapted and adapted cells were processed to remove adaptors, and filtered to obtain reads with a quality higher than Q20, by using the CLC Genomics Workbench software. Then, the filtered reads were aligned onto the reference genome of L. ferriphilum DSM 14647 [NCBI accession num- ber: PGK00000000] by CLC Genomics Workbench software. Transcriptomic data was submit- ted to European Bioinformatic Institute database (ArrayExpress Accession: E-MTAB-11136) Raw counts for each ORFs features were subjected to differential analysis with statistical R software, using the DESeq2 package [20]. A gene was considered differentially expressed with a p-value < 0.05. The assignment of genes to a functional category was carried using the Go Feat Tool and the public Gene Ontology (GO) database [21]. Oxygen consumption The oxygen consumption rate was determined by means of optodes (Fibox 3, PreSens-Preci- sion Sensing GmbH, Regensburg, Germany) [22]. In short, fresh iron-grown 100 mL-cultures of L. ferriphilum DSM 14647 were harvested by centrifugation at 8,000 x g for 15 min, the supernatant was removed, and the pelleted cells were resuspended in 0.1 mL the remaining growth medium, before being added to a 3-mL cuvette containing 2.6 mL of DSM 882 culture medium pH 1.8 with 0 or 180 mM NaCl for non-adapted and adapted cell cultures, respec- tively. Afterwards, 0.15 mL of ferrous iron solution were added to the cuvettes (final concen- tration 72 mM), and the suspension mixed cautiously. The cuvette was then carefully closed with a glass lid. An oxygen-sensing optode spot had previously been embedded inside the mea- suring cuvette. Fibre-optics located outside the cuvette on the opposite side of the oxygen sen- sor spot were connected with a 4-channel fiber–optics oxygen meter (Firesting O2), also equipped with a receptacle for a temperature sensor. The optode signal was evaluated using the software Pyro Oxygen Logger. Due to the strong temperature dependence of fluorescence, measurements were performed in a thermostatic cabinet (UVP Hybridizer HB-1000) at 37˚C. Optode measurements were performed in triplicate using biological replicates. Analysis of intracellular (hydroxy)ectoine content The compatible solutes ectoine and hydroxyectoine were quantified by HPLC analysis, using an Ultimate 3000–2015 HPLC (Thermo Scientific) system with a 250 mm × 4.6 mm Hypurity Aquastar C-18 column with particle size of 5 μm (Thermo Scientific), as described previously [4]. Chromatography was performed with a gradient of two solutions as mobile phase—eluent A (0.8 mM KH2PO4, 6.0 mM Na2HPO4, pH 7.6) and eluent B (acetonitrile)—at a flow rate of 1.0 mL min-1 at 25˚C. The presence of compatible solutes was monitored at 215 nm by a UV/ VIS detector. The retention times of ectoine and hydroxyectoine were determined using com- mercially available compounds (purity � 95%, Sigma-Aldrich) as standards. Intracellular ectoine and hydroxyectoine content was calculated as ng mg-1 of wet biomass, using a calibra- tion curve. Determination of ROS levels The intracellular level of total ROS was measured in non-adapted and adapted cultures using the fluorescent probe 2’,7’-dichlorodihydrofluorescein diacetate (H2DCFDA) according to Ferrer et al. [23]. Fluorescent emission values were normalized to the total protein concentra- tion. Protein quantification was performed by the colorimetric Bradford assay [24]. Since ROS determination included a last incubation step with the fluorescent probe under neutral pH conditions, the viability of the cell cultures was tested. For this purpose, a control was PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 4 / 18 PLOS ONETranscriptomic of Na-Cl adapted Leptospirillum ferriphilum performed by incubating cells in 100 mM potassium phosphate buffer pH 7.4 without the probe and then re-inoculating them into fresh medium as described [4]. Statistical analysis Statistical analysis was performed using the one-way ANOVA test followed by Tukey’s test in GraphPad Prism 5. The differences were considered to be significant at p < 0.05. Results and discussion Characterization of growth and NaCl-tolerance of L. ferriphilum DSM 14647 adapted to 180 mM NaCl The adaptation of L. ferriphilum to 180 mM NaCl led eventually to a culture with the same cell density (8 x 107 cells mL-1) as the non-adapted cell culture (Fig 1). However, salt approxi- mately tripled the time of cellular duplication (td) from 6 to 17 h. A retarding effect on growth rate and iron oxidation has been observed in different studies of NaCl-susceptible acidophilic microorganisms, including L. ferriphilum and other species (At. ferrooxidans and S. thermosul- fidooxidans) [14,25]. It is important to highlight that although the addition of ectoine favored the sequential acclimation of L. ferriphilum to 180 mM NaCl (data not shown), the adapted cell culture could grow steadily without ectoine supplementation, indicating that cells were physiologically adapted to face this stress condition. It has been widely reported that decreasing the external pH contributes significantly to the toxicity of chloride in this species and in other acidophilic bacteria [4,15]. In agreement with this, Fe2+ oxidation in the presence of NaCl is highly influenced by the pH of the growth medium [5]. Thus, in order to evaluate the tolerance of NaCl-adapted cells, we determined the MIC of adapted and non-adapted cell cultures exposed to a range of pH values. As shown in Table 1, the MIC of the adapted culture was higher than that of the non-adapted culture. In addition, the MIC significantly increased as the pH of the medium increased within the range of 1.4–3.0. However, it was also observed that at a higher pH of the medium, the difference of Fig 1. Growth of L. ferriphilum DSM 14647. Curves represent the growth of adapted and exposed to 180 mM NaCl (grey circles), and non-adapted and non-exposed (white circles) cell cultures. Data represent the average of 3 independent experiments. Error bars represent standard deviation. Initial cell density, 1x106 cells mL-1. https://doi.org/10.1371/journal.pone.0267316.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 5 / 18 PLOS ONETranscriptomic of Na-Cl adapted Leptospirillum ferriphilum Table 1. Minimum inhibitory concentration of NaCl in L. ferriphilum DSM 14647 adapted to 180 mM NaCl at dif- ferent external pH (pHex). pHex 1.4 1.8 2.4 3.0 NaCl MIC [mM] L. f. non-adapted 175 L. f. adapted to 180 mM NaCl 350 225 350 400 375 425 500 https://doi.org/10.1371/journal.pone.0267316.t001 the MIC between adapted and non-adapted cultures was lower. For example, at pH 1.4 the adapted culture had a MIC value twice (350 mM) that of the non-adapted culture (175 mM), while at pH 3.0, the MIC of the adapted culture was just 25% higher (500 mM) than that of the non-adapted culture (400 mM). The results clearly show that the prior adaptation of L. ferri- philum conferred a higher tolerance against NaCl, but that this tolerance was more noticeable at a low external pH. Transcriptomic profile of L. ferriphilum DSM 14647 adapted to 180 mM NaCl Screening of differentially expressed genes (DEG). The differential expression analysis was performed comparing cultures adapted to 180 mM NaCl versus non-adapted non-exposed cell cultures as described in Materials and Methods. In this analysis, 99 out of 2736 genes showed significant differential expression (p<0.05) of which 69 (2.5%) and 30 (1.1%) were up- regulated and down-regulated, respectively. Table 2 lists the genes that were up-regulated (excluding 43 ORFs predicted as hypothetical proteins, S1 Table) in a range of 4.1- to 91.7– fold change. Table 3 shows the down-regulated genes (excluding 8 ORFs predicted as hypo- thetical proteins, S2 Table) in a range of -4.3 to -9.3–fold change. Classification of genes by their functionality revealed a number of genes involved in metabolism and energy conserva- tion, the cell envelope, transport and osmoregulation, and stress response and signal transduc- tion, among others. -) and protons in the reaction: CO2 + H2O , Metabolism and energy conservation. The adaptation of L. ferriphilum to 180 mM NaCl resulted in the identification of a number of DEGs related to metabolism and energy conserva- tion. A significant increase in the expression of a carbonic anhydrase (CA, 8.1-fold) was observed in the adapted culture. This metalloenzyme catalyzes the reversible hydration of car- bon dioxide to form bicarbonate ions (HCO3 HCO3− + H+ [26]. In autotrophic bacteria that fix CO2 through the Calvin-Benson-Bassham cycle, CA is involved in the transport and supply of CO2 to Rubisco (D-ribulose 1,5-bispho- sphate carboxylase/oxygenase) in the carboxysome [27]. However, since this enzyme produces and uses protons and bicarbonate ions, it also plays a key role in the regulation of pH [28]. In acidophiles, genes encoding CA and the carboxysomal shell proteins have been described in At. ferrooxidans and At. thiooxidans [29–31]. Moreover, in At. ferrooxidans, the expression of the cbb5 operon that encodes the inorganic carbon transporter SulP and CA is dependent on the CO2 concentration regimen [31]. In L. ferriphilum and other leptospirilli, carbon fixation is performed by the reductive tricarboxylic acid cycle (RTCA) [29] in which, as far as it is known from the literature, CA does not seem to play a role. Thus, the predicted CA of L. ferri- philum could play a major role by contributing towards neutralizing the acidification of the cytoplasm that is expected to occur in the presence of chloride. In this way, the up-regulation of the CA-encoding gene could represent a direct strategy of cellular pH homeostasis. The con- tribution of CA to this purpose deserves to be experimentally addressed. PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 6 / 18 PLOS ONETable 2. Up-regulated genes in L. ferriphilum DSM 14647 adapted to 180 mM NaCl in relation to non-adapted non-exposed control cells. Accession number Gene product Metabolism and energy conservation Fold change Transcriptomic of Na-Cl adapted Leptospirillum ferriphilum KGA94808.1 WP_036082816.1 KGA94222.1 Cell envelope WP_014961534.1 WP_081938081.1 Transport and osmoregulation WP_036082891.1 Stress response WP_036083168.1 WP_052157908.1 WP_023524701.1 KGA93200.1 WP_036079670.1 KGA93006.1 KGA94361.1 Signal transduction WP_049713715.1 WP_036081469.1 WP_161781719.1 WP_036081415.1 Others KGA93962.1 WP_036081724.1 KGA94115.1 WP_036083266.1 WP_161781749.1 WP_036081132.1 WP_036080943.1 WP_020859441.1 WP_036082283.1 Carbonic anhydrase (CA) Cytochrome c oxidase subunit CcoN Sulfide:quinone reductase (Sqr) Regulator of protease activity HflC Fatty acid desaturase Outer membrane efflux protein TolC Flavohemoprotein Cytochrome c peroxidase Heat-shock protein Hsp20 Transcriptional Regulator IscR FAD-dependent pyridine nucleotide-disulfide oxidoreductase Radical SAM domain protein Mercuric reductase, MerA Diguanylate cyclase/phosphodiesterase DSF synthase (RpfF) Diguanylate cyclase/phosphodiesterase Diguanylate cyclase/phosphodiesterase Transposase Phage related integrase Methyl-accepting chemotaxis protein Methyl-accepting chemotaxis protein Periplasmic serine protease DO (HtrA) Flagellin protein FlaB DNA-binding protein HU Prokaryotic ubiquitin-like protein Pup Shufflon-specific DNA recombinase a: Transcriptomic data can be found in EBI database (https://www.ebi.ac.uk/) as indicated in Materials and Methods. b: Values correspond to the average fold change of 3 biological replicates. https://doi.org/10.1371/journal.pone.0267316.t002 8.1 4.9 4.6 6.6 4.9 7.4 14.3 10.9 8.1 7.2 5.6 5.2 4.3 11.2 10.7 6.6 5.2 7.7 5.4 5.4 5.1 4.9 4.4 4.3 4.2 4.2 Genes coding for proteins from electron-transport chains such as cytochrome c oxidase CcoN subunit (4.9-fold) and sulfide:quinone reductase Sqr (4.6-fold) were also significantly up-regulated. CcoN is the component of the cbb3-type cytochrome oxidase, a complex enzyme of the respiratory chain which has previously been reported in Leptospirillum spp. [32]. CcoN is the catalytic subunit of the enzyme in charge of the four-electron reduction of molecular oxygen to water, a process which is coupled to translocation of protons across the membrane [33]. The Sqr enzyme could play a role in the detoxification of endogenously generated H2S, a common product of cysteine metabolism that negatively impacts the redox status of bacterial cells [34,35]. The enzyme obtains electrons from H2S oxidation and transfers them to the qui- none pool, thus increasing the activity of the electron-transfer chain. The increase of both cyto- chrome c oxidase and Sqr activities should increase the respiratory rate of this bacterium to PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 7 / 18 PLOS ONETable 3. Down-regulated genes in L. ferriphilum DSM 14647 adapted to 180 mM NaCl in relation to non-adapted non-exposed control cells. Accession numbera Metabolism and cell envelope Gene product Fold change Transcriptomic of Na-Cl adapted Leptospirillum ferriphilum WP_036081541.1 WP_036081614.1 WP_036081511.1 WP_036081618.1 WP_036081521.1 WP_036081553.1 WP_036081550.1 WP_036081600.1 WP_052157773.1 WP_036081557.1 WP_036081546.1 WP_036081519.1 Transport and osmoregulation WP_036080892.1 WP_036080895.1 WP_036080909.1 WP_036081492.1 WP_020859429.1 WP_020859430.1 WP_020859431.1 Stress response WP_036080906.1 WP_036080898.1 WP_052157774.1 Glycosyl transferase, group 1 family protein Glutamine-fructose-6-phosphate aminotransferase Glycosyl transferase family 2 protein UDP-glucose dehydrogenase UTP-glucose-1-phosphate uridylyltransferase Undecaprenyl-phosphate galactose phosphotransferase Polysaccharide export protein dTDP-glucose 4,6-dehydratase Glycosyltransferase involved in cell wall biosynthesis Tyrosine-protein kinase EpsD Polysaccharide deacetylase Eight transmembrane protein EpsH Outer membrane efflux protein TolC RND efflux transporter RND family efflux transporter MFP subunit ABC transporter ATP-binding protein MdlB Diaminobutyrate-2-oxoglutarate transaminase (EctB) L-ectoine synthase (EctC) Ectoine hydroxylase (EctD) Cobalt-zinc-cadmium resistance protein CzcA Two component sigma54 specific transcriptional regulator Sigma-54 dependent transcriptional regulator a: Transcriptomic data can be found in EBI database (https://www.ebi.ac.uk/) as indicated in Materials and Methods. b: Values correspond to the average fold change average of 3 biological replicates. https://doi.org/10.1371/journal.pone.0267316.t003 -5.3 -5.7 -5.3 -6.4 -9.0 -4.8 -4.9 -5.0 -6.0 -6.9 -8.4 -9.3 -5.1 -5.4 -6.2 -5.9 -6.5 -5.6 -8.0 -6.8 -4.3 -10.2 provide energy (ATP), reducing power (NAD(P)H), and mainly the possibility of extruding protons from the cytoplasm to avoid acidification induced by chloride exposure [4,14]. In order to evaluate whether adapted cells showed a higher respiratory rate, the oxygen consump- tion of non-adapted and adapted cells of L. ferriphilum exposed to 180 mM NaCl was mea- sured. As shown in Fig 2, non-adapted L. ferriphilum exposed to 180 mM NaCl was not able to respire. Interestingly, the O2 consumption rate in adapted cell cultures treated with 180 mM NaCl was significantly greater than that of non-adapted untreated cells (1.2 x 10−9 versus 2.2 x 10−9 ppm O2 s-1cell-1; p<0.01). This result supports the idea that up-regulation of electron- transport chain genes contributes towards the increase in the oxygen respiratory activity in adapted cells exposed to NaCl. A similar effect was observed in a proteomic study of Ac. pros- perus in which cytochrome c1, rusticyanin and ATP synthase subunit b were over-expressed in the presence of 500 mM NaCl [14], indicating that proton extrusion by respiration may be a widely distributed chloride response mechanism in acidophiles. Cell envelope. One of the up-regulated genes codes for the regulator HflC (6.6-fold) which modulates the FtsH protease and may serve to maintain quality control of some mem- brane proteins [36,37]. Additionally, a gene coding for a fatty acid desaturase, which belongs to a group of enzymes in charge of double-bond insertion at specified positions of fatty acyl chains, necessary for membrane-lipid fluidity [38], was up-regulated (4.9-fold). In PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 8 / 18 PLOS ONETranscriptomic of Na-Cl adapted Leptospirillum ferriphilum Fig 2. Effect of NaCl adaptation on oxygen consumption rate in L. ferriphilum DSM 14647. The measurements were carried out in adapted (A) cells exposed to 180 mM NaCl and non-adapted (NA) cells exposed to 0 or 180 mM NaCl. The data represent the average of 3 independent experiments. Error bars represent standard deviation. Statistical analysis was carried out by an ANOVA test. https://doi.org/10.1371/journal.pone.0267316.g002 Synechocystis, strains overexpressing a desaturase gene were found to be more robust under salt stress conditions [39]. In addition, a correlation between the unsaturation of fatty acids in membrane lipids and tolerance to salt stress in this genus and other bacteria has been reported [39,40]. For L. ferriphilum, the up-regulation of the fatty acid desaturase gene suggests an increase in the unsaturated/saturated fatty acid ratio. Whether this confers higher fluidity to the membrane in salt stress compared to normal conditions, or it is to compensate a salt- induced decrease in the fluidity and thus ensure a fluid membrane at high salt, remains to be determined. Several genes involved in carbohydrate metabolism had lower expression in adapted versus non-adapted cells. Among them were genes encoding two glycosyl transferases (-5.3 fold), a UDP-glucose dehydrogenase (-6.4 fold) and a UTP-glucose-1-phosphate uridylyltransferase (-9.0 fold) which are directly related to the synthesis of glycosaminoglycans, critical precursors of peptidoglycans and other cell-surface polymers, such as lipopolysaccharides [41–43]. Another significantly repressed gene under high-salt conditions was glutamine-fructose- 6-phosphate aminotransferase/glucosamine-6-phosphate synthase (-5.7 fold), a dimeric enzyme that catalyzes the first step in hexosamin metabolism, converting D-fructose-6-phos- phate (Fru6P) and glutamine (Gln) into D-glucosamine-6-phosphate (GlcN6P) and glutamate [44]. The end product of the hexosamine pathway, uridine diphosphate N-acetylglucosamine (UDP-GlcNAc), plays an important role as a precursor of peptidoglycan and glycolipids [45]. Other genes related with the biosynthesis of the cell envelope that were down-regulated in L. ferriphilum grown at 180 mM NaCl encode undecaprenyl-phosphate galactose phospho- transferase (-4.8 fold) and dTDP-glucose 4,6-dehydratase (-5.0 fold), two enzymes involved in the generation of intermediate nucleotide sugars for O-antigen polysaccharide biosynthesis in the biogenesis of the outer membrane [46,47]. Altogether, these findings suggest that synthesis of cell surface polymers such as peptidoglycan and lipopolysaccharide were diminished as a result of the physiological salt adaptation in L. ferriphilum. Abiotic stressors jeopardize the PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 9 / 18 PLOS ONETranscriptomic of Na-Cl adapted Leptospirillum ferriphilum integrity of peptidoglycans and other components of the cell envelope by introducing lesions, which must be rapidly repaired to prevent cell lysis [48]. As a consequence, upon osmotic stress induction, cells respond by upregulating the activity of enzymes or genes essential for cell wall synthesis [49]. Thus, based on these antecedents, we envisioned that adapted cells of L. ferriphilum exposed to 180 mM NaCl did not generate the corresponding response to osmotic stress. Transport and osmoregulation. The adaptation of L. ferriphilum to NaCl also resulted in the up-regulation of several genes encoding proteins related to cellular transport. These included the gene encoding TolC protein (7.4-fold), a key component of multidrug efflux sys- tems such as AcrAB-TolC, AcrEF-TolC, EmrAB-TolC and MacAB-TolC of the outer mem- brane, which are important for bacterial survival and oxidative stress responses in acidic environments [50,51]. Conversely, genes encoding several transporters were repressed. In agreement with decreas- ing the biosynthesis of surface polymers, the expression of a gene encoding a polysaccharide- transport protein implicated in the export of polysaccharides across the outer membrane [52] was significantly lower in salt-adapted cells (-4.9-fold). Genes encoding an outer-membrane efflux protein TolC (different from the one referred to above; -5.1-fold), two genes coding for RND (Resistance-Nodulation-Division) efflux transporters (-5.4 and -6.2-fold, respectively) that form complexes with AcrAB-TolC, and play a role in the active efflux of antimicrobial agents [53], and ABC transporter ATP-binding protein MdlB (-5.9 fold), which is an integral membrane protein named Mdl (Multidrug resistance-like) that actively transports molecules across the lipid membrane against a concentration gradient, were also reduced in expression [54,55]. Regarding osmoregulation, it was unexpected that 3 genes involved in the biosynthesis of (hydroxy)ectoine-diaminobutyrate-2-oxoglutarate transaminase (ectB, -6.5-fold), L-ectoine synthase (ectC, -5.6-fold) and ectoine hydroxylase (ectD, -8.0-fold)—were all significantly down- regulated. Since hydroxyectoine plays an important role in protecting the cells of L. ferriphilum against saline stress [4], we were interested in evaluating the intracellular content of ectoine and hydroxyectoine in adapted cells exposed to 180 mM NaCl. As shown in Fig 3, ectoine was not detected in either adapted or non-adapted cells. However, hydroxyectoine reached 20 nmol mg-1 of wet biomass (p<0.01) in non-adapted cells cultured without NaCl while it was not detected in extracts of L. ferriphilum adapted to 180 mM NaCl. Taken together, these results reinforce the idea that the 180 mM NaCl-adapted culture of L. ferriphilum does not develop an active response to osmotic stress based on the synthesis of compatible solutes. Interestingly, in non-adapted cells, the compatible solute-mediated response appears to be functionating, and in this way these cells could be actively responding to the osmotic challenge. Stress response. Presumed stress response genes that exhibited a significant increase in their transcript levels encoded the following proteins: a flavohemoprotein (14.3-fold), an enzyme able to reduce nitric oxide (NO) from reactive nitrogen species (RNS) [56]; a cyto- chrome c peroxidase (10.9-fold) able to reduce periplasmic hydrogen peroxide [57]; an FAD- dependent pyridine nucleotide-disulfide oxidoreductase (5.6-fold) which catalyzes disulfide bond formation and reduction [58,59]; and a radical S-adenosyl-methionine (SAM, 5.2-fold) precursor for the biosynthesis of the antioxidant cobalamin [23]. These data strongly suggest that in L. ferriphilum, antioxidant proteins form part of the mechanisms that are activated to enable this species to face the stress induced by NaCl, and thereby manage redox homeostasis under these conditions. In agreement with the induction of antioxidative proteins, in a previous study carried out by our research group, it was established that exposure to NaCl induced a severe condition of oxidative stress in L. ferriphilum, leading to an increase in intracellular ROS levels and PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 10 / 18 PLOS ONETranscriptomic of Na-Cl adapted Leptospirillum ferriphilum Fig 3. Effect of NaCl adaptation on the content of compatible solutes in L. ferriphilum DSM 14647. Ectoine and hydroxyectoine content was measured in adapted cells exposed to 180 mM NaCl (A) and non-adapted (NA) cells exposed to 0 or 180 mM NaCl. The data represent the average of 3 independent experiments. Error bars represent standard deviation. Statistical analysis was carried out by ANOVA and a T Test. N.D.: not detected. https://doi.org/10.1371/journal.pone.0267316.g003 activation of the antioxidant response [4]. In order to establish whether the adapted cells are able to maintain the redox balance, the intracellular ROS level was measured using a fluores- cent probe as described in Material and Methods. The measurements were performed in non- adapted and adapted cultures grown in DSMZ 882 medium supplemented (or not) with 180 mM NaCl. As shown in Fig 4, non-adapted cells exposed to 180 mM NaCl had significantly Fig 4. Effect of NaCl on ROS generation in L. ferriphilum. ROS were measured in adapted cells exposed to 180 mM NaCl (A) and non-adapted (NA) cells exposed to 0 or 180 mM NaCl. Cytoplasmic ROS content is expressed as relative fluorescence units (RFU) of the activated fluorescent probe H2DCFDA per mg of protein. The data represent the average of 3 independent experiments. Error bars represent standard deviation. Statistical analysis was carried out by ANOVA and a T Test. https://doi.org/10.1371/journal.pone.0267316.g004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 11 / 18 PLOS ONETranscriptomic of Na-Cl adapted Leptospirillum ferriphilum higher intracellular ROS levels (p<0.01) compared with the control condition (without NaCl). Interestingly, salt-adapted L. ferriphilum treated with 180 mM NaCl showed similar, and even slightly lower, intracellular ROS levels compared to the control without NaCl, suggesting that these cells maintain correct redox homeostasis. This condition is most likely managed through the up-regulation of the antioxidant mechanisms described above. Another gene that showed up-regulation (7.2-fold) encodes the IscR regulator, potentially involved in regulating the biosynthesis of [Fe-S] clusters of proteins [60]. The [Fe-S] clusters are susceptible to being oxidized by superoxide anions, thus releasing Fe2+, and thereby trig- gering Fenton chemistry and the generation of highly harmful hydroxyl radicals [16,23]. Therefore, these results imply that under high salt conditions, [Fe-S] clusters of proteins suffer oxidative damage and, in consequence, the cells respond through the activation of the biosyn- thesis pathway for [Fe-S] clusters. Interestingly, the merA gene that encodes a mercuric reductase was over-expressed (4.3-fold) under the high-NaCl regimen. In bacteria, the mercury-resistance (mer) genes are activated and repressed by the metalloregulatory MerR protein, which has a high degree of selectivity for mercury (Hg2+) but can additionally be partially stimulated by a variety of transi- tion metals such as Cd2+, Zn2+, Ag+, Au+, and Au3+ [61]. For example, in the metal-tolerant bacterium Cupriavidus metallidurans, the genes merA, merT, and merP were up-regulated when this bacterium was exposed to cadmium [62,63]. A similar phenomenon has been described in Nitrosomonas europaea, since the mer operon was also induced by cadmium [64]. Although merR was not up-regulated in our study, this gene was detected in the genome of L. ferriphilum and could contribute to regulate the transcriptional activity of the merA gene in response to mercury or other metals. We speculate that in L. ferriphilum, chloride stress could cause oxidation of metalloproteins releasing oxidized metals to the intracellular space that may activate merA transcription. In L. ferriphilum this response could be relevant, since it has a high content of cytochromes and [Fe-S] proteins [32,65] that could contribute to increasing the intracellular free iron and copper contents under stress conditions. Whether the mer operon has a role in protection and /or avoiding toxicity toward these metals should be elucidated. Interestingly, some genes related to stress responses were repressed. Such is the case for the gene coding for the cobalt-zinc-cadmium resistance protein CzcA (-6.8-fold), one of the three components of the CzcABC efflux pump [66]. This pump functions as a cation-proton anti- porter mediating resistance against divalent metals such as cadmium (Cd2+), zinc (Zn2+), and cobalt (Co2+), among others [67]. As chloride exposure is known to induce cytoplasmic acidifi- cation by favoring entry of protons into the cell, the response to this anion should involve strategies that contribute to keeping the intracellular pH closer to neutrality. Thus, repression of the czcA gene and eventual down-regulation of CzcABC pump activity in adapted L. ferri- philum could participate towards avoiding the entry of protons into the cytoplasm. Signal transduction. Among the genes that were up-regulated in the culture adapted to NaCl, we detected one gene encoding a diffusible signal factor (DSF) synthase (10.7-fold) and 3 genes coding for diguanylate cyclase phosphodiesterase (5.2, 6.6 and 11.2-fold). The protein DSF synthase (RpfF) synthesizes diffusible signal factors, widely conserved quorum sensing signals in many Gram-negative bacterial species that play important roles in regulating various biological functions such as biofilm formation, virulence, and antibiotic and stress resistance [68,69]. RpfF synthesizes DSF by dehydration of a 3-hydroxyacyl-acyl carrier protein (ACP) fatty acid intermediate and also cleaves the thioester bond linking DSF to ACP [70]. When DSFs reach a threshold concentration outside the cell, bacteria activate their cognate receptor RpfC, a hybrid membrane sensor kinase that phosphorylates the intracellular response regula- tor RpfG [70]. The activated RpfG possesses c-di-GMP phosphodiesterase activity, which PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 12 / 18 PLOS ONETranscriptomic of Na-Cl adapted Leptospirillum ferriphilum hydrolyzes c-di-GMP to produce GMP. The change in c-di-GMP level affects the transcrip- tional expression of target genes, thus configuring a physiological response or modulating a biological process [70]. Therefore, based on the up-regulation of genes encoding DSF synthase and diguanylate cyclase phosphodiesterase, it is possible to infer that adaptation of L. ferriphi- lum to NaCl involves the activation of a cellular response mediated by DSF signals and the sec- ond messenger c-di-GMP. However, the target genes that are modulated by this mechanism remain to be elucidated. Others. Other genes up-regulated by NaCl adaptation were two methyl-accepting chemo- taxis proteins (5.4-fold and 5.1-fold) and a flagellin protein FlaB (4.4-fold) which are related with movement of microorganisms in response to chemical gradients, and biosynthesis of fla- gella, respectively [71]. An effect of osmolarity challenges on flagellar function has previously been reported in bacteria. Specifically, in Desulfovibrio vulgaris, cells were observed to be highly motile when subjected to salt stress and several key chemotaxis genes were very highly and reproducibly up-regulated [72]. More recently, Escherichia albertii showed swimming motility when cultured at low osmotic pressure. Under this condition, the biosynthesis of fla- gella was also induced [73]. It has been predicted that flagellar induction increases E. albertii survival in intestinal epithelial cell cultures. Whether motility and flagellum assembly are acti- vated by NaCl exposure, and the corresponding impact of their activation on adaptation and fitness of leptospirilli should be addressed. Another group of genes overexpressed in the NaCl-adapted culture encodes a transposase (7.7-fold), a phage-related integrase (5.4-fold), a DNA-binding protein HU (4.3-fold) and a shufflon-specific DNA recombinase (4.2-fold). All are involved in bacterial DNA transaction systems including transposition and recombination, among others [74]. Therefore, genetic/ genomic modifications could underlie physiological stress responses and/or may pre-adapt a small subset of the population to face this environmental stress. Conclusions Despite its high chloride sensitivity, L. ferriphilum could be stably adapted to 180 mM NaCl. In adapted cells, the MIC and thus tolerance to NaCl increased considerably compared to non- adapted non exposed cells. The MIC of adapted and non-adapted cells was shown to be directly dependent on the pH of the medium, and so the comparison of tolerance to chloride or other anions in acidophilic microorganisms should be carried out whilst strictly monitoring the pH of the growth medium. Transcriptomic data and experimental validations showed that the most significant responses of L. ferriphilum to chloride adaptation included neutralization and/or expulsion of protons through activation of carbonic anhydrase, respiratory cytochrome c oxidase and sul- fide:quinone reductase. Thus, the regulation of pH homeostasis seems to play a key role in the adaptive response. Towards the same goal, a cation/proton antiporter system CzcA that extrudes cations through the entry of protons was down-regulated. In addition, the increase in respiratory activity and oxygen consumption correlated with activation of antioxidant responses in which genes encoding for ROS scavenging properties and biomolecule protection seem to play a relevant role in controlling the intracellular ROS level and the redox status of adapted cells. The response detected shows that oxidative stress is an important element of the toxicity induced by chloride, and this could largely explain the reason why iron-oxidizing microorganisms have been reported to be more sensitive to the presence of anions than sulfur- oxidizers or other acidophiles [75]. Under cultivation conditions, iron-oxidizing microorgan- isms are exposed to high concentrations of iron as an energy substrate, while sulfur oxidizers are exposed to trace concentrations of this element that is used only as a micronutrient. Since PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 13 / 18 PLOS ONETranscriptomic of Na-Cl adapted Leptospirillum ferriphilum ferrous iron can trigger Fenton chemistry, its presence in high concentrations leads to a higher risk of redox stress, making the microorganisms more sensitive to other oxidative stress elici- tors. Chloride adaptation also correlated with a predicted increase in chemotaxis and biosyn- thesis of flagella, and predicted cellular communication and signaling via DSFs and c-di-GMP. Finally, an induction of genetic/genomic modifications by transposition and/or recombination also seemed to form part of the adaptive response to NaCl exposure. Although there was an increase in the activity of the electron-transport chain that likely led to an increase in ATP and NAD(P)H synthesis, carbohydrate metabolism and synthesis of polysaccharide polymers of the cell surface seemed to suffer significant decreases. Surprisingly, the canonical osmotic stress response did not appear to be necessary in salt-adapted cells, since genes for biosynthesis of the compatible solutes ectoine and hydroxyectoine were down-regulated, and only hydro- xyectoine could be detected and only in non-adapted cells without NaCl. Our results suggest that L. ferriphilum might have a response to long-term NaCl exposure that is different from other bacteria since it does not involve the upregulation of canonical mechanisms for facing osmotic stress. This study thus provides an important reference for future studies on NaCl adaptation in acidophilic bacteria. Supporting information S1 Table. Up-regulated hypothetical genes in L. ferriphilum NaCl-adapted cells. (XLSX) S2 Table. Down-regulated hypothetical genes in L. ferriphilum NaCl-adapted cells. (XLSX) Acknowledgments We thank Dr. Luis Valenzuela, Instituto Nacional de Tecnologı´a de los Alimentos (INTA), Universidad de Chile, for his help with data analyses. Author Contributions Conceptualization: Michael Schlo¨mann, Gloria Levica´n. Formal analysis: Javier Rivera-Araya, Gloria Levica´n. Funding acquisition: Renato Cha´vez, Michael Schlo¨mann, Gloria Levica´n. Investigation: Javier Rivera-Araya, Thomas Heine. Methodology: Javier Rivera-Araya, Thomas Heine. Project administration: Michael Schlo¨mann, Gloria Levica´n. Visualization: Javier Rivera-Araya. Writing – original draft: Javier Rivera-Araya, Gloria Levica´n. Writing – review & editing: Michael Schlo¨mann, Gloria Levica´n. References 1. Christel S, Herold M, Bellenberg S, El Hajjami M, Buetti-Dinh A, et al. Multi-omics reveals the lifestyle of the acidophilic, mineral-oxidizing model species Leptospirillum ferriphilum. Appl Environ Microbiol. 2018; 84: e02091–17. https://doi.org/10.1128/AEM.02091-17 PMID: 29150517 PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 14 / 18 PLOS ONETranscriptomic of Na-Cl adapted Leptospirillum ferriphilum 2. Schippers A, Hedrich S, Vasters J, Drobe M, Sand W, Willscher S. Biomining: metal recovery from ores with microorganisms. Adv Biochem Eng Biotechnol. 2014; 141:1–47. https://doi.org/10.1007/10_ 2013_216 PMID: 23793914 3. Shiers DW, Collinson DM, Watling HR. Life in heaps: a review of microbial responses to variable acidity in sulfide mineral bioleaching heaps for metal extraction. Res Microbiol. 2016; 167(7): 576–86. https:// doi.org/10.1016/j.resmic.2016.05.007 PMID: 27283362 4. Rivera-Araya J, Pollender A, Huynh D, Schlo¨ mann M, Cha´ vez R, et al. Osmotic imbalance, cytoplasm acidification and oxidative stress induction support the high toxicity of chloride in acidophilic bacteria. Front Microbiol, 2019; 10: 2455. https://doi.org/10.3389/fmicb.2019.02455 PMID: 31736901 5. Falaga´ n C, Johnson DB. The significance of pH in dictating the relative toxicities of chloride and copper to acidophilic bacteria. Res Microbiol. 2018; 169(10): 552–557. https://doi.org/10.1016/j.resmic.2018. 07.004 PMID: 30031071 6. Alexander B, Leach S, Ingledew WJ. The relationship between chemiosmotic parameters and sensitiv- ity to anions and organic acids in the acidophile Thiobacillus ferrooxidans. J Gen Microbiol, 1987; 133: 1171–1179. https://doi.org/10.1099/00221287-133-5-1171 7. Zammit C, Mangold S, Mutch LA, Watling HR, Dopson M, et al. Bioleaching in brackish waters—effect of chloride ions on the acidophile population and proteomes of model species. Appl Microbiol Biotech- nol. 2012; 93(1): 319–329. https://doi.org/10.1007/s00253-011-3731-3 PMID: 22124722 8. Parro V, Moreno-Paz M, Gonza´ lez-Toril E. Analysis of environmental transcriptomes by DNA microar- rays. Environ Microbiol. 2007; 9: 453–464. https://doi.org/10.1111/j.1462-2920.2006.01162.x PMID: 17222143 9. Mosier A, Justice N, Bowen B, Baran R, Thomas B, et al. Metabolites associated with adaptation of microorganisms to an acidophilic, metal-rich environment identified by stable-isotope-enabled metabo- lomics. mBio. 2013; 4(2):e00484–12. https://doi.org/10.1128/mBio.00484-12 PMID: 23481603 10. Galleguillos P, Grail B, Hallberg K, Demergasso C, Johnson DB. Identification of trehalose as a compat- ible solute in different species of acidophilic bacteria. J Microbiol. 2018; 56, 727–733. https://doi.org/10. 1007/s12275-018-8176-2 PMID: 30267316 11. Rivera-Araya J, Huynh ND, Kaszuba M, Cha´vez R, Schlo¨ mann M, et al. Mechanisms of NaCl-tolerance in acidophilic iron-oxidizing bacteria and archaea: Comparative genomic predictions and insights. Hydrometallurgy. 2020; 194: 105334. https://doi.org/10.1016/j.hydromet.2020.105334 12. Kieft T, Spence S. Osmoregulation in Thiobacillus ferrooxidans: Stimulation of iron oxidation by proline and betaine under salt stress. Curr Microbiol. 1988; 17: 255–258. https://doi.org/10.1007/BF01571324 13. Guo X, Jiang C, Luo Y, Zhang M, Poetsch A, et al. Proteomic and molecular investigations revealed that Acidithiobacillus caldus adopts multiple strategies for adaptation to NaCl stress. Chin Sci Bull. 2014; 59(3): 301–309. https://doi.org/10.1007/s11434-013-0039-y 14. Dopson M, Holmes D, Lazcano M, McCredden T, Bryan C, et al. Multiple osmotic stress responses in Acidihalobacter prosperus result in tolerance to chloride ions. Front Microbiol. 2017; 7:2132. https://doi. org/10.3389/fmicb.2016.02132 PMID: 28111571 15. Suzuki I, Lee D, Mackay B, Harahuc L, Oh JK. Effect of various ions, pH, and osmotic pressure on oxi- dation of elemental sulfur by Thiobacillus thiooxidans. Appl Environ Microbiol. 1999; 65(11): 5163– 5168. https://doi.org/10.1128/AEM.65.11.5163-5168.1999 PMID: 10543839 16. 17. 18. Imlay JA. Iron-sulphur clusters and the problem with oxygen. Mol Microbiol. 2006; 59(4), 1073–1082. https://doi.org/10.1111/j.1365-2958.2006.05028.x PMID: 16430685 Ferrer A, Bunk B, Spro¨ er C, Biedendieck R, Valde´ s N, Jahn M, et al. Complete genome sequence of the bioleaching bacterium Leptospirillum sp. group II strain CF-1. J Biotechnol. 2016; 222: 21–22. https:// doi.org/10.1016/j.jbiotec.2016.02.008 PMID: 26853478 Zhu JY, Shi X, Lu H, Xia B, Li Y, et al. RNA-seq transcriptome analysis of extensor digitorum longus and soleus muscles in large white pigs. Mol Genet Genom. 2016; 291(2): 687–70. https://doi.org/10.1007/ s00438-015-1138-z PMID: 26520103 19. Hosseinpour B, Sepahvand S, Aliabad KK, Bakhtiarizadeh M, Imani A, et al. Transcriptome profiling of fully open flowers in a frost-tolerant almond genotype in response to freezing stress. Mol Genet Geno- mics. 2018; 293: 151–163. https://doi.org/10.1007/s00438-017-1371-8 PMID: 28929226 20. Love M, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014; 15: 550. https://doi.org/10.1186/s13059-014-0550-8 PMID: 25516281 21. Araujo FA, Barh D, Silva A, Guimarães L, Ramos R.T.J. GO FEAT: a rapid web-based function annota- tion tool for genomic and transcriptomic data. Sci. Rep. 2018; 8(1): 1794. https://doi.org/10.1038/ s41598-018-20211-9 PMID: 29379090 PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 15 / 18 PLOS ONETranscriptomic of Na-Cl adapted Leptospirillum ferriphilum 22. Giebner F, Eisen S, Schlo¨mann M, Schopf S. Measurements of dissolved oxygen in bioleaching reac- 23. tors by optode application. Hydrometallurgy. 2017; 168: 64–68. https://doi.org/10.1016/j.hydromet. 2016.08.001 Ferrer A, Rivera-Araya J, Zapata C, Norambuena J, Sandoval A´ , et al. Cobalamin protection against oxidative stress in the acidophilic iron-oxidizing bacterium Leptospirillum group II CF-1. Front Microbiol. 2016; 7: 748. https://doi.org/10.3389/fmicb.2016.00748 PMID: 27242761 24. Bradford M. A rapid and sensitive method for the quantitation of microgram quantities of protein utilizing the principle of protein-dye binding. Anal Biochem. 1976; 7: 248–254. https://doi.org/10.1006/abio. 1976.9999 PMID: 942051 25. Huynh D, Giebner F, Kaschabek SR, Rivera-Araya J, Levican G, et al. Effect of sodium chloride on Lep- tospirillum ferriphilum DSM 14647 and Sulfobacillus thermosulfidooxidans DSM 9293: Growth, iron oxi- dation activity and bioleaching of sulfidic metal ores. Minerals Engineering. 2019; 138: 52–59. https:// doi.org/10.1016/j.mineng.2019.04.033 26. Supuran C, Capasso C. An overview of the bacterial carbonic anhydrases. Metabolites. 2017; 7(4): 56. https://doi.org/10.3390/metabo7040056 PMID: 29137134 27. Badger MR, Price GD. CO2 concentrating mechanisms in cyanobacteria: molecular components, their diversity and evolution. J Exp Bot. 2003; 54(383): 609–622. https://doi.org/10.1093/jxb/erg076 PMID: 12554704 28. Berrino E, Supuran CT. Novel approaches for designing drugs that interfere with pH regulation. Expert Opin Drug Discov. 2019; 14(3):231–248. https://doi.org/10.1080/17460441.2019.1567488 PMID: 30681011 29. 30. Levica´n G, Ugalde JA, Ehrenfeld N, Maass A, Parada P. Comparative genomic analysis of carbon and nitrogen assimilation mechanisms in three indigenous bioleaching bacteria: predictions and validations. BMC Genomics. 2008; 9(1): 581. https://doi.org/10.1186/1471-2164-9-581 Zhang X, Liu Z, Wei G, Yang F, Liu X. In silico genome-wide analysis reveals the potential links between core genome of Acidithiobacillus thiooxidans and its autotrophic lifestyle. Front Microbiol. 2018; 9: 1255. https://doi.org/10.3389/fmicb.2018.01255 PMID: 29937764 31. Esparza M, Jedlicki E, Gonza´ lez C, Dopson M, Holmes DS. Effect of CO2 onccentration on uptake and assimilation of inorganic carbon in the extreme acidophile Acidithiobacillus ferrooxidans. Front Micro- biol. 2019; 10: 603. https://doi.org/10.3389/fmicb.2019.00603 PMID: 31019493 32. Levica´n G, Go´mez M, Cha´vez R, Orellana O, Moreno-Paz M, et al. Comparative genomic analysis reveals novel facts about Leptospirillum spp. cytochromes. J Mol Microbiol Biotechnol. 2012; 22: 94– 104. https://doi.org/10.1159/000338105 PMID: 22627128 33. Ducluzeau AL, Ouchane S, Nitschke W. The cbb3 oxidases are an ancient innovation of the domain bacteria. Mol Biol Evol. 2008; 25(6):1158–1166. https://doi.org/10.1093/molbev/msn062 PMID: 18353797 34. Xie Z, Liu Y, Bian J. Hydrogen sulfide and cellular redox homeostasis. Oxid Med Cell Longev. 2016; 2016: 6043038. https://doi.org/10.1155/2016/6043038 PMID: 26881033 35. Fu LH, Wei ZZ, Hu KD, Hu LY, Li YH, Chen XY, et al. Hydrogen sulfide inhibits the growth of Escherichia coli through oxidative damage. J Microbiol. 2018; 56(4):238–245. https://doi.org/10.1007/s12275-018- 7537-1 PMID: 29492867 36. Schumann W. FtsH–a single-chain charonin? FEMS Microbiol Rev. 1999; 23(1): 1–11. https://doi.org/ 10.1111/j.1574-6976.1999.tb00389.x PMID: 10077851 37. Bandyopadhyay K, Parua PK, Datta AB, Parrack P. Escherichia coli HflK and HflC can individually inhibit the HflB (FtsH)-mediated proteolysis of λCII in vitro. Arch Biochem Biophys. 2010; 501(2): 239– 243. https://doi.org/10.1016/j.abb.2010.06.030 PMID: 20599668 38. Garba L, Ali MSM, Oslan SN, Abdul Rahman RNZBR. Review on fatty acid desaturases and their roles in temperature acclimatisation. J Applied Sci. 2017; 17: 282–295. https://doi.org/10.3923/jas.2017.282. 295 39. Allakhverdiev SI, Kinoshita M, Inaba M, Suzuki I, Murata N. Unsaturated fatty acids in membrane lipids protect the photosynthetic machinery against salt-induced damage in Synechococcus. Plant Physiol. 2001; 125(4): 1842–1853. https://doi.org/10.1104/pp.125.4.1842 PMID: 11299364 40. Kim S, Ha J, Lee H, Lee S, Lee J, Choi Y, et al. Role of Pseudomonas aeruginosa DesB in adaptation to osmotic stress. J Food Prot. 2019; 82(8):1278–1282. https://doi.org/10.4315/0362-028X.JFP-18-507 PMID: 31298570 41. Schuman B, Alfaro JA, Evans SV. Glycosyltransferase structure and function. In: Peters T, editors. Bio- active conformation I. Topics in current chemistry. Springer, Berlin, Heidelberg; 2006. pp. 217–257. https://doi.org/10.1007/128_2006_089 PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 16 / 18 PLOS ONETranscriptomic of Na-Cl adapted Leptospirillum ferriphilum 42. Thoden JB, Holden HM. The molecular architecture of glucose-1-phosphate uridylyltransferase. Protein Sci. 2007; 16(3): 432–40. https://doi.org/10.1110/ps.062626007 PMID: 17322528 43. Chen J, Yang S. Catalytic mechanism of UDP-glucose dehydrogenase. Biochem Soc Trans. 2019; 47 (3): 945–955. https://doi.org/10.1042/BST20190257 PMID: 31189734 44. Chevreux G, Atmanene C, Lopez P, Ouazzani J, Van Dorsselaer A, et al. Monitoring the dynamics of monomer exchange using electrospray mass spectrometry: the case of the dimeric glucosamine-6- phosphate synthase. J Am Soc Mass Spectrom. 2011; 22(3): 431–439. https://doi.org/10.1007/ s13361-010-0054-z PMID: 21472562 45. Ramos-Aires J, Ple´siat P, Kocjancic-Curty L, Thilo Ko¨ hler T. Selection of an antibiotic-hypersusceptible mutant of Pseudomonas aeruginosa: identification of the GlmR transcriptional regulator. Antimicrob Agents Chemother. 2004; 48(3): 843–851. https://doi.org/10.1128/AAC.48.3.843-851.2004 PMID: 14982774 46. Allard S, Beis K, Giraud MF, Hegeman A, Gross J, et el. Toward a structural understanding of the dehy- dratase mechanism. Structure, 2002; 10 (1): 81–92. https://doi.org/10.1016/s0969-2126(01)00694-3 PMID: 11796113 47. Jorgenson MA, Young KD. Interrupting biosynthesis of O antigen or the lipopolysaccharide core pro- duces morphological defects in Escherichia coli by sequestering undecaprenyl phosphate. J Bacteriol. 2016; 198: 3070–3079. https://doi.org/10.1128/JB.00550-16 PMID: 27573014 48. Mueller EA, Levin PA. Bacterial cell wall quality control during environmental stress. mBio. 2020; 11(5): e02456–20. https://doi.org/10.1128/mBio.02456-20 PMID: 33051371 49. Egan AJF, Errington J, Vollmer W. Regulation of peptidoglycan synthesis and remodelling. Nat Rev Microbiol. 2020; 18(8): 446–460. https://doi.org/10.1038/s41579-020-0366-3 PMID: 32424210 50. Xu Y, Moeller A, Jun SY, Le M, Yoon BY, et al. Assembly and channel opening of outer membrane pro- tein in tripartite drug efflux pumps of Gram-negative bacteria. J Biol Chem. 2012; 287(15): 11740– 11750. https://doi.org/10.1074/jbc.M111.329375 PMID: 22308040 51. Lee JJ, Wu YC, Kuo CJ, Hsuan SL, Chen TH. TolC is important for bacterial survival and oxidative stress response in Salmonella enterica serovar Choleraesuis in an acidic environment. Vet Microbiol. 2016; 193: 42–48. https://doi.org/10.1016/j.vetmic.2016.08.006 PMID: 27599929 52. Yuan B, Cheng A, Wang M. Polysaccharide export outer membrane proteins in Gram-negative bacte- ria. Future Microbiol. 2013; 8(4): 525–35. https://doi.org/10.2217/fmb.13.13 PMID: 23534363 53. Nikaido H, Takatsuka Y. Mechanisms of RND Multidrug Efflux Pumps. Biochim Biophys Acta. 2009; 1794(5): 769–781. https://doi.org/10.1016/j.bbapap.2008.10.004 PMID: 19026770 54. Allikmets R, Gerrard B, Court D, Dean M. Cloning and organization of the abc and mdl genes of Escher- ichia coli: relationship to eukaryotic multidrug resistance. Gene. 1993; 136(1–2): 231–236. https://doi. org/10.1016/0378-1119(93)90470-n PMID: 7904973 55. Moussatova A, Kandt C, O’Mara ML, Tieleman DP. ATP-binding cassette transporters in Escherichia coli. Biochim Biophys Acta. 2008; 1778(9): 1757–1771. https://doi.org/10.1016/j.bbamem.2008.06.009 PMID: 18634750 56. Staerck C, Gastebois A, Vandeputte P, Calenda A, Larcher G, et al. Microbial antioxidant defense enzymes. Microb Pathog. 2017; 110: 56–65. https://doi.org/10.1016/j.micpath.2017.06.015 PMID: 28629723 Zapata C, Paillavil B, Cha´ vez R, A´ lamos P, Levica´ n G. Cytochrome c peroxidase (CcP) is a molecular determinant of the oxidative stress response in the extreme acidophilic Leptospirillum sp. CF-1. FEMS Microbiol Ecol. 2017 93(3). https://doi.org/10.1093/femsec/fix001 PMID: 28087802 57. 58. Wang C, Wesener SR, Zhang H, Cheng YQ. An FAD-dependent pyridine nucleotide-disulfide oxidore- ductase is involved in disulfide bond formation in FK228 Anticancer Depsipeptide. Chem Biol. 2009; 16 (6): 585–593. https://doi.org/10.1016/j.chembiol.2009.05.005 PMID: 19549597 59. Gonza´ lez D, A´ lamos P, Rivero M, Orellana O, Norambuena J, et al. Deciphering the role of multiple thioredoxin fold proteins of Leptospirillum sp. in oxidative stress tolerance. Int J Mol Sci. 2020; 21(5): 1880. https://doi.org/10.3390/ijms21051880 PMID: 32164170 60. Py B, Barras F. Building Fe-S proteins: bacterial strategies. Nat Rev Microbiol. 2010; 8: 436–446. https://doi.org/10.1038/nrmicro2356 PMID: 20467446 61. Ralston DM, O’Halloran TV. Ultrasensitivity and heavy-metal selectivity of the allosterically modulated MerR transcription complex. Pro Natl Acad Sci U S A. 1990; 87: 3846–3850. https://doi.org/10.1073/ pnas.87.10.3846 PMID: 2187194 62. Rojas LA, Ya´ ñez C, Gonza´lez M, Lobos S, Smalla K, et al. Characterization of the metabolically heavy metal-resistant Cupriavidus metallidurans strain MSR33 generated for mercury remediation. PLoS One. 2011; 6: e17555. https://doi.org/10.1371/journal.pone.0017555 PMID: 21423734 PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 17 / 18 PLOS ONETranscriptomic of Na-Cl adapted Leptospirillum ferriphilum 63. Alviz-Gazitua P, Fuentes-Alburquenque S, Rojas LA, Turner RJ, Guiliani N, et al. The response of Cupriavidus metallidurans CH34 to cadmium involves inhibition of the initiation of biofilm formation, decrese in intracellular c-di-GMP levels, and a novel metal regulated phosphodiesterase. Front. Micro- biol. 2019; 10: 1499. https://doi.org/10.3389/fmicb.2019.01499 PMID: 31338076 64. Park S, Ely RL. Candidate stress genes of Nitrosomonas europaea for monitoring inhibition of nitrifica- tion by heavy metals. Appl Environ Microbiol. 2008; 74(17): 5475–82. https://doi.org/10.1128/AEM. 00500-08 PMID: 18606795 65. Ferrer A, Orellana O, Levica´ n G. Oxidative stress and metal tolerance in extreme acidophiles. In: Qua- trini R, Johnson D. editors. Acidophiles: Life in extremely acidic environments. Caister Academic Press; 2016. pp. 63–76. 66. Silver S. Bacterial resistances to toxic metal ions—a review. Gene. 1996; 179(1): 9–19. https://doi.org/ 10.1016/s0378-1119(96)00323-x PMID: 8991852 67. Nies DH. The cobalt, zinc, and cadmium efflux system CzcABC from Alcaligenes eutrophus functions as a cation-proton antiporter in Escherichia coli. J Bacteriol. 1995; 177(10), 2707–2712. https://doi.org/ 10.1128/jb.177.10.2707-2712.1995 PMID: 7751279 68. Ryan RP, Fouhy Y, Garcia BF, Watt SA, Niehaus K, et al. Interspecies signalling via the Stenotropho- monas maltophilia diffusible signal factor influences biofilm formation and polymyxin tolerance in Pseu- domonas aeruginosa. Mol Microbiol. 2008; 68(1): 75–86. https://doi.org/10.1111/j.1365-2958.2008. 06132.x PMID: 18312265 69. Deng Y, Lim A, Lee J, Chen S, An S, et al. Diffusible signal factor (DSF) quorum sensing signal and structurally related molecules enhance the antimicrobial efficacy of antibiotics against some bacterial pathogens. BMC Microbiol. 2014; 14(1): 51. https://doi.org/10.1186/1471-2180-14-51 PMID: 24575808 70. Ionescu M, Baccari C, Da Silva AM, Garcia A, Yokota K, et al. Diffusible signal factor (DSF) synthase RpfF of Xylella fastidiosa is a multifunction protein also required for response to DSF. J bacteriol. 2013; 195(23): 5273–5284. https://doi.org/10.1128/JB.00713-13 PMID: 24056101 71. Stock JB, Baker MD. Chemotaxis. In: Schaechter M, editor. Encyclopedia of microbiology. Academic Press; 2009. pp. 71–78. https://doi.org/10.1016/b978-012373944-5.00068–7 72. Mukhopadhyay A, He Z, Alm EJ, Arkin AP, Baidoo AE, Borglin SC, et al. Salt stress in Desulfovibrio vul- garis Hildenborough: an integrated genomics approach J Bacteriol. 2006; 188(11):4068–4078. https:// doi.org/10.1128/JB.01921-05 PMID: 16707698 73. 74. Ikeda T, Shinagawa T, Ito T, Ohno Y, Kubo A, Nishi J, et al. Hypoosmotic stress induces flagellar bio- synthesis and swimming motility in Escherichia albertii. Commun Biol. 2020; 3: 87. https://doi.org/10. 1038/s42003-020-0816-5 PMID: 32111956 Frost L, Leplae R, Summers A, Toussaint A. Mobile genetic elements: the agents of open source evolu- tion. Nat Rev Microbiol. 2005; 3, 722–732. https://doi.org/10.1038/nrmicro1235 PMID: 16138100 75. Rea SM, Mcsweeney NJ, Degens BP, Morris C, Siebert HM, Kaksonen AH. Salt-tolerant microorgan- isms potentially useful for bioleaching operations where fresh water is scarce. Miner Eng. 2015; 75: 126–132. https://doi.org/10.1016/j.mineng.2014.09.011 PLOS ONE \| https://doi.org/10.1371/journal.pone.0267316 April 29, 2022 18 / 18 PLOS ONE
10.1371_journal.pone.0269830	RESEARCH ARTICLE Dynamical intervention planning against COVID-19-like epidemics Gabriele OlivaID 1, Martin Schlueter2, Masaharu Munetomo2, Antonio Scala3,4,5,6 1 Unit of Automatic Control, Department of Engineering, Università Campus Bio-Medico di Roma, Rome, Italy, 2 Information Initiative Center, Hokkaido University, Sapporo, Japan, 3 CNR-ISC, Applico Lab, Roma, Italy, 4 Centro Ricerche Enrico Fermi, Roma, Italy, 5 Big Data in Health Society, Roma, Italy, 6 Global Health Security Agenda - GHSA, Roma, Italy * g.oliva@unicampus.it (GO); antonio.scala@cnr.it (AS) Abstract COVID-19 has got us to face a new situation where, for the lack of ready-to-use vaccines, it is necessary to support vaccination with complex non-pharmaceutical strategies. In this paper, we provide a novel Mixed Integer Nonlinear Programming formulation for fine- grained optimal intervention planning (i.e., at the level of the single day) against newborn epidemics like COVID-19, where a modified SIR model accounting for heterogeneous popu- lation classes, social distancing and several types of vaccines (each with its efficacy and delayed effects), allows us to plan an optimal mixed strategy (both pharmaceutical and non- pharmaceutical) that takes into account both the vaccine availability in limited batches at selected time instants and the need for second doses while keeping hospitalizations and intensive care occupancy below a threshold and requiring that new infections die out at the end of the planning horizon. In order to show the effectiveness of the proposed formulation, we analyze a case study for Italy with realistic parameters. Introduction The ongoing COVID-19 pandemics, with its huge toll in terms of deaths and economic dam- age, represents an unparalleled global threat to human society as a whole. As of May 2021, reportedly more than 155 million COVID-19 cases have been identified, with more than three million deaths [1]. To face such a threat, governments have initially reacted via non-pharma- ceutical interventions, i.e., by enforcing strict social distancing [2–5]. Then, with an unprece- dented effort by human society as a whole, a wide variety of vaccines have been developed in a remarkably narrow time span [6–10]; such a rapid development has been possible also due to the extensive reliance on bioinformatics [7] and artificial intelligence [11]. Notably, the effec- tiveness and geographical distribution of such a plethora of vaccines has proven to be highly heterogeneous (e.g., see [12] and references therein). Notice that, to date, no universally acknowledged cure has been identified; the case of the debate regarding therapies based on Remdesivir represents an illustrative example in this sense [13, 14]. Therefore, to date, the con- trol knobs available to governments amount to just social distancing (e.g., lockdowns, limiting affluence to shops, wearing masks, etc.) and vaccination. In the literature, optimization tools a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Oliva G, Schlueter M, Munetomo M, Scala A (2022) Dynamical intervention planning against COVID-19-like epidemics. PLoS ONE 17(6): e0269830. https://doi.org/10.1371/journal. pone.0269830 Editor: Maria Vittoria Barbarossa, Frankfurt Institute for Advanced Studies, GERMANY Received: December 13, 2021 Accepted: May 30, 2022 Published: June 14, 2022 Copyright: © 2022 Oliva et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the article and its Supporting information files. Funding: Please note that prof. Masaharu Munetomo received funding by the Japan Society for the Promotion of Science (JSPS) under the funding scheme KAKENHI, Grant Number JP20K11967. We declare that the funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 1 / 21 PLOS ONECompeting interests: The authors have declared that no competing interests exist. to face epidemics, such as OpenMalaria [15, 16] and STDSIM [17], have proved their effective- ness in analyzing and planning illness containment [18–20]. Dynamical intervention planning against COVID-19-like epidemics However, in the case of COVID-19, the unavailability of reliable information with adequate level of detail requires reliance on simpler approaches. Among several other options, SIR mod- els [21, 22] represent a reasonable choice in terms of predictive capability and simplicity. Inter- estingly, such models have been applied to describe different epidemics such as SARS [23] Influenza A (H1N1) [24], Measles [25] and Hepatitis C [26]. Moreover, compartmental mod- els in general have proven useful to model quite different epidemics scenarios e.g., Chlamydia trachomatis, antiviral treatment in the case of HIV, nosocomial infections and transmission of antibiotic-resistant pathogens [27]. For this reason, such model allows to understand the avail- able degrees of freedom, i.e., the policies that can be put in place to react to the epidemics, even in the absence of detailed quantitative predictions [28]. Indeed, several control [29–31] and optimization [29, 32–35] approaches have been developed, based on simple epidemics models such as the SIR. In particular, it is worth mentioning: [35], where a coarse-grained and static optimization framework for selecting the amount of vaccines to be allocated to different popu- lation classes with the aim of ending the epidemics is given; [36], where the authors focus on vaccination of essential workers; [37] where the authors allocate vaccines to age classes in order to optimize cost functionals such as deaths or hospitalization, but do not guarantee the end of the epidemics at the end of the considered time horizon nor consider different vaccines with different efficacies at the same time. Other examples include [38, 39], where optimization of the supply chain underlying the vaccine delivery is considered. Notably, based on epidemio- logical data from the UK together with estimates of vaccine efficacy, [40] provides a framework to conduct “what if” analysis of the evolution of the disease under different interventions. However, while being beneficial for high-level policy making, does not translate into an opera- tive plan for the administration of pharmaceutical and non-pharmaceutical interventions. Moreover, the workin [41] provides a model predictive approach to the problem where inter- ventions at time t are selected based on the foreseen effect in the next future, following a “receding horizon” approach. However, although quite detailed, the epidemics model underly- ing such framework amounts to a single population of individuals and does not distinguish between age classes. To the best of our knowledge, to date no formulation is able to support the fine-grain, oper- ative and dynamical planning of interventions, while accounting for a wide range of phenom- ena that are peculiar of epidemics like COVID-19, e.g., different types of vaccines, the need for a second dose, the capacity of the healthcare system in terms of regular and intensive care hos- pitalization, the availability of vaccines in batches at selected time instants. In this paper, we fill this gap by providing a novel optimization formulation that aims at implementing an optimal intervention plan to fight newborn epidemics like COVID-19, i.e., epidemics characterized by two key factors: (i) high infection rate and (ii) high stress posed on the healthcare system and/or society in terms of intensive care occupancy, deaths or economic consequences. Specifically, we first develop a modified discrete-time SIR model for heteroge- neous population classes (e.g., age or geographical classes) that accounts for the effect of social distancing and vaccination. In more detail, we assume the ratio at which individuals in two classes infect each other can be reduced by enforcing tailored social distancing measures. Moreover, we consider several types of vaccines, each characterized by their efficacy as well as the delay required for the vaccination to be effective. In this view, we assume that, after the vac- cine takes effect, a fraction of the population becomes immune. Notice that we explicitly take into account the possibility to plan for a first and a second dose of the vaccines. Based on the proposed variation of the SIR model, we develop an optimization formulation that aims at planning social distancing measures and vaccinations at the level of the single day PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 2 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics in order to reach the end of the epidemics and to guarantee that the state reached is robust, in that new infections die out. In doing so, we enforce constraints accounting for aspects such as the need of a second dose, the delayed and partial effect of multiple types of vaccines, the requirement that congestions of the healthcare system are avoided. Briefly, we show that the proposed formulation amounts to a Mixed Integer Nonlinear Pro- gramming (MINLP) problem. We point out that the term MINLP is used to denote a class of optimization problems where some of the variables being selected are constrained to be inte- ger-valued while some other can be real-valued (i.e., Mixed Integer (MI)), the objective func- tion and/or some of the constraints are nonlinear functions (i.e., Nonlinear (NL)). In the context of optimization, “program” or “programming” (P) can be regarded as a synonym of optimization problem or optimization formulation. Notably, the class of MINLP problems is known to be computationally hard to solve (e.g., see [42]); therefore, exact solution of the problem is a nontrivial task, thus calling for approximation strategies to be put in place. Materials and methods Notation We denote vectors by boldface lowercase letters and matrices with uppercase letters and we refer to the (i, j)-th entry of a matrix A by Aij. We represent by 0n and 1n vectors with n compo- nents, all equal to zero and to one, respectively. Moreover, we use 0n×m, 1n×m to denote an n×m matrix with just zero and one entries, respectively. We use square brackets to denote the arguments of a function, e.g., we use f[x, y] to denote a function with arguments x and y. For the sake of brevity, where understood, we abbreviate a function of one or more arguments by f[�]. Given a vector x 2 Rn we denote by diag[x] the n × n diagonal matrix having diag[x]ii = xi, for all i 2 {1, . . ., n}. On the same footings, given a matrix A 2 Rn � Rn we denote by diag[A] the n dimensional vector having diag[A]i = Aii, for all i 2 {1, . . ., n}. We denote by � the Hada- mard (i.e., entry-wise) matrix product between matrices A and B with the same dimensions, i.e., the matrix C = A � B is such that Cij = Aij Bij; analogously, the Hadamard product between c = a � b between two vectors a,b is such that ci = ai bi. Remember that Hadamard product is commutative; moreover, notice that a � b = diag[a]b = diag[b]a. SIR epidemics model Pn In this section we briefly review the SIR Epidemics model; the interested reader is referred to [35, 43] for further details. Let us consider a population of N individuals divided in n classes (e.g., by age or geographical area); we denote by Nℓ the population in the ℓ-th class with N ¼ ‘¼1 N‘. Moreover, let us indicate with sℓ[t], iℓ[t], rℓ[t] the fraction of susceptible, infec- tious and removed individuals in the ℓ-th class at time t and with s[t], i[t], r½t� 2 Rn the stack of such variables for all classes. In the following, we assume that s[0], i[0], r[0] 2 [0, 1]n and s[0] + i[0] + r[0] = 1n. The SIR equations for such an heterogeneous population are given by 8 >>>< >>>: @ts½t� ¼ (cid:0) s½t� � Bi½t� @ti½t� ¼ s½t� � Bi½t� (cid:0) g � i½t� ð1Þ @tr½t� ¼ g � i½t�; where B is the n × n transmission matrix, Bij being the rate at which a susceptible individual of class i meets an infectious individual of class j and becomes infected, while the vector g 2 Rn collects the rates γi at which infectious individuals in the i-th class are removed from the PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 3 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics infection cycle. Notably, the above choice for the initial conditions guarantees that s[t], i[t], r[t] 2 [0, 1]n and s[t] + i[t] + r[t] = 1n for all time instants t. End of the epidemics. Within the SIR model, an epidemic ends when i[t] = 0n, i.e., when s½t� þ r½t� ¼ 1n; such states are also called end-of-epidemic states. At this point, let us define n = [N1, . . ., Nn]T and let us consider the overall amount of infectious individuals I[t] = iT[t]n; we have that @tI½t� ¼ @tiT½t�n ¼ i½t�TðBT diag ½n�s½t� (cid:0) diag ½n�gÞ: In this view, since i[t] � 0, the total number of infected individuals I[t] is guaranteed to decrease with time irrespective of the particular value of i[t] if and only if BT diag[n]s[t] − diag[n]γ < 0n, i.e., if and only if [35, 44] Rs � 1n; ð2Þ where R ¼ diag ðgÞ(cid:0) 1 di ag ½n�(cid:0) 1BT di ag ½n� is linked to the next generation matrix [43] that characterizes the stability of an end-of-epidemic state respect to infections and allows to calcu- late the basic reproduction number R0 indicating the theoretical rate of new infections that an infectious individual could generate. Modeling assumptions and limits of the SIR model. The simplicity of the SIR model allows to design a scenario based on a limited number of parameters; it is thus one of the first models used to understand newborn epidemics. However, SIR models can sometime oversim- plify the complex disease process. As an example, SIR models imply “full mixing”, i.e., the assumption that all individuals in the population are equally likely to be in contact with each other. To this respect, the heterogeneous SIR corrects such an issue by introducing classes and considering the heterogeneity in their contact rate. Also, we have employed a simplified SIR model with fixed populations, although in the original formulation it could account also for migration, births or deaths; such an approach is justified in the initial phase of an epidemic, where the time horizon is limited and variation in population can be disregarded. When an epidemic becomes endemic, SIR models can be easily extended to SIRS models where recov- ered individuals can become again susceptible. Furthermore, if there is a waiting time for an infected person to become infectious, SIR models can be extended to SEIR models by intro- ducing an extra compartment E (i.e. “exposed”) that accounts for such an issue. In a “receding horizon” approach, where the model parameters are periodically adjusted to reflect the evolv- ing knowledge on the epidemic, it is reasonable to resort to the SIR model during the first iteration, since its parameters are the simplest to estimate. Eventually, at later iterations, it is possible not only adjust the parameters, but also to switch to more sophisticated models (SEIR or even SEIRS if the situation becomes endemic) with the proceeding of time. Our framework easily allows to switch from SIR to SEIR or SEIRS model just by adding extra compartments; notice that, by defining by a[t] = i[t] + e[t] the fraction of infected (i.e., exposed or infectious) individuals, the constrains ensuring the dampening of the epidemic for all classes (i.e., @a[t]<0) retain the same simple linear form of Eq (2), i.e., they depend only on the fraction of infectious individuals [44]. Modeling interventions within the SIR model In this section we modify the SIR model in order to explicitly account for possible interven- tions, namely, social distancing measures (e.g., adoption of personal protection equipment (PPE) and lockdowns) and vaccination. In view of later developments in the paper, it is conve- nient to first express the SIR model in discrete-time form. Notice that exact discretization of a PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 4 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics nonlinear differential equation with constant step size Δt would be in the form @tx½t� ¼ f ½x½t�� Z ðkþ1ÞDt x½ðk þ 1ÞDt� ¼ x½kDt� þ f ½x½t��dt: kDt ð3Þ However, given the complexity of the above exact method, it is convenient to consider an approximated relation. In particular, in this paper we resort to the Euler forward approxima- tion, i.e., we set thus obtaining � � � � @tx½t� t¼kDt � x½ðk þ 1ÞDt� (cid:0) x½kDt� Dt ; x½ðk þ 1ÞDt� � x½kDt� þ Dtf ½x½kDt��: In other words, we approximate the continuous-time SIR model in Eq (1) via the following discrete-time equations 8 >>>< >>>: s½ðh þ 1ÞDt� ¼ s½hDt� (cid:0) Dts½hDt� � Bi½hDt� i½ðh þ 1ÞDt� ¼ i½hDt� þ Dts½ðh þ 1ÞDt� � Bi½hDt� (cid:0) Dtg � i½ðh þ 1ÞDt� ð4Þ r½ðh þ 1ÞDt� ¼ r½hDt� þ Dtg � i½k�: and we point out that, to avoid numerical instability as a result of the discretization, we choose Δt = 0.01[day] for the parameters used in our case study (Other approaches allowing larger step size without causing instability could be considered, such as Euler backward integration, trapezoidal integration or Runge-Kutta methods; however, in this paper we opted for the Euler forward integration for the sake of simplicity). Let us now incorporate two different types of intervention in the above discrete-time SIR model, accounting for the adoption of social distancing measures and for vaccination. Notably, in the following, we consider interventions such as vaccinations and social distancing mea- sures that are planned at the level of the single day; as discussed next, such interventions will reflect in the discrete-time SIR model by assuming that the interventions remain constant over the day. As a consequence, such daily interventions will be indexed on a daily basis, while the variables in the discrete-time SIR model are indexed by hΔt. Moreover, we use the iterator k to denote the k-th day and, where needed, with a slight abuse of notation we use the iterator k to denote the value assumed by a variable of interest at the end of the k-th day, e.g., s[k] = s[hΔt], with h = k/Δt, while we point out that the day corresponding to the time instant hΔt is given by bhΔtc. Social distancing interventions. In order to model the effect of social distancing mea- sures in the SIR model, we observe that interventions such as lockdowns, limiting access to shops or imposing the adoption of PPEs, has the effect to reduce the rate at which susceptible individuals meet infectious individuals and/or become infected, modeled by the coefficients Bij. In order to model the effect of social distancing measures at the k-th day, let us define the social distancing intensity Eℓj[k] 2 [0, e], where e is the maximum allowed intensity and e � 1, as the intensity of the social distancing measures put in place for the ℓ-th and j-th class, e.g., Eℓj[k] = 0 means no measure is implemented, while Eℓj[k] = e means the maximum effort is PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 5 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics spent in avoiding contacts between the ℓ-th and j-th classes. Notice that, by definition, we have that Eℓj[k] = Ejℓ[k]. The above social distancing intensity coefficients account for the different strategies put in place. For instance, if the classes represent geographical regions, then a large value of Eℓj[k] implies large limitation of moving from the ℓ-th to the j-th one, while intermedi- ate values of Eℓj[k] could be used to model a scenario where mobility is permitted for work and health circumstances. Conversely, if we consider age classes, then a situation where Eℓj[k] is large for all j could be used to model age-targeted lockdowns (e.g., for the elderly people). As a result of the choice of Eℓj[k], we consider time-varying terms Bℓj[k] with the following structure or, in a compact form B‘j½k� ¼ B‘j ð1 (cid:0) E‘j½k�Þ B½k� ¼ B � ð1n�n (cid:0) E½k�Þ; ð5Þ ð6Þ where the n × n matrix E[k] collects the entries Eℓj[k]. In other words, Bℓj[k] corresponds to the nominal Bℓj when no intervention is implemented and reaches zero in the case of a complete lockdown. Vaccination. Let us now model the effect of vaccination on the discrete-time SIR model. In particular, we assume m different types of vaccines are available and we assume each vaccine j has an efficacy ηj 2 [0, 1] after a single dose, while after the second dose the efficacy rises to ηj + Δηj, with Δηj 2 [0, 1 − ηj]. Notice that the second dose is not required for all types of vac- cines; we model this aspect by resorting to a coefficient ( 1; 0; �j ¼ if the second dose is required for the j(cid:0) th vaccine otherwise: Moreover, for each type j of vaccine we assume a time window of tI j days is required for j to denote the time window the vaccine to take effect after the first dose, while we use wj � tI between the first and the second dose (if required) and tII the administration of the second dose and the reach of complete effect. In other words, an administration of vaccine j on the k-th day has an initial effect on day k þ tI effect on day k þ wj þ tII ing which vaccinated individuals are still exposed to the infection. j , while τI, τII are the times estimated from pharmacological trials dur- j to denote the time window between j and a complete In order to model the effect of vaccination, let us indicate with Xℓj[k] the units of first doses of vaccines of the j-th type that are injected to the ℓ-th class of population at the k-th day and let X½k� 2 Nn�m be the matrix with integer entries collecting such variables. Moreover, let Yℓj[k] denote the amount of units of second dose of vaccines of the j-th type that are injected to the ℓ-th class of population at the k-th day. In this view, the contribution Δrℓ[k] at day k to the number of removed individuals belong- ing to the ℓ-th class as a result of vaccination satisfies N‘Dr‘½k� ¼ Xm X‘j½k (cid:0) tI j �Zj þ Xm Y‘j½k (cid:0) tII j �DZj; j¼1 j¼1 i.e., the contribution of the j-th vaccine to Δrℓ[k] corresponds to the fraction of individuals that were vaccinated tI j days before with the first dose of the j-th vaccine, weighted by its efficacy ηj, plus the fraction of individuals that were vaccinated tII j days before with the second dose of the PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 6 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics j-th vaccine, weighted by the residual efficacy Δηj. In other words, Δrℓ[k] is given by Dr‘½k� ¼ 1 N‘ Xm X‘j½k (cid:0) tI j �Zj þ Xm Y‘j½k (cid:0) tII j �DZj j¼1 j¼1 ! where, for the sake of consistency, we assume X[�], Y[�] are zero when their argument is negative. Finally, we assume that Y‘j½k� � �jX‘j½k (cid:0) wj�; ð7Þ ð8Þ i.e., if required (ϕj = 1), the units of second dose of the j-th type of vaccine injected at the k-th day must not trespass the units of first dose injected χj days before; otherwise (ϕj = 0), no sec- ond dose is considered. Notice that, being Eq (8) an inequality, the second dose is not manda- tory, and it is possible to implement policies where only a fraction of the population receiving the first dose receives also the second as suggested by the UK study SIREN [45]. Resulting SIR model. To conclude the section, let us show the expression of the discrete- time SIR model where the above interventions are explicitly considered. In particular, as a result of the social distancing intervention, matrix B is replaced by the matrix B[k] in Eq (6); moreover, in order to take into account the effect of vaccination, we assume Δr[k] is subtracted at each day k from the fraction of susceptible individuals, and is simultaneously added to the removed ones, without influencing the fraction of infectious individuals. We reiterate that the effect of the interventions at day k is mediated by the sampling time Δt; in other words, the dis- crete-time SIR model becomes 8 >>>< >>>: s½ðh þ 1ÞDt� ¼ s½hDt� (cid:0) Dts½hDt� � B½k�i½hDt� (cid:0) DtDr½k� i½ðh þ 1ÞDt� ¼ i½hDt� þ Dts½ðh þ 1ÞDt� � B½k�i½hDt� (cid:0) Dtg � i½ðh þ 1ÞDt� ð9Þ r½ðh þ 1ÞDt� ¼ r½hDt� þ Dtg � i½k� þ DtDr½k� or, in a compact form z½ðh þ 1ÞDt� ¼ f ðz½hDt�; Dr½k�; E½k�Þ; k ¼ bhDtc; ð10Þ where z[�] = [sT[�], iT[�], rT[�]]T. Optimization formulation The above SIR model with explicit intervention terms is the natural cornerstone for the plan- ning of such interventions. In particular, we assume a finite-time horizon of kmax days and we consider a scenario where at the 0-th day the epidemics is described by given initial conditions s[0], i[0], r[0] 2 [0, 1]n with s[0] + i[0] + r[0] = 1n. The aim of the proposed formulation is to plan the different interventions to be put in place to guarantee the reach of the end of the epidemics on day kmax, i.e., we want to enforce dynam- ical constraints that represent the evolution of the proposed variation of the SIR model (Eq (9), with B[k] and Δri[k] defined as in Eqs (6) and (7), respectively), together with the requirement that the SIR model reaches the herd immunity surface. The latter requirement is equivalent to enforcing a constraint in the form s½kmax� þ r½kmax� ¼ 1n; ð11Þ ensuring that an end-of-epidemic state is reached; at the same time, we want to guarantee that PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 7 / 21 PLOS ONE Dynamical intervention planning against COVID-19-like epidemics new infections die out. This requirement, as discussed above, is equivalent to enforcing a con- straint in the form of Eq (2). Notably, B[k] is time varying; however, when the final planning instant kmax is reached, it is reasonable to assume that non-pharmaceutical interventions are discontinued and E[kmax] = 0n × n; thus, the conditions for avoiding epidemic overburst is Rs½kmax� � 1n: ð12Þ Let us now discuss the choice variables of the proposed model; specifically, the model aims at identifying the units X[k]�0n × n and Y[k] � 0n × n of first and second doses of vaccine to be injected on the k-th day and the intensity of the social distancing measures E[k] 2 [0, e]n × n on the k-th day, for all k 2 {1, . . ., kmax}. Notice that, as discussed above, the latter variables must satisfy E½k� ¼ ET½k�; 8k � kmax: ð13Þ Let us now focus on aspects related to vaccination. In order to plan for such intervention, we consider a situation where vaccines become available in batches. Specifically, we assume there are specific days k1; . . . ; kwmax q½k� 2 Rm to denote the vector collecting the total units of vaccines received as of day k for each type of vaccine. in which batches of vaccines are received, and we use In order to guarantee that the vaccination plan is sound, we need to impose that the cumu- lative units of vaccine that are injected as of day k do not trespass the received ones, for each type, i.e., Xk ðX½h� þ Y½h�ÞT1n � q½k�; 8k: h¼0 ð14Þ Notice that, in order to guarantee that second doses do not trespass the first ones, we con- sider the constraint in Eq (8); moreover, to guarantee that the overall amount of doses does not exceed the population in each class, we consider a constraint in the form Xkmax k¼1 ðX½k� þ Y½k�Þ1m � n: ð15Þ Finally, let us assume that a maximum overall number lh of daily inpatient beds, lsh of which being intensive care inpatient beds, are available. In this view, in order to enforce that the amount of hospitalizations and intensive care hospitalizations do not overcome the limits, we consider constraints in the form and sT h diag ½n�i½k� � lh; 8k � kmax sT shdiag ½n�i½k� � lsh; 8k � kmax; where the vectors σh and σsh collect the hospitalization and severe hospitalization rates for each class, respectively, and diag[n]i[k] is the vector collecting the population of infectious individu- als in each class. Let us now discuss the objective function of the proposed formulation. In particular, we aim to minimize the cumulative intensity of the the social distancing measures over the PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 8 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics considered time horizon, i.e., Xkmax Xn Xn E‘j½k�: k¼0 ‘¼1 j¼1 Notice that, within any optimization formulation, minimizing the objective function is sec- ondary to constraint satisfaction. Therefore, within the proposed formulation, reaching of the herd immunity and avoiding the collapse of the healthcare system represent a priority with respect to the minimization of the overall intensity of the social intervention. In other words, solutions that have small objective function value but violate the constraints will be deemed unfeasible and will be discarded by any solver. Overall, the proposed formulation consists of the following Mixed Integer Nonlinear Programming (MINLP) problem. min Xkmax Xn Xn E‘j½k� k¼0 ‘¼1 j¼1 subject to 8 z½ðh þ 1ÞDt� ¼ f ðz½hDt�; Dr½k�; E½k�Þ; 8h � kmax=Dt ðIÞ ðIIÞ ðIIIÞ ðIVÞ ðVÞ ðVIÞ ðVIIÞ ðVIIIÞ ðIXÞ ðXÞ ðXIÞ ðXIIÞ ðXIIIÞ ðXIVÞ ðXVÞ >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>< >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>: ; 8‘ � n; k � kmax ð16Þ B½k� ¼ B � ð1n�n (cid:0) E½k�Þ; 8k � kmax Dr‘½k� ¼ Pm j¼1 X‘j½k (cid:0) tI Pm j �Zj þ N‘ j¼1 Y‘j½k (cid:0) tII j �DZj s½kmax� þ r½kmax� ¼ 1n; Rs½kmax� � 1n; Pk h¼0 ðX½h� þ Y½h�ÞT1n � q½k�; 8k � kmax Pkmax k¼0 ðX½k� þ Y½k�Þ1m � n; Y‘j½k� � �jX‘j½k (cid:0) wj�; 8k � kmax; ‘ � n; j � m E½k� ¼ ET½k�; 8k � kmax h diag ½n�i½k� � lh; 8k � kmax sT shdiag ½n�i½k� � lsh; 8k � kmax sT r½k�; s½k�; i½k� 2 ½0; 1�n; 8k � kmax E½k� 2 ½0; e�n�n; 8k � kmax X½k�; Y½k� 2 Rn�m �0 ; 8k � kmax X½k�; Y½k� integer ; 8k � kmax: In other words, constraints (I)–(III) model the requirement that the fraction of susceptible, infectious and removed individuals evolve according to the proposed SIR model accounting for the interventions in terms of social distancing and vaccination. Constraint (IV) accounts for reaching an end-of-epidemic state, while Constraint (V) guarantees that new infections die out. Constraint (VI) and (VII) guarantee that the amount of used doses of vaccine do not PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 9 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics overcome the available ones or the overall population, respectively. Constraint (VIII) enforces that the second doses (if required) are injected after the adequate time window. Constraint (IX) prescribes that E(k) is symmetric, thus implying that the social distancing effort reducing the influence of the i-th class on the j-th one has a specular effect on the influence of the j-th class on the i-th one. Constraints (X) and (XI) guarantee that the regular and intensive care hospitalizations do not overcome the maximum limit. Finally, constraints (XII)–(XV) guaran- tee the well-posedness of the variables considered in the formulation. Approximation strategy. Notice that the proposed formulation amounts to a Mixed Integer Nonlinear Programming problem. In particular, we observe that the model requires a nontrivial amount of variables and constraints (i.e., O(max{n/Δt, n2, nm}) for each day of plan- ning. Moreover, we observe that the problem is nonconvex (i.e., considering the nonlinear equality constraints corresponding to the SIR model as two inequality constraints, there is no way both are convex). However, since the units of vaccines involved in the planning are expected to be large, it makes sense to attempt to reduce complexity by considering a continu- ous relaxation, i.e., dropping integrity constraints. However, also in the case of a convex objec- tive function and a continuous relaxation, the problem has high chances to be NP-Hard (e.g., see [46, 47]), thus calling for approximated solutions. In this paper, our strategy to calculate an approximated solution is to resort to an approxi- mated solver. In fact, we observe that s[�], i[�], r[�], Δr[�] are actually functions of the variables E[k], X[k], Y[k], even though it is nontrivial to express this dependency in a closed form. Therefore, our strategy is to consider only the variables E[k], X[k], Y[k] and to express the con- straints and the objective function in an algorithmic way, resorting to an approximated solver. Specifically, we use the MIDACO optimization software which implements an extension of the evolutionary Ant Colony Optimization meta-heuristic [48] and which has been developed especially for highly non-linear real-world applications. See [49, 50] for a focus of the perfor- mance of MIDACO software with respect to the state of the art. Note that the suggested strategy is independent of a particular solver, but the non-convex nature of the optimization problem suggests an evolutionary approach, like genetic algorithms [51]. Furthermore, the dimensionality of the resulting MINLP in the next case study is very large-scale, consisting of 76650 decision variables and 18295 constraints, and therefore requires a solver that can handle such dimensionality. Finally, we point out that, since in our implementation we chose to evaluate the variables s[�], i[�], r[�], Δr[�] as a function of the vari- ables E[k], X[k], Y[k], a positive consequence is that the step size Δt used to discretize the SIR model has no effect on the overall number of choice variables, which is one of the major sources of complexity for the solution of MINLP formulations (e.g., see [52]). Computational setting The optimization with MIDACO was conducted on an Intel1Xeon1CPU E7 2860 @ 2.27GHz. The CPU runtime for the optimization was fixed to five days. All MIDACO parame- ters were used by their default values, that means that a feasiblity accuracy of 0.001 was used for all individual constraints listed in Eq (16). Results In this section, we test the effectiveness of the proposed formulation by considering a case study with realistic parameters consistent with the current COVID-19 pandemics and relative vaccines. Specifically, we focus on Italy and we identify the optimal vaccination policies over a one-year time horizon, considering 15 age classes (see Table 1) and three types of vaccines. Specifically, Table 2 reports the information regarding the efficacy ηj, the delay required to PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 10 / 21 PLOS ONETable 1. Population in the different age classes (Source: [60]). COVID-19 hospitalization, severe hospitalization and death rates as of April 2021 (source: [59]). Dynamical intervention planning against COVID-19-like epidemics Age Class Population Hospitalization Severe Hospitalization 00–04 05–09 10–14 15–19 20–24 25–29 30–34 35–39 40–44 45–49 50–54 55–59 60–64 65–69 70+ 2645566 2769974 2932459 2968742 3041263 3281737 3531873 3877837 4387315 5060898 5068741 4869741 4102571 3554615 10297032 https://doi.org/10.1371/journal.pone.0269830.t001 10.7% 10.7% 5.81% 5.81% 6.28% 6.28% 8.84% 8.84% 11.97% 11.97% 16.86% 16.86% 27.33% 27.33% 34.70% 0.31% 0.31% 0.23% 0.23% 0.32% 0.32% 0.77% 0.77% 1.91% 1.91% 3.59% 3.59% 6.79% 6.79% 3.27% Death 0.10% 0.10% 0.10% 0.10% 0.10% 0.10% 0.14% 0.14% 0.26% 0.26% 0.57% 0.57% 2.73% 2.73% 14.80% j , the delay between doses χj, the efficacy of the second appreciate the effect of the first dose tI dose Δηj and the delay required to appreciate the effect of the second dose tII cines considered mimic real vaccines, and the parameters are estimates based on data in [53– 55]. Notably, we assume that the three considered types of vaccine are available only in batches, at specific time instants and in limited amount for each batch, as summarized in Table 3. For simplicity, it is assumed that at regime batches reach between six million and nine million of doses per trimester; such figures are consistent with what has been planned and deployed in Italy [56]. j . The fictional vac- Moreover, we consider a scenario where only a small fraction (i.e., 0.01%) of the age class in thee range 35 − 39 years is initially infected and we assume the maximum daily inpatient beds are lh = 1000, while the maximum daily intensive care inpatient beds are lsh = 100. Notice that, for the sake of simplicity, we allow vaccination for all age groups, even though Italian regulation does not yet allow COVID-19 vaccination under the age of 5. Parameter tuning In order to tune our formulation, we consider the country contact matrix K (see Fig 1), as esti- mated in [57] for Italy. In particular, only physical contacts have been considered. Notice that the element Kij of a contact matrix from [57] can be considered proportional to the probability that an individual in the i-th age class meets an individual in the j-th; thus, B = Λ � K where Table 2. Efficacy of the vaccines considered in the proposed case study. Vaccine A mimics BNT162b2 (Pfizer & BioNTech); vaccine B mRNA-1273 (Moderna) and vaccine C ChAdOx1 nCoV-2019 (University of Oxford/AstraZe- neca). The source for the estimates are: [53–55]. Delay for effect (1st dose) tI Efficacy (1st dose) ηj j [days] Delay between doses χj [days] Efficacy (2nd dose) Δηj j [days] Delay for effect (2nd dose) tII Vaccine A Vaccine B Vaccine C 0.89 15 28 0.06 15 0.89 15 28 0.05 15 0.70 15 84 0.20 15 https://doi.org/10.1371/journal.pone.0269830.t002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 11 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics Table 3. Units of Vaccines of each type that are assumed to become available in batches at specific days. Day 0 30 60 90 120 150 180 210 240 270 300 330 Vaccine A Vaccine B Vaccine C 500000 500000 1000000 2000000 2000000 2000000 2000000 2000000 2000000 3000000 3000000 3000000 500000 500000 1000000 2000000 2000000 2000000 2000000 2000000 2000000 3000000 3000000 3000000 500000 500000 1000000 2000000 2000000 2000000 2000000 2000000 2000000 3000000 3000000 3000000 https://doi.org/10.1371/journal.pone.0269830.t003 Λij is the probability that a contact between i and j results in an infection. In this case study, we will use a constant Λij = λ; analogously, we will use a constant γ. To fix the parameters, we will consider a basic reproduction number (i.e., the potential number of new infected generated by one case) R0 = 3—a value that has been estimated for COVID19 in France [58]. Since for an heterogeneous compartmental model of the form of Eq (1) the role of the basic reproduction number is played by k R k [22, 43], we can rescale λ to obtain a basic reproduction number equivalent to the observed one: l ¼ gR0 k K k ; Fig 1. Elements of the matrix K of physical contacts among age classes in Italy (source: [57]). For the sake of readability, the colors of the cells corresponds to ln(Kij). https://doi.org/10.1371/journal.pone.0269830.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 12 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics i.e., k R k¼k l diag ðgÞ(cid:0) 1 diag ðnÞ(cid:0) 1K T di ag ðnÞ k¼ R0: ð17Þ The other parameters required to tune the proposed model are the rates of hospitalization, of severe hospitalization, and of death; such parameters can be found in the ECDC ninth risk assessment update for COVID-19 in the EU/EEA and the UK [59] and are reported in Table 1. Experimental results Let us now discuss the experimental results from the computational point of view. Fig 2 shows the results of MIDACO in terms of objective function value and overall violation of the con- straints, plotted against the number of candidate solutions evaluated by MIDACO. As shown by the figure, we observe that a feasible solution is obtained in about 6 × 106 evaluations. As for the objective function, we observe that while the solution is infeasible there is a relevant reduction over time; in particular, we reach a steady solution after about 9 × 106 evaluations. Overall, these results suggest the reach of a local minimum. Having discussed the computational aspects, let us now focus on the structure of the found solution. Figs 3–6 report the structure of the interventions encoded by the found solution. Specifi- cally, according to Fig 3, it can be noted that the the social distancing measures are initially quite intense, and only at the end of the planning horizon there is a partial reduction. Fig 4 shows how vaccine usage is distributed based on the type of vaccine. According to the figure, there is no noticeable difference; this is likely the effect of the scarcity of vaccines in our sce- nario. Moreover, Fig 5 (as well as Fig 6, where the same data is aggregated and smoothed to improve readability) shows that, in the early stages of the planning, due to the scarcity of vac- cines, there is a preference for vaccinating individuals in the age range 20–69 years over young and elderly people; notably, such an age range receives a more or less steady amount of vac- cines over time. This stems from the fact that, as discussed above, in the proposed formulation the objective of minimizing the intensity of the social distancing is secondary to constraint sat- isfaction, i.e., less restrictive social distancing measures can be considered only if they allow Fig 2. Objective function value and overall constraint violation for the solution found using MIDACO, plotted against the number of candidate solutions evaluated. https://doi.org/10.1371/journal.pone.0269830.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 13 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics Fig 3. Intervention plan corresponding to the found solution in terms of the intensity of social distancing measures plotted against time. https://doi.org/10.1371/journal.pone.0269830.g003 Fig 4. Intervention plan corresponding to the found solution in terms of the units of the different types administered for each day. https://doi.org/10.1371/journal.pone.0269830.g004 Fig 5. Intervention plan corresponding to the found solution in terms of the units of vaccines administered to the different age classes for each day. https://doi.org/10.1371/journal.pone.0269830.g005 PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 14 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics Fig 6. Intervention plan corresponding to the found solution in terms of the units of vaccines administered to the different age classes for each day. The plot aggregates the age classes into the young (0–19 years), middle-age (19–69 years) and elderly (�70 years) macro-classes. To improve readability, data has been smoothed using a 30-day moving average filter. https://doi.org/10.1371/journal.pone.0269830.g006 the reach the herd immunity and prevent the collapse of the healthcare system. In other words, candidate solutions where strict social distancing was released earlier than the identified solu- tion have been discarded by the solver due to some violation of the constraints. Fig 7 shows the evolution of the proposed SIR model accounting for the effect of social dis- tancing and vaccines, as a result of the interventions planned within the found solution. It can be noted that, due to the strict social distancing measures, only a small fraction of the popula- tion becomes infected, with a noticeable peak for the age class in the range 10 − 14 years in cor- respondence to the softening of the lockdown measures (i.e., around day k = 320). Notice that the fraction of susceptible individuals is slowly eroded due to vaccination, while the fraction of removed has a consequent slow growth due to the resulting immunization. Fig 7. Evolution of the proposed SIR model accounting for the effect of social distancing and vaccines, based on the found solution. The first, second and third row of plots correspond to the fraction of susceptible, infected and removed individuals, respectively, while the k-th column of plots corresponds to the k-th age class. https://doi.org/10.1371/journal.pone.0269830.g007 PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 15 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics Fig 8. Intensity of lockdown within the found solution for the different age classes and for selected days over the considered time horizon. The intensity is shown with a blue to yellow scale, where blue represents no social distancing and yellow a complete lockdown. https://doi.org/10.1371/journal.pone.0269830.g008 Fig 8 breaks down the social distancing measures by age class; it can be noted that most of the considered time horizon all age classes are strongly restrained in their interaction. Then, around the end of the planning horizon the lockdown is significantly lifted for the age class in the range 10–14 years (from which the peak in the infected fraction of this age class). Finally, Fig 9 shows the results of the planning in terms of deaths, hospitalization and inten- sive care occupancy. It can be noted that the solution found corresponds to a situation where the capacity in terms of regular and intensive care beds is not reached, thus avoiding the col- lapse of the healthcare system. Fig 9. Deaths, hospitalizations and intensive care occupancies corresponding to the found solutions, plotted for each day. https://doi.org/10.1371/journal.pone.0269830.g009 PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 16 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics Conclusion In this paper we develop a fine-grained model to support the plan for intervention in order to contrast newborn epidemics. The proposed approach is particularly suitable for infections like the ongoing COVID-19 epidemic, characterized by high infection rate and able to pose the healthcare system and society under stress in terms of intensive care occupancy, deaths or eco- nomic consequences. Moreover, the approach allows to plan interventions that blend large- scale non-pharmaceutical interventions along with the pharmaceutical ones. Specifically, we build up the planning on two two main types of intervention, namely, non-pharmaceutical (essentially social distancing measures) and vaccination. In order to model the effect of such interventions, we develop a variation of the SIR epidemics model with heterogeneous popula- tion; specifically, we assume social distancing intensity to reflect into a reduction of the infec- tion rates, while vaccination to have a partial and delayed immunization effect. Notably, we consider several population classes, several vaccines with different efficacy and with partial and delayed effect, the possibility of a second dose, the availability of vaccines in batches, the need of reaching the herd immunity and the requirement to avoid congestions in the health- care system. Interestingly, besides representing a detailed, day-to-day planning, the proposed approach also provides useful insights from the clinical and policy-making point of view. In fact, the plan identified by the proposed methodology suggests that, initially, the scarcity of vaccines should be faced by enforcing a strict social distancing, and that vaccination priority should be given to the elderly and “middle-age” population over the younger one. The pro- posed model exhibits a nontrivial degree of complexity, and the identification of efficient approximated ways to solve it represents a challenging task. Yet, the proposed model repre- sents a remarkably descriptive framework, and future work will be mainly devoted to incorpo- rate other important perspectives for policy and decision makers, such as geographical [61], economical [62] and logistic aspects [63], social equity in the vaccine distribution [64], and skepticism of the population towards vaccines [65]. A last envisaged research perspective is related to the time duration of the planning. In fact, due to the change of the overall epidemiological or pharmaceutical situation, a yearly time horizon could be deemed excessively long; yet, the reach of the herd immunity requires a suffi- ciently wide time frame. To this end, a viable future work direction is to adopt a “receding horizon” perspective [41, 66], where the model is updated after a given period of time (e.g., one month or three months) and a new planning is executed starting from the epidemiological situation at that time. In particular, as new evidence regarding the prevalence of new strains is gathered, the model could be updated (e.g., changing the basic reproduction number [67, 68], adding new compartments such as the fraction of asymptomatic individuals [69, 70], consider- ing re-infections [71], etc.). Moreover, as new vaccines are developed (or discontinued, as in the case of the Vaxzevria vaccine in Italy [72]) and their effectiveness is better assessed with respect to the circulating variants of the virus, the model can be updated accordingly (e.g., requirement of a booster dose, change in effectiveness, change in the time between doses, etc.). In other words, while the proposed methodology could be considered an open loop approach, we foresee its extension to a closed loop approach. This, of course, raises interesting research questions about the trade-off between the frequency of the update and the computational demands that will be addressed in future work. Supporting information S1 Data. (TXT) PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 17 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics Author Contributions Conceptualization: Gabriele Oliva, Antonio Scala. Data curation: Martin Schlueter, Antonio Scala. Investigation: Gabriele Oliva, Antonio Scala. Methodology: Gabriele Oliva, Antonio Scala. Software: Martin Schlueter. Supervision: Masaharu Munetomo. Validation: Martin Schlueter, Masaharu Munetomo, Antonio Scala. Visualization: Gabriele Oliva. Writing – original draft: Gabriele Oliva, Antonio Scala. Writing – review & editing: Martin Schlueter, Masaharu Munetomo, Antonio Scala. References 1. COVID-19 map—John Hopkins Univeristy;. https://coronavirus.jhu.edu/map.html. 2. 3. Fauci AS, Lane HC, Redfield RR. Covid-19—navigating the uncharted; 2020. Lau H, Khosrawipour V, Kocbach P, Mikolajczyk A, Schubert J, Bania J, et al. The positive impact of lockdown in Wuhan on containing the COVID-19 outbreak in China. Journal of travel medicine. 2020;. https://doi.org/10.1093/jtm/taaa037 PMID: 32181488 4. Prather KA, Wang CC, Schooley RT. Reducing transmission of SARS-CoV-2. Science. 2020; 368 (6498):1422–1424. https://doi.org/10.1126/science.abc6197 PMID: 32461212 5. Bazant MZ, Bush JW. A guideline to limit indoor airborne transmission of COVID-19. Proceedings of the National Academy of Sciences. 2021; 118(17). https://doi.org/10.1073/pnas.2018995118 6. Le TT, Andreadakis Z, Kumar A, Roma´n RG, Tollefsen S, Saville M, et al. The COVID-19 vaccine devel- opment landscape. Nat Rev Drug Discov. 2020; 19:305–306. https://doi.org/10.1038/d41573-020- 00073-5 7. Chukwudozie OS, Duru VC, Ndiribe CC, Aborode AT, Oyebanji VO, Emikpe BO. The relevance of bio- informatics applications in the discovery of vaccine candidates and potential drugs for COVID-19 treat- ment. Bioinformatics and Biology Insights. 2021; 15:11779322211002168. https://doi.org/10.1177/ 11779322211002168 PMID: 33795932 8. Cleve M. What the lightning-fast quest for Covid vaccines means for other diseases. Nature. 2021; 589:16–8. 9. Kim JH, Marks F, Clemens JD. Looking beyond COVID-19 vaccine phase 3 trials. Nature medicine. 2021; 27(2):205–211. https://doi.org/10.1038/s41591-021-01230-y PMID: 33469205 10. Umakanthan S, Chattu VK, Ranade AV, Das D, Basavarajegowda A, Bukelo M. A rapid review of recent advances in diagnosis, treatment and vaccination for COVID-19. AIMS Public Health. 2021; 8(1):137. https://doi.org/10.3934/publichealth.2021011 PMID: 33575413 11. Keshavarzi Arshadi A, Webb J, Salem M, Cruz E, Calad-Thomson S, Ghadirian N, et al. Artificial intelli- gence for COVID-19 drug discovery and vaccine development. Frontiers in Artificial Intelligence. 2020; 3:65. https://doi.org/10.3389/frai.2020.00065 12. Francis AI, Ghany S, Gilkes T, Umakanthan S. Review of COVID-19 vaccine subtypes, efficacy and geographical distributions. Postgraduate Medical Journal. 2022; 98(1159):389–394. https://doi.org/10. 1136/postgradmedj-2021-140654 PMID: 34362856 13. Beigel JH, Tomashek KM, Dodd LE, Mehta AK, Zingman BS, Kalil AC, et al. Remdesivir for the treat- ment of Covid-19—preliminary report. New England Journal of Medicine. 2020; 383(19):1813–1836. https://doi.org/10.1056/NEJMoa2007764 PMID: 32445440 14. Norrie JD. Remdesivir for COVID-19: challenges of underpowered studies. The Lancet. 2020; 395 (10236):1525–1527. https://doi.org/10.1016/S0140-6736(20)31023-0 PMID: 32423580 15. Smith T, Killeen GF, Maire N, Ross A, Molineaux L, Tediosi F, et al. Mathematical modeling of the impact of malaria vaccines on the clinical epidemiology and natural history of Plasmodium falciparum PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 18 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics malaria: Overview. The American journal of tropical medicine and hygiene. 2006; 75(2_suppl):1–10. https://doi.org/10.4269/ajtmh.2006.75.2_suppl.0750001 PMID: 16931810 16. OpenMalaria Software; 2008. https://github.com/SwissTPH/openmalaria. 17. Van der Ploeg CP, Van Vliet C, De Vlas SJ, Ndinya-Achola JO, Fransen L, Van Oortmarssen GJ, et al. STDSIM: a microsimulation model for decision support in STD control. Interfaces. 1998; 28(3):84–100. https://doi.org/10.1287/inte.28.3.84 18. van Vliet C, Meester EI, Korenromp EL, Singer B, Bakker R, Habbema JDF. Focusing strategies of con- dom use against HIV in different behavioural settings: an evaluation based on a simulation model. Bulle- tin of the World Health Organization. 2001; 79:442–454. PMID: 11417040 19. Stuckey EM, Stevenson JC, Cooke MK, Owaga C, Marube E, Oando G, et al. Simulation of malaria epi- demiology and control in the highlands of western Kenya. Malaria journal. 2012; 11(1):1–14. https://doi. org/10.1186/1475-2875-11-357 PMID: 23107070 20. Steen R, Hontelez JA, Veraart A, White RG, de Vlas SJ. Looking upstream to prevent HIV transmission: can interventions with sex workers alter the course of HIV epidemics in Africa as they did in Asia? Aids. 2014; 28(6):891–899. https://doi.org/10.1097/QAD.0000000000000176 PMID: 24401648 21. Kermack WO, McKendrick AG. A contribution to the mathematical theory of epidemics. Proceedings of the royal society of london Series A, Containing papers of a mathematical and physical character. 1927; 115(772):700–721. 22. Diekmann O, Heesterbeek H, Britton T. Mathematical tools for understanding infectious disease dynamics. In: Mathematical Tools for Understanding Infectious Disease Dynamics. Princeton Univer- sity Press; 2012. 23. Chen Q. Application of SIR model in forecasting and analyzing for SARS. Beijing da xue xue bao Yi xue ban = Journal of Peking University Health sciences. 2003; 35:75–80. PMID: 12914225 24. Laguzet L, Turinici G. Individual vaccination as Nash equilibrium in a SIR model with application to the 2009–2010 influenza A (H1N1) epidemic in France. Bulletin of Mathematical Biology. 2015; 77 (10):1955–1984. https://doi.org/10.1007/s11538-015-0111-7 25. Bjørnstad ON, Finkensta¨dt BF, Grenfell BT. Dynamics of measles epidemics: estimating scaling of transmission rates using a time series SIR model. Ecological monographs. 2002; 72(2):169–184. https://doi.org/10.1890/0012-9615(2002)072%5B0169:DOMEES%5D2.0.CO;2 26. Ku¨ hnert D, Stadler T, Vaughan TG, Drummond AJ. Simultaneous reconstruction of evolutionary history and epidemiological dynamics from viral sequences with the birth–death SIR model. Journal of the Royal Society Interface. 2014; 11(94):20131106. https://doi.org/10.1098/rsif.2013.1106 PMID: 24573331 27. Kretzschmar M, Wallinga J. Mathematical models in infectious disease epidemiology. In: Modern infec- tious disease epidemiology. Springer; 2009. p. 209–221. 28. Scala A, Flori A, Spelta A, Brugnoli E, Cinelli M, Quattrociocchi W, et al. Time, space and social interac- tions: exit mechanisms for the Covid-19 epidemics. Scientific reports. 2020; 10(1):1–12. https://doi.org/ 10.1038/s41598-020-70631-9 29. Behncke H. Optimal control of deterministic epidemics. Optimal control applications and methods. 2000; 21(6):269–285. https://doi.org/10.1002/oca.678 30. Nowzari C, Preciado VM, Pappas GJ. Analysis and control of epidemics: A survey of spreading pro- cesses on complex networks. IEEE Control Systems Magazine. 2016; 36(1):26–46. https://doi.org/10. 1109/MCS.2015.2495000 31. Liu F, Buss M. Optimal control for heterogeneous node-based information epidemics over social net- works. IEEE Transactions on Control of Network Systems. 2020; 7(3):1115–1126. https://doi.org/10. 1109/TCNS.2019.2963488 32. Duijzer LE, van Jaarsveld WL, Wallinga J, Dekker R. Dose-optimal vaccine allocation over multiple pop- ulations. Production and Operations Management. 2018; 27(1):143–159. https://doi.org/10.1111/poms. 12788 PMID: 32327917 33. Bubar KM, Reinholt K, Kissler SM, Lipsitch M, Cobey S, Grad YH, et al. Model-informed COVID-19 vac- cine prioritization strategies by age and serostatus. Science. 2021; 371(6532):916–921. https://doi.org/ 10.1126/science.abe6959 PMID: 33479118 34. Enayati S, O¨ zaltın OY. Optimal influenza vaccine distribution with equity. European Journal of Opera- tional Research. 2020; 283(2):714–725. https://doi.org/10.1016/j.ejor.2019.11.025 35. Scala A, Oliva G. Optimal vaccination based on simple, yet effective, linear constraints and vaccines with different effectiveness. Socio-Economic Planning Sciences. Submitted;. 36. Buckner JH, Chowell G, Springborn MR. Dynamic prioritization of COVID-19 vaccines when social dis- tancing is limited for essential workers. Proceedings of the National Academy of Sciences. 2021; 118 (16). https://doi.org/10.1073/pnas.2025786118 PMID: 33811185 PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 19 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics 37. Matrajt L, Eaton J, Leung T, Brown ER. Vaccine optimization for COVID-19: Who to vaccinate first? Sci- ence Advances. 2021; 7(6):eabf1374. https://doi.org/10.1126/sciadv.abf1374 PMID: 33536223 38. Sinha P, Kumar S, Chandra C. Strategies for ensuring required service level for COVID-19 herd immu- nity in Indian vaccine supply chain. European journal of operational research. 2021;. 39. Nagurney A. Supply chain game theory network modeling under labor constraints: Applications to the Covid-19 pandemic. European Journal of Operational Research. 2021; 293(3):880–891. https://doi.org/ 10.1016/j.ejor.2020.12.054 40. Moore S, Hill EM, Tildesley MJ, Dyson L, Keeling MJ. Vaccination and non-pharmaceutical interven- tions for COVID-19: a mathematical modelling study. The Lancet Infectious Diseases. 2021; 21(6):793– 802. https://doi.org/10.1016/S1473-3099(21)00143-2 PMID: 33743847 41. Parino F, Zino L, Calafiore GC, Rizzo A. A model predictive control approach to optimally devise a two- dose vaccination rollout: A case study on COVID-19 in Italy. International journal of robust and nonlinear control. 2021;. https://doi.org/10.1002/rnc.5728 PMID: 34908815 42. Burer S, Letchford AN. Non-convex mixed-integer nonlinear programming: A survey. Surveys in Opera- tions Research and Management Science. 2012; 17(2):97–106. https://doi.org/10.1016/j.sorms.2012. 08.001 43. Brauer F, Castillo-Chavez C, Castillo-Chavez C. Mathematical models in population biology and epide- miology. vol. 2. Springer; 2012. 44. Oliva G, Bonfigli S, Scala A. The geometry of herd immunity. In preparation;. 45. Hall VJ, Foulkes S, Charlett A, Atti A, Monk EJ, Simmons R, et al. SARS-CoV-2 infection rates of anti- body-positive compared with antibody-negative health-care workers in England: a large, multicentre, prospective cohort study (SIREN). The Lancet. 2021; 397(10283):1459–1469. https://doi.org/10.1016/ S0140-6736(21)00675-9 46. Sherali HD, Adams WP. A reformulation-linearization technique for solving discrete and continuous nonconvex problems. vol. 31. Springer Science & Business Media; 2013. 47. Tawarmalani M, Sahinidis NV. Convexification and global optimization in continuous and mixed-integer nonlinear programming: theory, algorithms, software, and applications. vol. 65. Springer Science & Business Media; 2013. 48. Dorigo M, Stu¨ tzle T. Ant colony optimization: overview and recent advances. Handbook of metaheuris- tics. 2019; p. 311–351. https://doi.org/10.1007/978-3-319-91086-4_10 49. Schlu¨ ter M, Gerdts M, Ru¨ ckmann JJ. A numerical study of MIDACO on 100 MINLP benchmarks. Opti- mization. 2012; 61(7):873–900. https://doi.org/10.1080/02331934.2012.668545 50. Schlueter M, Erb SO, Gerdts M, Kemble S, Ru¨ ckmann JJ. MIDACO on MINLP space applications. Advances in Space Research. 2013; 51(7):1116–1131. https://doi.org/10.1016/j.asr.2012.11.006 51. Deb K, Pratap A, Agarwal S, Meyarivan T. A fast and elitist multiobjective genetic algorithm: NSGA-II. IEEE transactions on evolutionary computation. 2002; 6(2):182–197. https://doi.org/10.1109/4235. 996017 52. Kronqvist J, Bernal DE, Lundell A, Grossmann IE. A review and comparison of solvers for convex MINLP. Optimization and Engineering. 2019; 20(2):397–455. https://doi.org/10.1007/s11081-018- 9411-8 53. Polack FP, Thomas SJ, Kitchin N, Absalon J, Gurtman A, Lockhart S, et al. Safety and efficacy of the BNT162b2 mRNA Covid-19 vaccine. New England Journal of Medicine. 2020;. https://doi.org/10.1056/ NEJMoa2034577 PMID: 33301246 54. Baden LR, El Sahly HM, Essink B, Kotloff K, Frey S, Novak R, et al. Efficacy and safety of the mRNA- 1273 SARS-CoV-2 vaccine. New England journal of medicine. 2021; 384(5):403–416. https://doi.org/ 10.1056/NEJMoa2035389 PMID: 33378609 55. Knoll MD, Wonodi C. Oxford–AstraZeneca COVID-19 vaccine efficacy. The Lancet. 2021; 397 (10269):72–74. https://doi.org/10.1016/S0140-6736(20)32623-4 56. OpenData on Italian Vaccines;. https://github.com/italia/covid19-opendata-vaccini. 57. Mossong J, Hens N, Jit M, Beutels P, Auranen K, Mikolajczyk R, et al. Supporting materials for Social contacts and mixing patterns relevant to the spread of infectious diseases. PLoS medicine. 2008; 5(3): e74. 58. Salje H, Tran Kiem C, Lefrancq N, Courtejoie N, Bosetti P, Paireau J, et al. Estimating the burden of SARS-CoV-2 in France. Science. 2020; 369(6500):208–211. https://doi.org/10.1126/science.abc3517 59. Rapid Risk Assessment: Coronavirus disease 2019 (COVID-19) in the EU/EEA and the UK– ninth update;. https://bit.ly/39Kz9FO 60. US Census Bureau;. https://www.census.gov/. PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 20 / 21 PLOS ONEDynamical intervention planning against COVID-19-like epidemics 61. Galeazzi A, Cinelli M, Bonaccorsi G, Pierri F, Schmidt AL, Scala A, et al. Human mobility in response to COVID-19 in France, Italy and UK. Scientific reports. 2021; 11(1):1–10. https://doi.org/10.1038/s41598- 021-92399-2 PMID: 34162933 62. Bonaccorsi G, Pierri F, Cinelli M, Flori A, Galeazzi A, Porcelli F, et al. Economic and social conse- quences of human mobility restrictions under COVID-19. Proceedings of the National Academy of Sci- ences. 2020; 117(27):15530–15535. https://doi.org/10.1073/pnas.2007658117 63. Organization WH, et al. COVID-19 vaccination: supply and logistics guidance: interim guidance, 12 February 2021. World Health Organization; 2021. 64. Wu JH, John SD, Adashi EY. Allocating vaccines in a pandemic: the ethical dimension. The American Journal of Medicine. 2020; 133(11):1241–1242. https://doi.org/10.1016/j.amjmed.2020.06.007 PMID: 32653419 65. Briand SC, Cinelli M, Nguyen T, Lewis R, Prybylski D, Valensise CM, et al. Infodemics: A new challenge for public health. Cell. 2021; 184(25):6010–6014. https://doi.org/10.1016/j.cell.2021.10.031 PMID: 34890548 66. Agachi PS, Cristea MV, Csavdari AA, Szilagyi B. 2. Model predictive control. In: Advanced Process Engineering Control. De Gruyter; 2016. p. 32–74. 67. 68. Ito K, Piantham C, Nishiura H. Relative instantaneous reproduction number of Omicron SARS-CoV-2 variant with respect to the Delta variant in Denmark. Journal of medical virology. 2022; 94(5):2265– 2268. https://doi.org/10.1002/jmv.27560 PMID: 34967453 Ito K, Piantham C, Nishiura H. Predicted dominance of variant Delta of SARS-CoV-2 before Tokyo olympic games, Japan, July 2021. Eurosurveillance. 2021; 26(27):2100570. https://doi.org/10.2807/ 1560-7917.ES.2021.26.27.2100570 69. Di Giamberardino P, Iacoviello D, Papa F, Sinisgalli C. Dynamical evolution of COVID-19 in Italy with an evaluation of the size of the asymptomatic infective population. IEEE Journal of Biomedical and Health Informatics. 2020; 25(4):1326–1332. https://doi.org/10.1109/JBHI.2020.3009038 70. Kheifetz Y, Kirsten H, Scholz M. On the parametrization of epidemiologic models–lessons from model- ling COVID-19 epidemic. arXiv preprint arXiv:210911916. 2021;. 71. Wilhelm A, Widera M, Grikscheit K, Toptan T, Schenk B, Pallas C, et al. Reduced neutralization of SARS-CoV-2 omicron variant by vaccine sera and monoclonal antibodies. MedRxiv. 2021;. 72. European Centre for Disease Prevention EC, Control. Overview of EU/EEA country recommendations on COVID-19 vaccination with Vaxzevria, and a scoping review of evidence to guide decision-making. 2021. PLOS ONE \| https://doi.org/10.1371/journal.pone.0269830 June 14, 2022 21 / 21 PLOS ONE
10.1371_journal.pone.0253540	RESEARCH ARTICLE Amelioration of 5-fluorouracil-induced intestinal mucositis by Streptococcus thermophilus ST4 in a mouse model Siou-Ru Shen1,2, Wei-Jen Chen2,3, Hui-Fang Chu2, Shiuan-Huei Wu2, Yu-Ru Wang1,4, Tang-Long ShenID 1,4* 1 Center for Biotechnology, National Taiwan University, Taipei, Taiwan, 2 Syngen Biotech Co., Ltd., Tainan, Taiwan, 3 Graduate Institute of Management, Minghsin University of Science and Technology, Hsinchu, Taiwan, 4 Department of Plant Pathology and Microbiology, National Taiwan University, Taipei, Taiwan * shentl@ntu.edu.tw Abstract Intestinal mucositis is a commonly encountered toxic side effect in patients undergoing 5-fluorouracil (5-FU)-based chemotherapy. Numerous studies have shown that probiotics enable improving chemotherapy-induced intestinal mucositis, but the beneficial effects of probiotics differ depending on the strain. Therefore, in the present studies we suggest that S. thermophilus ST4 separated from raw milk may assess mucoprotective activity in 5-FU-induced intestinal mucositis. In our causal-comparative study design, fifteen mice were randomized assigned into three groups (n = 5/each group): control group, 5-FU group and 5-FU+S. thermophilus ST4 group. The control group was orally administrated saline only, and the 5-FU group was followed by intraperitoneal injection of 5-FU for 3 days after 10-day saline administration, and the 5-FU+S. thermophilus ST4 group was intragastrically subjected for S. thermophilus ST4 once per day during the whole experi- ment, starting from the first day of the experiment, followed by 5-FU intraperitoneal injec- tion for 3 days after 10-day S. thermophilus ST4 pretreatment. Diarrhea score, pro- inflammatory cytokines serum levels, intestinal histopathology and short chain fatty acid were assessed. Here, we demonstrated the beneficial effects of S. thermophilus ST4 derived from raw milk against 5-FU-induced intestinal mucositis, including body weight reduction, appetite loss and diarrhea. Intrinsically, S. thermophilus ST4 effectively main- tained epithelium structure in small intestines and colons as well as reduced the intestinal inflammation. Besides, S. thermophilus ST4 significantly increased the expression of acetic acid, reinforcing the muco-protective effects. In conclusion, our results demon- strate that S. thermophilus ST4 supplementation ameliorates 5-FU-induced intestinal mucositis. This suggests probiotic may serve as an alternative therapeutic strategy for the prevention or management of 5-FU-induced mucositis in the future. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Shen S-R, Chen W-J, Chu H-F, Wu S-H, Wang Y-R, Shen T-L (2021) Amelioration of 5- fluorouracil-induced intestinal mucositis by Streptococcus thermophilus ST4 in a mouse model. PLoS ONE 16(7): e0253540. https://doi. org/10.1371/journal.pone.0253540 Editor: Young Hoon Jung, Kyungpook National University, REPUBLIC OF KOREA Received: January 12, 2021 Accepted: June 7, 2021 Published: July 26, 2021 Copyright: © 2021 Shen et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. Funding: This study was partially supported by Ministry of Science and Technology (109-2321-B- 002-005) to T-LS. The funder had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. This study also received funding from Syngen Biotech Co., Ltd. In the form of salaries to S-RS, W-JC, H- FC, and S-HW. The specific roles of these authors PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 1 / 15 PLOS ONEare articulated in the ‘author contributions’ section. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors declare that this study received funding from Syngen Ministry of Science and Technology. Syngen Biotech Co., Ltd. Provides conceptualization, experimental materials, and data analysis. The specific roles of these authors are articulated in the ‘author contributions’ section. SRS, WJC, HFC, and SHW are employees of Syngen Biotech Co., Ltd. that markets Streptococcus thermophilus ST4 products and holds patents and trademarks, other authors declare that there are no conflicts of interest. This does not alter our adherence to PLOS ONE policies on sharing data and materials. Abbreviations: 5-FU, 5-fluorouracil; TNF-α, tumor necrosis factor-α; IL-1β, interleukin-1β; IL-6, interleukin-6; SCFAs, short-chain fatty acids; DSS, dextran sulfate sodium; IBD, inflammatory bowel disease; ROS, reactive oxygen species; GPR43, G- protein-coupled receptor 43. Streptococcus thermophilus protects intestinal mucositis Introduction 5-FU is widely used for treatments of a range of cancers, including colorectal cancer, pancre- atic cancer and breast cancers, whereas it frequently causes intestinal mucositis. Intestinal mucositis (mucosal barrier injury) characterized by a decrease in villi length and the disrup- tion of crypt cell homeostasis is attributed to a common toxic side effect of 5-FU [1]. This side effect causes severe diarrhea, malabsorption, morphological mucosal damage and severe infec- tion, which limits the safety and clinical applications using 5-FU as a chemotherapy. In fact, villus blunting and disrupted crypts are often seen in the small intestine upon chemotherapy due primarily to an upregulation in apoptosis and a downregulate in proliferation [2]. Simi- larly, in the colon, 5-FU administration significantly confers shortening the colon length, pre- sumably the shortened colon is closely associated with severe diarrhea. [3]. Moreover, 5-FU- induced intestinal mucositis increases the production of pro-inflammatory cytokines, such as tumor necrosis factor-α (TNF-α), interleukin-1β (IL-1β), and interleukin-6 (IL-6), which are the hallmarks of mucositis inflammation [3, 4] and responsible for initiating inflammation in response to tissue injury and infection during chemotherapy. The suppression of inflamma- tion and efficient healing abilities of the mucosa are beneficial for the maintenance of homeo- stasis in response to gut damage. In fact, upon an increase in mucosal repaired capability, certain cytokines involved in the repaired process had been shown to diminish the severity of intestinal mucositis both in animal models and clinical trials [5, 6]. Therefore, the strategic intervention used to block inflammatory processes or to maintain gut homeostasis are of great beneficial for 5-FU-induced intestinal mucositis. The term probiotics is defined as “live micro-organisms which, when administered in ade- quate amounts, confer a health benefit on the host” [7]. They are commonly existed in fer- mented milks, yogurts and cheese or dietary supplements usually in the dehydrated form [8]. Recently, accumulative evidence supports that probiotic supplements beneficial effects for human and animal health, especially in improvements of intestinal functionalities and preven- tion of inflammation intestinal diseases [9]. Eventually, probiotics is known to exhibit anti- inflammatory effects through increases in the production of short-chain fatty acids (SCFAs), mainly acetate, propionate, and butyrate, stemming from carbohydrates, fibers, and polyphe- nols fermented by gut microbiota [10, 11]. Functionally, in the presence of SCFAs, a range of positive effects have been demonstrated, including maintenance of the normal structure, integ- rity and function of the intestine [12], modulation of the colonic and intracellular environment [13], and fuel for the intestinal epithelial cells, promotion of colonic epithelial cells prolifera- tion and gene expression [14, 15]. Streptococcus thermophilus is a gram-positive, lactic acid production, and ovoid-shape bacte- rium appearing in pairs or in short chains. Taking advantage of its lactose digestion activity, S. thermophilus has been utilized to improve individuals with lactose intolerance [16] in addition to other activities such as antioxidation [17], immune modulation [18], gastrointestinal epithe- lium homeostasis [19], prevention of chronic gastritis [20], attenuation of diarrhea [21, 22], alle- viation of ulcer and inflammation [23] and so on. Some studies have shown that the administration of S. thermophilus strain can reduced some parameters of mucositis in animal model induced by chemotherapy, such as prevent weight loss, attenuate the diarrhea and intesti- nal damage [24, 25]. Although S. thermophilus has been granted as a good probiotic in varied intestinal inflammatory models, for example, alleviation of colitis symptoms in a dextran sulfate sodium (DSS) model [26, 27], the beneficial effects of probiotics S. thermophilus differ in origins and/or strains. In the current study, we evaluated the mucoprotective activity of S. thermophilus ST4 in a 5-FU-induced intestinal mucositis. Our results warrant the development of probiotic supplements for chemotherapeutic side effects in related to gastrointestinal mucositis. PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 2 / 15 PLOS ONEStreptococcus thermophilus protects intestinal mucositis Materials and methods Preparation of 5-FU and S. thermophilus ST4 5-FU was purchased from Sigma (St. Louis, MO, USA). The preparation of 5-FU was firstly dissolved in saline at a concentration of 5 mg/mL, and then sterile filtered through a 0.2 μm syringe filter. 5-FU was injected intraperitoneally at a single dose of 50 mg/kg/day for 3 conse- cutive days to cause intestine mucositis. S. thermophilus ST4 was isolated in raw milk and pro- vided by Syngen Bio-Tech Co., Ltd. (Tainan, Taiwan). S. thermophilus ST4 was diluted in sterile water and administered by oral gavage. The mice received 100 μL of suspension con- taining 5×108 CFU of the probiotics cocktail daily for 17 days as described in Fig 1A. In vivo experiments Animal care. Five-week-old male BALB/cByJNarl (BALB/c) mice were purchased from the National Laboratory Animal Center (Taipei, Taiwan). The mice were housed in a climate- controlled environment (23 ± 2˚C, relative humidity of 50 ± 5%) with a 12 h of light/dark cycle and allowed free access to food and water ad libitum. The mice adapted to the environ- ment for 2 weeks. The animal experimental protocol used in the current study was reviewed and approved by the Institutional Animal Care and Use Committee of the National Taiwan Fig 1. S. thermophilus ST4 attenuates 5-FU-induced intestinal mucositis. (A) Experimental design for the animal study. The group of 5-FU+S. thermophilus ST4 indicates that S. thermophilus ST4 was intragastrically subjected for pretreatment once per day for 10 days, followed by 5-FU (50 mg/kg) intraperitoneal injection once daily for 3 days, and then continuing S. thermophilus ST4 (5×108 CFU/day) intragastrical administration once daily for additional 4 days. The group of 5-FU is the mice were only treated with 5-FU but no S. thermophilus ST4. Mice without any treatment of both 5-FU and S. thermophilus ST4 were as the control group. Arrows indicate the dates for the injection of 5-FU. (B) Body weight is shown as a percentage of initial body weight in a diary base. (C) Diary food intake was measured diary for each group. (D) The severity of diarrhea is scored using the four-grade scale (0–3) starting from 5-FU treatment toward sacrifice. Data are present as mean ± SD. The data with different superscripted letters are significantly different based on the one-way ANOVA (p<0.05). https://doi.org/10.1371/journal.pone.0253540.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 3 / 15 PLOS ONEStreptococcus thermophilus protects intestinal mucositis University according to the principles of the 3Rs (Replacement, Reduction and Refinement). The experimental design tried to mimic the 5-FU-indcued mucositis by treating mice with low dosage of 5-FU along with or without S. thermophilus ST4 to evaluate the multiple effects, such as body weight change, diarrhea, inflammation, histopathology etc, of the mice on definitive mucositis in comparison with the saline control. When the loss of body weight was reached up to 20% compared to that measured at the point before the 5-FU treatment, the experimental animal was euthanized. Five mice for these three groups are enough for statistical analyses of data collected. In vivo intestinal mucositis model. The experimental set-up is illustrated in Fig 1A and the mice were randomized to one of three groups (n = 5/group) including control group, the 5-FU-induced intestinal mucositis group, and 5-FU+S. thermophilus ST4 (5×108 CFU/day) group using random number tables to achieve randomization. Intestinal mucositis was induced on the 11th to 13th days by intraperitoneal injection of 5-FU (50 mg/kg/day) for 5-FU group and 5-FU+S. thermophilus ST4 group. An injection of saline into mice were used as the control group. Mice were euthanized on the 18th day. Disease severity was assessed daily by measuring body weight and diarrhea status, the latter was graded based on the stool consistency: 0 (normal, normal stool or absent); 1 (slight, slightly wet and soft stool); 2 (moderate, wet and unformed stool with moderate perianal stain- ing of the coat); and 3 (severe, watery stool with severe perianal staining of the coat) [21]. All of data were quantitative collection. For example, fecal samples from each mouse were individ- ually, i.e. each mouse was placed a single clean cage with no bedding waiting for defacating 2–3 fecal pellets, collected, recorded and labelled in a test tube before they were stored in -80˚C. There are two variables in the current study, 5-FU and S. thermophilus ST4. In fact, 5-FU treatment enables inducing the phenomenon of mucositis and S. thermophilus ST4 is the test factor in effect on ameliorating the mucositis. The amounts (concentrations) of both vari- ables have been tested in our preliminary study (data not shown). Here, we grouped the mice into 3 groups, saline treatment, 5-FU alone, and 5-FU+S. thermophilus ST4, respectively, to evaluate the effect of S. thermophilus ST4 on 5-FU-induced mucositis. Then, the mice were sac- rificed under anesthesia to collect their entire small intestines (mainly jejunum tissue) and colons (range from cecum to rectum) after removal of fact tissue and colon length (range excludes the cecum) were measured accordingly. Morphology and histopathology analysis. For the assessment of pathological changes, the small intestines and colons were fixed in 10% formaldehyde solution and embedded in par- affin. Sections with 3–5 μm in thickness were cut, deparaffinized, rehydrated, stained with hematoxylin and eosin (H&E). The morphological alteration and inflammatory cell infiltration were examined under a light microscope (DL 882096, LWScientific, Inc., Georgia, USA). The photos were taken at a final magnification of 100× and 400×. Immunohistochemical staining of the small intestine. Sections (3 μm thick) were cut and mounted on a glass slide and followed by deparaffinization in xylene, rehydration in a graded ethanol series, antigen retrieval using microwave and then endogenous peroxidase inactivation in immersing specimens in 3% H2O2. To perform immunostaining, slides were incubated at 4˚C for overnight with rabbit polyclonal anti-F4/80 antibody (Abcam, cat.# ab6640) at 1: 1000 dilution and followed by another incubation with HPR-conjugated anti- rabbit secondary antibody (Jackson lab) at 1:2000 dilution at room temperature for 1 hour. After washes, sections were developed with diaminobenzidine (DAB) and counterstained with hematoxylin, and then dehydrated in a graded alcohol series, cleared in xylene, and placed with coverslips. Positive F4/80 cells were counted using Image J1 software (Rasband, W.S., ImageJ, U. S. National Institutes of Health, Bethesda, MD, USA). PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 4 / 15 PLOS ONEStreptococcus thermophilus protects intestinal mucositis Pro-inflammatory cytokines analysis. To quantify cytokines, blood was first collected from the heart immediately after mice were sacrificed. Blood samples were then centrifuged to obtain serum. Concentrations of pro-inflammatory cytokines (TNF-α, IL-1, and IL-6) in serum were measured using corresponding enzyme-linked immunosorbent assay kits (ELISA; Thermo Fisher Scientific, Inc.) according to the manufacturer’s instruction. Results are expressed as pg/mL. Short-chain fatty acids (SCFAs) analysis of feces. SCFAs were quantified according to the previously published description [28]. Briefly, fresh fecal samples were collected and homogenized with cold saline (NaCl 0.9%, w/v) at a ratio of 1:10 (w/v) prior to centrifugation for 10 min at 1,000 g. Collected supernatants were acidified with 20 μL of 50% (w/v) sulfuric acid containing isocaporic acid as an internal standard. Then, SCFAs were mixed and extracted by diethyl ether. The extracted samples were directly injected into the GC column (Agilent J and W HP-INNO Wax GC Column, 30 m, 0.25 mm id, 0.25 μm film thickness). The GC system consisted of an Agilent 7890A (Agilent Technologies, Palo Alto, CA, USA), equipped with a flame ionization detector. Helium as carried gas was used for the separation at a flow rate of 7 mL/min. One μL of each sample was injected with an injector temperature of 140˚C and the detector temperature 250˚C. The conditions were as follows: oven temperature, initially held at 80˚C for 1 min and then raised to 140˚C at a rate of 20˚C/min, then held at 140˚C for another 1 min, and raised again to 220˚C at a rate of 20˚C/min, and lastly held at 220˚C for 2 more mins. Statistical analysis All data were analyzed by GraphPad Prism 6.0 software and presented as means ± SD from the indicated number of independent experiments. Comparisons of statistical significance between experimental groups were determined by one-way analysis of variance (ANOVA) using SPSS for Windows (version 12.0) of variance followed by Duncan’s multiple range method. Results Effect of S. thermophilus ST4 treatment on 5-FU-induced body weight loss, food intake and diarrhea To evaluate the protective effect of S. thermophilus ST4 on 5-fluorouracil-induced intestinal mucositis, mice were oral gavage with or without S. thermophilus ST4 prior to 5-FU-induced intestinal mucositis and monitored their body weight loss, food intake and diarrhea dairy (Fig 1A). On contrary to the control group, the 5-FU group mice resulted in significantly decreased body weight (about 20%), food intake (about 50%), diarrhea (score from 0 up to 2.5), reluc- tance to move and hair loose (data not shown). Nevertheless, we found that the administration of S. thermophilus ST4 apparently ameliorated the aforementioned clinical symptoms of 5-FU- induced intestinal mucositis toward no significant difference compared to the control group as shown in Fig 1B–1D. Here, the daily evaluation of intestinal mucositis induced by 5-FU was followed by weight reduction throughout the experimental period. Compared to the control group, gradual body weight loss was observed in mice treated with 5-FU, and the mean body weight (relative mean body weight change (%) was recorded daily and expressed as mean from baseline at day 11th = 0%) was reduced 19.59% of initial body weight on the 18th day. Intri- cately, our data showed that the S. thermophilus ST4 obviously rescued 5-FU-induced impairment in the severity of 5-FU-induced intestinal mucositis resulting from the reductions in body weight loss only 1.85% as shown in Fig 1B. In addition, the food intake was irreversibly by 5-FU+S. thermophilus ST4 compared to the 5-FU treated group reduced by 50% (Fig 1C), PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 5 / 15 PLOS ONEStreptococcus thermophilus protects intestinal mucositis reaching to the same as the control group. Diarrhea was assessed by using the Bowen’s score system. We classified diarrhea into four grades according to the stool consistency: 0, normal stool; 1, slightly wet and soft stool indicating mild diarrhea; 2, wet and unformed stool indicat- ing moderate diarrhea; and 3, watery stool indicating severe diarrhea. accordingly, we observed and recorded the diarrhea score of each mouse starting on the day of 5-FU treatment. As shown in Fig 1D, diarrhea scores exhibited a significant reduction and recovery in the 5-FU +S. thermophilus ST4 treatment compared to the 5-FU-induced group from the most severe day 14th to day 15th, and the mean diarrhea score changed from 1.0 to 0.1 and from 2.5 to 2.50, respectively (Fig 1D). Taken together, these data clearly present the potential effect of S. ther- mophilus ST4 on protecting the 5-FU-induced intestinal mucositis in a mouse model. Effects of S. thermophilus ST4 on 5-FU-induced intestinal mucositis To further evaluate histopathological characteristics, we firstly examined the length of the colon among the different treated groups. In our experiments, the average colon length (7.5 cm) of 5-FU treated mice was significantly shortened compared to that (8.9 cm) of the con- trol group (Fig 2A), whereas the colon length shorten was markedly alleviated and remained at the average of 9.2 cm by treatment with S. thermophilus ST4 even under 5-FU treated condition. Furthermore, we examined the change of small intestines treated with 5-FU Fig 2. Histopathological examination of 5-FU-induced intestinal mucositis with or without S. thermophilus ST4 treatment. Three groups of mice as indicated were sacrificed according to the experiment design subjected for measuring (A) colon length and (B and C) for H&E staining of the small intestine and colon after excised, sectioned, and then stained, respectively. The representative images (100X and 400X magnifications) of each group are presented. Data are present as mean ± SD. The data with different superscripted letters are significantly different based on the one-way ANOVA (p<0.05). https://doi.org/10.1371/journal.pone.0253540.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 6 / 15 PLOS ONEStreptococcus thermophilus protects intestinal mucositis followed with or without S. thermophilus ST4 at the microscopic level (Fig 2B and 2C). Mor- phological analyses indicated that the villus heights to crypts depths of 5-FU treated mice were dramatically shortened in comparison with the control mice, while the reduction of small intestine villus heights and crypts depths was markedly alleviated by treatment with S. thermophilus ST4 as shown in Fig 2B. Our results suggested that the 5-FU-induced mucosi- tis in small intestines could be prevented by 5-FU+S. thermophilus ST4. Likewise, the obvi- ous differences were observed in colon morphology after the induction of mucositis compared to the control group and 5-FU group. 5-FU resulted in the epithelial and crypt damage in colon tissues, and the crypt depth was dramatically decreased by the mucositis as shown in Fig 2C. As expected, 5-FU+S. thermophilus ST4 administration exhibited protec- tive effects on colon damage. S. thermophilus ST4 suppressed 5-FU-induced pro-inflammatory cytokines To understand the mechanistic nature of S. thermophilus ST4 suppressed 5-FU-induced mucositis, we examined expression of inflammatory factors in the presence or absence of S. thermophilus ST4 in a 5-FU-induced mouse model. Indeed, inflammation is known to be involved in the pathogenesis of mucositis, and the pro-inflammatory cytokines are con- sidered as important factors and potential targets for treatment of mucositis. As shown in Fig 3A–3C, in 5-FU-induced mice, the pro-inflammatory cytokines concentrations of TNF-α (34.78 pg/mL), IL-1β (66.02 pg/mL) and IL-6 (460.68 pg/mL) were significantly Fig 3. Reduction of the inflammatory effect by S. thermophilus ST4 in the 5-FU-induced intestinal mucositis mouse model. Despite of elevated expressions of TNF-α (A), IL-1β (B) IL-6 (C) and F4/80 (a marker for macrophage, D and E) induced by 5-FU, S. thermophilus ST4 enabled significantly diminishing the increase of these inflammatory factors in serum or in the intestine (D) and the colon (E) of the 5-FU-induced mucositis mice. Data are present as mean ± SD. The data with different superscripted letters are significantly different based on the one-way ANOVA (p<0.05). https://doi.org/10.1371/journal.pone.0253540.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 7 / 15 PLOS ONEStreptococcus thermophilus protects intestinal mucositis increased in comparison with the control group (TNF-α at 7.10 pg/mL, IL-1β at 46.52 pg/ mL and IL-6 at 34.17 pg/mL), respectively. Similar to the control group, the 5-FU+S. ther- mophilus ST4 group significantly reduced the concentrations of TNF-α, IL-1β and IL-6 at 4.52, 52.51, and 48.53 pg/mL, respectively, compared to the 5-FU group. In agreement with the above result, we also found that the S. thermophilus ST4 group gave rise to less macrophage infiltrations in the small intestine and the colon as shown in Fig 3D and 3E, in compared to those in 5-FU- induced mice, from 56.52% to 17.58%. To be noted, 5-FU treatment impaired mucosal epithelium and disrupted crypt-villus structures, which was accompanied with increase cellular infiltration and macrophage (F4/80 stain) aggregation. in contrast, 5-FU+S. thermophilus ST4 administration exhibited protective effects on intestinal mucositis. Effects of SCFAs production by the S. thermophilus ST4 on 5-FU-induced intestinal mucositis To elucidate the essential metabolite in correlation with the anti-inflammatory effect of the S. thermophilus ST4 on 5-FU-induced intestinal mucositis in the mouse model, we investigated the fecal concentrations of various short-chain fatty acids among the different groups. Here, we further revealed a higher concentration of acetic acid in the stool of the 5-FU+S. thermophi- lus ST4 group (78.86 mmol/g) than that in the control group and 5-FU group (71.28 mmol/g and 70.69 mmol/g, respectively) as shown in Fig 4, implicating this short-chain fatty acid rather than other SCFAs, including propionic acid and butyric acid, critical for the protective effect of S. thermophilus ST4 on intestinal mucositis. Fig 4. S. thermophilus ST4 promotes SCFAs production of the 5-FU-induced mouse. (A-C) HPLC profiles and (D) concentration (μmol/g) of SCFAs in feces collected from three groups. Isocaproic acid as an internal standard in HPLC. Data are present as mean ± SD of 5 mice. The data with different superscripted letters are significantly different based on the one-way ANOVA (p<0.05). https://doi.org/10.1371/journal.pone.0253540.g004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 8 / 15 PLOS ONEStreptococcus thermophilus protects intestinal mucositis Discussion In this present study, we recap 5-FU injected mice resulting in body weight loss, reduced food intake, and diarrhea in our experimental model, which is related to the negative effects of intestinal mucositis as previous reports [3, 29, 30]. Apparently, we showed that S. thermophilus ST4 enabled attenuating the 5-FU-induced body weight loss indicated by the improvement in food intake. Additionally, S. thermophilus ST4 also alleviated the diarrhea of 5-FU-indcued mucositis mice. A previous study evaluated S. thermophilus strain (NCIMB 41856) is con- cluded to promote maintenance of mucosal barrier function and, therefore, indirectly reduce immune stimulation and inflammation in a DSS mouse model to ameliorate signs of colitis in an iron-rich condition [27]. However, our current study does obviously see the additional dif- ferences in histopathology and cytokine production between probiotics treatment group or non-treatment group in a 5-FU-induced mucositis mouse model. We performed the effect of 5-FU on S. thermophilus ST4 cytotoxicity in vitro, none of which induced cytotoxicity in the S. thermophilus ST4 that we tested [S1 Appendix]. In addition, we tried to allow the S. thermophi- lus ST4 colonized on the intestine before 5-FU-induced mucositis to provide a protective effect. In contrast, the Bailey et al administrated DSS to induce colitis at the beginning in com- bination with treatment of S. thermophilus strain (NCIMB 41856). Thus, these studies provide more beneficial effects of S. thermophilus strains on mucositis while the probiotics are applied to the animal as earlier. The intestinal epithelium is a single-cell layer composing the largest and most important barrier whose primary function is to assist in absorption of nutrients across the epithelial lin- ing in addition to maintaining a physical barrier against the external environment [31]. Effi- cient absorption is enhanced within the small intestine via finger-like projections called villi, which greatly augment the surface area in contact with luminal contents [32]. The intestinal barrier is maintained by epithelial cells joined by tight junctions on their lateral boarders, thus the selective passage of luminal contents across the cell being formed [32]. To be noted, crypts are flask-like structures around the base of villi containing proliferative units necessary for maintaining epithelial integrity of throughout the intestine [32]. In contrast, intestinal shorten- ing is a phenomenon which is depicted both in experimental colitis model [33] and chemo- therapy induced mucositis [34], suggesting destruction or inflammation in intestine. For example, 5-FU-induced mucositis effect on intestinal morphology is often characterized by decreased crypt length, blunting and fusion of villi, enterocytes hyperplasia and increased apo- ptosis [35], thereby resulting in the disruption of mucosal integrity as well as the shape changes of villus and crypt parameters. The gastrointestinal crypt epithelium is particularly vulnerable to chemotherapy drug toxicity with symptoms including nausea and vomiting, abdominal pain, distension, and diarrhea [36]. Nevertheless, studies have reported that the oral adminis- tration of S. thermophilus TH-4 at a dose of 109 CFU/mL, partially attenuated methotrexate- induced small intestinal damage [24]. Moreover, [25] showed that live TH-4 and supernatant partially normalized mitotic count and histological severity score while 5-FU-induced intesti- nal mucositis. Morphologically, the broken villi in the small intestine and colon were dramatically increased upon the induction of intestinal mucositis in the current study, while the loss of crypt structure was occurred in colon segments. We demonstrated that the structures of villus and crypt in small intestine and colon were significantly protected by S. thermophilus ST4 treatment at 5×108 CFU/day for 17 days, in which S. thermophilus ST4 might remain mucosal growth during chemotherapy. Rather than the direct injury to intestinal basal stem cells, 5-FU-induced intestinal mucosi- tis also leads to a consequence of complex biological events, including reactive oxygen species PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 9 / 15 PLOS ONEStreptococcus thermophilus protects intestinal mucositis (ROS) generation, immune cells infiltration, and pro-inflammatory cytokines over-production [37, 38]. Pro-inflammatory cytokines, such as TNF-α, IL-1β, and IL-6, play a role in amplifying the severity of chemotherapy-induced intestinal mucositis [39]. In fact, TNF-α is a key factor in the caveolin-1-mediated internalization of occluding, which elevates gut permeability [40]. Besides, TNF-α is a very potent matrix metalloproteinases produced by neutrophils and acti- vated macrophages in contribution to the epithelial ulceration and sub-mucosal destruction [41]. Numerous studies have revealed elevated productions of the inflammasome-dependent cytokines IL-1β and IL-18 during clinical and experimental chemotherapy-induced mucositis [42]. Consistently, IL-1β enhances intestinal mucositis pathogenesis by triggering apoptosis of intestinal crypt epithelial cells via p53-mediated upregulation of p21 and p27 [43]. Moreover, [44] substantiated the importance of IL-6 in 5-FU-induced small intestine injury using an IL- 6-/- mouse model, also indicating the caspase-3-dependent pathway is involved in IL6-me- diated apoptosis in the ileum and colon in response to 5-FU treatment. On the other hand, IL- 6 is recognized as an important mediator of gut dysfunction in IBD [45] in concordance with an increase in IL-6 signaling often observed in the inflamed mucosa of IBD [46]. In fact, expressions of TNF-α, IL-1β and IL-6 were also found to be increased in response to the 5-FU treatment, indicating that inflammatory cytokines play a key role in the pathogenesis of muco- sitis induced by chemotherapy and radiotherapy. Therefore, it is believed that TNF-α, IL-1β and IL-6 are involved in mucositis and have been the targets of inhibition [4]. Herein, we showed that TNF-α, IL-1β and IL-6 were significantly increased following 5-FU treatment, while these increases were attenuated by S. thermophilus ST4 treatment. In combination of mucositis phenotypes, S. thermophilus ST4 serving as a regimen enables attenuating the sever- ity of intestinal mucositis induced by 5-FU through the inhibition of pro-inflammatory cyto- kines expression. Numerous intestinal diseases are characterized by immune cell activation in association with the detriment of epithelial barrier function. Based on the model of gastrointestinal muco- sitis reported by [47], infiltrated phagocytes, such as macrophages and neutrophils, at inflamed sites, are thought to be responsible for the formation of ROS, which can subsequently alter the localization of tight junction components such as ZO-1 and occluding [48]. In fact, macro- phage infiltration is a common feature of inflammation in the chemotherapy-induced intestine [4]. In consistence with the above, severe damages in small intestine and colon manifested by villus deformation, loss, and atrophy were accompanied with enhanced macrophage infiltra- tion in mice by 5-FU treatment. In contrast, S. thermophilus ST4 enables decreasing the infil- tration of macrophage into distal mucosa and protecting the structural integrity of small intestine and colon tissue. SCFAs (short chain fatty acids) can be resulted from bacterial fermentation in the intestine to provide energy for colonic mucosal epithelial cells and essential for the development and mediation of the intestinal barrier function [49, 50]. We showed that S. thermophilus ST4 sig- nificantly increased fecal acetic acid concentration, one of common and functional SCFAs. Studies have revealed that acetate enables inhibiting fat accumulation in adipose tissue and then results in suppression of metabolic inflammation via insulin signaling in adipocytes in rodents [51]. In addition, activation of G-protein-coupled receptor 43 (GPR43) by acetate markedly protects against gut inflammation in the model of colitis [52]. Reportedly, studies have observed that Bifidobacteria can protect the host against lethal infection via the produc- tion of acetate [53]. [54] revealed that acetic acid increased DNA synthesis in a human colonic epithelial cell line from adenocarcinoma (LS-123 cells) and a non-transformed small intestinal cell line from germ-free rats (IEC-6 cells). There are several limitations in this study. One limitation is that we focused on the histologi- cal effects of small intestines and colon, other parts of the gastrointestinal tract such as stomach PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 10 / 15 PLOS ONEStreptococcus thermophilus protects intestinal mucositis specimens were not examined. Also this study did not address the possible mechanisms by which S. thermophilus ST4 might exert their beneficial outcomes to sustain tight junction pro- teins and transepithelial electrical resistance. These areas should be investigated in the future. On the other hand, further studies should also focus on the exploration of S. thermophilus ST4-derived compounds in effect on tight junction expression and intestinal permeability should be conducted to better elucidate the mechanisms underlying maintenance of intestinal barrier functions since that we observed a major effect on ameliorating diarrhea in response to the 5-FU treatment. More clinical works are needed to demonstrate the beneficial effects of S. thermophilus ST4 in the management of 5-FU-induced intestinal mucositis. Conclusion The present work showed that oral administration of probiotic S. thermophilus ST4 enables attenuating 5-FU-induced intestinal mucositis. The S. thermophilus ST4 markedly improved the clinical complications such as weight loss, food intake and diarrhea score. Furthermore, the S. thermophilus ST4 treatment significantly mitigated the histological damage compared to the 5-FU-induced intestinal mucositis. Similarly, the biochemical changes such as inflamma- tory cytokines such as TNF-α, IL-1β and IL-6 were markedly alleviated attenuated by the S. thermophilus ST4 treatment. Furthermore, acetic acid in SCFAs were significantly enhanced by S. thermophilus ST4. In summary, the present results speculated that the positive effect of S. thermophilus ST4 is partially if not all attributed by an increase in the content of acetic acid in correlation with maintenance of inflammatory homeostasis as well as preservation of intestinal permeability (Fig 5). Based on these findings, our results indicated that oral administration of probiotic S. thermophilus ST4 can ameliorate 5-FU-induced intestinal mucositis in a mouse mucositis model. Accordingly, it is of indicative that probiotics may be a potential alternative therapeutic strategy for the prevention or management of 5-FU-induced intestinal mucositis in the future. More clinical works are required to demonstrate the beneficial effects of S. ther- mophilus ST4 and elucidate the correct dosing regimens in the management of 5-FU-induced intestinal mucositis. Fig 5. A schematic model for S. thermophilus ST4 attenuating 5-FU-induced intestinal mucositis. Elevation of SCFAs, for example acetic acid, in intestine enables diminishment of inflammation as well as maintenance of barrier function in intestine and colon, therefore resulting in increased appetite and in reducing diarrhea in the 5-FU-induced intestinal mucositis. https://doi.org/10.1371/journal.pone.0253540.g005 PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 11 / 15 PLOS ONEStreptococcus thermophilus protects intestinal mucositis Supporting information S1 Appendix. The effect of 5-FU on S. thermophilus ST4 cytotoxicity. The 5-FU did not apparently cause cytotoxicity on S. thermophilus ST4 at the test concentration of 0.5~10 μM. (DOCX) S1 Table. The effect of 5-FU on S. thermophilus ST4 cytotoxicity. (TIF) Acknowledgments We thank Drs. Hsin-Yi Lo and Pei-Yu Chu for discussion and assistance in some of the experiments. Author Contributions Conceptualization: Wei-Jen Chen, Shiuan-Huei Wu, Tang-Long Shen. Data curation: Siou-Ru Shen. Formal analysis: Siou-Ru Shen, Yu-Ru Wang, Tang-Long Shen. Funding acquisition: Wei-Jen Chen, Tang-Long Shen. Investigation: Siou-Ru Shen. Methodology: Shiuan-Huei Wu. Project administration: Shiuan-Huei Wu, Tang-Long Shen. Resources: Wei-Jen Chen, Hui-Fang Chu, Shiuan-Huei Wu, Tang-Long Shen. Supervision: Wei-Jen Chen, Hui-Fang Chu, Shiuan-Huei Wu, Tang-Long Shen. Writing – original draft: Siou-Ru Shen, Yu-Ru Wang, Tang-Long Shen. Writing – review & editing: Tang-Long Shen. References 1. Zhang T, Lu SH, Bi Q, Liang L, Wang YF, Yang XX, et al. Volatile oil from amomi fructus attenuates 5- fluorouracil-induced intestinal mucositis. Front Pharmacol. 2017; 8:786. https://doi.org/10.3389/fphar. 2017.00786 PMID: 29170638 2. Keefe DMK, Brealey J, Goland GJ, Gummins AG. Chemotherapy for cancer causes apoptosis that pre- cedes hypoplasia in crypts of the small intestine in humans. Gut. 2000; 47(5):632–637. https://doi.org/ 10.1136/gut.47.5.632 PMID: 11034578 3. Li HL, Lu L, Wang XS, Qin LY, Wang, Qiu SP, et al. Alteration of gut microbiota and inflammatory cyto- kine/chemokine profiles in 5-Fluorouracil induced intestinal mucositis. Front Cell Infect Microbiol. 2017; 26(7):455. https://doi.org/10.3389/fcimb.2017.00455 PMID: 29124041 4. Sultani M, Stringer AM, Bowen JM, Gibson RJ. Anti-inflammatory cytokines: important immunoregula- tory factors contributing to chemotherapy-induced gastrointestinal mucositis. Chemother Res Pract. 2012; 2012:490804. https://doi.org/10.1155/2012/490804 PMID: 22973511 5. Logan RM, Stringer AM, Bowen JM, Yeoh AS, Gibson RJ, Sonis ST, et al. The of pro-inflammatory cyto- kines in cancer treatment-induced alimentary tract mucositis: pathobiology, animal models and cyto- toxic drugs. Cancer Treat Rev. 2007; 33:448–460. https://doi.org/10.1016/j.ctrv.2007.03.001 PMID: 17507164 6. De Koning BA, van Dieren JM, Lindenbergh-Kortleve DJ, van der Sluis M, Matsumoto T, Yamaguchi K, et al. Contributions of mucosal immune cells to methotrexate-induced mucositis. Int Immunol. 2006; 18:941–949. https://doi.org/10.1093/intimm/dxl030 PMID: 16636014 7. La Fata G, Weber P, Mohajeri MH. Probiotics and the gut immune system: indirect regulation. Probiotics Antimicrob Proteins. 2018; 10(1):11–21. https://doi.org/10.1007/s12602-017-9322-6 PMID: 28861741 PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 12 / 15 PLOS ONEStreptococcus thermophilus protects intestinal mucositis 8. Stanton C, Gardiner G, Meehan H, Collins K, Fitzgerald G, Lynch PB, et al. Market potential for probiot- ics. Am J Clin Nutr. 2001; 73(2 Suppl):476S–483S. https://doi.org/10.1093/ajcn/73.2.476s PMID: 11157361 9. Ohland CL, MacNaughton WK. Probiotic bacteria and intestinal epithelial barrier function. Am J Physiol Gastrointest Liver Physiol. 2010; 298(6):807–819. https://doi.org/10.1152/ajpgi.00243.2009 PMID: 20299599 10. Kim H, Krenek KA, Fang C, Minamoto Y, Markel ME, et al. Polyphenolic derivatives from mango (Man- gifera Indica L.) modulate fecal microbiome, short-chain fatty acids production and the HDAC1/AMPK/ LC3 axis in rats with DSS-induced colitis. J Funct Foods. 2018; 48:243–251. https://doi.org/10.1016/j. jff.2018.07.011 11. Zeng H, Umar S, Rust B, Lazarova D, Bordonaro M. Secondary bile acids and short chain fatty acids in the colon: a focus on colonic microbiome, cell proliferation, inflammation, and cancer. Int J Mol Sci. 2019; 20(5):1214. https://doi.org/10.3390/ijms20051214 PMID: 30862015 12. Markowiak-Kopeć P, Śliżewska K. The effect of probiotics on the production of short-chain fatty acids by human intesitnal microbiome. Nutrients. 2020; 12(4):1107. https://doi.org/10.3390/nu12041107 PMID: 32316181 13. Wong JM, De Souza R, Kendall CW, Emam A, Jenkins DJ. Colonic health: fermentation and short chain fatty acids. J Clin Gastroenterol. 2006; 40(3):235–243. https://doi.org/10.1097/00004836- 200603000-00015 PMID: 16633129 14. Gamet L, Daviaud D, Denis-Pouxviel C, Remesy C, Murat JC. Effects of short-chain fatty acids on growth and differentiation of the human colon-cancer cell line HT29. Int J Cancer. 1992; 52(2):286– 289. https://doi.org/10.1002/ijc.2910520222 PMID: 1521915 15. Roy MJ, Dionne S, Marx G, Qureshi I, Sarma D, Levy E, et al. In vitro studies on the inhibition of colon cancer by butyrate and carnitine. Nutrition. 2009; 25(11–12):1193–1201. https://doi.org/10.1016/j.nut. 2009.04.008 PMID: 19619983 16. Rizkalla SW, Kabir JLM, Chevalier A, Pacher N, Slama, G. Chronic consumption of fresh but not heated yogurt improves breath-hydrogen status and short-chain fatty acid profiles: a controlled study in healthy men with or without lactose maldigestion. Am J Clin Nutr. 2000; 72(6):1474–1479. https://doi.org/10. 1093/ajcn/72.6.1474 PMID: 11101474 17. Bruno-Ba´rcena JM, Andrus JM, Libby SL, Klaenhammer TR, Hassan HM. Expression of heterologous manganese superoxide dismutase gene in intestinal Lactobaciili provides protection against hydrogen peroxide toxicity. Appl Environ Microbiol. 2004; 70(8):4702–4710. https://doi.org/10.1128/AEM.70.8. 4702-4710.2004 PMID: 15294805 18. Delorme C. Safety assessment of dairy microorganisms: Streptococcus thermophilus. Int J Food Micro- biol. 2008; 126(3):274–277. https://doi.org/10.1016/j.ijfoodmicro.2007.08.014 PMID: 17822794 19. Resta-Lenert S, Barrett KE. Live probiotics protect intestinal epithelial cells from the effects of infection with enteroinvasive Escherichia Coli (EIEC). Gut. 2003; 52(7):988–997. https://doi.org/10.1136/gut.52. 7.988 PMID: 12801956 20. Zheng B, van Bergenhenegouwen J, van de Kant HJ, Folkerts G, Garssen J, Vos AP, et al. Specific pro- biotic dietary supplementation leads to different effects during remission and relapse in murine chronic colitis. Benef Microbes. 2016; 7(2):205–213. https://doi.org/10.3920/bm2015.0037 PMID: 26645352 21. Bowen JM, Stringer AM, Gibson RJ, Yeoh ASJ, Hannam S, Keefe DMK. VSL#3 probiotic treatment reduces chemotherapy-induced diarrhea and weight loss. Cancer Biol Ther. 2007; 6(9):1449–1454. https://doi.org/10.4161/cbt.6.9.4622 PMID: 17881902 22. Delia P, Sansotta G, Donato V, Frosina P, Salatino A, Messina G, et al. Use of probiotics for prevention of radiation-induced diarrhea. World J Gastroenterol. 2007; 13(6):912–915. https://doi.org/10.3748/ wjg.v13.i6.912 PMID: 17352022 23. Bojrab G. Composition, of L. bulgaricus and S. thermophilus, for the treatment of gastrointestinal disor- ders, hyperlipidemia, autoimmune diseases, and obesity. European patent application. 2001; 1 (177):794. 24. Tooley KL, Howarth GS, Lymn KA, Lawrence A, Butler RN. Oral ingestion of Streptococcus thermophi- lus diminishes severity of small intestinal mucositis in methotrexate treated rats. Cancer Biol Ther. 2006; 5(6):593–600. https://doi.org/10.4161/cbt.5.6.2659 PMID: 16627985 25. Whitford EJ, Cummins AG, Butler RN, Prisciandaro LD, Fauser JK, Yazbeck R, et al. Effects of Strepto- coccus thermophilus TH-4 on intestinal mucositis induced by the chemotherapeutic agent, 5-fluorouracil (5-FU). Cancer Biol Ther. 2009; 8(6):505–511. https://doi.org/10.4161/cbt.8.6.7594 PMID: 19305160 26. Chen Y, Zhang M, Ren F. A role of exopolysaccharide produced by Streptococcus thermophilus in the intestinal inflammation and mucosal barrier in Caco-2 monolayer and dextran sulphate sodium-induced experimental murine colitis. Molecules. 2019; 24(3):513. https://doi.org/10.3390/molecules24030513 PMID: 30708992 PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 13 / 15 PLOS ONEStreptococcus thermophilus protects intestinal mucositis 27. Bailey JR, Vince V, Williams NA, Cogan TA. Streptococcus thermophilus NCIMB 41856 ameliorates signs of colitis in an animal model of inflammatory bowel disease. Benef Microbes. 2017; 8(4):605– 614. https://doi.org/10.3920/BM2016.0110 PMID: 28618865 28. Huang YL, Chu HF, Dai FJ, Yu TY, Chau CF. Intestinal health benefits of the water-soluble carbohy- drate concentrate of wild grape (Vitis thunbergia) in hamsters. J Agric Food Chem. 2012; 60(19):4854– 4858. http://doi.org/10.1021/jf300942p PMID: 22546171 29. Kato S, Hayashi S, Kitahara Y, Nagasawa K, Aono H, Shibata J, et al. Saireito (TJ-114), a Japanese tra- ditional herbal medicine, reduces 5-fluorouracil-induced intestinal mucositis in mice by inhibiting cyto- kine-mediated apoptosis in intestinal crypt cells. PLoS One. 2015; 10(1):e0116213. http://doi.org/10. 1371/journal.pone.0116213 PMID: 25565296 30. Keefe DM. Intestinal mucositis: mechanisms and management. Curr Opin Oncol. 2007; 19:323–327. http://doi.org/10.1097/CCO.0b013e3281214412 PMID: 17545794 31. Groschwitz KR, Hogan SP. Intestinal barrier function: molecular regulation and disease pathogenesis. J Allergy Clin Immunol. 2009; 124(1):3–20. http://doi.org/10.1016/j.jaci.2009.05.038 PMID: 19560575 32. Bowen JM, Gibson RJ, Gummins AG, Keefe DMK. Intestinal mucositis: the role of the Bcl-2 family, p53 and caspases in chemotherapy-induced damage. Support Care Cancer. 2006; 14(7):713–731. https:// doi.org/10.1007/s00520-005-0004-7 PMID: 16453135 33. Bao S, Carr EDJ, Xu YH, Hunt NH. Gp91(phox) contributes to the development of experimental inflam- matory bowel disease. Immunol Cell biol. 2011; 89(8):853–860. http://doi.org/10.1038/icb.2011.4 PMID: 21321580 34. Cool JC, Dyer JL, Xian CJ, Butler RN, Geier MS, Howarth GS. Pre-treatment with insulin-like growth factor-I partially ameliorates 5-fluorouracil-induced intestinal mucositis in rats. Growth Horm IGF Res. 2005; 15(1):72–82. https://doi.org/10.1016/j.ghir.2004.12.002 PMID: 15701575 35. Cinausero M, Aprile G, Ermacora P, Basile D, Vitale MG, Fanotto V, et al. New frontiers in the pathobiol- ogy and treatment of cancer regimen-related mucosal injury. Front Pharmacol. 2017; 8(8):354. https:// doi.org/10.3389/fphar.2017.00354 PMID: 28642709 36. Vanhoecke B, Bateman E, Mayo B, Vanlancker E, Stringer A, Thorpe D, et al. Dark agouti rat model of chemotherapy-induced mucositis: establishment and current state of the art. Exp Biol Med (Maywood). 2015; 240(6):725–741. https://doi.org/10.1177/1535370215581309 PMID: 25966981 37. Sonis ST. Regimen-related gastrointestinal toxicities in cancer patients. Curr Opin Support Palliat Care. 2010; 4(1):26–30. https://doi.org/10.1097/SPC.0b013e328335fb76 PMID: 20040879 38. Lu Z, Ding L, Lu Q, Chen YH. Claudins in intestines: distribution and functional significance in health and diseases. Tissue Barriers. 2013; 1(3):e24978. https://doi.org/10.4161/tisb.24978 PMID: 24478939 39. Sonis ST. The pathobiology of mucositis. Nat Rev Cancer. 2004; 4(4):277–284. http://doi.org/10.1038/ nrc1318 PMID: 15057287 40. Marchiando AM, Shen L, Graham WV, Weber CR, Schwarz BT, Austin 2nd JR, et al. Caveolin-1-depen- dent occludin endocytosis is required for TNF-induced tight junction regulation in vivo. J Cell Biol. 2010; 189(1):111–126. https://doi.org/10.1083/jcb.200902153 PMID: 20351069 41. Da Silva AP, Pollett A, Rittling SR, Denhardt DT, Sodek J, Zohar R. Exacerbated tissue destruction in DSS-induced acute colitis of OPN-null mice is associated with downregulation of TNF-alpha expression and non-programmed cell death. J Cell Physiol. 2006; 208(3):629–639. https://doi.org/10.1002/jcp. 20701 PMID: 16741956 42. Melo MLP, Brito GA, Soares RC, Carvalho SBLM, Silva JV, Soares PMG, et al. Role of cytokines (TNF- alpha, IL-1beta and KC) in the pathogenesis of CPT-11-induced intestinal mucositis in mice: effect of pentoxifylline and thalidomide. Cancer Chemother Pharmacol. 2008; 61(5), 775–784. https://doi.org/ 10.1007/s00280-007-0534-4 PMID: 17624531 43. Wang X, Zhu S, Qian L, Gao J, Wu M, Gao J, et al. IL-1Ra selectively protects intestinal crypt epithelial cells, but not tumor cells, from chemotoxicity via p53-mediated upregulation of p21 (WAF1) and p27 (KIP1). Pharmacol Res. 2014; 82:21–33. https://doi.org/10.1016/j.phrs.2014.03.007 PMID: 24681512 44. Jin XL, Zimmers TA, Zhang ZX, Pierce RH, Koniaris LG. Interleukin-6 is an important in vivo inhibitor of intestinal epithelial cell death in mice. Gut. 2010; 59(2):186–196. https://doi.org/10.1136/gut.2008. 151175 PMID: 19074180 45. Vermeire S, Van Assche G, Rutgeerts P. C-reactive protein as a marker for inflammatory bowel dis- ease. Inflamm Bowel Dis. 2004; 10(5):661–665. https://doi.org/10.1097/00054725-200409000-00026 PMID: 15472532 46. Nikolaus S, Waetzig GH, Butzin S, Ziolkiewicz M, AI-Massad N, Thieme F, et al. Evaluation of interleu- kin-6 and its soluble receptor components Sil-6R and sgp130 as markers of inflammation in inflamma- tory bowel diseases. Int J Colorectal Dis. 2018; 33(7):927–936. https://doi.org/10.1007/s00384-018- 3069-8 PMID: 29748708 PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 14 / 15 PLOS ONEStreptococcus thermophilus protects intestinal mucositis 47. Sonis ST, Tracey C, Shklar G, Jenson J, Florine D. An animal model for mucositis induced by cancer chemotherapy. Oral Surg Oral Med Oral Pathol. 1990; 69(4):437–443. https://doi.org/10.1016/0030- 4220(90)90376-4 PMID: 2326035 48. Rao RK, Basuroy S, Rao VU, Karnaky KJ, Gupta A. Tyrosine phosphorylation and dissociation of occlu- din-ZO-1 and E-cadherin-beta-catenin complexes from the cytoskeleton by oxidative stress. Biochem J. 2002; 368(Pt 2):471–481. https://doi.org/10.1042/BJ20011804 PMID: 12169098 49. Sakata T. Stimulatory effect of short-chain fatty acids on epithelial cell proliferation in the rat intestine: a possible explanation for trophic effects of fermentable fiber, gut microbes and luminal trophic factors. Br J Nutr. 1987; 58(1):95–103. https://doi.org/10.1079/bjn19870073 PMID: 3620440 50. Peng L, He Z, Chen W, Holzman IR, Lin J. Effects of butyrate on intestinal barrier function in a Caco-2 cell monolayer model of intestinal barrier. Pediatr Res. 2007; 61(1):37–41. http://doi.org/10.1203/01. pdr.0000250014.92242.f3 PMID: 17211138 51. Kimura I, Ozawa K, Inoue D, Imamura T, Kimura K, Maeda T, et al. The gut microbiota suppresses insu- lin-mediated fat accumulaiton via the short-chain fatty acid receptor GPR43. Nat Commun. 2013; 4:1829. https://doi.org/10.1038/ncomms2852 PMID: 23652017 52. Maslowski KM, Vieira AT, Ng A, Kranich J, Sierro F, Yu D, et al. Regulation of inflammatory responses by gut microbiota and chemoattractant receptor GPR43. Nature. 2009; 461(7268):1282–1286. https:// doi.org/10.1038/nature08530 PMID: 19865172 53. Fukuda S, Toh H, Hase K, Oshima K, Nakanishi Y, Yoshimura K, et al. Bifidobacteria can protect from enteropathogenic infection through production of acetate. Nature. 2011; 469(7331):543–547. https:// doi.org/10.1038/nature09646 PMID: 21270894 54. Olaya J, Neopikhanov V, Uribe A. Lipopolysaccharide of Escherichia coli, polyamines, and acetic acid stimulate cell proliferation in intestinal epithelial cells. In Vitro Cell Dev Biol Anim. 1999; 35(1):43–48. https://doi.org/10.1007/s11626-999-0042-4 PMID: 10475254 PLOS ONE \| https://doi.org/10.1371/journal.pone.0253540 July 26, 2021 15 / 15 PLOS ONE
10.1371_journal.pone.0270817	RESEARCH ARTICLE Astrocytes and pericytes attenuate severely injured patient plasma mediated expression of tight junction proteins in endothelial cells Preston StaffordID Jamie HadleyID, Patrick Hom, Terry R. SchaidID, Mitchell J. CohenID* ☯, Sanchayita Mitra☯, Margot Debot, Patrick Lutz, Arthur StemID, Division of GITES, Department of Surgery, University of Colorado Anschutz Medical Campus, Aurora, Colorado a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 ☯ These authors contributed equally to this work. * Mitchell.cohen@cuanschutz.edu Abstract OPEN ACCESS Citation: Stafford P, Mitra S, Debot M, Lutz P, Stem A, Hadley J, et al. (2022) Astrocytes and pericytes attenuate severely injured patient plasma mediated expression of tight junction proteins in endothelial cells. PLoS ONE 17(7): e0270817. https://doi.org/10.1371/journal.pone.0270817 Editor: Ma´ria A. Deli, Eo¨tvo¨s Lora´nd Research Network Biological Research Centre, HUNGARY Received: March 22, 2022 Accepted: June 20, 2022 Published: July 5, 2022 Copyright: © 2022 Stafford et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Funding: MC Trans-agency Research Consortium for Trauma Induced Coagulopathy (TACTIC UM1- HL120877) from the National Heart, Lung and Blood Institute, NIH. https://grants.nih.gov/grants/ guide/rfa-files/RFA-HL-13-025.html The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Blood Brain Barrier (BBB) breakdown is a secondary form of brain injury which has yet to be fully elucidated mechanistically. Existing research suggests that breakdown of tight junction proteins between endothelial cells is a primary driver of increased BBB permeability follow- ing injury, and intercellular signaling between primary cells of the neurovascular unit: endo- thelial cells, astrocytes, and pericytes; contribute to tight junction restoration. To expound upon this body of research, we analyzed the effects of severely injured patient plasma on each of the cell types in monoculture and together in a triculture model for the transcriptional and translational expression of the tight junction proteins Claudins 3 and 5, (CLDN3, CLDN5) and Zona Occludens 1 (ZO-1). Conditioned media transfer studies were performed to illuminate the cell type responsible for differential tight junction expression. Our data show that incubation with 5% human ex vivo severely injured patient plasma is sufficient to pro- duce a differential response in endothelial cell tight junction mRNA and protein expression. Endothelial cells in monoculture produced a significant increase of CLDN3 and CLDN5 mRNA expression, (3.98 and 3.51 fold increase vs. control respectively, p<0.01) and CLDN5 protein expression, (2.58 fold change vs. control, p<0.01), whereas in triculture, this increase was attenuated. Our triculture model and conditioned media experiments suggest that conditioned media from astrocytes and pericytes and a triculture of astrocytes, pericytes and endothelial cells are sufficient in attenuating the transcriptional increases of tight junc- tion proteins CLDN3 and CLDN5 observed in endothelial monocultures following incubation with severely injured trauma plasma. This data suggests that inhibitory molecular signals from astrocytes and pericytes contributes to prolonged BBB breakdown following injury via tight junction transcriptional and translational downregulation of CLDN5. Introduction Traumatic Brain Injury (TBI) is one of the leading causes of mortality worldwide, with mortal- ity rates as high as 30% and significant morbidity in survivors [1,2]. Following trauma, treating PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 1 / 19 PLOS ONECompeting interests: The authors have declared that no competing interests exist. Endothelial cell tight junction expression in non-TBI trauma primary brain injury and mitigating secondary brain injury is the primary focus. Separate from direct TBI, non-brain injured trauma patients also display brain injury like symptomatol- ogy. In both cases, (direct TBI and indirect TBI) microvascular injury and neuroinflammation are thought to drive this pathology and have become a target for research and therapeutics [3– 5]. A key component of this microvascular inflammatory injury is breakdown of the Blood Brain Barrier (BBB) which is associated with morbidity and mortality following TBI [4], and observationally following severe non-TBI traumatic injury. Central to targeted therapeutic treatment for BBB dysfunction is understanding the pathology underlying cellular responses to traumatic injury. In this we look to address brain pathology in non-TBI injured patients, by examining BBB function after non-TBI trauma. The Blood Brain Barrier regulates transcytotic movement of molecules between the vascula- ture and neuronal space. The BBB is composed of the neurovascular unit, comprising three main cell types–endothelial cells, astrocytes, and pericytes. These cell types participate in the formation of a physical barrier composed of tight junction proteins Claudin 3 (CLDN3), Clau- din 5 (CLDN5), Occludin, Junctional Adhesion Molecules (JAM1, JAM2, JAM3), and the anchor proteins Zona Occludens 1, 2, and 3, (ZO-1, ZO-2, ZO-3) expressed and localized within and between the endothelial cells [5]. This physical barrier is the principle homeostatic regulator between the CNS and peripheral vasculature [6] and performs the essential functions of facilitating the transfer of nutrients, regulating ion stasis, and blocking noxious molecules from flowing into the neuronal extracellular space [5]. BBB dysfunction results in adverse patient outcomes linked to transcytotic leakage of fluids, proteins, and other molecules, lead- ing to intraneuronal toxicity and homeostatic imbalance [7–9]. Central to BBB integrity is the maintenance of tight junction proteins which can be degraded through protease activity following TBI [10]. BBB degradation and subsequent leak of inflammatory soluble factors into intraneuronal space has been demonstrated to occur as soon as 30 minutes post-injury, with maximal leakage occurring in a biphasic manner starting as early as 4 hours post-injury, and continued permeability up to 30 days post-injury [11–13]. Based on these findings [6,14–18] we focused on the expression of three tight junction proteins essential to BBB permeability maintenance, CLDN3, CLDN5 and ZO-1. To elucidate this mechanism in the context of trauma, we incubated cells in monoculture and in triculture with severely injured patient plasma and performed conditioned media exchange studies to probe the contribution of each cell type to tight junction expression. We hypothesized that plasma from non-TBI traumatically injured patients leads to a loss of junc- tional integrity between endothelial cells via transcriptional down regulation of the major gap junction proteins Claudins 3, 5 and ZO-1. Methods Patient plasma Normal pooled plasma (George King Bio-medical) phenotypes were verified by the manufac- turer and used as our healthy plasma negative control. Experimental plasma was collected from severely injured trauma patients upon arrival to the emergency department at a Level 1 trauma center. Plasma samples were collected, and patient demographics described in accor- dance with a sampling protocol approved by the Colorado Multiple Institutional Review Board (COMIRB#13–3087); all subjects were informed, and collection performed under waiver of consent. (Tables 1 and 2) [19] Patient demographics were collected, and injury sever- ity score and base deficit was assessed [20–22]. To address limited ex vivo plasma volumes, two separate groups of pooled trauma plasma were created by combining individual patient samples. The severely injured patient plasma was pooled based upon the patient’s injury PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 2 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma Table 1. Trauma plasma pool for transcription experiments described in Figs 1 and 2. Trauma Plasma Pool Age ISS BD Blunt/Penetrating 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 62 40 43 35 62 23 24 38 38 23 36 41 50 38 26 37 34 24 19 55 38 25 25 34 24 41 66 25 43 42 17 19 16 29 25 26 42 33 25 29 -9.7 Blunt -12.7 Blunt -24.7 Blunt -11 Blunt -8.3 Blunt -7 Blunt -10 Penetrating -9.3 Blunt -7 Penetrating -6.3 Penetrating -10 Penetrating -7 Penetrating -10 Blunt -13 Blunt -10 Penetrating -13.6 Penetrating -25 Blunt -12.4 Penetrating -10.3 Blunt -13 Blunt Median 37.5 27.5 -10.0 Table 1- Pooled plasma samples from severely injured patients used for transcription experiments in Figs 1 and 2. Samples were comprised of equivalent volumes from each patient pooled together, 500 μL from each patient. Both individual patients and final pooled plasma were selected with the cutoff criteria of ISS<15, BD<-6. https://doi.org/10.1371/journal.pone.0270817.t001 severity score (ISS), and base deficit (BD). ISS is used to determine the extent of injury severity, and BD is used to determine tissue hypoperfusion [20–22]. Severe injury is any sample from a patient with an ISS>15 and BD<-6. Table 2. Trauma plasma pool used for ELISA. Trauma Plasma Pool Age ISS BD Blunt/Penetrating 2 2 2 2 2 2 2 2 2 2 33 44 28 24 31 27 57 54 43 36 34 22 29 59 16 30 75 25 25 41 -6 Blunt -13 Blunt -17 Blunt -13 Blunt -11 Blunt -19.3 Penetrating -27.5 Penetrating -24.5 Blunt -24.7 Penetrating -27.6 Penetrating Median 34.5 29.50 -18.2 Table 2- Pooled plasma samples from severely injured patients used for translation experiments in Figs 3 and 4. Samples were comprised of equivalent volumes from each patient pooled together, 500 μL from each patient. Both individual patients and final pooled plasma were selected with the cutoff criteria of ISS<15, BD<-6. https://doi.org/10.1371/journal.pone.0270817.t002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 3 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma Cell culture Immortalized Human Cerebral Microvascular Endothelial cells (hCMEC/D3) and complete cell growth media (EndoGRO-MV) were obtained from EMD Millipore and cultured as described by the manufacturer. Primary human astrocytes with complete astrocyte media and primary human brain vascular pericytes with complete pericyte growth media were purchased from Science Cell and cultured in astrocyte and pericyte specific media respectively, as described by the manufacturer. Purity and phenotype of each cell type were verified by manu- facturer with certificate of analysis provided upon delivery. HCMEC/D3 cells cultured with less than 12 passages from passage indicated by manufacturer were used for all experiments. Human astrocytes and pericytes cultured with less than 8 passages from passage indicated by manufacturer were used for all experiments. Triculture Model on Insert: The triculture model was an adaptation of previously published protocols from Hatherell et. al. 2011 and Stone et. al. 2019 [23,24]. In brief, transwell 12 well cell inserts with a 0.4 μm 0.9 cm2 PET membrane (Falcon) were coated for one hour with a 50–50 mixture of 1% rat-tail Collagen Type II and 1% Poly-L-Lysine on the basal surface. The coating was left applied for an hour before being removed by vacuum pipette. To distinguish between the apical and basal (top and bottom) surface of the cell insert membrane, we denoted them as follows. The membrane facing up inside the well of the insert is denoted as apical. The membrane facing down on the outside of the insert exposed to the culture well plate is denoted as basal. For monocultures, endothelial cells were plated on the apical membrane at a concentration of 300k cells; The Pericytes and Astrocytes were plated on the basal membrane at a concentra- tion of 300k and 100k respectively. Astrocytes and pericytes plated on the basal side were sup- plemented with additional media to a volume total 300μL. For the triculture, the basal side was plated first, with pericytes added first and given 30min to adhere, followed by the addition of astrocytes and another 30 min incubation to allow for adherence of the cells to the membrane. Following the one-hour incubation, the cell inserts were inverted and placed into a 12 well cell culture plate. For the triculture, following the inversion into the 12 well culture plate, 300K endothelial cells were plated on the apical surface. The apical surface was supplemented with 1 mL growth media and the basal surface with 3 mL of growth media and incubated for 48 hours in a humidified chamber at 37 C with 5% CO2. After the incubation period of 48 hours, media was removed from the culture wells and fresh media supplemented with 5% severely patient plasma or healthy plasma was added to the apical side of the insert and incubated for 4 or 6 hours. Cells sub-cultured for each experiment were supplemented with media according to manufacturer protocol. Following seeding of cells on inserts for each experiment and in order to avoid alterations in results due to differences in media type, all experimental cell cultures were supplemented with a 1:1:1 mixture of endothe- lial, astrocyte, and pericyte media. The severely injured patient plasma for each experimental replicate was composed of pooled patient plasma from patients with non-TBI traumatic inju- ries ISS> 15 and BD<-6. Conditioned Media (CM) Model: Endothelial, astrocyte, and pericyte monocultures, and tricultures were incubated with or without 5% severely injured patient plasma for 1 hour to produce cell type specific conditioned media. Conditioned media in the presence and absence of trauma plasma respectively (CM+TP; CM-TP) from each cell type was harvested. In order to interrogate the effect of other cell types on endothelial cells that have been exposed to severely injured patient plasma; Endothelial monocultures were supplemented with severely injured patient plasma to 5% working concentration followed immediately by supplementa- tion of cell type specific conditioned media (CM+TP; CM-TP) to 10% CM working PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 4 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma concentration. Naïve control endothelial cells were allowed to grow in their normal growth media throughout the experiment in absence of conditioned media and plasma. RNA extraction and cDNA synthesis RNA extraction was performed using a Qiagen RNeasy extraction kit. RNA samples were con- centrated using a speed vac, 40 C for 60 minutes, and quantified spectrophotometrically for cDNA synthesis using a Biotek Synergy H1 microplate reader. 1.5 μg total RNA was used for each cDNA synthesis reaction with QuantaBios qScript cDNA supermix using an adapted form of QuantaBios protocol. 60uL reactions were performed using the following protocol run on a Bio-Rad thermal cycler: 5 minutes at 25 C, 30 min at 42 C, 5 min at 85 C, cool down and hold at 4 C. qPCR qPCR protocols were adapted and optimized from recommended Quantstudios protocols. All RT-qPCR was performed on a QuantStudio 3 Real-Time PCR System (Applied Biosystems). All primer probe pairs were purchased from Applied Biosystems- GAPDH (Hs99999905_m1, Assay Location 229), CLDN3 (Hs00265816_s1, Assay Location 807), CLDN5 (Hs00533949_s1, Assay Location 1713), and ZO-1 (Hs01551861_m1, Assay Location 1762) for TaqMan based qPCR. The master mix used was comprised of 10uL of TaqMan fast advanced master mix (Applied Biosystems), 1uL of combined primer probes, and 3uL of RNAse free water per reac- tion. 6uL cDNA mix from cDNA synthesis was added to each well containing 14uL of master mix. The 20uL reactions were run on a 96 well plate in a Quantstudios 3 RT-PCR system with the following protocol: 2 minutes at 50 C, 10 minutes at 95 C, 20 seconds at 95 C and 1 minute at 60 C for 40 cycles. Ct values of the gene of interest were normalized to endogenous GAPDH control. Fold changes were produced from normalized Ct values compared to normalized experimental controls Whole cell lysate processing Following the removal of the media, each cell insert was washed with 1X PBS. Cells were then trypsinized with 0.25% Trypsin EDTA for 10 mins. The cell pellet was collected by centrifuga- tion at 100 g for 5 mins. Pellets were washed in 1X PBS three times and suspended in 50 μL 1X PBS. The whole cell lysate (WCL) was prepared as follows. The suspended pellets were placed in a Diagenode Biorupter for 7.5 minutes with sustained pulse intervals, 30 second on, 30 sec- ond off. The Lysate was then spun down at 10,000 rpm for 5 minutes. The supernatant was col- lected, and the pellet discarded. Protein concentrations in cell lysates were measured using Bradford assay. A standard curve was created by diluting 2mg/mL of BSA to 50, 25, 20, 10, 5, and 2.5 ug/mL in nanopure water. Samples were diluted 1:150; 150uL of the standards and diluted sample were added to a flat bottom 96 well plate in duplicate along with 150uL of Coo- massie plus protein reagent and incubated at room temperature for 10 min. Optical Density was measured at 595 nm using a Biotek Synergy H1 microplate reader. Protein concentration was determined using a 4 parametric nonlinear regression standard curve. ELISA Quantification of CLDN3 and CLDN5 in whole cell lysates were performed using a sandwich ELISA kit in accordance with the manufacturer’s protocol. (Antibodies Online Sandwich ELISA Assay Kits for CLDN3, ABIN6954828, and CLDN5, ABIN6962352) 4ug of WCL per sample was prepared and diluted in 200uL of PBS and added to the ELISA wells per PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 5 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma manufacturer’s protocol. Optical Density was measured using a Biotek Synergy H1 microplate reader at 450 nm. A 4-parametric nonlinear regression standard curve was used to determine protein concentration. The fold change was normalized against control samples. Data analysis GraphPad Prism 9 and Microsoft Excel 2016 were used for statistical analysis. qRT-PCR gene expression analysis was calculated using the 2-ΔΔCt method (delta-delta Ct method). Control values were set to a value of 1 compared to sample values following 2-ΔΔCt calculations. All data was represented as fold change over control. Statistical analysis was performed with GraphPad Prism 9 software using a two-way ANOVA test with Tukey’s multiple comparison post-hoc analysis. Experimental samples for non-conditioned media runs were compared against their own naïve controls. Experimental samples for conditioned media experiments were compared to both their naïve control and to each condition type respectively (CM+TP; CM-TP). P values of less than 0.05 were considered statistically significant. Results Severely injured patient plasma induces differential transcription expression of CLDN3 and CLDN5 We initially examined transcriptional responses of junctional proteins in the BBB in response to ex vivo trauma plasma. The pooled plasma used in the transcription studies had a mean ISS of 31.2 and BD of -11.5. Taqman based quantitative RT-PCR was used to determine the tran- scriptional expression of CLDN3, CLDN5, and ZO-1. Endothelial, astrocyte, and pericyte monocultures, and the triculture showed no significant change in mRNA transcriptional expression of CLDN3, CLDN5, or ZO-1 following incubation with healthy plasma (Fig 1A– 1C) Endothelial monocultures, had significant increases in transcriptional expression of tight junction protein CLDN3 (3.98 fold increase vs. control, p<0.001) and CLDN5 (3.6 fold increase vs. control, p<0.001) (Fig 1A and 1B) following a four hour incubation with trauma plasma from severely injured patients. This fold increase vs. control in expression of CLDN3 was not observed in astrocyte monoculture, pericyte monoculture, or the triculture. (Fig 1A) Similarly, CLDN5 transcriptional response also did not change vs. control values in astrocyte monoculture, pericyte monoculture, or the triculture. (Fig 1B) ZO-1 expression did not show any significant change vs. control for cells grown in monoculture or in the triculture model. (Fig 1C) To further expound upon these findings, a conditioned media experiment was per- formed to delineate the specific cell types contributing to the downregulation of the tight junc- tion proteins in the triculture vs. the endothelial monoculture. Addition of astrocyte, pericyte, or triculture conditioned media suppress endothelial monoculture transcriptional upregulation of CLDN3 and CLDN5 A conditioned media exchange study was performed in order to delineate the specific cell types of the neuro vascular unit contributing to transcriptional downregulation of CLDN3 and CLDN5 observed in the triculture model and whether that contribution is due to the release of soluble factors from those cells. (Fig 1A and 1B). The conditioned media exchange studies comprised of the following experimental group. Endothelial monocultures were incubated with conditioned media from a) endothelial monoculture (ECM+), b) astrocytes monoculture (ACM+), c) pericyte monoculture (PCM+) or d) Triculture (TCM+) in the presence or absence severely injured patient plasma. (CM+TP; CM-TP). The transcriptional PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 6 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 7 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma Fig 1. CLDN5, CLDN3, and ZO-1 mRNA expression following a 4 hour incubation with a working concentration of 5% plasma from healthy patients (HP) or 5% Trauma Plasma from severely injured patients (TP). (A) CLDN3 mRNA expression: The Endothelial monoculture incubated with TP showed a 3.98-fold increase vs. control, p<0.001, n = 6; Astrocyte monocultures, Pericyte monocultures, and the triculture did not show any significant change vs. control. All cell cultures incubated with HP showed no significant change vs. control. (B) CLDN5 mRNA expression: Endothelial monoculture showed a 3.51-fold increase vs. control, p<0.001, n = 6; Astrocyte monocultures, Pericyte monocultures, and the triculture did not show any significant fold change vs. control. All cell cultures incubated with HP showed no significant change vs. control. (C) ZO-1 mRNA expression: ZO-1 mRNA expression did not change significantly vs. control for either the monoculture or tricultures, n = 5. All cell cultures incubated with HP showed no significant change vs. control. Six experimental replicates were performed n = 6. �� Denotes p<0.01 �� Denotes p<0.001. https://doi.org/10.1371/journal.pone.0270817.g001 downregulation of CLDN3 and CLDN5 seen in the triculture model was also observed in endothelial monocultures following transfer of astrocyte and pericyte conditioned media. The endothelial monoculture which received ECM+/CM+TP; showed increased expression of CLDN3 (4.91-fold increase vs. control, p<0.01) and CLDN5 (2.94-fold increase vs. control, p<0.0001) (Fig 2A and 2B). Endothelial monocultures receiving CM-TP: ECM+, ACM+, PCM+, or TCM+ did not show any significant increase in mRNA expression for CLDN3 or CLDN5. (Fig 2A and 2B) Similarly, endothelial monocultures receiving CM+TP: ACM+, PCM+, or TCM+ did not show any significant increase in mRNA expression for CLDN3 and CLDN5. (Fig 2A and 2B) Endothelial monocultures that received either CM+TP or CM-TP: ACM+, PCM+ or TCM+ did not show any significant change in the expression of ZO-1 when compared to controls (Fig 2C). Severely injured patient plasma induces differential protein expression of CLDN5 In order to determine if the changes in the transcriptional response of the tight junction pro- teins CLDN3 and CLDN5 in the endothelial monoculture and triculture were also reflected in protein translation, sandwich ELISAs were performed. The pooled plasma used in the transla- tion studies had a mean ISS of 35.6 and BD of -18.36. The protein expression of CLDN3 and CLDN5 were assayed in whole cell lysate. The Endothelial monocultures show a significant increase in CLDN5 protein expression following both a 4 hour incubation (5.12 fold increase vs. control, p<0.001) and 6 hour incubation (2.58 fold change vs. control, p<0.01) with 5% severely injured patient plasma. (Fig 3B) We observed no significant fold changes occurred for CLDN5 protein expression in the Triculture, nor were there any significant fold changes in CLDN3 protein expression for either the endothelial monoculture or triculture (Fig 3A and 3B). Addition of astrocyte or pericyte conditioned media suppress endothelial monoculture protein upregulation of CLDN5 The translation of CLDN3 and CLDN5 proteins from the mRNA transcripts from endothelial monocultures following conditioned media exchange was assessed from whole cell lysate. The conditioned media exchange studies had the same experimental setup as the previous media exchange experiment (Fig 2) We observed no significant difference between the condition types or time points for any tested sample for CLDN3 protein expression. (Fig 4A) Endothelial monocultures receiving 6 hour ECM+/CM+TP presented with a significant increase in CLDN5 protein expression compared to endothelial monocultures receiving 6Hr ACM+/CM +TP (4.67 vs 3.03-fold change vs. control respectively, p<0.01) and 6Hr PCM+/CM+TP (4.67 vs. 3.26-fold change vs. control respectively, p< 0.05). (Fig 4B) Endothelial monocultures receiving 6Hr TCM+/CM+TP presented with a significant increase in CLDN5 protein PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 8 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 9 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma Fig 2. CLDN3, CLDN5, and ZO-1 mRNA expression in endothelial cell monocultures following a 4 hour incubation with cell type specific conditioned media. Endothelial Monoculture Conditioned Media, (ECM+), Astrocyte Monoculture Conditioned Media, (ACM+), Pericyte Monoculture Conditioned Media, (PCM+), and Triculture Conditioned Media, (TCM+), were prepared by either a 1-hour incubation with 5% severely injured patient plasma (CM+TP) or left as is with no plasma addition (CM-TP). All conditioned media was added to endothelial cell monocultures activated with severely injured patient plasma to a working concentration of 5%. Final working concentration of severely injured patient plasma and conditioned media in endothelial monocultures receiving conditioned media was 5% and 10% respectively. Each condition was compared to an endothelial cell monoculture naïve control (CM+TP—and CM-TP -). CM+TP wells were compared against one another to determine the differential effect each cell type’s conditioned media had. (A) CLDN3 mRNA expression: The endothelial monoculture receiving ECM+/CM+TP showed a 4.96-fold change increase vs. control, p<0.01, n = 2; ECM+/CM+TP produced a significant fold change increase in CLDN3 mRNA expression compared to ACM+/ CM+TP, (p<0.01), PCM+/CM +TP, (p<0.01), and TCM+/CM+TP, (p<0.01). Endothelial monocultures receiving ACM+, PCM+, or TCM+ did not show significant fold change vs. control for either CM-TP or CM+TP. Additionally, endothelial monocultures receiving ECM+/CM-TP showed no significant change in transcription. (B) CLDN5 mRNA expression: The Endothelial monoculture ECM+/CM+TP showed a 2.94-fold change increase vs. control, p<0.0001, n = 2; ECM+/CM +TP produced a significant fold change increase in CLDN5 mRNA expression compared to ACM+/CM+TP, (p<0.0001), PCM+/CM+TP, (p<0.0001), and TCM+/CM+TP (p<0.0001). Endothelial monocultures receiving ACM +, PCM+, or TCM+ did not show significant fold change vs. control for either CM+TP or CM-TP. Endothelial monocultures receiving ECM+/CM-TP showed no significant change in transcription. (C) ZO-1 mRNA expression: Endothelial monocultures receiving ACM+, ECM+, PCM+, or TCM+ did not show significant fold change vs. control for either CM+TP or CM-TP. Two experimental replicates of the conditioned media experiment were performed, n = 2. �� Denotes p<0.01. �� Denotes p<0.0001. https://doi.org/10.1371/journal.pone.0270817.g002 expression compared to endothelial monocultures receiving 6Hr ACM+/CM+TP (5.48 vs 3.03-fold change vs. control respectively, p<0.001) and 6Hr PCM+/CM+TP (5.48 vs. 3.26-fold change vs. control respectively, p<0.001). (Fig 4B) We observed no significant difference between the condition types or time points for any other tested sample (Fig 4B). Discussion Tight junction proteins are central to the maintenance of BBB integrity and increases in BBB permeability have a significant impact on patient outcomes [4,6]. BBB dysfunction has had lit- tle inquiry in the context of non-TBI critically injured patients despite their presentation of clinical sequela of brain injury. We demonstrate here that ex vivo plasma from severely injured non-TBI patients causes BBB breakdown accompanied by transcriptional and translational responses of central tight junction proteins of the neurovascular unit required to maintain high resistance across the BBB. Our data shows that the causes of non-TBI injured patient’s presentation of BBB dysfunction center on the expression, or lack thereof, of tight junction proteins key to BBB permeability. CLDN3, CLDN5, and ZO-1 are central to regulation of BBB permeability [16,17,25–29], and are essential in maintaining barrier integrity [7–9]. Loss of expression of these tight junc- tion proteins compromises barrier integrity by increasing transcellular permeability across the vasculature. Claudin 5 is implicated as a singular driver in multiple disease states including some neuroinflammation states [17] and plays major roles in disease states from carcinomas to schizophrenia and other endothelial barrier dysfunctions [25–27,30]. Following traumatic injuries, an overall downregulation and dysfunction of CLDN5 is observed, coupled directly with BBB permeability dysfunction [4,28,31–33]. There is also evidence that increased exoge- nous CLDN5 expression may increase the tightness of the junctions [8,33]. This made CLDN5 the primary target of interest for the transcriptional regulation among the three main cell types of the BBB following incubation with severely injured patient plasma. The importance of Claudin 3 as a major tight junction protein and its role in BBB permeability is not yet clearly defined [16,34]. However, because CLDN3 is present in the tight junction of brain endothelial cells and CLDN3’s definitive role in the BBB remains unclear, CLDN3 was PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 10 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma Fig 3. CLDN5, and CLDN3 protein expression following a 4- or 6-hour incubation with 5% plasma from severely injured patients. (A) CLDN3 protein expression: We observed no significant change in fold change of CLDN3 expression was observed in either the endothelial monoculture or triculture, n = 2 (B) CLDN5 protein expression: Endothelial monoculture showed a 5.12- and 2.58-fold change increase vs. control following 4 and 6 hours of incubation respectively with 5% severely injured plasma. p<0.001, p<0.05, n = 2; There is a significant difference between the fold change at 4 PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 11 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma hours of incubation compared to 6 hours of incubation following 5% severely injured patient plasma in the endothelial monoculture, p<0.01. The triculture did not show any significant fold change vs. control. One ELISA was performed with the combined whole cell lysate of three experimental replicates, n = 2. � Denotes p<0.05, �� Denotes p<0.01, �� Denotes p<0.001. https://doi.org/10.1371/journal.pone.0270817.g003 chosen as an ideal candidate to focus on in an in vitro traumatic injury model. Finally, ZO-1 was analyzed because of its central role in the anchoring of tight junction proteins to the actin filaments. ZO-1’s central role in facilitating foundational structure to tight junctions, coupled with BBB permeability disruptions following its re-localization or degradation from MMPs fol- lowing injury, made ZO-1 an ideal candidate to include in this study [29,35]. Our studies using an in vitro triculture model suggest inhibitory signals from astrocytes and pericytes induce transcriptional downregulation of CLDN3 and CLDN5 and translational downregulation of CLDN5 in brain endothelial cells within four hours of exposure to severely injured patient plasma. (Figs 1A, 1B and 4B) Furthermore, our observations suggest that astro- cytes and pericytes in separate monocultures release sufficient soluble factors to elicit inhibi- tion of tight junction expression in endothelial cells. These observations were upheld when endothelial cell monocultures were treated with conditioned media from monocultures of astrocyte or pericytes incubated with 5% severely injured patient plasma. (Figs 2A, 2B, 4A and 4B) The observed inhibition of the tight junction proteins within the triculture following incu- bation for four hours in severely injured patient plasma is specific to the transcription of CLDN3 and CLDN5 and the translation of CLDN5. We also demonstrate that the use of a healthy plasma negative controls made no significant impact on tight junction expression compared to our naïve control. Due to the insignificance of our healthy plasma’s impact on tight junction expression, the decision was made, with both experimental design and cost in mind, to exclude this condition from future trials. Likewise, ZO-1 showed no differences in transcriptional response (Figs 1C and 2C), leading to its exclusion from translation studies. Finally, CLDN3 expression did not show a robust translational response, as seen with CLDN5, compared to the observed transcriptional changes. (Fig 3A). This does not necessarily decrease the possibility of astrocyte or pericyte derived initial inhibitory signaling leading to decreased endothelial cell tight junction expression; but could support literature that suggests CLDN3 plays a less significant role in restoration and/or maintenance of BBB permeability than that of other tight junction proteins [36]. We observed a robust and significant transcriptional and translational expression of CLDN5 in endothelial monocultures following incubation with severely injured patient plasma which was not observed in tricultures incubated with severely injured plasma. The finding that plasma from severely injured patients downregulates expression of critical tight junction proteins within a triculture model is compelling and is suggestive of a critical role that pericytes and astrocytes play in normal physiological responses to severe traumatic injury via initiation of inhibitory cross talk between the three cell types. This inhibition of CLDN5 expression is further supported by conditioned media experiments. Endothelial monocultures which received conditioned media from astrocytes and pericytes following 6 hours of incuba- tion were observed to have significantly less CLDN5 protein expression compared to endothe- lial monocultures receiving conditioned media from endothelial monocultures or the triculture. A possible explanation for the inhibition of CLDN5 protein expression in endothe- lial monocultures receiving astrocyte conditioned media that was not observed in endothelial monocultures receiving triculture conditioned media is unmitigated release of Angiopoietin 2 (ANG2) by activated astrocytes. ANG2, a soluble mediator and a marker of endothelial dys- function [37], is present in copious amounts in trauma plasma and is associated with worse PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 12 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma Fig 4. CLDN3 and CLDN5 protein expression in endothelial cell monocultures following a transfer of conditioned. Endothelial Monoculture Conditioned Media, (ECM+), Astrocyte Monoculture Conditioned Media, (ACM+), Pericyte Monoculture Conditioned Media, (PCM+), and Triculture Conditioned Media, (TCM+), were prepared by either a 1 hour incubation with 5% severely injured patient PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 13 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma plasma (CM+TP) or left as is with no plasma additions made (CM-TP). Final working concentration of severely injured patient plasma and conditioned media in endothelial monocultures receiving conditioned media was 5% and 10% respectively. Each condition was compared to its Naïve control (CM+TP—and CM-TP -) receiving no conditioned media and no severely injured patient plasma to produce fold changes. CM+TP were compared against one another to determine the differential effect amongst the conditioned media types. Both 4 and 6 hour incubation times were assessed. � Denotes p<0.05, �� Denotes p<0.01, �� Denotes p<0.001 (9) CLDN3 protein expression: Endothelial monocultures did not present with any significant changes in CLDN3 protein expression for all conditions. (10) CLDN5 protein expression: All conditions tested resulted in a significant increase in the fold change of CLDN5 expression compared to the naïve control (not shown). The Endothelial monoculture receiving 6Hr ECM+/CM+TP showed a significant increase in fold change vs. control compared to 6Hr ACM +/CM+TP (p<0.01), and 6Hr PCM+/CM+TP (p<0.05) n = 2; The endothelial monoculture receiving 6Hr TCM+/CM+TP showed a significant increase in fold change compared to 6Hr ACM+/CM+TP (p<0.001), and 6Hr PCM+/CM+TP (p<0.001) n = 2. https://doi.org/10.1371/journal.pone.0270817.g004 clinical outcomes through destabilization of endothelial barriers, increasing leakage and edema. ANG2 acts as an antagonist to Angiopoietin 1 (ANG1) for its primary ligand, the receptor tyrosine kinase TIE2 [38]. TIE2 activation leads to dimerization and auto-phosphory- lation of its intracellular domain, subsequently activating the PI3K/AKT pathway [39–41]. Inactivation of the PI3K/AKT pathway via ANG2 leads to downregulation of CLDN5 expres- sion through recruitment of a repressor complex to CLDN5’s silencer binding domain [41,42]. While this mechanism is understood in the context of TBI and other pathologies, it is unknown what mechanism causes BBB dysfunction due to non-TBI trauma; but it is sugges- tive from literature that increased ANG2 expression following injury does contribute in some capacity. This underlying mechanism for the observed CLDN5 transcriptional and transla- tional results will be analyzed in follow-up studies probing the causal link, if any, between ANG2 signaling and non-TBI trauma related BBB dysfunction. Conclusion and future direction Clinical observations by physicians treating non-TBI traumatically injured patients suggest TBI like BBB dysfunction occurs regardless of injury location but correlates directly with level of injury and shock; reinforcing the need to understand the underpinnings BBB dysfunction [43]. Our findings lend a possible explanation for non-brain injury trauma presenting with similar symptoms to TBI by affecting the expression of tight junction central to maintenance of BBB integrity within the neurovascular unit. Permeability increases following neuroinflammatory stimuli [44,45]. An emerging body of evidence suggests pathology after TBI is driven by alterations in cellular crosstalk between endothelial cells astrocytes and pericytes [46] via the release of multiple biomolecules of inter- est including angiopoietin 2, endothelin-1, tumor necrosis factor- alpha, and matrix metallo- protease-9 [33,46–50]. The impact of these molecular mediators and proteases on tight junction expression following trauma, and the subsequent role this modulation of tight junc- tion expression has on BBB permeability, is not yet fully understood. Our data is consistent with a physiological process designed to prevent infection and maxi- mize repair of cellular structures within the interneuronal space [8,15,50–53]. Delaying BBB permeability restoration through the inhibition of tight junction proteins activate physiological responses to neuroinflammatory factors such as TNF-alpha, IL1-B, IL-6 and other cytokines released during injury and allows for the localization and infiltration of leukocytes across the barrier through ICAM and VCAM dependent processes [51,52]. This infiltration of leukocytes coupled with increased flow of other soluble factors between the vasculature and interneuronal space may provide better neuronal repair following minor brain injuries [8,15,51]. However, when the injury is more severe, this delay in endothelial barrier restoration prevents neuro- logic recovery and may lead to secondary brain injury. Our results, while observed in a physio- logical presentation closer to that of the BBB than other in vitro models, requires future in vivo work to strengthen these findings. This is due to inherent limitations of in vitro studies such as PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 14 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma lack of shear stress flow, absence of glial cells the complexities of the microenvironment the cells of the neurovascular unit encounter [54,55]. We do believe, however, that isolating the cells in vitro gives a better controlled environment to probe the exact cellular response to spe- cific conditions introduced via the severely injured patient plasma that cannot be controlled for in vivo [55]. The underlying molecular mechanisms leading to increased blood brain permeability fol- lowing traumatic injury are poorly enumerated and a comprehensive understanding of this process would provide novel approaches in designing future interventions to prevent the adverse effects of secondary brain injury. The findings in this study will guide our future work to include transcriptional and translational analysis of other tight junction proteins central to BBB permeability and identify soluble factors in plasma that cause perturbations in tight junc- tion transcription and translation. We expect that enhanced understanding of this astrocyte, pericyte, and endothelial cellular crosstalk will help identify new therapies for BBB pathologies following trauma and will provide useful insight in designing specific interventional strategies in patients with severe BBB dysfunction while creating a path forward for implementation of these therapies in personalized healthcare for traumatically injured patients. Supporting information S1 Data. Underlying data and statistics used to generate Fig 1. (XLSX) S2 Data. Underlying data and statistics used to generate Fig 2. (XLSX) S3 Data. Underlying data and statistics used to generate Fig 3. (XLSX) S4 Data. Underlying data and statistics used to generate Fig 4. (XLSX) Acknowledgments We thank the University Of Colorado Department Of Surgery, the division of GITES, and the Trauma research lab team for facilitating our research. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or other sponsors of the project. Author Contributions Conceptualization: Sanchayita Mitra, Margot Debot, Arthur Stem, Mitchell J. Cohen. Data curation: Preston Stafford, Sanchayita Mitra, Margot Debot, Patrick Lutz, Arthur Stem. Formal analysis: Preston Stafford, Sanchayita Mitra, Mitchell J. Cohen. Funding acquisition: Mitchell J. Cohen. Investigation: Preston Stafford, Sanchayita Mitra, Margot Debot, Patrick Lutz, Arthur Stem, Mitchell J. Cohen. Methodology: Preston Stafford, Sanchayita Mitra. Project administration: Preston Stafford, Sanchayita Mitra, Margot Debot, Mitchell J. Cohen. Resources: Preston Stafford, Sanchayita Mitra, Arthur Stem, Patrick Hom, Mitchell J. Cohen. PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 15 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma Software: Preston Stafford, Patrick Lutz. Supervision: Sanchayita Mitra, Margot Debot, Mitchell J. Cohen. Validation: Preston Stafford, Sanchayita Mitra, Mitchell J. Cohen. Visualization: Preston Stafford, Mitchell J. Cohen. Writing – original draft: Preston Stafford, Sanchayita Mitra, Mitchell J. Cohen. Writing – review & editing: Preston Stafford, Sanchayita Mitra, Margot Debot, Patrick Lutz, Arthur Stem, Jamie Hadley, Patrick Hom, Terry R. Schaid, Mitchell J. Cohen. References 1. Faul M, Wald MM, Xu L, Coronado VG. Traumatic brain injury in the United States; emergency depart- ment visits, hospitalizations, and deaths, 2002–2006. 2. Taylor C. A., Bell J. M., Breiding M. J., & Xu L. (2017). Traumatic brain injury–related emergency depart- ment visits, hospitalizations, and deaths—United States, 2007 and 2013. MMWR Surveillance Summa- ries, 66(9), 1. Price, L., Wilson, C., & Grant, G.: Blood–brain barrier pathophysiology following traumatic brain injury. In Translational research in traumatic brain injury. CRC Press/Taylor and Francis Group. 3. Beggs S, Liu XJ, Kwan C, Salter MW. Peripheral nerve injury and TRPV1-expressing primary afferent C-fibers cause opening of the blood-brain barrier. Molecular pain. 2010 Dec; 6(1):1–2. https://doi.org/ 10.1186/1744-8069-6-74 PMID: 21044346 4. Price L, Wilson C, Grant G. Blood–brain barrier pathophysiology following traumatic brain injury. Trans- lational research in traumatic brain injury. 2016. 5. Zhao Z., Nelson A. R., Betsholtz C., & Zlokovic B. V.: Establishment and dysfunction of the blood-brain barrier. Cell, 163(5), 1064–1078. https://doi.org/10.1016/j.cell.2015.10.067 PMID: 26590417 6. Alluri H, Wiggins-Dohlvik K, Davis ML, Huang JH, Tharakan B. Blood–brain barrier dysfunction following traumatic brain injury. Metabolic brain disease. 2015 Oct; 30(5):1093–104. 7. Milatz S, Krug SM, Rosenthal R, Gu¨ nzel D, Mu¨ller D, Schulzke JD, et al. Claudin-3 acts as a sealing component of the tight junction for ions of either charge and uncharged solutes. Biochimica et Biophy- sica Acta (BBA)-Biomembranes. 2010 Nov 1; 1798(11):2048–57. 8. Ohtsuki S, Sato S, Yamaguchi H, Kamoi M, Asashima T, Terasaki T. Exogenous expression of claudin- 5 induces barrier properties in cultured rat brain capillary endothelial cells. Journal of cellular physiology. 2007 Jan; 210(1):81–6. https://doi.org/10.1002/jcp.20823 PMID: 16998798 9. Itoh M, Furuse M, Morita K, Kubota K, Saitou M, Tsukita S. Direct binding of three tight junction-associ- ated MAGUKs, ZO-1, ZO-2, and ZO-3, with the COOH termini of claudins. The Journal of cell biology. 1999 Dec 13; 147(6):1351–63. https://doi.org/10.1083/jcb.147.6.1351 PMID: 10601346 10. Vajtr DA, Benada OL, Kukačka J, Průsˇa R, Housˇtˇava L, Toupalı´k PA, et al. Correlation of Ultrastructural Changes of Endothelial Cells and Astrocytes Occurring during Blood Brain Barrier Damage after Trau- matic Brain Injury with Biochemical Markers of Blood Brain Barrier Leakage and Inflammatory Response. Physiological research. 2009 Apr 1; 58(2). 11. Barzo´ P, Marmarou A, Fatouros P, Corwin F, Dunbar J. Magnetic resonance imaging—monitored acute blood-brain barrier changes in experimental traumatic brain injury. Journal of neurosurgery. 1996 Dec 1; 85(6):1113–21. https://doi.org/10.3171/jns.1996.85.6.1113 PMID: 8929504 12. Strbian D, Durukan A, Pitkonen M, Marinkovic I, Tatlisumak E, Pedrono E, et al. The blood–brain barrier is continuously open for several weeks following transient focal cerebral ischemia. Neuroscience. 2008 Apr 22; 153(1):175–81. https://doi.org/10.1016/j.neuroscience.2008.02.012 PMID: 18367342 13. Shapira Y, Setton D, Artru AA, Shohami E. Blood-brain barrier permeability, cerebral edema, and neu- rologic function after closed head injury in rats. Anesthesia and analgesia. 1993 Jul 1; 77(1):141–8. https://doi.org/10.1213/00000539-199307000-00028 PMID: 8317722 14. Winkler EA, Bell RD, Zlokovic BV. Central nervous system pericytes in health and disease. Nature neu- roscience. 2011 Nov; 14(11):1398–405. https://doi.org/10.1038/nn.2946 PMID: 22030551 15. Nitta T, Hata M, Gotoh S, Seo Y, Sasaki H, Hashimoto N, et al. Size-selective loosening of the blood- brain barrier in claudin-5–deficient mice. The Journal of cell biology. 2003 May 12; 161(3):653–60. https://doi.org/10.1083/jcb.200302070 PMID: 12743111 16. Steinemann A, Galm I, Chip S, Nitsch C, Maly IP. Claudin-1,-2 and-3 are selectively expressed in the epithelia of the choroid plexus of the mouse from early development and into adulthood while claudin-5 PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 16 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma is restricted to endothelial cells. Frontiers in neuroanatomy. 2016 Feb 22; 10:16. https://doi.org/10. 3389/fnana.2016.00016 PMID: 26941614 17. Greene C, Hanley N, Campbell M. Claudin-5: gatekeeper of neurological function. Fluids and Barriers of the CNS. 2019 Dec; 16(1):1–5. 18. Coyne CB, Gambling TM, Boucher RC, Carson JL, Johnson LG. Role of claudin interactions in airway tight junctional permeability. American Journal of Physiology-Lung Cellular and Molecular Physiology. 2003 Nov; 285(5):L1166–78. https://doi.org/10.1152/ajplung.00182.2003 PMID: 12909588 19. Coleman JR, Moore EE, Samuels JM, Ryon JJ, Nelson JT, Olson A, et al. Whole blood thrombin gener- ation is distinct from plasma thrombin generation in healthy volunteers and after severe injury. Surgery. 2019 Dec 1; 166(6):1122–7. https://doi.org/10.1016/j.surg.2019.07.014 PMID: 31522748 20. Quah S. International encyclopedia of public health. Academic Press; 2016 Oct 6. 21. Hodgman EI, Morse BC, Dente CJ, Mina MJ, Shaz BH, Nicholas JM, et al. Base deficit as a marker of survival after traumatic injury: consistent across changing patient populations and resuscitation para- digms. The journal of trauma and acute care surgery. 2012 Apr; 72(4):844. https://doi.org/10.1097/TA. 0b013e31824ef9d2 PMID: 22491595 22. Cohen MJ, Serkova NJ, Wiener-Kronish J, Pittet JF, Niemann CU. 1H-NMR-based metabolic signa- tures of clinical outcomes in trauma patients—beyond lactate and base deficit. Journal of Trauma and Acute Care Surgery. 2010 Jul 1; 69(1):31–40. 23. Hatherell K, Couraud PO, Romero IA, Weksler B, Pilkington GJ. Development of a three-dimensional, all-human in vitro model of the blood–brain barrier using mono-, co-, and tri-cultivation Transwell mod- els. Journal of neuroscience methods. 2011 Aug 15; 199(2):223–9. https://doi.org/10.1016/j.jneumeth. 2011.05.012 PMID: 21609734 24. Stone NL, England TJ, O’Sullivan SE. A novel transwell blood brain barrier model using primary human cells. Frontiers in cellular neuroscience. 2019 Jun 6; 13:230. https://doi.org/10.3389/fncel.2019.00230 PMID: 31244605 25. Liebner S, Fischmann A, Rascher G, Duffner F, Grote EH, Kalbacher H, et al. Claudin-1 and claudin-5 expression and tight junction morphology are altered in blood vessels of human glioblastoma multi- forme. Acta neuropathologica. 2000 Jul; 100(3):323–31. https://doi.org/10.1007/s004010000180 PMID: 10965803 26. Paschoud S, Bongiovanni M, Pache JC, Citi S. Claudin-1 and claudin-5 expression patterns differenti- ate lung squamous cell carcinomas from adenocarcinomas. Modern pathology. 2007 Sep; 20(9):947– 54. https://doi.org/10.1038/modpathol.3800835 PMID: 17585317 27. Jiao H, Wang Z, Liu Y, Wang P, Xue Y. Specific role of tight junction proteins claudin-5, occludin, and ZO-1 of the blood–brain barrier in a focal cerebral ischemic insult. Journal of Molecular Neuroscience. 2011 Jun; 44(2):130–9. https://doi.org/10.1007/s12031-011-9496-4 PMID: 21318404 28. Doherty CP, O’Keefe E, Wallace E, Loftus T, Keaney J, Kealy J, et al. Blood–brain barrier dysfunction as a hallmark pathology in chronic traumatic encephalopathy. Journal of Neuropathology & Experimen- tal Neurology. 2016 Jul 1; 75(7):656–62. https://doi.org/10.1093/jnen/nlw036 PMID: 27245243 29. Fanning AS, Jameson BJ, Jesaitis LA, Anderson JM. The tight junction protein ZO-1 establishes a link between the transmembrane protein occludin and the actin cytoskeleton. Journal of Biological Chemis- try. 1998 Nov 6; 273(45):29745–53. https://doi.org/10.1074/jbc.273.45.29745 PMID: 9792688 30. Greene C, Kealy J, Humphries MM, Gong Y, Hou J, Hudson N, et al. Dose-dependent expression of claudin-5 is a modifying factor in schizophrenia. Molecular psychiatry. 2018 Nov; 23(11):2156–66. https://doi.org/10.1038/mp.2017.156 PMID: 28993710 31. 32. Farrell M, Aherne S, O’Riordan S, O’Keeffe E, Greene C, Campbell M. Blood-brain barrier dysfunction in a boxer with chronic traumatic encephalopathy and schizophrenia. Clinical neuropathology. 2019 Mar 1; 38(2):51. https://doi.org/10.5414/NP301130 PMID: 30574863 J Wen, S Qian, Q Yang, L Deng, Y Mo, Y Yu. Overexpression of netrin-1 increases the expression of tight junction-associated proteins, claudin-5, occludin, and ZO-1, following traumatic brain injury in rats. Experimental and therapeutic medicine. 2014 Sep 1; 8(3):881–6. https://doi.org/10.3892/etm.2014. 1818 PMID: 25120618 33. Nag S, Venugopalan R, Stewart DJ. Increased caveolin-1 expression precedes decreased expression of occludin and claudin-5 during blood–brain barrier breakdown. Acta neuropathologica. 2007 Nov 1; 114(5):459–69. https://doi.org/10.1007/s00401-007-0274-x PMID: 17687559 34. Dias MC, Coisne C, Lazarevic I, Baden P, Hata M, Iwamoto N, et al. Claudin-3-deficient C57BL/6J mice display intact brain barriers. Scientific reports. 2019 Jan 18; 9(1):1–6. 35. Ishrat T, Sayeed I, Atif F, Hua F, Stein DG. Progesterone and allopregnanolone attenuate blood–brain barrier dysfunction following permanent focal ischemia by regulating the expression of matrix PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 17 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma metalloproteinases. Experimental neurology. 2010 Nov 1; 226(1):183–90. https://doi.org/10.1016/j. expneurol.2010.08.023 PMID: 20816826 36. Berndt P, Winkler L, Cording J, Breitkreuz-Korff O, Rex A, Dithmer S, et al. Tight junction proteins at the blood–brain barrier: far more than claudin-5. Cellular and molecular life sciences. 2019 May; 76 (10):1987–2002. https://doi.org/10.1007/s00018-019-03030-7 PMID: 30734065 37. Ganter MT, Cohen MJ, Brohi K, Chesebro BB, Staudenmayer KL, Rahn P, et al. Angiopoietin-2, marker and mediator of endothelial activation with prognostic significance early after trauma?. Annals of sur- gery. 2008 Feb 1; 247(2):320–6. https://doi.org/10.1097/SLA.0b013e318162d616 PMID: 18216540 38. Beck H, Acker T, Wiessner C, Allegrini PR, Plate KH. Expression of angiopoietin-1, angiopoietin-2, and tie receptors after middle cerebral artery occlusion in the rat. The American journal of pathology. 2000 Nov 1; 157(5):1473–83. https://doi.org/10.1016/S0002-9440(10)64786-4 PMID: 11073808 39. Kim M, Allen B, Korhonen EA, Nitschke´ M, Yang HW, Baluk P, et al. Opposing actions of angiopoietin-2 on Tie2 signaling and FOXO1 activation. The Journal of clinical investigation. 2016 Sep 1; 126 (9):3511–25. https://doi.org/10.1172/JCI84871 PMID: 27548529 40. Beard RS Jr, Hoettels BA, Meegan JE, Wertz TS, Cha BJ, Yang X, et al. AKT2 maintains brain endothe- lial claudin-5 expression and selective activation of IR/AKT2/FOXO1-signaling reverses barrier dysfunc- tion. Journal of Cerebral Blood Flow & Metabolism. 2020 Feb; 40(2):374–91. https://doi.org/10.1177/ 0271678X18817512 PMID: 30574832 41. Gu H, Wang YQ, Zhao CH, Zhong XM, Yang JG. The decrease of Tie-2 receptor phosphorylation in microvascular endothelial cells is involved in early brain injury after subarachnoid hemorrhage. Artery Research. 2018 Sep 1; 23:45–51. 42. Morini MF, Giampietro C, Corada M, Pisati F, Lavarone E, Cunha SI, et al. VE-Cadherin–Mediated Epi- genetic Regulation of Endothelial Gene Expression. Circulation research. 2018 Jan 19; 122(2):231–45. https://doi.org/10.1161/CIRCRESAHA.117.312392 PMID: 29233846 43. Beggs S, Liu XJ, Kwan C, Salter MW. Peripheral nerve injury and TRPV1-expressing primary afferent C-fibers cause opening of the blood-brain barrier. Molecular pain. 2010 Dec; 6(1):1–2. https://doi.org/ 10.1186/1744-8069-6-74 PMID: 21044346 44. Yang C, Hawkins KE, Dore´ S, Candelario-Jalil E. Neuroinflammatory mechanisms of blood-brain barrier damage in ischemic stroke. American Journal of Physiology-Cell Physiology. 2019 Feb 1; 316(2): C135–53. https://doi.org/10.1152/ajpcell.00136.2018 PMID: 30379577 45. Thal SC, Neuhaus W. The blood–brain barrier as a target in traumatic brain injury treatment. Archives of medical research. 2014 Nov 1; 45(8):698–710. https://doi.org/10.1016/j.arcmed.2014.11.006 PMID: 25446615 46. Burda J. E., Bernstein A. M., & Sofroniew M. V.: Astrocyte roles in traumatic brain injury. Experimental neurology, 275, 305–315. https://doi.org/10.1016/j.expneurol.2015.03.020 PMID: 25828533 47. Rosa JM, Farre-Alins V, Ortega MC, Navarrete M, Lopez-Rodriguez AB, Palomino-Antolin A, et al. TLR4-pathway impairs synaptic number and cerebrovascular functions through astrocyte activation fol- lowing traumatic brain injury. British Journal of Pharmacology. 2021 Apr 8. https://doi.org/10.1111/bph. 15488 PMID: 33830504 48. Gorina R, Font-Nieves M, Ma´ rquez-Kisinousky L, Santalucia T, Planas AM. Astrocyte TLR4 activation induces a proinflammatory environment through the interplay between MyD88-dependent NFκB signal- ing, MAPK, and Jak1/Stat1 pathways. Glia. 2011 Feb; 59(2):242–55. https://doi.org/10.1002/glia.21094 PMID: 21125645 49. Bhowmick S, D’Mello V, Caruso D, Wallerstein A, Abdul-Muneer PM. Impairment of pericyte-endothe- lium crosstalk leads to blood-brain barrier dysfunction following traumatic brain injury. Experimental neurology. 2019 Jul 1; 317:260–70. https://doi.org/10.1016/j.expneurol.2019.03.014 PMID: 30926390 50. Bai Y, Zhu X, Chao J, Zhang Y, Qian C, Li P, et al. Pericytes contribute to the disruption of the cerebral endothelial barrier via increasing VEGF expression: implications for stroke. PloS one. 2015 Apr 17; 10 (4):e0124362. https://doi.org/10.1371/journal.pone.0124362 PMID: 25884837 51. Lazear HM, Daniels BP, Pinto AK, Huang AC, Vick SC, Doyle SE, et al. Interferon-λ restricts West Nile virus neuroinvasion by tightening the blood-brain barrier. Science translational medicine. 2015 Apr 22; 7 (284):284ra59–. https://doi.org/10.1126/scitranslmed.aaa4304 PMID: 25904743 52. Camire RB, Beaulac HJ, Willis CL. Transitory loss of glia and the subsequent modulation in inflamma- tory cytokines/chemokines regulate paracellular claudin-5 expression in endothelial cells. Journal of neuroimmunology. 2015 Jul 15; 284:57–66. https://doi.org/10.1016/j.jneuroim.2015.05.008 PMID: 26025059 53. Honda M, Nakagawa S, Hayashi K, Kitagawa N, Tsutsumi K, Nagata I, et al. Adrenomedullin improves the blood–brain barrier function through the expression of claudin-5. Cellular and molecular neurobiol- ogy. 2006 Mar 1; 26(2):109–18. https://doi.org/10.1007/s10571-006-9028-x PMID: 16763778 PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 18 / 19 PLOS ONEEndothelial cell tight junction expression in non-TBI trauma 54. Haseloff RF, Dithmer S, Winkler L, Wolburg H, Blasig IE. Transmembrane proteins of the tight junctions at the blood–brain barrier: structural and functional aspects. InSeminars in cell & developmental biology 2015 Feb 1 (Vol. 38, pp. 16–25). Academic Press. 55. Helms HC, Abbott NJ, Burek M, Cecchelli R, Couraud PO, Deli MA, et al. In vitro models of the blood– brain barrier: an overview of commonly used brain endothelial cell culture models and guidelines for their use. Journal of Cerebral Blood Flow & Metabolism. 2016 May; 36(5):862–90. https://doi.org/10. 1177/0271678X16630991 PMID: 26868179 PLOS ONE \| https://doi.org/10.1371/journal.pone.0270817 July 5, 2022 19 / 19 PLOS ONE
10.1371_journal.pone.0245201	RESEARCH ARTICLE A new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Yanmei YangID Qi Zhang1,2 1,2, Xianqi HuID 1,2, Pei LiuID 1,2, Li Chen3, Huan Peng4, Qiaomei Wang1,2, 1 College of Plant Protection, Yunnan Agricultural University, Kunming, Yunnan Province, China, 2 State Key Laboratory for Conservation and Utilization of Bio-Resources in Yunnan, Yunnan Agricultural University, Kunming, Yunnan Province, China, 3 Wheat Research Institute, Shanxi Academy of Agricultural Sciences, Linfen, Shanxi Province, China, 4 State Key Laboratory for Biology of Plant Disease and Insect Pests, Institute of Plant Protection, Chinese Academy of Agricultural Science, Beijing, China xqh@ynau.edu.cn Abstract An unknown root-knot nematode was found at high density on grape roots collected from Yunnan Province. Morphometric traits and measurements, isozyme phenotypes, and molecular analysis clearly differentiated this nematode from previously described root-knot nematodes. This new species is described, illustrated and named Meloidogyne vitis sp. nov. The new species can be distinguished from other Meloidogyne spp. by a unique combina- tion of characters. Females display a prominent neck, an excretory pore is located on the ventral region between 23rd and 25th annule behind lips, an EP/ST ratio of approximately 2.5 (1.98–2.96), a perineal pattern with two large and prominent phasmids, and a labial disc fused with the medial lips to form a dumbbell-shaped structure. Males display an obvious head region, a labial disc fused with the medial lips to form a dumbbell-shaped structure, no lateral lips, a prominent slit-like opening between the labial disc and medial lips, a distinct sunken appearance of the middle of the medial lips, and four incisures in the lateral field. Second-stage juveniles are characterized by a head region with slightly wrinkled mark, a labial disc fused with the medial lips to form a dumbbell-shaped structure, a slightly sunken appearance of the middle of the medial lips, a slit-like amphidial openings between the labial disc and lateral lips, and four incisures in the lateral field. The new species has rare Mdh (N3d) and Est phenotypes (VF1). Phylogenetic analysis based on ITS1-5.8S-ITS2, D2D3 fragments of rDNA, and coxI and coxII fragments of mtDNA sequences clearly separated the new species from other root-knot nematodes, and the closest relative was Meloidogyne mali. Meloidogyne mali was collected for amplifying these sequences as mentioned above, which were compared with the corresponding sequences of new species, the result showed that all of these sequences with highly base divergence (48–210 base divergence). More- over, sequence characterized amplified region (SCAR) primers for rapid identification of this new species were designed. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Yang Y, Hu X, Liu P, Chen L, Peng H, Wang Q, et al. (2021) A new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan. PLoS ONE 16(2): e0245201. https://doi.org/ 10.1371/journal.pone.0245201 Editor: Ebrahim Shokoohi, University of Limpopo, SOUTH AFRICA Received: May 15, 2020 Accepted: December 19, 2020 Published: February 3, 2021 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0245201 Copyright: © 2021 Yang et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All sequences are available from the GenBank database (accession number: MN816222, MN816223, MN816224, MN816225, MN816226, MN814829, MN814830, PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 1 / 25 PLOS ONEMN814831, MT012386 and MT012387). Other relevant data are within the paper and its Supporting Information files. Funding: This work was supported by grants from the National Key Research and Development Project (Nos. 2018YFD0201202; 2017YFD0200601). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. A new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Introduction Root-knot nematodes (RKNs), belonging to family Heteroderidae, are the most important cat- egories of plant-parasitic nematode and parasitize a large number of plant species [1]. After being parasitized, plants often exhibit severely reduced production, causing significant eco- nomic losses. More than 90 species of RKNs have been described to date [2, 3], which are able to infect virtually any species of higher plant and have a near cosmopolitan distribution [4]. In recent years, new RKN species have been found in woody plants such as coffee, olive and kiwi- fruit [5–7]. Vitis vinifera (Vitaceae, Vitis L.) is one of the most widely grown fruit crops in many areas of the world [8]. It is a woody vine plant whose fruits can be eaten raw as a fresh fruit or made into dried fruit, juices and wine; thus, it is of great economic value. Grape cultivation is believed to have originated in Armenia, near the Caspian Sea in Russia, from where it spread westward to Europe and eastward to Iran and Afghanistan [9]; at present, it is widely grown in tropical, temperate and subtropical regions worldwide. China is home to the most abundant grape genetic resources in the world, where grape cultivation has been conducted for thou- sands [10]. Yunnan Province is one of the major provinces for the grape industry in China. As of the end of 2014, the total output value of grape in Yunnan exceeded 7 billion yuan; thus, grape plays an important role in the agricultural industry in this province [11]. However, vari- ous pathogens, including plant-parasitic nematodes, pose a serious threat to the production of grapes worldwide, and RKNs are one of the important factors restricting grape production [12, 13]. Root-knot nematode infestation of grape has been documented in southern Australia, South Africa, France, the United States and other countries [14]. Grapesroot system are well developed and is the target of RKN infection, and both seedlings and adult plants can be harmed. The destroyed fibrous roots and root hairs initially show slight swelling. In later stages, the diseased roots become rotten, which directly affects the absorption of water and nutrients by the root system and results in great reductions in grape yield and fruit quality. In Australia, almost all vineyards on sandy soils are infected by RKNs [15], and four species (Meloidogyne incognita, Meloidogyne arenaria, Meloidogyne javanica and Meloidogyne hapla) were found to damage grapes in southern Australia [16]. Meloidogyne incognita and M. hapla are common grape root pests [17], and Liu and Zhang (2017) reported that grapes from the Huaihai economic zone were infected by M. incognita [18]. Meloidogyne javanica is the pre- dominant RKN in Australian vineyards [12], and M. hapla is abundant and widespread in Washington’s semiarid vineyards [19]. Meloidogyne incognita, M. arenaria, M. javanica and M. hapla are the main species parasitizing grapes [14, 20]. However, three other Meloidogyne species, Meloidogyne nataliei, Meloidogyne ethiopica and Meloidogyne thamesi, also infect grapes [21–23]. RKNs can severely reduce grape yield, causing significant economic losses. Li et al. (2006) reported that M. incognita was the most common RKN in vineyards, where it could reduce the yield of susceptible grape varieties by approximately 80% and that of resistant grape varieties by approximately 40% [24]. Furthermore, RKNs can also interact with bacteria, fungi and other pathogens to further increase damage to grapes. Given that grapes are seriously damaged by RKNs, the accurate identification of pathogen species could be crucial for designing effective prevention and control strategies. However, in China, there have been few studies on grape RKNs in recent years, with problems such as unclear distributions, unclear species and a lack of prevention and control technologies. In addition, the RKNs that parasitize grapes are not completely clear, and some reported results may need to be further discussed. In Yunnan, we found a high density of RKN-infected grapes, and the pathogenic species was different from previously described RKNs. The morphological PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 2 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan characteristics, such as the perineal pattern, of this species were very similar to those of Meloi- dogyne mali. Therefore, morphological, biochemical and molecular biological methods were used to identify this unknown species. Materials and methods Nematode population Samples of grape roots and rhizosphere soils were collected from vineyards in Luliang County, Yunnan Province. Female and egg samples were extracted from the root tissues of grapes. Sec- ond-stage juveniles (J2s) were collected from hatching eggs. A portion of the J2s were inocu- lated on cucumber roots, and the other portion were heat-killed and fixed using a 4% solution of formaldehyde for morphological observation and measurement. Male samples were obtained from the cucumber roots. Some of the males were used for scanning electron micros- copy (SEM), and others were heat-killed and fixed using a 4% solution of formaldehyde for morphological observation and measurement. The J2 samples of M. mali (used for compari- son) were provided by Peng Huan, Institute of Plant Protection, Chinese Academy of Agricul- tural Sciences. Making perineal patterns Perineal patterns of female adults were made following the method of Xie Hui (2000) [25]. Specifically, female adults were selected from grape root-knot tissue under an anatomical microscope, and a hard plastic consisting of 45% lactic acid solution was used to make an impression of the perineal cuticular pattern with a scalpel. Then, the perineal pattern was cleaned with a 45% lactic acid solution, placed on a glass slide and covered with coverslip, using pure glycerine as a floating carrier. Light microscopy (LM) All nematode samples were observed and examined under a Carl Zeiss Axio Vert. A1 inverted microscope. All samples were measured using the de Man indices [26], and the measurements were expressed in micrometers. Scanning electron microscopy (SEM) The samples of female adults, males and J2s were prepared following the methods of Eisenback et al. (1980) [27] and Eisenback and Hirschmann (1979) [28]. The perineal pattern was made following the method of Eisenback et al. (1980) [27], with slight modification. Double fixation with 3.5% glutaraldehyde and 1% osmic acid was employed. Specifically, live samples of female adults, males and J2s were cleaned with ddH2O and fixed with a 3.5% glutaraldehyde solution for more than 48 h in a 4˚C refrigerator. After that, they were washed with phosphate-buffered saline (PBS) 3 times, fixed with 1% osmic acid for 2 h, washed with PBS 3 times, dehydrated in a graded ethanol series (30%-100%), critical-point dried, and coated with gold. Observation was performed under a Hitachi S-3000N scanning electron microscope (Japan). Isozyme phenotype electrophoresis Isozyme electrophoresis was carried out following the methods of Esbenshade and Trianta- phyllou (1985) [29] and Esbenshade and Triantaphyllou (1990) [30]. Phenotypes were observed for esterases (Est) and malate dehydrogenase (Mdh). Five young egg-laying females of M. vitis sp. nov. and five young egg-laying females from a previously identified population of M. javanica (used for comparison) were prepared and placed in microtubes containing PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 3 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Table 1. The primers used in the research. Primers code 18S 26S D2A D3B cox1F cox1R C2F3 1108 https://doi.org/10.1371/journal.pone.0245201.t001 Primer sequence (50-30) TTGATTACGTCCCTGCCCTTT TCCTCCGCTAAATGATATG ACAAGTACCGTGAGGGAAAGTTG TCGGAAGGAACCAGCTACTA TGGTCATCCTGAAGTTTATG CTACA ACATAATAAGTATCATG GGTCAATGTTCAGAAATTTGTGG TACCTTTGACCAATCACGCT References Vrain et al. (1992) De Ley et al. (1999) Trinh et al. (2019) Powers et al. (1993) 10 μL mixed liquor consisting of 20% sucrose, 2% Triton X-100 and 0.02% bromophenol blue, the nematodes were broken with a sterile dissecting needle and the enzyme solution can be used immediately or stored at -20˚C refrigerator until use. Electrophoresis was carried out in separating and stacking gels consisted of 7% and 3% polyacrylamide, respectively, 0.75 mm thick, with Tris-glycine buffer (PH8.7) in a Mini-PROTEAN1 Tetra Cell apparatus (Bio-Rad). Voltage was maintained at 80 volts for the first 30 minutes, the following was maintained at 150 volts of the separation period until the bromophenol blue dye had migrated to approxi- mately 0.5 cm ahead of the bottom of the gel. Gels was stained with Mdh stain solution for Mdh and with Est stain solution for Est, the preparation of Mdh and Est stain solution follow- ing the method of Esbenshade and Triantaphyllou (1985) [31]. DNA extraction, PCR amplification and sequencing DNA was extracted from a single female adult and a large number of J2s following the method described by Adam et al. (2007) [32] and stored in a -80˚C refrigerator until use. Two rDNA fragments (ITS1-5.8S-ITS2 and D2D3) and two mtDNA fragments (partial coxI and coxII 16S rRNA) of M. vitis sp. nov. and M. mali were amplified. The primer pairs 18S/26S [33], D2A/ D3B [34], cox1F/cox1R [5] and C2F3/1108 [35] were used to amplify the ITS1-5.8S-ITS2 and D2D3 fragments of rDNA and the coxI and coxII fragments of mtDNA, respectively. The primer sequences are listed in Table 1. All of the polymerase chain reactions (PCRs) were per- formed in 25.00 μL mixed solution containing template DNA (2.50 μL), 10× PCR buffer (Mg2 +, plus, 2.50 μL), dNTPs (mixture, 2.00 μL), forward and reverse primers (10 μmol/L, 1.00 μL respectively), Taq DNA polymerase (5 U/μL, 0.25 μL), and ddH2O (15.75 μL). All the reagents used in the PCRs were purchased from TransGen Biotech Company. The PCR amplification procedure is provided in Table 2. After the amplification reaction, 5.00 μL PCR product was mixed with 1.00 μL 6× loading buffer (purchased from TaKaRa Company) and electrophoresed in a 1% Tris-acetate-ethylenediaminetetraacetic acid (TAE)-buffered agarose gel. PCR products were excised from the gel and purified using the EasyPure Quick Gel Extraction Kit (purchased from TransGen Biotech Company). The recovered product was ligated with pmD18 cloning Table 2. The PCR amplification procedure of primers for the research. Primers Pre degeneration Response parameter (35 cycle) Final extension 18S/26S D2A/D3B cox1F/cox1R C2F3/1108 94˚C 4 min 94˚C 4 min 94˚C 4 min 94˚C 4 min https://doi.org/10.1371/journal.pone.0245201.t002 Degeneration 94˚C 30 s 94˚C 30 s 94˚C 30 s 94˚C 30 s Annealing 55˚C 45 s 60˚C 40 s 54˚C 30 s 51˚C 30 s Extension 72˚C 1 min 72˚C 1 min 72˚C 1 min 72˚C 1 min 72˚C 10 min 72˚C 10 min 72˚C 10 min 72˚C 10 min PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 4 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan vector (purchased from TaKaRa Company) and transformed into DH5α competent cells (pur- chased from TaKaRa Company). The positive clones were selected and sequenced. Phylogenetic analyses and sequence alignment The obtained sequences were compared with those from other nematodes available in the GenBank database using the BLAST homology search program. ITS1-5.8S-ITS2, D2D3 of rDNA, and partial coxI and coxII 16S rRNA of mtDNA sequences from Meloidogyne spp. were selected for phylogenetic reconstruction. The ITS1-5.8S-ITS2 (JX015432.1) sequence of Rotylenchus buxophilus and the D2D3 (AY589364.1) sequence of Ditylenchus halictus and the coxI (EF617356.1) sequence of Romanomermis wuchangensis and complete mitochondrial genome (FN313571.1) sequence of Radopholus similis were used as outgroup taxa. A phyloge- netic tree was generated based on the neighbor-joining (NJ) method in MEGA 5.1 to analyze the phylogenetic relationships and genetic distances of nematodes. The phylograms were boot- strapped 1000 times to assess the degree of support for phylogenetic analysis. DNAMAN software was used to compare the ITS1-5.8S-ITS2, D2D3 of rDNA, and partial coxI and coxII 16S rRNA of mtDNA sequences between M. vitis sp. nov. and M. mali, and analysis the sequence divergence. Designing SCAR primers The rDNA ITS1-5.8S-ITS2 sequences of M. vitis sp. nov. amplified in this research were sub- mitted to the GenBank database for BLAST alignment, and a specific sequence in this region was selected to design specific primers in Primer 5.0 software. The primers Mv-F (5-CTGGT TCAGGGTCATTTATAAAC-3) and Mv-R (5-TATACGCTTGTGTGGATGAC-3) were used for PCR amplification of M. vitis sp. nov. The PCRs were performed in a 25.00 μL mixed solution containing template DNA for M. vitis sp. nov. (2.50 μL), 10× PCR buffer (Mg2+, plus, 2.50 μL), dNTPs (mixture, 2.00 μL), forward and reverse primers (10 μmol/L, 1.00 μL respectively), Taq DNA polymerase (5 U/μL, 0.25 μL), and ddH2O (15.75 μL). The amplification procedure was as follows: 4 min at 94˚C, 35 cycles of 30 s at 94˚C, 30 s at 53˚C and 30 s at 72˚C, and a final incubation of 10 min at 72˚C. Amplification products were separated by electrophoresis in a 1% TAE-buffered agarose gel and visualized under ultraviolet light. Nomenclatural acts The electronic edition of this article conformed to the requirements of the amended Interna- tional Code of Zoological Nomenclature, and hence the new names contained herein are avail- able under that Code from the electronic edition of this article. This published work and the nomenclatural acts it contains have been registered in ZooBank, the online registration system for the ICZN. The ZooBank LSIDs (Life Science Identifiers) can be resolved and the associated information viewed through any standard web browser by appending the LSID to the prefix “http://zoobank.org/”. The LSID for this publication is: urn:lsid:zoobank.org:pub: CFA0651C- DDD9-4DA8-A5B2-AF24C7ED8699. The electronic edition of this work was published in a journal with an ISSN and has been archived and is available from the following digital reposi- tories: PubMed Central, LOCKSS. Results Meloidogyne vitis sp. nov. Yang, Hu, Liu, Chen, Peng, Wang & Zhang sp. nov. urn: lsid: zoo- bank. org: act: 0163D840-A867-4F03-9C9C-1F879452E34B. PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 5 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Disease symptoms More than 90% of the grape roots collected from vineyards in Luliang County, Yunnan Prov- ince, were seriously damaged by RKNs. The symptoms of lightly nematode-infected plants were not obvious. However, the severely nematode-infected plants presented symptoms of plant dwarfing, leaf yellowing and shedding, little fruit, declining and low growth. The roots were atrophied and distorted, with severe root knots and other symptoms. Both the axial roots and branch roots were damaged. The surface of the infected roots presented numerous galls with white or milky white eggs. Eggs occurred either on the outside of the root galls or within the root galls. The aged roots were rotten and had become necrotic. Adult female heads were found associated with the xylem (Fig 1A and 1B). Description of Meloidogyne vitis sp. nov. Female (n = 25). The morphometric measurements are shown in Table 3. Holotype (female in glycerin). Body length = 822.99 μm, maximum body width = 531.80 μm, stylet length = 15.25 μm, stylet knob width = 4.29 μm, stylet knob height = 1.85 μm, distance from base of stylet to dorsal esophageal gland opening (DEGO) = 5.32 μm, metacorpus length = 45.50 μm, metacorpus width = 39.49 μm, anterior end to center of metacarpus = 77.64 μm, distance from anterior end to excretory pore = 38.82 μm, distance from anterior end to excretory pore/length of stylet (EP/ST) = 2.5. Morphological characters. The body is pear-shaped and milky white, with a prominent and variably sized neck. The neck has blurry annuluses, the abdomen has slight bulges, the poste- rior part of the body is round, and the anal region has no protuberances (Figs 2K, 2L and 3E). The stylet is developed, with a straight cone and columnar shaft, the stylet knobs are oblate, the metacorpus is round or ovoid and the valve is developed and obvious, the opening of the dorsal esophageal gland orifice is hook-like, an excretory pore is located in the posterior por- tion of the stylet knobs (Figs 2J and 3F), and the EP/ST ratio is approximately 1.98–2.96 μm. Under SEM, the labial disc is ovoid-squared, slightly raised on the medial lips, and fused with Fig 1. Root symptom of diseased Vitis vinifera. (A) and (B) all show the root symptom of diseased Vitis vinifera, the arrow of fig show eggs of root-knot nematodes. https://doi.org/10.1371/journal.pone.0245201.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 6 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Table 3. Morphometrics of Meloidogyne vitis sp. nov. Holotype Females Paratype Females Character n Body length Body width Stylet length Stylet knobs width Stylet knobs height DEGO Metacorpus length Metacorpus width Head region height Head region width Distance from anterior end to center of metacarpus Distance from anterior end to excretory pore Tail length Hyaline tail length Anal body diameter Spicules length Gubermaculum length 822.99 531.80 15.25 4.29 1.85 5.32 45.50 39.49 / / 77.64 38.82 / / / / / a (Body length/ Body width) 1.55 c (Body length/ Tail length) d (Tail length / Anal body diameter) Tail length/Hyaline tail length / / / All measurements are in μm and shown in the form: mean ± s.d. (range). https://doi.org/10.1371/journal.pone.0245201.t003 25 958.99 ± 132.32 (822.99–1245.16) 609.00 ± 43.63 (531.80–688.11) 15.73 ± 3.68 (8.11–26.58) 4.44 ± 0.96 (2.74–5.95) 2.08 ± 0.48 (1.32–3.32) 4.13 ± 0.84 (2.59–5.32) 42.72 ± 7.05 (23.01–51.53) 37.03 ± 5.81 (21.11–42.86) / / 72.75 ± 12.70 (44.17–86.28) 38.82 ± 4.15 (34.33–44.80) / / / / / 1.58 ± 0.2 (1.30–1.95) / / / Males 10 1330.42 ± 179.15 (1032.23–1593.38) 36.75 ± 6.15 (25.69–43.94) 19.31 ± 1.71 (17.02–21.39) 3.50 ± 0.62 (2.65–4.67) 2.54 ± 0.29 (2.23–3.19) 3.30 ± 0.52 (2.35–3.91) 17.95 ± 1.63 (15.61–20.43) 9.23 ± 0.73 (7.92–10.38) 5.20 ± 0.39 (4.70–5.76) 10.41 ± 1.27 (8.33–12.32) 99.31 ± 5.88 (90.96–108.73) 133.33 ± 4.94 (126.49–140.81) 12.86 ± 0.77 (11.81–14.23) / 24.05 ± 1.81 (21.68–26.68) 30.88 ± 2.59 (27.86–35.75) 10.23 ± 1.86 (8.15–14.88) 36.79 ± 5.96 (30.67–50.15) 103.59 ± 13.78 (81.72–127.65) 0.54 ± 0.03 (0.48–0.57) / J2s 26 396.85 ± 18.34 (353.36–425.76) 16.19 ± 1.93 (12.81–22.43) 13.33 ± 0.32 (12.74–14.11) 1.56 ± 0.31 (1.21–2.22) 1.24 ± 0.18 (0.98–1.69) 1.35 ± 0.31 (1.02–2.01) 10.37 ± 1.21 (8.14–12.18) 6.94 ± 0.64 (5.67–8.15) / / 54.89 ± 1.99 (50.8–58.62) 41.55 ± 2.13 (37.46–45.31) 57.43 ± 3.92 (47.01–63.77) 12.16 ± 1.74 (9.72–15.73) 12.76 ± 1.91 (10.15–17.11) / / 24.76 ± 2.51 (18.98–28.44) 6.95 ± 0.65 (6.15–8.77) 4.58 ± 0.60 (3.39–5.34) 4.81 ± 0.76 (3.44–6.08) PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 7 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Fig 2. Line drawings of Meloidogyne vitis sp. nov. Male (A-F) A: Entire body of male; B, C: Anterior region of male; D, E: Tail region of male; F: Lateral field of male. Second-stage juveniles (G-I) G: Entire body of second-stage juveniles; H: Anterior region of second- stage juveniles; I: Tail region of second-stage juveniles. Female (J-O) J: Anterior region of female; K, L: Entire body of female; M: Eggs of female; N, O: Perineal pattern of female. (Scale bars: A, K, L = 200 μm; B-E, I, H, G, M-O = 20 μm; G = 100 μm; F = 10 μm). https://doi.org/10.1371/journal.pone.0245201.g002 the medial lips to form a dumbbell-shaped structure; there are no obvious lateral lips, and the oral aperture is slit-like and located in the middle of the labial disc, surrounded by six inner labial sensilla (Fig 4B). An excretory pore is located on ventrally region between 23rd and 25th annule behind lips (Fig 4A and 4E). The stylet is cone-shaped and sharply pointed (Fig 4D). The perineal pattern of female adults is round to ovoid with a moderately high dorsal arch and smooth and fine striae that are extremely dense and faint; lateral fields are not clearly visi- ble, and there are no lateral lines, however, a few specimens have slight striae on two shoulders or wings in the lateral field; two phasmids are large, prominent and round, with a diameter that can account for 2–5 annular striae, and seemingly eye-shaped; straight lines of two phas- mids are parallel or nearly parallel to the vulva; the vulval slit is wide and seemingly mouth- PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 8 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Fig 3. Light micrographs of Meloidogyne vitis sp. nov. Male (A-D) A: Entire body of male; B: Anterior region of male; C, D: Tail region of male. Female (E-J) E: Entire body of female; F: Anterior region of female; G: Partial region of female; H, I: Perineal pattern of female; J: Eggs of female. Second-stage juveniles (K-O) K, O: Tail region of second-stage juveniles; L: Entire body of second-stage juveniles; M, N: Anterior region of second-stage juveniles (Scale bars: A, E = 200 μm; B-D, F, H, I, J, K, M-O = 20 μm; G, L = 100 μm). https://doi.org/10.1371/journal.pone.0245201.g003 shaped; the anal fold is clearly visible and seemingly nose-shaped; the whole perineal pattern is seemingly monkey-face-shaped; the area of the vulva and anus is smooth, with no striae (Figs 2N, 2O, 3H and 3I). The distance between two phasmids is wider than or equal to the length of the vulval slit; however, in very few specimens, this value is slightly smaller. The vulva-anus distance is short: 19.94±1.63 (17.35–23.48) μm. The vulva-phasmid distance is 27.98 ± 2.33 (23.28–33.04) μm. The anus-phasmid distance is 6.84 ± 1.49 (4.73–9.79) μm. The morphology of the perineal pattern under SEM is consistent with that under LM, but SEM shows more morphological details of the anus and vulva, and the striae are clearer (Fig 4C). Male (n = 10). The morphometric measurements are shown in Table 3. Morphological characters. The body is vermiform and variable in length, and the anterior end of the body tapers off (Figs 2A and 3A). The head cap is obvious and slightly separated from the body; the stylet is developed and has an obvious boundary with the stylet shaft; the stylet knot is oblate-spheroidal; the metacorpus is vertically ovoid; and the valve is obvious (Figs 2C and 3B). The tail is mostly straight and short with a humped end; the spicules are PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 9 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Fig 4. Scanning electron microscope photographs of Meloidogyne vitis sp. nov. Female (A-E) A: anterior end in lateral view; B: anterior end in face view; C: perineal pattern; D: anterior end in lateral view and see the stylet; E: excretory pore. Second-stage juvenile (F-I) F: anterior end in lateral view; G: anterior end in face view; H: lateral field; I: tail region. Male (J-O) J: lateral field; K: tail region; L: anterior end in lateral view; M: anterior end in face view; N: excretory pore; O: tail region. (Scale bars: A, C, O = 20 μm; B, D, M = 3 μm; F, G = 2 μm; E, L, H, N = 5 μm; J, I, K = 10 μm). https://doi.org/10.1371/journal.pone.0245201.g004 developed, arch-shaped, and slightly curved; and the gubernaculum is obvious and curved- moon shaped (Figs 2D, 2E, 3C and 3D). Under SEM, the head region lacks annulus; the labial disc is horizontally ovoid-squared, slightly raised on the medial lips, slightly wider than the medial lips and fused with the medial lips to form a dumbbell-shaped structure; there are no lateral lips; a prominent slit-like opening between the labial disc and medial lips is observed; a distinct depression appears in the middle of the medial lips; and the oral aperture is slit-like and located in the middle of the labial disc, surrounded by six inner labial sensilla (Fig 4L and PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 10 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan 4M). The lateral field consists of four incisures forming 3 lateral bands, which are full of reticu- lar striae (Fig 4J). Lateral incisures extend to the tail end of the body, and in addition to the tail end, the annulus passes through incisures. The spicules resemble a figure eight, and the tip end slightly curves to form a hook-like shape (Fig 4K and 4O). The excretory pore is an irregular pore located in the depression of the cuticle, where it spans two annuluses, and the regular body annuluses are interrupted around the excretory pore (Fig 4N). J2 (n = 26). The morphometric measurements are shown in Table 3. Morphological characters. The body is vermiform and slender and tapers at both ends, but more towards the tail than the anterior end (Figs 2G and 3L). The body annulations are not obvious. The stylet is straight, slender, and sharply pointed and has an obvious boundary with the stylet shaft; the stylet knobs are obvious and spherical; and the metacorpus is ovoid and clearly visible (Figs 2H, 3M and 3N). The tail is variable, exhibits a range of variation in tail fields and is conical and constricted; the hyaline tail is short (Figs 2I, 3K and 3O). The anus is difficult to distinguish except under a high-power oil immersion objective lens. The intestinal contents were too much, so the rectum and caudal sensory organ were difficult to observe. Under SEM, the head region is not smooth and slightly folded; the labial disc appears round, slightly raised on the medial lips and fused with the medial lips to form a dumbbell-shaped structure, with a slightly sunken appearance in the middle of the medial lips; a prominent slit- like amphidial opening is located between the labial disc and lateral lips; the oral aperture is round and located in the middle of the labial disc, surrounded by six inner labial sensilla (Fig 4F and 4G). The lateral field forms 3 lateral bands delimited by four incisures (Fig 4H). The anal opening is elliptical and located in the cuticular depression in the tail of the body (Fig 4I). The excretory pore is irregular and located in the cuticular depression, and the regular body annuluses are interrupted around the excretory pore. Egg (n = 20). Eggs of female adults are oval-shaped (Figs 2M and 3J). Twenty eggs were measured in clear water. The egg length was 86.52–110.69 μm (mean: 98.67 μm, standard error: 6.44), and the egg width was 30.56–39.31 μm (mean: 34.73 μm, standard error: 2.38). Taxonomic summary Type host. Grape (Vitis vinifera L., Vitis L., Vitaceae). Type locality. Luliang County, Yunnan Province, China (25˚07’ N, 103˚78’ E). Etymology. The specific epithet refers to the host plant on which this new species was found. Additionally, because of its ability to seriously infect cultivated grapes (V. vinifera L.), we suggest the name Meloidogyne vitis sp. nov. Type material. The holotype female and paratypes, males, perineal patterns and J2s were deposited in the nematode collection of the authors’ Laboratory of Plant Nematology, College of Plant Protection, Yunnan Agricultural University, China. Specific ITS1-5.8S-ITS2, D2D3 rDNA, coxI rRNA and coxII 16S rRNA mtDNA sequences were deposited in GenBank with accession numbers MN816222.1, MN816223.1, MN816225.1, MN816226.1, MN814829.1, MN814830.1, MT012386.1, and MT012387.1, respectively. Diagnosis and relationships Meloidogyne vitis sp. nov. can be distinguished from other Meloidogyne spp. by a unique com- bination of several morphological characters. The perineal pattern of female adults is round or ovoid, with large and prominent phasmids. Females have a prominent neck, an excretory pore is located on ventrally region between 23rd and 25th annule behind lips, and the EP/ST ratio is approximately 2.5 (1.98–2.96 μm). The male has a prominent head region, the labial disc is fused with the medial lips to form a dumbbell-shaped structure, a wide slit is located between PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 11 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan the labial disc and medial lips, the tail is blunt and round, the gubernaculum is prominent and arch-shaped, and the lateral field consists of four incisures. J2s are characterized by their head region is not smooth and slightly folded, the labial disc is fused with the medial lips to form a dumbbell-shaped structure, the hyaline tail is short and constricted, the relatively small c value (6.15–8.77 μm), and the lateral fields have four incisures. In addition, M. vitis sp. nov. has unique ITS1-5.8S-ITS2 and D2D3 of rDNA, coxI and coxII 16S rRNA of mtDNA sequences. Because of the large and prominent phasmids in the perineal pattern in female adults, M. vitis sp. nov. is similar to M. mali [36], Meloidogyne artiellia [37], Meloidogyne floridensis [38], Meloidogyne naasi [39], M. nataliei [18], Meloidogyne shunchangensis [40], Meloidogyne kongi [41], Meloidogyne dimocarpus [42], and Meloidogyne thailandica [43]. Meloidogyne vitis sp. nov. differs from M. mali in that the perineal pattern of the female shows a moderately high dorsal arch rather than a low and flat dorsal arch and there are no lateral lines rather than clear single or double lateral lines in the lateral fields, the DEGO of male and J2 are smaller (2.35– 3.91 vs. 6.00–13.00 μm and 1.02–2.01 vs. 4–6 μm, respectively), the J2 tail is longer (47.01– 63.77 vs. 30–34 μm), and the J2 c value is smaller (6.15–8.77 vs. 12–15 μm). Meloidogyne vitis sp. nov. differs from M. artiellia in the lip region of the female (the lateral lip is not obvious rather than appearing as six almost equally sized lips), the greater body length (822.99–1245.16 vs. 650–760 μm) and body width (531.80–688.11 vs. 340–460 μm) in females, the perineal pat- tern of the female being round or ovoid rather than the general outline pattern resembling a figure eight, the different male lip region (the labial disc and medial lips are fused into a dumb- bell-shaped structure instead of the lip region appearing as six nearly equally sized lips), its smaller male DEGO (2.35–3.91 vs. 5.00–7.00 μm), its longer J2 tail (47.01–63.77 vs. 24.5 μm), its smaller J2 stylet length and c value (12.74–14.11 vs. 14–16 μm and 6.15–8.77 vs. 13–16 μm, respectively). Meloidogyne vitis sp. nov. differs from M. floridensis in that its perineal pattern of the female has no lateral lines rather than faint lateral lines, it has narrower stylet knobs in males (2.65–4.67 vs. 5.00–6.00 μm), and it has a smaller J2 DEGO (1.02–2.01 vs. 2.50– 3.00 μm). Meloidogyne vitis sp. nov. differs from M. nassi in that the posterior of the female is smooth rather than presenting a slight protuberance, the excretory pore of the female is situ- ated behind instead of slightly in front of the stylet knobs, the longer females body length (822.99–1245.16 vs. 455–705 μm) and body width (531.80–688.11 vs. 227–398 μm), and it lacks the four or five small and vesicle-like structures grouped irregularly round in front of the metacorpus that can be found in M. nassi male and J2. Meloidogyne vitis sp. nov. differs from M. nataliei in that the posterior of the female is smooth rather than having a slight protuber- ance; the perineal pattern of the female has no lateral lines instead of two separated ropelike striae extending laterally from the vulval and anal areas and forming a lateral field; the stylet length, DEGO, and spicule length of male are smaller (17.02–21.39 vs. 28.40–29.20 μm, 2.35– 3.91 vs. 4.0–6.5 μm, and 27.86–35.75 vs. 41.3–44.3 μm respectively); and the body length, stylet length, and DEGO of J2 are smaller (353.36–425.76 vs. 539.00–641.00 μm, 12.74–14.11 vs. 21.9–22.8 μm, and 1.02–2.01 vs. 3.0–4.3 μm, respectively). Meloidogyne vitis sp. nov. differs from M. shunchangensis in the female lip region (the lateral lip is not obvious rather than the lip region appearing as six lips), the perineal pattern of females having no striae rather than occasionally having lines of striae between the anus and vulva and having no incisures rather than sometimes having obvious double striae in the lateral field, the greater J2 tail length and transparent tail length (47.01–63.77 vs. 20.3–28.6 μm and 9.72–15.73 vs. 3.4–4.2 μm, respec- tively), the smaller J2 DEGO and c value (1.02–2.01 vs. 2.6–3.7 μm and 6.15–8.77 vs. 11.9– 16.60 μm, respectively). Meloidogyne vitis sp. nov. differs from M. kongi in that the female’s posterior is smooth rather than possessing a prominent protuberance, the perineal pattern is no line striae vs full of line striae between the anus and vulva, the female body is longer (822.99–1245.16 vs. 610.6–820.3 μm), the DEGO of J2 is smaller (1.02–2.01 vs. 3.9–5.8 μm), PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 12 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan there are no lateral lips rather than two semicircular lateral lips in the head region in males, the stylet length and DEGO of male are smaller (17.02–21.39 vs. 22.00–23.90 μm and 2.35–3.91 vs. 5.80–7.50 μm, respectively), and the DEGO of J2 is smaller (1.02–2.01 vs. 3.9–5.8 μm). Meloi- dogyne vitis sp. nov. differs from M. dimocarpus in that the perineal pattern of the female has no lateral lines instead of double or single striae in the lateral field and no striae instead of mul- tiple continuous striae between the anus and vulva, the phasmids are large instead of small, and the male and J2 DEGO are smaller (2.35–3.91 vs. 4.50–5.75 μm and 1.02–2.01 vs. 2.25– 3.75 μm, respectively). Meloidogyne vitis sp. nov. differs from M. thailandica in that the peri- neal pattern of the female lacks the radial structures underneath the pattern area that are char- acteristic of M. thailandica, the J2 DEGO and hyaline tail length are smaller (1.02–2.01 vs. 2. 5–3.5 μm and 9.72–15.73 vs. 15–20 μm, respectively). Meloidogyne vitis sp. nov. can also be distinguished from several other Meloidogyne species infecting grape, including M. incognita, M. javanica, M. arenaria, M. hapla, M. ethiopica and M. thamesi, by the large and prominent phasmids in the perineal pattern of females. Isozyme analysis The isozyme electrophoretic analysis of young egg-laying females of M. vitis sp. nov. showed three rare Mdh bands (Fig 5A, MV lane) and one rare Est band migrating rapidly in the gel (Fig 5B, MV lane), which did not occur in the Mdh and Est phenotypes of M. javanica. The Mdh and Est bands of M. vitis sp. nov. have not been reported in other RKNs. According to the relative mobility (Rm) values and referring to the naming method of Esbenshade and Tri- antaphyllou (1985) [29], the Mdh band was named N3d, and the Est band was named VF1. PCR product electrophoresis Size of the PCR amplification bands in different fragments of M. vitis sp. nov. and M. mali as fol- lows: for both of the ITS1-5.8S-ITS2 fragments was approximately 870 bp, both of the 28S D2D3 fragments was approximately 770 bp, both of the coxI mtDNA fragments was approximately 400 bp; the coxII mtDNA fragments of M. vitis sp. nov. was approximately 550 bp (Fig 6). Molecular characterization Amplification and sequencing of the ITS1-5.8S-ITS2 fragment of rDNA from the females and J2s of M. vitis sp. nov. and J2s of M. mali revealed that the sequence sizes were 877 bp, 877 bp, and 873 bp, respectively. The GenBank accession numbers are MN816222.1 and MN816223.1 for the female and J2 of M. vitis sp. nov., respectively, and MN816224.1 for the J2 of M. mali. The ITS1-5.8S-ITS2 fragment identities were 100% for the female and J2 of M. vitis sp. nov. A BLAST search of M. vitis sp. nov. revealed that the most similar sequence was that of M. mali (GenBank accession numbers KR535971.1, JX978229.1, and JX978228.1), with an identity of only 87%. The sequence of M. mali identified in this research was most similar to sequences of M. mali in GenBank, with identities ranging from 97% (accession number KR535971.1) to 99% (accession numbers JX978225.1, JX978229.1, and JX978228.1). Phylogenetic trees (52 sequences in total) showed that the female and J2 of M. vitis sp. nov. formed a well-supported clade with high bootstrap support (100%) and were closest to M. mali from GenBank (acces- sion numbers KR535971.1, JX978229.1, and JX978228.1) and M. mali identified in this research (accession number MN816224.1), all of which formed one group with high bootstrap support (95%). Meloidogyne vitis sp. nov. was clearly separated from other species (Fig 7). Sen- quence alignment of ITS1-5.8S-ITS2 rDNA between female of M. vitis sp. nov. and J2 of M. mali identified in this research showed that the identity was only 77.09%, and the sequences were thus highly diverged (210-base divergence) (S1 Fig). PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 13 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Fig 5. Malate dehydrogenase and esterase phenotype patterns obtained with electrophoresis of protein homogenates from five young egg-laying females of Meloidogyne vitis sp. nov. (lane Mv) and five young egg-laying females of the Meloidogyne javanica reference population (lane Mj). (A) Malate dehydrogenase patterns. (B) esterase patterns. (C) Relative mobility (Rm) of malate dehydrogenase bands. (D) Relative mobility (Rm) of esterase bands. https://doi.org/10.1371/journal.pone.0245201.g005 Amplification and sequencing of the D2D3 fragment of 28S rDNA from females and J2s of M. vitis sp. nov. revealed sequence sizes both were 775 bp. The GenBank accession numbers are MN816225.1 and MN816226.1 for the female and J2 of M. vitis sp. nov., respectively. The D2D3 fragment identities were 99.61% for the female and J2 of this new species. A BLAST search of M. vitis sp. nov. revealed that the most similar sequence was that of M. mali (Gen- Bank accession numbers KX430177.1, JX978226.1, JX978227.1, and KF880398.1), with a 93% identity. Phylogenetic trees (45 sequences in total) showed that the female and J2 of M. vitis sp. nov. formed a well-supported clade with high bootstrap support (99%). Meloidogyne vitis sp. nov. is most closely related to M. mali from GenBank (accession numbers KF880398.1, JX978227.1, KF880399.1, KX430177.1, and JX978226.1) and formed a sister group with this species with high bootstrap support (98%). Meloidogyne vitis sp. nov. was clearly separated from other species (Fig 8). Squence alignment of D2D3 28S rDNA between female of M. vitis PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 14 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Fig 6. PCR electropherogram for different fragments of Meloidogyne vitis sp. nov. and Meloidogyne mali. M: 2000 DNA marker; CK: The negative control consisting of water; Lanes 1–2: The ITS1-5.8S-ITS2 region of Meloidogyne vitis sp. nov. and Meloidogyne mali, respectively; Lanes 3–4: The D2/D3 region of Meloidogyne vitis sp. nov. and Meloidogyne mali, respectively; Lanes 5–6: The coxI region of Meloidogyne vitis sp. nov. and Meloidogyne mali, respectively; Lane 7: The coxII region of Meloidogyne vitis sp. nov. https://doi.org/10.1371/journal.pone.0245201.g006 sp. nov. and M. mali from GenBank (accession number KX430177.1) showed that the identity was 93.81% and the sequences were highly divergent (48-base divergence) (S2 Fig). The sequence sizes of the coxI fragments of 16S rRNA from the female and J2s of M. vitis sp. nov. and J2s of M. mali were 413 bp, 413 bp, and 417 bp, respectively. The GenBank accession numbers are MN814829.1 and MN814830.1 for the female and J2 of M. vitis sp. nov., respectively, and MN814831.1 for the J2 of M. mali. The sequence identities were 99.76% for the female and J2 of M. vitis sp. nov. A BLAST search of M. vitis sp. nov. revealed the highest match with the sequences of Meloidogyne ichinohei (GenBank accession number KY433448.1) and M. exigua (GenBank accession numbers MH128478.1, MH128477.1, and MH128476.1), with identities of 85%. The sequences of M. mali identified in this research were most similar to those of M. mali from GenBank (accession numbers KM887146.1, KM887145.1, KY433450.1 and KY433449.1), with 99% identities. Phylogenetic trees (34 sequences in total) showed that the female and J2 of M. vitis sp. nov. formed a well-supported clade with high bootstrap support (99%). Meloidogyne vitis sp. nov. was most closely related to M. mali from GenBank (accession numbers KM887146.1, KY433450.1, KM887145.1, and KY433449.1) and M. mali identified in this research (accession number MN814831.1), and they formed a sister group. Meloidogyne vitis sp. nov. was clearly sepa- rated from other species (Fig 9). Sequence alignment of coxI 16S rRNA between the female of M. vitis sp. nov. and J2 of M. mali identified in this research showed that the identity was 84.26% and the sequences were highly divergent (67-base divergence) (S3 Fig). The coxII 16S rRNA sequences from the female and J2s of M. vitis sp. nov. were 545 bp and 540 bp in size, respectively. The GenBank accession numbers are MT012386.1 for the female and MT012387.1 for the J2 of M. vitis sp. nov. The identities were 99.63% for the female and J2 of M. vitis sp. nov. A BLAST search revealed that M. vitis sp. nov. is most closely related to M. mali (GenBank accession number KC112913.1), with an identity of 81%. Phylogenetic trees (53 sequences in total) showed that the female and J2 of M. vitis sp. nov. formed a well-sup- ported clade with high bootstrap support (100%). Meloidogyne vitis sp. nov. was most closely related to M. mali from GenBank (accession number KC112913.1) and formed one PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 15 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Fig 7. Phylogenetic relationships of Meloidogyne vitis sp. nov. with other root-knot nematodes based on ITS1- 5.8S-ITS2 sequences. Numbers to the left of the branches are bootstrap values for 1000 replications. https://doi.org/10.1371/journal.pone.0245201.g007 monophyletic clade with moderate bootstrap support (74%) (Fig 10). Sequence alignment of coxII 16S rRNA between female of M. vitis sp. nov. and M. mali from GenBank (accession number KC112913.1) showed that the identity was only 78.58% and the sequences were highly divergent (118-base divergence) (S4 Fig). PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 16 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Fig 8. Phylogenetic relationships of Meloidogyne vitis sp. nov. with other root-knot nematodes based on D2/D3 sequences of 28S rDNA. Numbers to the left of the branches are bootstrap values for 1000 replications. https://doi.org/10.1371/journal.pone.0245201.g008 SCAR-PCR analysis Individual females of previously identified and purified populations of M. incognita, M. java- nica, M. arenaria, M. hapla, and Meloidogyne enterolobii were used for comparison. DNA was extracted from individual females of M. incognita, M. javanica, M. arenaria, M. hapla, M. enterolobii and M. vitis sp. nov. The ITS1-5.8S-ITS2 fragment of the six RKN species was amplified using the primers 18S/26S. In all populations, a single band with a size of approxi- mately 760–900 bp was amplified (Fig 11A). However, using the primers Mv-F/Mv-R to PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 17 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Fig 9. Phylogenetic relationships of Meloidogyne vitis sp. nov. with other root-knot nematodes based on coxI-rRNA genes sequences. Numbers to the left of the branches are bootstrap values for 1000 replications. https://doi.org/10.1371/journal.pone.0245201.g009 amplify the same six templates mentioned above, species-specific fragments of approximately 170 bp were amplified only in M. vitis sp. nov., no fragments were observed for templates from the other five RKN species (Fig 11B). The species-specific product of M. vitis sp. nov. was recy- cled, cloned and sequenced, resulting in a 174 bp sequence, which was deposited in the Gen- Bank database for BLAST alignment, no similar sequences were found. All results indicated that the Mv-F/Mv-R primers were specific and reliable. Discussion Reliable detection and identification technology is necessary for the protection of agricultural pro- duction systems against quarantine nematodes worldwide [44]. In the past, RKNs were identified mainly by morphological observations. Although observations of the perineal pattern of female adults is the primary method used for morphological identification, there is some intraspecific variability in this pattern due to differences in host and nutrition, differences between young females and female adults and other factors. Therefore, identification results are often uncertain. PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 18 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Fig 10. Phylogenetic relationships of Meloidogyne vitis sp. nov. with other root-knot nematodes based on coxII-16S rRNA genes sequences. Numbers to the left of the branches are bootstrap values for 1000 replications. https://doi.org/10.1371/journal.pone.0245201.g010 The perineal pattern of Meloidogyne inornata is similar to that of M. incognita, making it difficult to distinguish these two species [45], and the perineal pattern of pre-adult M. javanica resembles that of Meloidogyne africana adults [46]. In addition, the perineal pattern of RKNs varies greatly PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 19 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan Fig 11. PCR amplification of the supplied RKNs with the Mv-F/Mv-R primers. M: 2000 DNA marker; CK: The negative control consisting of water. A: Lanes 1–6, The ITS1-5.8S-ITS2 region of Meloidogyne vitis sp. nov., Meloidogyne incognita, Meloidogyne javanica, Meloidogyne arenaria, Meloidogyne hapla and Meloidogyne enterolobii, respectively. B: Lanes 7–12, The amplification results of root-knot nematode species-specific PCRs of Meloidogyne vitis sp. nov., Meloidogyne incognita, Meloidogyne javanica, Meloidogyne arenaria, Meloidogyne hapla and Meloidogyne enterolobii, respectively. https://doi.org/10.1371/journal.pone.0245201.g011 across generations; for example, M. javanica has a total variation rate of 22.6%; the variation is mainly caused by nonobvious incisures and the formation of shoulder protuberances, which make it difficult to distinguish this RKN from M. arenaria [47]. Traditional morphological meth- ods face considerable challenges in the identification of RKNs due to intraspecific variation and interspecies similarity [48]. Therefore, PCR-based methods and biochemical methods are becom- ing increasingly important in the diagnosis of Meloidogyne spp. The technology of isozyme (in particular Est and Mdh) electrophoretic is a relatively old method for the identification of Meloidogyne spp., but it is still used by many researchers to identify some RKNs. This technique remains as an effective methodology with which to unam- biguously identify and differentiate Meloidogyne megadora [49]. In this research, the band phe- notypes of malate dehydrogenase from the new species were different from those of other RKNs reported to date; the new species produced three bands and showed the N3d phenotype. The esterase of the new species migrated rapidly in the gel, showing a VF1 phenotype, which is the same esterase phenotype observed for M. naasi, Meloidogyne exigua and M. thailandica; however, all of these species have different malate dehydrogenase phenotypes [39, 50]. Never- theless, the isozyme analyses applied in species identification are limited because they are effec- tive only when egg-laying females are available [51]. PCR-based methodologies are of ever-increasing importance in species diagnostics and phylogenetics within the genus Meloidogyne [52]. Phylogenetic analyses of rDNA sequences are considered a reliable diagnostic approach and are commonly used to identify and compare certain RKNs [53]. Based on rDNA-PCR identification, the ITS fragment is possibly the most widely used genetic marker for living organisms and the most commonly used species-level marker used for organisms (plants, protists, and fungi) [52]. This fragment has been widely used to identify RKNs. However, Powers et al. [54] and Blok et al. [55] found that the ITS1- 5.8S-ITS2 sequences of M. incognita, M. javanica, and M. arenaria were extremely conserved. Landa et al. (2008) [56] also found that the 18S sequences of rDNA from M. hispanica and M. ethiopica were very similar. Although M. hispanica and M. ethiopica can be clearly differenti- ated by their D2D3 28S ribosomal DNA sequences [56], M. incognita, M. javanica, and M. are- naria cannot [5]. Therefore, identification of some RKN species with rDNA-PCR technology remains difficult. The mitochondrial genome (mtDNA) provides a rich source of genetic markers for species identification [57], including parasitic nematode identification, because of their high mutation rates and maternal inheritance [58], mtDNA-PCR technology is an alter- native method for precise identification of Meloidogyne species, to study intraspecific PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 20 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan variability and to follow maternal lineages [59]. The mitochondrial genes coxI and coxII have been widely used as DNA markers in various large organismal groups in the animal kingdom [60]. In terms of resolution, coxI is more capable of discriminating between species than either of the rRNA genes [61]. Sequence characterized amplified region (SCAR) markers have proven to be a very sensitive and reliable tool for the identification of RKNs and to provide an easy and rapid assessment of a large number of samples by a simple visual evaluation of gels [62]. The SCAR-PCR technique is more sensitive than other existing molecular techniques [63], providing a rapid species identification approach for turfgrass RKNs independent of morphol- ogy [64]. An increasing number of species-specific primers are being designed for the identifi- cation of RKNs [65–67]. In the present research, the species-specific primer pair Mv-F/Mv-R was designed based on the sequence of the rDNA ITS1-5.8S-ITS2 fragment in M. vitis sp. nov. so that it could provide a simple and rapid method for identifying this new species. In this research, PCR amplification of ITS1-5.8S-ITS2 rDNA, D2D3 28S rDNA, and mtDNA (coxI and coxII) was used to identify RKNs. The nematode collected from grape was different from previously described RKNs. The ITS1-5.8S-ITS2 and D2D3 sequences of rDNA and the coxII sequence of mtDNA were compared with the corresponding fragments in RKNs available from GenBank. The most similar species was M. mali, with similarities of 87%, 93%, and 81%, respectively. The coxI mtDNA sequence was most similar to that of M. ichinohei, with a similarity of 85%. The phylogenetic tree based on ITS1-5.8S-ITS2 rDNA, D2D3 28S rDNA, and mtDNA (coxI and coxII) sequences showed that the new species was closely related to M. mali and clearly distinguished from other RKNs. In summary, both the morphological and molecular characteristics revealed that the new RKN from grape is sufficiently different from the RKNs described to date to be considered a new RKN species. Thus, this new species was named M. vitis sp. nov. according to its host resource. Meloidogyne vitis sp. nov. and M. mali are similar in morphology and have a close molecular relationship; they may have evolved from the same ancestral species. Almost all grape roots were infected by RKNs in the vineyard investigated in this study, and the aged roots were decayed and necrotic due to RKN infection, which indicated that the RKNs in this vineyard had existed there for many years. However, their origin is still unknown, and they may have been introduced from outside. The phylogenetic tree based on various fragments shows that the new species has independent evolutionary trends; it is either indigenous in some regions of the world as an ancient species or has recently evolved and been widely spread by agriculture. High-density RKNs undoubtedly pose a serious threat to grape production. This species may be restricted to Luliang County, Yunnan Province, and will seriously damage grapes there by causing severe root knots, dwarfed plants and reduced fruit production. More surveys are needed to clarify the distribution of M. vitis sp. nov., and further research will also be necessary to determine its host range, pathogenesis, and control strategies. Supporting information S1 Fig. Sequence alignment of Meloidogyne vitis sp. nov. and Meloidogyne mali identified in this research based on ITS1-5.8S-ITS2 sequences. (1 = Meloidogyne vitis sp. nov., 2 = Meloidogyne mali). (TIF) S2 Fig. Sequence alignment of Meloidogyne vitis sp. nov. identified in this research and Meloidogyne mali from GenBank (KX430177.1) based on D2D3 sequences of 28S rDNA. (1 = Meloidogyne vitis sp. nov., 2 = Meloidogyne mali). (TIF) PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 21 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan S3 Fig. Sequence alignment of Meloidogyne vitis sp. nov. and Meloidogyne mali identified in this research based on coxI-rRNA genes sequences. (1 = Meloidogyne vitis sp. nov., 2 = Meloidogyne mali). (TIF) S4 Fig. Sequence alignment of Meloidogyne vitis sp. nov. identified in this research and Meloidogyne mali from GenBank (KC112913.1) based on coxII-16S rRNA genes sequences. (1 = Meloidogyne vitis sp. nov., 2 = Meloidogyne mali). (TIF) S1 Raw images. (PDF) Author Contributions Conceptualization: Yanmei Yang, Xianqi Hu, Li Chen. Data curation: Yanmei Yang, Xianqi Hu, Pei Liu. Formal analysis: Yanmei Yang, Xianqi Hu, Pei Liu. Funding acquisition: Xianqi Hu. Investigation: Yanmei Yang, Qi Zhang. Methodology: Yanmei Yang, Xianqi Hu, Pei Liu, Huan Peng, Qiaomei Wang. Resources: Huan Peng. Software: Yanmei Yang, Pei Liu, Qiaomei Wang. Supervision: Xianqi Hu. Writing – original draft: Yanmei Yang. Writing – review & editing: Xianqi Hu, Pei Liu. References 1. Ali MA, Azeem F, Li H, Bohlmann H. Smart parasitic nematodes use multifaceted strategies to parasit- ize plants. Front Plant Sci. 2017; 8: 1–21. https://doi.org/10.3389/fpls.2017.00001 PMID: 28220127; PubMed Central PMCID: PMC5632807. 2. Niu JH. [The preliminary study for MiMsp40, a secretary protein gene of southern root-knot nematode (Meloidogyne incognita)]. PhD Thesis. Beijing: China Agricultural University. 2011. Chinese. 3. Wu BS. [Western Henan tobacco root-knot nematode species identification and its losses estimated]. Master’s Thesis. Changsha: Hunan Agricultural University. 2013. Chinese. 4. Elling AA. Major Emerging Problems with Minor Meloidogyne Species. Phytopathology. 2013; 103(11): 1092–1102. https://doi.org/10.1094/PHYTO-01-13-0019-RVW PMID: 23777404. 5. Trinh QP, Le TML, Nguyen TD, Nguyen HT, Liebanas G, Nguyen TAD. Meloidogyne daklakensis n. sp. (Nematoda: Meloidogynidae), a new root-knot nematode associated with Robusta coffee (Coffea cane- phora Pierre ex A. Froehner) in the Western Highlands, Vietnam. J Helminthol. 2019; 93(2): 242–254. https://doi.org/10.1017/S0022149X18000202 PMID: 29619918 6. Ali N, Tavoillot J, Mateille T, Chapuis E, Besnard G, Bakkali AEI, et al. A new root-knot nematode Meloi- dogyne spartelensis n. sp. (Nematoda: Meloidogynidae) in northern Morocco. Eur J Plant Pathol. 2015; 143(1): 25–42. https://doi.org/10.1007/s10658-015-0662-3 7. Tao Y, Xu CL, Yuan CF, Wang HH, Lin BR, K Zhuo, et al. Meloidogyne aberrans sp. nov. (Nematoda: Meloidogynidae), a new root-knot nematode parasitizing kiwifruit in China. PLoS One. 2017; 12(8): 1– 22. https://doi.org/10.1371/journal.pone.0182627 PMID: 28854186; PubMed Central PMCID: PMC5576651. PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 22 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan 8. Rahmani M, Bakhshi D, Qolov M. Impact of pruning severity and training systems on red and white seedless table grape (Vitis vinifera) qualitative indices. Aust J Crop Sci. 2015; 9(1): 55–61. 9. Krithika V, Naik R, Pragalyaashree. Functional properties of grape (Vitis vinifera) seed extract and pos- sible extraction techniques-A review. Agricultural Reviews. 2015; 36(4): 313–320. https://doi.10.18805/ ag.v36i4.6668 10. Gao Y, Yang L F, Gao X T. [The present situation, trend and development proposal of the Chinese grape industry]. HeBei Fruits. 2007; (2): 1–2. Chinese. https://doi.CNKI:SUN:HBGS.0.2007-02-001 11. Zhang W, Zhang YH, Lu XY, Bai MD, Kong WX, Ma CH, et al. [Study on the present situation and coun- termeasures of Yunnan grape industry development]. Chinese tropical agriculture. 2015; (4): 25–28. Chinese. https://doi.CNKI:SUN:ZGRD.0.2015-04-008 12. Smith BP, Morales NB, Thomas MR, Smith HM, Clingeleffer PR. Grapevine rootstocks resistant to the root-knot nematode Meloidogyne javanica. Aust J Grape Wine Res. 2017; 23(1): 125–131. https://doi. 10.1111/ajgw.12242 13. Smith HM, Smith BP, Morales NB, Moskwa S, Clingeleffer PR, Thomas MR. SNP markers tightly linked to root knot nematode resistance in grapevine (Vitis cinerea) identified by a genotyping-by-sequencing approach followed by sequenom MassARRAY validation. PLoS One. 2018; 13(2): 1–27. https://doi. org/10.1371/journal.pone.0193121 PMID: 29462210; PubMed Central PMCID: PMC5819801. 14. Nicol JM, Stirling GR, Rose BJ, May P, Heeswijck RV. Impact of nematodes on grapevine growth and productivity: current knowledge and future directions, with special reference to Australian viticulture. Aust J Grape Wine Res. 1999; 5(3): 109–127. https://doi.10.1111/j.1755-0238.1999.tb00295.x 15. Yang Y, Jittayasothorn Y, Chronis D, Wang X, Cousins P, Zhong GY. Molecular characteristic -s and efficacy of 16D10 siRNAs in inhibiting root-knot nematode infection in transgenic Grape hairy roots. PLoS One. 2013; 8(7): 1–13. https://doi.org/10.1371/journal.pone.0069463 PMID: 23874962; PubMed Central PMCID: PMC3712915. 16. Stirling GR, Cirami RM. Resistance and tolerance of grape rootstocks to South Australian populations of root-knot nematode. Austt J Exp Agric Anim Husb. 1984; 24: 277–282. https://doi.10.1071/ ea9840277 17. 18. Zhu H, Sun Q, Du Y, Gao Z, Zhai H. Detection of grape phylloxera on grapevine roots with diagnostic polymerase chain reaction methods targeted to the internal transcribed space region 2 nuclear gene. Aust J Grape Wine Res. 2014; 21(1): 1–4. https://doi.10.1111/ajgw.12111 Liu Y, Zhang H. [Present situation and control strategy of grape root-knot nematode disease in the Huai- hai economic zone]. Anhui Agri Sci Bull. 2017; 23(12): 76–79. Chinese. https://doi.CNKI:SUN: AHNB.0.2017-12-035 19. Howland AD, Skinkis PA, Wilson JH, Riga E, Pinkerton JN, Schreiner RP, et al. Impact of grapevine (Vitis vinifera) varieties on reproduction of the northern root-knot nematode (Meloidogyne hapla). J Nematol. 2015; 47(2): 141–147. PMID: 26170476; PubMed Central PMCID: PMC4492289. 20. Esmenjaud D, Bouquet A. Selection and application of resistant germplasm for grapevine nematodes management. In: Ciancio A, Mukerji KG, editors. Integrated Management of Fruit Crops and Forest Nematodes. Integrated Management of Plant Pests and Diseases, vol 4. Springer, Dordrecht. 2009. p.195– 214. https://doi.org/10.1111/j.1755-0998.2008.02411.x PMID: 21564611 21. Golden AM, Rose LM, and Bird GW. Description of Meloidogyne nataliei n. sp. (Nematoda: Meloidogy- nidae) from Grape (Vitis labrusca) in Michigan, with SEM observations. J Nematol. 1981; 13 (3): 393– 400. PMID: 19300781 22. Carneiro RMDG, Almeida MRA, Cofcewicz ET, Magunacelaya JC, Aballay E. Meloidogyne ethiopica, a major root-knot nematode parasitizing Vitis vinifera and other crops in Chile. Nematology. 2007; 9(5): 635–641. https://doi.10.1163/156854107782024794 23. McLeod RW. and Khair GT. Male intersexes in Meloidogyne thamesi. Nematologica. 1973; 19(4): 561– 562. https://doi.10.1163/187529273X00583 24. Li HY, Yang GD, Shu HR, Yang YT, Ye BX, Nishida I et al. Colonization by the arbuscular mycorrhizal fungus Glomus versiforme induces a defense response against the root-knot nematode Meloidogyne incognita in the grapevine (Vitis amurensis Rupr.), which includes transcriptional activation of the class III chitinase gene VCH3. Plant Cell Physiol. 2006; 47(1): 154–163. https://doi.org/10.1093/pcp/pci231 PMID: 16326755 25. Xie H. [Taxonomy of Plant Nematodes]. 1st ed. Hefei: Anhui Science and Technology Press. 2000. Chinese. 26. Xie H. [Taxonomy of Plant Nematodes]. 2nd ed. Beijing: High Education Press. 2005. Chinese. 27. Eisenback JD, Hirschmann H, and Triantaphyllou AC. Morphological comparison of Meloidogyne female head structures, perineal patterns, and stylets. J Nematol. 1980; 12(4): 300–313. https://doi.10. 2307/1380340 PMID: 19300707; PubMed Central PMCID: PMC2618025. PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 23 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan 28. Eisenback JD, Hirschmann H. Morphological comparison of second-stage juveniles of several Meloido- gyne species (root-knot nematodes) by scanning electron microscospy. Scan Electron Microsc. 1979; 3(3): 223–230. 29. Esbenshade PR, Triantaphyllou AC. Use of enzyme phenotypes for identification of Meloidogyne Spe- cies. J Nematol. 1985; 17(1): 6–20. PMID: 19294051; PubMed Central PMCID: PMC2618420. 30. Esbenshade PR, Triantaphyllou AC. Isozyme phenotypes for the identification of Meloidogyne Species. J Nematol. 1990; 22(1): 10–15. PMID: 19287683 31. Esbenshade PR, Triantaphyllou AC. Electrophoretic methods for the study of root-knot nematode enzymes. In: Barker KR, Carter CC, Sasser JN, editors. An Advanced Treatise on Meloidogyne, Vol- ume II Methodology. North Carolina: North Carolina State University Graphics; 1985. p.115–123. 32. Adam MAM, Phillips MS, Blok VC. Molecular diagnostic key for identification of single juveniles of seven common and economically important species of root-knot nematode (Meloidogyne spp.). Plant Pathol. 2007; 56(1): 190–197. https://doi.10.1111/j.1365-3059.2006.01455.x 33. Vrain TC, Wakarchuk DA, Laplantelevesque AC, Hamilton R I. Intraspecific rDNA restrietion fragment length polymorphisrms in the Xiphinema americanum group. Fundam. Appl. Nematol. 1992; 15(6): 563–573. https://doi.10.1159/000156634 34. De Ley P, Felix MA, Frisse LM, Nadler SA, Sternberg PW, Thomas WK. Molecular and morphological characterization of two reproductively isolated species with mirror-image anatomy (Nematoda: Cepha- lobidae). Nematology. 1999; 1(6): 591–612. https://doi.10.1163/156854199508559 35. Powers TO, Harris TS. A polymerase chain reaction method for identification of five major Meloidogyne species. J Nematol. 1993; 25(1): 1–6. https://doi.10.1186/1756-8722-2-42 PMID: 19279734; PubMed Central PubMed Central PMCID: PMC2619349. 36. 37. Itoh Y, Ohshima Y, Ichinohe M. A root-knot nematode, Meloidogyne mali n. sp. on apple-tree from Japan (Tylenchida: Heteroderidae). Appl Entomol Zool. 1969; 4(4): 194–202. https://doi.10.1303/aez. 4.194 Franklin MT. A British root-knot nematode, Meloidogyne artiellia n.sp. J Helminthol. 1961; 35: 85–92. https://doi.org/10.1017/s0022149x00017600 PMID: 13701441. 38. Handoo ZA, Nyczepir AP, Esmenjaud D, Beek JGVD, Castagnone-Sereno P, Carta LK, et al. Morpho- logical, Molecular, and differential-host characterization of Meloidogyne floridensis n. sp. (Nematoda: Meloidogynidae), a root-knot nematode parasitizing peach in florida. J Nematol. 2004; 36(1): 20–35. PMID: 19262784; PubMed Central PMCID: PMC2620741. 39. Franklin MT. A root-knot nematode, Meloidogyne naasi n.sp. on field crops in England and Wales. Nematologica. 1965, 11(1): 79–86. https://doi.10.1163/187529265X00500 40. Wu Y. [Identification of a new species of root-knot nematode parasitizing citrus in Fujian province, China]. Master’s Thesis. Fuzhou: Fujian Agriculture and Forestry University. 2011. Chinese. https://doi. 10.7666/d.y1878300 41. Yang BJ, Wang QL, Feng RZ. [Meloidogyne Kongin.sp.(Nematoda: Meloidogynidae) a root-knot nema- tode parasitizing citrus sp. in guangxi China]. Journal of Guangxi Agricultural College. 1988; 7(3): 1–9. Chinese. https://doi.CNKI:SUN:GXNB.0.1988-03-000 42. Liu GK, Zhang SS. [Description of a new species of root-knot nematodes, Meloidogyne dimocarpus n. sp.]. Journal of Shenyang Agricultural University. 2001; 32(3): 167–172. Chinese. https://doi.CNKI: SUN:SYNY.0.2001-03-002 43. Handoo ZA, Skantar AM, Carta LK, Erbe EF. Morphological and Molecular characterization of a new root-knot nematode, Meloidogyne thailandica n. sp. (Nematoda: Meloidogynidae), Parasitizing Ginger (Zingiber sp.). J Nematol. 2005; 37(3): 343–353. PMID: 19262883; PubMed Central PMCID: PMC2620980. 44. Kiewnick S, Frey J E, Braun-Kiewnick A. Development and validation of LNA-based quantitative real- time PCR assays for detection and identification of the root-knot nematode Meloidogyne enterolobii in complex DNA backgrounds. Phytopathology. 2015; 105(9): 1245–1249. https://doi.org/10.1094/ PHYTO-12-14-0364-R PMID: 25775103. 45. Carneiro RMDG, Correa VR, Almeida MRA, Gomes ACMM, Deimi AM, Castagnone-Sereno P, et al. Meloidogyne luci n. sp. (Nematoda: Meloidogynidae), a root-knot nematode parasitising different crops in Brazil, Chile and Iran. Nematology. 2014; 16(3): 289–301. https://doi.10.1163/15685411-00002765 46. 47. Janssen T, Karssen G, Topalovic O, Coyne D and Bert W. Integrative taxonomy of root-knot nematodes reveals multiple independent origins of mitotic parthenogenesis. PLoS One. 2017; 12(3): 1–31. https:// doi.org/10.1371/journal.pone.0172190 PMID: 28257464; PubMed Central PMCID: PMC5336219. Liao JL, Jiang H, Sun LH, Hu XQ, Zhang ZR, Wang XR. [Identification of species and race of root-knot nematodes on crops in southern china]. Journal of Huazhong Agricultural University. 2003; 22(6): 544– 548. Chinese. PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 24 / 25 PLOS ONEA new root-knot nematode, Meloidogyne vitis sp. nov. (Nematoda: Meloidogynidae), parasitizing grape in Yunnan 48. Baidoo R, Joseph S, Mengistu TM, Brtto JA, Mcsorley R, Stamps RH, et al. Mitochondrial haplotype- based identification of root-knot nematodes (Meloidogyne spp.) on Cut Foliage Crops in Florida. J Nematol. 2016; 48(3): 193–202. https://doi.org/10.21307/jofnem-2017-027 PMID: 27765993; PubMed Central PMCID: PMC5070932. 49. Maleita CM, Almeida MSF, Vovlas N, Abrantes I. Morphological, Biometrical, Biochemical, and Molecu- lar characterization of the coffee root-knot nematode Meloidogyne megadora. Plant Dis. 2016; 100(8): 1725–1734. https://doi.org/10.1094/PDIS-01-16-0112-RE PMID: 30686217 50. Ge JJ. [Chinese entry phytosanitary pests]. 1st ed. Beijing: China agricultural press. 2018. Chinese. 51. Smith T, Brito JA, Han H, Kaur R, Cetintas R, Dickson DW. Identification of the peach root-knot nema- tode, Meloidogyne floridensis, using mtDNA PCR-RFLP. Nematropica. 2015; 45(1): 138–143. 52. Perry RN, Moens M, Starr JL. Root-knot Nematodes. CABI Publishing, 2009. 53. Castillo P, Vovlas N, Troccoli A, Liebanas G, Rius JEP, Landa BB. A new root-knot nematode, Meloido- gyne silvestris n.sp.(Nematode: Meloidogynidae), parasitizing European holly in Northern Spain. Plant Pathol. 2009; 58: 606–619. https://doi.10.1111/j.1365-3059.2008.01991.x 54. Powers TO, Todd TC, Burnell AM, Murray PCB, Fleming CC, Szalanski AL, et al. The rDNA internal transcribed spacer region as a taxonomic marker for nematodes. J Nematol. 1997; 29(4): 441–450. https://doi.10.1016/S0022-1910(97)00079-6 PMID: 19274180; PubMed Central PMCID: PMC2619808. 55. Blok VC, Phillips MS, Fargette M. Comparison of sequences from the ribosomal DNA intergenic Region of Meloidogyne mayaguensis and other major tropical root-knot nematodes. J Nematol. 1997; 29(1): 16–22. https://doi.10.1006/jipa.1996.4645 PMID: 19274129; PubMed Central PMCID: PMC2619761. 56. Landa BB, Rius JEP, Vovlas N, Carneiro RMDG, Maleita CMN, Abrantes IMO, et al. Molecular charac- terization of Meloidogyne hispanica (Nematoda, Meloidogynidae) by phylogenetic analysis of genes within the rDNA in Meloidogyne spp. Plant Dis. 2008; 92(7): 1104–1110. https://doi.org/10.1094/PDIS- 92-7-1104 PMID: 30769531 57. Rubinoff D and Holland BS. Between two extremes: mitochondrial DNA is neither the panacea nor the nemesis of phylogenetic and taxonomic inference. Syst Biol. 2005; 54(6): 952–961. https://doi.org/10. 1080/10635150500234674 PMID: 16385775. 58. Hu M and Gasser RB. Mitochondrial genomes of parasitic nematodes-progress and perspectives. Trends Parasitol. 2006; 22(2): 78–84. https://doi.org/10.1016/j.pt.2005.12.003 PMID: 16377245. 59. Garcia L E, Sanchez-Puerta MV. Comparative and evolutionary analyses of Meloidogyne spp. based on mitochondrial genome sequences. PLoS One. 2015; 10(3): 1–14. https://doi.org/10.1371/journal. pone.0121142 PMID: 25799071; PubMed Central PMCID: PMC4370701. 60. Kiewnick S, Holterman M, Elsen SVD, Megen HV, Frey JE, Helder J. Comparison of two short DNA bar- coding loci (COI and COII) and two longer ribosomal DNA genes (SSU & LSU rRNA) for specimen iden- tification among quarantine root-knot nematodes (Meloidogynespp.) and their close relatives. Eur J Plant Pathol. 2014; 140(1): 97–110. https://doi.10.1007/s10658-014-0446-1 61. Ahmed M, Sapp M, Prior T, Karssen G, Back MA. Technological advancements and their importance for nematode identification. Soil. 2016; 2(2): 257–270. https://doi.10.5194/soil-2-257-2016 62. Muniz MFS, Campos VP, Castagnone-Sereno P, Castro JMC, Almeida MRA, Carneiro RMDG. Diver- sity of Meloidogyne exigua (Tylenchida: Meloidogynidae) populations from coffee and rubber tree. Nematology. 2008; 10(6): 897–910. https://doi.10.1163/156854108786161418 63. Fourie H, Zijlstra C, Mcdonald AH. Identification of root-knot nematode species occurring in South Africa using the SCAR-PCR technique. Nematology. 2001; 3(7): 675–680. https://doi.10.1163/ 156854101753536046 64. Ye WM, Zeng YS, Kerns J. Molecular Characterisation and diagnosis of root-knot nematodes (Meloido- gyne spp.) from turfgrasses in North Carolina, USA. PLoS One. 2015; 10(11): 1–16. https://doi.org/10. 1371/journal.pone.0143556 PMID: 26599462; PubMed Central PMCID: PMC4658007. 65. Htay CC, Peng H, Huang WK, Kong LG, He WT, Holgado R, et al. The development and molecular characterization of a rapid detection method for rice root-knot nematode (Meloidogyne graminicola). Eur J Plant Pathol. 2016; 146(2): 281–291. https://doi.10.1007/s10658-016-0913-y 66. Randig O, Bongiovanni M, Carneiro RMDG, Castagnone-Sereno P. Genetic diversity of root-knot nem- atodes from Brazil and development of SCAR markers specific for the coffee-damaging species. Genome. 2002; 45(5): 862–870. https://doi.org/10.1139/g02-054 PMID: 12416618. 67. Zijlstra C, Hoof RV, Donkers-Venne D. A PCR test to detect the cereal root-knot nematode Meloidogyne naasi. Eur J Plant Pathol. 2004; 110(8): 855–860. https://doi.10.1007/s10658-004-2492-6 PLOS ONE \| https://doi.org/10.1371/journal.pone.0245201 February 3, 2021 25 / 25 PLOS ONE
10.1371_journal.pone.0283829	RESEARCH ARTICLE Medication adherence and associated factors among psychiatry patients at Asella Referral and Teaching Hospital in Oromia, Ethiopia: Institution based cross sectional study Dinkinesh Begna GudetaID 1, Kassech Leta1, Birhanu Alemu2, Usha Rani Kandula2 1 Department of Psychiatry, College of Health Science and Medicine, Arsi University, Asella, Ethiopia, 2 Department of Nursing, College of Health Science and Medicine, Arsi University, Asella, Ethiopia dinkinesh.begna@yahoo.com Abstract Background a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Gudeta DB, Leta K, Alemu B, Kandula UR (2023) Medication adherence and associated factors among psychiatry patients at Asella Referral and Teaching Hospital in Oromia, Ethiopia: Institution based cross sectional study. PLoS ONE 18(4): e0283829. https://doi.org/10.1371/journal. pone.0283829 Editor: Muhammad Junaid Farrukh, UCSI University, MALAYSIA Received: October 29, 2022 Accepted: March 16, 2023 Published: April 13, 2023 Copyright: © 2023 Gudeta et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The data used for this manuscript are available upon request from the ethics committee of Arsi University College of Health Science and Medicine (pr.arsiu@arsiun.edu. et). Because the dataset file contains personal information, it may not be ethical to publicly deposit the data. Funding: The Arsi University’s Research and Dissemination Office (Project number: CoHS/R/ 0091/2021/2022) provided funding for this study. Medication adherence is the first and main determinant of treatment success. It is defined by world health organization as “the degree to which the person’s behavior corresponds to the agreed recommendations from a health care provider”. Non-adherence is a multi-facto- rial phenomenon that can result from five major interacting factors. These are health team and health system-related factors; patient-related factors; therapy-related factors; socio- economic factors; and condition-related factors. The prevalence of non-adherence in mental illness was found to be 40% to 60% world wide. In developing countries, the magnitude of poor adherence is expected to increase. So this study aimed to assess medication adher- ence status and its associated factors among psychiatric patients in Asella Referral and Teaching Hospital in Oromia, Ethiopia. Methods An institution-based cross-sectional study was conducted from March 18, 2022 to May 25, 2022, with a total sample of 422 patients. Medication adherence was measured by a modi- fied version of the medication adherence rating scale in the psychiatric setting to determine treatment adherence status, and unstructured questionnaires were assessed by interview- ing the patient. Additional data concerning the medication-taking behavior of the patient was collected from caregivers. Bivariate logistic regression was performed to see the association between each explanatory variable and the outcome variable. The odds ratio and 95% confi- dence interval were used to see the association between treatment adherence and the strength of the link. Results A total of 395 study participants were interviewed, making a response rate of 93.6%. The prevalence of treatment adherence was 246(62.3%). Medication adherence show high association with lifetime alcohol use [AOR: 3.18, 95% CI:1.31–7.72] compared to those who PLOS ONE \| https://doi.org/10.1371/journal.pone.0283829 April 13, 2023 1 / 14 PLOS ONEThe funding sources had no input towards the planning of the study, collecting and analyzing the data, choosing to publish, or writing the report. Competing interests: The authors have declared that no competing interests exist. Abbreviations: ARTH, Assela Referral and Teaching Hospital; ART, Anti Retro Viral Therapy; COVD, Corona Virus Disease; ETB, Ethiopian Birr; MMI, Major Mental Illness; OPD, Outpatient Department; SPSS, Software Package for Social Science; TB, Tuberculosis; VIF, Variance inflation factor; WHO, World Health Organization. Medication adherence and associated factors had no alcohol use histroy, and perceived stigma [AOR (95% CI: 2.31 (1.01–5.31)] com- pared with those who had no perceived stigma, where as adherence show low association with having slight or superficial insight about illness [AOR (95% CI: 0.25 (0.12–0.53)] com- pared to those who reported cured off their illness and belief in medication [AOR: 0.36, 95% CI: 0.16–0.81)] compared to those who didn’t belief in the medication they are taking. Conclusion The prevalence of mediation adherence was found to be lower. In this study, factors such as having the slight insight or poor insight about their illness and belief in the medication decreased medication adherence, whereas having an alcohol use history in their lifetime and perceived stigma increased medication adherence. For a better health outcome, aware- ness creation at an insight level needs to be worked on by psychiatric professionals working on the follow-up psychiatric patients at psychiatry clinic of Assela Referral and Teaching Hospital to enable them to well adhere to their medication. Introduction Medication adherence is the frist and main determinant of treatment success. It is defined by world health organization as “the degree to which the person’s behavior corresponds to the agreed recommendations from a health care provider”. Non-adherence to medicine is a major public health problem that is termed an "invisible epidemics”. The prevalence of non-adherence in mental illness was found to be 40% to 60% world wide [1]. In developing countries, the magnitude of poor adherence is expected to increase, and its impact will be very high. In Ethiopia, the prevalence of non-adherence was reported as 39.3%– 55.5%, and many factors were significantly associated with treatment adherence [3, 6–9]. Non adherence to medication continues to be a significant challenge for patients suffering from mental illnesses, physicians, and the healthcare system [1]. It is well known that many patients have difficulty in following treatment recommendations, especially with mental illness [2]. This is due to some differences from other areas of medicine, such as chronicity with a time of exacerbation and relapse, poor insight into the illness process, patients feeling better and cog- nitive disturbance, and alternative treatment choices which is the traditional one [3–5]. In par- ticular, in severe mental illnesses like schizophrenia, it is a common and critical issue that needs a solution [2]. Poor adherence is a multi-factorial phenomenon which can result from five major interact- ing factors; health team and health system related, patient related, therapy related, social /eco- nomic and clinical factors [2]. According to some research findings poor adherence in mental illness is associated with poor social support, higher self-stigma, perceived spiritual causation of mental illness and being employed [3, 4]. Of these health system related factors may vary from country to country or even from area to area within a given country where the factors may include lack of health insurance coverage, due to different barrier to high quality health care and having lower socio economic status [6]. Inability to well adhere to a medication regimen may not affect only the patient, it also affect health care effectiveness. Broadly it may result in poor health, higher healthcare costs, relapse of illnesses, increase risk of suicide, increased risk of dependency, presence of inter PLOS ONE \| https://doi.org/10.1371/journal.pone.0283829 April 13, 2023 2 / 14 PLOS ONEMedication adherence and associated factors episodic symptoms, frequent hospital admission, difficulty to return back to previous func- tionality, increased risk of absents and rebound effect and increase rate of mortality [1]. Because of the multifactorial reasons for non-adherence, also in our country, many risk fac- tors might exist because of the difference in socio-demographic and economic factors, patient- related, health care and health system related, and clinical-related factors. By this time, people even fear going to health facilities due to the COVID-19 pandemic, so it is also the other hin- dering factor. Despite the existence of solutions for treatment non-adherence and its determinants, since it is perceived as a minor health issue even if it leads to a variety of undesirable health conse- quences, there is a scarcity of information on the area of mental illness in our country, particu- larly in the Arsi Zone, Oromia region to our knowledge of searching. Therefore, we were interested in determining medication adherence status and factors associated with mental ill- ness. The current study will have a significant contribution in that it will help to set strategies to overcome perceptual and practical barriers to adherence. Since it is an indicator of health- care effectiveness, it can also enhance the flow of patient visits to psychiatry clinics. It involves good patient outcomes and support health policy-making issues. Methods and materials Study setting The study was conducted at Asella Referral and Teaching Hospital’s (ARTH) outpatient psy- chiatric clinic. Asella is a town in Ethiopia’s Oromia regional state, and it is the capital city of Arsi zone. It is 175 kilometers North of Addis Ababa, the country’s capital. ARTH is a well- known government institution in Asella town, and it is a teaching and tertiary level hospital that provides inpatient and outpatient health services for over 3.5 million people in Ethiopia’s North East [7]. More than 60,000 patients got psychiatric service from this hospital. This care is given by 1 psychiatrist, 4 BSC psychiatric nurses, and 2 MSc psychiatry professionals. There is no inpatient psychiatric service in this hospital. Study period and design A hospital-based cross-sectional study was conducted in the psychiatric OPD of ARTH from March 18, 2022 to May 25, 2022. Source and study population The source population were patients who came for follow-up to take medication at Asella Referral and Teaching Hospital. All psychiatric patients who came to take their treatment dur- ing the data collection period are included in the study population. Eligibility criteria The study included all patients aged 15 and above who came for follow-up at the psychiatric clinic and whose charts clearly indicated a diagnosis of major mental illness, who were calm and took their treatment on their second visit and on follow-up. On the other hand patients who were not able to communicate because of their illness were excluded. Study variables The treatment adherence status of psychiatric patients, which was assessed by a validated and modified version of the medication adherence rating scale and classified as adherent or non- adherent, was a dependent variable. whereas socio-demographic and economic factors (age, PLOS ONE \| https://doi.org/10.1371/journal.pone.0283829 April 13, 2023 3 / 14 PLOS ONEMedication adherence and associated factors gender, income, occupation, education, religion, ethnicity, area of residence, availability of social support), patient-related factors (smoking, alcohol use, other substance use, patient atti- tude, belief on medication, fear of stigma, level of satisfaction, and fear of COVID-19 transmis- sion), health-system-related factors (time duration for follow-up, creation of awareness on treatment adherence, availability of medication) and clinical factors such as duration of the ill- ness, insight level, severity of illness, side effect of medication, dosage, frequency per day) were independent variables. Operational definition. • Major mental illness: mental illness with the diagnosis of major depressive disorder, schizo- phrenia, Bipolar disorders, brief psychotic disorder and schizophreniform • Medication adherence status: Medication adherence was measured by modified version of medication adherence rating scale with [8] with a total of 8 item where item 1 to 7 have choice of Yes = 1, No = 0 whereas item number 8 have choices A to E, A = 0, B to E = 1. Those who score 8 out of 8 are termed adherents, where a score of less than 8 indicates non- adherence • Poor adherent- score of less than 6 (0–5) • Partially adherent–score of 6 to 7 • Adherent– 8 out of 8 score • Psychiatry patient: patient with major mental illnesses • Associated factor: factor that lead the patient for non-adherence • Patients: clients 15 years old and above Sample size estimation/determination/ The sample size was calculated using the single population proportion formula. By using the proportion of treatment non-adherence for major mental illness in concurrent cases and for associated factors (since there is no similar study in Ethiopia up to the knowledge of the inves- tigators) of 50%, and 95% confidence level, 5% tolerable margin of error and a possible non- response rate of 10%, the final sample size was 422. Sampling procedure Simple randam sampling technique was used. During the data collection dates, when patients’ charts were brought to the clinic, they were checked for inclusion criteria. The charts were dif- ferentiated from new patients, then, by lottery method or randomly, we chose those who would be involved in the study. If a patient is absent while the chart is brought, another patient who fulfills the criteria is added. The patients on follow-up recorded in the record book were the sampling frame of this study. Data collection procedure Data was prepared in English, then it was translated in to local language, Afan Oromo and Amharic and back to English. Data was collected by the preference of patients by either of the two language from respondents. Primary data was collected from individual patients. These were structured questions where the contents were socio-demographic and economic factors; clinical-related factors; health system-related factors; and patient-related factors. In addition to PLOS ONE \| https://doi.org/10.1371/journal.pone.0283829 April 13, 2023 4 / 14 PLOS ONEMedication adherence and associated factors this, four questions concerning unresolved reasons for not properly taking medication at fam- ily level and possible solutions they expect were developed to be gathered from care givers. For those patients who came alone, questionnaires designed to be collected from caregivers were collected by phone. Secondary data was gathered from the patient’s chart by using the already prepared checklist that contains 7 items in English. The data was collected by six psychiatric professionals. The principal and coinvestigators provided supervision during the data collec- tion period. Data quality assurance Medication adherence was measured by validated and modified version of the medication adherence rating scale with internal consistency (Cronbach’s alpha) = 0.75 and reliability of 0.83) [8]. Additionally, other questionnaires were prepared after reviewing literature in English. Then it was translated into Amharic and Afan Oromo and back to English to ensure consistency. Data collection took place by Afan Oromo and Amharic languages according to the language fluency of patients. Before starting the data collection, a pre-test of questionnaires was done on 10% of psychiatric patients attending service at Adama General Hospital to pre- vent cross-contamination of information, and it was performed in order to check the language clarity and consistency of the questionnaire. Corrections to words on the questionnaire that were ambiguous for respondents were made after the pre-test. There was supervision. The questions’ completeness was checked and corrected before the patients left the mental health clinic. For uniformity and accuracy, the collected data was entered into a computer by utilizing a double data entry procedure. Data processing and analysis The data was cleaned for completeness, coded, entered, and analyzed using the Statistical Pack- age for Social Science (SPSS) version 24.00. The summary of the descriptive analysis was calcu- lated and displayed as frequency and percentage (categorical data), and mean for continuous data. Bivariate analysis was performed to see the association between the independent and the outcome variable. Multi-collinearity was checked to see if there was an association between independent variables and those with a variance inflation factor (VIF) value of 1 to 5 were taken, as they have no correlation, so they were used in the final model. Variables with a p- value of less than 0.25 in the bivariate analysis were candidate variables for multivariable analy- sis. To identify significant factors in the model, first the variables were filtered by using the backward LR, and then, after some variables were removed, the remaining variables were pro- cessed by the enter method. Model fitness was assessed using the Hosmer and Lemeshow test. After adjustment, an independent variable with a p value less than or equal to 0.05 was consid- ered significantly associated with the outcome variable. It was reported using an adjusted odds ratio with a 95% confidence interval. Ethical consideration The Arsi University College of Health and Medical Science Ethical Review Committee granted us ethical approval. All adult participants provided verbal informed consent, and minor partic- ipants provided informed consent after receiving a detailed description of the study from their parents/legal guardians. All methods were carried out in accordance with the applicable guide- lines and regulations. Patients identified as non-adherent groups were advised to properly take their medication, and family members caring for the patients were given health information. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283829 April 13, 2023 5 / 14 PLOS ONEMedication adherence and associated factors Results Socio-demographic and economic factors A total of 395 patients were interviewed, which makes a response rate of 93.6%. Two hundred twenty-nine (58%) of the respondents were females. The minimum age was 15, whereas the maximum age was 78, with a mean age of 33 and a standard deviation of +11.7. A large num- ber of the respondents lie in the age group of 30–39, which accounts for 127 (32.2%). Most of them are Muslim in religion and Oromo in ethnicity; 235 (59.5%) and 322 (81.5%) respec- tively. Most were from rural areas (56.7%). Two hundred thirty-five (59.5%) of respondents said they had only learned in elementary school, while 43 (10.9%) said they had a diploma or higher. The mean income was 1110.30 Ethiopian Birr (ETB) with a standard deviation of +- 2065.55ETB. Almost two thirds of the respondents had no well-known income; they account for 293 (74.2%), whereas 3 (0.8%) had the maximum range of income in the group, which lies between 10,055 and 11,305 ETB. More than half of the respondents were married 206 (52.2%). Regarding the presence of an individual to remind them to take their medication, 358 (90.6%) reported that they had an individual who reminded them to take their medication (Table 1). Patient related factors Among the respondents, 41 (10.4%), 34 (8.6%) and 38 (9.6%) had a history of Khat, tobacco and alcohol use in their lifetimes, respectively. Other substances like cannabis and marijuana were rarely used by 12 (3%) of the study participants. In terms of patients’ belief in the medica- tion they are taking, 354 (89.6%) believed that the medication could cure them. 264 (66.8%) of Table 1. Frequency distribution of respondents by socio demographic and economic factors in Asella Referral and Teaching Hospital 2022. Variables Categories Frequency Percent (%) Sex Age Religion Ethnicity Residence Education Marital status Living With Male Female 15–19 20–29 30–39 40–49 50–59 Christian Muslim Oromo Amhara and other Urban Rural Have no formal education Grade 1–8 Grade 9–12 Diploma and above Single Married Other (Divorced and widowed) Alone With family Other (with non-supportive individual) https://doi.org/10.1371/journal.pone.0283829.t001 166 229 32 125 127 62 49 160 235 322 73 171 224 62 235 55 43 154 206 35 36 315 44 42 58 8.1 31.6 32.2 15.7 12.5 40.5 59.5 81.5 18.5 43.3 56.7 15.7 59.5 13.9 10.9 39 52.2 8.9 9.1 89.9 11 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283829 April 13, 2023 6 / 14 PLOS ONETable 2. Frequency distribution of respondents by patient related factors in Asela Referral and Teaching Hospital 2022. Medication adherence and associated factors Variables Life time Chat Use Current Chat use Life time Alcohol use Current Alcohol use Life time Tobacco use Current Tobacco use Other substance (Marijuana and Shisha) use Belief in medication Reason for taking medication Perceived Stigma Having fear of CVD 19 while coming for follow up Had history of stopping Follow up due to CVD 19 Current fear of CVD19 transmission Feature plan to Continuing treatment with the pandemic Satisfaction https://doi.org/10.1371/journal.pone.0283829.t002 Categories Frequency Present Yes No Yes No Yes No Yes No Yes No Yes No Yes No Have belief /positive/ Have no belief/negative/ Good Poor Yes No Yes No Yes No Yes No Yes I will stop Satisfied Unsatisfied 41 354 31 364 38 357 37 358 34 361 26 369 12 383 354 41 264 131 37 358 24 371 10 385 26 369 388 7 375 20 10.4 89.6 7.8 92.2 9.6 90.4 9.4 90.6 8.6 91.4 6.6 93.4 3 97 89.6 10.4 66.8 33.2 9.4 90.6 6.1 93.9 2.5 97.5 6.6 93.4 98.2 1.8 94.9 5.1 the respondents have a good understanding of the reason they took the medication. 37 (9.4%) of them reported perceived stigma, while the rest did not. During the time of the CVD-19 pandemic, only 24 (6.1%) reported fear of coming for follow up, whereas 10 (2.5%) had a history of cutting down their medication due to fear of transmission. Even iff the pan- demic will continue, 388 people (98.2%) said they would keep taking their medication on follow up. Only 20 (5.1%) were unsatisfied with the service they obtained from the psychia- try OPD (Table 2). Health system related factors Among the study participants, 131 (33.2%) reported that there is a problem with medication accessibility in the hospital pharmacy and 145 (36.7%) of the respondents reported that the cost is expensive. Only 86 (21.8%) reported that they waited more than two hours while they came for their follow up. 263 (66.6%) and 344 (87.1%) were confirmed as they got information from professionals working in the clinic on the side effects of medication and medication adherence (Table 3). PLOS ONE \| https://doi.org/10.1371/journal.pone.0283829 April 13, 2023 7 / 14 PLOS ONEMedication adherence and associated factors Table 3. Frequency distribution of respondents by health system related factors in Asella Referral and Teaching Hospital 2022. Variables Medication availability problem Cost of medication in the hospital pharmacy Categories Yes No Expensive Cheap Unknown I use insurance Costly level of the medication in the hospital pharmacy Extremely expensive Follow up waiting time Information on side effect of medication by professionals Information about medication adherence by professionals https://doi.org/10.1371/journal.pone.0283829.t003 Medium cost Balanced cost Unknown I use insurance Above 2hour Below 2hour Yes I got No I didn’t got Yes I got No I didn’t got Number Percent 131 264 145 208 42 53 101 199 42 86 309 263 132 344 51 33.2 66.8 36.7 52.7 10.6 13.4 25.6 50.4 10.6 21.8 78.2 66.6 33.4 87.1 12.9 Clinical factors According to the information obtained from their chart, 255 (64.6%) were diagnosed as cases of schizophrenia and taking atypical antipsychotics, 151 (38.2%) and 122 (30.9%) were on con- ventional antipsychotics. One hundred sixty-nine (42.8%) of respondents had slight or superfi- cial insight about their illness, and 78 (19.7%) reported being cured of their illness. Concerning severity estimation by the patients, 271 (68.6%) reported that their illness was severe and 49 (12.4%) reported that it is mild. Thirty patients have different reports of side effects currently which was also reported by professional observation. Medical illness comor- bidity accounted for 24 (6.1%) of the 27 (6.6%) with high epilepsy comorbidity (Table 4). Information taken from caregivers to obtain additional data All respondents were interviewed face-to-face and by phone to add additional information that may be associated with the medication-taking behavior of the respondents. Accordingly, among 395 close family members, 132 (33.4%) reported that the most common reason for not properly taking the medication was a lack of insight and hesitation as a result. For the question asked the difficult and unsolved issue by them not to properly take medication, 130 (32.9%) of care givers reported lack of insight and hesitation as a result was found the difficult and still unsolved problem, followed by care givers’ report that there is no difficult problem that accounts for 109 (27.6%). For the possible solution for proper medication taking behavior, 130 (32.9%) of caregivers reasoned that free and sufficient medication accessibility would be a pos- sible solution, followed by strong family support at 84 (21.3%) (Table 5). Prevalence of medication adherence status Fig 1 indicates the prevalence of respondents’ medication adherence status at Assela Referral and Teaching Hospital in Ethiopia. Among the respondents, 149 (37.7%) were non-adherent and 246 (62.3%) of them were adherent to medication. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283829 April 13, 2023 8 / 14 PLOS ONETable 4. Frequency distribution of respondents by clinical factors in Asella Referral and Teaching Hospital 2022. Variables Categories Frequency Number Percent Medication adherence and associated factors Duration of illness Insight Severity Rank by respondents 6–58 59–111 112–164 165–217 218–480 Slight (superficial) insight Poor /no insight Say cured Severe Moderate Mild Diagnosis of the respondent Schizophrenia Other psychotic disorder MDD and other DD Bipolar disorder Anxiety disorders 22 Medication the respondent is taking Atypical antipsychotics Medication side effect Medication frequency Type of side effect Type of Comorbid illness Conventional antipsychotics Antidepressant Combination of antipsychotics and antidepressant Combination of antipsychotics and Mood stabilizer Yes No Once per day Twice per day Other option (injectable, once per month) Tremor (hand and body) Weight gain Other (dry mouth, drowsiness and sedation) Drawling of saliva, muscular rigidity Other (Malignant) No side effect yet Diabetes Miletus Hypertension Epilepsy and other neurologic disorder Other (Liver, kidney and heart problem) No comorbid illness 159 104 47 44 41 169 148 78 271 75 49 255 34 59 25 151 122 36 69 17 60 335 345 41 9 7 9 4 1 9 365 5 2 12 5 371 40.3 26.3 11.9 11.1 10.4 42.8 37.5 19.7 68.6 19 12.4 64.6 8.6 14.9 6.3 5.6 38.2 30.9 9.1 17.5 4.3 15.2 84.8 87.3 10.4 2.3 1.8 2.3 1 0.3 2.3 92.4 1.3 0.5 3 1.3 93.9 MDD = Major depressive Disorder, DD Depressive disorder https://doi.org/10.1371/journal.pone.0283829.t004 Factors associated with treatment adherence Lifetime alcohol use, belief in medication, perceived stigma, and illness insight were among the variables that showed a significant association with treatment adherence. The odds of med- ication adherence among respondents who had a lifetime alcohol use history [AOR (95%CI); 3.18 (1.31–7.72)] were 3 times higher than their counterparts. Individuals who believed in PLOS ONE \| https://doi.org/10.1371/journal.pone.0283829 April 13, 2023 9 / 14 PLOS ONETable 5. Frequency distribution of respondents by their caregivers stated factors at Asella Referral and Teaching Hospital 2022. Variable Categories Frequency Number Percent Medication adherence and associated factors The most critical problem for not taking medication properly financial problem Distance, lack of transportation Hopelessness Forgetting lack of insight, including hesitate to take medication lack of understanding about treatment adherence personal issues like poor social support and substance use religion (like using holly water) lack of cure belief cured because of side effect No reason Financial problem Distance, lack of transportation Belief cured Personal issues (poor social support, substance use and forgetting) lack of insight, hesitate Lack of understanding Religion Lack of cure No difficult reason Because of side effect Giving health information Free and enough medication Shortening waiting time I do no family support deep examination like test with machine Follow up should be long enough Difficult and unsolved problems Possible solutions set by givers https://doi.org/10.1371/journal.pone.0283829.t005 46 10 6 59 132 9 12 3 3 4 10 101 51 14 7 67 130 9 1 2 109 5 159 130 3 11 84 3 5 11.6 2.5 1.5 14.9 33.4 2.3 3 0.8 0.8 1 2.5 25.6 12.9 3.5 1.8 17 32.9 2.3 0.3 0.5 27.6 1.3 40.3 32.9 0.8 2.8 21.3 0.8 1.3 their medication were 65% less likely to be adherent than those who did not believe in their medication [AOR (95%CI); 0.36 (0.16–0.81)]. Respondents who had perceived stigma had 2 times higher chance of adherence to their medication compared to those who had no per- ceived stigma [AOR (95%CI); 2.31 (1.01–5.31)]. Individuals who had slighht insight compared to those who reported currently getting a cure from their illness were less adherent [AOR (95%CI); 0.25 (0.12–0.53)]. The educational status and medication availability problems in the hospitals’ pharmacy were found insignificant in the final model. (Table 6). Discussion The aim of this study was to assess the prevalence of treatment adherence and its associated fac- tors. The prevalence of treatment adherence was found to be 62.3%. This prevalence is nearly sim- ilar to a study conducted in Jimma, Nigeria with a prevalence of 60.5% and 55.7% respectively [3, 9]. Where as it was high compared with studies conducted in Bahir Dar at different years of study [10–12] and systematic review and Meta-analysis summary prevalence [11]. The possible reasons for the difference can be explained by differences in client background, time of the study and the PLOS ONE \| https://doi.org/10.1371/journal.pone.0283829 April 13, 2023 10 / 14 PLOS ONEMedication adherence and associated factors Fig 1. Prevalence of medication adherence status of respondents in Assela Referral and Teaching Hospital. https://doi.org/10.1371/journal.pone.0283829.g001 tool used for assessing adherence to medication. There are different factors for treatment adher- ence according to different findings because of variation in different aspects across the world, including setting, clients’ background, tools used, and other factors [11, 13]. In the current study, factors significantly associated with treatment adherence were found to be: life-time alcohol use, perceived stigma, belief about medication, and insight about illness. In this study, lifetime alcohol use history had 3 times the odds of adherence compared to those who had no history of using alcohol in their lifetime. It contradicts a systematic review of the literature, which found substance abuse to be a significant factor in medication non-adherence among patients with various major mental illnesses [4, 14] and excessive alcohol use is related to a lower medication adherence rate [15]. The possible reason for the difference might be due to respondents’ change in attitude toward medication taking as a result of getting help and advice from psychiatric professionals to stack on their medication. Those who had perceived stigma were 2 times more adherent com- pared to those who had no perceived stigma. This can be explained by the respondents’ inner motivation to be cured of their illness so that people cannot stigmatize them. Belief about medica- tion was the other factor which showed a significant association in which clients who believed in their medication were less adherent compared to those who had no belief in their medication. This can be due to an arbitrary answer by respondents to a closed-ended question. This is con- trary to the study finding in New York, which states that higher motivation for mental health treatment and recovery support was related to greater adherence [15, 16]. The difference might be due to fear of responding "I cannot believe in the medication you are giving to me." to the clini- cian. Some of the respondents may say yes only to prevent further questions. Concerning insight, PLOS ONE \| https://doi.org/10.1371/journal.pone.0283829 April 13, 2023 11 / 14 PLOS ONEMedication adherence and associated factors Table 6. Multivariate logistic regression model for selected factors with treatment adherence status of respondents at Asella Referral and Teaching Hospital 2022. Variables Education Life time Alcohol use Belief in medication perceived stigma Medication availability problem Insight about illness Categories level of adherence COR (95%CI) p value AOR (95%CI) P Value Adherent Non adherent No formal education grade 1–8 grade 9–12 diploma and above Yes No Yes No Yes No Yes No Slight insight Poor/No insight Say cured 29 157 32 28 31 215 213 33 28 218 90 156 108 71 67 33 78 23 15 7 142 141 8 9 140 41 108 61 77 11 0.47 (0.21–1.04)* 0.06 0.58 (0.24–1.41) .82 (1.07–0.54) 0.75 (0.32–1.7) 1 2.92 (1.25–6.82)* 0.01 1 0.82 1.27 (0.59–2.72) 0.48 0.97 (0.39–2.39) 1 3.18 (1.31–7.72)* 1 0.36 (0.16–0.81)* 0.01 0.35 (0.15–0.83)* 1 1 1.99 (0.91–4.36) 0.08* 2.31 (1.01–5.31)* 1 1 1.52 (0.97–2.36) 0.06* 1.55 (0.95–2.50) 1 0.29 (0.14–0.59)* 0.15 (0.07–0.3) 1 1 0.0011 0.25 (0.12–0.53)* 0.001 0.12 (0.06–0.26)* 0.23 0.52 0.95 0.01 0.01 0.04 0.07 0.00 0.00 *- significantly associated AOR- Adjusted Odds Ratio Adjusted factors- life time alcohol use, belief in medication, perceived stigma and insight about illness https://doi.org/10.1371/journal.pone.0283829.t006 those respondents who had slight insight and those who had no insight were less adherent to their medication than those who reported they got more cure from their illness. This is supported by studies’ findings which found that poor insight is one of the significant factors in medication non-adherence [4, 5, 9, 10, 14]. It is not consistent with the findings of study which found patients feeling better [4] and expectation of cure rather than continuing treatment leads for treatment non adherence [17]. In this study, additional data was gathered from family members who were caring for the patient to search for possible associated factors for medication non-adherence. This is due to the multifaceted factors that caused patients to deviate from their medication. So, unless care givers are included, it is impossible to solve the problem of non-adherence, which is recom- mended by some researchers [11, 15, 18]. Therefore, this finding part was supportive of the cli- ents’ data. They stated that the most critical problem for not taking medication properly was a lack of insight, which accounted for 33% of the total. Still, this is the problem that most family members did not have a solution for (33%). While family members asked for possible solutions for their significant other to adhere to their medication, (40%) reported giving health informa- tion for the patient followed by free and enough medication from the hospital (33%) might help to solve the current non-adherence issue, according to the caregiver’s opinion. Generally, it is impossible to conclude that adherence can be improved without the involvement of patients and significant others in their family members [18]. Limitation Because this study is cross-sectional type, it cannot determine the exact cause of treatment nonadherence and it cannot indicate non adherence effects on the patients. Secondary data were extracted from patients’ charts but not assessed from the patient. Some people were sim- ply saying yes for some of the close ended question. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283829 April 13, 2023 12 / 14 PLOS ONEMedication adherence and associated factors Strength Data were collected by professionals by using local language, translation of questions were done by concerned language department professionals and health information was given for non adherent patients after knowing the result. Conclusion The prevalence of medication adherence among people with major mental illness was not sat- isfactory. Factors like alcohol use, belief in medication, having perceived stigma and insight were found to be associated with medication adherence status. Alcohol use history and per- ceived stigma increase medication adherence, whereas having superficial insight, poor insight, and belief in medication decrease medication adherence. Recommendation Improving the medication adherence of people with major mental illness in the study area and tackling those associated factors will have a better contribution to the good outcome of patients. As a result, health care facilities, professionals, patients, and family members need to collaborate to address these problems. Awareness creation at an insight level needs to be worked on by psychiatry professionals working on follow-up of patients at ARTH psychiatry clinic to enable patients to well adhere to their medication. Deep understanding of adherence and true belief in their medication will help to increase their adherence level. Discussing patient-specific factors with the patient and significant other family members will further address the problem of adherence need to be addressed by psychiatry professionals working in this clinic. Acknowledgments We would like to thank the psychiatry professionals at Adama General Hospital and Asella Referral and Teaching Hospital for their assistance during data collection. Author Contributions Conceptualization: Dinkinesh Begna Gudeta, Birhanu Alemu. Data curation: Dinkinesh Begna Gudeta, Kassech Leta, Usha Rani Kandula. Formal analysis: Dinkinesh Begna Gudeta, Usha Rani Kandula. Funding acquisition: Dinkinesh Begna Gudeta. Investigation: Dinkinesh Begna Gudeta. Methodology: Dinkinesh Begna Gudeta. Project administration: Dinkinesh Begna Gudeta. Supervision: Kassech Leta, Birhanu Alemu, Usha Rani Kandula. Validation: Dinkinesh Begna Gudeta, Kassech Leta, Birhanu Alemu. Writing – original draft: Dinkinesh Begna Gudeta. Writing – review & editing: Dinkinesh Begna Gudeta, Kassech Leta, Birhanu Alemu, Usha Rani Kandula. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283829 April 13, 2023 13 / 14 PLOS ONEMedication adherence and associated factors References 1. Chapman SCE, Horne R. Medication nonadherence and psychiatry. Curr Opin Psychiatry. 2013; 26: 446–452. https://doi.org/10.1097/YCO.0b013e3283642da4 PMID: 23880592 2. Sabate´ E, World Health Organization. Adherence to long-term therapies: evidence for action. World Health Organization; 2003. 3. Mekuriaw B, Abera M, Tessema W, Agenagnew L, Dawud B, Abdisa E, et al. Medication Non-adher- ence and Use of Traditional Treatment Among Adult Psychiatric Patients in Jimma Town Treated at Jimma University Teaching Hospital Psychiatric Clinic. Community Based Cross-sectional Study, 2016. J Psychiatry. 2017; 21: 1–7. https://doi.org/10.4172/2378-5756.1000434 4. Press D. Psychiatrists ‘ awareness of adherence to antipsychotic medication in patients with schizo- phrenia: results from a survey conducted across Europe, the Middle East, and Africa. 2013; 121–132. https://doi.org/10.2147/PPA.S37534 PMID: 23390361 5. Ahmad I, Khalily MT, Hallahan B. Reasons associated with treatment non-adherence in schizophrenia in a Pakistan cohort. Asian J Psychiatr. 2018. https://doi.org/10.1016/j.ajp.2017.07.024 PMID: 28738259 6. Americans A. Psychotropic Medication Nonadherence Among United States Latinos: A Comprehensive Literature Review. 2009; 60: 21–23. https://doi.org/10.1176/appi.ps.60.2.157 PMID: 19176409 7. Asella Referral and Teaching Hospital. 2016. 8. Cuevas C De, Pe W. Psychometric properties of the eight-item Morisky Medication Adherence Scale (MMAS-8) in a psychiatric nate. 2015. https://doi.org/10.1016/j.ijchp.2014.11.003 PMID: 30487829 9. Access O. Prevalence and predictors of sub-optimal medication adherence among patients with severe mental illnesses in a tertiary psychiatric facility in Maiduguri, North-eastern Nigeria. 2015; 8688: 1–11. https://doi.org/10.11604/pamj.2015.21.39.6664 PMID: 26405475 10. Gebeyehu DA, Mulat H, Bekana L, Asemamaw NT, Birarra MK, Takele WW, et al. Psychotropic medi- cation non-adherence among patients with severe mental disorder attending at Bahir Dar Felege Hiwote Referral hospital, north west Ethiopia, 2017. BMC Res Notes. 2019;12. https://doi.org/10.1186/ s13104-019-4126-2 PMID: 30808408 11. Semahegn A, Torpey K, Manu A, Assefa N, Tesfaye G, Ankomah A. Psychotropic medication non- adherence and its associated factors among patients with major psychiatric disorders: A systematic review and meta-analysis. Syst Rev. 2020; 9: 1–18. https://doi.org/10.1186/s13643-020-1274-3 PMID: 31948489 12. Coskun S, Bagcivan G. Associated factors with treatment adherence of patients diagnosed with chronic disease: Relationship with health literacy. Appl Nurs Res. 2020; 151368. https://doi.org/10.1016/j.apnr. 2020.151368 PMID: 33092934 13. Ha¨ge A, Weymann L, Bliznak L, Ma¨ rker V, Mechler K, Dittmann RW. Non-adherence to Psychotropic Medication Among Adolescents–A Systematic Review of the Literature. 2018; 46: 69–78. 14. Higashi K, Medic G, Littlewood KJ, Diez T, Granstro¨ m O, de Hert M. Medication adherence in schizo- phrenia: Factors influencing adherence and consequences of nonadherence, a systematic literature review. Ther Adv Psychopharmacol. 2013; 3: 200–218. https://doi.org/10.1177/2045125312474019 PMID: 24167693 15. Magura S, Mateu PF, Rosenblum A. Mental Health and Substance Use Risk factors for medication non- adherence among psychiatric patients with substance misuse histories. 2013; 37–41. https://doi.org/10. 1080/17523281.2013.839574 PMID: 25309623 16. Gurmu AE, Abdela E, Allele B, Cheru E, Amogne B. Rate of Nonadherence to Antipsychotic Medica- tions and Factors Leading to Nonadherence among Psychiatric Patients in Gondar University Hospital, Northwest Ethiopia. 2014;2014. 17. Teferra S, Hanlon C, Beyero T, Jacobsson L, Shibre T. Perspectives on reasons for non-adherence to medication in persons with schizophrenia in Ethiopia: A qualitative study of patients, caregivers and health workers. BMC Psychiatry. 2013; 13: 1–9. https://doi.org/10.1186/1471-244X-13-168 PMID: 23773362 18. Miasso AI, Celso P, Telles P, Vedana G, Shasanmi R, Escobar FR, et al. Issues in Mental Health Nurs- ing Adherence to Psychotropic Medications and Associated Factors in Primary Health Care Adherence to Psychotropic Medications and Associated Factors in Primary Health Care. 2016;2840. https://doi. org/10.1080/01612840.2016.1214854 PMID: 27548290 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283829 April 13, 2023 14 / 14 PLOS ONE
10.1371_journal.pone.0283965	RESEARCH ARTICLE Study on the correlation between basketball players’ multiple-object tracking ability and sports decision-making Qifeng GouID 1,2¤, Sunnan LiID 2¤* 1 College of Physical Education, Northwest Normal University, Lanzhou, Gan Su, China, 2 College of Physical Education and Sports, Beijing Normal University, Beijing, China ¤ Current address: College of Physical Education and Sports, Beijing Normal University, Beijing, China * hanceli99@126.com a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 Abstract Background OPEN ACCESS Citation: Gou Q, Li S (2023) Study on the correlation between basketball players’ multiple- object tracking ability and sports decision-making. PLoS ONE 18(4): e0283965. https://doi.org/ 10.1371/journal.pone.0283965 Editor: Goran Kuvačić,University of Split, CROATIA Received: November 4, 2022 Accepted: March 21, 2023 Published: April 5, 2023 Copyright: © 2023 Gou, Li. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting information files. Funding: The present study was supported by the National Social Science Foundation of China (No. 22BTY055). The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Players’ multiple-object tracking (MOT) ability is very important in basketball because it may affect players’ sports decision-making (SDM), thus affecting the results of the game. The purpose of this study was to investigate the differences between expert and novice basket- ball players in MOT ability and SDM and to explore the correlation between basketball play- ers’ visual attention and SDM. Methods A total of 48 female basketball players (24 categorized in the expert group and 24 in the nov- ice group) participated in the MOT task in Experiment 1 and the basketball 3 vs. 3 games in Experiment 2. Experiment 1 examined the difference in dynamic visual attention character- istics between expert players and novice players by changing the tracking number. Experi- ment 2 examined the differences between expert players and novice players through the SDM of basketball 3 vs. 3 games. Sports decisions were evaluated by basketball experts. MOT ability and SDM ability were analyzed through Pearson correlation. Results The overall MOT accuracy of expert players (64.6%) and novice players (55.7%) was signifi- cantly different (χ2 = 59.693, P = 0.000). There was no significant difference in accuracy when tracking 2–3 targets (P > 0.05), but there was a significant difference in accuracy when tracking 4–6 targets (P < 0.05). The overall SDM accuracy of expert players (91.6%) and novice players (84.5%) was significantly different (χ2 = 31.975, P = 0.000). There was no significant difference between expert players and novice players in the accuracy of drib- bling decision-making (P > 0.05), but there was a significant difference in the accuracy of passing decision-making and shooting decision-making (P < 0.01). When tracking 4–5 tar- gets, the tracking score was positively correlated with the passing decision score and PLOS ONE \| https://doi.org/10.1371/journal.pone.0283965 April 5, 2023 1 / 12 PLOS ONEThe correlation between MOT ability and SDM dribbling decision score of expert players, and the tracking score of novice players was posi- tively correlated with the passing decision score (r > 0.6, P < 0.01). Conclusions First, the tracking accuracy of expert players was significantly higher than that of novice players, especially when tracking 4–6 targets. As the number of targets increased, accuracy decreased. Second, the accuracy of expert players’ SDM was significantly higher than that of novice players, especially in passing decision-making and shooting decision-making. Expert players exhibited fast and accurate SDM. Third, there was a correlation between MOT ability and SDM performance. The MOT ability of 4–5 targets was positively correlated with passing decision-making, which was statistically significant. The correlation between the MOT ability and SDM performance of expert players was greater and more significant. Having too many targets to track (more than 6) interfered with players’ decisions. Introduction "Decision" means consciously choosing from different possible actions [1]. Players’ profes- sional sports skills and speed are closely related to high-level performance in visual attention tasks [2]. In particular, the ability to make decisions under time pressure is crucial to the results of a match [3]. Efficient visual attention and rapid working memory processing at key nodes of the competition are the prerequisites for making accurate decisions during a match [4]. The first step of sports decision-making (SDM) is perception; neurons transmit informa- tion to the central nervous system that enable a person to make decisions [5]. From the per- spective of operation, the dynamics of perception, action and cognition can be described at two levels of analysis [6]. The first-level analysis determines information variables that guide behavior [7]. The second-level analysis characterizes the time evolution of this behavior [8]. The effectiveness of SDM refers to whether the decision-making performance conforms to the current situation and solves problems [5]. Sanchez et al. used questionnaires and interviews to study how high-level female basketball players make decisions in competitions, and the results showed that players are experts in making decisions during competitions [1]. More research required subjects to watch images or videos about which they would have to make decisions to reflect the accuracy of SDM, and the results showed that players made faster and more accurate decisions after viewing them [9], but there are few studies on on-site sports decisions. An eye tracker was used to show which areas players prefer to focus on when making sports decisions (e.g., an open area without offensive and defensive players [10]). Some studies have also used electroencephalogram (EEG), functional near-infrared spectroscopy (fNIRS) and functional magnetic resonance imaging (fMRI) techniques, and the prefrontal cortex plays governs motor decision-making [11]. An increasing number of studies have taken attention control as a driving factor for the limitation of visual short-term and working memory ability [12]. The ability to remember previously seen information in a short time is critical to many cognitive and perceptual processes and is called visual working memory (VWM) [12]. More than 80% of brain information comes from visual attention [13]. Good attention quality is key for players to effectively implement sports skills; it helps players to more effectively and selectively pay attention to, recognize and interpret visual information and provides sufficient information for further decision-making and responses [14]. In ball games, when and how long do players PLOS ONE \| https://doi.org/10.1371/journal.pone.0283965 April 5, 2023 2 / 12 PLOS ONEThe correlation between MOT ability and SDM turn to other more important clues? The ability to predict the direction and speed of the ball according to clues provided by the opponent’s action is often the key to victory [5]. Team ball games require players to be able to pay attention to the changes in multiple teammates, oppo- nents, and their positions at the same time; therefore, multiple-object tracking (MOT) ability is very important for playing team ball games [15]. Improving the attention ability of basketball players to promote the accuracy of players’ decision-making has become the focus of coaches and sports research experts in various countries [10]. We need to pay attention to and track the motion of multiple objects at the same time, a process referred to as MOT [16]. MOT combines the selectivity, limited capacity and subjec- tive effort of attention and can reflect multiple aspects of attention, such as selective attention, distributive attention, and continuous attention [17]. MOT performance is affected by many factors, such as the number of targets and nontargets [18] and the speed of object movement [19]. The research method used most is to divide the subjects into expert and novice groups and guide their training to identify the gap between the expert and novice groups [20]. Man- gine et al. conducted research on 12 National Basketball Association (NBA) players using 3D MOT tasks and found that players with good visual tracking speed performed better in games [21], and football players’ visual intervention improved SDM [22]. Rugby defenders performed better than offensive players and college students in MOT tasks [23]. With an increasing num- ber of targets, the tracking performance of the basketball players in the expert group was better than that in the novice group, the tracking performance of the guard was the best, and those playing the center position performed slightly better than those in the forward position [24]. Qiu et al. found that compared with intermediate or nonplayers, excellent basketball players showed better tracking performance when tracking 3 or 4 targets. However, no significant dif- ferences were found between intermediate players and nonplayers. In addition, no difference was observed among the three groups when tracking 2 targets [25]. When the number of tar- gets increased to 6 targets, basketball players had better tracking performance than college stu- dents who did not play sports [2]. Maarseveen et al. studied the sports decisions of high-level basketball players in situ of specific 3 vs. 3 basketball matches and found that players began to check possible choices earlier when deciding to shoot or pass the ball than when making break- throughs [26]. Skilled players aimed at the ball holder when defending and used peripheral vision to observe the movement of other players [27]. However, there are few studies on female subjects [28], yet some scholars found that expert players had no obvious advantage in MOT ability [20], team players in specific positions had no advantage in MOT tasks [29], and 3D MOT training did not significantly improve SDM [30,31]. More research supports that professional players have better MOT ability [17,32]. The ability of players to track multiple targets may be related to long-term skill training [10,33]. 3D MOT and decision-making task performance are related [34]. 3D MOT task training improves the decision-making accuracy of football players [33]. However, few people have studied the relationship between the number of items tracked during MOT tasks and SDM [10]. There- fore, this study used the expert-novice paradigm to verify the relationship between the number of MOT tasks tracked and SDM. This study assessed the relationship between MOT ability and SDM. Experiment 1 used MOT tasks to compare the dynamic visual attention characteristics of female expert and nov- ice basketball players. Experiment 2 compared the SDM of female expert and novice basketball players. Then, the relationship between the dynamic visual attention characteristics of basket- ball players and SDM was evaluated. We hypothesized that female expert basketball players’ MOT ability and SDM would be better than those of novice players and that there is a correla- tion between MOT ability and SDM ability. The correlation between the accurate score of PLOS ONE \| https://doi.org/10.1371/journal.pone.0283965 April 5, 2023 3 / 12 PLOS ONEThe correlation between MOT ability and SDM MOT and the accurate score of SDM of expert players and novice players is different; there- fore, it may be possible that the correlation of expert players is greater. Materials & methods Participants This research used GPower 3.1.9.7 (Germany) software to estimate the sample size and set the effect size as ηp2 = 0.03 and two-tailed α = 0.05 [35], and a statistical power of 0.80 was reached for 40 subjects [36]. Controlling for attrition, we selected 48 subjects (24 in each group) [2]. Forty-eight subjects were divided into either an expert group or a novice group according to their basketball experience level. The expert players were selected from the Northwest Division of the China University Basketball League Division One [10]. The age range of expert players was 18–22 years (mean: 20.31; SD: 2.25 years), with more than 9 years of experience per player (mean: 9.78; SD: 2.43 years) and over 15 hours (mean: 15.02; SD: 2.86 hours) of training per week in the past year. The novice group consisted of 24 students from the sports basketball elec- tive course of Northwest Normal University, aged 18–21 years (mean: 19.35; SD: 1.62 years), who had not participated in basketball training. All participants were female, right-handed, with normal or corrected vision, and the average game video time per week did not exceed 4 hours [37]. All subjects participated in the experiment and received compensation for their time. The study was approved by the Ethics Committee of Northwest Normal University (No. 20210812). All of the participants provided written informed consent prior to the start of the experiment. Apparatus In Experiment 1, a Thinkpad E15 laptop was used. The program environment was Windows 10, the display screen was 15.6 inches (34.4 cm × 19.4 cm), the screen resolution was 1920 × 1080 pixels, and the refresh rate was 60 Hz. MATLAB R2020b (The MathWorks Inc., Natick, MA, USA) software and Psychtoolbox 3 (version: 3.0.17) were used to program the MOT experiment. At the beginning of the experiment, a "+" symbol appeared in the middle of a gray background (38.20˚× 21.3˚) for 1000 ms. Then, the screen displayed 12 white balls (0.66˚ in diameter), of which 2–6 balls changed from white to blue, flashed 3 times (2 s) and were marked as targets. The balls that did not flash were considered as nontargets. After that, all the balls returned to their original white color. Then, the balls started to move indepen- dently at a speed of 5˚/s. During this process, there may have been occlusion between the balls. After 10 s, the balls stopped moving, and the same number of balls as the target were randomly selected and marked in red. The subject needed to judge how many targets were marked red, press the corresponding number key, and then begin the next test (e.g., in a 4-target condition, there were 4 randomly selected items in red at the end of tracking time, and the subject had to determine if any of them were actual targets; then, participants had to press keys correspond- ing to 0–4). In Experiment 2, two cameras (Canon, HFG50) were used to record the basketball 3 vs. 3 match. The camera was placed on the auditorium of the stadium, approximately 10 meters away from the competition site, such that it could capture the entire competition area. Players were identified by their jerseys and numbers, and the video was analyzed by Storm Video soft- ware (version: 5.81.0202.1111). Design In Experiment 1, the MOT task was used to examine the tracking performance under different target numbers. A mixed experimental design of 2 (groups) × 5 (number of targets) was used. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283965 April 5, 2023 4 / 12 PLOS ONEThe correlation between MOT ability and SDM The groups (experts and novices) were the variables between the subjects. The number of tar- gets (2–6) was an intrasubject variable, and the dependent variable was the tracking accuracy. The accuracy algorithm was defined as the percentage that each subject correctly selected the target in all experimental trials. In Experiment 2, the basketball 3 vs. 3 match was judged by three fixed basketball referees. SDM was evaluated by two basketball experts according to the game video and coding standards [33]. For the evaluation results with objections, a third bas- ketball expert was invited to discuss and assess. Before the experiment, 5 other experts unani- mously agreed that the coding standard was valid. One point represented an accurate decision, 0 points represented an inaccurate decision, and decisions that were both accurate and inaccu- rate were not coded. To avoid the impact of rapid responses on accuracy, we did not require subjects to respond quickly to ensure accuracy [25]. The experiment was accompanied by 1 tester to ensure completion. Procedure Experiment 1 was conducted in the laboratory of Northwest Normal University from August 25, 2021, to August 28, 2021. To familiarize the subjects with the experimental process, there were 8 preexperiments before the formal experiment. The distance between the subjects and the screen was approximately 60 cm. The experiment was composed of 5 blocks, each of which consisted of 2–6 targets. Each block had 30 trials, for a total of 150 trials. In each block, a white screen was displayed for 1 minute in the middle to alleviate participant eye fatigue. The experi- mental sequence was balanced within the subjects, and the whole experiment lasted approxi- mately 27 minutes. The experimental process is shown in Fig 1. The basketball 3 vs. 3 match in Experiment 2 was held in the Basketball Hall of Northwest Normal University from August 29, 2021, to September 1, 2021. Before the experiment, all sub- jects were asked to know the rules of 3 vs. 3 matches very well. Participants were randomly assigned to 8 teams in the same group, and each team was composed of 3 participants. Eight teams met randomly in 28 matches (no two teams met twice), and each match was 5 minutes. During the entire test, each person participated in 7 games, totaling 35 minutes. Statistical analysis The data were recorded in and collected by MATLAB R2020b software. SPSS 26.0 (SPSS, Inc., Somers, New York, USA) software was used to compare the differences in MOT ability and SDM performance and evaluate the correlation between MOT ability and SDM across the two Fig 1. MOT task process. https://doi.org/10.1371/journal.pone.0283965.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283965 April 5, 2023 5 / 12 PLOS ONEThe correlation between MOT ability and SDM groups. Pearson’s χ-squared test was used to compare the accuracy of the MOT and SDM per- formance, and Pearson’s correlation coefficient was used to compare the correlation between the MOT and SDM performance. An ɑ level of 0.05 was considered statistically significant for all of the statistical comparisons. SDM ability was coded using standardized tools [33,38]. Then, the scores of each player in each game were added to calculate the total score. The corre- lation was calculated by the SDM accuracy and the accuracy for the MOT task. The scoring cri- teria were as follows: 1 point for each accurate result and 0 points for each inaccurate result [33]. Results Accuracy of visual tracking number trials (AVTNs) Pearson’s χ-squared test results of Experiment 1 showed that there was a significant difference in the total MOT accuracy between expert players (64.6%) and novice players (55.7%) (χ2 = 59.693, P = 0.000). When tracking 2–6 targets, the accuracy of expert players was higher than that of novice players, and the accuracy decreased with increasing tracking load. There was no significant difference in accuracy when tracking 2–3 targets (P > 0.05), but there was a signifi- cant difference in accuracy when tracking 4–6 targets (P < 0.05), as shown in Table 1. Accuracy of SDM trials (ASDMs) In Experiment 2, the Pearson χ-squared test results showed that the overall SDM accuracy of expert players (91.6%) and novice players (84.5%) was significantly different (χ2 = 31.975, P = 0.000). The overall decision-making speed of the expert players was faster than that of the novice players (0.74 times per second vs. 0.52 times per second, respectively), and the accuracy of the expert players in the three techniques of passing, dribbling and shooting was higher than that of the novice players. There was no significant difference in the accuracy of dribble decision-making between expert players and novice players (P > 0.05). The study showed that the number of times dribble decision-making occurred was greatly different across the three techniques, and the expert players had more than twice as many decisions resulting in drib- bling as the novice group (658 times vs. 278 times, respectively). There were also significant differences between passing decision-making and shooting decision-making accuracy (P < 0.01), as shown in Table 2. Table 1. Comparison of AVTNs between the expert group and novice group. Accurate (%) Expert group 672 (93.3) 576 (80.0) 432 (60.0) 405 (56.3) 240 (33.3) 2325 (64.6) Novice group 654 (90.8) 546 (75.8) 384 (53.3) 324 (45.0) 96 (13.3) 2004 (55.7) χ2 P value 3.086 3.632 6.516 18.228 80.497 59.693 0.079 0.057 0.011 0.000 0.000 0.000** Tracking load 2 targets 3 targets 4 targets 5 targets 6 targets Total Notes. * P < 0.05. P < 0.01. https://doi.org/10.1371/journal.pone.0283965.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283965 April 5, 2023 6 / 12 PLOS ONETable 2. Comparison of ASDMs between the expert group and novice group. The correlation between MOT ability and SDM Decision criteria Passing Dribbling Shooting Total Notes. P < 0.01. Accurate (%) Expert group 457 (96.4) 624 (94.8) 339 (80.9) 1420 (91.6) Novice group 399 (89.3) 263 (94.6) 256 (70.7) 918 (84.5) χ2 P value 17.938 0.021 11.112 31.975 0.000 0.886 0.001 0.000 https://doi.org/10.1371/journal.pone.0283965.t002 Pearson correlation between MOT ability and SDM performance Pearson’s correlation coefficient between the accuracy score of visual tracking number trials (SVTNs) and the accuracy score of SDM (SSDMs) was calculated [2]. The size of each correla- tion coefficient used the following evaluation criteria: < 0.1 = trivial; 0.1–0.3 = small; 0.3– 0.5 = moderate; and 0.5–0.7 = large [39]. The analysis showed that when tracking 2 targets, there was a medium positive correlation between the SVTN and the shooting SSDM of expert players (r = 0.403, P = 0.051), and there was no correlation between the SVTN and the shoot- ing SSDM of novice players (r = -0.018, P = 0.935). When tracking 3 targets, there was a medium negative correlation between the SVTN and expert players’ dribbling SSDM (r = -0.307, P = 0.144), and there was no correlation between the SVTN and novice players’ drib- bling SSDM (r = 0.043, P = 0.843). When tracking 4–5 targets, the SVTN was positively corre- lated with the passing SSDM and dribbling SSDM of expert players, and the SVTN was positively correlated with the passing SSDM of novice players, which was statistically signifi- cant (r > 0.6, P < 0.01), indicating that the tracking ability of a large number of targets is sig- nificantly correlated with accurate passing decisions. When tracking 6 targets, for expert players, there was a small negative correlation between the SVTN and dribbling SSDM and between the SVTN and shooting SSDM (0.1 < r <0.3, P > 0.05), and for novice players, there was a small negative correlation between the SVTN and passing SSDM, between the SVTN and dribbling SSDM and between the SVTN and shooting SSDM (0.1 < r < 0.3, P > 0.05). This shows that tracking with more than 6 targets has an interference effect on SDM, as shown in Table 3. Table 3. Correlation between MOT ability and SDM performance among basketball players. Target Number Pearson correlation Expert group Novice group Passing Dribbling Shooting Passing Dribbling Shooting -0.235 0.270 -0.140 0.513 0.711 0.000 0.667 0.000 0.005 0.982 -0.197 0.357 -0.307 0.144 0.704 0.000 0.666 0.000 -0.191 0.370 0.403 0.051 -0.206 0.334 0.089 0.679 -0.056 0.794 -0.231 0.277 0.145 0.499 0.100 0.642 0.785 0.000 0.724 0.000 -0.199 0.350 0.131 0.541 0.043 0.843 0.215 0.313 0.104 0.628 -0.203 0.341 -0.018 0.935 0.088 0.681 -0.008 0.970 -0.016 0.941 -0.193 0.367 r value P value r value P value r value P value r value P value r value P value 2 targets 3 targets 4 targets 5 targets 6 targets Notes. P < 0.01. https://doi.org/10.1371/journal.pone.0283965.t003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283965 April 5, 2023 7 / 12 PLOS ONEThe correlation between MOT ability and SDM Discussion The purpose of this study was to investigate the differences between visual attention and SDM among female expert and novice basketball players and determine the correlation between visual attention and SDM. The results verified our hypothesis. Compared with novice players, expert players had advantages in visual attention and SDM, and visual attention and SDM were related. In the dynamic movement of team ball games, the ability to "read the game" dis- tinguishes skilled players from unskilled players [40]. Expectations and decisions represent the ability of the human brain to extract meaningful contextual information from visual scenes and are crucial for high-level sports performance [33]. Eye movement control depends on two interactive yet distinct processes: attentional decision-making to allocate value to information sources and SDM to flexibly link selected information with actions [41]. Physical activity is generally believed to enhance brain plasticity and improve cognitive and executive functions. Experts have been proven to be superior to subelite and novice players in sports-specific tasks, including the utilization of advanced visual cues [42], visual search strategies [10], anticipation and decision-making skills [5]. Voss and his colleagues showed that professional knowledge of sports is related to high-level performance in processing speed and visual attention [28]. The superior perception of more highly skilled players may be attributed to their skills or abilities, rather than their greater experience or exposure to tasks [40]. Our research results showed that when tracking 2–6 targets, the tracking accuracy of expert players was significantly higher than that of novice players (64.6% vs. 55.7%, respectively). With an increasing number of visual tracking objects, the accuracy of tracking objects decreased among all participants. This high- lights that the capacity of attention resources is limited; as the number of targets increases, the resources allocated to each target decrease, which leads to a decline in tracking accuracy [18]. When tracking 4–6 targets, the accuracy of the expert players was significantly higher than that of the novice players, and the expert players had an MOT advantage. This finding supported the research results of Qiu and Li [11,24,25] but is different from the research results of Jin and Ji [2,20]. The visual search strategy involved in the MOT process may be closely related to the strategy of motion experts in extracting visual information from actions [33]. In basketball, players monitor the movements and positions of other players (teammates and opponents) to quickly make good passes, which is similar to tracking multiple objects in MOT [2]. When players have to navigate 1 vs. 1 and 2 vs. 2 situations, they feel confident and need to "read" the defense before deciding what to do [1]. The expert players exhibited faster and higher accuracy in SDM when compared to novice players (0.74 times per second vs. 0.52 times per second, 91.6% vs. 84.5%, respectively). The study showed that the number of times dribble decision- making occurred had the greatest difference across the three techniques, and the expert players had more than twice as many as the novice players (658 times vs. 278 times, respectively). This outcome may be due to the low dribbling skills of novice players. On-court performance is the result of the combination of SDM and technical skill level. SDM is the premise, technical level is the basis, and on-court performance is the result [5]. Our study found that when tracking 4–5 targets, the SVTN was positively correlated with passing SSDM, and the correlation was statistically significant. When tracking 4–5 targets, the SVTN of the expert players and novice players was positively correlated with the passing SSDM, and the SVTN of the expert players was larger and more significant. These findings reflect the visual attention advantage of expert players. When tracking 6 targets, the SVTN was negatively correlated with SSDM, which indicated that having to attend to too many tracking targets (more than 6) interfered with players’ decisions. This finding is different from some research results [29,30]. Good visual attention is a key component of players’ accurate SDM, and it is more obvious in high-level competitions. In the 3 vs. 3 competition, the number of PLOS ONE \| https://doi.org/10.1371/journal.pone.0283965 April 5, 2023 8 / 12 PLOS ONEThe correlation between MOT ability and SDM perceptual information sources is not so high, and the role of peripheral vision is more empha- sized to distribute attention more evenly according to specific situations [40]. This tells us that attention training should be applied to physical education teaching and training. Compared with dribbling and shooting, 3D MOT training can significantly improve players’ passing deci- sions [33]. Ten sessions of 3D MOT training improved attention, visual information process- ing speed and working memory. In addition, 3D MOT training can be transferred to social- related tasks [43]. Neurological studies have demonstrated the role of 3D MOT training in enhancing the cognitive function of healthy young people [44]. MOT technology reported the activation of brain regions involved in the attention process [45]. These regions include the parietal and frontal lobe regions of the cortex, which are considered to be the regions responsi- ble for attention shifts and eye movement [46]. When reading a situation during a competi- tion, players activate the movement perception process in the dorsal and ventral areas [47]. Attention tracking of multiple elements during 3D MOT training may overlap with the brain network required in the decision-making process. Future imaging or EEG research will help us explain this neural process [33]. This study provided an effective reference for the transfer of SDM from the laboratory to the live situation and solves the correlation between visual attention and SDM. The limitation of the study was that the subjects were only categorized into either an expert group or a novice group. The addition of a group of subjects with medium skills would enhance the study. Future research can explore the mechanism by which visual attention training influences SDM by combining event-related potentials (ERPs), eye movement instruments and fMRI. Conclusions Our hypotheses are that female expert basketball players are superior to novice players in MOT ability and SDM and that MOT ability is related to SDM ability. Our results support these hypotheses. (1) The tracking accuracy of expert players was significantly higher than that of novice players, especially when tracking 4–6 targets. As the number of targets increased, accuracy decreased. (2) The accuracy of expert players’ SDM was significantly higher than that of novice players, especially in passing decision-making and shooting decision-making. Expert players exhibited fast and accurate SDM. (3) There was a correlation between MOT ability and SDM ability. The MOT ability of 4–5 targets was positively correlated with passing decision- making, which was statistically significant. The correlation between MOT ability and SDM of expert players was greater and more significant than that of novice players. Having to attend to too many tracking targets (more than 6) interfered with players’ decisions. Supporting information S1 Data. (XLSX) S1 Appendix. Decision coding tool table. (DOCX) Acknowledgments Everyone who contributed significantly to this study has been listed. Author Contributions Conceptualization: Qifeng Gou, Sunnan Li. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283965 April 5, 2023 9 / 12 PLOS ONEThe correlation between MOT ability and SDM Formal analysis: Sunnan Li. Funding acquisition: Sunnan Li. Methodology: Qifeng Gou, Sunnan Li. Project administration: Sunnan Li. Software: Qifeng Gou. Supervision: Sunnan Li. Writing – original draft: Qifeng Gou, Sunnan Li. Writing – review & editing: Qifeng Gou, Sunnan Li. References 1. Sa´ nchez ACJ, Calvo AL, Buñuel PSL, Godoy SJI. Decision-making of spanish female basketball team players while they are competing. Rev Psicol Deporte. 2009; 18(3): 369–373. https://dialnet.unirioja.es/ servlet/articulo?codigo=6140406 ISSN:1132-239X. 2. Jin P, Li XW, Ma B, Guo HB, Zhang ZX, Mao LJ. Dynamic visual attention characteristics and their rela- tionship to match performance in skilled basketball players. PeerJ. 2020; 8(13): e9803. https://doi.org/ 10.7717/peerj.9803 PMID: 32879809. 3. Roca A, Williams AM. Does decision making transfer across similar and dissimilar sports? Psychol Sport Exerc. 2017; 31: 40–43. https://doi.org/10.1016/j.psychsport.2017.04.004 ISSN:1469-0292. 4. Lex H, Essig K, Knoblauch A, Schack T. Cognitive representations and cognitive processing of team- specific tactics in soccer. PLoS One. 2015; 10(2): e0118219. https://doi.org/10.1371/journal.pone. 0118219 PMID: 25714486. 5. Ren J. Motor control and learning. Beijing: Higher Education Press; 2019. 6. Arau´jo D, Davids K, Hristovski R. The ecological dynamics of decision making in sport. Psychol Sport Exerc. 2006; 7(6): 653–676. https://doi.org/10.1016/j.psychsport.2006.07.002 ISSN:1469-0292. 7. Warren W, Kay B, Zosh W, Duchon A, Sahuc S. Optic flow is used to control human walking. Nat Neu- rosci. 2001; 4(2): 213–216. https://doi.org/10.1038/84054 PMID: 11175884. 8. Fajen BR, Warren WH. Behavioral dynamics of steering, obstacle avoidance, and route selection. J Exp Psychol Hum Percept Perform. 2003; 29(2): 343–362. https://doi.org/10.1037/0096-1523.29.2. 343 PMID: 12760620. 9. Wang B. Research on motor intuition. Wuhan: Central China Normal University Press; 2006. 10. Jin P. The visual attention characteristics of basketball players and its relationship to match perfor- mance. (Education PhD Thesis), Shanghai University of Sport; 2020. 11. Qiu FH. Investigation of brain function and structural features in elite basketball players’ during multiple object tracking. (Education PhD Thesis), Shanghai University of Sport; 2019. 12. Astle DE, Scerif G. Interactions between attention and visual short-term memory (VSTM): What can be learnt from individual and developmental differences? Neuropsychologia. 2011; 49(6): 1435–1445. https://doi.org/10.1016/j.neuropsychologia.2010.12.001 PMID: 21185321. 13. Robinson-Riegler B. Cognitive Psychology (Ling CX, Trans). Beijing: Posts and Telecom Press; 2020. 14. Alves H, Voss MW, Boot WR, Deslandes A, Cossich V, Salles JI, et al. Perceptual-cognitive expertise in elite volleyball players. Front Psychol. 2013; 4(36): 1–9. https://doi.org/10.3389/fpsyg.2013.00036 PMID: 23471100. 15. Faubert J, Sidebottom L. Perceptual-cognitive training of athletes. J Clin Sport Psychol. 2012; 6(1): 85– 102. https://doi.org/10.1123/jcsp.6.1.85 ISSN: 1932-9261. 16. Pylyshyn ZW, Storm RW. Tracking multiple independent targets: evidence for a parallel tracking mech- anism. Spat Vis. 1988; 3(3): 179–197. https://pubmed.ncbi.nlm.nih.gov/3153671/ https://doi.org/10. 1163/156856888x00122 17. Faubert J. Professional athletes have extraordinary skills for rapidly learning complex and neutral dynamic visual scenes. Sci Rep. 2013; 3(1): 1154–1157. https://doi.org/10.1038/srep01154 PMID: 23378899. 18. Alvarez GA, Franconeri SL. How many objects can you track? Evidence for a resource-limited attentive tracking mechanism. J Vis. 2007; 7(13): 1–10. https://doi.org/10.1167/7.13.14 PMID: 17997642. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283965 April 5, 2023 10 / 12 PLOS ONEThe correlation between MOT ability and SDM 19. Meyerhoff HS, Papenmeier F, Jahn G, Huff M. Not Flexible enough: Exploring the temporal dynamics of attentional reallocations with the multiple object tracking paradigm. J Exp Psychol Hum Percept Per- form. 2016; 42(6): 776–787. https://doi.org/10.1037/xhp0000187 PMID: 26689311. 20. Ji CX, Liu JY. Non target inhibition mechanism in multiple object tracking task of badminton player. Jour- nal of Tianjin University of Sport. 2015; 30(2): 152–156. ISSN: 1005-0000. 21. Mangine GT, Hoffman JR, Wells AJ, Gonzalez AM, Rogowski JP, Townsend JR, et al. Visual tracking speed is related to basketball-specific measures of performance in NBA players. J Strength Cond Res. 2014; 28(9): 2406–2414. https://doi.org/10.1519/JSC.0000000000000550 PMID: 24875429. 22. Bahdur K. The effect of a physical, psychological and visual intervention on decision-making in football players. (Sport Science PhD Thesis), University of Johannesburg (South Africa); 2015. 23. Martı´n A, Sfer AM, D’Urso Villar MA, Barraza JF. Position affects performance in multiple-object track- ing in rugby union players. Front Psychol. 2017; 8: 1–8. https://doi.org/10.3389/fpsyg.2017.01494 PMID: 28951725. 24. Li J. Influence of the number of objects on basketball players’ performance in visual tracking task. Jour- nal of Tianjin University of Sport. 2012; 27(2): 133–137. ISSN: 1005-0000. 25. Qiu FH, Pi YL, Liu K, Li XP, Zhang J, Wu Y. Influence of sports expertise level on attention in multiple object tracking. PeerJ. 2018; 6: e5732. https://doi.org/10.7717/peerj.5732 PMID: 30280051. 26. Van Maarseveen MJJ, Savelsbergh GJP, Oudejans RRD. In situ examination of decision-making skills and gaze behaviour of basketball players. Hum Mov Sci. 2018; 57: 205–216. https://doi.org/10.1016/j. humov.2017.12.006 PMID: 29253742. 27. Ryu D, Abernethy B, Mann DL, Poolton JM, Gorman AD. The role of central and peripheral vision in expert decision making. Perception. 2013; 42(6): 591–607. https://doi.org/10.1068/p7487 PMID: 24422243. 28. Voss MW, Kramer AF, Basak C, Prakash RS, Roberts B. Are expert athletes ‘expert’ in the cognitive laboratory? A meta-analytic review of cognition and sport expertise. Appl Cognitive Psych. 2010; 24(6): 812–826. https://doi.org/10.1002/acp.1588 ISSN: 0888-4080. 29. Memmert D, Simons DJ, Grimme T. The relationship between visual attention and expertise in sports. Psychol Sport Exerc. 2009; 10(1): 146–151. https://doi.org/10.1016/j.psychsport.2008.06.002 ISSN: 1469-0292. 30. Harenberg S, McCarver Z, Worley J, Murr D, Vosloo J, Kakar R, et al. The effectiveness of 3D multiple object tracking training on decision-making in soccer. Sci Med Footb. 2022; 6(3): 355–362. https://doi. org/10.1080/24733938.2021.1965201 PMID: 35862166. 31. Komarudin K, Mulyana M, Berliana B, Purnamasari I. Neuro tracker three-dimensional multiple object tracking (3D-MOT): A tool to improve concentration and game performance among basketball athletes. Ann Appl Sport Sci. 2021; 9(1): e946. https://doi.org/10.29252/aassjournal.946 ISSN: 2322-4479. 32. Zhang XM, Liao YG, Ge CL. Performances of general college students and athletes in multiple-object tracking task. Journal of Beijing Sport University. 2008; 31(4): 504–506. ISSN: 1007-3612. 33. Romeas T, Guldner A, Faubert J. 3D-multiple object tracking training task improves passing decision- making accuracy in soccer players. Psychol Sport Exerc. 2016; 22: 1–9. https://doi.org/10.1016/j. psychsport.2015.06.002 ISSN: 1469-0292. 34. Romeas T, Chaumillon R, Labbe´ D, Faubert J. Combining 3D-MOT with sport decision-making for per- ceptual-cognitive training in virtual reality. Percept Mot Skills. 2019; 126(5): 922–948. https://doi.org/ 10.1177/0031512519860286 PMID: 31272277. 35. Cohen J. Statistical power analysis. Curr Dir Psychol Sci. 1992; 1(3): 98–101. https://doi.org/10.1111/ 1467-8721.ep10768783 ISSN: 0963-7214. 36. Lakens D. Performing high-powered studies efficiently with sequential analyses. Eur J Soc Psychol. 2014; 44(7): 701–710. https://doi.org/10.1002/ejsp.2023 ISSN: 0046-2772. 37. Sungur H, Boduroglu A. Action video game players form more detailed representation of objects. Acta Psychol. 2012; 139(2): 327–334. https://doi.org/10.1016/j.actpsy.2011.12.002 PMID: 22266223. 38. Gabbett TJ, Carius J, Mulvey M. Does improved decision-making ability reduce the physiological demands of game-based activities in field sport athletes? J Strength Cond Res. 2008; 22(6): 2027– 2035. https://doi.org/10.1519/JSC.0b013e3181887f34 PMID: 18978606. 39. Hopkins WG, Marshall SW, Batterham AM, Hanin J. Progressive statistics for studies in sports medicine and exercise science. Med Sci Sports Exerc. 2009; 41(1): 3–13. https://doi.org/10.1249/MSS. 0b013e31818cb278 PMID: 19092709. 40. Williams AM. Perceptual skill in soccer: Implications for talent identification and development. J Sports Sci. 2000; 18(9): 737–750. https://doi.org/10.1080/02640410050120113 PMID: 11043899. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283965 April 5, 2023 11 / 12 PLOS ONEThe correlation between MOT ability and SDM 41. Gottlieb J, Balan P. Attention as a decision in information space. Trends Cogn Sci. 2010; 14(6): 240– 248. https://doi.org/10.1016/j.tics.2010.03.001 PMID: 20399701. 42. Ward P, Williams AM, Bennett SJ. Visual search and biological motion perception in tennis. Res Q Exerc Sport. 2002; 73(1): 107–112. https://doi.org/10.1080/02701367.2002.10608997 PMID: 11926480. 43. Legault I, Faubert J. Perceptual-cognitive training improves biological motion perception: Evidence for transferability of training in healthy aging. Neuroreport. 2012; 23(8): 469–473. https://doi.org/10.1097/ WNR.0b013e328353e48a PMID: 22495038. 44. Parsons B, Magill T, Boucher A, Zhang M, Zogbo K, Be´ rube´ S, et al. Enhancing cognitive function using perceptual-cognitive training. Clin EEG Neurosci. 2014; 47(1): 37–47. https://doi.org/10.1177/ 1550059414563746 PMID: 25550444. 45. Howe PD, Horowitz TS, Morocz IA, Wolfe J, Livingstone MS. Using fMRI to distinguish components of the multiple object tracking task. J Vis. 2009; 9(4): 1–11. https://doi.org/10.1167/9.4.10 PMID: 19757919. 46. Culham JC, Brandt SA, Cavanagh P, Kanwisher NG, Dale AM, Tootell RBH. Cortical fMRI activation produced by attentive tracking of moving targets. J Neurophysiol. 1998; 80(5): 2657–2670. https://doi. org/10.1152/jn.1998.80.5.2657 PMID: 9819271. 47. Goodale MA, Milner AD. Separate visual pathways for perception and action. Trends Neurosci. 1992; 15(1): 20–25. https://doi.org/10.1016/0166-2236(92)90344-8 PMID: 1374953. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283965 April 5, 2023 12 / 12 PLOS ONE
10.1371_journal.pone.0263296	RESEARCH ARTICLE Lateral access mechanism of LPA receptor probed by molecular dynamics simulation Rieko Suenaga1, Mizuki Takemoto1¤, Asuka Inoue2, Ryuichiro IshitaniID Osamu Nureki1* 1, 1 Department of Biophysics and Biochemistry, Graduate School of Science, The University of Tokyo, Bunkyo-ku, Tokyo, Japan, 2 Graduate School of Pharmaceutical Sciences, Tohoku University, Aramaki, Aoba-ku, Sendai, Miyagi, Japan ¤ Current address: Preferred Networks, Inc., Otemachi, Chiyoda-ku, Tokyo, Japan ishitani@bs.s.u-tokyo.ac.jp (RI); nureki@bs.s.u-tokyo.ac.jp (ON) Abstract G-protein-coupled receptors (GPCR) are a family of membrane receptors that play impor- tant roles in the regulation of various physiological phenomena. LPA receptors (LPA1-6) are members of the class A GPCRs, which transduce a lysophosphatidic acid (LPA) signal across the cell membrane and evoke various responses, including cellular survival, prolifer- ation, differentiation, and migration. The crystal structure of LPA6 revealed a gap between its transmembrane helices (TMs), which is opened toward the membrane side. This led to the proposal of the “lateral access model,” in which its lipophilic ligand directly enters the binding pocket through the gap structure at the membrane. In this study, we performed molecular dynamics (MD) simulations and Markov state model (MSM) analyses of LPA6 and LPA, to elucidate the long timescale dynamics of the ligand binding process. The results from the 71.4-μs MD simulation suggested that the flexibility of the TMs constituting the gap structure enables the lateral entrance of the ligand, and the key interactions between the receptor and ligand facilitate the transition state of the ligand binding process. Introduction G-protein-coupled receptors (GPCRs) are a family of membrane receptors with a conserved motif of seven transmembrane alpha helices, which transmit extracellular signals into the cell. Triggered by the binding of extracellular ligands, GPCRs activate intracellular heterotrimeric G proteins, thereby evoking downstream signaling cascades. The ligands of GPCRs include amines, peptides, nucleic acids, and lipids, which play essential roles in the regulation of vari- ous physiological phenomena. In addition to their importance, their properties as receptors on cellular membranes make them promising drug targets. Lysophosphatidic acid (LPA) is a lipid mediator that induces various cellular responses, including survival, proliferation, differentiation, and migration [1]. The LPA receptors form a group of class A GPCRs, including LPA1-6. Among these receptors, LPA6 is reportedly impor- tant for hair formation [2, 3] and cancer progression [4], and thus regarded as a potential tar- get for anti-cancer agents. Recently, our group reported the 3.2 Å crystal structure of LPA6 [5] a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Suenaga R, Takemoto M, Inoue A, Ishitani R, Nureki O (2022) Lateral access mechanism of LPA receptor probed by molecular dynamics simulation. PLoS ONE 17(2): e0263296. https://doi.org/10.1371/journal.pone.0263296 Editor: Claudio M. Soares, Universidade Nova de Lisboa Instituto de Tecnologia Quimica e Biologica, PORTUGAL Received: July 9, 2021 Accepted: January 17, 2022 Published: February 3, 2022 Copyright: © 2022 Suenaga et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting information files. The whole trajectory files generated in the MD simulation have been deposited in the Zenodo server under digital object identifier: 10.5281/ zenodo.5533040. Funding: R.I. Grant-in-Aid for Scientific Research (B) (25291011), Japan Society for the Promotion of Science, No O.N. Grant-in-Aid for Specially Promoted Research (16H06294), Japan Society for the Promotion of Science, No A.I. PRIME grant PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 1 / 20 PLOS ONE(JP18gm5910013), Japan Agency for Medical Research and Development (AMED), No A.I. LEAP grant (JP18gm0010004), Japan Agency for Medical Research and Development (AMED), No A. I. KAKENHI grant (17K08264), Japan Society for the Promotion of Science, No. Competing interests: The authors have declared that no competing interests exist. Molecular dynamics simulation of LPA receptor (Fig 1a), which revealed a “gap” open toward the lipid bilayer between transmembrane helix (TM) 4 and TM5. A monoolein molecule was observed in this gap, and thus we proposed that the monoolein molecule binds to the gap by mimicking the acyl chain of LPA. In addition, based on the arrangements of the transmembrane helices and important acidic residues, we suggested that this crystal structure represents a pre-activation state, in which the receptor is bound to LPA, but has not adopted the active conformation. The lipophilic LPA may laterally access the ligand binding pocket from the side of the receptor, by passing through this gap. We also proposed some models of the LPA-LPA6 complex in the pre-activation state, based on the docking simulation [5]. Similar gap structures have also been observed in other lipid-accepting GPCRs, including sphingosine 1-phosphate receptor 1 (S1P1) [6], cannabinoid receptor 1 (CB1) [7, 8], thromboxane A2 receptor (TP) [9], prostaglandin E receptor 4 (EP4) [10], prosta- glandin D2 receptor (DP2) [11], human GPR40 receptor (hGPR40), also known as free fatty- acid receptor 1 (FFAR1) [12, 13], and platelet-activating-factor receptor (PAFR) [14]. There- fore, the lateral access through a gap structure might be a common mechanism among GPCRs for lipophilic ligands. Although it is difficult to experimentally verify the lateral access mechanism of lipophilic ligands from the viewpoint of structural biology, molecular dynamics (MD) simulations can provide important insights into the ligand binding process. In this study, we performed a series of all-atom molecular dynamics simulations of the ligand binding process of LPA6. Since the timescale of the ligand binding process may exceed those accessible by the conventional MD simulation method, we utilized the Markov state model (MSM) analysis [15] to capture the long timescale and rare events. MSM describes molecular kinetics as memoryless probabilistic transitions between a set of conformational states, and has been successfully applied to solve several problems, including protein folding [16], the activation mechanism of the β2-adrener- gic receptor [17], and the ligand binding process of a soluble protein [18]. Our present results, reconstructed from the 71.4 μs MD simulations, successfully visualized the LPA binding pro- cess. Furthermore, the results provided detailed insights into the lateral access mechanism of the lipophilic ligand, including the transition state formation and the mechanism of ligand entrance into the binding pocket of LPA6. Results Markov state model of the ligand binding process To reconstruct the long timescale dynamics of the ligand binding pathway, we tried to build a Markov State Model (MSM) from multiple short MD simulations [15]. MSM construction requires a sufficient number of short MD simulations, starting from initial states and covering from the ligand-unbound state to the ligand-bound state. In our previous study, we obtained three LPA6 ligand binding models by a docking simulation [5] (Fig 1b). Firstly, we analyzed three docking models (Models 1, 2, and 3). In Models 1 and 2, the LPA head group is deeply accomodated in the binding pocket and interacts with residues on TM1, 2, and 7. In contrast, in Model 3, the head group only interacts with the residues on TM5 and 6, which are exposed on the extracellular side. As a result, the interactions between the LPA head group and the basic residues of the receptor are different. In Model 1, the head group is bound to three resi- dues, K261.31, R832.60 and K1885.32. In Model 2, the head group contacts K261.31, K1885.32, R2676.62 and R2817.32, and in Model 3, the head group is only bound to K1885.32 and R2676.62. While our previous functional analysis showed that K261.31, R832.60 and R2817.32 are important for LPA6 activation [5], the interactions with these residues are absent in Model 3. Therefore, we used Models 1 and 2 as the initial structures of the simulation. PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 2 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor Fig 1. Structure of LPA6 and LPA6-LPA docking model. a) Overall crystal structure of zebrafish LPA6 (Protein Data Bank (PDB) ID: 5XSZ), viewed from the membrane plane. In the left panel, the whole structure is shown as a ribbon model, and the disordered region of ECL2 is illustrated as a dashed line. In the right panel, the whole structure is shown as a surface model from the same viewpoint as the left panel. The gap structure between TM4 and TM5 is indicated with the blue rectangle. b) The LPA6-LPA docking models (Models 1, 2, and 3) viewed from the extracellular side (left), and the membrane plane (right). The protein is shown as a ribbon model and the LPA (2-LPA(18:2)) molecule is depicted by a CPK model. The conserved positively charged residues are shown as stick models. https://doi.org/10.1371/journal.pone.0263296.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 3 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor We next obtained the sequential trajectory from the ligand-bound state to the unbound state by a “steered MD” simulation, in which the ligand was slowly separated from the receptor with external force. We then conducted non-biased MD simulations starting from the 379 and 335 snapshot structures of the steered MD simulations of Models 1 and 2, respectively. Finally, we collected the 37.9 μs and 33.5 μs MD trajectories for Models 1 and 2, respectively. Although the trajectories from Model 1 and Model 2 were independently analyzed, we hereafter discuss the results from Model 1, unless otherwise noted, since the results from Models 1 and 2 share numerous similarities (S2 Fig). To build the MSM for these simulations, we analyzed the MD simulation trajectories, as shown in Fig 2. The RMSD values to the “reference structures” were calculated using the ligand non-hydrogen atoms and protein atoms interacting with the ligand, thereby forming the feature vectors. These feature vectors were then compressed using the principal component analysis (PCA), and were finally subjected to the clustering and MSM Fig 2. Schematic flow-chart of the procedure used in this work for MSM construction. https://doi.org/10.1371/journal.pone.0263296.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 4 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor analysis. The robustness of the analysis was verified by the bootstrap method (S1 Fig). Detailed descriptions of the simulation and MSM construction are provided in the S1 File. Definition of macrostates We classified the 995 microstates into seven “macrostates” by the PCCA+ method [19]. The microstate structure with the largest population in each microstate was visualized as its repre- sentative structure (Fig 3a). These macrostates were then divided into three groups (i.e., disso- ciated, partially-bound, and bound groups). The ligand-unbound dissociated group was classified into two macrostates, Macrostates 1 and 2 (Fig 3a). In Macrostate 2, the ligand molecule was adjacent to the receptor gap. Particu- larly, in its representative structure, the ligand head group was trapped by the positive residue K185 on extracellular loop 2 (ECL2) (Fig 3b). In the partially-bound group, which includes Macrostates 3, 4, and 5, the ligand was par- tially bound within the receptor gap, and a small number of interactions were formed. In Macrostate 3, the ligand head group was only bound to K185ECL2 and K1885.32, and did not interact with any functionally important basic residues [5]. In Macrostates 4 and 5, the ligand head group formed interactions with R2676.62, but the acyl chain was not fully accommodated within the receptor gap (Fig 3b). Therefore, this group may include transition states between the ligand-bound and -unbound states. In the bound group, which includes Macrostates 6 and 7, the ligand was fully accommo- dated by the receptor and formed interactions with the functionally important basic residues (Fig 3a). In Macrostate 7, the ligand formed the deepest interactions with the receptor, and the head group was bound to K1885.32, R2676.62 and R2817.32, whereas it was bound to only K1885.32 and R2676.62 in Macrostate 6 (Fig 3b). In the following sections, we will discuss these seven representative structures. Energy landscape and transition path analysis of the ligand binding process We calculated the microstate population distribution in the equilibrium state from the MSM, and then estimated the free energy landscape by projecting it onto the two-dimensional plane spanned by the PC axes obtained by the PCA of the 500-dimensional feature vectors (Fig 4). In the PCA, the cumulative variance ratio of the PCs was over 97% from PC1 to 3 (Fig 4a), show- ing that the ligand binding process can be explained by these three PCs. Thus, we hereafter focused on them. We first projected the microstate population distribution on the plane spanned by the PC1 and 2 axes, and plotted the cluster centers on it (Fig 4b). The plot revealed two low-energy basins at the low- and high-PC1 regions, which may correspond to stable states in the ligand binding process. The detailed analysis suggested that PC1 is simply related to the ligand-recep- tor distance, with high and low PC1 values corresponding to dissociated and bound states, respectively. In contrast, PC2 is correlated with the relaxation process after removing the bias- ing potential of steered MD, with high and low PC2 values corresponding to the biased and relaxed states, respectively (see S1 File). The low free-energy basins around the low PC2 regions suggested that the sampling of the relaxed conformations was sufficient to provide an unbiased view of the ligand binding dynamics. We next projected the probability distribution on the plane spanned by the PC1 and 3 axes, and plotted the cluster centers (Fig 5). In the PC1-3 plane, the basin around the “bound” group in the PC1-2 plane was separated into two basins, the low-PC3 and high-PC3 basins, which correspond to Macrostates 6 and 7, respectively (Fig 5a). The detailed analysis suggested PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 5 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor Fig 3. Macrostates and MSM groups calculated from the present simulation results. a) The macrostates observed in the present simulation. The size of the circle corresponds to the population of the microstate. The transitions between the macrostates are indicated by the gray arrows. The thickness of the arrows is proportional to the transition probability. The structures of the centers of the largest clusters are shown with cartoon and CPK models. b) Close-up views of representative structures of the macrostates. The interactions of the ligand-ligand recognition residues in each structure are shown as ball and stick models. https://doi.org/10.1371/journal.pone.0263296.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 6 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor Fig 4. Free energy landscape projected on the plane spanned by the PC1 and PC2 axes. a) The explained variance ratios of PC1 to PC5 (gray bar chart), and their cumulative values (gray line). b) Free energy landscape projected on the PC1-PC2 plane. In the left panel, the centers of the clusters and the initial structures are plotted as light and dark blue circles, respectively. The sizes of the light blue circles are proportional to the populations of the clusters. In the middle panel, the centers of the clusters are color-coded according to their macrostates. In the right panel, the center of each microstate is plotted on the free energy map. The groups of the macrostates defined in Fig 3 and the main text (i.e., dissociated, partially-bound, and bound groups) are indicated by magenta lines. c) The relationships between the PC1 values and the distances between the LPA head group and the conserved positive residues (K1885.32, R2676.62, R2817.32, K261.31, and R832.60). The distances are defined as those between the nearest oxygen atom of the head group and the nitrogen atom of the basic residues. d) The ligand conformation and the receptor structure with the highest PC2 value (PC2 = 3.57). The ligand and receptor are shown as CPK and cartoon models, respectively. The definition of the ligand vector and its z value are depicted as the dashed line and blue arrow, respectively. e) The relationship between the PC2 value and the z value of the ligand shown in (d). The difference of the distributions of the ligand z values in the higher and lower PC2 regions (red and blue for PC2 > 0 and < 2, respectively) is shown as a histogram in the right panel. https://doi.org/10.1371/journal.pone.0263296.g004 that the PC3 axis corresponds to the process for deeply accommodating the ligand head group within the basic pocket formed around R2817.32 (see S1 File). Reconstruction of the ligand binding/unbinding trajectory Using the MSM, we reconstructed a 100-μs trajectory starting from the ligand-unbound state (Fig 6, S1 Movie). First, we analyzed the transitions among the macrostates (Fig 6a), and then the LPA binding to the receptor, by measuring the distance between the phosphate group of PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 7 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor Fig 5. Free energy landscape projected on the plane spanned by the PC1 and PC3 axes. a) Free energy map projected on the PC1-PC3 plane. In the left panel, the centers of the clusters and the initial structures are plotted as light and dark blue circles, respectively. The sizes of the light blue circles are proportional to the populations of the clusters. In the middle panel, the centers of the clusters are color-coded according to their microstate. In the right panel, the positions of the central structures of the macrostates are indicated by circles. A basin in the free energy map with a small PC1 value corresponds to a dissociated state, that with an intermediate PC1 value represents a partially-bound state to the central region with high energy, and that with a large PC1 value corresponds to a bound state. Macrostate 6 corresponds to a basin with a low PC3 value, and 7 corresponds to a basin with a high PC3 value. b) Relationship between the PC3 value and the distance between the LPA head group and R2817.32. The definition of the distance is the same as in Fig 4. https://doi.org/10.1371/journal.pone.0263296.g005 LPA and the center of the receptor (Fig 6b). The results showed that the reconstructed trajec- tory includes both the ligand binding and dissociation events, which occur on the order of 10 μs. Our previous structural analysis of LPA6 suggested that the membrane-embedded LPA molecule laterally enters the binding pocket, through the gap formed between TM4 and TM5 (TM4-5 gap) [5]. To investigate the dynamics of the TM4-5 gap, we measured the distance between the Cα atoms of T161 and V195. In the ligand-unbound state, the width of the TM4-5 gap fluctuated from 4.4 Å to 15.8 Å (Fig 6c), revealing its flexibility. In the snapshot structure with the smallest gap width, the TM4-5 gap is completely closed, and the ligand binding site is occluded from the lipid membrane (Fig 6e). In contrast, in the ligand-bound state, the fluctua- tions of the width of the TM4-5 gap are smaller than those in the ligand-unbound state. The minimum width was 6.9 Å, which is slightly wider than those of the ligand-unbound state (Fig 6c). The gap width histogram also revealed the differences in the flexibility between the ligand- bound and -unbound states (Fig 6d). In the transition state between the ligand-bound and PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 8 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor Fig 6. Reconstituted 100-μs trajectory of the LPA binding process. a) Plot of the transitions among Macrostates 1–6 of the reconstituted trajectory. The color code corresponds to that used in Fig 3. b) Plot of the transition of the distance between the LPA head group and the center of mass of the receptor protein. Distances less than 25 Å are highlighted in light blue. c) The transition of the TM4-TM5 distance is plotted in light blue, and the moving average between the four frames is plotted in blue. d) The distribution of distances between TM4 and TM5 in the ligand-unbound (gray) and -bound (blue) states. e) The structures with the smallest (orange) and largest (green) TM4-TM5 distances in the ligand-unbound state are shown by cartoon models and cross- sectional views of the electron density surfaces. f) Plot of the distances between the LPA head group and the conserved positive residues (K185, K1885.32, R2676.62, R2817.32, K261.31, and R832.60) around 25 μs and 35 μs. Areas with distances of 2.5 Å or less are highlighted in light blue and light green backgrounds. https://doi.org/10.1371/journal.pone.0263296.g006 PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 9 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor -unbound states, in which the head group of LPA enters the pocket of the receptor, the width of the TM4-5 gap was about 12 Å. We next analyzed the interactions between the ligand and the binding pocket residues, focusing on the ligand binding process observed around 30 μs in the reconstituted trajectory (Fig 6f). In the trajectory, the ligand head group initially formed a weak interaction with K185 on the ECL2 loop, as observed in the representative structure of Macrostate 2 (Fig 3b). Simul- taneously, the head group also formed an interaction with K1885.32; however, its frequency was lower than that of K185. This stage (25–28.5 μs) mainly corresponds to the dissociated group (Macrostates 1 and 2) in the macrostate clusters. The head group then enters the TM4-5 gap, and forms interactions with K1885.32 and R2676.62. The interaction with R2676.62 is partic- ularly stable and maintained throughout the ligand-bound state, suggesting its importance for the first step in the ligand binding process. In contrast, the interaction between the acyl chain and the TM4-5 gap has not been stably formed yet. Next, at 1.5 μs after binding with R2676.62, the head group is deeply accommodated within the pocket and interacts with R2817.32 and K261.31. The interaction with R2817.32 induces a slight inward shift of the extracellular side of TM7. The acyl chain of LPA is also accommodated in the TM4-5 gap. This stage corresponds to the bound group (Macrostates 6 and 7) of the macrostate clusters. In contrast, we did not observe any interactions between the head group and R832.60, which were observed in the ini- tial docking pose of Model 1 (Fig 1b). Thus, the resulting bound state of the reconstituted tra- jectory is similar to Model 2, in which no interactions are formed between the head group and R832.60. The macrostates and their relationships (Fig 3), as well as the reconstructed trajectory (Fig 4), suggested that K185 on ECL2 is another key residue in the ligand binding process. K185 forms a salt bridge with the phosphate moiety of the ligand head group in Macrostates 2 and 3, before forming the transition state in Macrostate 5 (Fig 3b). Afterwards, it dissociates from the ligand in Macrostates 5, 6, and 7 (Fig 3b). Therefore, the present results suggest that the posi- tive charge of K185 anchors the ligand head group, thereby facilitating the formation of the transition state in the ligand binding process. To examine the importance of K185, we mea- sured the signal transduction activities of single mutants (K185A and K185E) and double mutants (K185A/K188A and K185E/K188E) of the zebrafish LPA6 receptor. However, none of these mutations had a large impact on the activity (Fig 7). The effect of anchoring by K185 may be less critical than that of the activation for signal transduction, which makes it difficult to confirm the contribution of K185 by the assay method used in this study. Further analyses are required to corroborate the role of K185 in the ligand anchoring. Discussion In this study, we successfully reproduced the pathway from the ligand-unbound state to the pre-activation state by MSM, based on the all-atom MD simulation. Our results support the lateral access model of the LPA ligand, and provide further detailed insights into the mecha- nism of ligand access to the LPA6 receptor. Particularly, the present simulation results suggest the importance of the flexibility of the TM4-5 gap, which forms the entrance of the receptor pocket. In the course of the simulation, the TM4-5 gap spontaneously opens and closes, allow- ing the ligand to enter the pocket of the receptor. The width distribution of the TM4-5 gap structure ranges from 8 to 12 Å. After ligand binding, the TM4-5 gap is stabilized in its open conformation by interactions with the ligand. Similar conformational changes in the gap have been reported for other lipid mediator GPCRs. For example, cysteinyl leukotriene receptor 1 (CysLT1R) also has a gap between TM4-5, which is similar to that of LPA6 [20]. The 1-μs MD simulation of CysLT1R demonstrated that the transition between the gap-open and gap-closed PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 10 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor Fig 7. Functional analysis of LPA mutants designed based on the present study. a) The receptor activities were examined using the alkaline- phosphatase tagged TGF-ɑ shedding assay. b) Activities of LPA and its mutants expressed as RAi (Emax/EC50 relative to WT). Data are mean ± s.e.m. (n = 3 or 4). �P < 0.05; ��P < 0.01; ��P < 0.001, one-way ANOVA with Dunnett’s post hoc test. NS, not significant. https://doi.org/10.1371/journal.pone.0263296.g007 conformations spontaneously occurred, suggesting the relatively low energy barrier between them [20]. Furthermore, the binding of the membrane lipid molecule (POPC) reportedly PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 11 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor stabilizes the gap in its open conformation [20]. The importance of the gap flexibility has also been mentioned for cannabinoid receptor 1 (CB1), although unlike LPA6, the gap is between TM1 and TM7, based on the comparison of the agonist- and antagonist-bound crystal struc- tures, as well as the MD simulation results [21]. The MD simulations of CB1 also suggested that the binding of POPC in the gap stabilizes it in the open conformation [21]. Thus, the flexi- bility of the TM gap may be an important feature of GPCRs with lipophilic ligands, across the differences in the gap-forming TMs. It is interesting to note that the crystal structure of LPA6 was determined without an agonist or antagonist, while all of the other GPCRs with the gap structure were determined in the agonist- or antagonist-bound states [5]. In the crystal struc- ture of LPA6, the monoolein molecule was observed in its TM4-5 gap, and may have stabilized the gap structure for crystallization. The present simulation provides novel insight into the ligand binding process. The transi- tion state in the binding process is particularly interesting, and can be analyzed from the energy landscape calculated in the present simulation. The transition state may correspond to the state between the unbound- and bound-states with the lowest energy barrier. The energy landscape projected onto PC axes 1–3 suggests that such a saddle point in the energy landscape may exist around the region with PC1 of −5 ~ 0, PC2 of 0 ~ 1, and PC3 of −1 ~ 0. Thus, we selected the three largest clusters from the region (clusters A, B, and C), and compared their representative structures (Fig 8a). Given that i) the direction of the PC1 increase corresponds Fig 8. Transition state of the ligand binding process. a) The structures contained in the three clusters corresponding to the transition states, clusters A, B and C, are plotted on the PC1-PC2 and PC1-PC3 planes, respectively. The central structure of Macrostate 6 is indicated by pink stars. b) The central structures of clusters A, B and C, and the central structure of Macrostate 6 are shown by cartoon and ball and stick models, respectively. https://doi.org/10.1371/journal.pone.0263296.g008 PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 12 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor to the ligand binding process, and ii) the PC1 value increases in the order of A, B, and C, we can assume that the conformational change toward the LPA binding occurs in this order. All of these clusters belong to Macrostate 5 (Fig 5a), which is one of the partially-bound groups. In clusters A, B, and C, the head group of LPA is accommodated in the binding pocket and forms interactions with K1885.32 and R2676.62 (Fig 8b). In contrast, the acyl chain tail of LPA lacks interactions with the receptor in cluster A, and instead partially interacts with the TM4-5 gap in clusters B and C (Fig 8b). The acyl chain tail interacts with the two aromatic residues Y107 and Y111 on TM3, and its terminus is still unstable due to the lack of interactions with the receptor (Fig 8b). Thus, after the head group binds, the acyl chain tail may first contact Y107 and Y111, thereby forming the binding transition state. After the terminus of the acyl chain tail is accommodated in the TM4-5 gap, the head group may be pushed deeply into the binding pocket to form the pre-activation state (Macrostate 6; Fig 8b). Taken together, the present sim- ulations suggest that Y107 and Y111, which form part of the TM4-5 gap, could play a key role in the transition state formation in the ligand binding process. In addition, these aromatic residues are well conserved not only in phylogenetically-related GPCRs, such as P2Y15 and P2Y126, but also in PAFR7 and CysLT1R2, possessing the TM4-5 gap. Accordingly, the ligand binding process and transition mechanism might be conserved in these GPCRs. The macrostates and their relationships (Fig 3), as well as the reconstructed trajectory (Fig 4), suggested that K185 on ECL2 is another key residue in the ligand binding process. K185 forms a salt bridge with the phosphate moiety of the ligand head group in Macrostates 2 and 3, before forming the transition state in Macrostate 5 (Fig 3b). Afterwards, it dissociates from the ligand in Macrostates 5, 6, and 7 (Fig 3b). Therefore, the present results suggest that the posi- tive charge of K185 anchors the ligand head group, thereby facilitating the formation of the transition state in the ligand binding process. To examine the importance of K185, we mea- sured the signal transduction activities of the single mutants (K185A and K185E), as well as the double mutants (K185A/K188A and K185E/K188E), of the zebrafish LPA6 receptor. How- ever, none of these mutations had a large impact on the activity (S1 Fig). The effect of anchor- ing by K185 may be less critical than that of the activation for signal transduction, which makes it difficult to confirm the contribution of K185 by the assay method used in this study. Further analyses are required to clarify the role of K185 in the ligand anchoring. In our previous study, the crystal structure of LPA6 and the structure-based functional analysis suggested that the functionally important basic residues (K261.31, R832.60, R2676.62, R2817.32) form the putative binding site for the ligand phosphate group. However, given the locations of these basic residues in the crystal structure and the size of the phosphate group, it is impossible for all of these residues to interact simultaneously with the ligand. Therefore, it has been hypothesized that some of these residues form the initial binding site for the ligand phosphate group in the pre-activation state, and then the large conformational changes of TM6 and 7 bring these residues near the ligand phosphate group to form the final binding site, thereby activating the downstream signaling pathway. This hypothetical model raises the ques- tion of which residues are involved in forming this initial binding site in the pre-activation state. In the present study, we performed two simulations based on Models 1 and 2, in which K261.31, R832.60 and K1885.32 (Model 1), and K261.31, K1885.32, R2676.62 and R2817.32 (Model 2) are involved in the initial binding site formation. Despite the different initial binding manners of LPA, both simulations resulted in the similar binding mode of the LPA head group to Model 2; i.e., R2676.62 and R2817.32, but not R832.60, are involved in the initial binding site (Fig 3; Macrostates 6 and 7). Therefore, we propose the updated model of the pre-activation state based on the present simulation (Fig 9), in which the initial binding site of the pre-activation state is formed by K1885.32, R2676.62, and R2817.32. PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 13 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor Fig 9. A schematic model of the LPA6 activation process proposed in this study. https://doi.org/10.1371/journal.pone.0263296.g009 In conclusion, we reconstructed the ligand-binding process using MSM, based on the all- atom MD simulation with the 71.4 μs total length. The results provided detailed insights into the lateral access mechanism of the lipophilic ligand, including the transition state formation. In addition, the results suggested the recognition manner of the phosphate moiety of the ligand head group, by the initial binding site in the pre-activation state. In this study, we only focused on the process from the ligand-unbound to pre-activation state. We also ignored the effects of lipids other than POPC (i.e., cholesterol, etc.), which could have some impact on the ligand-binding and activation processes. Further functional, structural and computational analyses of the active state of LPA6, using a more realistic lipid system, will be required to fur- ther clarify the complete activation mechanism of LPA6. Methods Molecular dynamics simulation tool and force field All MD simulations were performed using GROMACS ver. 5.0.7 [22] and the CHARMM36 [23] force field. PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 14 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor System setup For the initial structures of the simulation, three LPA6-LPA docking models (Models 1, 2 and 3) were used. All docking models were embedded within a 1-palmitoyl-2-oleoyl-sn-glycero- 3-phosphocholine (POPC) bilayer, using the MemProtMD pipeline [24]. According to the original article, the N- and C-termini were capped with N-acetyl and N-methyl amide groups, respectively. Each simulation system was a 100 Å3 cube, solvated with 150 mM NaCl and TIP3 water molecules. The systems were equilibrated first for 0.1 ns under NVT conditions, with 10 kJ mol-1 Å-2 restraints for all heavy atoms (all atoms except hydrogen) of both the protein and ligand. Finally, the systems were equilibrated for 5.0 ns under NPT conditions with the same restraints, followed by a 100 ns equilibration without any restraints. For all of the following simulations, the LINCS algorithm was used for atom bond calculations. The system tempera- ture was kept at 310 K by a Nose´-Hoover thermostat. Long-range electrostatic interactions were calculated by the particle mesh Ewald method. Steered MD simulation To sample the initial structures for building MSM, steered MD simulations were conducted using the three equilibrated systems described above. The simulations were conducted under 10 kJ mol-1 Å-2 restraints for all Cα atoms of the protein, and the protein and the ligand. A 10 kJ mol-1 Å-2 harmonic potential was applied between their centers of mass, pulling the ligand off at a constant velocity of 0.4 Å/ns. The temperature was kept at 310 K, and the pressure was 1 bar. On two models, 5 simulations were performed for 50 ns, and coordinates for all atoms were sampled every 10 ps, to obtain 15,000 structures (= 2 models x 50 ns x 5 runs / 10 ps) in total. Finally, 714 structures (= 379 from Model 1 + 335 from Model 2) were randomly sam- pled from the above structures. Molecular dynamics simulation All atoms of the sampled structures were assigned random velocities, and equilibrated for 500 ps under 10 kJ mol-1 Å-2 restraints for all Cα atoms of the protein and all atoms of the ligand. Using these initial structures, the main simulations were conducted for 100 ns without any restraints, and the coordinates for all atoms were sampled every 10 ps, to obtain 7,140,000 structures (= 714 initial structures×100 ns/10 ps) in total. The equilibration and the production run were both performed at 310K, 1 bar. Data analysis and visualization tools We used MDtraj ver.1.8.0 [25] to handle the coordinates or trajectories and MSMBuilder ver. 3.8 [26] for featurization, clustering, building the MSM and reconstructing the pathway. We used MSMExplorer ver.1.1.0 [27] for the free energy calculation in the principle components space. All molecular graphics were illustrated with CueMol2 ver. 2.2.3.443 (http://www. cuemol.org/) and all movies were generated with VMD ver. 1.9.3 [28]. Plot graphics were gen- erated with seaborn (https://seaborn.pydata.org/) and Matplotlib [29]. Markov state model construction All obtained structures were grouped into three datasets according to the original docking models. The MSM in each dataset was constructed by the following procedure. All structures were classified into 500 clusters with K-medoid clustering of the root-mean-square deviation (RMSD) values of all atoms, with the MiniBatchKMedoids module of MSMBuilder. The RMSD values were calculated for all atoms of the ligand and the 35 interacting residues: D23, PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 15 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor K26, R83, V103, F106, Y107, M110, Y111, L115, G157, S158, T161, G162, L165, F179, E180, F182, S183, S184, K185, E186, W187, K188, L191, V195, I198, V201, G202, L260, Y263, R267, C274, E277, R281 and Y284. The interactions include hydrogen bond and hydrophobic inter- actions, and the range of the interaction distance was 1.5 Å–5.0 Å. Using all atoms, including these 37 residues and the ligand molecule, the RMSD values to these 500 reference structures were calculated with the LandMarkRMSDFeaturizer of MSMBuilder. These feature values were compressed into 10 dimensions by the principal component analysis (PCA) implemented in MSMBuilder. Structures were classified into 1,000 clusters by K-means clustering using 10 principal components (PC1–PC10), by the MiniBatchKMeans module of MSMBuilder. In that process, we transformed the MD trajectories into the 1,000 cluster transition trajectories. The Markov state model (MSM) describes the dynamics of a system as a series of memory- less, probabilistic transitions between a set of states [15]. On the memoryless premise, the probability of the transition from state i to state j in a lag time of τ is described by a transition- probability matrix, T(τ) = {Tij(τ)}. For tuning the approximate lag time, the implied timescales, indicating how quickly the process reaches equilibrium, were calculated by the following for- mula: ti ¼ (cid:0) t ln li where ti indicates the ith slowest implied timescale, determined from the ith largest eigenvector of T(τ), λi. In theory, if the implied timescales ti converge as the lag time τ increases, then the model satisfies the Markov assumption [15]. In this study, the lag time was determined to be 10 ns after the tuning. Finally, the MSM was built based on the cluster transition trajectories by the MarkovStateModel module of MSMBuilder. Statistical analysis Kinetically related microstates were grouped into 5 metastable states (macrostates) using PCCA+ 20. Ligand-unbound states were grouped into one macrostate. In this macrostate, the microstate with the largest population was chosen. The center of mass structure of this micro- state was extracted using a k-d tree algorithm [30]. Using this structure as the starting state, we reconstructed a 100 μs (= lag time 10 ns x 10,000 steps) trajectory by the sample_msm module of MSMBuilder. Transition path theory (TPT) is a way to extract the highest-flux pathways of the system from an estimated MSM [31]. To calculate the TPT path, we determined the final states by extracting the center of mass (using the k-d tree algorithm) of the largest cluster in the third dump on the PC1–PC3 energy map. TPT paths were generated by the TPT modules of MSMBuilder. Ligand-induced TGFα shedding assay The ligand responses of the LPA6 mutants were determined by the TGFα shedding assay, which measures LPA6-mediated G12/13 signaling as described previously [5, 32]. Briefly, HEK293FT cells at a concentration of 2 x 105 cells/ml were seeded in 4 ml of 10% FCS- and penicillin/streptomycin-containing DMEM (complete DMEM) per 6-cm culture dish and placed in a CO2 incubator at 37ºC for 1 day. A modified pEGFPC1 vector (Clontech) encoding the wild-type or a mutant LPA6, consisting of a haemagglutinin signal sequence, FLAG epitope tag and zebrafish LPA6 (residues 1–312) [5], was transfected together with a pCAGGS plasmid encoding alkaline phosphatase (AP)-tagged TGFα (AP-TGFα; human codon-optimized) into HEK293FT cells (seeded 1-day before the transfection in 4 ml per 6-cm culture dish at a cell PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 16 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor concentration of 2 x 105 cells/ml) by using a polyethylenimine (PEI) transfection reagent (400 ng LPA6 plasmid, 1 μg AP-TGFα plasmid, and 8 μl of 1 mg/ml PEI solution per 6-cm culture dish). After one day of culture, the transfected cells were harvested by trypsinization, neutral- ized with complete DMEM, washed once with Hank’s Balanced Salt Solution (HBSS) contain- ing 5 mM HEPES (pH 7.4), and resuspended in 10 ml of HEPES-containing HBSS. The cell suspension was seeded at a volume of 80 μl (per well hereafter) in a 96-well culture plate (cell plate), in which 10 μl of 30 μM Ki16425 (Adooq), an antagonist for LPA1 and LPA3, was pre- dispensed. After a 30-min incubation in a CO2 incubator, a 10 μl aliquot of serially diluted linoleoyl LPA (Echelon Biosciences; 10X, diluted in 0.01% bovine serum albumin (BSA) and 5 mM HEPES-containing HBSS) was added to the cells in triplicate and incubated for 1 h. The cell plate was then centrifuged, and 80 μl of conditioned media (CM) was transferred to an empty 96-well plate (CM plate). The AP reaction solution (80 μl, containing 10 mM p-nitro- phenylphosphate (p-NPP), 120 mM Tris–HCl (pH 9.5), 40 mM NaCl, and 10 mM MgCl2) was dispensed into both the cell plates and the CM plates. The absorbance at 405 nm (A405) of the plates was measured with a microplate reader (SpectraMax 340 PC384, Molecular Devices), before and after a 1 h incubation at room temperature. Ligand-induced AP-TGFα release was calculated as described previously [32]. Unless otherwise noted, the background AP-TGFα release signals in the empty vector (mock)-transfected cells were subtracted from those in the LPA6-expressing cells. Using the Prism 8 software (GraphPad Prism), the LPA6-dependent AP-TGFα release signals were fitted to a four-parameter sigmoidal concentration-response curve, from which the EC50 and Emax values were obtained. The relative Emax/EC50 value, also known as the relative intrinsic activity (RAi), a dimensionless parameter [33], was logarithmi- cally transformed (Log RAi) and used to indicate receptor activity. Enzyme-induced TGFα shedding assay The responses of the LPA6 mutants to PA-PLA1α were determined by a modified TGFα shed- ding assay, as reported previously [5]. In brief, HEK293FT cells in the growth phase were sus- pended in Opti-MEM I Reduced Serum Medium (Thermo Fisher Scientific) at a cell concentration of 4 x 105 cells/ml, seeded in a 96-well plate (80 μl per well), and placed in a CO2 incubator. On the same day, the transfection solution (per well hereafter) was prepared by mixing 8 ng of the LPA6 plasmid, 20 ng of the AP-TGFα plasmid, and a titrated volume of the human PA-PLA1α pCAGGS plasmid (wild-type or the catalytically inactive S154A mutant), with 0.2 μl of 1 mg/ml PEI and 20 μl Opti-MEM I Reduced Serum Medium. The empty pCAGGS plasmid was used to balance the equal volumes of transfected plasmid. After adding the transfection solution (20 μl), the cells were cultured for 1 day. The 96-well plate was centri- fuged at 190g for 2 min, and the supernatant (80 μl) was transferred to an empty 96-well plate. The AP activities in the cell plate and the CM plate were measured and calculated as described above, except the measurement interval was 15 min. The background AP-TGFα release signal in the absence of the PA-PLA1α plasmid was subtracted from those in the PA-PLA1α-express- ing cells. Flow cytometry analysis HEK293FT cells were seeded in a 12-well culture plate (1 ml per well) and transfected with plasmids (100 ng LPA6 plasmid and 250 ng AP-TGFα plasmid), as described above. One day after transfection, the cells were harvested with 0.53 mM EDTA-containing Dulbecco’s PBS (D-PBS). The cell suspension was transferred to a 96-well V-bottom plate and fluorescently labeled with an anti-FLAG epitope (DYKDDDDK) tag monoclonal antibody (Clone 1E6, Fuji- Film Wako Pure Chemicals; 10 μg/ml diluted in 2% goat serum- and 2 mM EDTA-containing PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 17 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor D-PBS (blocking buffer)) and a goat anti-mouse IgG secondary antibody conjugated with Alexa Fluor 488 (ThermoFisher Scientific; 10 μg/ml in diluted in blocking buffer). After wash- ing with D-PBS, the cells were resuspended in 200 μl of 2 mM EDTA-containing D-PBS and filtered through a 40-μm filter. The fluorescent intensity of single cells was quantified by an EC800 flow cytometer equipped with a 488 nm laser (Sony). The fluorescent signal derived from Alexa Fluor 488 was recorded in the FL1 channel, and flow cytometry data were analyzed with the FlowJo software (FlowJo). Live cells were gated with a forward scatter (FS-Peak-Lin) cutoff of 390, setting a gain value of 1.7. Values of mean fluorescence intensity (MFI) from approximately 20,000 cells per sample were used for analysis. Supporting information S1 Fig. Analysis of the MSM construction by the bootstrap method. From the Model 1 tra- jectory set (379 trajectories), 300 trajectories (~80%) were sampled with replacement, and 5 datasets (Sets 1–5) were created. For each dataset, we constructed the MSM models and classi- fied them into the macrostates, as described in the Methods section. The microstates of MSM are plotted on the PC1-PC3 plane. The sizes of the circles are proportional to the populations of the clusters, and the centers of the clusters are color-coded according to their macrostates, as in Fig 5a. The results gave similar distributions of the macrostates, showing the robustness of the MSM constructed in this study. (PDF) S2 Fig. Similarity between the Model 1 and Model 2 simulation results. The histograms of the distances between the ligand head group (phosphate oxygen atoms) and the important basic residues (side-chain nitrogen atoms) of the receptor are plotted for the results of the Model 1 and Model 2 simulations. The resulting distributions overlap well, suggesting that the interactions and structural ensembles of the two simulations converged to a similar distribu- tion, despite the differences in their initial docking poses. (PDF) S1 Movie. Movie showing the reconstituted 100-μs trajectory of the LPA binding process. The ligand and conserved basic residues are shown in space-filling and stick models, respec- tively. (MP4) S1 File. (DOCX) Acknowledgments Computations of MD simulations were partially performed on the NIG supercomputer at ROIS National Institute of Genetics. We thank Reiya Taniguchi for helpful discussions and advice, and Kayo Sato, Yumiko Sugawara, Shigeko Nakano and Ayumi Inoue (Tohoku Uni- versity) for their assistance with plasmid preparation, maintenance of cultured cells and cell- based GPCR assays. Author Contributions Conceptualization: Rieko Suenaga, Mizuki Takemoto, Ryuichiro Ishitani, Osamu Nureki. Formal analysis: Asuka Inoue. Funding acquisition: Asuka Inoue, Ryuichiro Ishitani. PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 18 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor Investigation: Rieko Suenaga, Mizuki Takemoto, Ryuichiro Ishitani. Methodology: Rieko Suenaga, Mizuki Takemoto, Asuka Inoue, Ryuichiro Ishitani. Project administration: Ryuichiro Ishitani, Osamu Nureki. Supervision: Ryuichiro Ishitani, Osamu Nureki. Validation: Rieko Suenaga, Mizuki Takemoto, Ryuichiro Ishitani, Osamu Nureki. Visualization: Rieko Suenaga, Mizuki Takemoto. Writing – original draft: Rieko Suenaga, Asuka Inoue, Ryuichiro Ishitani, Osamu Nureki. Writing – review & editing: Rieko Suenaga, Asuka Inoue, Ryuichiro Ishitani, Osamu Nureki. References 1. Shimizu T. Lipid Mediators in Health and Disease: Enzymes and Receptors as Therapeutic Targets for the Regulation of Immunity and Inflammation. Annu. Rev. Pharmacol. Toxicol. 49, 123–150 (2009). https://doi.org/10.1146/annurev.pharmtox.011008.145616 PMID: 18834304 2. Kazantseva A., et al. Human Hair Growth Deficiency Is Linked to a Genetic Defect in the Phospholipase Gene LIPH. Science 314, 2004–2007 (2006). https://doi.org/10.1126/science.1133276 PMID: 17095700 3. Pasternack S. M., et al. G protein-coupled receptor P2Y5 and its ligand LPA are involved in mainte- nance of human hair growth. Nat. Genet. 40, 329–334 (2008). https://doi.org/10.1038/ng.84 PMID: 18297070 4. Ketscher A., et al. LSD1 controls metastasis of androgen-independent prostate cancer cells through PXN and LPAR6. Oncogenesis 3, (2014). https://doi.org/10.1038/oncsis.2014.34 PMID: 25285406 5. Taniguchi R., et al. Structural insights into ligand recognition by the lysophosphatidic acid receptor LPA6. Nature 548, 356–360 (2017). https://doi.org/10.1038/nature23448 PMID: 28792932 6. Hanson M. A., et al. Crystal structure of a lipid G protein-coupled receptor. Science 335, 851–855 (2012). https://doi.org/10.1126/science.1215904 PMID: 22344443 7. Hua T., et al. Crystal Structure of the Human Cannabinoid Receptor CB 1. Cell 167, 750–762.e14 (2016). https://doi.org/10.1016/j.cell.2016.10.004 PMID: 27768894 8. Shao Z., et al. High-resolution crystal structure of the human CB1 cannabinoid receptor. Nature 540, 602–606 (2016). https://doi.org/10.1038/nature20613 PMID: 27851727 9. 10. Fan H., et al. Structural basis for ligand recognition of the human thromboxane A 2 receptor. Nat. Chem. Biol. 15, 27–33 (2019). https://doi.org/10.1038/s41589-018-0170-9 PMID: 30510189 Toyoda Y., et al. Ligand binding to human prostaglandin E receptor EP 4 at the lipid-bilayer interface. Nat. Chem. Biol. https://doi.org/10.1038/s41589-018-0131-3 PMID: 30510193 11. Wang L., et al. Structures of the Human PGD 2 Receptor CRTH2 Reveal Novel Mechanisms for Ligand Recognition. Mol. Cell 72, 48–59.e4 (2018). https://doi.org/10.1016/j.molcel.2018.08.009 PMID: 30220562 12. Srivastava A., et al. High-resolution structure of the human GPR40 receptor bound to allosteric agonist TAK-875. Nature 513, 124–127 (2014). https://doi.org/10.1038/nature13494 PMID: 25043059 13. Ho J. D., et al. Structural basis for GPR40 allosteric agonism and incretin stimulation. Nat. Commun. 9, 1–11 (2018). 14. Cao C., et al. Structural basis for signal recognition and transduction by platelet-activating-factor recep- tor. Nat. Struct. Mol. Biol. 25, 488–495 (2018). https://doi.org/10.1038/s41594-018-0068-y PMID: 29808000 15. Prinz J. H., et al. Markov models of molecular kinetics: Generation and validation. J. Chem. Phys. 134, (2011). https://doi.org/10.1063/1.3565032 PMID: 21548671 16. Lane T. J., Shukla D., Beauchamp K. A. & Pande V. S. To milliseconds and beyond: Challenges in the simulation of protein folding. Curr. Opin. Struct. Biol. 23, 58–65 (2013). https://doi.org/10.1016/j.sbi. 2012.11.002 PMID: 23237705 17. Bai Q., et al. Ligand induced change of β2 adrenergic receptor from active to inactive conformation and its implication for the closed/open state of the water channel: Insight from molecular dynamics simula- tion, free energy calculation and Markov state model analysis. Phys. Chem. Chem. Phys. 16, 15874– 15885 (2014). https://doi.org/10.1039/c4cp01185f PMID: 24962153 PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 19 / 20 PLOS ONEMolecular dynamics simulation of LPA receptor 18. Decherchi S., Berteotti A., Bottegoni G., Rocchia W. & Cavalli A. The ligand binding mechanism to purine nucleoside phosphorylase elucidated via molecular dynamics and machine learning. Nat. Com- mun. 6, 6155 (2015). https://doi.org/10.1038/ncomms7155 PMID: 25625196 19. Ro¨blitz S. & Weber M. Fuzzy spectral clustering by PCCA+: Application to Markov state models and data classification. Adv. Data Anal. Classif. 7, 147–179 (2013). 20. Luginina A., et al. Structure-based mechanism of cysteinyl leukotriene receptor inhibition by antiasth- matic drugs. Sci. Adv. 5, eaax251, (2019). https://doi.org/10.1126/sciadv.aax2518 PMID: 31633023 21. Krishna Kumar K., et al. Structure of a Signaling Cannabinoid Receptor 1-G Protein Complex. Cell 176, 448–458.e12 (2019). https://doi.org/10.1016/j.cell.2018.11.040 PMID: 30639101 22. Abraham M. J., et al. Gromacs: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. SoftwareX 1– 2, 19–25 (2015). 23. Best R. B., et al. Optimization of the additive CHARMM all-atom protein force field targeting improved sampling of the backbone φ, ψ and side-chain χ1 and χ2 Dihedral Angles. J. Chem. Theory Comput. 8, 3257–3273 (2012). https://doi.org/10.1021/ct300400x PMID: 23341755 24. Stansfeld P. J., et al. MemProtMD: Automated Insertion of Membrane Protein Structures into Explicit Lipid Membranes. Structure 23, 1350–1361 (2015). https://doi.org/10.1016/j.str.2015.05.006 PMID: 26073602 25. McGibbon R. T., et al. MDTraj: A Modern Open Library for the Analysis of Molecular Dynamics Trajecto- ries. Biophys. J. 109, 1528–1532 (2015). https://doi.org/10.1016/j.bpj.2015.08.015 PMID: 26488642 26. Beauchamp K. A., et al. MSMBuilder2: Modeling conformational dynamics on the picosecond to milli- second scale. J. Chem. Theory Comput. 7, 3412–3419 (2011). https://doi.org/10.1021/ct200463m PMID: 22125474 27. Herna´ndez C.X., Harrigan M.P., Sultan M.M. & Pande V.S. MSMExplorer: Data Visualizations for Bio- molecular Dynamics. J. Open Source Softw. 2, 188 (2017). 28. Humphrey W., Dalke A. & Schulten K. VMD: Visual molecular dynamics. J. Mol. Graph. 14, 33–38 (1996). https://doi.org/10.1016/0263-7855(96)00018-5 PMID: 8744570 29. Hunter J. D. Matplotlib: A 2D graphics environment. Comput. Sci. Eng. 9, 99–104 (2007). 30. Bentley J. L. Multidimensional Binary Search Trees Used for Associative Searching. Commun. ACM 18, 509–517 (1975). 31. Noe´, F., Schu¨ tte, C., Vanden-eijnden, E., Reich, L. & Weikl, T. R. Constructing the equilibrium ensemble of folding pathways from short off-equilibrium simulations. (2009). 32. Inoue A., et al. TGFα shedding assay: an accurate and versatile method for detecting GPCR activation. Nat. Methods 9, 1021–1029 (2012). https://doi.org/10.1038/nmeth.2172 PMID: 22983457 33. Ehlert F. J., et al. A simple method for estimation of agonist activity at receptor subtypes: comparison of native and cloned M3 muscarinic receptors in guinea pig ileum and transfected cells. J. Pharmacol. Exp. Ther. 289, 981–992 (1999). PMID: 10215678 PLOS ONE \| https://doi.org/10.1371/journal.pone.0263296 February 3, 2022 20 / 20 PLOS ONE
10.1371_journal.pone.0261838	RESEARCH ARTICLE Community perceptions towards invasion of Prosopis juliflora, utilization, and its control options in Afar region, Northeast Ethiopia Wakshum ShiferawID 1, Sebsebe Demissew2, Tamrat Bekele2, Ermias Aynekulu3 1 College of Agricultural Sciences, Natural Resources Management, Arba Minch University, Arba Minch, Ethiopia, 2 College of Natural and Computational Sciences, Addis Ababa University, The National Herbarium, Addis Ababa, Ethiopia, 3 World Agroforestry Centre (ICRAF), Nairobi, Kenya waaqsh@yahoo.com Abstract This study aimed to assess community perceptions towards invasion of Prosopis juliflora, utilization, and its control options in Afar region, Northern Ethiopia. Using purposive sam- pling and stratified random methods, 20 members of key informants and 154 households from four sites of Awash Fentale and Amibara Districts were selected. For data analysis, we used Kruskal Wallis non-parametric tests of K independent samples. About 30% of respon- dents in Amibara and 29% in Awash Fentale reported that Prosopis juliflora was largely introduced into their landscape by livestock. It showed that 29% of the respondents in Awash Fentale and 41% in Amibara responded that Prosopis juliflora largely invaded and affected rangelands. Morevover, about 1% of respondents in Awash Fentale and 14% in Amibara argued that Prosopis juliflora hindered movements of livestock. In addition, 30% of respondents in Amibara and 29% in Awash Fentale believe that Prosopis juliflora was largely dispersed by livestock. It showed that 20% of households in Awash Fentale and 41% in Amibara have the notion that Prosopis juliflora majorly impacted rangelands. Whereas 1.3% of respondents in Awash Fentale and 14% in Amibara argued that Prosopis juliflora have hampered the movement of livestock. Thus, the afromentioned findings are implica- tions for management of rangelands. With regard to the control of Prosopis juliflora inva- sions, 12% of respondents in Awash Fentale and 33% in Amibara District tried control its expansion by fire. About 10% of respondents in Awash Fentale and 9% in Amibara district managed Prosopis juliflora expansion by its utilization, whereas, in Awash Fentale (11%) and Amibara (8%) households indicated that invasion of Prosopis juliflora could be con- trolled by mechanical methods. It is advisable to do some managerial work to reverse these impacts as perceived by local communities in the study area to avert the aggressive prolifer- ation of Prosopis juliflora in the region. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Shiferaw W, Demissew S, Bekele T, Aynekulu E (2022) Community perceptions towards invasion of Prosopis juliflora, utilization, and its control options in Afar region, Northeast Ethiopia. PLoS ONE 17(1): e0261838. https://doi. org/10.1371/journal.pone.0261838 Editor: Tunira Bhadauria, Feroze Gandhi Degree College, INDIA Received: June 7, 2021 Accepted: December 12, 2021 Published: January 25, 2022 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0261838 Copyright: © 2022 Shiferaw et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting information files. PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 1 / 20 PLOS ONEFunding: The authors received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. Pastoralists’ perception towards the invasion of Prosopis juliflora Introduction Prosopis juliflora (Sw.) DC. (hereafter P. juliflora) is a shrub or tree species native to Mexico, Central, and Northern America. From its native ranges, P. juliflora spread to Africa, Asia, and Austria [1]. In Africa, P. juliflora was first introduced in Senegal in 1822 and continued to establish in other countries at different times [2]. In Ethiopia, it was first introduced in the 1970s to restore degraded lands [3]. After its introduction, P. juliflora started to aggressively expand. It has been become an invasive or noxious weed in several African countries including Kenya, Ethiopia, Sudan, Senegal, and South Africa [1,7]. In the introduced countries, P. juli- flora expansion increased. For example, PENHA [4] reported that the global cover of P. juli- flora as a whole was 50 million hectares and P. juliflora covered about 5 million hectares in Africa in 2014 while Shiferaw et al. [5] reported a land cover of 1.20x106 ha by P. juliflora. It was encroaching at a rate of 3.11 x 104 ha yr-1 and constituted 12.3% of the land surface in the Afar region. Pittroff [6] also reported that P. juliflora cover was more than 1.80 x 106 ha of Afar region. It has now become an invasive or noxious weed in several African countries including Kenya, Ethiopia, Sudan, Senegal, and South Africa [1,7]. Thus, community perception will play a significant role in rangeland management and lay the conceptual foundation for the management of the invasion by P. juliflora. Study by Dafalla [8] indicats that educated people are more supportive of the eradication of P. juliflora than people that are not educated. This is debatable! Less educated people are to some degree dependent only on wood of P. juliflora for livelihood support. In contrary, in Ethiopia, local people have negative attitudes towards P. juliflora. They believe that the species has replaced economically important pasture and farmlands, and threatening pastoral and agro-pastoral livelihoods. It is also thought to have impacted human and animal health. The species is has become a threat to road traffic, and water infrastructures. In addition, the invasion has turned into a major driver of biodiversity loss in the invaded regions [3]. Inadequate management practices like prevention of its invasion into rangelands by local development interventions and social conflicts over grazing lands facilitated the invasion of P. juliflora in Afar region [9–11]. The perceptions of Afar pastoralists concerning the P. juliflora invasion are negative because it impacts their livelihoods and environment they inhabit [12]. Likewise, the majority of the households in the Gewane district of Afar region have not appre- ciated the positive and significant association of P. juliflora with their income diversifications [13]. Palatable grasses including Chrysopogon plumulosus Hochst., Cenchrus ciliaris L., Setaria verticillata (L.), and other valuable woody species such as Acacia tortilis (Frossk.), A. senegal (L.) Willd., A. nilotica (L.) Willd. ex. Del. was being replaced by inavsaion of the species. Thus, the present study aimed to assess community perceptions towards (i) the intoruciton and inva- sion (ii) the socio-economic values of the species, and (iii) controlling options of P. juliflora in Afar region of Ethiopia. Materials and methods Description of the study area Amibara District is located in between altitudes of 741 and 746 m.a.s.l. It is located between 9˚ 190 4400 N and 40˚ 100 5200 E, whereas Awash Fentale is located at 700 and 1000 m.a.s.l. and 9˚ 100 0000 N and 40˚ 030 3300 E. The mean annual temperature for the Awash Fentale District was 27 ± 2˚C, while the mean minimum was 16.7 ± 1.97˚C. The mean maximum temperature was 37.8 ± 2.1˚C (Fig 1a). The mean annual temperature for Amibara District was 26.8 ± 4˚C, whereas the mean minimum temperature was 13.8 ± 4.3˚C and the mean maximum was 38.2 ± 2.3˚C (Fig 1). The study PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 2 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora Fig 1. (a) Thirty-one year climate diagram for Awash Fentale District and (b) Fifteen year climate diagram for Amibara District [16]. https://doi.org/10.1371/journal.pone.0261838.g001 areas are located within semiarid and arid agro-ecologies of Ethiopia. The annual precipitation of Awash Fentale and Amibara districts was 490 ± 34 mm and 416 ± 31 mm respectively (Fig 1). A total population of 83, 851 and 40,901 was living in Amibara and Awash Fentale respec- tively [14]. Ninety percent of Afar people are pastoralists, while another 10% are considered as agro-pastoralists [15]. Afar region is characterized by desert and semi-desert scrubland, Acacia-Commiphora woodland, and bush land vegetation types [17]. This study was conducted in Acacia-Commi- phora woodland and desert and semi-desert scrubland with vegetation subtype Acacia-Com- miphora woodland and bushland. The characteristic herbaceous vegetation consisted of Chrysopogon, Sporobolus, Dactyloctenium, cymbopogon, and Cynodon species. The woody veg- etation was mainly composed of Acacia Senegal (L.) Willd., Acacia oerfota (Forssk.) Schweinf., Acacia nilotica (L.) Willd. ex. Del., Acacia tortilis (Frossk.) Hayne, Acacia mellifera (Vahl) Benth., Acalypha acrogyna Pax, Cadaba rotundifolia Forssk., Dobera glabra (Forssk.) Poir., Grewia tenax (Forssk.) Fiori, Salvadora persica L., Balanites aegyptiaca (L.) Del., and Ziziphus spina-christi L. [17]. Data collection Socio-demographicof household characteristics. We used semistructured and struc- tured questionnaires to collect through key informants interview and household survey. Pri- mary data were being collected through discussion with key informants and sampled households using pre-tested semistructured and structured questionnaires. For data collection, a three stage sampling method i.e., districts were purposely selected while household respon- dents were randomly selected. We took random households for the household from two sites from each district namely Diduba and Kebena from Awash Fentale and Kurkura and Andido from Amibara disricts [18]. The respondents for households were randomly selected females and males that were heads of the family. The memebers of the key informants were selected from district experts that were related to natural resources management experts, team leaders of district natural resources manament, office leaders of each district agricultural and natural resources, and adiminstrators of each district. Diduba and Kurkura sites were less affected by P. juliflora compared to Kebena and Andido. A toral of 154 household which is 5% of total households was selected for household PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 3 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora survey from the four sites. The selected households were from lightly, moderately, and highly invaded sites. Sample households from total households in the study sites were then stratified into wealth, sex and, age categories and then selected using simple random sampling technique from the total households in the sites. Based on the information from key informants households which had > 10 camels, > 20 cattle and > 60 small ruminants (goats and sheep) were categorized as rich. Those owning 1–10 camels, 5–10 cattle, and 10–60 small ruminants were categorized as medium households while hoseholds with no camels, < 5 cattle and < 10 small ruminants were categorized as poor households [19]. Data analysis Qualitative and quantitative methods were used during the data analyses. The data were not normally distributed, thus non-parametric tests of K independent samples of Kruskal Wallis of χ2 for mean separtion was used. The empirical multinomial logit model for this study was specified as [20]: yi ¼ fðx1; x2 . . . xnÞ Where yi, the dependent variable, is the wealth or socioeconomic status of pastoralists, xi’s are the included explanatory variables. The dependent variable (yi) is defined as follows: 1 for the poor pastoralists, 2 for the middle-class pastoralists, and 3 for the rich pastoralists. yi is also defined as 1 for male and 2 for female pastoralists; 1 as the youth (1–18 years), 2 for adult class (19–60 years), and 3 for old class (> 65 years) of pastoralists. Aside from the wealth status, other variables initially considered for inclusion in the model include age, sex, and education status of the agropastoralists and pastoralists. Then, explanatory vs. response variables were used for the empirical valuations, all the analyses were done using the XX and YY procedures of descriptive statstics in SPSS Software [21]. Households were significantly different among their opinions regarding questions raised such as: what were their mode of living, how P. juliflora was introduced, why it was introduced into their sites, preferred site for P. juliflora regeneration, benefits they get from P.juliflora, use of P. juliflora for traditional medicine, the preparation of P. juliflora for traditional medicine for human disease, livestock disease, preparation method for traditional medicine for livestock diseases from P. juliflora (P < 0.05). However, the rests of the households’ perceptions didn’t show significant (P > 0.05) (S2 Table). The variable definitions and measurements are given in S1 Table. For assessing community awareness towards the invasion of P. juliflora, 32% from Awash Fentale and 66.2% households from Amibara District were used for the interview. Among the respondents, 71% and 29% were male and female households, respectively (Table 1). Of which 1%, 91%, and 8% of house- holds fall in young, middle, and elder age groups, respectively. The family size of households ranged from 0–13 and mean value of 6. About 95% of the respondents had no formal educa- tion, 1% had primary educations, and 2% attended secondary and post-secondary education. Among the respondents, 27% were representatives of peasant associations and also had differ- ent positions in government offices. Results Histrory of P. juliflora introduction We found that about 30% of the respondent in Amibara and 29% in Awash Fentale districts thought that P. juliflora was introduced to Afar by livestock. About 19% of the respondents in PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 4 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora Table 1. Household characteristics of Awash Fenatle and Amibara Districts. Explanatory variable Mean of household characteristics SE Minimum Maximum District Site Sex Age Wealth Relationship to household head Household type Household number Education of household head Education of household member https://doi.org/10.1371/journal.pone.0261838.t001 1.7 2.8 1.3 2.1 1.4 1.8 1.6 5.7 0.1 0.8 223.2 154.0 226.5 285.2 211.0 116.7 142.6 93.9 216.9 163.8 1 1 1 1 1 1 1 0 0 0 2 4 2 3 3 8 6 13 3 3 df 1 3 1 2 2 6 5 13 3 5 χ2 16.23 19.25 24.96 230.7 83.08 525.36 354.75 79.27 400.23 161.66 Amibara and 3% Awash Fentale districts thought argued that P. juliflora was introduced to by a foreigner came to Amibara area. A small proportion of the respondents in Amibara (6.5%) and Awash Fentale (1.3%) thought that the species was introduced by local people (Table 2). About 22% of key inormatns in Amibara and 33% in Awash Fentale thought that the species was introduced by a foreigner (Fig 2). Purposes of P. juliflora introduction About 31% of the respondents in Amibra and 10% in Awash Fentale reported that fuelwood was the main reason for the introduction of P. juliflora in Afar region (Fig 2). Households in Amibara (18%) and Awash Fentale (5%) reported that the species was introduced for shade purpose (Fig 2). About 7% of the respondents in Amibara and 9% in Awash Fentale reported that the species was introduced for the purpose of soil and water conservation (Fig 2). How- ever, 12% of the respondents in Amibra and 17% in Awas Fentale districts did not know the purpose of its introduction. According to the key inforamts, 11% in Amibara and 22% in Awash Fentale confirmed that P. juliflora was introduced for fuelwood purposes. About 33% of the key informants in Ami- bara and 22% in Awash Fentale reported that the species was introduced for shade purpose (Fig 3). Table 2. Agents for the introduction of P. juliflora into Awash Fentale and Amibara Didtricts. Response Local people Natural Foreigners Livestock Wild animals Others Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Frequency 2 10 0 5 5 29 44 46 0 1 0 11 % 1.3 7 0.0 3 3 19 29 30 0.0 0.6 0.0 7 https://doi.org/10.1371/journal.pone.0261838.t002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 5 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora Fig 2. Household responses on the reasons for the introduction of P. juliflora into Awash Fenatle and Amibara Districts. https://doi.org/10.1371/journal.pone.0261838.g002 Socio-economic values of P. juliflora Regarding the opinions of households, the use of P. juliflora for traditional medicine, prepara- tions of human traditional medicine from plant parts, human disease cured by P. juliflora, live- stock diseases cured by P. juliflora, and preparations of livestock traditional medicine from plant parts had shown that significant variations among respondents (P < 0.05) across Dis- tricts (S2 Table). Of households, about 49% of the respondents in Amibara and 12% in Awash Fentale district reported that they got different benefits from P. juliflora (Table 3). Use of P. julflora for medicinal purposes About 10% of households in Awash Fentale and 9% in Amibara District used P. juliflora for traditional medicines. It was reported that 8% of respondents in Awash Fentale and 5% in Amibara District used P. juliflora for curing human injury that pricked by its thorns. Only 8% and 4% of respondents in Awash Fentale and Amibara District perceived that livestock injures were cured by traditional medicine of P. juliflora (Table 4). Preparation of traditional medicine from P. juliflora For preparation of traditional medicine curing human diseases (wounds), 8% and 5% of households in Awash Fentale and Amibara responded that it cured by crushing its leaf part. But, the majority of households in Awash Fentale (25%) and Amibara (59%) didn’t know how to prepare traditional medicines for human diseases, whereas 8% and 4% of households in PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 6 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora Fig 3. Key informants responses for purposes of introduction of P. juliflora into Awash Fenatle and Amibara Districts. https://doi.org/10.1371/journal.pone.0261838.g003 Awash Fentale and Amibara responded that livestock wounds cured by pounding its leaf part. But, the majority of households in Awash Fentale (26%) and Amibara (62%) didn’t know how to prepare its traditional medicines for curing livestock (Table 4). Effects of P. juliflora Larger number of repondents in Amibara (41%) and Awash Fentale (20%) reported that P. juliflora affected them most by invading and degrading the health of their rangelands, whereas a few hoseholds in the study districts reported that P. juliflora blocked roads e.g., human and livestock roads (Table 5). Effect of P. juliflora invasion and land use changes About 16% of households in Awash Fenatle and 27% in Amibara District showed that conver- sions of land use land cover were caused by invasion of P. juliflora, whereas 11% in Awash Fenatle District and 16% in Amibara were caused by farm land expansion. In addition, 0.6% of respondents in Awash Fentale and 6% in Amibara Distrcits reported that the cause for conver- sions of land uses were drought or shortage of rainfall (Table 6). Households responded that 13% in Awash Fenatle and 35% in Amibara District changed from woodlands into P. juliflora, whereas 16% in Awash Fentale and 27% in Amibara district changed from grazing lands into P. juliflora cover. Moreover, 3% in Awash Fentale and 27% households in Amibara District reveald that water bodies changed into the P. juliflora cover (Fig 4). PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 7 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora Table 3. Perceptions of households on benefits of P. juliflora in Awash Fenatle and Amibara Districts. Benefits Shade for livestock Shade for people Fuel wood Furniture House construction Live fence Soil and Water Conservation Ameliorating effects Shelterbelt Fodder Combating desertification Medicinal values District Frequency Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara 9 11 2 4 10 66 1 0 3 6 10 2 1 1 0 1 0 1 2 0 5 1 9 10 % 5.8 7.1 1.3 2.6 6.5 42.9 0.6 0 1.9 3.9 6.5 1.3 0.6 0.6 0 0.6 0 0.6 1.3 0 3.2 0.6 5.8 6.5 https://doi.org/10.1371/journal.pone.0261838.t003 Effects of P. juliflora invasion and its control Of the hoseholds, 3% in Awash Fentale and 6% in Amibara had the attitudes that P. juliflora formed impenetrable thicket which hindered the easy movements of human beings around. Results also show that 11% of key informants in Awash Fentale and 22% in Amibara District argued that prime grazing lands of Afar region were invaded by P. juliflora (Fig 5). It was argued that 55% of households in Amibara and 27% in Awash Fenatle Districts reported that P. juliflora had negative impacts on biodiversity. However, the rest 11% in Amibara and 7% in Awash Fentale Districts had positive impacts on biodiversity. In addition, the majority of households (45%) indicated P. juliflora affected livestock in Amibara District and 27% in Awash Fentale District (Table 7). As a result, 33% of key informants in Awash Fentale and 22% in Amibara confirmed that livestock production and productivity reduced. In Awash Fentale and Amibara District, 7% and 11% of households agreed that P. juliflora impacted biodiversity. About 44% and 22% of key informants in Awash Fentale and Amibara also confirmed as an invasion of P. juliflora had negative impacts on biodiversity in terms of the reduction of species diversity and the rest 11% and 22% of key informants in Awash Fentale and Amibara perceived that P. juliflora didn’t affect species diversity (Fig 5). About 27% of households in Awash Fentale and 45% in Amibara District responded that livestock were the most affected by P. juliflora invasions. In addition, 7% and 20% of respon- dents in Awash Fentale and Amibara responded that plants affected by P. juliflora invasions. 21% and 38% of household respondents in Awash Fentale and Amibara District responded PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 8 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora Table 4. The use of P. juliflora for traditional medicine in Awash Fenatle and Amibara Districts. Do you think P. juliflora used for traditional medicine? Response District Frequency Yes No I do not know Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Which part of P. juliflora used for traditional medicine? Seeds or pods Awash Fentale Leaf Amibara Awash Fentale Amibara Preparation of P. juliflora for traditional medicine for human cure? Pounding Awash Fentale I don’t know Amibara Awash Fentale Amibara Human diseases cured by traditional medicine of P. juliflora? Wound Awash Fentale I do not know Amibara Awash Fentale Amibara Livestock diseases cured by traditional medicine of P. juliflora? Wound Awash Fentale I don’t know Amibara Awash Fentale Amibara 14 14 34 85 4 3 14 14 38 88 13 11 39 91 13 8 39 93 12 6 40 95 Preparation method for traditional medicine for livestock diseases from P. juliflora? Awash Fentale Pounding 12 I don’t know Amibara Awash Fentale Amibara 7 40 94 https://doi.org/10.1371/journal.pone.0261838.t004 Table 5. Impacts of P. juliflora in Awash Fenatle and Amibara Districts. % 10 9 22 55 3 2 9 9 25 57 8 7 25 59 8 5 25 60 8 4 26 62 8 5 26 61 Districts Frequency Effects Woody weedy in agricultural lands Invasion into grazing lands Blocking roads of livestock Blocking roads of human beings Invasion of water courses, drying rivers and water tables Awash Fentale Lack of aesthetic value due to its monoculture https://doi.org/10.1371/journal.pone.0261838.t005 Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara 12 7 30 63 2 22 2 9 0 1 5 0 % 8 5 20 41 1.3 14 1.3 6 0 0.6 1.3 0 PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 9 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora Table 6. The causes of land use land cover conversions in Awash Fentale and Amibara Districts. Causes District Frequency Livestock production beyond the carrying capacity/overgrazing Awash Fentale Farm land expansion Anthropogenic Invasion of P. juliflora Moisture stress Drought or shortage of rainfall Toxic effects of P.juliflora https://doi.org/10.1371/journal.pone.0261838.t006 Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara 17 25 4 0 0 1 24 42 2 0 1 9 2 24 % 11 16 3 0.0 0.0 0.6 16 27 1.3 0.0 0.6 6 1.3 16 that P. juliflora had impacts on plants. About 12% hoseholds (in Awash Fentale) and 33% (in Amibara) had attitudes that livestock were among those affected by the invasion of P. juliflora (Table 8). In Awash Fentale (10%) and Amibara Districts (9%) households had forwarded their sug- gestions that invasion of P. juliflora could overcome through management by utilization; whereas, in Awash Fentale (12%) and Amibara (33%) households indicated that invasion of P. juliflora could be controlled by fire. The other 11% households in Awash Fenatle and 8% in Amibara District reported that invasion of P. juliflora could be controlled by mechamical methods such as: mechanical control by cutting mature trees, mechanical control by uprooting Fig 4. Conversion of land uses due to P. juliflora inavsion in last 10 years in Awash Fentale and Amibara Districts. https://doi.org/10.1371/journal.pone.0261838.g004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 10 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora Fig 5. Responses of key informants about impacts of P. juliflora in Awash Fenatle and Amibara Districts. https://doi.org/10.1371/journal.pone.0261838.g005 mature trees, mechanical control by cutting juvenile stems, and mechanical control by uproot- ing juvenile stems (Fig 6). About 22% key informants (in Awash Fentale) and 44% (in Amibara) had attitudes that livestock were among those affected by the invasion of P. juliflora. Furthermore, in Awash Table 7. Types of impacts on biodiversity in Awash Fenatle and Amibara Districts. What kinds of impacts biodiversity on biodiversity? Positive Negative District Awash Fentale Amibara Awash Fentale Amibara What type of biodiversity affected by P. juliflora? Awash Fentale Livestock Plants Human beings Amibara Awash Fentale Amibara Awash Fentale Amibara Which will be the most affected biodiversity by P. juliflora? Livestock Awash Fentale Plants Human beings Wild animals Amibara Awash Fentale Amibara Awash Fentale Amibara Awash Fentale Amibara https://doi.org/10.1371/journal.pone.0261838.t007 Frequency 10 17 42 85 42 69 10 31 0 2 47 69 5 29 0 1 0 3 % 7 11 27 55 27 45 7 20 0 1.3 31 45 3 19 0 0.6 0 2 PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 11 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora Table 8. Perceptions of key informants on impacts of P. juliflora on biodiversity in Awash Fenatle and Amibara Districts. What kinds of impacts of P. juliflora on biodiversity? Response District Frequency Positive Negative Awash Fentale Amibara Awash Fentale Amibara What type of biodiversity affected by P. juliflora? Awash Fentale Livestock Human beings Plants Amibara Awash Fentale Amibara Awash Fentale Amibara What type of biodiversity most affected by P. juliflora? Livestock Awash Fentale Plants Amibara Awash Fentale Amibara https://doi.org/10.1371/journal.pone.0261838.t008 1 2 4 2 2 4 1 0 2 0 3 4 2 1 % 11 22 44 22 22 44 11 0 22 0 33 44 22 11 Fig 6. Responses of households about the control of invasion of P. juliflora in Awash Fentale and Amibara Districts. Note: CU = Manage by utilization, MCCMS = Mechanical control by cutting mature stems, MCUMS = Mechanical control by uprooting mature stems, MCCJS = Mechanical control by cutting juvenile stems, MCUJS = Mechanical control by uprooting juvenile stems. https://doi.org/10.1371/journal.pone.0261838.g006 PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 12 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora Fig 7. Responses of key informants about the control of invasion of P. juliflora in Awash Fentale and Amibara Districts. https://doi.org/10.1371/journal.pone.0261838.g007 Fentale (22%) and Amibara (11%) key informants indicated that invasion of P. juliflora reduced plant species diversity. In Awash Fentale (56%) and Amibara districts (44%) key infor- mants had forwarded their suggestions that invasion of P. juliflora could overcome through management by utilization. Unless the invasion of the species can be controlled in the future, in the Awash Fentale (22%) and in Amibara District (44%) key informants informed that prime grazing lands will be overtaken by its invasion (Fig 7). Discussion Lifestyle of households The majority of households in Awash Fentale were agro-pastoralists than those in Amibara District. The reason could be due to the existence of wetlands around Kebena River in Kebena site of Awash Fentale that most households engaged in farming and livestock rearing activities. Similarly, Tegegn [22] indicated that most households in the Gewane district of Ethiopia were agro-pastoral ways of living. Few households in the study areas engaged in non-farm activities and obtained their incomes from daily labor wages and petty trades shopping in small scale. Similar findings by Shackleton [23] state that households engaged in non-farm activities obtained their incomes in other engagements including employment in government institutions. Moreover, in the arid Kalahari of South Africa, Shackleton [24] argued that few households engaged in non-farm activities and earned their incomes from daily labor activities. Households’ perceptions to the introduction history of P. juliflora In the study sites, the reasons for the introduction of P. juliflora varied among households and key informants. The reasons could be due to little knowledge on who introduced P. juliflora and its means introduction into a new inavsion [25]. Reports also show that the introduction of P. juliflora was not clear in Afar region in particular and Africa in general [26]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 13 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora Similar findings by [27] reveal that introduction of most exotic plant species were related to lifestyle changes of communities and contradictions in views and perceptions. Mwangi and Swallow [28] argued that purposes of introduction of P. juliflora into most countries of Africa remain unclear and the local people were not engaged in its introduction. Contrary to the local communities’ perceptions, research reports [29] and [30] indicate that P. juliflora was intro- duced by workers in Middle Awash Irrigation Project in Afar region of Ethiopia. Households’ perceptions towards the uses of P. juiflora The majority of key informants in both District s stated that introduction of P. juliflora was for purpose of shade. Some scholars such as [26] and [30] suggested more general views indicating the introduction was to combat desertification. According to the findings of this study, P. juli- flora is used for shade purposes either for livestock or human beings [24,26,30–32]. On the other hand, a study by [12] shows that P. juliflora was used as livestock feed. Use of P. julflora for medicinal purposes Unlike the present study, research report by Shackleton [33] reveals that farmers’ used P. juliflora for shade purposes greater than used for fuelwood and medicine. However, findings by Shackle- ton [34] show that the use of P. juliflora for medicinal purposes is greater that for purposes of construction and land rehabilitation. On the other hand, Shackleton [33] reported that the use of P. juliflora for medicine was greater the uses of P. juliflora for fodder, fuelwood, and shade pur- poses, but less than that of pods for food of children and sloving job opportunities. In contrast to the present findings, Shackleton [34] reported that most households used P. juliflora for medici- nal purposes. Likewise, studies by Argaw [35] and Ibrahim [36] argue that P. juliflora used as ver- satile medicine and important and a potential candidate for deriving phytomedicines. According to Preeti [36]) and Henciya [37], the curative powers of traditional medicines for human beings or livestock could be due to high contents of extracts such as flavonoids, tannins, alkaloids, quinones, or phenolic compounds that demonstrate potentials in various antimicro- bial activities such as analgesic, anthelmintic, and antibiotic constituents. In Ethiopia, few liter- atures had reported on the parts of P. juliflora that were used for traditional medicine, on the methods of its preparation for human beings or livestock diseases, or about its curing effect. Walter [38] also reported on the use of the nectar and pollen of P. juliflora to produce honey. The author further argued that P. juliflora didn’t rank equally with those of indigenous species that had been known by people for thousands of years for their medical powers in India. A study by Wise [39] shows from the Northern Cape of South Africa that pod of P. juliflora had medicinal properties and is being used for medicinal purposes. In addition, research report by Shackleton [32] reveals that pods used to produce organic medicine were being gath- ered from local traders of cities, towns, and villages across South Africa’s Northern Cape. Besides, Pasiecznik [1] reported that all parts of P. juliflora tree had uses of traditional medi- cines. They further reported on the occasional medicinal use of P. juliflora by Native Ameri- cans to treat the epidemic diseases of Old World Origin. Effects of P. juliflora The majority of households reported that invasion of P. juliflora invaded more into rangelands in Amibara than Awash Fentale District. The reasons could be due to the initial introduction of the species into Amibara District. Similarly, Harnet [40] reported that the invasion of P. juli- flora into both dry season and wet season rangelands and roadsides in the Eastern lowlands of Eritrea. A study by Argaw [34] also shows that the overall analysis of responses showed range- lands were among the impacts of P. juliflora. In addition, Ndhlovu [41] also suggested that PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 14 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora wider areas of rangelands covered by P. juliflora invasions and reduced its grazing capacity by 34% in South Africa. Similar research reports confirmed that intermingling growth and thorny natures of P. juli- flora could be due to the formation of thicket that blocked access roads for animals or human beings [34,42–44]. Invasion of P. juliflora into rangelands made blocked the roadsides and resulted in the losses of prime grazing lands. Thus, the production and productivity of live- stock in the P. juliflora invaded areas were being reduced. Several researchers such as [45–47] and [5] also confirmed that the invasion of P. juliflora resulted in the loss of prime grazing lands and ultimately the causes for the reduction of livestock production and productivity. Overall, less positive attitudes were reflected by households about the impacts of P. juliflora than negative impacts, and this might be due to a lack of full scale awareness about the impacts of P. juliflora. Moreover, most of the respondents lost their livestock that the health of animals by P. juliflora and few of them used P. juliflora for income sources. These types of views were long-standing issues of hot debates in numerous other countries in the tropics and sub-tropics. A similar view in Sudan shows that the benefits from P. juliflora completely outweigh its detri- mental effects in this particular area [48]. The individuals in the tropics and subtropics accepted P. juliflora as positive where native tree species were not growing in their area. Studies have shown that people’s views to be shaped by the negative or positive attributes of the invasive plant species [25]. Respondents who had large number of livestock were likely to be more aware of the negative impacts of P. juliflora than the positive effects [12]. On the con- trary, several types of research show that P. juliflora has negative impacts on plant diversity [39,40,49,50]. On the other hand, Maundu [51] argued that P. juliflora was an aggressive weed with both strong positive and negative attributes. The negative perceptions towards P. juliflora slowly changed back to positive attitudes for the tree after communities recognized on how to exploit the tree for their economic and social benefits [52]. The variations in the views of the stakeholders on the impacts of P. juliflora on biodiversity might be due to the difference in their knowledge and practices regarding P. julidlora. Most scholars argued that P. juliflora affected biodiversity [26,34,39,50]. Most researchers reported that the invasion of P. juliflora influenced mostly plants [53–56]. Large proportions of households in Awash Fentale District were of the opinion that human beings were being injured by the long thorns of P. juliflora [34,48,56], whereas higher propor- tions of households in Amibara District show that impenetrable thicket of P. juliflora blocked roadsides that hindered the easy movement of both humans and animals [11,22,28,34]. The variations in the responses were due to the difference in the severity of invasions of P. julflora and the level of knowledge level between households in the two Districts. A report by Seid [12] shows that burning of P.juliflora was the most widely employed method to control the invasion of P. juliflora. Control of P. juliflora by burning was in line with the arguments of most key informants in the Amibara District. Most key informants in Awash Fentale on the other hand, tried to control P. juliflora invasion using chemical methods. The study by [33] reveals that control of P. juliflora by utilization that is charcoal making, fuel- wood, and contruction purpose was the best method to minimize the invasion P. juliflora. In the future, if the expansion of P. juliflora is not controlled, most households will predict that it will largely invade and affect prime grazing lands [33]. Factors shaping knowledges and perceptions of households The living modes of the communities were also likely to affect invasion of P. juliflora. Most of the time and in all places of Afar region, the communities might also hear the negative sides of P. juliflora than positive by extension agents and experts in their sites. PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 15 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora The most palatable part of P. juliflora by animals was leaf of P. juliflora and the most toxic plant part killing their animals after their consumptions [1]. The reasons could be due to a per- manent weakening of the ability to digest cellulose in pods. This might also be due to the high sugar content of the pod that depressed the rumen bacterial cellulose activity and finally killing the animal [43,56]. In this study, the decline in the number of livestock feeds could be due to the decline in the capacity of rangelands for livestock grazing through suppressing and displac- ing important indigenous forage species. Similar results were reported by [3,28,41,50] from their studies in Kenya, South Africa, and Ethiopia indicate the places or countries!. Most non- formal educated households in Awash Fenatle and primary school educated households in Amibara argued that P. juliflora was being introduced by NGOs and an individuals which sim- ilar to [3,18]. Preferresd sites for invasion of P. juliflora In Amibara and Awash Fentale District s, non-formal, primary, secondary, and tertiary-edu- cated households ranked rangelands > homestead > mechanized farmlands > roadsides with respect to preferred sites for the establishment of P. juliflora respectively. Findings by [24,30,33] showed that rangelands were changed more into P. juliflora than other land-use types. The reason could be due to greater probability of seed dispersal by fecal droppings and abiotic suitability in the rangelands and in homesteads (e.g., moisture and high organic matter accumulation). Conclusion The perceptions of local communities towards invasive species depend on the level of commu- nities’ awareness, knowledge, and practices in their socio-economic and ecological uses. The benefits of invasive species outweigh their negative effects. When this is the case, local commu- nities tend to embrace species and retain them an ecosystem. The majority of households in Awash Fentale and few households in Amibara District fol- low the agro-pastoral mode of living. The majority of the households specialized on on-farm activities or livestock production for their income sources. A few households in the study areas engaged in off farm activities e.g., daily laborers and petty trades. The majority of households viewed that livestock were the main causes for the spread of P. juliflora in their areas. Few peo- ple in Amibara district were of the opinion that a foreigner working in the Middle Awash Irri- gation project was the culprit for the introduction of P. juliflora into the Afar region. In general, the household use of P. juliflora is minimal in the study area. The majority of households were not using P. juliflora at all. Some households used P. juliflora as shade for live- stock. A few households used P.juliflora for traditional medicines to cure their livestock and human beings. The leaf of P. juliflora was the part mainly used than other parts of the plant. The severe invasion of rangelands was indicative of the fact that livestock production and productivity were ultimately hampered in the study area, thus making agro-pastoral livelihood at risk. P. juliflora also blocked roads and pathways so much that accessibility of grazing areas became more difficult for livestock and human beings. The majority of households in both Districts were of the view that P. juliflora had no impacts on biodiversity particularly on plants. Moreover, the species has encroached into lower topographic areas: homesteads, and wetland areas. People had made attempts to control P. juliflora using fire, mechanical clear-cutting, and control through utilization (e.g., charcoal making, utilization of its woods for different purposes). Thus, it’s advisable in ussing P. juliflora to redue its further invasion into rangelands and other landuses in the region. PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 16 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora Supporting information S1 Table. Variables definition and measurements [57]. (DOCX) S2 Table. Effects of site variation on perceptions of communities towards introduction, use, and effects of P. juliflora in Amibara and Awash Fentale Woredas. (DOCX) S1 File. (DOCX) Author Contributions Conceptualization: Wakshum Shiferaw, Ermias Aynekulu. Data curation: Wakshum Shiferaw. Formal analysis: Wakshum Shiferaw, Ermias Aynekulu. Funding acquisition: Wakshum Shiferaw. Investigation: Wakshum Shiferaw, Ermias Aynekulu. Methodology: Wakshum Shiferaw, Ermias Aynekulu. Project administration: Wakshum Shiferaw, Sebsebe Demissew, Tamrat Bekele. Resources: Wakshum Shiferaw. Software: Wakshum Shiferaw. Supervision: Wakshum Shiferaw, Tamrat Bekele, Ermias Aynekulu. Validation: Wakshum Shiferaw, Sebsebe Demissew, Tamrat Bekele, Ermias Aynekulu. Visualization: Wakshum Shiferaw, Sebsebe Demissew, Tamrat Bekele. Writing – original draft: Wakshum Shiferaw. Writing – review & editing: Wakshum Shiferaw, Tamrat Bekele, Ermias Aynekulu. References 1. Pasiecznik NM, Felker P, Harris PJC, Harsh LN, Cruz G, Tewari JC, et al (2001). The Prosopis juliflora —Prosopis juliflora pallida Complex: A Monograph, HDRA, Coventry. p172, ISSBN: 0 905343 301, UK. 2. 3. Jama, B, Zeila A (2005). Agroforestry in the drylands of Africa: a call to action, ICRAF Working Paper— no. 1. Nairobi, World Agroforestry Centre. Tessema TY (2012). Ecological and Economic Dimensions of the Paradoxical Invasive Species- Proso- pis juliflora and Policy Challenges in Ethiopia. Journal of Economics and Sustainable Development, 3 (8), ISSN 2222-2855, www.iiste.org. 4. PENHA (2014). Bereket Tsegay, Livingstone, J., Zeremariam Fre (Eds) In Proceedings of a Regional Conference, Exploring Prosopis juliflora Management and Policy Options in the Greater Horn of Africa, Addis Ababa, Ethiopia. 5. Shiferaw H, Schaffner U, Bewuket W, et al. (2019b). Modeling the current fractional cover of an invasive alien plant and drivers of its invasion in dryland ecosystems. Sci. Rep. 9, 1576. 6. Pittroff W. Invasive alien species: the threat to sustainable livelihoods and ecosystems health. In: Poli- cies and Realities-Needs for Environmental Rehabilitation. In: Proccedings of Biological Society of Ethi- opia. 2019; Addis Ababa, Ethiopia. 7. Shackleton RT, Le Maitre DC, Pasiecznik NM, Richardson DM. Prosopis: a global assessment of the biogeography, benefits, impacts, and management of one of the world’s worst woody invasive plant taxa. AoB PLANTS. 2014; 6: plu027; https://doi.org/10.1093/aobpla/plu027 PMID: 24899150 PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 17 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora 8. Dafalla TAM. An Approach towards Mesquite Management in Kassala State. MSc Thesis, University of Khartoum. 2007; Sudan, Khartoum, pp14-140. 9. Tilahun M, Angassa A, Abebe A, Mengistu A. Perception and attitude of pastoralists on the use and con- servation of rangeland resources in the Afar region, Ethiopia. Ecological Processes. 2016; 5(18):1–10. 10. Shiferaw W, Demissew S, Bekele T. Invasive alien plant species in Ethiopia: ecological impacts on bio- diversity a review paper. Int J Mol Biol Open Access. 2018; 3(4):171–178. 11. Rogers P, Nunan F, Fentie Abiy Addisu. Reimagining invasions: The social and cultural impacts of Pro- sopis juliflora on pastoralists in southern Afar, Ethiopia. Pastoralism: Research, Policy, and Practice. 2017; 7(22):1–13. 12. Dubale A. Invasive Plants and Food Security: the case of Prosopis juliflora in the Afar region of Ethiopia, Prepared for IUCN by FARM-Africa. 2008; Addis Ababa, Ethiopia. 13. Seid JM. Household perception about Prosopis juliflora and its effect on pastoral livelihood diversifica- tion strategy: the case of Gewane District in Afar Region State, Ethiopia. International Journal of Eco- nomics, Commerce and Research (IJECR). 2012; 2 (3): 32–64. 14. CSA. Population Projection of Ethiopia for All Regions at District Level from 2014–2017. 2013; Addis Ababa, Ethiopia. 15. Wakie T, Evangelista P, Laituri M. Mapping Current and Potential Distribution of Non- Native Prosopis juliflora in the Afar Region of Ethiopia. Plos One. (2014); 3: 9(11). https://doi.org/10.1371/journal.pone. 0112854 PMID: 25393396 16. Shiferaw W., Bekele T., Demissew S., Aynekulu E. Relationship between Prosopis juliflora invasion and livelihood diversification in the South Afar region, Northeast Ethiopia. Regional Sustainability, 2020; 1(1):82–92. 17. Friis I, Demissew S Breugel PV. Atlas of potential vegetation of Ethiopia. The Royal Danish Academy of Sciences and Letters. 2010; Biologiske Skrifter, 58, pp.1–315. 18. Shiferaw W, Demissew S, Bekele T, Aynekulu E. Effects of Prosopis juliflora invasions on land use/ cover change in South Afar region, Northeast Ethiopia. Advance Research Journal of Multidisciplinary Discoveries. 2019a; 34(5):26–43. 19. Phiria D, Franzel S, Mafongoya P, Jered I, Katanga R, Phiri S. Who is using the new technology? The association of wealth status and gender with the planting of improved tree fallows in Eastern Province, Zambia. Agricultural Systems. 2004; 79(2):131–144. 20. Ojiako IA, Manyong VM, Ezedinma C, Asumugha GN. Determinants of Wealth and Socioeconomic Sta- tus of Rural Households: An Application of Multinomial Logit Model to Soybean Farmers in Northern Nigeria. J Soc Sci. 2009; 19(1): 31–39. 21. 22. IBM Corporation (1986, 2016). IBM SPSS Statistics 24 Core System User’s Guide, USA printing. Tegegn, G.G. (2008). Experiences on Prosopis juliflora Management Case of Afar Region. Report for Farm Africa. 23. Shackleton S, Kirby D, Gambiza J. Invasive plants—friends or foes? Contribution of prickly pear (Opun- tia ficus-indica) to livelihoods in Makana Municipality, Eastern Cape, South Africa. Development South- ern Africa. 2011; 28(2): 177–193. 24. Shackleton SE, Shackleton RT. Local knowledge regarding ecosystem services and disservices from invasive alien plants in the arid Kalahari, South Africa. Journal of Arid Environments. 2018; 159: 22–33. 25. Pasiecznik N, Choge SK., Trenchard L, Harris P. Improving food security in famine-prone areas using invasive and underutilized Prosopis trees. Food Chain. 2012; 2(2): 197–206. 26. Koyira ST (2015). Impact of Prosopis juliflora L. (Fabaceae) on Plant Biodiversity at Alledeghi Wildlife Reserve and Surrounding Local Community, Ethiopia. MSc Thesis. Addis Ababa University, Ethiopia. 27. Reda A, Tewelde F (2017). Households Perception of Invasive Alien Plants Species in Tigray, Ethiopia. Am-Euras. J. Agric. & Environ. Sci., 17 (6): 471–481. 28. Mwangi E, Swallow B. Prosopis juliflora Invasion and Rural Livelihoods in the Lake Baringo area of Kenya. Conservation and Society. 2008; 6(2): 130–140. 29. Shiferaw H, Teketay D, Nemomissa S. Some Biological Characteristics That Foster the Invasion of Pro- sopis juliflora (Sw.) DC. in Middle Awash Rift Valley Area, North-Eastern Ethiopia. Journal of Arid Envi- ronments. 2004; 58:135–154. 30. Wakie W, Laituri M, Evangelista PH. Assessing the distribution and impacts of Prosopis juliflora through participatory approaches. Applied Geography. 2016; 66:132–143. 31. Wilgen BWV, Forsyth GG, Maitre DCL, Wannenburgh A, Kotze JDF, Berg EVD, et al. An assessment of the effectiveness of a large, national-scale invasive alien plant control strategy in South Africa. Biolog- ical Conservation. 2012; 148: 28–38. PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 18 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora 32. Shackleton RT, Maitre DCL, Wilgen BWV, Richardson DM. Use of non-timber forest products from inva- sive alien Prosopis juliflora (mesquite) and native trees in South Africa: implications for management. Forest Ecosystems. 2015b; 2(16): 1–11. 33. Shackleton RT, Maitre DCL, Richardson DM (2015a). Stakeholder perceptions and practices regarding Prosopis juliflora (mesquite) invasions and management in South Africa. Ambio, 44:569–581, https:// doi.org/10.1007/s13280-014-0597-5 PMID: 25547521 34. Argaw T (2015). Impacts of Utilizing Invasive Prosopis juliflora (SWARTZ) DC. Rural Household Econ- omy at Gewane District, Afar Regional State, North-Eastern Ethiopia. Journal of Economics and Sus- tainable Development, 6(5):81–97. 35. Ibrahim M, Nadir M, Ali A, Ahmed V, Rasheed M. Phytochemical Analysis of Prosopis juliflora Swartz DC. Pak. J. Bot. 2013; 45(6): 2101–2104. 36. Preeti K, Avatar SR., Mala A. Pharmacology and Therapeutic Application of Prosopisjuliflora: A Review. Journal of Plant Sciences. 2015; 3(4): 234–240. 37. Henciya S, Seturaman P, Rathinam AJ, Tsai Y, Nikam R, Wu Y, et al. Biopharmaceutical potentials of Prosopis spp. (Mimosaceae, Leguminosae). Journal of Food and Drug Analysis. 2017; 25:187–196. https://doi.org/10.1016/j.jfda.2016.11.001 PMID: 28911536 38. Walter K. Prosopis juliflora, an Alien among the Sacred Trees of South India. Academic dissertation, Tropical Forestry Reports. 2011; 38, University of Helsinki. 39. Wise RM, Wilgen BWV, Maitre DCL. Costs, benefits and management options for an invasive alien tree species: The case of mesquite in the Northern Cape, South Africa. Journal of Arid Environments. 2012; 84: 80–90. 40. Harnet B. The Ecological and Socioeconomic Role of Prosopis juliflora in Eritrea, An Analytical Assess- ment within the Context of Rural Development in the Horn of Africa. Ph.D. Dissertation, 2008; zur Erlan- gung des Grades. 41. Ndhlovu T, Milton-Dean SJ, Esler KJ. Impact of Prosopis (mesquite) invasion and clearing on the graz- ing capacity of semiarid Nama Karoo rangeland, South Africa. African Journal of Range and Forage Sci- ence. 2011; 28:129–137. 42. Mitiku GS. The Ecological Distribution and Socio-Economic impacts of Prosopis juliflora (Sw.) DC. in the Amibara District, Afar National Regional State, MSc Thesis, 2008; Addis Ababa University. 43. Mussa MA, Abdulkerim JU, Regasa T. Prosopis juliflora (Sw.) DC Distribution, Impacts, and available Control Methods in Ethiopia. Tropical and Subtropical Agroecosystems. 2017; 20(1): 75–89. 44. Patnaik P, Abbasi T, Abbasi SA. Prosopis (Prosopis juliflora): blessing and bane. Tropical Ecology. 2017; 58(3): 455–483. 45. FAO (2006). Invasion of Prosopis juliflora in India, In Problems posed by the introduction of Prosopis juliflora species in selected countries, Italy Rome, pp.13. 46. Syomiti M. How Prosopis juliflora can be economically rewarding to pastoral communities in Kenya’s rangeland. East Africa TIRI Research, TRB-13 (2015). Kenya Agricultural Research Institute, 2015; Kenya. 47. 48. Zergaw YA, Anke J., Detle M. (2015) Ecosystem engineer unleashed: Prosopis juliflora threatening eco- system services? Regional Environmental Change, 15: 155–167, https://doi.org/10.1007/s10113-014- 0616-x Laxen J. Is prosopis a curse or a blessing?–An ecological-economic analysis of an invasive alien tree species in Sudan. PhD Dissertation, 2007; University of Helsinki. pp. 4. 49. Singh G, Rathod TR, Mutha S, Upadhyaya S, Bala N. Impact of Different Tree Species Canopy on Diversity and Productivity of Understory Vegetation in Indian Desert. International Society for Tropical Ecology, Tropical Ecology. 2008; 49(1):13–23. 50. Shackleton RT, LeMaitre DC, VanWilgen BW, Richardson DM. The impact of invasive alien Prosopis species (mesquite) on native plants in different environments in South Africa. South African Journal of Botany. 2015c; 97: 25–31. 51. Maundu P, Kibet S, Morimoto Y, Imbumi M, Adeka R. Impact of Prosopis juliflora on Kenya’s semi-arid and arid ecosystems and local livelihoods, Biodiversity. 2009; 10:2–3, 33–50. 52. Namaemba EW. Policy Outcomes and Community Perceptions of a Dryland Invasive Species: A Case Study of Prosopis juliflora in Baringo County, Kenya. MSc Thesis, 2013; The Hague, The Netherlands. pp. 1–53. 53. El-Keblawy A., Al-Rawai A. (2007). Impacts of the invasive exotic Prosopis juliflora (Sw.) D.C. on the native flora and soils of the UAE. Plant Ecology.2007; 190:23–35. PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 19 / 20 PLOS ONEPastoralists’ perception towards the invasion of Prosopis juliflora 54. Getachew S, Demissew S, Woldemariam T (2012). Allelopathic Effects of the Invasive Prosopis juliflora (Sw.) DC on Selected Native Plant Species in Middle Awash, Southern Afar Rift of Ethiopia. Manage- ment of Biological Invasions. 2007; 3(2):105–114. 55. Mehari ZH. The Invasion of Prosopis juliflora and Afar Pastoral Livelihoods in the Middle Awash area of Ethiopia, Ecological Processes. 2015; 4(13). 56. Riet-Correa F, Medeiros RMT, Pfister JA, Mendonc¸a FS. Toxic plants affecting the nervous system of ruminants and horses in Brazil. Pesq. Vet. Bras. 2017; 37(12):1357–1368. 57. Shiferaw W, Demissew S, Bekele T, Aynekulu E. Relationship between Prosopis juliflora invasion and livelihood diversification in the South Afar region, Northeast Ethiopia. Regional Sustainability. 2020; 1 (1): 82–92. PLOS ONE \| https://doi.org/10.1371/journal.pone.0261838 January 25, 2022 20 / 20 PLOS ONE
10.1371_journal.pone.0284036	RESEARCH ARTICLE Migration background and COVID-19 related intensive care unit admission and mortality in the Netherlands: A cohort study 1,2,3, Meriem Khairoun2, Martine van Stigt Thans1, Danielle Meeder1, Gurbey OcakID Hazra Moeniralam1,4, Friedo W. Dekker3, Marianne C. Verhaar2, Willem Jan W. Bos1,5, Karin A. H. Kaasjager6 1 Department of Internal Medicine, Sint Antonius Hospital, Nieuwegein, The Netherlands, 2 Department of Nephrology and Hypertension, University Medical Center Utrecht, Utrecht, The Netherlands, 3 Department of Clinical Epidemiology, Leiden University Medical Center, Leiden, The Netherlands, 4 Department of Intensive Care, Sint Antonius Hospital, Nieuwegein, The Netherlands, 5 Department of Internal Medicine, Leiden University Medical Center, Leiden, The Netherlands, 6 Department of Internal Medicine, University Medical Center Utrecht, Utrecht, The Netherlands g.ocak@antoniusziekenhuis.nl Abstract Background Since the beginning of the SARS-CoV-2 pandemic, studies have been reporting inconsis- tently on migration background as a risk factor for COVID-19 outcomes. The aim of this study was to evaluate the association between migration background and clinical outcomes with COVID-19 in the Netherlands. Methods This cohort study included 2,229 adult COVID-19 patients admitted in two Dutch hospitals between February 27, 2020 and March 31, 2021. Odds ratios (ORs) for hospital admission, intensive care unit (ICU) admission and mortality with 95% confidence intervals (CIs) were calculated for non-Western (Moroccan, Turkish, Surinamese or other) persons as compared with Western persons in the general population of the province of Utrecht (the Netherlands) as source population. Furthermore, among hospitalized patients, Hazard ratios (HRs) with 95% CIs for in-hospital mortality and intensive care unit (ICU) admission were calculated using Cox proportional hazard analyses. Hazard ratios were adjusted for age, sex, body mass index, hypertension, Charlson Comorbidity Index, chronic corticosteroid use before admission, income, education and population density to investigate explanatory variables. Results Of the 2,229 subjects, 1,707 were of Western origin and 522 were of non-Western origin. There were 313 in-hospital deaths and 503 ICU admissions. As compared with persons with a Western origin in the general population of the province of Utrecht, the ORs for non-West- ern persons was 1.8 (95% CI 1.7–2.0) for hospitalization, 2.1 (95% CI 1.7–2.5) for ICU admission and 1.3 (95% CI 1.0–1.7) for mortality. Among hospitalized patients, HR for ICU a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Ocak G, Khairoun M, van Stigt Thans M, Meeder D, Moeniralam H, Dekker FW, et al. (2023) Migration background and COVID-19 related intensive care unit admission and mortality in the Netherlands: A cohort study. PLoS ONE 18(4): e0284036. https://doi.org/10.1371/journal. pone.0284036 Editor: Patricia Rezende do Prado, Federal University of Acre (UFAC), BRAZIL Received: November 14, 2022 Accepted: March 22, 2023 Published: April 5, 2023 Copyright: © 2023 Ocak et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Unfortunately, due to the General Data Protection Regulation (GDPR) of EU and the connected Dutch data protection legislation we are not allowed to share individual patient data even in a pseudonymized format (data contact: University Medical Center Utrecht, WAG/ mb/20/016983, mvianen@umcutrecht.nl). Funding: The authors received no specific funding for this work. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284036 April 5, 2023 1 / 12 PLOS ONECompeting interests: The authors have declared that no competing interests exist. admission was 1.1 (95% CI 0.9–1.4) and 0.9 (95% CI 0.7–1.3) for mortality for non-Western hospitalized persons as compared with hospitalized patients of Western origin after adjustment. Migration background and COVID-19 Conclusion Non-Western persons, including Moroccan, Turkish and Surinamese subjects, had increased risks of hospital admission, ICU admission and COVID-19 related death on a pop- ulation level. Among hospitalized COVID-19 patients, no association was found between migration background and ICU admission or mortality. Introduction Studies have been reporting inconsistently about the presence of ethnic disparity in terms of COVID-19 severity and outcomes [1–9]. Large COVID-19 cohort studies across different con- tinents reported higher rates of intensive care unit (ICU) admission [1, 2] and in-hospital death [1–4] among ethnic minority groups compared with white persons. In contrast, several other studies with similar study populations in terms of size and ethnic diversity did not find these associations [5–9]. The role of ethnicity has been the subject of research during previous pandemics. During the 2009/2010 H1N1 influenza outbreak, data analyses from across the United Kingdom showed increased mortality rates for people of non-White origin compared to those of White origin [10]. In addition, actively engaging local ethnic populations turned out to be one of the key factors in controlling the Ebola pandemic in West Africa in 2014 [11]. Given these find- ings, identification of the vulnerable ethnic groups would be essential for the implementation of targeted prevention measures in order to gain optimal control of the COVID-19 pandemic and future pandemics. So far, the majority of studies on the relationship between migration background and COVID-19 outcomes originates from the United States [12, 13] or the United Kingdom [14]. Only one recent study from the Amsterdam and its neighboring city Almere in the Nether- lands investigated the association between migration background and COVID-19 outcomes [15]. Other studies from other Western European countries on the association between migra- tion background and hospital admission, ICU and mortality are lacking. Furthermore, few studies adjusted for both socioeconomic and clinical factors in the association between migra- tion background and COVID-19 outcomes [9, 12–14]. The aim of this study was to investigate the association between migration background and COVID-19 related hospital admission, ICU admission and death on a population level. Fur- thermore, we examined the relationship between migration background and length of hospital stay ICU admission and in-hospital mortality among patients hospitalized with COVID-19. Material and methods Study design and population This observational cohort study was performed among patients hospitalized with COVID-19 in the two largest hospitals in the province of Utrecht in the Netherlands; the Sint Antonius Hospital (non-academic teaching hospital) and the University Medical Center Utrecht (aca- demic hospital). In the Sint Antonius Hospital, patients were included between March 5, 2020 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284036 April 5, 2023 2 / 12 PLOS ONEMigration background and COVID-19 and January 9, 2021. In the University Medical Center Utrecht, the inclusion period was from February 27, 2020 until March 31, 2021. The study population comprised adult patients aged 18 years or older who had a polymer- ase chain reaction proven COVID-19 diagnosis. Patients without a Dutch residential address (n = 3) or without any available information on socioeconomic status (n = 1) were excluded. This study was conducted in accordance with the Declaration of Helsinki. The Medical Research Ethics Committee decided that institutional review board approval was not required (University Medical Center Utrecht), as patients were not subjected to interventions. Patient identification numbers were visible to the researchers in order to access electronic medical files. All data were pseudonymized before analysis. Follow-up started at hospitalization date and ended with either death or discharge from hospital. Migration background Migration background was classified according to the categorization used by the Dutch Cen- tral Bureau of Statistics [16]. Patients were grouped into ‘Western origin’ (Europe, North America, Oceania, Indonesia, Japan) or ‘non-Western origin’ (Turkey, Africa, South America and Asia). The non-Western group was further disaggregated into patients of Moroccan, Turkish and Surinamese origin, as these patients represent the three largest minority groups in the Netherlands. The remaining non-Western persons were categorized as other non-Western origin. For each patient, medical records were viewed to collect data on birth place to assess migration background. For patients for whom country of birth was unknown, surname or physician-reported information available in the electronic health record was used to indicate the migration background [15]. A person who was born in Africa, Latin America, Asia or Tur- key were classified as a non-western background according to the Dutch Central Bureau of Statistics [16]. Demographic and clinical data Data on age and sex were collected through medical record review. Clinical explanatory vari- ables included the individual comorbidities scored in the Charlson Comorbidity Index [17]. Furthermore, data on body mass index, hypertension and chronic use of systemic corticoste- roids (before admission) were collected. Hypertension was scored present if patients were using any antihypertensive treatment on admission. Data on socioeconomic status (income, education and population density) was available on environmental-level socioeconomic status and was not available for the individual patients. Using demographic registries by the Dutch Central Bureau of Statistics based on the postal code of the patient’s residential address [18], patient were categorized into tertiles based on the level of minimum income households, the level of education and population density. Outcomes The primary outcome of our study was in-hospital mortality. Secondary outcomes were ICU admission and length of hospital stay. Statistical analysis Continuous variables were reported as medians with interquartile ranges and categorical vari- ables with counts and percentages. Person-days of follow-up were counted from the date of hospitalization to the date of death or hospital discharge. Patients were censored at the end of PLOS ONE \| https://doi.org/10.1371/journal.pone.0284036 April 5, 2023 3 / 12 PLOS ONEMigration background and COVID-19 observation date (January 9, 2021 for the Sint Antonius Hospital and March 31, 2021 for the University Medical Center Utrecht) or when transferred to another hospital. To investigate the association between migration background and COVID-19 related hos- pital admission, ICU admission and death on a population level, the distribution of the migra- tion background in the general population of the province of Utrecht (the Netherlands) (reference) were compared with the distribution in patients who were admitted to the hospi- tals, ICU and patients who died due to COVID-19. For this purpose, odds ratios (ORs) with 95% confidence intervals (CIs) were calculated for COVID-19 related hospital admission, ICU admission and death. In addition, groups with different migration backgrounds were compared after admission to the hospital. Cox proportional hazard regression models were used to evaluate the associa- tion between migration background and ICU admission and in-hospital mortality. Crude haz- ard ratios (HR) with 95% confidence intervals (CIs) were calculated, with the patient group of Western origin as the reference category. Furthermore, crude and adjusted Beta Coefficients with 95% CIs were calculated using linear regression to investigate the association between migration background and length of hospital stay and ICU stay. To investigate the potential role of explanatory variables in the association between migration background and outcomes (mortality, ICU admission or length of hospital stay), we adjusted the regression analyses in four models. In model 1, HRs were adjusted for age and sex only. In model 2, we also adjusted for body mass index, hypertension, Charlson Comorbidity Index and chronic use of systemic corticosteroids (before admission). In model 3, environmental socioeconomic status (income and education) was added as potential explanatory variable. In the final model, HRs were fully adjusted for age, sex, comorbidities, medication use, income, education and population den- sity. Data analyses were performed using IBM SPSS Statistics for Windows, Version 26.0 (IBM Corp, Armonk, NY, USA). Results Baseline characteristics Of the 2,355 patients, 2,229 patients met the inclusion criteria and were included in this study (Fig 1). Of the included patients, 1,707 (76.6%) were of Western origin and 522 (23.4%) were of non-Western origin. Of the 522 patients of non-Western origin, 224 were Moroccan, 105 were Turkish, 89 were Surinamese and 104 were from other non-Western origins. Table 1 shows the baseline characteristics. Non-Western persons were younger, had a lower income and lower level of education and were living in more densely populated areas than patients of Western origin. Furthermore, non-Western persons had a higher body mass index, had more often diabetes mellitus and less often cardiovascular diseases, chronic obstructive pulmonary disease and malignancies than patients of Western origin. In addition, persons of non-Western origin scored lower on the Charlson Comorbidity Index than patients of Western origin. The median follow-up was 7 (IQR 4–13) days. Follow-up ended for 1,697 (76.1%) patients with discharge from the hospital, 313 (14.0%) patients died during admission, 140 (6.3%) patients were transferred to another hospital and 79 (3.5%) patients were still admitted at the end of follow-up. Migration background of the general population and hospitalized patients In the general population of the province of Utrecht in the Netherlands (source population), 1,160,618 persons had a Western origin and 194,216 were of non-Western origin in 2020 (Table 2). Of the non-Western persons, 30% were Moroccan, 16% were Turkish, 10% Suri- namese and 45% had another non-Western origin. As compared with persons with a Western PLOS ONE \| https://doi.org/10.1371/journal.pone.0284036 April 5, 2023 4 / 12 PLOS ONEMigration background and COVID-19 Fig 1. Study population flowchart. https://doi.org/10.1371/journal.pone.0284036.g001 origin, the OR for non-Western persons was 1.8 (95% CI 1.7–2.0) for hospitalization with an OR of 2.6 (95% CI 2.3–3.0) for Moroccan subjects, an OR of 2.3 (95% CI 1.9–2.8) for Turkish subjects and an OR of 3.1 (95% CI 2.5–3.9) for Surinamese subjects. The OR for Non-Western persons as compared with the Western population was 2.1 (95% CI 1.7–2.5) for ICU admis- sion. The OR for ICU admission was 2.9 (95% CI 2.1–3.8) for Moroccan subjects, 2.4 (95% CI 1.6–3.7) for Turkish subjects and 3.5 (95% CI 2.3–5.4) for Surinamese subjects. For in hospital death, ORs were also increased for non-Western persons (OR 1.3 (95% CI 1.0–1.7)), including Moroccan subjects (OR 2.1 (95% CI 1.4–3.1)), Turkish subjects (OR 1.6 (95% CI 1.0–2.9)) and Surinamese subjects (OR 3.0 (95% CI 1.7–5.3)). We did not find increased risks of hospitaliza- tion, ICU admission or in-hospital mortality in non-Western persons other than Moroccan, Turkish or Surinamese subjects. Migration background and mortality among hospitalized patients There were 313 in-hospital deaths, of which 258 among patients of western origin and 55 among non-Western persons (Table 3). Non-Western background was not associated with an PLOS ONE \| https://doi.org/10.1371/journal.pone.0284036 April 5, 2023 5 / 12 PLOS ONEMigration background and COVID-19 Table 1. Baseline characteristics stratified for migration background. Patients Age (years) (IQR) Sex, male (%) Body mass indexa (kg/m2) (IQR) Income of households Low Intermediate High Level of education Low Intermediate High Population density High Intermediate Low Charlson Comorbidity Index (IQR) Cardiovascular disease (%) Diabetes mellitus (%) COPD (%) Malignancy (%) Hypertension (%) Chronic use of systemic corticosteroids (%) Total N = 2,229 Western origin N = 1,707 Non-Western origin N = 522 65.9 (56.2–76.0) 1,305 (58.5) 27.7 (24.7–31.3) 67.8 (58.7–76.8) 1,011 (59.2) 27.4 (24.5–31.0) 59.0 (48.8–70.5) 294 (56.3) 28.6 (25.5–31.9) 749 (33.6) 779 (34.9) 701 (31.4) 744 (33.4) 749 (33.6) 736 (33.0) 745 (33.4) 745 (33.4) 739 (33.2) 3 (1–5) 579 (26.0) 553 (24.8) 248 (11.1) 328 (14.7) 988 (44.3) 189 (8.5) 419 (24.5) 669 (39.2) 619 (36.3) 486 (28.5) 619 (36.3) 602 (35.3) 418 (24.5) 615 (36.0) 674 (39.5) 3 (2–5) 485 (28.4) 351 (20.6) 225 (13.2) 290 (17.0) 782 (45.8) 157 (9.2) 320 (63.2) 110 (21.1) 82 (15.7) 258 (49.4) 130 (24.9) 134 (25.7) 327 (62.6) 130 (24.9) 65 (12.5) 2 (1–4) 94 (18.0) 202 (38.7) 23 (4.4) 38 (7.3) 206 (39.5) 32 (6.1) SES, socioeconomic status; COPD, chronic obstructive pulmonary disease. aBody mass index missing in 179 patients https://doi.org/10.1371/journal.pone.0284036.t001 increased HR as compared with persons with a Western origin (HR 0,7, 95% CI 0.5–1.1). After adjustment for explanatory variables including age, sex, body mass index, hypertension, Charl- son Comorbidity Index, chronic use of systemic corticosteroids before admission, environ- mental socioeconomic status and population density, the HR was 0.9 (95% CI 0.7–1.3) for non-Western hospitalized persons as compared with hospitalized patients of Western origin. There was no association between the subcategories of patients of non-Western origin and the Table 2. Association between migration background and COVID-19 related hospital admission, ICU admission and mortality. General population Hospital Admission ICU Admission Mortality Hospital Admission ICU admission Mortality N = 1,354,834 N = 2,229 N = 503 N = 313 OR (95%CI) OR (95%CI) OR (95%CI) Western Non-Western 1,160,618 194,216 Moroccan Turkish Surinamese Other 57,563 30,783 19,441 86,429 1,707 522 224 105 89 104 374 129 53 24 22 30 258 55 27 11 13 4 1 (Ref) 1 (Ref) 1 (Ref) 1.8 (1.7–2.0) 2.1 (1.7–2.5) 1.3 (1.0–1.7) 2.6 (2.3–3.0) 2.9 (2.1–3.8) 2.1 (1.4–3.1) 2.3 (1.9–2.8) 2.4 (1.6–3.7) 1.6 (1.0–2.9) 3.1 (2.5–3.9) 3.5 (2.3–5.4) 3.0 (1.7–5.3) 0.8 (0.7–1.1) 1.1 (0.7–1.6) 0.2 (0.1–0.6) ICU, intensive care unit; AKI, OR, odds ratio; CI, confidence interval. https://doi.org/10.1371/journal.pone.0284036.t002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284036 April 5, 2023 6 / 12 PLOS ONEMigration background and COVID-19 Table 3. Association between migration background and in-hospital death. Number of events HR (95%CI) Western origin N = 1,707 Non-western origin N = 522 Moroccan Turkish N = 224 N = 105 Surinamese N = 89 Other N = 104 Crude aModel 1 Adjusted bModel 2 Adjusted cModel 3 Adjusted dModel 4 Adjusted 258 1 (ref) 55 27 11 13 4 0.7 (0.5–1.1) 0.8 (0.5–1.2) 0.8 (0.4–1.5) 1.1 (0.6–2.0) 0.2 (0.1–0.6) 1 1 (ref) 1.0 (0.7–1.3) 0.9 (0.6–1.3) 1.3 (0.7–2.3) 1.8 (0.9–3.1) 0.4 (0.1–1.1) 1 1 (ref) 1.0 (0.7–1.3) 0.9 (0.6–1.3) 1.2 (0.7–2.3) 1.6 (0.9–2.9) 0.5 (0.2–1.3) 1 1 (ref) 1.0 (0.7–1.4) 0.8 (0.6–1.3) 1.3 (0.7–2.3) 1.7 (0.9–2.9) 0.5 (0.2–1.3) 1 1 (ref) 0.9 (0.7–1.3) 0.8 (0.5–1.2) 1.1 (0.6–2.2) 1.5 (0.9–2.7) 0.4 (0.1–1.2) HR, hazard ratio; CI, confidence interval. aModel 1 Adjusted: Hazard ratio adjusted for age and sex. bModel 2 Adjusted: Hazard ratio adjusted for age, sex, body mass index, hypertension, Charlson Comorbidity Index and use of systemic corticosteroids. cModel 3 Adjusted: Hazard ratio adjusted for age, sex, body mass index, hypertension, Charlson Comorbidity Index, use of systemic corticosteroids and environmental socioeconomic status (proportion of minimum income households and proportion with low level of education). * dModel 4 Adjusted: Hazard ratio adjusted for age, sex, body mass index, hypertension, Charlson Comorbidity Index, use of systemic corticosteroids, environmental socioeconomic status and population density. https://doi.org/10.1371/journal.pone.0284036.t003 risk of in-hospital death. Adjusted HRs for patients of Moroccan, Turkish and Surinamese ori- gin were 0.8 (95% CI 0.5–1.2), 1.1 (95% CI 0.6–2.2) and 1.5 (95% CI 0.9–2.7), respectively. Migration background and ICU admission among hospitalized patients Of the 503 (22.6%) patients who were admitted to the ICU, 374 were from Western and 129 were from non-Western origin (Table 4). Patients of non-Western origin did not have an increased HR of ICU admission compared with the Western reference group (adjusted HR 1.1 (95% CI 0.9–1.4)). Subgroup analysis in non-Western persons showed similar HRs. Table 4. Association between migration background and ICU admission. Number of events HR (95%CI) Western origin N = 1,707 Non-western origin N = 522 Moroccan Turkish N = 224 N = 105 Surinamese N = 89 Other N = 104 374 129 53 24 22 30 Crude aModel 1 Adjusted bModel 2 Adjusted cModel 3 Adjusted dModel 4 Adjusted 1 (ref) 1.2 (0.9–1.5) 1.2 (0.9–1.6) 1.1 (0.7–1.6) 1.2 (0.8–1.9) 1.4 (0.9–2.0) 1 1 (ref) 1.1 (0.9–1.4) 1.1 (0.8–1.5) 1.0 (0.6–1.5) 1.1 (0.7–1.8) 1.3 (0.9–1.9) 1 1 (ref) 1.1 (0.9–1.4) 1.1 (0.8–1.4) 1.0 (0.6–1.5) 1.1 (0.7–1.7) 1.3 (0.9–1.9) 1 1 (ref) 1.1 (0.9–1.4) 1.1 (0.8–1.5) 1.0 (0.6–1.5) 1.1 (0.7–1.8) 1.3 (0.9–1.9) 1 1 (ref) 1.1 (0.9–1.4) 1.1 (0.8–1.5) 1.0 (0.6–1.5) 1.2 (0.8–1.8) 1.3 (0.9–1.9) HR, hazard ratio; CI, confidence interval. aModel 1 Adjusted: Hazard ratio adjusted for age and sex. bModel 2 Adjusted: Hazard ratio adjusted for age, sex, body mass index, hypertension, Charlson Comorbidity Index and use of systemic corticosteroids. cModel 3 Adjusted: Hazard ratio adjusted for age, sex, body mass index, hypertension, Charlson Comorbidity Index, use of systemic corticosteroids and environmental socioeconomic status (proportion of minimum income households and proportion with low level of education). * dModel 4 Adjusted: Hazard ratio adjusted for age, sex, body mass index, hypertension, Charlson Comorbidity Index, use of systemic corticosteroids, environmental socioeconomic status and population density. https://doi.org/10.1371/journal.pone.0284036.t004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284036 April 5, 2023 7 / 12 PLOS ONEMigration background and COVID-19 Migration background and length of hospital stay The median length of hospital stay was 7 days (interquartile range (IQR) 4–13). Of patients who did not stay at the ICU, the medium length of stay was 6 days (IQR 3–9). It took a medium length of 2 days (IQR 0–4) before patients were admitted at the ICU. The medium length of stay at the ICU was 15 days (IQR 8–26). No differences in length of hospital stay or ICU stay were seen between patients of Western and non-Western origin (Table 5), nor in subgroup analysis between Western and Moroccan, Turkish and Surinamese patients. Discussion In this cohort study, the main finding was that non-Western persons, including Moroccan, Turkish and Surinamese subjects, had increased odds ratios of hospital admission, ICU admis- sion and death related to COVID-19 when the distribution of these non-Western persons in the hospital were compared with the source population. In contrast, there were no differences when the comparison between non-Western persons and Western persons were restricted to hospitalized patients. Non-Western origin as compared with patients of Western origin among hospitalized patients was neither associated with in-hospital mortality, nor with ICU admission or length of hospital or ICU stay. Several previous studies showed conflicting results in the association between ethnicity and mortality [1–9, 12–14, 19–21]. An important reason for the differences between studies in the mortality risk, could be differences in the source population of the studies [1–9, 12–14, 19–21]. The results are probably different when using the general population as source population instead of hospitalized patients. Indeed, a recent meta-analysis showed that in hospitalized patients in the United States [13], there was no higher risk of mortality in the ethnic minority groups than white patients. In addition, it was shown that ethnic minorities were more likely to be hospitalized than white Americans [13]. Another recent study in the Netherlands showed that the risk of COVID-19 hospitalization was higher in all ethnic minority groups compared to the Dutch, but the risk of adverse outcomes after hospitalization was similar [15]. This is in line with our finding that patients with a migration background have an increased hospitaliza- tion risk and that among hospitalized patients, there is no association between migration back- ground and mortality. The association between migration background and ICU admission has also been investi- gated by several other studies. Most of the studies describing the association between ethnicity and ICU admission originate from the United States [13]. Although several studies showed increased risks of ICU admission [5, 22], other studies did not [23, 24]. A study from Canada showed that Asian and black immigrants had a higher risk of ICU admission compared with white persons [25]. A study from the United Kingdom showed that South Asian ethnicity was associated with a reduced risk of admission to ICU [26]. In contrast, another study from the United Kingdom showed that black, Asian and other ethnic minorities had increased risks of ICU admission [27]. There are only a few studies that compared the length of hospital stay between ethnic groups [5, 28, 29]. These studies did not find an association between length of hospital stay and ethnicity, which is in line with our study. A study in the United States found similar median length of hospital stay in black (6 days) and white (7 days) subjects [5]. Furthermore, a study from the United Kingdom found no differences in the total length of hospital admission or length of ICU stay between different ethnic groups and white persons [28]. In addition, there were no differences in the length of ICU stay between white (median stay of 18 days) and non-white (median stay of 18 days) subjects in a study from an ICU population from the United States [29]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284036 April 5, 2023 8 / 12 PLOS ONEMigration background and COVID-19 Table 5. Association between migration background and length of hospital stay. Days (IQR) Beta Coefficients (95% confidence interval) Crude aModel 1 Adjusted bModel 2 Adjusted cModel 3 Adjusted dModel 4 Adjusted Length of hospital stay Western origin N = 1,707 Non-western origin Moroccan Turkish N = 522 N = 224 N = 105 Surinamese N = 89 Other N = 104 ICU stay Western origin Non-western origin Moroccan Turkish Surinamese Other N = 374 N = 129 N = 53 N = 24 N = 22 N = 30 7 (4–13) 7 (4–13) 6 (3–13) 7 (4–12) 6 (4–13) 7 (4–12) 0 (ref) 0 (ref) 0 (ref) 0 (ref) 0 (ref) -0.5 (-1.7–0.7) 0.2 (-1.1–1.4) 0.2 (-1.0–1.5) 0.2 (-1.1–1.5) 0.4 (-1.0–1.7) 0 (-1.8–1.8) 0.5 (-1.3–2.3) 0.5 (-1.3–2.3) 0.3 (-1.6–2.1) 0.5 (-1.3–2.4) -1.7 (-4.1–0.7) -1.2 (-3.7–1.2) -0.9 (-3.3–1.5) -1.0 (-3.5–1.4) -0.7 (-3.2–1.9) -1.7 (-4.3–0.9) -1.2 (-3.8–1.4) -1.1 (-3.8–1.5) -1.2 (-3.9–1.4) -0.9 (-3.6–1.7) 0.9 (-1.6–3.3) 1.8 (-0.7–4.4) 1.8 (-0.7–4.4) 1.8 (-0.7–4.4) 2.0 (-0.6–4.6) 15 (8–27) 0 (ref) 0 (ref) 0 (ref) 0 (ref) 0 (ref) 16 (10–25) 1.3 (-2.2–4.8) 2.2 (-1.5–5.8) 2.3 (-1.3–6.0) 2.2 (-1.6–6.0) 2.6 (-1.4–6.6) 17 (12–28) 4.3 (-0.8–9.5) 4.3 (-0.9–9.5) 4.4 (-0.8–9.6) 3.7 (-1.7–9.1) 4.4 (-1.2–9.9) 14 (9–19) -3.8 (-10.6–3.1) -3.4 (-10.4–3.6) -3.4 (-10.5–3.6) -3.4 (-10.7–3.9) -2.1 (-9.6–5.3) 16 (10–23) -2.9 (-10.1–4.2) -2.5 (-9.7–4.8) -1.8 (-9.0–5.5) -1.7 (-9.1–5.7) -0.7 (-8.1–6.8) 15 (6–40) 3.0 (-3.4–9.4) 4.2 (-2.4–10.8) 4.2 (-2.5–10.8) 4.3 (-2.4–11.0) 4.8 (-2.0–11.6) aModel 1 Adjusted: Beta Coefficients adjusted for age and sex. bModel 2 Adjusted: Beta Coefficients adjusted for age, sex, body mass index, hypertension, Charlson Comorbidity Index and use of systemic corticosteroids. cModel 3 Adjusted: Beta Coefficients adjusted for age, sex, body mass index, hypertension, Charlson Comorbidity Index, use of systemic corticosteroids and environmental socioeconomic status (proportion of minimum income households and proportion with low level of education). * dModel 4 Adjusted: Beta Coefficients adjusted for age, sex, body mass index, hypertension, Charlson Comorbidity Index, use of systemic corticosteroids, environmental socioeconomic status and population density. https://doi.org/10.1371/journal.pone.0284036.t005 There could be several potential explanations for the differences in COVID-19 related out- comes between non-Western persons and persons of Western origin as shown in our study. Firstly, it could be that the higher prevalence of unknown underlying comorbid conditions in the non-Western persons makes this population more vulnerable for COVID-19 and adverse outcomes than the Western population [15]. Second, social and cultural differences, including scientific mistrust and adherence to COVID-19 measures and number of persons in a house- hold, could play an important role in our results [30]. Differences in vaccination between non- Western persons and persons of Western origin could not be an explanation in our study, as the inclusion period of our study ended at the beginning of the vaccination programme in the Netherlands. The risks were not increased for non-Western subjects other than Moroccan, Turkish and Surinamese subjects. This group is diverse and includes relatively young, healthy and highly educated expatriates from non-Western countries. We also showed that once patients were hospitalized, there were no differences in COVID- 19 related outcomes between migration backgrounds. There could be several explanations for this finding. An explanation could be that in this selection of patients with severe COVID-19 requiring hospitalization, origin of patients is not critical in causing adverse COVID-19 out- comes. Another explanation could be that all hospitalized patients have the same standard quality care which could decrease differences in adverse COVID-19 outcomes between West- ern and non-Western persons. In addition, there could be selection bias introduced by PLOS ONE \| https://doi.org/10.1371/journal.pone.0284036 April 5, 2023 9 / 12 PLOS ONEMigration background and COVID-19 differences in hospital presentation between Western and non-Western persons. We had no data on differences of service utilization between Western and non-Western persons. The results of this study could have important implications for public policy. Since non- Western persons, including Moroccan, Turkish and Surinamese subjects, had increased risks of hospital admission, ICU admission and death related to COVID-19, preventive strategies (including increasing adherence to covid-19 preventive measures and increasing vaccination acceptance) targeting these high-risk groups in the general population could significantly decrease disease burden. An important step is the identification of these non-Western persons. Collecting data about migration background by health providers could help for this purpose. However, this is a sensitive political issue in many European countries. Another important aspect is to develop and provide adapted educational materials for non-Western persons. Fur- thermore, trusted local sources should be involved to deliver important public health messages. Additionally, vaccination programmes adapted to the needs of these non-Western persons could help to improve vaccination rates as preventive strategy. Future studies should investi- gate the effectiveness of preventive strategies in non-Western persons. The major strengths of the present study are the large number of patients with information on migration background, clinical data, sociodemographic data and outcome data including length of hospital stay, ICU admission and mortality. This study also has limitations. First, migration background was determined on the basis of data in the medical records. Therefore, it could be that we have misclassified patients. We grouped patients with a Western back- ground into one group to minimalize misclassification. We believe that it is more likely that a Western background is not reported than a non-Western background. Second, socioeconomic status and population density data were derived from the postal code of the residential address as patient-reported data were not available. In the Netherlands there is no national index of deprivation, but household income and educational level are generally considered as reliable indicators. These environmental-based data have been used in previous research in the Neth- erlands to assess socioeconomic status [31, 32]. Third, we had no information about comor- bidities and other explanatory variables of patients with COVID-19 who were not hospitalized. Therefore, we could not investigate the role of these explanatory variables in the comparison with the general population. Finally, small risk differences between subgroups of non-Western persons might have been missed due to insufficient power. In conclusion, non-Western persons, including Moroccan, Turkish and Surinamese sub- jects, had increased risks of hospital admission, ICU admission and in-hospital death related to COVID-19 on a population level. In contrast, there were no differences in ICU admission and death when the comparison between non-Western persons and Western persons was restricted to hospitalized patients. Acknowledgments We gratefully acknowledge the Sint Antonius Hospital and University Medical Center Utrecht for their efforts in providing the patient data. Author Contributions Conceptualization: Gurbey Ocak, Meriem Khairoun, Martine van Stigt Thans, Danielle Meeder. Formal analysis: Gurbey Ocak, Meriem Khairoun, Martine van Stigt Thans, Danielle Meeder. Investigation: Gurbey Ocak, Meriem Khairoun, Martine van Stigt Thans, Danielle Meeder. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284036 April 5, 2023 10 / 12 PLOS ONEMigration background and COVID-19 Methodology: Gurbey Ocak, Meriem Khairoun. Supervision: Hazra Moeniralam, Friedo W. Dekker, Marianne C. Verhaar, Willem Jan W. Bos, Karin A. H. Kaasjager. Writing – original draft: Gurbey Ocak, Meriem Khairoun, Martine van Stigt Thans, Danielle Meeder. Writing – review & editing: Hazra Moeniralam, Friedo W. Dekker, Marianne C. Verhaar, Willem Jan W. Bos, Karin A. H. Kaasjager. References 1. Qeadan F, VanSant-Webb E, Tingey B, et al. Racial disparities in COVID-19 outcomes exist despite comparable Elixhauser comorbidity indices between Blacks, Hispanics, Native Americans, and Whites. Sci Rep 2021; 11(1):8738 https://doi.org/10.1038/s41598-021-88308-2 PMID: 33888833 2. Mathur R, Rentsch CT, Morton CE, et al. Ethnic differences in SARS-CoV-2 infection and COVID-19- related hospitalisation, intensive care unit admission, and death in 17 million adults in England: an observational cohort study using the OpenSAFELY platform. Lancet 2021; 397(10286):1711–1724 https://doi.org/10.1016/S0140-6736(21)00634-6 PMID: 33939953 3. Baqui P, Bica I, Marra V, et al. Ethnic and regional variations in hospital mortality from COVID-19 in Bra- zil: a cross-sectional observational study. Lancet Glob Health 2020; 8(8):e1018–e1026 https://doi.org/ 10.1016/S2214-109X(20)30285-0 PMID: 32622400 4. Navaratnam AV, Gray WK, Day J, et al. Patient factors and temporal trends associated with COVID-19 in-hospital mortality in England: an observational study using administrative data. Lancet Respir Med 2021; 9(4):397–406 https://doi.org/10.1016/S2213-2600(20)30579-8 PMID: 33600777 5. Price-Haywood EG, Burton J, Fort D, et al. Hospitalization and Mortality among Black Patients and White Patients with Covid-19. N Engl J Med 2020; 382(26):2534–2543 https://doi.org/10.1056/ NEJMsa2011686 PMID: 32459916 6. Wang Z, Zheutlin A, Kao YH, et al. Hospitalised COVID-19 patients of the Mount Sinai Health System: a retrospective observational study using the electronic medical records. BMJ Open 2020; 10(10): e040441 https://doi.org/10.1136/bmjopen-2020-040441 PMID: 33109676 7. Fabiani M, Mateo-Urdiales A, Andrianou X, et al. Epidemiological characteristics of COVID-19 cases in non-Italian nationals notified to the Italian surveillance system. Eur J Public Health 2021; 31(1):37–44 https://doi.org/10.1093/eurpub/ckaa249 PMID: 33416859 8. Gopal Rao G, Allen A, Papineni P, et al. Cross-sectional observational study of epidemiology of COVID-19 and clinical outcomes of hospitalised patients in North West London during March and April 2020. BMJ Open 2021; 11(2):e044384 https://doi.org/10.1136/bmjopen-2020-044384 PMID: 33602712 9. Raharja A, Tamara A, Kok LT. Association Between Ethnicity and Severe COVID-19 Disease: a Sys- tematic Review and Meta-analysis. J Racial Ethn Health Disparities 2021; 8(6):1563–1572 https://doi. org/10.1007/s40615-020-00921-5 PMID: 33180278 10. 11. Zhao H, Harris RJ, Ellis J, et al. Ethnicity, deprivation and mortality due to 2009 pandemic influenza A (H1N1) in England during the 2009/2010 pandemic and the first post-pandemic season. Epidemiol Infect 2015; 143(16):3375–3383 https://doi.org/10.1017/S0950268815000576 PMID: 25850904 Laverack G, Manoncourt E. Key experiences of community engagement and social mobilization in the Ebola response. Glob Health Promot 2016; 23(1):79–82 https://doi.org/10.1177/1757975915606674 PMID: 26518037 12. Magesh S, John D, Li WT, et al. Disparities in COVID-19 Outcomes by Race, Ethnicity, and Socioeco- nomic Status: A Systematic-Review and Meta-analysis. JAMA Netw Open 2021; 4(11):e2134147 https://doi.org/10.1001/jamanetworkopen.2021.34147 PMID: 34762110 13. Agyemang C, Richters A, Jolani S, et al. Ethnic minority status as social determinant for COVID-19 infection, hospitalisation, severity, ICU admission and deaths in the early phase of the pandemic: a meta-analysis. BMJ Glob Health 2021; 6(11) https://doi.org/10.1136/bmjgh-2021-007433 PMID: 34740916 14. Bann D, Villadsen A, Maddock J, et al. Changes in the behavioural determinants of health during the COVID-19 pandemic: gender, socioeconomic and ethnic inequalities in five British cohort studies. J Epi- demiol Community Health 2021; 75(12):1136–1142 https://doi.org/10.1136/jech-2020-215664 PMID: 34039660 15. Collard D, Stronks K, Harris V, et al. Ethnic Differences in Coronavirus Disease 2019 Hospitalization and Hospital Outcomes in a Multiethnic Population in the Netherlands. Open Forum Infect Dis 2022; 9 (6):ofac257 https://doi.org/10.1093/ofid/ofac257 PMID: 35783684 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284036 April 5, 2023 11 / 12 PLOS ONEMigration background and COVID-19 16. https://opendata.cbs.nl/statline/#/CBS/nl/dataset/37713. 17. Charlson ME, Pompei P, Ales KL, et al. A new method of classifying prognostic comorbidity in longitudi- nal studies: development and validation. J Chronic Dis 1987; 40(5):373–383 https://doi.org/10.1016/ 0021-9681(87)90171-8 PMID: 3558716 18. https://opendata.cbs.nl/#/CBS/nl/dataset/84583NED/table?ts=1633959860289. 19. Yates T, Summerfield A, Razieh C, et al. A population-based cohort study of obesity, ethnicity and COVID-19 mortality in 12.6 million adults in England. Nat Commun 2022; 13(1):624 https://doi.org/10. 1038/s41467-022-28248-1 PMID: 35110546 20. Feldman JM, Bassett MT. Variation in COVID-19 Mortality in the US by Race and Ethnicity and Educa- tional Attainment. JAMA Netw Open 2021; 4(11):e2135967 https://doi.org/10.1001/jamanetworkopen. 2021.35967 PMID: 34812846 21. Mackey K, Ayers CK, Kondo KK, et al. Racial and Ethnic Disparities in COVID-19-Related Infections, Hospitalizations, and Deaths: A Systematic Review. Ann Intern Med 2021; 174(3):362–373 https://doi. org/10.7326/M20-6306 PMID: 33253040 22. Golestaneh L, Neugarten J, Fisher M, et al. The association of race and COVID-19 mortality. EClinical- Medicine 2020; 25:100455 https://doi.org/10.1016/j.eclinm.2020.100455 PMID: 32838233 23. Kim L, Garg S, O’Halloran A, et al. Risk Factors for Intensive Care Unit Admission and In-hospital Mor- tality Among Hospitalized Adults Identified through the US Coronavirus Disease 2019 (COVID-19)- Associated Hospitalization Surveillance Network (COVID-NET). Clin Infect Dis 2021; 72(9):e206–e214 https://doi.org/10.1093/cid/ciaa1012 PMID: 32674114 24. McCarty TR, Hathorn KE, Redd WD, et al. How Do Presenting Symptoms and Outcomes Differ by Race/Ethnicity Among Hospitalized Patients With Coronavirus Disease 2019 Infection? Experience in Massachusetts. Clin Infect Dis 2021; 73(11):e4131–e4138 https://doi.org/10.1093/cid/ciaa1245 PMID: 32827436 25. Passos-Castilho AM, Labbe AC, Barkati S, et al. Outcomes of hospitalized COVID-19 in Canada: impact of ethnicity, migration status, and country of birth ethnicity, migration and COVID-19 in-hospital outcomes in Canada. J Travel Med 2022 26. Santorelli G, McCooe M, Sheldon TA, et al. Ethnicity, pre-existing comorbidities, and outcomes of hos- pitalised patients with COVID-19. Wellcome Open Res 2021; 6:32 https://doi.org/10.12688/ wellcomeopenres.16580.2 PMID: 34522788 27. Thomas DR, Orife O, Plimmer A, et al. Ethnic variation in outcome of people hospitalised during the first COVID-19 epidemic wave in Wales (UK): an analysis of national surveillance data using Onomap, a name-based ethnicity classification tool. BMJ Open 2021; 11(8):e048335 https://doi.org/10.1136/ bmjopen-2020-048335 PMID: 34408047 28. Apea VJ, Wan YI, Dhairyawan R, et al. Ethnicity and outcomes in patients hospitalised with COVID-19 infection in East London: an observational cohort study. BMJ Open 2021; 11(1):e042140 https://doi. org/10.1136/bmjopen-2020-042140 PMID: 33455936 29. Lazar MH, Moscoso EE, Jennings JH. Outcomes in Medical Intensive Care Patients Housed in Geo- graphically Distant Units. J Intensive Care Med 2020; 35(12):1471–1475 https://doi.org/10.1177/ 0885066619835488 PMID: 30836814 30. Kazemian S, Fuller S, Algara C. The role of race and scientific trust on support for COVID-19 social dis- tancing measures in the United States. PLoS One 2021; 16(7):e0254127 https://doi.org/10.1371/ journal.pone.0254127 PMID: 34242275 31. Timmermans EJ, Veldhuizen EM, Snijder MB, et al. Neighbourhood safety and smoking in population subgroups: The HELIUS study. Prev Med 2018; 112:111–118 https://doi.org/10.1016/j.ypmed.2018.04. 012 PMID: 29654838 32. Snijder MB, Galenkamp H, Prins M, et al. Cohort profile: the Healthy Life in an Urban Setting (HELIUS) study in Amsterdam, The Netherlands. BMJ Open 2017; 7(12):e017873 https://doi.org/10.1136/ bmjopen-2017-017873 PMID: 29247091 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284036 April 5, 2023 12 / 12 PLOS ONE
10.1371_journal.pone.0284204	RESEARCH ARTICLE Do network synergies facilitates the realization of M&A motivation?: From the perspective of network node degree and strength change Ziqiao Zhang1, Qiusheng ZhangID 2, Chaofeng Li3 1 School of Accounting, Luoyang Institute of Science and Technology, Luoyang, China, 2 School of Economics and Management, Beijing Jiaotong University, Beijing, China, 3 China North Industries Corporation, Beijing, China 200901000639@lit.edu.cn Abstract The current evaluation of M&A performance lacks consideration of M&A motives. In this paper, we theoretically analyse and empirically test the effect of network synergy generated by M&A on the degree of realization of corporate M&A motives and the mechanism of its effect by constructing an equity network between a listed company and its subsidiaries within the company. The results show that the greater the variation of internal network node degree and strength, the more beneficial it is to promote the degree of realization of corpo- rate M&A motivation; the results of further mechanism of action tests show that the variation of network node degree and strength has significant effects on economies of scale, econo- mies of scope, and transaction costs; Furthermore, the heterogeneity test finds that the effect of variation of network node degree and strength to promote the degree of realization of corporate M&A motivation is more significant in the case of non-cash payment method and related M&A. This paper extends the study of complex networks to the field of M&A and uniquely explains the paradox of the "high failure rate" of M&A and the increasing activity of M&A activities from the perspective of network synergy, which helps to rationalize the M&A behavior of enterprises and further regulate the M&A behavior of listed companies by regu- latory authorities. 1 Introduction Known as one of the most important strategic investment methods for enterprises to use the capital market to improve quality and efficiency, gain network synergies, and cultivate compet- itive advantages in the market, mergers and acquisitions (M&A) have had a significant impact on the development of a modern industrial system and on the high-quality development of the Chinese economy in recent years. According to the "Review of China’s M&A market in 2020 and prospect in 2021’ released by PwC", the deal value of M&A activities in China increased by 30% in 2020 to reach $7,338 billion US dollars, the highest level since 2016. However, scholarly a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Zhang Z, Zhang Q, Li C (2023) Do network synergies facilitates the realization of M&A motivation?: From the perspective of network node degree and strength change. PLoS ONE 18(4): e0284204. https://doi.org/10.1371/journal. pone.0284204 Editor: Valentina Diana Rusu, Alexandru Ioan Cuza University: Universitatea Alexandru Ioan Cuza, ROMANIA Received: November 17, 2022 Accepted: March 24, 2023 Published: April 20, 2023 Copyright: © 2023 Zhang et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Funding: This study was supported by Project Item No.1: Research on the Realization of M&A Motivation and Mechanism Based on Network Synergy (National Natural Science Foundation of China, ID: 72072009), and Project Item No.2: Research on M&A Performance of Listed Companies under the Requirements of High-quality PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 1 / 21 PLOS ONEDevelopment: A Case Study of Henan Province (Soft Science Project of Henan Provincial Science and Technology Department, ID: 222400410652). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Do network synergies facilitates the realization of M&A motivation? research indicates that the growing M&A activity is accompanied by a high rate of failure [1]. M&A activity frequently fails to generate value for shareholders and, in some cases, can detract from the value of the firm [2, 3]. This paradox has become a scientific issue at the forefront of M&A [4]. The mainstream approach to evaluate M&A performance is to compare cumulative abnor- mal stock returns, long-term shareholder value, and changes in financial metrics before and after the acquisition. However, these approaches may overlook the "original intent" of M&A motives and result in inaccurate conclusions. Several Scholars attempted to evaluate M&A per- formance from the perspective of M&A motives, and Brouthers et al. [5] further argued that the success and failure of M&A should be measured by examining the extent to which M&A motives are realized. In existing studies, some researchers have made different classifications of M&A motives, however, they just used single indicators or synthetic indicators obtained through factor analysis and principal component analysis instead of combining all types of motives to evaluate the performance [6, 7]. Even though some researchers have attempted to evaluate M&A performance through different types of M&A motives, they have only con- structed evaluation systems theoretically without thorough testing [8, 9]. Most of the M&A performance information from existing tests was obtained either through questionnaires ask- ing executives to rate relevant indicators, asking reviewers to comment on relevant indicators, or using hierarchical analysis, which are all empowerment methods that assign weights to rele- vant indicators [10, 11]. However, these approaches cannot accurately reflect the M&A perfor- mance, and there is currently no empirical test for large samples or research results on the mechanisms that influence the degree of realization of corporate M&A motivation. Under the background of the new era, enterprise networking features become increasingly distinct, and competition among businesses have transformed into competition among busi- nesses’ economic networks. The enterprise economic network is a complex system jointly formed by an enterprise’s internal economic relationship based on control relationships and inter-firm economic networks embedded in enterprises. Since M&A is the merging and rear- rangement of two economic systems, and the network economy can modify the structure of the enterprise economic network before and after the merger to build a network economy, the network economy becomes the most important source of synergy effects of M&A in the enter- prise economic network. However, in recent years, goodwill has been frequently “mined” by companies, including failures to meet performance expectations for high premium M&A and high failure rates for M&A. The reasons for these phenomena include, but are not limited to, insufficient understanding of the economics of corporate M&A and insufficient network econ- omy theory and applied research to clearly guide the practice of business combinations. There- fore, it is a great theoretical and practical significance to investigate the influence of network synergies generated by changes in the acquirers’ economic network before and after mergers and acquisitions on the realization of M&A motivations. The purpose of this paper is to examine whether network synergy has an impact on the degree of M&A motivation realization from the perspective of network node degree and strength change in M&A affairs occurring in China’s capital market. Its aims to provide theo- retical and empirical evidence on the rationality of the degree of M&A motivation as a crite- rion for determining the success or failure of M&A. (2) To analyze and explain the relative contribution of network synergy on the degree of M&A motivation. (3) To guide M&A prac- tice and provide useful theoretical support to further regulate corporate M&A behavior. The remainder of this paper are organized as follows: Section 2 conducts theoretical analy- sis and develops our research hypothesis. Section 3 describes our empirical models. we present the date source, sample selection, and the main empirical results of the simulations and give PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 2 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? some discussions in the fourth section. Finally, we show the relevant conclusions and the out- look of this work in the fifth section. 2 Methodology and hypothesis development 2.1 Economic networks and network synergies For Chinese businesses operating in a transitional economy and a traditional relational society, the informal system of social networks is critical for corporate governance and resource alloca- tion. The application of social networks in enterprises promotes the establishment and devel- opment of economic networks. Through interconnection, businesses build networks to acquire scarce resources such as knowledge and information through a collection of mutual relationships [12, 13]. The capacity of organizations to access network resources is determined by their positions in the network. Firms are driven to move in the network to find more resources in new and suitable places as part of the growth process, which promotes the evolu- tion of economic networks in terms of structure and economic strength. According to Burt [14], network centrality and structural holes influence a firm’s position in the network. Firms with higher network centrality also have higher authority and reputation, and firms occupying structural hole positions tend to have rich, diverse, and differentiated resources such as knowl- edge and information. In addition, the more important the network position is, the higher the level of business credit financing of the firm [15]. With the development of science and tech- nology as well as the intensification of competition, economic network relationships have become increasingly complex. Thus, the node degree and strength have become the main indi- cators to measure the features of complex networks. The node degree measures the position of nodes in the network, which reflects the ability of nodes to obtain resources in the network and the breadth of cooperation between subjects. Usually, the larger the node degree is, the stronger its "intermediary role" in connecting other nodes in the network and the more obvi- ous its role in the whole network. And the strength reflects how closely the node cooperates with other nodes in the network, which reflects the depth of cooperation between subjects. Furthermore, the topology of economic networks affects the efficiency of resource flow between nodes. As the topology of the economic network changes, the degree and strength of nodes in the economic network will also change, and the non-uniform network characteristics of the economic network cause resources to flow to high-dimensional nodes, which promotes the establishment of new economic links between nodes and nodes to generate a network economy and improves the enterprise competitiveness through network synergy capabilities [16]. 2.2 Network synergy and M&A motivation realization Mergers and acquisitions generate network synergies by changing the topology of economic networks. The synergistic effect of M&A is considered to be the "1+1>2" effect of the matching relationship between the merging parties. In other words, the overall performance of the group of firms is better than the simple sum of the performance of the individual parts. To achieve synergies in a business acquisition, the acquirer must either be able to limit the threats of existing and potential competitors in input markets, production processes, or output mar- kets; or it must be able to develop new markets or encroach on competitors’ markets, render- ing competitors unable to respond [17]. M&A is the merging and reorganization of two corporate economic systems, and the development of M&A can result in changes in the corpo- rate economic network’s structure both before and after the merger. By combining the nodes of the acquirer and the acquiree, the inter-organizational network can be significantly reshaped, allowing the acquirer to gain control of the acquiree’s nodes, resulting in a more PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 3 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? favorable structure for the post-merger company, thus creating synergies [18]. The network synergy of M&A is represented in the closeness of nodes in the network and the efficiency of network organizations. Specifically, the closeness of nodes is reflected by comparing firms’ internal and external transaction costs in the form of equity strength and contractual strength. And the efficiency of network organizations is reflected by economies of scale, economies of scope, and the effectiveness of network operations brought by the resources of individual nodes in the network. The network synergies generated by M&A can have a significant impact on the realization of M&A motives. In the context of the "new normal" of China’s economy, explicit M&A moti- vation is crucial for firms to acquire innovation capabilities, which directly affects the realiza- tion of M&A strategies [19]. Currently, a rising number of scholars tend to use the medium- to long-term test of M&A performance to determine whether M&A motives are realized. M&A implies the collapse or fusion of pre-merger nodes and the acquirer’s ability to legally inherit and control the acquiree’s relationships, all of which have a significant impact on the acquirer’s structural position, particularly if the acquirer is looking for ways to improve its position. According to Sun et al. [20], the network location centrality, network location structure hole, and network capacity are the primary parameters to determine the network power. The net- work synergy of M&A promotes the optimization of overall resource allocation efficiency by influencing the position of nodes in the network so that enterprises could receive economic benefits from cost reduction and specialized division of labor to reduce various uncertainties in daily business activities [21]. Following the M&A, both parties’ internal resources are reallo- cated according to corporate strategy, and both parties’ relationship networks are reshaped according to strategic needs to achieve a superior network position and generate value and benefits for the combined economic system. Specifically, on the one hand, the change in the enterprise’s position in the network generates a structural advantage in terms of the increasing efficiency of the enterprise’s information acquisition and resource allocation; On another hand, the enterprise’s overall structure of the external economic network changes, and the sta- bility of the whole enterprise network is enhanced, which means that the risk resistance of enterprise is further improved, leading to improved post-merger business performance. Based on the discussion above, the first hypothesis of this paper is stated as follows: Hypothesis 1: The network synergy generated by M&A can improve the performance of the firm. The network synergy generated by M&A can produce a "resource effect" that will enhance an enterprise’s resource endowment and ability to utilize resources, generate economies of scale and scope, and reduce internal and external transaction costs. According to resource- based theory, a company will obtain more resources and capabilities if it has a unique network structure. The subjects in the nodes of the relationship network, which are also called "social knots" in the network, can use the linkage paths in the network to exchange and disseminate information. The closer to the center of the network, the more useful information can be accessed, and the more breadth and depth of information can be obtained [22]. Therefore, acquirers seek to collaborate with enterprises with a distinctive and potentially economically advantageous network structure in their local area, either as partners or subsidiaries [23]. Moreover, if there are highly similar resources between firms, they will have an opportunity to benefit from economies of scale [24]. More concretely, firms can share R&D, technologies, procurement, production, operations, marketing plans, distribution channels, unified manage- ment, and sales forces to achieve economies of scale [25]. Furthermore, the advantages of net- work organization are that M&A can encourage specialized division and collaboration of labor among different firms and coordinate and combine strengths and specialties while eliminating PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 4 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? their disadvantages and deficiencies and increasing marginal rewards, which creates econo- mies of scale for the network organization as a whole [26]. Thus, hypothesis 2–1 is: Hypothesis 2–1: The network synergies generated by M&A can enhance economies of scale. By establishing inter-firm connections, firms can break down information barriers and effectively help them gain access to a variety of complementary resources that are not available internally [27–29]. Complementary resources provide opportunities for firms to gain econo- mies of scope, create synergies and develop new resources and corresponding skills [24]. Dong [30] argued that communication and interaction between nodes in a network are complex and various in form, containing both feedback of knowledge, information, material, and energy and sharing of technology, information, management, and experience, forming a beneficial sit- uation of complementary resources. Thereby, the relationship between nodes in the network will be strengthened in the process of cooperation and interaction, as well as the ability of the network to create value will be improved. M&A can rapidly change the existing network topol- ogy structure, which allows companies to acquire and utilize key complementary resources outside the enterprise, and enables them to operate a variety of different but related products at the same time. The stronger the complementarity between the products of the two parties, the smaller the differences in the requirements for technical equipment, management, and the quality of the personnel involved. With the increasing diversification of product categories, companies can achieve economies of scope and synergy effects, leading to improved resource utilization efficiency and enhanced competitive advantages. Thus, hypothesis 2–2 is: Hypothesis 2–2: The network economy generated by M&A can enhance the economy of scope. According to institutional economics, enterprise networks, alongside market and bureau- cracy, is one way of resource allocation. The enterprise network is comprised of nodal firms that are linked by complementary strengths and resource sharing [31–33]. Through the inti- mate interaction of nodes in an enterprise economic network, nodes can receive resources such as information and knowledge from the external environment [34]. In the fluctuating and complex network environment, the company uses the effectiveness of relationships in its position to cross the barriers of ineffective relationships, thereby establishing a "build-govern" network position to achieve innovative development [35–37]. Networks are self-organizing alternatives to markets, which help to improve the efficiency of dedicated resources, at the same time, the trust mechanisms in networks can also help to save transaction costs [38]. Spe- cifically, M&A can significantly reshape inter-organizational networks and accomplish net- work synergy by changing the association between nodes in the network to accelerate the flow of information and reduce the accessing cost of resources within the organization. Thus, com- panies with strong ties in the economic network can rapidly acquire resources such as infor- mation and knowledge of other companies. The closer a company is to the center of the economic network, the more information transmission channels it will have, the faster the speed of obtaining resources, the stronger the bargaining power with partners, and the lower the transaction cost. Based on the statements above, hypothesis 2–3 is: Hypothesis 2–3: The network synergies generated by M&A can reduce transaction costs. 3 Research design 3.1 Sample selection and data This paper selects listed companies in Shanghai and Shenzhen A-shares that underwent merg- ers and acquisitions from 2007–2019 as the research sample, and the required data are PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 5 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? obtained from the Guotaian database (CSMAR) and the Wind database (Wind). The selection criteria are as followed: (1) exclude M&A companies treated by ST, ST, and PT; (2) exclude listed companies in the financial and insurance industries; (3) exclude samples of asset divesti- ture, asset replacement, debt restructuring, share repurchase and other types of restructuring; (4) retain the sample of M&A companies whose acquirers are listed companies and select only those with the M&A progress mark "completed "; (5) exclude samples with M&A amounts less than 1 million yuan; (6) for the same company with multiple M&As completed in the same year, retaining the first M&A event completed by the company in that year to reduce the inter- action between different M&A events; (7) exclude samples with incomplete key data; (8) Win- sorize the tails of all continuous variables at the 1% and 99% quantiles to avoid the adverse effects of some samples with extreme values on the empirical results. After the above screening, a final sample of 2201 M&A is obtained. 3.2 Model specification and key variables The main regression test models are: Realization ¼ a0 þ a1DN þ a2Control þ Realization ¼ a0 þ a1Ds þ a2Control þ X Industry þ X Year þ ε X Industry þ X Year þ ε ð1Þ ð2Þ In the model, Realization is the dependent variable that measures the degree of realization of the M&A motives, and ΔROE is used to examine the change in performance of the acquirer after the M&A. The difference between the acquirer’s ROE value in the year following the completion of the M&A (the first and second years) and the acquirer’s ROE value in the year preceding the completion of the M&A is used to determine the degree of realization of the M&A motive. The variation of node degree (N) and strength (S), referred to ΔN and ΔS, in the internal and external networks of listed companies are proxy variables for network synergy. The node degree (N) refers to the sum of the number of nodes connected to a node, which reflects the node’s ability to obtain resources in the network. In general, the higher the node degree is, the stronger the "intermediary role" of the node in connecting other nodes in the complex network. The calculation of node degree is Ni = ∑j2k aij. Therefore, the calculation of change in node degree ΔN before and after the merger is: ΔN = (node degree of listed companies in the year of M&A completion—node degree of listed companies in the year before M&A completion) /nodal degree of listed companies in the year before M&A completion. The node strength (S) refers to the sum of all edge weights connected to the nodes, reflect- ing the closeness of cooperation with other subjects, and is an important characteristic variable for establishing cooperation and communication between enterprises and other network sub- jects [39–41]. The strength of network relationships is directly related to the ability of subjects to obtain various types of heterogeneous resources such as technology and knowledge. In gen- eral, the higher the node degree, the higher the "status" of the node in the network. The calculation of node strength is Si = ∑j2k wij. Therefore, the calculation of change in node strength ΔS before and after the merger is: ΔS = (node strength of listed companies in the year of M&A completion–node strength of listed companies in the year before M&A completion) /node strength of listed companies in the year before M&A completion PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 6 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? The reason for using proportional calculation is that the change in the acquirer’s position in the network depends on its initial position. For example, a decrease in node degree of 0.25 is more meaningful for a company with an initial node degree of 0.5 (50% change) than for a company with an initial node degree of 0.75 (33% change). This paper constructs an equity network formed by a listed company and its subsidiaries, so the node degree is the number of the listed company’s subsidiaries and the node strength is the sum of the listed company’s shareholdings in subsidiaries. Referring to the studies of Chen et al. [42], Liu et al. [43], and Pan et al. [44], the following variables are controlled in the regression model: firm size (Size), firm solvency (Lev), cash flow (OCF), the proportion of independent directors (Indep), board size (Board), dual position sta- tus (Dual), first largest shareholder (Top1), related transactions (Relevance), type of M&A tar- get (Type), nature of ownership of the M&A party (SOE), industry (Industry) and year (Year). The definition of these control variables is shown in Table 1. 4 Empirical results 4.1 Descriptive statistics and correlation analysis Panel A in Table 2 illustrates the M&A transaction characteristics of the sample. following Panel A, cash payment M&As accounted for 63.83%, and non-cash payment M&As accounted for 36.17%, indicating that cash payment is the main payment method in the Chinese M&A market; related-party M&As accounted for 35.85%, and non-related M&As accounted for 64.15%. M&As of asset restructuring accounted for 28.62%, and M&As of non-major asset restructuring accounted for 71.38%; The proportion of equity bids in M&A targets reached Table 1. Definition and interpretation of each variable. Dependent variables Independent variables Variable Name Degree of realization of M&A motives Variation of node degree Variation of node strength Control variables Company Size Corporate solvency Cash Flow Percentage of independent directors Board Size Dual job situation Percentage of shareholding of the largest shareholder Variable Symbols Realization ΔN ΔS Size Lev OCF Indep Board Dual Top1 Variable Explanation The difference between the acquirer’s ROE in the year after the completion of the acquisition (year 1 / year 2) and the ROE in the year before the completion of the acquisition (Nodal degree of listed companies in the year of M&A completion—Nodal degree of listed companies in the year before M&A completion)/Nodal degree of listed companies in the year before M&A completion (Strength of listed companies in the year of M&A completion—Strength of listed companies in the year before M&A completion)/ Strength of listed companies in the year before M&A completion Logarithm of total assets at the end of the year Total liabilities / total assets Net cash flow from operating activities/total assets Number of independent directors/number of directors Number of Board of Directors The chairman and general manager both take1, otherwise take0 Number of shares held by the company’s largest shareholder/total number of shares of the company Related Transactions Relevance The M&A related M&Athat takes the value of 1, otherwise it is 0 Types of M&A subjects Nature of property rights of mergers and acquisitions Industry Annual Type SOE Industry Year Equity subjects are taken1, non-equity subjects are taken 0 The merging party is a state-owned holding company that takes the value of 1, otherwise, it is 0 According to the industry sector in which the acquirer is in (2012 year CSRC industry classification standard) Year of Corporate M&A https://doi.org/10.1371/journal.pone.0284204.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 7 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? Table 2. Descriptive statistics of the sample. M&A Payment Method Form of M&A transaction Panel A: M&A transaction characteristics of the sample Unrelated Transactions Whether major asset restructuring Major Asset Restructuring Non-major asset restructuring Cash payment Number of samples 1405 Non-Cash Payments 796 Related Transactions 789 Percentage (%) 63.83% 36.17% 35.85% Subject Classification Equity underlying 2089 Non-Equity Underlying 112 Number of samples 228 Percentage (%) 94.91% 5.09% 10.36% Horizontal M&A Vertical M&A Hybrid M&A 1412 64.15% M&A Type 630 28.62% 1204 54.70% 650 29.53% Stock Market Industry Panel B: Sample Stock Market and Industry Distribution Mainboard SMB and GEM Manufacturing Information transmission, software and information technology services Wholesale and retail trade Number of samples 869 1332 1470 Percentage (%) 39.48% 60.52% 66.79% https://doi.org/10.1371/journal.pone.0284204.t002 196 8.91% 99 4.50% 1571 71.38% Other 119 5.41% Other 436 19.81% 94.91%, and the proportion of non-equity bids was only 5.09%; among the types of M&As, horizontal mergers, and acquisitions accounted for 10.36%, vertical mergers and acquisitions accounted for 54.70%, accounting for more than half of the total sample, mixed mergers accounted for 29.53%, and other types of mergers accounted for 5.41%. Panel B illustrates the stock market and industry distribution of the sample. Based on Table 2, M&A in the main board market accounted for 39.48%, and M&A in the SME board and GEM board accounted for 60.52%; the manufacturing industry had the largest sample with 66.79%, M&A in the infor- mation transmission, software, and information technology service industry accounted for 8.91%, and M&A in the wholesale and retail industry accounted for 4.50%. Table 3 shows the descriptive statistics of the main variables in the model. The results show that in the equity network, the means of the degree of realization of M&A motivation in the year of M&A completion, one year after completion, and two years after completion are -0.0004, -0.013, and -0.037 respectively, which indicated that the M&A motives of the compa- nies in the sample are generally not realized, and the M&A has not improved the performance level of the enterprises. The mean of the changes in the node degree of listed companies is 0.6890, indicating that the number of subsidiaries of listed companies after the completion of M&A is greater than the number of subsidiaries of listed companies before M&A. The mean of the changes in the strength of listed companies is 0.7190, indicating that the strength of control of listed companies over subsidiaries after the completion of M&A is greater than the strength of control of listed companies over subsidiaries before M&A. Table 4 shows the correlation analysis of the main variables. Overall, there is no serious col- linearity problem among the control variables. 4.2 Main regression results This section examines the effects of changes in the node degree and strength of the firm’s inter- nal network on the degree of realization of the M&A motives. The regression results for the year in which the M&A was completed are shown in Table 5. Firstly, the effects of changes in PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 8 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? Table 3. Descriptive statistics of the main variables. Realizationt Realizationt+1 Realizationt+2 ΔN ΔS Size Lev OCF Indep Board Dual Top1 Relevance Type SOE N 2201 2201 2201 2201 2201 2201 2201 2201 2201 2201 2201 2201 2201 2201 2201 Mean -0.0004 -0.0130 -0.0370 0.6888 0.7192 22.1280 0.4062 0.0406 0.3757 8.4607 0.3012 Variance Minimum value 1/4 Median 0.0784 0.1230 0.1790 1.1329 1.2625 1.0826 0.1899 0.0628 0.0545 1.5369 0.4589 -0.3314 -0.6680 -1.0360 -0.4545 -0.4986 20.1370 0.0632 -0.1436 0.3333 5 0 -0.0290 -0.0400 -0.0550 0.1190 0.1124 21.3640 0.2535 0.0036 0.3333 7 0 Median -0.0016 -0.0020 -0.0090 0.3333 0.3329 21.9780 0.3945 0.0392 0.3333 9 0 3/4 Median Maximum value 0.0243 0.0310 0.0310 0.7778 0.7917 22.7400 0.5469 0.0775 0.4286 9 1 0.3279 0.3600 0.3630 7 8 25.5390 0.8424 0.2081 0.5714 13 1 34.0950 14.8260 8.5400 22.3000 32.1900 43.6400 74.9600 0.3585 0.9491 0.2622 0.4797 0.2198 0.4399 0 0 0 0 1 0 0 1 0 1 1 1 1 1 1 https://doi.org/10.1371/journal.pone.0284204.t003 node degree ΔN and changes in strength ΔS of listed companies on the degree of realization of corporate M&A motives are examined individually, shown in columns (1) and (5). Secondly, the effects of changes in node degree ΔN and changes in strength ΔS of listed companies on the degree of realization of corporate M&A motives are examined jointly with the control vari- ables, shown in columns (2) and (6). Thirdly, the effects are examined again by controlling Industry and Year, shown in columns (3) and (7). The last one examines the effects of adopting cluster analysis on Industry, shown in columns (4) and (8). The results show that the change in node degree (ΔN) and the change in node strength (ΔS) have a significant positive effect on the degree of realization of corporate M&A motivation (Realizationt) in the year in which the M&A is completed. The regression results for one year after the completion of M&A are presented in Table 6. The results show that after controlling for year and industry and clustering the industry, the Lev OCF Indep Board Dual Top1 Relevance Type SOE 1 -0.103 -0.019 0.142* -0.120* 0.116* 0.167* -0.017 0.291* 1 -0.034 0.069* -0.019 0.147* 0.034 -0.019 0.041* 1 -0.559* 0.079* 0.029 -0.042 -0.001 -0.052 1 -0.160* 0.025 0.110* -0.041* 0.276*** 1 -0.035* -0.123* 0.026 -0.247* 1 0.128* -0.111* 0.222* 1 -0.055* 0.283* 1 -0.121* 1 Table 4. Correlation analysis. Size Lev OCF Indep Board Dual Top1 Relevance Type SOE Size 1 0.543* 0.005 -0.029 0.256* -0.189* 0.138* 0.250* -0.046 0.392*** Note * indicates P < 0.1 indicates P < 0.05 * indicates P < 0.01. https://doi.org/10.1371/journal.pone.0284204.t004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 9 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? Table 5. Multivariate regression results of the variation of node degree and strength within listed companies and Realizationt. (1) Single 0.0050* (3.40) -0.0039 (-1.98) 0.0052 2201 (2) more 0.0052* (3.49) 0.0012 (0.62) 0.0303* (2.87) 0.1194* (4.41) 0.0126 (0.34) -0.0008 (-0.57) 0.0033 (0.88) -0.0000 (-0.29) 0.0045 (1.22) 0.0065 (0.84) -0.0001 (-0.03) -0.0536 (-1.20) 0.0210 2201 (3) Control 0.0052* (3.47) -0.0001 (-0.07) 0.0263 (2.34) 0.1268* (4.56) 0.0170 (0.45) -0.0003 (-0.19) 0.0031 (0.83) 0.0000 (0.19) 0.0038 (1.03) 0.0023 (0.30) 0.0040 (0.85) Control Control -0.1243 (-2.49) 0.0465 2201 ΔN ΔS Size Lev OCF Indep Board Dual Top1 Relevance Type SOE Year Industry _cons R2 N (4) Industry Clustering 0.0052* (4.00) (5) Single (6) more (7) (8) Control Industry Clustering 0.0050* 0.0051* 0.0048* (3.82) -0.0040 (-2.11) 0.0066 2201 (3.79) 0.0011 (0.57) 0.0302* (2.86) 0.1181* (4.37) 0.0137 (0.37) -0.0008 (-0.54) 0.0032 (0.86) -0.0000 (-0.32) 0.0044 (1.18) 0.0066 (0.87) -0.0002 (-0.04) -0.0523 (-1.17) 0.0220 2201 (3.57) -0.0003 (-0.13) 0.0264 (2.34) 0.1255* (4.51) 0.0175 (0.47) -0.0002 (-0.16) 0.0030 (0.80) 0.0000 (0.18) 0.0038 (1.03) 0.0024 (0.31) 0.0039 (0.83) Control Control -0.1205 (-2.42) 0.0468 2201 0.0048* (3.64) -0.0003 (-0.16) 0.0264 (2.19) 0.1255* (2.99) 0.0175 (0.50) -0.0002 (-0.15) 0.0030* (1.97) 0.0000 (0.27) 0.0038 (0.81) 0.0024 (0.40) 0.0039* (1.92) Control Control -0.1205* (-2.02) 0.0468 2201 -0.0001 (-0.08) 0.0263 (2.17) 0.1268* (3.01) 0.0170 (0.47) -0.0003 (-0.18) 0.0031** (2.15) 0.0000 (0.30) 0.0038 (0.81) 0.0023 (0.39) 0.0040* (1.94) Control Control -0.1243* (-2.07) 0.0465 2201 Note , , contribute to significant levels of 10%, 5%, and 1% respectively. https://doi.org/10.1371/journal.pone.0284204.t005 change in node degree of listed companies (ΔN) has a significant positive effect on the degree of realization of corporate M&A motivation (Realizationt+1) one year after M&A completion. Under the univariate tests, the change in node strength (ΔS) of listed companies positively affects the degree of realization of corporate M&A motivation (Realizationt+1) one year after M&A completion, and it is statistically significant at the 10% level of significance. After con- trolling for year and industry and clustering the industries, the change in node strength (ΔS) of listed companies still has a positive effect on the degree of realization of corporate M&A moti- vation (Realizationt+1) one year after M&A completion, and it is also statistically significant at the 10% level of significance. The regression results for 2 years after the completion of M&A is presented in Table 7. The results show that the change in the node degree of listed companies (ΔN) and the change in node strength (ΔS) have a negative but insignificant effect on the degree of realization of PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 10 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? Table 6. Multivariate regression results of the variation of node degree and strength of listed companies with Realizationt+1. (1) Single 0.0034 (1.48) (2) more 0.0033 (1.42) (3) Control 0.0036 (1.53) (4) Industry Clustering 0.0036 (2.28) (5) Single (6) more (7) (8) Control Industry Clustering -0.0009 (-0.28) 0.0030 (0.18) 0.1906* -0.0015 (-0.44) -0.0059 (-0.34) 0.1856* (4.48) -0.0044 (-0.07) 0.0023 (1.07) 0.0048 (0.81) 0.0001 (0.45) 0.0101* (1.75) 0.0011 (0.09) 0.0039 (0.56) -0.0329 (-0.47) 0.0148 2201 (4.26) -0.0093 (-0.16) 0.0025 (1.15) 0.0052 (0.89) 0.0001 (0.53) 0.0085 (1.46) -0.0040 (-0.34) 0.0075 (1.03) Control Control -0.0584 (-0.75) 0.0460 2201 -0.0015 (-0.55) -0.0059 (-0.29) 0.1856* (4.12) -0.0093 (-0.33) 0.0025 (0.90) 0.0052 (1.58) 0.0001 (0.46) 0.0085* (3.42) -0.0040 (-0.45) 0.0075 (1.35) Control Control -0.0584 (-0.78) 0.0460 2201 -0.0151*** (-4.92) 0.0010 2201 0.0037* (1.77) 0.0034 (1.61) -0.0009 (-0.30) 0.0029 0.0029 (1.41) -0.0016 (-0.47) -0.0059 (0.18) 0.1899* (-0.33) 0.1845* (4.47) -0.0034 (-0.06) 0.0024 (1.08) 0.0047 (0.80) 0.0001 (0.44) 0.0100* (1.72) 0.0012 (0.10) 0.0040 (0.57) -0.0323 (-0.46) 0.0151 2201 (4.24) -0.0095 (-0.16) 0.0026 (1.16) 0.0051 (0.87) 0.0001 (0.54) 0.0086 (1.48) -0.0040 (-0.33) 0.0074 (1.01) Control Control -0.0553 (-0.71) 0.0458 2201 -0.0153*** (-5.10) 0.0014 2201 0.0029* (1.82) -0.0016 (-0.59) -0.0059 (-0.28) 0.1845* (4.10) -0.0095 (-0.34) 0.0026 (0.92) 0.0051 (1.53) 0.0001 (0.46) 0.0086* (3.49) -0.0040 (-0.45) 0.0074 (1.33) Control Control -0.0553 (-0.74) 0.0458 2201 ΔN ΔS Size Lev OCF Indep Board Dual Top1 Relevance Type SOE Year Industry _cons R2 N Note , , contribute to significant levels of 10%, 5%, and 1% respectively. https://doi.org/10.1371/journal.pone.0284204.t006 corporate M&A motivation (Realizationt+2) two years after the completion of M&A. The results for the year of M&A, the year after 1 year, and the year after 2 years show that the effect of network economic synergy becomes less pronounced as the integration of M&A takes lon- ger. Overall, combining the regression results of the year of M&A completion, one year after, and two years after, the internal network economy generated by M&A can improve M&A per- formance and promote the realization of corporate M&A motives, therefore, the results are consistent with the Hypothesis 1. 4.3 Heterogeneity tests 4.3.1 Group test by M&A payment method. Table 8 examines the impact of M&A pay- ment methods on corporate M&A performance. The results show that there is a significant PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 11 / 21 PLOS ONETable 7. Multivariate regression results of the variation of node degree and strength of listed companies with Realizationt+2. Do network synergies facilitates the realization of M&A motivation? (1) Single -0.0016 (-0.47) (2) more -0.0017 (-0.50) -0.0089 (-1.99) 0.0109 (0.45) 0.1701* (2.75) 0.0083 (0.10) 0.0007 (0.22) -0.0104 (-1.21) 0.0007* (2.77) 0.0038 (0.45) 0.0069 (0.39) 0.0251 (2.47) -0.0358* (-8.03) 0.0001 2201 0.1045 (1.02) 0.0154 2201 (3) Control -0.0020 (-0.59) -0.0064 (-1.36) 0.0017 (0.07) 0.1756* (2.79) -0.0062 (-0.07) -0.0004 (-0.13) -0.0079 (-0.93) 0.0006 (2.13) 0.0034 (0.40) 0.0001 (0.01) 0.0294* (2.79) Control Control 0.0190 (0.17) 0.0607 2201 ΔN ΔS Size Lev OCF Indep Board Dual Top1 Relevance Type SOE Year Industry _cons R2 N (4) Industry Clustering (5) Single (6) more (7) (8) Control Industry Clustering -0.0020 (-0.53) -0.0064 (-1.21) 0.0017 (0.06) 0.1756 (2.24) -0.0062 (-0.11) -0.0004 (-0.14) -0.0079 (-1.25) 0.0006 (2.43) 0.0034 (0.35) 0.0001 (0.01) 0.0294 (2.46) Control Control 0.0190 (0.18) 0.0607 2201 -0.0024 (-0.79) -0.0027 (-0.89) -0.0089 (-1.98) 0.0108 (0.45) 0.1699* (2.75) 0.0060 (0.07) 0.0006 (0.20) -0.0104 (-1.21) 0.0007* (2.80) 0.0042 (0.50) 0.0067 (0.38) 0.0249 (2.45) -0.0352* (-8.03) 0.0003 2201 0.1059 (1.04) 0.0156 2201 -0.0027 (-0.88) -0.0064 (-1.35) 0.0017 (0.06) 0.1755* (2.79) -0.0080 (-0.09) -0.0005 (-0.14) -0.0079 (-0.93) 0.0006 (2.15) 0.0037 (0.44) -0.0000 (-0.00) 0.0293* (2.78) Control Control 0.0192 (0.17) 0.0609 2201 -0.0027 (-0.72) -0.0064 (-1.22) 0.0017 (0.06) 0.1755 (2.25) -0.0080 (-0.14) -0.0005 (-0.16) -0.0079 (-1.25) 0.0006 (2.49) 0.0037 (0.39) -0.0000 (-0.00) 0.0293** (2.47) Control Control 0.0192 (0.18) 0.0609 2201 Note , , contribute to significant levels of 10%, 5%, and 1% respectively. https://doi.org/10.1371/journal.pone.0284204.t007 difference in the impact of network synergy on the degree of M&A motivation realization under non-cash payment and cash payment, with the use of non-cash payment method (mainly reflected in share-based payment M&A), the change in nodal degree (ΔN) and the change in node strength (ΔS) significantly contribute to the degree of realization of corporate M&A motivation and are statistically significant at the 1% level. It indicates that the original shareholders of the target company will become strategic investors, which makes the benefit bundling mechanism between the M&A parties stronger and the network economy has a greater impact on the degree of M&A motivation realization. When cash payment is used, the change in node degree (ΔN) and the change in strength (ΔS) have a negative but statistically insignificant effect on the degree of realization of corporate M&A motivation. 4.3.2 Group test by whether it is a related transaction. The related transaction also impacts the M&A activities significantly. As shown in Table 9, the internal network changes PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 12 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? (5) Non-Cash (6) Cash (7) Non-Cash (8) Cash 0.0072* (3.97) -0.0006 (-0.28) -0.0010 (-0.28) 0.0194 796 -0.0043* (-1.88) 0.0001 1405 0.0067*** (3.57) 0.0001 (0.02) 0.0378* (1.74) 0.0959* (1.77) 0.0846 (1.31) -0.0002 (-0.10) 0.0059 (0.85) 0.0001 (0.23) 0.0067 (1.02) 0.0420*** (2.64) 0.0117 (1.44) Control Control -0.0911 (-0.70) 0.0778 796 -0.0012 (-0.58) -0.0006 (-0.24) 0.0245* (1.85) 0.1379*** (4.26) -0.0232 (-0.50) -0.0003 (-0.19) 0.0016 (0.36) 0.0000 (0.10) -0.0042 (-0.85) -0.0101 (-1.16) -0.0010 (-0.18) Control Control -0.0607 (-0.66) 0.0651 1405 (4) Cash -0.0001 (-0.06) -0.0006 (-0.23) 0.0243* (1.84) 0.1379* (4.26) -0.0230 (-0.50) -0.0004 (-0.20) 0.0016 (0.36) 0.0000 (0.08) -0.0041 (-0.82) -0.0100 (-1.14) -0.0008 (-0.14) Control Control -0.0618 (-0.68) 0.0649 1405 Table 8. Group test by M&A payment method. (1) Non-Cash 0.0072* (3.52) (2) Cash -0.0006 (-0.26) (3) Non-Cash 0.0069*** (3.23) -0.0000 (-0.00) 0.0381* (1.75) 0.0979* (1.81) 0.0799 (1.24) -0.0003 (-0.13) 0.0058 (0.84) 0.0001 (0.29) 0.0073 (1.11) 0.0428*** (2.68) 0.0120 (1.48) Control Control -0.0873 (-0.67) 0.0751 796 -0.0004 (-0.11) 0.0153 796 -0.0043* (-1.85) 0.0000 1405 ΔN ΔS Size Lev OCF Indep Board Dual Top1 Relevance Type SOE Year Industry _cons R2 N Note , , contribute to significant levels of 10%, 5%, and 1% respectively. https://doi.org/10.1371/journal.pone.0284204.t008 resulting from M&A in the sample of connected transactions significantly contribute to the degree of realization of corporate M&A motives, and the results are statistically significant. This indicates that there are fewer corporate cultural differences between the M&A subjects that have related transactions, which facilitates the smooth transfer of information between the acquiring parties and faster realization of M&A synergies to achieve the realization of M&A motives [45, 46]. However, M&A with unrelated transactions face a higher degree of informa- tion asymmetry, which makes it difficult to generate network synergies. 4.4 Robustness test 4.4.1 Changing the dependent variable. Under this test, the regressions are conducted using ΔROA and ROA in the year of M&A completion as substitute variables for the degree of realization of corporate M&A motives respectively, and the remaining variables are kept PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 13 / 21 PLOS ONETable 9. Group test by whether they are related transactions. Do network synergies facilitates the realization of M&A motivation? (1) (2) (3) (4) (5) (6) (7) (8) Non-associated -0.0017 (-0.81) Associations 0.0095* (4.44) Non-associated -0.0005 (-0.22) Associations 0.0098* (4.46) Non-associated Associations Non-associated Associations -0.0015 (-0.58) 0.0426* (3.14) 0.0939* (2.81) -0.0108 (-0.24) 0.0004 (0.26) 0.0057 (1.35) -0.0000 (-0.23) 0.0034 (0.34) 0.0003 (0.05) Control Control -0.0636 (-0.69) 0.0721 1412 0.0002 (0.06) -0.0005 (-0.03) 0.1708* (3.46) 0.0612 (0.94) -0.0005 (-0.20) -0.0006 (-0.08) 0.0001 (0.52) 0.0048 (0.38) 0.0063 (0.82) Control Control -0.0437 (-0.43) 0.0723 789 -0.0019 (-1.00) 0.0094* (4.99) -0.0021 (-0.93) 0.0007 1412 -0.0039 (-1.13) 0.0307 789 -0.0018 (-0.94) -0.0016 (-0.64) 0.0427* (3.15) 0.0939* (2.81) -0.0122 (-0.27) 0.0004 (0.26) 0.0057 (1.36) -0.0000 (-0.19) 0.0031 (0.32) 0.0001 (0.02) Control Control -0.0593 (-0.64) 0.0727 1412 0.0096* (4.97) 0.0002 (0.06) 0.0003 (0.01) 0.1690*** (3.44) 0.0623 (0.96) -0.0006 (-0.23) -0.0007 (-0.09) 0.0001 (0.48) 0.0044 (0.35) 0.0060 (0.79) Control Control -0.0428 (-0.43) 0.0780 789 -0.0023 (-0.99) 0.0005 1412 -0.0035 (-0.98) 0.0245 789 ΔN ΔS Size Lev OCF Indep Board Dual Top1 Type SOE Year Industry _cons R2 N Note , , contribute to significant levels of 10%, 5%, and 1% respectively. https://doi.org/10.1371/journal.pone.0284204.t009 constant. The regression results are shown in Appendix 1, 2 in S1 Appendix (Robustness test), and the results are still consistent with the hypothesis. 4.4.2 Adding new control variables. The control variables of M&A transaction size and payment method are added to test the effects of changes in the node degree (ΔN) and changes in node strength (ΔS) of listed companies on the degree of realization of corporate M&A motives. The regression results are shown in Appendix 3 in S1 Appendix (Robustness test), and the results still support the hypothesis. 5 Supplementary test To test the impacts of network synergy effects on economies of scale, economies of scope, and transaction costs due to changes in economic network nodes, models (3)-(8) are constructed to test hypotheses H2-1 - H2-3: Scale ¼ b0 þ aDN þ bControl þ X Industry þ X Year þ ε1 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 ð3Þ 14 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? Scale ¼ b0 þ aDS þ bControl þ X Industry þ X Year þ ε2 Scpoe ¼ b0 þ aDN þ bControl þ X Industry þ X Year þ ε3 Scpoe ¼ b0 þ aDS þ bControl þ Cost ¼ b0 þ aDN þ bControl þ X X Industry þ Industry þ X X Year þ ε4 Year þ ε5 Cost ¼ b0 þ aDS þ bControl þ X Industry þ X Year þ ε6 ð4Þ ð5Þ ð6Þ ð7Þ ð8Þ 5.1 Measurement of economies of scale For the measure of economies of scale (Scale), referring to the study of Zhang and He [47], after M&A, the acquirer and the acquiree become one enterprise, and the two supplier rela- tionship networks are synthesized into one supplier relationship network. Facing the same suppliers before, the acquirer has more voice after M&A, and the original suppliers are more willing to establish stronger cooperative relationships with the enterprise. The enterprise’s bar- gaining power over suppliers is enhanced, and its reliance on the top few suppliers is reduced, thus reducing procurement costs and effectively reducing the risk of market fluctuations. Therefore, the change in the ratio of the top five suppliers’ procurement amount to the com- pany’s annual procurement amount before and after the M&A is used to measure the economy of scale, which is calculated as follows: Scale = Proportion of the annual procurement amount of the top five suppliers in the year of completion of the merger and acquisition—Proportion of the annual procurement amount of the top five suppliers in the year before the completion of the merger and acquisition If the result is positive, it indicates that the dependence of the company on the supplier increases after the completion of the M&A. If the result is negative, it indicates that the depen- dence of the company on the supplier decreases, and the scale effect increases. The greater the decrease in this value, the greater the scale effect. 5.2 Measurement of economies of scope The economy of scope (Scope) in this paper is measured by using the revenue Herfindahl index and the revenue entropy index, which are widely used in the existing literature. The entropy index is calculated based on the proportion of revenue from each industry to total rev- enue based on the “Industry Classification Guidelines for Listed Companies revised by the China Securities Regulatory Commission” in 2012, the Wind database, and the segment reporting information disclosed in the annual reports of enterprises. The income Herfindahl index refers to the squared sum of the proportion of each industry’s income to the total income of the enterprise, and the larger this index is, the less diversified the enterprise is. Therefore, this study uses the change in entropy index and Herfindahl index before and after the M&A, which is calculated as follows: Scope1 = Entropy index in the year of M&A completion -Entropy index in the year before M&A completion Scope2 = Herfindahl index in the year of M&A completion—Herfindahl index in the year before M&A completion PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 15 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? 5.3 Measurement of transaction costs For the measurement of transaction cost savings, Ang et al. [48] used the overhead cost indica- tor; and Xia and Liu [49] used the sum of selling expenses, administrative expenses, and finan- cial expenses as a percentage of total assets to measure for a more comprehensive exploration of the components of transaction costs and to eliminate the interference of firm size. In this paper, regarding the above studies, the change in the sum of selling expenses, administrative expenses, and financial expenses as a percentage of total assets is used to measure transaction costs. The calculation is as follows: Cost = Selling, administrative and financial expenses as a percentage of total assets in the year in which the acquisition is completed—Selling, administrative and financial expenses as a percentage of total assets in the year before the completion of the acquisition The regression results of the effects of changes in node degree (ΔN) and changes in node strength (ΔS) on economies of scale, economies of scope, and transaction costs in the eco- nomic network of listed companies are shown in Table 10. Columns (1), (3), (5), and (7) show that the change in the node degree (ΔN) of listed com- panies significantly impacts the economies of scale, economies of scope, and transaction costs, where the change in the node degree (ΔN) of listed companies promotes economies of scale and scope and reduces transaction costs. On other hand, Columns (2), (4), (6), and (8) show that the change in node strength (ΔS) of listed companies promotes economies of scale and scope but has no significant effect on transaction costs. 6 Conclusion This paper constructs an equity network between listed companies and their subsidiaries, the- oretically, analyses, and empirically tests the impact of network synergy on the degree of reali- zation of corporate M&A motives, and draws the following conclusions using M&A activities in Shanghai and Shenzhen A-shares as a research sample from 2007 to 2019. (1) Mergers and acquisitions, as a method of merging and reshaping two economic networks, can lead to changes in the degree and strength of internal and external network nodes of enterprises. Gen- erally, the greater the changes, the greater the network synergy effect it generates, which is con- ducive to promoting the realization of enterprises’ M&A motives. In non-cash payment methods and related M&A, changes in the degree and strength of network nodes have more significant effects on the degree of realization of corporate M&A motives. The network syn- ergy generated by M&A can significantly enhance M&A performance, generate economies of scale and scope, and save transaction costs. This paper provides a novel explanation for the frontier issue of the paradox of M&A’s "high failure rate" and increasing activity. By organically linking M&A motives and M&A net- work synergies to explain M&A behavior and effects, it provides new evidence for established studies to judge M&A success or failure in terms of whether M&A motives are realized. More- over, it enriches the existing research on factors influencing the realization degree of M&A motivation. Based on the changes in the nodal degree and strength of corporate economic net- works, a complex systems approach is used to construct a model to examine the relative contri- bution of M&A synergies on the realization degree of M&A motivation, forming a useful addition to the existing literature. The findings of this study helps enterprises planning to implement M&A strategies to clar- ify their M&A motives, make rational judgments on network synergies, scientifically carry out M&A valuation and avoid excessive M&A premiums. Meanwhile, it helps the relevant regula- tory authorities to formulate policies for further regulating the M&A behavior and disclosing M&A-related information of listed companies. Additionally, the findings are conducive to the PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 16 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? Table 10. Regression results of network economies and economies of scale, economies of scope, and transaction cost. (2) Scale (1) Scale -1.6672* (-5.49) (3) Scope1 0.0330* (4.04) (4) Scope1 (5) Scope2 -0.0197* (-4.20) (6) Scope2 -1.4959* 0.0207* -0.0123*** 0.2402 (0.53) 4.1594* (1.76) -0.9597 (-0.16) -7.9010 (-0.98) -0.3720 (-1.19) -0.0439 (-0.06) -0.0070 (-0.28) 0.3768 (0.50) 2.5387 (1.45) -1.8749* (-1.90) Control Control -1.5238 (-0.12) 0.0411 1.7976 1503 (-5.48) 0.2840 (0.63) 4.2400* (1.79) -0.6823 (-0.11) -7.8325 (-0.97) -0.3823 (-1.22) -0.0215 (-0.03) -0.0066 (-0.26) 0.3841 (0.51) 2.6062 (1.49) -1.8548* (-1.88) Control Control -2.6205 (-0.21) 0.0411 1.7952 1503 -0.0028 (-0.30) -0.0160 (-0.31) -0.0641 (-0.52) 0.0228 (0.14) -0.0022 (-0.37) 0.0242 (1.38) 0.0001 (0.10) 0.0122 (0.74) 0.0393 (1.12) -0.0006 (-0.03) Control Control -0.0609 (-0.18) 0.0507 1.9158 1364 (3.00) -0.0038 (-0.41) -0.0158 (-0.31) -0.0856 (-0.70) 0.0091 (0.05) -0.0023 (-0.37) 0.0221 (1.26) 0.0001 (0.19) 0.0163 (0.99) 0.0395 (1.12) -0.0024 (-0.12) Control Control -0.0335 (-0.10) 0.0455 1.7091 1364 0.0015 (0.27) 0.0100 (0.34) 0.0254 (0.36) -0.0082 (-0.08) 0.0003 (0.09) -0.0151 (-1.50) -0.0001 (-0.22) -0.0035 (-0.37) -0.0191 (-0.94) 0.0031 (0.27) Control Control 0.0397 (0.20) 0.0505 1.9077 1364 (-4.20) 0.0021 (0.38) 0.0099 (0.34) 0.0383 (0.54) 0.0002 (0.00) 0.0004 (0.10) -0.0139 (-1.38) -0.0001 (-0.31) -0.0060 (-0.63) -0.0192 (-0.95) 0.0042 (0.36) Control Control 0.0233 (0.12) 0.0448 1.6816 1364 (7) Cost -0.0009** (-1.99) -0.0020* (-1.90) 0.0128 (2.27) 0.0301 (2.25) 0.0165 (0.86) 0.0006 (0.90) 0.0007 (0.37) 0.0001* (8) Cost 0.0000 (0.23) -0.0021* (-1.94) 0.0126 (2.24) 0.0302 (2.25) 0.0173 (0.90) 0.0007 (0.96) 0.0005 (0.26) 0.0001* (1.72) -0.0049* (1.74) -0.0052* (-2.63) -0.0031 (-0.83) 0.0012 (0.54) Control Control -0.0066 (-0.23) 0.0685 2.4456 1200 (-2.80) -0.0035 (-0.93) 0.0013 (0.60) Control Control -0.0091 (-0.32) 0.0654 2.3258 1200 ΔN ΔS Size Lev OCF Indep Board Dual Top1 Relevance Type SOE Year Industry _cons R2 F N Note , , ** contribute to significant levels of 10%, 5%, and 1% respectively. https://doi.org/10.1371/journal.pone.0284204.t010 formation of a modern industrial system and the high-quality development of China’s econ- omy as well. Furthermore, it will facilitate the formation of macroeconomic networks, includ- ing the Eurasian "connectivity strategy", and the proper implementation of M&A activities of group companies. Due to the limitations of information disclosure and data acquisition, the economic net- work constructed in this paper is based on the formal relationship between enterprises. How- ever, the informal relationship between enterprises may also play an important role in the strategic practice of enterprises. Future research can consider building an economic network based on the informal relationship between enterprises. In addition, for the sake of simplifying the model and highlighting the research focus, the relevant factors affecting enterprise network collaboration in the model may not be comprehensive enough. Subsequent research can PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 17 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? consider the degree of realization of M&A motivation based on the impairment of M&A good- will, and further, deepen the research on the intermediary effect of the impact mechanism. 7 Discussions The increasingly active M&A activity is accompanied by the "high failure rate". This paradox has become one of the problems that has long puzzled the theoretical and practical fields [4]. Some scholars have suggested that one of the reasons may be caused by empirical research is using an inaccurate measure of performance [50, 51]. Up to now, accumulation of abnormal returns of stocks, long-term shareholder value and changes in financial indicators before and after implementation are commonly used to evaluate M&A performance. While a number of researchers have acknowledged the shortcomings of using financial ratios or stock value to measure merger performance, Brouthers [5] suggest a better measure of merger success or fail- ure is the degree to which the merger achieves these predetermined objectives. Therefore, Some scholars have tried to evaluate M&A performance from the perspective of M&A motiva- tion, but when measuring M&A performance, they did not combine various motives to evalu- ate, or use questionnaires, index evaluation or other empowerment methods to evaluate the M&A motivation system. It cannot truly reflect the performance of M&A activities. There are no research results clarify the complicated mechanism that affect the realization of M&A moti- vation. By combining the respective nodes of the acquirer and the target, M&A can signifi- cantly reshape the inter-organizational network, thus creating synergies which will affect the efficiency of resource allocation, and then affects the economic performance of enterprises [18]. Learn from the previous research, this study constructs an equity network to explain the realization of M&A motivation and the complicated mechanism that why the realization is affected based on the idea of network synergies. The results show that the greater the variation of internal network node degree and strength, the more beneficial it is to promote the degree of realization of corporate M&A motivation. Furthermore, the network synergy generated by M&A can significantly enhance M&A performance, generate economies of scale and scope, and save transaction costs. Based on network collaboration, this paper uses the realization degree of achievement of M&A motivation as the criterion for determining M&A success or failure, which not only expands the research field of complex networks, but also reasonably and uniquely explains the paradox of "high failure rate" and increasingly active M&A activities from a new per- spective. This makes up for the shortcomings of existing studies that commonly used M&A performance indicators and their measurement methods may be detached from the M&A motivation, resulting in biased judgment results. Furthermore, we adopt a large sample sta- tistical analysis method to measure the relative contribution of M&A synergy to the degree of M&A motivation, which enriches the research results on the factors influencing M&A motivation. As a basic theoretical research result, this study is conducive to guiding M&A behavior to be more rational, and also to guiding M&A application research. As stated in the article, although the paper has achieved certain theoretical contributions and practical implications, there are still limitations and shortcomings that need subsequent improvement. Supporting information S1 Dataset. (XLSX) PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 18 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? S1 Appendix. (DOCX) Author Contributions Conceptualization: Ziqiao Zhang. Formal analysis: Ziqiao Zhang, Qiusheng Zhang. Funding acquisition: Ziqiao Zhang. Investigation: Chaofeng Li. Methodology: Ziqiao Zhang. Project administration: Qiusheng Zhang. Resources: Qiusheng Zhang. Validation: Chaofeng Li. Writing – original draft: Ziqiao Zhang. Writing – review & editing: Ziqiao Zhang. References 1. Bauer F.; Matzler K. Antecedents of M&A success: The role of strategic complementarity, cultural fit, and degree and speed of integration. Strategic Management Journal, 2014, 35: 269–291. https://doi. org/10.1002/smj.2091 2. Ferris S.F.; Jayaraman N.; Sabherwal S. CEO overconfidence and international merger and acquisition activity. Journal of Financial and Quantitative Analysis, 2013, 48(1): 137–164. https://doi.org/10.1017/ S0022109013000069 3. Rao-Nicholson R, Khan Z.; Akhtar P.; Merchant H. The impact of leadership on organizational ambidex- terity and employee psychological safety in the global acquisitions of emerging market multinationals. International Journal of Human Resource Management, 2016, 27(20): 2461–2487. https://doi.org/10. 1080/09585192.2016.1204557 4. Renneboog L.D.R.; Vansteenkiste C. Failure and success in mergers and acquisitions. Journal of Cor- porate Finance, 2019, 58: 650–699. https://doi.org/10.2139/ssrn.3434256 5. Brouthers K. D., van Hastenburg P., van den Ven J. If most mergers fail why are they so popular? Long Range Planning, 1998, 31(3): 347–353. https://doi.org/10.1016/S0024-6301(98)80002-2 6. Rabier M.R. Acquisition motives and the distribution of acquisition performance. Strategic Management Journal, 2017, 38(13): 2666–2681. https://doi.org/10.1002/smj.2686 7. Zhang X. M&A motivation and M&A performance of Chinese enterprises: A study based on DEA and TSFA. Management Modernization, 2015, 35(2): 87–89. https://doi.org/10.3969/j.issn.1003-1154. 2015.02.030 8. Angwin D.N. Motive archetypes in mergers and acquisitions (M&A): The implications of a configu- rational approach to performance. Advances in Mergers & Acquisitions, 2007, 6: 77–105. https://doi. org/10.1016/S1479-361X(07)06004-8 9. Zhou S.; Wen H. Research on M&A performance evaluation system based on industry evolution and M&A motivation. Accounting Research, 2013, (10): 75–82. CNKI:SUN:KJYJ.0.2013-10-010 10. Hassan I.; Ghauri P.; Mayrhofer U. Merger and acquisition motives and outcome assessment. Thunder- bird International Business Review, 2018, 60(4): 709–718. https://doi.org/10.1002/tie.21967 11. Yin A.; Ma X.; Yu P. M&A performance evaluation based on M&A motivation——Taking China eastern airlines’ M&A of Shanghai airlines as an example. Financial accounting Monthly, 2018, (11): 107–114. https://doi.org/10.19641/j.cnki.42-1290/f.2018.11.020 12. Fu Z.; Zeng S. Industrial agglomeration: A case study from the Perspective of enterprise social Network. Management World, 2008, 12: 83–92. https://doi.org/10.19744/j.cnki.11-1235/f.2008.12.010 13. Yi X.J.; Zhang B.; Yang R.D.; Yang B.Y. Family social network and farmers’ saving behavior: an empiri- cal study based on China’s rural areas. Management World, 2012, (5): 43–51. https://doi.org/10.19744/ j.cnki.11-1235/f.2012.05.004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 19 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? 14. Burt R.S. Models of network structure. Annual Review of Sociology, 1980, 6: 79–141. https://doi.org/ 10.1146/annurev.so.06.080180.000455 15. 16. Zhang Y. The position of independent directors’ relationship network and corporate commercial credit financing——Based on the perspective of degree centrality and structural hole. Journal of Zhongnan University of Economics and Law, 2021, (2): 40–52. https://doi.org/10.19639/j.cnki.issn1003-5230. 20201010.001 Li F.; Hu Z. Research on influencing factors and mechanisms of network capability of agricultural enter- prises. Research Management, 2018, 39(12): 141–151. https://doi.org/10.19571/j.cnki.1000-2995. 2018.12.017 17. Mark L.S. The Synergy Trap: How Companies Lose the Acquisition Game. New York: Simon and Schuster. 1997, 288p. 18. Hernandez E.; Shaver J.M. Network synergy. Administrative Science Quarterly, 2019, 64(1): 171–202. https://doi.org/10.1177/0001839218761369 19. Tong Y.; Wang Q.; Zeng Y.; Hua C. Mergers and acquisitions motivation, financing decisions, and inno- vation output of mergers and acquisitions. Accounting Research, 2020, 5: 104–116. 20. Sun Y.; Wang J.; Chen N. Research on influencing factors of enterprise network power based on WSR. Management Review, 2021, 33(5): 163–173. https://doi.org/10.14120/j.cnki.cn11-5057/f.20201231. 005 21. Guo Q. Research on Network Synergy Effect of Multimodal Logistics Enterprise Mergers and Acquisi- tions. PhD Thesis. Beijing: Beijing Jiaotong University. 2011. 22. Zhang C.; Luo W.; Li M. The relationship network of independent directors and the quality of financial reporting in the audit committee. Audit and Economic Research, 2022, (1): 42–52. 23. Shi W.; Sun S.; Pinkham B.; Peng M.W. Domestic alliance network to attract foreign partners: Evidence from international joint ventures in China. Journal of International Business Studies, 2014, 45(3): 338– 362. https://doi.org/10.1057/jibs.2013.71 24. Harrison J.S.; Hitt M.A.; Hoskisson R.E.; Ireland R.D. Resource complementarity in business combina- tions: Extending the logic to organizational alliances. Journal of Management, 2001, 27(6): 679–690. https://doi.org/10.1177/014920630102700605 25. Buzzell R.D., Gale B.T. The PIMS Principles: Linking Strategy and Performance. Free Press, New York. 1987. 26. Niu J. Current situation and realization approach of enterprise scale economy. Wealth Times, 2020, 2: 94–95. CNKI:SUN:GXHZ.0.2020-02-069 27. Jiang Y.; Li X. Broadband cancellation method in an adaptive co-site interference cancellation system. International Journal of Electronics, 2022, 109(5): 854–874. https://doi.org/10.1080/00207217.2021. 1941295 28. Ni Q.; Guo J.; Wu W.; Wang H.; Wu J. Continuous influence-based community partition for social net- works. IEEE Transactions on Network Science and Engineering, 2022, 9(3): 1187–1197. https://doi. org/10.1109/TNSE.2021.3137353 29. Salancik G.R.; Pfeffer J. A social information processing approach to job attitudes and task design. Administrative Science Quarterly, 1978, 23(2): 224–253. https://doi.org/10.2307/2392563 PMID: 10307892 30. Dong W. Research on formation mechanism of alliance combination—based on the perspective of com- plex network. Industrial Technology Economics, 2017, 36(5): 126–130. https://doi.org/10.6049/kjjbydc. 2018110457 31. Wang H.; Cui Z.; Liu R.; Fang L.; Sha Y. A multi-type transferable method for missing link prediction in heterogeneous social networks. IEEE Transactions on Knowledge and Data Engineering, 2023, https:// doi.org/10.1109/TKDE.2022.3233481 32. Li T.; Li Y.; Hoque M.A.; Xia T.; Tarkoma S.; Hui P. To what extent we repeat ourselves? Discovering daily activity patterns across mobile app usage. IEEE Transactions on Mobile Computing, 2022, 21(4): 1492–1507. https://doi.org/10.1109/TMC.2020.3021987 33. Sun G.; Li T. Research on the digital transformation path of enterprise network under the background of digital economy. Science and Science and Technology Management, 2021, 1: 128–145. 34. Gulati R.; Nohria N.; Zaheer A. Strategic networks. Strategic Management Journal, 2000, 21: 203–215. https://doi.org/10.12691/jbms-3-2-4 35. Zhang L.; Hou X.R.; Wang K.W.; Zhang Q.; Jia Y.B. Crossing the barriers of invalid relationships: A study on the dynamic construction mechanism of enterprise network locations. Research Management, 2020, 41(6): 11. https://doi.org/10.19571/j.cnki.1000-2995.2020.06.026 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 20 / 21 PLOS ONEDo network synergies facilitates the realization of M&A motivation? 36. 37. 38. Liu X.; Tong D.; Huang J.; Zheng W.; Kong M.; Zhou G. What matters in the e-commerce era? Model- ling and mapping shop rents in Guangzhou, China. Land Use Policy, 2022, 123: 106430. https://doi. org/10.1016/j.landusepol.2022.106430 Lei W.; Hui Z.; Xiang L.; Zelin Z.; Xu-Hui X.; Evans S. Optimal remanufacturing service resource alloca- tion for generalized growth of retired mechanical products: Maximizing matching efficiency. IEEE Access, 2021, 9: 89655–89674. https://doi.org/10.1109/ACCESS.2021.3089896 Jiang Q. Ask Smith and Coase: How is the Internet possible? Research on Financial Issues. 2018, 10: 21–30. https://doi.org/10.19654/j.cnki.cjwtyj.2018.10.002 39. Barrat A.; Barthe´ lemy M.; Pastor-Satorras R.; Vespignani A. The architecture of complex weighted net- works. Proceedings of the National Academy of Sciences, 2004, 101(11): 3747–3752. https://doi.org/ 10.1073/pnas.0400087101 PMID: 15007165 40. Grant R.M. Prospering in dynamically-competitive environments: Organizational capability as knowl- edge integration. Organization Science, 1996, 7: 375–387. https://doi.org/10.1287/orsc.7.4.375 41. Li M.; Su J.; Hu C. The influence of enterprise network location and relationship strength on technologi- cal innovation performance in industry-university-research cooperation. Science and Technology Prog- ress and Countermeasures, 2020, 37(14): 118–124. https://doi.org/10.6049/kjjbydc.Q201908747 42. Chen S.; Jiang G.; Lu C. Director connection, target company selection and mergers and acquisitions performance——Based on the research perspective of information asymmetry between mergers and acquisitions. Management World, 2013, 12: 117–132. https://doi.org/10.19744/j.cnki.11-1235/f.2013. 12.011 43. Liu C.; Li S.; Sun L. Do independent directors have consulting functions?——An empirical study on the functions of independent directors in out-of-place mergers and acquisitions. Management World, 2015, 3:124–136. https://doi.org/10.19744/j.cnki.11-1235/f.2015.03.012 44. Pan A.; Liu X.; Wu Q. Cross-ownership M&A, institutional environment and debt financing cost of private enterprises. Journal of Accounting Research, 2019, 5: 3–10. CNKI:SUN:KJYJ.0.2019-05-001 45. Wu H.; Liu S. Corporate culture and technological innovation efficiency from the perspective of related M&A. Journal of Zhongnan University of Economics and Law, 2018, (3): 65–72. https://doi.org/10. 19639/j.cnki.issn1003-5230.2018.0036 46. Wu C.; Tang Q. Does related-party mergers and acquisitions have information transmission effects? ——Based on the perspective of corporate social capital. Audit and Economic Research, 2016, 31(2): 81–90. https://doi.org/10.3969/j.issn.1004-4833.2016.02.009 47. Zhang Q.; He P. Research on the merger of external economic networks of enterprises. Friends of Accounting, 2015, (12): 7–15. CNKI:CDMD:2.1015.561660 48. Ang J.S.; Cole R.A.; Lin J.W. Agency costs and ownership structure. Journal of Finance, 2000, 55, 81– 106. https://doi.org/10.1111/0022-1082.00201 49. Xia J.; Liu C. Administrative approval reform, transaction costs and China’s economic growth. Manage- ment World, 2017, (4): 47–59. https://doi.org/10.19744/j.cnki.11-1235/f.2017.04.005 50. Berkovitch E.; Narayanan M.P. Motives for takeovers: An empirical investigation. Journal of Financial and Quantitative analysis, 1993, 28: 347–362. https://doi.org/10.2307/2331418 51. Ingham H.; Kran I.; Lovestam A. Mergers and profitability: A managerial success story?* Journal of Management Studies, 1992, 29(2): 195–208. https://doi.org/10.1111/j.1467-6486.1992.tb00660.x PLOS ONE \| https://doi.org/10.1371/journal.pone.0284204 April 20, 2023 21 / 21 PLOS ONE
10.1371_journal.pone.0283908	RESEARCH ARTICLE Identification of Ndfip1 as a novel negative regulator for spatial memory formation associated with increased ubiquitination of Beclin 1 and PTEN Wei-Lun Hsu1, Yun-Li Ma1, Yan-Chu Chen1, Yen-Chen Liu1, Kuang-Min Cheng1,2, Eminy H. Y. Lee1* 1 Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan, 2 Graduate Institute of Life Sciences, National Defense Medical Center, Taipei, Taiwan * eminy@gate.sinica.edu.tw Abstract Long-term memory formation requires de novo RNA and protein synthesis. By using the dif- ferential display-polymerase chain reaction strategy, we have presently identified the Nedd4 family interacting protein 1 (Ndfip1) cDNA fragment that is differentially expressed between the slow learners and the fast learners from the water maze learning task in rats. Further, the fast learners show decreased Ndfip1 mRNA and protein expression levels than the slow learners. Spatial training similarly decreases the Ndfip1 mRNA and protein expression lev- els. Conversely, the Ndfip1 conditional heterozygous (cHet) mice show enhanced spatial memory performance compared to the Ndfip1flox/WT control mice. Result from co-immuno- precipitation experiment indicates that spatial training decreases the association between Ndfip1 and the E3 ubiquitin ligase Nedd4 (Nedd4-1), and we have shown that both Beclin 1 and PTEN are endogenous ubiquitination targets of Nedd4 in the hippocampus. Further, spatial training decreases endogenous Beclin 1 and PTEN ubiquitination, and increases Beclin 1 and PTEN expression in the hippocampus. On the other hand, the Becn1 condi- tional knockout (cKO) mice and the Pten cKO mice both show impaired spatial learning and memory performance. Moreover, the expression level of Beclin 1 and PTEN is higher in the Ndfip1 cHet mice compared with the Ndfip1flox/WT control mice. Here, we have identified Ndfip1 as a candidate novel negative regulation for spatial memory formation and this is associated with increased ubiquitination of Beclin 1 and PTEN in the hippocampus. Introduction It is well documented that long-term memory formation requires de novo RNA and protein synthesis. Previous studies have shown that inhibition of mRNA and protein synthesis impairs long-term memory formation in rats [1,2]. These results suggest that gene expression in neu- rons associated with learning plays an important role in memory formation. New protein a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Hsu W-L, Ma Y-L, Chen Y-C, Liu Y-C, Cheng K-M, Lee EHY (2023) Identification of Ndfip1 as a novel negative regulator for spatial memory formation associated with increased ubiquitination of Beclin 1 and PTEN. PLoS ONE 18(4): e0283908. https://doi.org/10.1371/journal. pone.0283908 Editor: Stephen D. Ginsberg, Nathan S Kline Institute, UNITED STATES Received: October 11, 2022 Accepted: March 20, 2023 Published: April 6, 2023 Copyright: © 2023 Hsu et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and are shown in the Supporting Information, S6 Fig. Funding: Dr. Eminy H.Y. Lee received the Grants from the Ministry of Science and Technology (MOST109-2320-B-001-027 and MOST110-2320- B-001-006-MY2), and research fund from the Institute of Biomedical Sciences, Academia Sinica, Taiwan to support this work. The funders do not play a role in this study. The URL of MOST is: PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 1 / 25 PLOS ONEhttps://www.nstc.gov.tw/ The URL of IBMS, Academia Sinica is: https://www.ibms.sinica.edu. tw/ch/index.php. Competing interests: The authors have declared that no competing interests exist. Ndfip1 impairs spatial memory synthesis and proper translational control were also demonstrated to be necessary for both long-lasting synaptic plasticity and long-term memory formation [3]. In addition, learning was shown to allow the activated synapses to recruit newly synthesized glutamate α-amino- 3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors in the hippocampus [4]. Different genes are involved in different forms of learning and memory and the pleiotropic effect may exist with disruption of a given gene [5]. Even for similar types of learning, gene expression could be different depending upon the animal species studied. Therefore, various strategies have been adopted to address the issue of gene expression associated with long-term memory formation in the past. For example, Castellucci et al. have identified several genes that are associated with long-term sensitization of gill-withdrawal reflex in Aplysia using two- dimensional gel analysis [6]. In a mammalian study using microarray analysis, 140 candidate genes from the rat hippocampus have been identified that are associated with water maze learning [7]. Moreover, we have previously identified the integrin-associated protein (IAP) gene that is associated with memory formation of inhibitory avoidance learning in rats, and this was performed by using the differential display-polymerase chain reaction (DD-PCR) strategy [8]. Later on, by using the same method, we have identified 98 cDNA fragments from the dorsal hippocampus that are differentially expressed between the fast-learning rats and slow-learning rats associated with water maze learning, and one of these cDNA fragments encodes the serum- and glucocorticoid-inducible kinase (sgk) gene [9]. We further demon- strate that SGK facilitates both spatial memory formation and long-term potentiation (LTP) in rats [10,11]. In addition to the sgk gene, we also identified another gene named the protein inhibitor of activated STAT1 (pias1) gene. The expression level of PIAS1 is also higher in fast learners than slow learners, and knockdown of endogenous PIAS1 expression impairs spatial memory formation [12]. In addition to protein synthesis, protein degradation has been shown to also play a role in memory consolidation [13–15] and the ubiquitin-proteasome system is an important mecha- nism mediating protein degradation involved in memory processing [16]. Other than the sgk gene and pias1 gene reported above, we also identified other cDNA fragments that are associ- ated with spatial memory formation in our earlier study [9]. One of these cDNA fragments encodes the rat Nedd4 (neural-precursor-cell-expressed developmentally down-regulated 4) family interacting protein 1 (Ndfip1) gene. Ndfip1 is an adaptor protein for the Nedd4 family of E3 ubiquitin ligases, including Nedd4, and that Ndfip1 and Nedd4 together identify target proteins for ubiquitination [17]. Ndfip1 is also necessary for the secretion of Nedd4 family pro- teins [18]. Ndfip1 is involved in various cellular functions. In the central nervous system, Ndfip1 was shown to regulate divalent metal transporter 1 (DMT1) and iron homeostasis and prevents metal toxicity [19,20]. Further, Ndfip1 is required for the development of neuronal dendrites and spines and it is associated with neuronal survival [21,22]. But the role of Ndfip1 in mammalian learning and memory function is not known. Here, we examined the role and molecular mechanism of Ndfip1 in spatial learning and memory formation. The class III PI3-kinase (PI3K-III) complex is suggested to involve in a few cellular pro- cesses and Beclin 1 is believed to form the core of the PI3K-III complex for recruitment of reg- ulatory proteins [23]. In addition, Beclin 1 was found as a polyubiquitination target of Nedd4 and Nedd4 regulates Beclin 1 stability through ubiquitination modification [24]. Further, inhibitory avoidance learning was shown to increase the level of Beclin 1 [25]. In this study, we wished to examine whether Beclin 1 is an endogenous ubiquitination target of Nedd4 involved in spatial learning and memory formation and whether Beclin 1 expression is regulated by Ndfip1. On the other hand, phosphatase and tensin homology deleted on chromosome 10 (PTEN) is a tumor suppressor and a polyubiquitination target of Nedd4, and Nedd4-mediated PTEN PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 2 / 25 PLOS ONENdfip1 impairs spatial memory ubiquitination results in PTEN proteasomal degradation [26]. Further, both LTP and long- term depression are dysregulated in Nse-Cre Pten cKO mice with deletion of Pten in dentate granule neurons [27], and that contextual fear memory is impaired in PTENα (an isoform of PTEN)-deficient mice [28]. Thus, we also wished to examine whether PTEN is an endogenous ubiquitination target of Nedd4 associated with spatial learning and memory formation and whether PTEN expression is regulated by Ndfip1. Materials and methods Differential display PCR (DD-PCR) DD-PCR was performed in a previous study and the cDNA fragment examined here was also obtained previously [9]. Briefly, 80 arbitrary random primers (H-AP1BH-AP80, RNAimage Kit) were purchased from GenHunter (Nashville, TN). The reverse transcribed (RT) products of dorsal hippocampal tissues from three fast learners and slow learners were subjected to dif- ferent amplification reactions by using these primers according to the procedures described previously [9]. Differentially expressed cDNA fragments were resolved from the sequencing gels and cloned into the PCR 2.1 TA vector (Invitrogen, Carlsbad, CA). Animals Adult male Sprague-Dawley rats (250–350 g) were purchased from the BioLASCO, Taiwan. The Ndfip1flox/WT mice and Ndfip1flox/flox mice were generated from targeted ES cell with replaced Exon 3 of Ndfip1 with the same exon flanked by loxP sites, but the majority of mice we obtained are the Ndfip1flox/WT mice. The Beclin 1flox/flox mice (strain name: Becn1tm1.1Yue/ J, stock number: 028794) and Pten flox/flox mice (strain name: B6.129S4-Ptentm1Hwu/J, stock number: 006440) were purchased from Jackson Laboratory (Bar Harbor, ME). Animals were mated, bred and maintained on a 12/12 h light/dark cycle (light on at 8:00 am) at the Animal Facility of the Institute of Biomedical Sciences (IBMS), Academia Sinica with food and water continuously available. Experimental procedures follow the Guidelines of Animal Use and Care of the National Institute of Health and were approved by the Animal Committee of IBMS, Academia Sinica. Generation of Ndfip1flox/WT conditional heterozygous (cHet) mice Ndfip1 knockout-first embryonic stem (ES) cell was obtained from the European Mouse Mutant Cell Repository (EuMMCR, clone number: HEPD0764_3_A03). The recombinant allele contains a FRT-hBactP-Neo-FRT-loxP cassette and a loxP site upstream and down- stream of Exon 3 Ndfip1, respectively. Targeted ES cells were injected into blastocysts to gener- ate chimeric mice from Transgenic Core Facility of Academia Sinica. The chimeric mice were identified by genotyping PCR using primer Ndfip1-F1, Ndfip1-F2 and Ndfip1-R to detect the loxP site upstream of Exon 3 and using primers Neo-F and Ndfip1-R to detect the FRT-flanked selection cassette. These primer sequences are shown in Table 1. For deletion the FRT- hBactP-Neo-FRT selection cassette upstream Exon 3 of Ndfip1, male chimeric mice containing FRT-hBactP-Neo-FRT-loxP cassette were mated with female Act-Flpe mice (stock number: 003800, Jackson lab) to generate Ndfip1flox/WT mice, which were backcrossed with wild-type C57BL/6 mice for at least eight generations and confirmed by genotyping PCR using Ndfip1- F1, Ndfip1-F2, and Ndfip1-R primers. After eight generations, male and female Ndfip1flox/WT mice were inbred to generate the Ndfip1flox/WT mice and Ndfip1flox/flox mice. However, most of the animals we obtained from inbreeding are the Ndfip1flox/WT mice and we have therefore used this genotype of mice for the present study. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 3 / 25 PLOS ONENdfip1 impairs spatial memory Table 1. List of all the primers used. Primer Ndfip1-Forward-1 Ndfip1-Forward-2 Neo-Forward Ndfip1-Reverse Nedd4-1 cDNA-Forward Nedd4-1 cDNA-Reverse Pten cDNA-Forward Pten cDNA-Reverse Human ubiquitin cDNA-Forward Human ubiquitin cDNA-Reverse Ndfip1 siRNA-1 sense Ndfip1 siRNA-1 antisense Ndfip1 siRNA-2 sense Ndfip1 siRNA-2 antisense Nedd4 siRNA sense Nedd4 siRNA antisense Cre vector-Forward Cre vector-Reverse GFP vector-Forward GFP vector-Reverse Ndfip1-Q-PCR-Forward Ndfip1-Q-PCR-Reverse HRPT-Q-PCR-Forward HPRT-Q-PCR-Reverse https://doi.org/10.1371/journal.pone.0283908.t001 Primer Sequences (5’-3’) CTTACTTGCTGCACCATCTGGCCAG GTCGAGATATCTAGACCCAGCTTTC AGCGAGCACGTACTCGGATG CAAGCTCTAGTCCAGCTTAGGCAAC ATGCAAGCTTGCCACCATGAGCTCGGACATGGCAGCC ATGCGCGGCCGCCATCAACCCCATCAAAGCCCTG ATCGGAATTCATGACAGCCATCATCAAA ATCGAAGCTTTCAGACTTTTGTAATTTG ATCGCCATGGATGCAGATCTTCGTGAAGAC ATCGGGATCCTTAGACACCCCCCCTCAAGC UUUGGUCUCUCUCUAAUUAtt UAAUUAGAGAGAGACCAAAtt GCAUUCCUCUUUAACUGGAtt UCCAGUUAAAGAGGAAUGCtt UCAAUCGCCAUCUGAAGUUUAUCCtt AGUUAGCGGUAGACUUCAAAUAGGtt ATCGGAATTCCCAAAGAAGAAGAGAAAGGTTATGTCCAATTTACTGACC ATCGGCGGCCGCCTAATCGCCATCTTCCAG ATCGAGTACTGCCACCATGGAGATCGAGTGCCGCATC ATCGGGTACCGGCGAAGGCGATGGGGGTC GCTACAACACTGCCCAGCTA ACCCAATCCAGTTGAAGAGG GCCGACCGGTTCTGTCAT TCATAACCTGGTTCATCATCACTAATC Genotyping With the primers used above, the Ndfip1+/+ wild-type mice have a PCR product of 305 bp in length, the Ndfip1flox/WT mice have PCR products of 305 bp and 180 bp in length, and the Ndfip1flox/flox mice have a PCR product of 180 bp in length. The parameters used for PCR are: 95˚ C for 30 s, 60˚ C for 30 s, 72˚ C for 10 s for 36 cycles, followed by a final elongation at 72˚ C for 90 s. The PCR product was analyzed on a 2% agarose gel. Drugs and drug infusion to the mouse hippocampus N-methyl-D-aspartate (NMDA) was purchased from Tocris Bioscience (St. Louis, MO, USA) and was dissolved in PBS in a concentration of 8 mM immediately before injection. SP600125 was purchased from Sigma-Aldrich (St. Louis, MO) and was first dissolved in 100% DMSO (Sigma-Aldrich) and further diluted with PBS to a final concentration of 1 μg/μl in 45% DMSO. SP600125 was prepared immediately before injection. A volume of 0.2 μl was injected to each side of the mouse CA1 area. The injection rate was 0.1 μl/min. Plasmid DNA construction Different plasmids (with different tags) were constructed as that described previously [29]. Briefly, for construction of the V5-tagged Nedd4-1 plasmid, full-length Nedd4-1 was cloned by amplifying the rat hippocampal Nedd4-1 cDNA (accession # NM_012986.1) with forward and reverse primers shown in Table 1. The PCR product was sub-cloned between the HindIII and PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 4 / 25 PLOS ONENdfip1 impairs spatial memory NotI sites of the mammalian expression vector pcDNA3.1-V5 vector. For construction of the Flag-tagged PTEN plasmid, full-length Pten was cloned by amplifying the rat hippocampal Pten cDNA (accession # NM_031606.1) with forward and reverse primers shown in Table 1. The PCR product was sub-cloned between the EcoRI and HindIII sites of the mammalian expression vector pCMVTag2B vector. The Flag-Beclin 1 (Product ID: OHu13549D) and Flag- Ndfip1 (Product ID: OHu20100D) plasmids were purchased from GenScript (Piscataway, NJ). For construction of the His-tagged Ubiquitin plasmid, full-length ubiquitin was cloned by amplifying the human ubiquitin cDNA (accession # NM_021009.7) according to that of Liu et al. [30] with forward and reverse primers shown in Table 1. The PCR product was sub- cloned between the NcoI and BamHI sites of the mammalian expression vector CMV-3 × His tag vector. Plasmid DNA and small interference RNA (siRNA) transfection Different plasmids (with different tags) and Ndfip1 siRNA (or control siRNA) were transfected to HEK293T cells for determination of Beclin 1 and PTEN ubiquitination. HEK293T cells were maintained in Dulbecco’s modified Eagle’s medium containing 10% fetal bovine serum and incubated at 37˚C in a humidified atmosphere with 5% CO2. Two sets of Ndfip1 siRNA were used in HEK293T cells and sequences for the sense and antisense strands are shown in Table 1. The Negative Control siRNA was used as the control. They were all synthesized from MDBio, Inc (Taipei, Taiwan). Duplex siRNA was diluted with DEPC-treated water to 50 μM, and two duplex siRNAs were used in combined Ndfip1 siRNA-1 and Ndfip1 siRNA-2 (equal amount). Different amounts of plasmid DNA and Ndfip1 siRNA (100 pmol) transfections were made by using the Lipofectamine 2000 reagent (Invitrogen, Carlsbad, CA) in 6-well cul- ture plates according to the manufacturer’s protocols. Immunoprecipitation (IP) and western blot were conducted 48 h after plasmid DNA or Ndfip1 siRNA transfection. In addition, Nedd4 siRNA (10 pmol) or control siRNA was and transfected to the rat CA1 area for exami- nation of endogenous Beclin 1 ubiquitination and PTEN ubiquitination 48 h later. Transient siRNA transfection was conducted using the non-viral transfection agent polyethyleneimine (PEI) (0.1 mM) and we have previously demonstrated that PEI does not produce toxicity to hippocampal neurons [31]. The sequences for Nedd4 siRNA sense and antisense strands are shown in Table 1. Lenti-NLS-Cre construction and injection to the mouse hippocampus The lenti-NLS-Cre vector was prepared based on the construct we already have as described previously [29]. For construction of GFP-2A-NLS-Cre lentiviral vector, full-length Cre recom- binase cDNA was added with nuclear localization signal (NLS) using PCR-amplification and cloned into pLenti-Tri-cistronic (ABM, Richmond, BC, Canada) to obtain a bicistronic vector expressing both GFP and NLS-Cre. The primer sequences used for the Cre vector are shown in Table 1. The PCR product was subcloned between the EcoRI and NotI sites of the lentiviral vector pLenti-Tri-cistronic (ABM). The GFP construct was cloned by amplifying the GFP gene from pLenti-CMV-GFP-2A-Puro-Blank (ABM) and subcloned into the pLenti-Tri-cis- tronic vector between ScaI and KpnI sites, upstream of the 2A peptide (a self-processing viral peptide bridge) and NLS-Cre sequences. The primer sequences used for the GFP vector are shown in Table 1. For lentivirus packaging, HEK293LTV cells (Cell Biolabs, San Diego, CA) were transfected with 1.5 μg of psPAX2 (Addgene plasmid #12260), 0.5 μg of pMD2.G (Addgene plasmid #12259), and 2 μg of pLenti-GFP-2A-NLS-Cre (or 2 μg of pLenti- CMV-GFP-2A-Puro-Blank (ABM) coding for GFP as control) using 10 μl of Lipofectamine 2000 (Invitrogen) in 6-well cell culture dishes. Lentiviral particles were collected using the PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 5 / 25 PLOS ONENdfip1 impairs spatial memory speedy lentivirus purification solution (ABM) according to the manufacturer’s protocols. Cell culture medium containing lentiviral particles can be harvested for two to three times at 12 h interval until 36 h after transfection, and it was kept at 4˚C for the collecting period. The col- lected culture medium was further clarified by centrifugation at 2,500 x g for 10 min and fil- trated through a 0.45 μm syringe filter. The speedy lentivirus purification solution (ABM) was added into filtrated supernatant (1:9, v/v) containing lentiviral particles and mixed thoroughly by inversion. The lentiviral supernatant was centrifuged at 5,000 x g at 4˚C for 10 min. Super- natant was then discarded and the viral pellet was re-suspended in ice cold PBS. After titration, the viral stock was stored at -80˚C in aliquots. The lentivirus titer was determined by lentivirus qPCR Titer Kit (ABM) according to the manufacturer’s protocols. The final concentration of the lentiviral vector used for injection to the hippocampus is 5 x 108 IU/ml. For lentiviral vector injection, mice were anesthetized with pentobarbital (50 mg/kg, i.p.) and subjected to stereotaxic surgery without cannulation. Lentiviral vector was directly injected to their CA1 area bilaterally at the following coordinates: -1.8 mm posterior to the bregma, ±1.3 mm lateral to the midline, and -2.1 mm ventral to the skull surface. A volume of 0.2 μl was injected to each side of the CA1 area. The infusion rate was 0.1 μl/min. Spatial learn- ing started two weeks after the lentiviral vector injection. Quantitative real-time PCR (Q-PCR) Total RNA from CA1 tissue was isolated by using the RNAspin mini kit (GE Healthcare). The cDNA was generated from total RNA with Superscript III reverse transcriptase (Invitrogen). Real-time PCR analysis was performed by using the ABI PRISM 7500 real-time PCR system with Power SYBR Green PCR Master Mix according to the instruction manual (Applied Bio- systems, Foster City, CA). The forward and reverse primer sequences used for Ndfip1 are shown in Table 1. The primer sequences used for HRPT are also shown in Table 1. The cycle threshold (Ct) value and data were analyzed by using the 7500 system Sequence Detection Software (Applied Biosystems). Quantitative analysis of Ndfip1 gene expression was normal- ized to that of HPRT gene expression. Immunoprecipitation and western blot Immunoprecipitation (IP) and western blot for hippocampal tissue lysate were conducted according to that described previously [32]. Briefly, cell lysate, mice and rat CA1 tissue were lysed by brief sonication in RIPA lysis buffer containing 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 2 mM EDTA, 1% IGEPAL CA-630 and 20 mM N-ethylmaleimide (Catalog No. E3876- 5G, Sigma-Aldrich). One tablet of protease inhibitor cocktail (Catalog No. 05892791001, cOm- plete ULTRA Tablets, Mini, EDTA-free, EASYpack, Roche, Mannheim, Germany) and one tablet of phosphatase inhibitor (Catalog No. 04906837001, PhosSTOP, Roche) were added to each 10 ml of the RIPA lysis buffer. For IP of Flag-PTEN and Flag-Beclin 1, the clarified lysate (0.5 mg) was immunoprecipitated with 3 μl of anti-Flag M2 antibody (Catalog No. F1804, Sigma-Aldrich, St. Louis, MO) at 4˚C for overnight. The protein G magnetic beads (30 μl, 50% slurry, GE Healthcare, Chicago, IL) were added to the IP reaction product to catch the immune complex at 4˚C for 3 h. For IP of Ndfip1, the clarified lysate (0.5 mg) was immuno- precipitated with 6 μl of anti-Ndfip1 antibody (Catalog No. Ab236892, Abcam, Cambridge, UK) at 4˚C for overnight. The protein A magnetic beads (60 μl, 50% slurry, GE Healthcare, Chicago, IL) were added to the IP reaction product to catch the immune complex at 4˚C for 3 h. The immune complex on beads were washed three times with washing buffer containing 20 mM HEPES (pH 7.4), 150 mM NaCl, 1 mM EDTA, 1% IGEPAL CA-630, 1 mM DTT, 50 mM β-glycerophosphate, 50 mM NaF, 10 mg/ml PMSF, 4 μg/ml aprotinin, 4 μg/ml leupeptin and PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 6 / 25 PLOS ONENdfip1 impairs spatial memory 4 μg/ml pepstatin and were subjected to 8% or 12% SDS-PAGE followed by transferring onto the PVDF membrane (Catalog No. IPVH00010, Millipore, Bedford, MA). Western blot was conducted using the following antibodies: anti-Ndfip1 (1:3000; Catalog No. Ab236892, Abcam, Cambridge, UK), anti-Nedd4 (1:2000; Catalog No. 5344S, Cell Signaling, Danvers, MA), anti-Beclin 1 (1:2000; Catalog No. 3738S, Cell Signaling), anti-PTEN (1:2000; Catalog No. 9188S, Cell Signaling), anti-ubiquitin (1:3000, Catalog No. 3936, Cell Signaling), anti-His (1:4000, Catalog No. 66005-Ig, Proteintech, Rosemont, IL), anti-Flag M2 (1:8000; Sigma- Aldrich), anti-V5 (1:8000; Catalog No. MCA2895, AbD Serotec, Kidlington, UK) and anti- actin (1:200000; Millipore; Catalog No. MAB1501) antibodies. The secondary antibody used was HRP-conjugated goat-anti rabbit IgG antibody or HRP-conjugated goat-anti mouse IgG antibody (1:8000, Catalog No. 111-035-003 and 115-035-003, Jackson ImmunoResearch, West Grove, PA). The secondary antibody used for co-IP experiment was HRP-conjugated goat- anti rabbit IgG light chain antibody (1:6000, Catalog No. 112-035-175, Jackson ImmunoRe- search) or HRP-conjugated goat-anti mouse IgG light chain antibody (1:6000, Catalog No. 115-035-174, Jackson ImmunoResearch). Membrane was developed by reacting with chemilu- minescence HRP substrate (Millipore) and was exposed to the LAS-3000 image system (Fuji- film, Tokyo, Japan) for visualization of protein bands. The protein bands were quantified by using the NIH Image J Software. Immunohistochemistry For immunohistochemical staining of GFP in the CA1 area of the mouse brain, mice were anesthetized with pentobarbital (50 mg/kg, i.p.) and perfused with ice-cold PBS, followed by 4% paraformaldehyde. Brains were removed and post-fixed in 20% sucrose/4% paraformalde- hyde solution for 20–48 h. Brains were then frozen, cut into 30-μm sections on a cryostat and mounted on gelatin-coated slides. Brain sections were rinsed with 1 X PBS for 10 min and per- meabilized with pre-cold EtOH/CH3COOH (95%:5%) for 10 min, followed by 1 X PBS for 10 min for three times. The sections were pre-incubated in a blocking solution containing 3% bovine serum albumin (BSA) and 0.1% Triton X-100 in 1 X TBS for 1 h, followed by 1 X PBS for 10 min for three times. For examination of lentiviral vector transduction, brain sections containing the CA1 area were prepared for visualization of GFP (green) fluorescence. For immunofluorescence detection of the nucleus, tissue sections were added with 20 μl of the Fluoromount-G mounting medium with DAPI (SouthernBiotech, Birmingham, AL). Photo- micrographs were taken using a Zeiss LSM700 confocal microscope. Water maze learning For spatial acquisition adopted in the present study, the water maze used was a plastic, circular pool, 1.2 m in diameter and 25 cm in height that was filled with water (25 ± 2˚C) to a depth of 16 cm. A circular platform of 10 cm in diameter was placed at a specific location away from the edge of the pool. The top of the platform was submerged 0.6 cm below the water surface. Water was made cloudy by adding milk powder. Distinctive, visual cues were set on the wall. For spatial acquisition, animals were subjected to three trials a day (as one session) with one given early in the morning, one given in the early afternoon and the other one given in the late afternoon. The acquisition procedure lasted for 5 days (for 5 sessions) and a total of 15 trials were given. For these trials, animals were placed at different starting positions spaced equally around the perimeter of the pool in a random order. Animals were given 60 sec to find the platform. If an animal could not find the platform within 60 sec, it was guided to the platform and was allowed to stay on the platform for 20 sec. The time that each animal took to reach the platform was recorded as the escape latency. A probe trial of 60 sec was given on day 6 to test PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 7 / 25 PLOS ONENdfip1 impairs spatial memory their memory retention. Animals were placed in the pool with the platform removed and the time they spent in each quadrant (target quadrant, left quadrant, opposite quadrant and right quadrant), the total distance travelled in the target quadrant and their swim speed were recorded. For spatial training, the procedures were the same as that for spatial acquisition except that training lasted for two consecutive days (for Ndfip1 mRNA measure) or three consecutive days (for Ndfip1 protein measure). Animals were sacrificed at the end of the last training trial. For screening of the fast-learning rats and slow-learning rats, the same criteria used in a previous study were adopted here [9]. Briefly, animals that reached the escape latency smaller than 30 sec by the end of the third training session were designated as the fast-learning rats (fast learners). Animals that did not reach this escape latency until the end of the seventh train- ing session were designated as the slow-learning rats (slow learners). For the visible platform learning experiment, a flag was mounted on the platform and the platform was 2.5 cm above the water surface so the animals can visualize the flag and indentify the location of the platform. In addition, milk powder was not added to the swimming pool to make it cloudy. Statistical analysis Spatial acquisition data were analyzed with two-way analysis of variance (ANOVA) with repeated measure followed by post-hoc Newman-Keuls multiple comparisons (represented by q value). Probe trial performance (time spent in the target quadrant and distance travelled in the target quadrant) and all biochemical data were analyzed with the Student’s t-test. Values of p < 0.05 were considered significant (* p < 0.05, p < 0.01, * p ≦ 0.001). Results Identification of the Nedd4 family interacting protein 1 (Ndfip1) gene from DD-PCR By using DD-PCR we have previously identified 98 cDNA fragments that are differentially expressed in the dorsal hippocampus between fast learners and slow learners from the water maze learning task [9]. When a specific primer set H-AP53 (5’-end primer sequence as 5’- AAGCTTCCTCTAT-3’) and H-T11C (3’-end primer sequence as 5’-AAGCTTTTTTTTTTTC- 3’) was used, one identified cDNA fragment of 253 bp in length (designated as G9-1-1) showed 100% sequence homology to the 3’-end region of the rat Ndfip1 gene (data accession number for Ndfip1: NM_001013059.1) (Fig 1A and 1B). This gene was categorized as one of the unknown genes in our previous study [9]. The expression level of this gene is lower in the fast learners than slow learners (Fig 1A). Spatial training decreases Ndfip1 expression in the rat hippocampus To examine the role of Ndfip1 in spatial learning, we first screened another batch of rats and obtained a separate group of fast learners and slow learners using the same criteria and proce- dures adopted from our previous study (Fig 1C) [9]. We then examined the Ndfip1 mRNA level in one side of the CA1 area from these fast learners and slow learners using quantitative real-time PCR (Q-PCR). The result revealed that Ndfip1 mRNA level is higher in the slow learners than fast learners (Fig 1D). Another side of the CA1 tissue from the same animals was subjected to western blot determination of Ndfip1 protein expression. Results revealed that the Ndfip1 protein level is similarly higher in slow learners than fast learners (Fig 1E). The above results suggest that Ndfip1 expression is negatively associated with spatial acquisition. Based on these results, we expect that spatial training should decrease the expression level of Ndfip1. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 8 / 25 PLOS ONENdfip1 impairs spatial memory Fig 1. Identification of the Ndfip1 gene and Ndfip1 expression is decreased by spatial training. (A) DD-PCR of hippocampal RNA associated with water maze learning in rats. One cDNA fragment (indicated by the arrow) was differentially expressed between the fast learners and the slow learners. (B) Alignment of the sequence of G9-1-1 (the arbitrary primers used) with rat Ndfip1. The numbers correspond to the Ndfip1 cDNA sequences. Vertical lines indicate identity and 100% sequence homology was found between these two. (C) Acquisition performance of fast learners (N = 6), slow learners (N = 6), and control rats (N = 22) from the water maze learning task. FL: Fast learners; SL: Slow learners. Data are expressed as mean values. (D) Ndfip1 mRNA level in the CA1 area is lower in fast learners than slow learners (t1,10 = 4.57, p = 0.001). (E) Ndfip1 protein expression in the CA1 area is lower in fast learners than slow learners (t1,10 = 12.43, p < 0.001). (F) Ndfip1 mRNA level in the CA1 area is lower in trained rats than the non- trained (swimming control) rats (t1,10 = 8.15, p < 0.001). (G) Ndfip1 protein expression in the CA1 area is lower in trained rats than the non-trained rats (t1,10 = 13.79, p < 0.001). N = 6 each group. Data are expressed as individual values and mean ± SEM. # p≦0.001. https://doi.org/10.1371/journal.pone.0283908.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 9 / 25 PLOS ONENdfip1 impairs spatial memory To test this hypothesis, a different batch of rats was randomly divided to the trained group or the non-trained group. Animals in the trained group were subjected to regular water maze learning, as described in the Method Section. Animals in the non-trained group swam for the same period of time for each trial as that of the trained animals except that the visual cues and platform were both removed. Animals were sacrificed after training and their CA1 tissue was dissected out for Ndfip1 mRNA and protein level determination. Result from Q-PCR analysis revealed that spatial training decreased Ndfip1 mRNA level (Fig 1F). Meanwhile, spatial train- ing also decreased Ndfip1 protein expression level (Fig 1G). Spatial learning is known to acti- vate the glutamate NMDA receptor. If spatial training downregulated the expression of Ndfip1, it is expected that NMDA receptor activation would yield similar result. This specula- tion was examined here. Rats were randomly divided to two groups and received acute intra- hippocampal PBS or NMDA (8 mM) injection. For the purpose of evaluating the effectiveness of NMDA injection, animals were sacrificed 30 min after PBS or NMDA injection and their CA1 tissue was dissected out and subjected to western blot determination of MAPK/ERK phosphorylation level. Ndfip1 expression was also determined in the same tissue lysate. Result revealed that NMDA decreased Ndfip1 expression as soon as 30 min after injection. Mean- while, it increased the phosphorylation level of ERK1 and ERK2, but the expression level of ERK1 and ERK2 was not altered (S1 Fig). Spatial learning and memory is enhanced in Ndfip1 cHet mice To further examine the role of Ndfip1 in spatial learning and memory formation, we have gen- erated the Ndfip1flox/WT mice as described in the Method Section. The strategy for generating the Ndfip1 cHet mice is shown in Fig 2A. For the purpose of obtaining the Ndfip1 cHet mice, one group of Ndfip1flox/WT mice received intra-hippocampal lenti-GFP-2A-NLS-Cre vector transduction. The other group of Ndfip1flox/WT mice received intra-hippocampal lenti-GFP- vector transduction and served as the control group. All the animals were subjected to water maze learning two weeks after lentiviral vector transduction. The schedule for lentivirus trans- duction, behavioral testing and biochemical assays is shown in Fig 2B. The expression of GFP (green) in the mouse CA1 area after lentivirus transduction is shown in Fig 2C. Results revealed that spatial acquisition performance was significantly improved in Ndfip1 cHet mice compared to the Ndfip1flox/WT mice (Fig 2D). These animals were subjected to probe trial test the next day after the last acquisition trial. Results revealed that Ndfip1 cHet mice spent more time in the target quadrant than that of the Ndfip1flox/WT mice (Fig 2E). They also made more travelling in the target quadrant than the Ndfip1flox/WT mice did (Fig 2F). But the swim speed of these two groups of mice is similar (S2A Fig). The same animals were also subjected to visible platform learning after the probe trial test. The result showed that the visible platform performance between these two groups of mice is not differ- ent (S2B Fig). These results indicated that the visual and motor functions were not altered in Ndfip1 cHet mice. Animals were sacrificed at the end of visible platform learning and their CA1 tissue was dissected out for determination of Ndfip1 protein expression. Result revealed that Ndfip1 expression level was significantly decreased in the Ndfip1 cHet mice compared to the Ndfip1flox/WT mice, but the GFP expression level is similar between these two groups of mice, indicating that the lentiviral vector was effectively transducted to these animals in similar amount (Fig 2G). Spatial training decreases the association between Ndfip1 and Nedd4 and decreases endogenous Beclin 1 ubiquitination in the hippocampus As mentioned above, Ndfip1 is an adaptor protein for Nedd4 [17], and we have found that Ndfip1 expression is decreased by spatial training; here we examined the relationship between PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 10 / 25 PLOS ONENdfip1 impairs spatial memory Fig 2. Spatial learning and memory is improved in Ndfip1 conditional heterozygous (cHet) mice. (A) Schematic illustration showing the strategy of generating the Ndfip1flox/WT mice. The inserted cassette is composed of a flippase recombination enzyme recognition target (FRT), neomycin resistance gene and Cre recombinase recognition target (loxP). The first loxP site is followed by the human β-actin promoter driving the neomycin resistance gene, a second FRT site and a second loxP site. A third loxP site is inserted downstream of the targeted Exon 3. Ndfip1 Exon 3 is flanked by loxP sites. Mice with Ndfip1 floxed allele were generated by crossing male chimeric mice containing FRT- neo-FRT-loxP cassette with female Act-Flp (flippase) mice. Ndfip1flox/WT mice were inbred to obtain Ndfip1flox/WT mice or conditional Ndfip1 (Ndfip1flox/flox) mice. Lenti-GFP vector or lenti-GFP-2A-NLS-Cre vector was transducted to the CA1 area of Ndfip1flox/WT mice. (B) Schedule of lentivirus transduction to Ndfip1flox/WT mice, behavioral testing and tissue dissection. (C) Immunohistochemistry showing the location of lenti-GFP vector transduction and GFP expression (green color) in the CA1 area. DAPI staining is shown in blue color. Scale bar is 100 μm (left panel). A picture at a higher magnification is shown in the right panel. Scale bar is 100 μm. (D) Ndfip1flox/WT mice that received lenti-GFP vector transduction (Ndfip1flox/WT mice) or lenti-GFP-2A-NLS-Cre transduction (Ndfip1 cHet mice) were PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 11 / 25 PLOS ONENdfip1 impairs spatial memory subjected to water maze learning two weeks later. The Ndfip1 cHet mice showed a better acquisition performance than the Ndfip1flox/WT mice (F1,16 = 24.66, p < 0.001). (E) Probe trial performance (t1,16 = 2.27, p < 0.05 for the target quadrant) and (F) Distance travelled in the target quadrant for the probe trial test (t1,16 = 2.39, p < 0.05) from the same animals as that in (D). The Ndfip1 cHet mice showed a better retention performance and travelled more in the target quadrant than the Ndfip1flox/WT mice (G) The Ndfip1 cHet mice showed a decreased Ndfip1 expression level than the Ndfip1flox/WT mice (t1,16 = 18.46, p < 0.001). The GFP expression level is similar for these two groups of mice. N = 9 each group. Data are expressed as individual values and mean ± SEM. * p < 0.05 and * p < 0.001. https://doi.org/10.1371/journal.pone.0283908.g002 Ndfip1 and Nedd4 associated with spatial learning. Co-IP experiment was conducted to exam- ine this issue. We first carried out a control experiment. The CA1 tissue lysates from non- trained and trained rats were immunoprecipitated with IgG and immunoblotted with anti- Nedd4 and anti-Ndfip1 antibodies. Results showed that other than the light-chain and heavy- chain bands, no specific band was observed (Fig 3A). Next, the CA1 tissue lysates from differ- ent non-trained and trained rats were immunoprecipitated with anti-Ndfip1 antibody and immunoblotted with anti-Nedd4 and anti-Ndfip1 antibodies. Result indicated that Ndfip1 is associated with Nedd4 in the hippocampus, but this association is significantly decreased in trained animals compared with the non-trained animals. On the other hand, the Ndfip1 expression level is consistently decreased in both the IP product and the lysates from trained animals, but the Nedd4 expression level in the lysate is not altered (Fig 3B). Based on these results, we hypothesized that Ndfip1 may impair spatial memory through its interaction with Nedd4 and subsequent ubiquitination of Nedd4 target proteins, whereas these target proteins play a role in facilitating spatial memory formation. As described above, Beclin 1 is an ubiquitination target of Nedd4 [24], here we first exam- ined whether Ndfip1 regulates Beclin 1 ubiquitination by Nedd4. Different plasmids together with V5-Nedd4 (or V5-vector) plasmid were co-transfected to HEK293T cells. Cell lysate was immunoprecipitated with anti-Flag antibody and immunoblotted with anti-His antibody. Result revealed that enhanced Beclin 1 ubiquitination was observed when V5-Nedd4 was trans- fected to the cells compared with V5-vector transfection (S3A Fig). But Beclin 1 ubiquitination by Nedd4 was markedly decreased when Ndfip1 siRNA was co-transfected to HEK293T cells compared with control siRNA transfection (S3B Fig). Next, we examined whether Beclin 1 is an endogenous ubiquitination target of Nedd4 in the hippocampus and whether endogenous Beclin 1 ubiquitination is altered by spatial training. To examine the first issue, control siRNA or Nedd4 siRNA was transfected to the rat CA1 area. The CA1 tissue lysate was immunoprecip- itated with anti-Beclin 1 antibody and immunoblotted with anti-ubiquitin antibody. Results showed that endogenous Beclin 1 ubiquitination level is markedly decreased by Nedd4 siRNA transfection. Nedd4 siRNA transfection also effectively decreased Nedd4 expression (Fig 3C). Next, we aimed to identify the role of Beclin 1 in spatial learning. We first conducted a con- trol experiment. The CA1 tissue lysates from non-trained and trained rats were immunopre- cipitated with IgG and immunoblotted with anti-ubiquitin antibody. Results showed that other than the IgG heavy-chain, no specific band was observed (Fig 3D). The CA1 tissue lysates from different non-trained and trained rats were immunoprecipitated with anti-Beclin 1 antibody and immunoblotted with anti-ubiquitin antibody. Result showed that endogenous Beclin 1 ubiquitination level is significantly decreased in trained animals compared with the non-trained controls (Fig 3E). Spatial training increases Beclin 1 expression and spatial learning and memory is impaired in Becn1 cKO mice Based on the result that spatial training decreases endogenous Beclin 1 ubiquitination in the hippocampus, we expect that spatial training should increase the expression level of Beclin 1 in PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 12 / 25 PLOS ONENdfip1 impairs spatial memory Fig 3. Spatial training decreases the association between Ndfip1 and Nedd4 and decreases endogenous Beclin 1 ubiquitination in the hippocampus. (A) The CA1 tissue lysates from non-trained and trained rats were immunoprecipitated with IgG and immunoblotted with anti-Nedd4 and anti-Ndfip1 antibodies to serve as the control group. No specific band was identified. (B) The CA1 tissue lysates from different non-trained and trained rats were immunoprecipitated with anti-Ndfip1 antibody and immunoblotted with anti-Nedd4 and anti-Ndfip1 antibodies. The association between Ndfip1 and Nedd4 is decreased in trained rats. The tissue lysate was also subjected to western blotting using anti-Ndfip1 and anti-Nedd4 antibodies. The Ndfip1 expression level is decreased in both the IP product and the lysate in trained rats. The quantified results from four independent experiments are shown in the right panel (t1,6 = 6.18 for Nedd4 expression, p < 0.001 and t1,6 = 8.23 for Ndfip1 expression, p < 0.001). (C) Control siRNA or Nedd4 siRNA (10 pmol) was transfected to the rat CA1 area. The CA1 tissue lysates from these animals were immunoprecipitated with anti-Beclin 1 antibody and immunoblotted with anti-ubiquitin and anti-Beclin 1 antibodies. Endogenous Beclin 1 ubiquitination level is decreased in Nedd4 siRNA-transfected rats (t1,6 = 5.78, p = 0.001). The same cell lysates were also subjected to western blot determination of Nedd4 expression, and Nedd4 expression level is decreased in Nedd4 siRNA-transfected rats (t1,6 = 9.26, p < 0.001). N = 4 each group. (D) The CA1 tissue lysates from non-trained and trained rats were immunoprecipitated with IgG and immunoblotted with anti-ubiquitin antibody to serve as the control group. No specific band was identified. (E) The CA1 tissue lysates from different non-trained and trained rats were immunoprecipitated with anti-Beclin 1 antibody and immunoblotted with anti-ubiquitin and anti- Beclin 1 antibodies. Endogenous Beclin 1 ubiquitination level is decreased in trained rats (t1,6 = 11.29, p < 0.001). N = 4 each group. Data are expressed as individual values and mean ± SEM. * p < 0.001. Ub: Ubiquitin. https://doi.org/10.1371/journal.pone.0283908.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 13 / 25 PLOS ONENdfip1 impairs spatial memory the hippocampus. This issue was examined here. The CA1 tissue lysates from non-trained and trained animals described above (from Fig 1F) were subjected to western blot determination of Beclin 1 expression. Result revealed that Beclin 1 expression level was significantly increased by spatial training (Fig 4A). Next, we examined the role of Beclin 1 in spatial memory forma- tion. The Becn1flox/flox mice were randomly divided to two groups. One group of mice received intra-hippocampal lenti-GFP-2A-NLS-Cre transduction and the other group of mice received intra-hippocampal lenti-GFP vector transduction. They were then subjected to water maze learning. The injection and behavioral testing paradigm is the same as that described in Fig 2B. Result revealed that spatial acquisition performance was significantly impaired in Becn1 cKO mice (Fig 4B). In addition, the Becn1 cKO mice spent less time in the target quadrant (Fig 4C) and they also made less travelling in the target quadrant during the probe trial test (Fig 4D). But the swim speed of these two groups of mice is similar (S2C Fig). These animals were also subjected to visible platform learning after the probe trial test. Result showed that the acquisi- tion performance between the Becn1 cKO mice and Becn1 loxp mice was not different (S2D Fig). This result indicated that the visual and motor functions were not altered in Becn1 cKO mice. Animals were sacrificed at the end of visible platform learning and their CA1 tissue was dissected out for determination of Beclin 1 expression. Western blot analysis indicated that Beclin 1 expression level is significantly decreased in Becn1 cKO mice, but GFP expression level is similar between these two groups of mice, indicating that the lentiviral vector was effec- tively transducted to these animals in similar amount (Fig 4E). Spatial training decreases endogenous PTEN ubiquitination and Pten cKO mice show impaired spatial learning and memory In this set of experiments, we examined whether spatial training alters endogenous PTEN ubi- quitination and the role of PTEN in spatial memory formation. To be related to Ndfip1, we first studied whether Ndfip1 regulates Nedd4-mediated PTEN ubiquitination. Different plas- mids together with V5-Nedd4 (or V5-vector) plasmid were co-transfected to HEK293T cells. Cell lysate was immunoprecipitated with anti-Flag antibody and immunoblotted with anti-His antibody. Result revealed that PTEN ubiquitination is increased when V5-Nedd4 plasmid was transfected compared with V5-vector transfection (S4A Fig). But PTEN ubiquitination by Nedd4 was markedly decreased when Ndfip1 siRNA was co-transfected to HEK293T cells compared with control siRNA transfection (S4B Fig). Next, we examined whether PTEN is an endogenous ubiquitination target of Nedd4 in the hippocampus. Control siRNA or Nedd4 siRNA was transfected to the rat CA1 area. The CA1 tissue lysate was immunoprecipitated with anti-PTEN antibody and immunoblotted with anti- ubiquitin antibody. Results showed that endogenous PTEN ubiquitination level is significantly decreased by Nedd4 siRNA transfection. Nedd4 siRNA transfection also decreased Nedd4 expression (Fig 5A). Next, we examined whether endogenous PTEN ubiquitination is associ- ated with spatial training. The hippocampal tissue lysates from non-trained and trained rats were immunoprecipitated with anti-PTEN antibody and immunoblotted with anti-ubiquitin antibody. Result revealed that endogenous PTEN ubiquitination level is significantly decreased in trained animals compared with the non-trained controls (Fig 5B). Based on the result that spatial training decreased endogenous PTEN ubiquitination in the hippocampus, we expect that spatial training should increase the expression level of PTEN in the hippocampus. We addressed this issue here. The CA1 tissue lysates from non-trained and trained animals described above (from Fig 1F) were subjected to western blot determination of PTEN expres- sion. Result showed that spatial training markedly increased the expression level of PTEN (Fig 5C). Lastly, we examined the role of PTEN in spatial memory formation by using the Pten PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 14 / 25 PLOS ONENdfip1 impairs spatial memory Fig 4. Spatial training increases Beclin 1 expression, and spatial learning and memory is impaired in Becn1 cKO mice. (A) Rats were subjected to water maze training or swimming control (non-trained) and were sacrificed three days later. Their CA1 tissue was subjected to western blot determination of Beclin 1 expression (N = 6 each group, the same animals in Fig 1F). Water maze training increased Beclin 1 expression level (t1,10 = 11.95, p < 0.001). (B) The Becn1 loxp mice (that received lenti-GFP vector transduction) and Becn1 cKO mice (that received lenti-GFP- 2A-NLS-Cre vector transduction) were subjected to water maze learning two weeks after lentivirus transduction. The Becn1 cKO mice showed impaired acquisition performance (N = 9 each group) (F1,16 = 33.75, p < 0.001). (C) Probe trial performance (t1,16 = 3.79, p < 0.01 for the target quadrant) and (D) Distance travelled in the target quadrant (t1,16 = 2.55, p < 0.05) from the same animals as that in (B). The Becn1 cKO mice showed impaired retention performance and made less travelling in the target quadrant than the Becn1 loxp mice. (E) The Beclin 1 expression level is decreased in the Becn1 cKO mice than the Becn1 loxp mice (t1,16 = 22.16, p < 0.001). The GFP expression level is similar for these two groups of mice. Data are expressed as individual values and mean ± SEM. * p < 0.05, p < 0.01 and * p < 0.001. https://doi.org/10.1371/journal.pone.0283908.g004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 15 / 25 PLOS ONENdfip1 impairs spatial memory Fig 5. Spatial training decreases endogenous PTEN ubiquitination and Pten cKO mice show impaired spatial learning and memory. (A) Control siRNA or Nedd4 siRNA (10 pmol) was transfected to the rat CA1 area. The CA1 tissue lysates from these animals were immunoprecipitated with anti-PTEN antibody and immunoblotted with anti- ubiquitin and anti-PTEN antibodies. Endogenous PTEN ubiquitination level is decreased in Nedd4 siRNA-transfected rats (t1,6 = 5.92, p = 0.001). The same cell lysates were also subjected to western blot determination of Nedd4 expression, and Nedd4 expression level is decreased in Nedd4 siRNA-transfected rats (t1,6 = 6.22, p < 0.001). N = 4 each group. (B) The CA1 tissue lysates from non-trained and trained rats were immunoprecipitated with anti-PTEN antibody and immunoblotted with anti-ubiquitin and anti-PTEN antibodies. Endogenous PTEN ubiquitination level is decreased in trained rats (t1,6 = 7.91, p < 0.001). N = 4 each group. (C) Rats were subjected to water maze training or swimming control (non-trained) and were sacrificed three days later. Their CA1 tissue was subjected to western blot determination of PTEN expression, and the PTEN expression level is increased in trained rats (t1,10 = 8.59, p < 0.001). N = 6 each group. (D) The Pten loxp mice (that received lenti-GFP vector transduction) and Pten cKO mice (that received lenti-GFP-2A-NLS-Cre vector transduction) were subjected to water maze learning two weeks after lentivirus transduction. The Pten cKO mice showed impaired acquisition performance (N = 9 each group) (F1,16 = 18.07, p < 0.001). (E) Probe trial performance (t1,16 = 3.36, p < 0.01 for the target quadrant) and (F) Distance travelled in the target quadrant (t1,16 = 4.23, p < 0.001) from the same animals as that in (D). The Pten cKO mice showed impaired retention performance and made less travelling in the target quadrant than the Pten loxp mice. (G) The PTEN expression level is decreased in the Pten cKO mice than the Pten loxp mice (t1,16 = 17.96, p < 0.001). The GFP expression level is similar for these two groups of mice. Data are expressed as individual values and mean ± SEM. p < 0.01 and * p≦0.001. Ub: Ubiquitin. https://doi.org/10.1371/journal.pone.0283908.g005 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 16 / 25 PLOS ONENdfip1 impairs spatial memory cKO mice. The Ptenflox/flox mice were randomly divided to two groups. One group of mice received intra-hippocampal lenti-GFP-2A-NLS-Cre transduction and the other group of mice received intra-hippocampal lenti-GFP vector transduction. They were then subjected to water maze learning as described above. Result revealed that the spatial acquisition performance was significantly impaired in Pten cKO mice (Fig 5D). These animals also spent less time in the tar- get quadrant (Fig 5E) and made less travelling in the target quadrant (Fig 5F) during the probe trial test. But the swim speed of these two groups of mice is similar (S2E Fig). These animals were also subjected to visible platform learning after the probe trial test. Result showed that the acquisition performance between the Pten cKO mice and Pten loxp mice was not different (S2F Fig). This result indicated that the visual and motor functions were not altered in Pten cKO mice. Animals were sacrificed at the end of visible platform learning and their CA1 tissue was dissected out for determination of PTEN expression. Western blot analysis revealed that PTEN expression level is significantly decreased in Pten cKO mice, but the GFP expression level is similar between these two groups of mice, indicating that the lentiviral vector was effec- tively transducted to these animals in similar amount (Fig 5G). Beclin 1 expression and PTEN expression are increased in Ndfip1 cHet mice The above results together showed that spatial learning and memory performance is enhanced in Ndfip1 cHet mice whereas it is impaired in Becn1 cKO and Pten cKO mice. These results also showed that Ndfip1 promotes Nedd4-mediated ubiquitination of Beclin 1 and PTEN. Although we have shown that spatial training decreased endogenous Beclin 1 and PTEN ubi- quitination in the hippocampus, it is not known whether Beclin 1 and PTEN are downstream effectors of Ndfip1 in Ndfip1-mediated memory impairment. To examine this issue, the CA1 tissue lysates from Ndfip1flox/WT control mice and Ndfip1 cHet mice that have been subjected to water maze learning (in Fig 2) were subjected to western blot determination of Beclin 1 and PTEN expression. Results revealed that both Beclin 1 expression level (Fig 6A and 6B) and PTEN expression level (Fig 6A and 6C) are significantly increased in Ndfip1 cHet mice com- pared to Ndfip1flox/WT mice. Discussion In the present study, we have found that the Ndfip1 cDNA fragment is differentially expressed between the fast learners and the slow learners from the water maze learning task with the fast learners showing a lower expression level. Spatial training decreases Ndfip1 mRNA and pro- tein expression in the hippocampus, whereas the Ndfip1 cHet mice show improved acquisition and retention performance. Both Beclin 1 and PTEN are endogenous ubiquitination targets of Nedd4 in the hippocampus. Spatial training decreases the association between Ndfip1 and Nedd4; meanwhile, it decreases endogenous Beclin 1 and PTEN ubiquitination in the hippo- campus. Consistent with these observations, spatial training increases the expression level of Beclin 1 and PTEN, but Becn1 cKO and Pten cKO mice both show impaired spatial learning and memory. Further, Beclin 1 and PTEN expression level is increased in the Ndfip1 cHet mice compared to the Ndfip1flox/WT control mice. Ubiquitination is an important post-translational modification that regulates various cellu- lar processes and physiological functions. The Nedd4 family E3 ubiquitin ligase is one of the WW-HECT domain E3 ubiquitin ligases that recognize the PY motif on other proteins [33]. Ndfip1 is a PY-containing protein and is an adaptor protein for Nedd4 family proteins. Thus, Ndfip1 is believed to promote Nedd4-mediated ubiquitination. In the present study, we have found that the association between Ndfip1 and Nedd4 is decreased in the hippocampus of PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 17 / 25 PLOS ONENdfip1 impairs spatial memory Fig 6. Beclin 1 expression and PTEN expression are increased in Ndfip1 cHet mice. (A) The CA1 tissue lysates from animals in Fig 2 (D-G) were subjected to western blot determination of Beclin 1 and PTEN expression. Both Beclin 1 and PTEN expression level is increased in Ndfip1 cHet mice than Ndfip1flox/WT mice. (B) The quantified result of Beclin 1 expression in Ndfip1flox/WT mice and Ndfip1 cHet mice (N = 9 each group) (t1,16 = 19.61, p < 0.001). (C) The quantified result of PTEN expression in Ndfip1flox/WT mice and Ndfip1 cHet mice (N = 9 each group) (t1,16 = 16.17, p < 0.001). Data are expressed as individual values and mean ± SEM. *** p < 0.001. https://doi.org/10.1371/journal.pone.0283908.g006 animals subjected to spatial training. It is conceivable that Ndfip1 activation of Nedd4 is also decreased in the hippocampus of trained animals. This speculation is partly supported by our findings that Ndfip1 siRNA decreased Nedd4-mediated ubiquitination of Beclin 1 and PTEN in the cell and that spatial training decreased endogenous Beclin 1 and PTEN ubiquitination in the hippocampus. It is also supported by our observations that Beclin 1 expression and PTEN expression are both increased in Ndfip1 cHet mice which showed enhanced spatial learning and memory performance. Although we have shown that Beclin 1 and PTEN are both the endogenous ubiquitination targets of Nedd4 in the hippocampus, our results do not exclude the possibility that Beclin 1 and PTEN are also the ubiquitination targets of other Nedd family proteins involved in Ndfip1-mediated memory impairment. Other than its role in the inhibition of tumorigenesis, Beclin 1 and post-translational modi- fications of Beclin 1 are well documented to play a role in autophagy regulation [24,34,35]. On PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 18 / 25 PLOS ONENdfip1 impairs spatial memory the other hand, autophagy in the hippocampus is suggested to be required for memory forma- tion in mice [36] and that inhibitory avoidance learning was shown to increase the levels of autophagy and a few lysosomal degradation proteins, including Beclin 1, in the rat hippocam- pus [25]. These results support our findings that Beclin 1 expression in the hippocampus is increased in animals subjected to water maze learning and that Becn1 cKO mice show impaired spatial learning and memory performance. But we further indicate that training- induced Beclin 1 expression is likely due to decreased Beclin 1 ubiquitination by spatial train- ing. Our results are also congruent with the role of Beclin 1 found in pathological memory [37]. For example, Beclin 1 expression level was found decreased in the brain of Alzheimer’s disease (AD) patients whereas overexpression of Beclin 1 reduces the accumulation of amy- loid-beta in an animal model of AD [38]. But few opposite results were also reported. For example, deficiency of activated C kinase was found to impair memory formation associated with upregulation of Beclin 1 [39]. The reason for this discrepancy requires clarification. PTEN is well known for its role as a tumor suppressor and Pten mutations were frequently found in various human cancers [40]. PTEN signaling also plays an important role in some neuronal functions and brain diseases [41]. In the present study, we have found that PTEN expression is important for spatial memory formation. Our results are consistent with the find- ings that PTEN deletion causes deficits in fear conditioning learning [42] and that spatial learning and LTP are impaired in PTENα-deficient mice [28]. Our results are also congruent with the reports that Pten cKO mice show dysregulated synaptic plasticity [27]. But we have provided novel mechanism that downregulation of PTEN expression is a downstream event of Ndfip1 signaling in association with Nedd4, and Ndfip1 expression is negatively regulated by spatial training. Further, we have found that spatial training decreased endogenous PTEN ubi- quitination in the hippocampus. However, the present results and results from the above stud- ies are incongruent with the observation that PTEN inhibition rescues cognitive impairment in APP/PS1 mice [43]. A possible explanation for this discrepancy is that for the latter study, the pharmacological tool adopted is targeted specifically at the PTEN and PDZ motif- dependent interactions, whereas for the study of Wang et al. [28], it is emphasized on the PTEN-CaMKII-NMDA receptor signaling. In addition, PTEN was found to dephosphorylate phosphatidylinositol 3,4,5-trisphosphate (PIP3) and negatively regulate phosphatidylinositol- 3-kinase (PI3K) signaling [44], but an early study has shown that PI3K activation is essential for fear memory formation [45]. These results do not conflict each other because PTEN could regulate signaling pathways other than PI3K signaling for memory processing, and the net result is facilitation of memory formation. The present results show that Ndfip1 impairs spatial learning and memory and this is prob- ably mediated through Nedd4-mediated ubiquitination of Beclin 1 and PTEN. These results implicate that Ndfip1 negatively regulates neuronal plasticity, but they are inconsistent with the observations that Ndfip1 is required for the development of neuronal dendrites and spines [46] and Ndfip1 is associated with neuronal survival upon brain injury [22]. Our results are also incongruent with the report that decreased Ndfip1 expression is associated with AD path- ogenesis [47]. We do not know the explanations for these discrepancies yet. It is possible that Ndfip1 interacts with different Nedd4 family proteins in these studies or that Ndfip1 may play different roles in different physiological and pathological conditions. It is also possible that Ndfip1 impairs learning and memory through Nedd4-mediated ubiquitination of other pro- teins in addition to Beclin 1 and PTEN. For example, alterations of AMPA receptor surface expression, trafficking and turnover are important mechanisms underlying synaptic plasticity [48,49], and AMPA receptor was found as an ubiquitination substrate of Nedd4 and AMPA receptor ubiquitination by Nedd4 results in decreased AMPA receptor surface expression and decreased excitatory synaptic transmission [50]. Further, we have previously shown that SGK PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 19 / 25 PLOS ONENdfip1 impairs spatial memory expression facilitates spatial learning and memory formation in rats [9]. SGK was ubiquiti- nated and degraded by Nedd4-2 [51] and that in part accounts for the low endogenous expres- sion level of SGK [52]. Based on these findings, it is reasonable to expect that downregulation of Ndfip1 by spatial training stabilizes SGK expression and facilitates memory formation. Moreover, Ndfip1 was found to recruit Nedd4-2 and mediates the ubiquitination of TrkB, a neurotrophin receptor that mediates brain-derived neurotrophic factor (BDNF) signaling [53], and conditional deletion of Ndfip1 increases TrkB expression in the hippocampus [54], whereas BDNF plays a critical role in mammalian learning and memory formation [55,56]. Because Ndfip1 is the adaptor protein for both Nedd4 (Nedd4-1) and Nedd4-2 [57], it is likely that more Nedd4 family substrate proteins are involved in spatial learning and memory forma- tion through the regulation by Ndfip1. On the other hand, another study has shown that spa- tial memory and LTP are both impaired in Nedd4 heterozygous mice [58]. We do not know the explanation for the discrepancy between this study and our study as well as the above stud- ies. It is possible that the effect of whole brain Nedd4 reduction is different from that of hippo- campal Nedd4 reduction in terms of spatial learning and memory processing. It is also possible that compensation mechanism may occur to Nedd4-2 that takes place the function of Nedd4 (Nedd4-1) due to reduced Nedd4 level in the brain. In the present study, we have adopted the Ndfip1flox/WT genotype of mice for experimentation. The reason is that most of the mice we obtained are the Ndfip1flox/WT mice and only few Ndfip1flox/flox mice were generated through inbreeding. In addition, the body size of the Ndfip1flox/flox mice was also smaller. We do not know the reason behind these phenomena, but it is not because DNA insertion of the loxP cassette prevents Ndfip1 protein expression because the Ndfip1 expression level is similar between the wild-type mice and the Ndfip1flox/WT mice (S5 Fig). In this study, the slow learners and non-trained rats both show a higher level of Ndfip1 expres- sion. A common mechanism may be due to the relatively low level of NMDA receptor activation because we have shown that NMDA administration downregulates Ndfip1 expression (S1 Fig). But the slow learners still gradually learn the task compared to the non-trained rats. There is a pos- sibility that some neurophysiological mechanisms, such as CA1 neuron field potential, might be different in these two groups of mice. Moreover, although we have found that spatial learning and memory is impaired in the Ndfip1 cHet mice compared to the Ndfip1flox/WT mice, the cellular and physiological mechanisms underlying Ndfip1-regulated memory impairment is still not known. It could be possible that synaptic plasticity and CA1 neuron field potential is increased in Ndfip1 cHet mice. It is also likely that Ndfip1 may indirectly downregulate the expression of certain pro- teins that are known to facilitate learning and memory formation, such as BDNF. The physiologi- cal role of Ndfip1 in negative regulation of memory formation requires further investigation. The same issue applies to the Becn1 cKO mice and Pten cKO mice, but the cellular mechanisms could be different from that of the Ndfip1 cHet mice. In summary, we have shown that Ndfip1 might be an important candidate gene in negative regulation of spatial memory formation, and this is associated with its interaction with Nedd4 and increased ubiquitination of Beclin 1 and PTEN in the hippocampus (Fig 7). Our results are consistent with the notion that protein degradation, other than protein synthesis, also underlies the mechanism of memory formation, but the physiological mechanisms underlying Ndfip1-regulated memory impairment remains to be elucidated. In addition, a previous study has shown that Ndfip1 is associated with AD [47]. Here, we have shown that Ndfip1 is involved in negative regulation of memory formation. In future studies, it is worth to examine the role and mechanism of Ndfip1 possibly involved in the pathogenesis or neuroprotection against AD. It is also worth to explore whether Ndfip1 may play a role in neurological disor- ders that are related to cognitive impairment, such as Rett syndrome. These studies might pro- vide novel therapeutic implication of cognitive impairment associated with these diseases. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 20 / 25 PLOS ONENdfip1 impairs spatial memory Fig 7. An illustration showing that spatial training induces decrease of Ndfip1 expression through NMDA receptor mediation, and it subsequently reduces the association between Ndfip1 and Nedd4, that results in decreased Beclin 1 and PTEN ubiquitination and enhanced spatial memory formation. https://doi.org/10.1371/journal.pone.0283908.g007 Supporting information S1 Fig. Acute NMDA administration decreases Ndfip1 expression in rat hippocampus. (TIF) S2 Fig. The swim speed for probe trial performance and visible platform learning perfor- mance of three genotypes of mice. (TIF) S3 Fig. Beclin 1 is an ubiquitination target of Nedd4 and Beclin 1 ubiquitination is decreased by Ndfip1 siRNA in HEK293T cells. (TIF) S4 Fig. PTEN is an ubiquitination target of Nedd4 and PTEN ubiquitination is decreased by Ndfip1 siRNA in HEK293T cells. (TIF) PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 21 / 25 PLOS ONENdfip1 impairs spatial memory S5 Fig. Ndfip1 expression level is not different between wild-type (WT) mice and Ndfip1flox/WT mice. (TIF) S6 Fig. (PDF) Acknowledgments Thanks are given to the Animal Core of IBMS and RNAi Core of Academia Sinica. Author Contributions Conceptualization: Wei-Lun Hsu, Yan-Chu Chen, Eminy H. Y. Lee. Data curation: Yun-Li Ma. Formal analysis: Wei-Lun Hsu, Yun-Li Ma, Yan-Chu Chen, Yen-Chen Liu, Kuang-Min Cheng. Funding acquisition: Eminy H. Y. Lee. Investigation: Wei-Lun Hsu, Yun-Li Ma, Yan-Chu Chen, Yen-Chen Liu, Kuang-Min Cheng. Methodology: Wei-Lun Hsu. Resources: Eminy H. Y. Lee. Supervision: Eminy H. Y. Lee. Validation: Wei-Lun Hsu, Eminy H. Y. Lee. Visualization: Eminy H. Y. Lee. Writing – original draft: Wei-Lun Hsu, Eminy H. Y. Lee. Writing – review & editing: Eminy H. Y. Lee. References 1. Davis HP, Squire LR. Protein synthesis and memory: a review. Psychol Bull. 1984; 96(3):518–559. PMID: 6096908. 2. Lee EHY, Hung HC, Lu KT, Chen WH, Chen HY. Protein synthesis in the hippocampus associated with memory facilitation by corticotropin-releasing factor in rats. Peptides. 1992; 13(5):927–937. https://doi. org/10.1016/0196-9781(92)90051-4 PMID: 1480516. 3. Santini E, Huynh TN, Klann E. Mechanisms of translation control underlying long-lasting synaptic plas- ticity and the consolidation of long-term memory. Prog Mol Biol Transl Sci. 2014; 122:131–167. https:// doi.org/10.1016/B978-0-12-420170-5.00005-2 PMID: 24484700. 4. Matsuo N, Reijmers L, Mayford M. Spine-type-specific recruitment of newly synthesized AMPA recep- tors with learning. Science. 2008; 319(5866):1104–1107. https://doi.org/10.1126/science.1149967 PMID: 18292343. 5. Tully T. Discovery of genes involved with learning and memory: an experimental synthesis of Hirschian and Benzerian perspectives. Proc Natl Acad Sci USA. 1996; 93(24):13460–13467. https://doi.org/10. 1073/pnas.93.24.13460 PMID: 8942957. 6. Castellucci VF, Kennedy TE, Kandel ER, Goelet P. A quantitative analysis of 2-D gels identifies proteins in which labeling is increased following long-term sensitization in Aplysia. Neuron. 1988; 1(4):321–328. https://doi.org/10.1016/0896-6273(88)90080-3 PMID: 3272738. 7. Cavallaro S, D’Agata V, Manickam P, Dufour F, Alkon DL. Memory-specific temporal profiles of gene expression in the hippocampus. Proc Natl Acad Sci USA. 2002; 99(25):16279–16284. https://doi.org/ 10.1073/pnas.242597199 PMID: 12461180. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 22 / 25 PLOS ONENdfip1 impairs spatial memory 8. Huang AM, Wang HL, Tang YP, Lee EH. Expression of integrin-associated protein gene associated with memory formation in rats. J Neurosci. 1998; 18(11):4305–4313. https://doi.org/10.1523/ JNEUROSCI.18-11-04305.1998 PMID: 9592107. 9. Tsai KJ, Chen SK, Ma YL, Hsu WL, Lee EHY. sgk, a primary glucocorticoid-induced gene, facilitates memory consolidation of spatial learning in rats. Proc Natl Acad Sci USA. 2002; 99(6):3990–3995. https://doi.org/10.1073/pnas.062405399 PMID: 11891330. 10. Ma YL, Tsai MC, Hsu WL, Lee EH. SGK protein kinase facilitates the expression of long-term potentia- tion in hippocampal neurons. Learn Mem. 2006; 13(2):114–118. https://doi.org/10.1101/lm.179206 PMID: 16585788. 11. 12. 13. 14. Tai DJC, Su CC, Ma YL, Lee EHY. SGK1 phosphorylation of IκB kinase α and p300 upregulates NF-κB activity and increases N-methyl-D-aspartate receptor NR2A and NR2B expression. J Biol Chem. 2009; 284(7):4073–4089. https://doi.org/10.1074/jbc.M805055200 PMID: 19088076. Tai DJC, Hsu WL, Liu YC, Ma YL, Lee EHY. Novel role and mechanism of protein inhibitor of activated STAT1 in spatial learning. EMBO J. 2011; 30(1):205–220. https://doi.org/10.1038/emboj.2010.290 PMID: 21102409. Fioravante D, Byrne JH. Protein degradation and memory formation. Brain Res Bull. 2011; 85(1– 2):14–20. https://doi.org/10.1016/j.brainresbull.2010.11.002 PMID: 21078374. Jarome TJ, Helmstetter FJ. Protein degradation and protein synthesis in long-term memory formation. Front Mol Neurosci. 2014; 7:61. https://doi.org/10.3389/fnmol.2014.00061 PMID: 25018696. 15. Cullen PK, Ferrara NC, Pullins SE, Helmstetter FJ. Context memory formation requires activity-depen- dent protein degradation in the hippocampus. Learn Mem. 2017; 24(11):589–596. https://doi.org/10. 1101/lm.045443.117 PMID: 29038220. 16. Jarome TJ, Devulapalli RK. The Ubiquitin-Proteasome System and Memory: Moving Beyond Protein Degradation. Neuroscientist. 2018; 24(6):639–651. https://doi.org/10.1177/1073858418762317 PMID: 29529924. 17. Mund T, Pelham HRB. Regulation of PTEN/Akt and MAP kinase signaling pathways by the ubiquitin ligase activators Ndfip1 and Ndfip2. Proc Natl Acad Sci USA. 2010; 107(25):11429–11434. https://doi. org/10.1073/pnas.0911714107 PMID: 20534535. 18. Putz U, Howitt J, Lackovic J, Foot N, Kumar S, Silke J, et al. Nedd4 family-interacting protein 1 (Ndfip1) is required for the exosomal secretion of Nedd4 family proteins. J Biol Chem. 2008; 283(47):32621– 32627. https://doi.org/10.1074/jbc.M804120200 PMID: 18819914. 19. Foot NJ, Leong YA, Dorstyn LE, Dalton HE, Ho K, Zhao L, et al. Ndfip1-deficient mice have impaired DMT1 regulation and iron homeostasis. Blood. 2011; 117(2):638–646. https://doi.org/10.1182/blood- 2010-07-295287 PMID: 20959604. 20. Howitt J, Putz U, Lackovic J, Doan A, Dorstyn L, Cheng H, et al. Divalent metal transporter 1 (DMT1) regulation by Ndfip1 prevents metal toxicity in human neurons. Proc Natl Acad Sci USA. 2009; 106 (36):15489–15494. https://doi.org/10.1073/pnas.0904880106 PMID: 19706893. 21. Hammond VE, Gunnersen JM, Goh CP, Low LH, Hyakumura T, Tang MM, et al. Ndfip1 is required for the development of pyramidal neuron dendrites and spines in the neocortex. Cereb Cortex. 2014; 24 (12):3289–3300. https://doi.org/10.1093/cercor/bht191 PMID: 23897647. 22. Sang Q, Kim MH, Kumar S, Bye N, Morganti-Kossman MC, Gunnersen J, et al. Nedd4-WW domain- binding protein 5 (Ndfip1) is associated with neuronal survival after acute cortical brain injury. J Neu- rosci. 2006; 26(27):7234–7244. https://doi.org/10.1523/JNEUROSCI.1398-06.2006 PMID: 16822981. 23. He C, Levine B. The Beclin 1 interactome. Curr Opin Cell Biol. 2010; 22(2):140–149. https://doi.org/10. 1016/j.ceb.2010.01.001 PMID: 20097051. 24. Platta HW, Abrahamsen H, Thoresen SB, Stenmark H. Nedd4-dependent lysine-11-linked polyubiquiti- nation of the tumour suppressor Beclin 1. Biochem J. 2012; 441(1):399–406. https://doi.org/10.1042/ BJ20111424 PMID: 21936852. 25. Pandey K, Yu XW, Steinmetz A, Alberini CM. Autophagy coupled to translation is required for long-term memory. Autophagy. 2021; 17(7):1614–1635. https://doi.org/10.1080/15548627.2020.1775393 PMID: 32501746. 26. Wang X, Trotman LC, Koppie T, Alimonti A, Chen Z, Gao Z, et al. NEDD4-1 is a proto-oncogenic ubiqui- tin ligase for PTEN. Cell. 2007; 128(1):129–139. https://doi.org/10.1016/j.cell.2006.11.039 PMID: 17218260. 27. Takeuchi K, Gertner MJ, Zhou J, Parada LF, Bennett MVL, Zukin RS. Dysregulation of synaptic plastic- ity precedes appearance of morphological defects in a Pten conditional knockout mouse model of autism. Proc Natl Acad Sci USA. 2013; 110(12):4738–4743. https://doi.org/10.1073/pnas.1222803110 PMID: 23487788. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 23 / 25 PLOS ONENdfip1 impairs spatial memory 28. Wang P, Mei F, Hu J, Zhu M, Qi H, Chen X, et al. PTENα modulates CaMKII signaling and controls con- textual fear memory and spatial learning. Cell Reports. 2017; 19(12):2627–2641. https://doi.org/10. 1016/j.celrep.2017.05.088 PMID: 28636948. 29. 30. Tai DJC, Liu YC, Hsu WL, Ma YL, Cheng SJ, Liu SY, et al. MeCP2 SUMOylation rescues Mecp2- mutant-induced behavioural deficits in a mouse model of Rett syndrome. Nat Commun. 2016; 7:10552. https://doi.org/10.1038/ncomms10552 PMID: 26842955. Liu SY, Ma YL, Hsu WL, Chiou HY, Lee EHY. Protein inhibitor of activated STAT1 Ser503 phosphoryla- tion-mediated Elk-1 SUMOylation promotes neuronal survival in APP/PS1 mice. Br J Pharmacol. 2019; 176(11):1793–1810. https://doi.org/10.1111/bph.14656 PMID: 30849179. 31. Chao CC, Ma YL, Lee EHY. BDNF enhances Bcl-xL expression through protein kinase CK2-activated and NF-κB-mediated pathway in rat hippocampus. Brain Pathol. 2011; 21(2):150–162. https://doi.org/ 10.1111/j.1750-3639.2010.00431.x PMID: 20731656. 32. 33. Liu YC, Hsu WL, Ma YL, Lee EHY. Melatonin induction of APP intracellular domain 50 SUMOylation alleviates AD through enhanced transcriptional activation and Aβ degradation. Mol Ther. 2021; 29 (1):376–395. https://doi.org/10.1016/j.ymthe.2020.09.003 PMID: 32950104. Ingham RJ, Gish G, Pawson T. The Nedd4 family of E3 ubiquitin ligases: functional diversity within a common modular architecture. Oncogene. 2004; 23(11):1972–1984. https://doi.org/10.1038/sj.onc. 1207436 PMID: 15021885. 34. Xu HD, Qin ZH. Chapter 5: Beclin 1, Bcl-2 and Autophagy. Autophagy: Biology and Diseases, Advances in Experimental Medicine and Biology 1206 (ed. Qin ZH), Science Press and Springer Nature Singa- pore Pte Ltd. 2019. 35. Hill SM, Wrobel L, Rubinsztein DC. Post-translational modifications of Beclin 1 provide multiple strate- gies for autophagy regulation. Cell Death Differ. 2019; 26(4):617–629. https://doi.org/10.1038/s41418- 018-0254-9 PMID: 30546075. 36. Glatigny M, Moriceau S, Rivagorda M, Ramos-Brossier M, Nascimbeni AC, Lante F, et al. Autophagy is required for memory formation and reverses age-related memory decline. Current Biol. 2019; 29 (3):435–448.e8 https://doi.org/10.1016/j.cub.2018.12.021 PMID: 30661803. 37. Salminen A, Kaarniranta K, Kauppinen A, Ojala J, Haapasalo A, Soininen H, et al. Impaired autophagy and APP processing in Alzheimer’s disease: The potential role of Beclin 1 interactome. Progress in Neu- robiology. 2013; 106–107:33–54. https://doi.org/10.1016/j.pneurobio.2013.06.002 PMID: 23827971. 38. Pickford F, Masliah E, Britschgi M, Lucin K, Narasimhan R, Jaeger PA, et al. The autophagy-related protein beclin 1 shows reduced expression in early Alzheimer disease and regulates amyloid β accumu- lation in mice. Journal of Clinical Investigation. 2008; 118(6):2190–2199. https://doi.org/10.1172/ JCI33585 PMID: 18497889. 39. 40. Zhu J, Chen X, Song Y, Zhang Y, Zhou L, Wang L. Deficit of RACK1 contributes to the spatial memory impairment via upregulating BECLIN1 to induce autophagy. Life Sci. 2016; 151:115–121. https://doi. org/10.1016/j.lfs.2016.02.014 PMID: 26930371. Li J, Yen C, Liaw D, Podsypanina K, Bose S, Wang SI, et al. PTEN, a putative protein tyrosine phospha- tase gene mutated in human brain, breast, and prostate cancer. Science. 1997; 275(5308):1943–1947. https://doi.org/10.1126/science.275.5308.1943 PMID: 9072974. 41. Endersby R, Baker SJ. PTEN signaling in brain: neuropathology and tumorigenesis. Oncogene. 2008; 27(41):5416–5430. https://doi.org/10.1038/onc.2008.239 PMID: 18794877. 42. Lugo JN, Smith GD, Morrison JB, White J. Deletion of PTEN produces deficits in conditioned fear and increases fragile X mental retardation protein. Learn Mem. 2013; 20(12):670–673. https://doi.org/10. 1101/lm.032839.113 PMID: 24241749. 43. Knafo S, Sa´ nchez-Puelles C, Palomer E, Delgado I, Draffin JE, Mingo J, et al. PTEN recruitment con- trols synaptic and cognitive function in Alzheimer’s models. Nature Neuroscience. 2016; 19(3):443– 453. https://doi.org/10.1038/nn.4225 PMID: 26780512. 44. Maehama T, Dixon JE. The tumor suppressor, PTEN/MMAC1, dephosphorylates the lipid second mes- senger, phosphatidylinositol 3,4,5-trisphosphate. J Biol Chem. 1998; 273(22):13375–13378. https:// doi.org/10.1074/jbc.273.22.13375 PMID: 9593664. 45. Lin CH, Yeh SH, Lin CH, Lu KT, Leu TH, Chang WC, et al. A role for the PI-3 kinase signaling pathway in fear conditioning and synaptic plasticity in the amygdala. Neuron. 2001; 31(5):841–851. https://doi. org/10.1016/s0896-6273(01)00433-0 PMID: 11567621. 46. Hammond VE, Gunnersen JM, Goh CP, Low LH, Hyakumura T, Tang MM, et al. Ndfip1 is required for the development of pyramidal neuron dendrites and spines in the neocortex. Cereb Cortex. 2014; 24 (12):3289–3300. https://doi.org/10.1093/cercor/bht191 PMID: 23897647. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 24 / 25 PLOS ONENdfip1 impairs spatial memory 47. Tian J, Zheng W, Li XL, Cui YH, Wang ZY. Lower expression of Ndfip1 is associated with Alzheimer dis- ease pathogenesis through decreasing DMT1 degradation and increasing iron influx. Front Aging Neu- rosci. 2018; 10:165. https://doi.org/10.3389/fnagi.2018.00165 PMID: 29937728. 48. Malinow R, Malenka RC. AMPA receptor trafficking and synaptic plasticity. Annu Rev Neurosci. 2002; 25:103–126. https://doi.org/10.1146/annurev.neuro.25.112701.142758 PMID: 12052905. 49. Collingridge GL, Isaac JT, Wang YT. Receptor trafficking and synaptic plasticity. Nat Rev Neurosci. 2004; 5(12):952–962. https://doi.org/10.1038/nrn1556 PMID: 15550950. 50. 51. Lin A, Hou Q, Jarzylo L, Amato S, Gilbert J, Shang F, et al. Nedd4-mediated AMPA receptor ubiquitina- tion regulates receptor turnover and trafficking. J Neurochem. 2011; 119(1):27–39. https://doi.org/10. 1111/j.1471-4159.2011.07221.x PMID: 21338354. Zhou R, Snyder PM. Nedd4-2 phosphorylation induces serum and glucocorticoid-regulated kinase (SGK) ubiquitination and degradation. J Biol Chem. 2005; 280(6):4518–4523. https://doi.org/10.1074/ jbc.M411053200 PMID: 15576372. 52. Brickley DR, Mikosz CA, Hagan CR, Conzen SD. Ubiquitin modification of serum and glucocorticoid- induced protein kinase-1 (SGK-1). J Biol Chem. 2002; 277(45):43064–43070. https://doi.org/10.1074/ jbc.M207604200 PMID: 12218062. 53. Chao MV. Neurotrophins and their receptors: a convergence point for many signalling pathways. Nat Rev Neurosci. 2003; 4(4):299–309. https://doi.org/10.1038/nrn1078 PMID: 12671646. 54. Murray SS, Wong AW, Yang J, Li Y, Putz U, Tan SS, et al. Ubiquitin regulation of trk receptor trafficking and degradation. Molecular Neurobiology. 2019; 56(3):1628–1636. https://doi.org/10.1007/s12035- 018-1179-5 PMID: 29911254. 55. Ma YL, Wang HL, Wu HC, Wei CL, Lee EH. Brain-derived neurotrophic factor antisense oligonucleotide impairs memory retention and inhibits long-term potentiation in rats. Neuroscience. 1998; 82(4):957– 967. https://doi.org/10.1016/s0306-4522(97)00325-4 PMID: 9466420. 56. Lubin FD, Roth TL, Sweatt JD. Epigenetic regulation of BDNF gene transcription in the consolidation of fear memory. J Neurosci. 2008; 28(42):10576–10586. https://doi.org/10.1523/JNEUROSCI.1786-08. 2008 PMID: 18923034. 57. Yang B, Kumar S. Nedd4 and Nedd4-2: closely related ubiquitin-protein ligases with distinct physiologi- cal functions. Cell Death Differ. 2010; 17(1):68–77. https://doi.org/10.1038/cdd.2009.84 PMID: 19557014. 58. Camera D, Coleman HA, Parkington HC, Jenkins TA, Pow DV, Boase N, et al. Learning, memory and long-term potentiation are altered in Nedd4 heterozygous mice. Behavioural Brain Research. 2016; 303:176–181. https://doi.org/10.1016/j.bbr.2016.01.054 PMID: 26821291. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283908 April 6, 2023 25 / 25 PLOS ONE
10.1371_journal.pone.0283318	RESEARCH ARTICLE Effects of integrated hospital treatment on the default mode, salience, and frontal- parietal networks in anorexia nervosa: A longitudinal resting-state functional magnetic resonance imaging study Motoharu GondoID Shu Takakura1, Kazufumi Yoshihara1,5, Chihiro Morita1, Makoto Yamashita1, Sanami Eto5, Nobuyuki Sudo1,5 1, Keisuke Kawai1,2, Yoshiya Moriguchi3, Akio HiwatashiID 4, 1 Department of Psychosomatic Medicine, Kyushu University Hospital, Fukuoka, Japan, 2 Department of Psychosomatic Medicine, Kohnodai Hospital, National Center for Global Health and Medicine, Chiba, Japan, 3 Department of Behavioral Medicine, National Institute of Mental Health, National Center of Neurology and Psychiatry, Tokyo, Japan, 4 Department of Clinical Radiology, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan, 5 Department of Psychosomatic Medicine, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan gondo.motoharu.015@m.kyushu-u.ac.jp Abstract The psychopathology of patients with anorexia nervosa has been hypothesized to involve inappropriate self-referential processing, disturbed interoceptive awareness, and excessive cognitive control, including distorted self-concern, disregard of their own starvation state, and extreme weight-control behavior. We hypothesized that the resting-state brain net- works, including the default mode, salience and frontal-parietal networks, might be altered in such patients, and that treatment might normalize neural functional connectivity, with improvement of inappropriate self-cognition. We measured resting-state functional mag- netic resonance images from 18 patients with anorexia nervosa and 18 healthy subjects before and after integrated hospital treatment (nourishment and psychological therapy). The default mode, salience, and frontal-parietal networks were examined using independent component analysis. Body mass index and psychometric measurements significantly improved after treatment. Before treatment, default mode network functional connectivity in the retrosplenial cortex and salience network functional connectivity in the ventral anterior insula and rostral anterior cingulate cortex were decreased in anorexia nervosa patients compared with those in controls. Interpersonal distrust was negatively correlated with salience network functional connectivity in the rostral anterior cingulate cortex. Default mode network functional connectivity in the posterior insula and frontal-parietal network functional connectivity in the angular gyrus were increased in anorexia nervosa patients compared with those in controls. Comparison between pre- and post-treatment images from patients with anorexia nervosa exhibited significant increases in default mode network func- tional connectivity in the hippocampus and retrosplenial cortex, and salience network func- tional connectivity in the dorsal anterior insula following treatment. Frontal-parietal network a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Gondo M, Kawai K, Moriguchi Y, Hiwatashi A, Takakura S, Yoshihara K, et al. (2023) Effects of integrated hospital treatment on the default mode, salience, and frontal-parietal networks in anorexia nervosa: A longitudinal resting-state functional magnetic resonance imaging study. PLoS ONE 18(5): e0283318. https://doi.org/10.1371/journal.pone.0283318 Editor: Kenji Hashimoto, Chiba Daigaku, JAPAN Received: September 20, 2022 Accepted: March 6, 2023 Published: May 30, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0283318 Copyright: © 2023 Gondo et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Due to the limitations of the consent provided by the subjects in our study, we cannot disclose the data to the public. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 1 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa functional connectivity in the angular cortex showed no significant changes. The findings revealed that treatment altered the functional connectivity in several parts of default mode and salience networks in patients with anorexia nervosa. These alterations of neural func- tion might be associated with improvement of self-referential processing and coping with sensations of discomfort following treatment for anorexia nervosa. Only researchers who have formally applied to and been approved by the human research ethics committee of Kyushu University Hospital can access the data (ijkseimei@jimu.kyushu-u.ac.jp). Funding: This work was supported by MEXT KAKENHI Grant Number JP26460910, JP15K08921, and AMED Grant Number JP23dm0307104. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Introduction Anorexia nervosa (AN) is characterized by extremely low body weight, intense fear of weight gain, body image distortion, and extreme weight-control behavior. Individuals suffering from this condition exhibit impaired cognition when self-evaluating their body weight and shape, an inability to recognize the dangers posed by their current low body weight and malnutrition [1], high levels of alexithymia [2], and disturbances in social emotional functioning [3, 4]. Inappropriate self-referential processing, disturbed interoceptive awareness, and excessive cognitive control have been hypothesized to constitutes the basis of the psychopathology of patients with AN [5]. Excessive concentration on body shape and weight can restrict the spheres of life that patients with AN are able to engage in, such as important normative age- graded experiences and introspection, resulting in interpersonal deficits [6]. Understanding the brain function exhibited by patients with AN may be helpful for clarify- ing the underlying neurobiological mechanisms involved in the disorder. There is a need to increase understanding of the comprehensive neural networks of these psychopathologies in patients with AN. Therefore, we focused on intrinsic network functional connectivity (FC) in resting-state functional magnetic resonance imaging (rsfMRI) analysis, which maps function- ally connected brain networks, on the basis of spontaneous non-task related fluctuations of blood oxygen level-dependent (BOLD) signals in the resting brain [7, 8]. Resting-state net- works (RSNs) are sets of brain areas exhibiting strong FC (the cross-correlation between BOLD signals in different regions in the resting brain), which play specific functional roles in brain activity at rest [9, 10]. It has been suggested that self-referential emotional processing occurs in the default mode network (DMN) [10], which comprises the precuneus, posterior cingulate cortex (PCC), retro- splenial cortex (RSC), medial prefrontal cortex, lateral temporal cortex, inferior parietal lobule, and hippocampal formation [11], parts of which have been generally related to emotion pro- cessing [12]. The DMN exhibits vigorous activity during rest [13]. The DMN is associated with self-processing and self-consciousness, and may thus be relevant to introspection [11, 14, 15]. The DMN is deactivated antagonistically when the frontal-parietal network (FPN) is active, including the dorsolateral prefrontal cortex and parietal cortices [16], which underlie executive functioning, such as working memory and goal-oriented (top-down) cognition [17–19]. Another important network, the salience network (SN), plays a role in switching between the DMN and FPN. The SN consists of the anterior insula (AI) and anterior cingulate cortex (ACC); the former detects salient events of interoceptive and exteroceptive sensation and emo- tion, whereas the latter facilitates coping with these events [20]. We focused on the DMN and associated networks (i.e., the SN and FPN) to reveal the neural pathology of self-referential function, interoceptive awareness, and cognitive regulation in patients with AN. In previous studies of AN, analyses of the DMN have reported different results in subjects with AN in different states at different disease stages [21–26]. Individuals with AN in the dis- ease state have been reported to exhibit less DMN FC in the precuneus, whereas individuals PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 2 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa who had recovered from AN exhibited no difference in DMN FC compared with healthy con- trols (HCs) [25]. Previous studies of SN in patients with AN have also reported different results, possibly depending on the disease stage. Network FC involving the ACC was reported to be reduced in patients with current AN [24, 25] and participants who had recovered from AN [25] compared with that in HCs, whereas some other studies indicated no alteration in the SN FC in partici- pants who had recovered from AN [23]. In contrast, previous studies of the FPN, representing executive cognitive control, have reported consistent results across disease stages in AN, such as higher FC between the FPN and angular gyrus in both AN patients and in patients who had recovered from AN compared with that in HCs [23, 27]. However, all of the studies mentioned above were cross-sectional. In addition, because current and recovered patients were included in different groups in previous studies, it remains unclear whether individual patients exhibit changes in networks according to the progression of the disease state at the within-subject level. According to the hypothesis that resting-state FC in patients with AN changes with treat- ment, two previous studies investigated FC in these patients before and after inpatient weight restoration treatment for 2–4 weeks [28, 29]. In adolescents and young adults with AN, using seed-based analysis, FC of the nucleus accumbens with the orbitofrontal cortex was found to be higher in patients with AN before treatment than in HCs, and was decreased after treatment compared with before treatment [28, 29]. FC between networks (SN-FPN/SN-DMN/ DMN-FPN) was not affected by the treatment, whereas the connectivity between SN and FPN was reduced in patients with AN relative to HCs [28]. The treatment strategy adopted in these two previous studies, however, was a short-term, inpatient physical treatment focusing on weight gain, raising the question of whether the treatment had a sufficient therapeutic effect on cognitive changes or psychopathology in patients with AN. Furthermore, the analyses relied only on between-network connectivity, which is not always interpretable. Thus, it remains unclear whether each individual network (DMN, SN, and FPN) exhibited functional changes. The purpose of the current study was to clarify the longitudinal treatment effect of structural cognitive behavioral therapy standardized for AN, called the “cognitive behavioral approach with behavioral limitation,” which is a type of reinforcement therapy [30] that is typically con- ducted for 3–5 months in inpatient treatment. The treatment effect was measured not only using behavioral psychometric scales but also with within-network FC in each of the three dif- ferent RSNs identified by independent component analyses (ICA) on rsfMRI data. We hypothe- sized that (1) within-network connectivity in pre-treatment AN patients would be altered in the DMN, SN, and FPN compared with HCs and post-treatment AN patients; (2) these alterations would be associated with individual psychological outcomes related to respective function (self- referential processing, interoception, and cognitive regulation); and (3) pre-treatment AN-spe- cific alterations of RSN would be improved after treatment. Using these combined methods may be helpful for identifying effective treatment targets for AN in the future. Materials and methods Ethics The current study was approved by the human research ethics committee at Kyushu Univer- sity Hospital. Written informed consent for the patients’ treatment and these studies was obtained from all participants and from their parents or guardians of minors. All procedures involved in this work complied with the ethical standards of the relevant national and institu- tional committees on human experimentation, and with the Helsinki Declaration of 1975, as revised in 2008. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 3 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa Participants All participants with AN and HCs were girls or women, right-handed, and aged between 15 and 50 years old. Thirty AN patients consented to participate in this study and received stan- dardized hospital treatment for AN between 2011 and 2015 at our hospital. Patients were diag- nosed with AN according to the DSM-IV by clinicians specialized in eating disorders using the Mini-International Neuropsychiatric Interview, and patients with severe depression, sui- cidal risk, personality disorders, schizophrenia, or alcohol dependence were excluded. Patients with mild or moderate depression, anxiety, or obsessive-compulsive symptoms were included because these symptoms are often comorbid with AN. Patients with AN were allowed to take their required medications. Data were acquired from patients with AN on the first day and the last day of treatment (pre-AN and post-AN, respectively). Eight patients dropped out of the treatment. The data of four patients were excluded from analysis because of poor registration of the fMRI data. Thus, the final longitudinal sample was 18 patients with AN (eight restricting type, 10 binge eating/purging type). Eighteen HC subjects were recruited from the local community and were required to have no history of eating disorder or other mental illness. HCs were also required not to take any medications. No longitudinal data were collected from HCs. Integrated hospital treatment We treated patients with AN with inpatient therapy called the “cognitive behavioral approach with behavioral limitation” [30]. In this therapeutic approach, patients consented to setting a target body weight and undergoing behavioral limitation (e.g., S1 Table). In addition, patients take part in psychological interviews for their behavioral problems, while they are undergoing nourishment therapy with behavioral limitation. Patients initially received small meals, for easy ingestion. However, if a patient was unable to intake the minimum amount required for nourishment (35 kcal/kg body weight), nasogastric feeding was administered, with patient consent, to compensate for the lack of oral feeding. After confirmation by therapists that a patient was able to ingest the whole meal without difficulty, the amount of the meal was increased gradually by approximately 200 kcal/day, and nasogastric feeding was gradually reduced. Behavioral limitation plays a role in promoting introspection by controlling external stimulation. Gradual removal of behavioral limitations leads to gradual adaptation to real life. S1 Table shows an example of the behavioral limitations and the schedule for lifting them. From the start of the behavioral limitation until the target body weight is reached, the behav- ioral limitations were lifted step by step as a reward for every kg of weight gain. When a patient reached the target body weight, the next stage of therapy began, providing a rehearsal of real life. In parallel with the therapy, patients received counseling twice a week to learn how to deal with maladjusted behavior, cognition, and emotion. Group therapy and family counseling were also conducted. Through this combined therapy, patients were expected to realize and correct erroneous notions regarding slenderness, eating behavior and interpersonal relation- ships. They acquired the ability to notice and express their own emotions, and to think about their lives and interpersonal relationships. As necessary, patients were prescribed anti-depres- sants, anti-anxiety and/or mood stabilizers for mood disturbance and anti-depressants and/or anti-psychotics for severe obsessive behavior. Demographic and psychometric measurements Each participant’s body mass index (BMI) was measured, and a subset of participants com- pleted psychometric questionnaires including the self-rating depression scale (SDS) [31] (pre- AN, n = 15; post-AN, n = 14; HC, n = 18), eating disorder inventory (EDI) (subscales: drive PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 4 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa for thinness, bulimia, body dissatisfaction, ineffectiveness, perfectionism, interpersonal dis- trust, interoceptive awareness, and maturity fears) [32] (pre-AN, n = 17; post-AN, n = 17; HC, n = 18), and the 20-item Toronto alexithymia scale (TAS-20) (subscales: difficulty in identify- ing feelings, difficulty in describing feelings, and externally oriented thinking) [33, 34] (pre- AN, n = 17; post-AN, n = 17; HC, n = 18). Group comparisons of these participant characteris- tics and psychometric measurements were conducted on available data using two-sample t- tests (“pre-AN vs. HC” and “post-AN vs. HC”). Paired-sample t-tests were used to analyze a longitudinal subset of patients who completed questionnaires at pre- and post- AN (“pre-AN vs. post-AN,” SDS, n = 12; EDI, n = 16; TAS-20, n = 16) with IBM SPSS Statistics Version 23.0 (IBM SPSS Inc., Chicago, IL, USA). Brain image data acquisition Imaging was performed with a PHILIPS Achieva 3-Tesla scanner (Best, Netherlands). Subjects lay in a supine position, with foam pads fixing the head and earplugs inserted into the ears to reduce head motion and scanner noise. Resting state was defined as when the subject was not engaging in any specific cognitive task during fMRI scanning [35]. During the acquisition of rsfMRI, the subjects were instructed to remain still, relax, close their eyes, and not think any- thing in particular. Although fMRI was performed at the first procedure with instructions not to fall asleep before the scan, a sleep scale was not used. We obtained the resting-state func- tional scans using an echo-planar imaging sequence with the following parameters: 32 axial slices, repetition time = 1,793 ms, echo time = 40 ms, fractional anisotropy = 90˚, slice thick- ness/gap = 3/1 mm, field of view = 210 × 210 mm, resolution = 3 × 3 × 4 mm, and 160 volumes in total (4 minutes 54 seconds). High-resolution three-dimensional magnetization-prepared rapid gradient-echo T1-weighted images were acquired for anatomical localization with the following imaging parameters: 200 sagittal slices, repetition time = 7.0 ms, echo time = 3.2 ms, fractional anisotropy = 9˚, slice thickness/gap = 1/0 mm, field of view = 256 × 240 mm, resolu- tion = 1 × 1 × 1 mm (6 minutes 31.7 seconds). No participants exhibited structural abnormali- ties during visual inspection. Image data preprocessing Preprocessing of resting-state functional brain images was performed using SPM12 (Statistical Parametric Mapping, Wellcome Trust, UCL, UK). The first 10 volumes of functional images were removed to eliminate the non-equilibrium effects of magnetization. The raw images were converted to the neuroimaging informatics technology initiative (NIFTI) format. Realignment, slice timing correction, and spatial co-registration were performed. According to the exclusion criteria for head motion correction in previous resting state fMRI analyses [36, 37], transla- tional motion parameters were verified to be less than 1 functional voxel. Rotation motion parameters were verified to be less than 2 degrees. Co-registered images were spatially normal- ized to Montreal Neurological Institute space [38] with a voxel size of 3 × 3 × 3 mm3 using a standard template in SPM12. The normalized images were then smoothed with an 8 mm full- width at half-maximum Gaussian kernel. Independent component analysis RSN consists of brain regions in which neural activities are temporally correlated and consid- ered to be functionally interconnected. To identify RSNs, an ICA was performed to decompose the rsfMRI voxel-by-voxel signals into temporally independent hemodynamic patterns distrib- uted in different regions [39] using the Group ICA fMRI Toolbox, which operates in Matlab. The number of components was estimated using minimum description length criteria [40]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 5 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa The dimensionality of the preprocessed rsfMRI data from each subject was reduced using principal component analysis. An ICA using the infomax algorithm was then applied to the data [41]. For each subject, this ICA generated a volumetric map for each component (compo- nent image), which contains the contribution of each component’s time course to the BOLD signal in each voxel. The individual component images were reconstructed (back-reconstruc- tion using the GICA algorithm) and converted to z-values [39], which were fed into subse- quent second-level between-subject analyses, as described below. Component selection The spatial distribution of RSNs was identified using a template based on previous studies [42] obtained from 90 functional regions of interest (fROIs, https://findlab.stanford.edu/functional_ ROIs.html). The DMN template comprised the medial prefrontal cortex, PCC, RSC, and medial temporal lobe. The SN template was composed of the AI and dorsal ACC. The FPN template included the bilateral parietal cortex and dorsolateral prefrontal cortex (DLPFC). We chose the component with the highest correlation with the respective template mask. Mapping network-related connectivity by groups and treatment stages To map the three different RSNs’ connectivity by groups and treatment stages, whole-brain voxel-by-voxel one-sample t-tests were performed on individual z-transformed component images for each selected component in HCs, pre-AN patients, and post-AN patients (signifi- cant at a family-wise error [FWE]-corrected peak-level threshold of p < 0.05) (SPM-12). These RSN maps were used to create the RSN masks in the following analysis. Group and treatment effect on FC in resting-state networks To specify the brain regions with AN-specific alterations in FC within each of the three differ- ent RSNs of interest, individual component images were compared between pre-AN patients and HCs with second-level group analyses using voxel-by-voxel two-sample t-tests (SPM-12). We employed a cluster defined by an FWE-corrected peak-level threshold of p < 0.05. To ensure that the differences selectively reflected the FC within the network of interest, these analyses were restricted within the respective RSN masks that were created by overlap between the gray matter mask and the conjunction of network-related connectivities in pre-AN patients and HCs (significant at a FWE-corrected peak-level threshold of p < 0.05). In an exploratory analysis to identify the brain regions exhibiting a treatment effect on FC within each of three different RSNs, individual component images were compared between the pre-AN and post-AN groups in second-level group analyses using voxel-by-voxel paired t- tests (SPM-12). We employed a cluster defined by an FWE-corrected peak-level threshold of p < 0.05. To ensure that the differences reflected the FC within the network of interest selec- tively, these analyses were restricted within the respective RSN masks that were created from overlap between the gray matter mask and the conjunction of network-related connectivities in the pre-AN and post-AN groups (significant at an FWE-corrected peak-level threshold of p < 0.05). We used the gray matter mask from WFU PickAtlas (http://fmri.wfubmc.edu/ software/PickAtlas). Definition of regions of interest To further investigate the detailed features of the RSN connectivity in brain regions involved in AN pathology, and to determine whether such pathological connectivity would be changed with treatment, we defined the regions of interest (ROIs) that were specific to AN pathology. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 6 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa The spheres (5 mm radius) centering the peak of significant clusters were detected by compari- son between pre-AN patients and HC, and the detected spheres were defined as the ROIs spe- cific to AN pathology. In some cases, regions detected by the comparison were assumed to be small and consisted almost exclusively of peaks. To show that the peaks were not false positives and to correct for this possibility, the regions surrounding the peaks, including the peaks, were defined as the ROI. Next, we calculated the mean of component contribution values within the voxels in ROIs in the individual component image for pre-AN patients, HC, and post-AN patients using Marsbar software [43]. These values represent the strengths of network connec- tivity to whole brain. The individual mean component values at these ROIs were used for the following two analyses: a multiple regression analysis to identify demographic and psychomet- ric indices related to RSN FC in these ROIs in each of pre-AN patients and HC, and an analysis of treatment effects on RSN FC in these ROIs by comparing the post-AN and pre-AN groups. Relationship of RSN FC in ROIs to psychometric measurements Multiple regression analysis was used to evaluate the demographic and psychometric factors that most strongly influenced differences in RSN FC between AN patients and HCs in AN-spe- cific ROIs. Independent variables were set from indices that were significantly different between pre-AN individuals and HCs in psychometric measurements (SDS, EDI subscales, and TAS-20 subscales). Covariates were set from demographic indices (age, education, BMI, disease duration, and medication use). For a subset of participants who completed all psycho- metric measurements (i.e., pre-AN, n = 15, and HC, n = 18), we set mean contribution values within the ROIs as the dependent variables, psychometric measurements as independent vari- ables selected with the stepwise method, and demographic indices as covariates with the forced entry method using SPSS. Treatment effects on RSN FC within AN-specific ROIs In addition to the exploratory analysis of the treatment effect on RSN FC, we investigated the treatment effect on network connectivity within the AN-specific ROIs in each of the different networks of interest. The mean values in the ROIs in individual component images were com- pared between pre-AN and post-AN patients using paired-sample t-tests (SPM-12, Marsbar), thresholded at a significance level of q < 0.05 with false discovery rate correction for multiple comparisons [44]. Additionally, correlation between a change of RSN FC in the ROIs and increase in BMI by treatment was calculated. Results Demographic indices and psychometric measurements There were no significant differences in age between AN patients and HCs. However, HCs had a longer duration of education than patients. The pre-AN group had significantly lower BMI values than both the post-AN and HC groups. The post-AN group had significantly higher BMI values than the pre-AN group, but had significantly lower BMI values than the HC group (Table 1). Although the post-AN group tended to recover body weight loss, they did not recover enough to reach the normal weight level. In almost all psychometric measurements, the pre-AN group had more pathological charac- teristics than the HC group (Table 1). The pre-AN group exhibited higher levels of depressive symptoms (SDS), higher ED pathology (EDI total and all eight subscales) and higher alexithy- mia (TAS-20 total and its two subscales). There was no difference in TAS-20 externally ori- ented thinking. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 7 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa Table 1. Demographic and psychometric characteristics of study participants. Group; mean ± SD t, p value Characteristics Age, [range], Y BMI, kg/m2 Education, Y Disease duration, M Treatment period, D SDS EDI, Drive for thinness Bulimia Body dissatisfaction Ineffectiveness Perfectionism Interpersonal distrust Interoceptive awareness Maturity fears Total TAS-20, DIF DDF EOT Total Medication use, no. Patients with anorexia nervosa, n = 18 Healthy controls, n = 18 Pre Post 28.5 ± 9.9, [15–47] 13.4 ± 1.7 13.2 ± 2.1 110 ± 109 52.7 ± 9.1 8.5 ± 6.3 7.2 ± 7.7 13.6 ± 3.8 16.5 ± 5.6 4.8 ± 4.3 7.4 ± 3.6 12.1 ± 5.5 10.0 ± 5.3 80.0 ± 30.0 23.5 ± 4.7 18.5 ± 3.6 20.2 ± 3.6 62.2 ± 8.3 12 16.0 ± 1.2 124 ± 47 40.6 ± 8.3 4.5 ± 3.9 1.2 ± 2.1 9.1 ± 5.0 10.0 ± 6.5 3.2 ± 3.4 5.6 ± 3.7 4.1 ± 4.6 8.1 ± 5.7 45.8 ± 24.1 16.7 ± 5.2 16.5 ± 3.0 19.2 ± 3.3 52.4 ± 8.9 12 28.4 ± 7.5, [22–44] 20.2 ± 1.8 15.5 ± 0.5 0 39.8 ± 8.0 2.6 ± 4.2 1.7 ± 2.0 9.1 ± 6.3 4.9 ± 4.4 1.9 ± 2.6 3.0 ± 2.0 1.7 ± 2.1 5.4 ± 2.6 29.8 ± 19.4 14.3 ± 5.8 13.7 ± 3.8 18.2 ± 3.5 46.2 ± 10.0 0 Pre vs HC 0.19, 0.985 Post vs HC Pre vs Post 11.46, < 0.001 8.03, < 0.001 9.94, < 0.001 4.46, < 0.001 4.34, < 0.001 3.27, 0.003 3.08, 0.006 2.55, 0.015 6.86, < 0.001 2.39, 0.023 4.54, < 0.001 7.30, < 0.001 3.13, 0.005 5.91, < 0.001 5.15, < 0.001 3.80, 0.001 1.67, 0.104 5.15, < 0.001 0.05, 0.959 1.44, 0.160 0.06, 0.952 0.00, 0.999 2.68, 0.011 1.31, 0.198 2.48, 0.021 1.97, 0.061 1.76, 0.09 2.16, 0.038 1.30, 0.204 2.40, 0.022 0.88, 0.386 1.93, 0.063 3.16, 0.009 2.52, 0.023 3.13, 0.007 3.76, 0.002 4.33, 0.001 2.12, 0.051 1.94, 0.071 6.92, < 0.001 2.63, 0.019 5.11, < 0.001 3.23, 0.003 1.93, 0.063 1.37, 0.180 3.16, 0.003 Pre: anorexia nervosa in pre-treatment, Post: anorexia nervosa in post-treatment, HC: healthy control SD: standard deviation, BMI: body mass index, SDS: self-rating depression scale, EDI: eating disorder inventory, TAS-20: 20-item Toronto alexithymia scale, DIF: Difficulty identifying feelings, DDF: Difficulty describing feelings, EOT: Externally oriented thinking Pre vs HC, Post vs HC; Two-sample t-tests. Psychometric measurements were analyzed in a subset of participants who completed SDS (Pre, n = 15; Post, n = 14; HC, n = 18), EDI (Pre, n = 17; Post, n = 17; HC, n = 18), TAS-20 (Pre, n = 17; Post, n = 17; HC, n = 18). Pre vs Post: Paired t-tests. Psychometric measurements were analyzed in a subset of patients who completed the SDS (n = 12), EDI (n = 16) and TAS-20 (n = 16) at both the Pre and Post time points. The details of medication use are shown in S2 Table. https://doi.org/10.1371/journal.pone.0283318.t001 The treatment effect was shown not only by a gain in body weight, but also by improvement of most psychometric measurements. Post-AN scores were significantly decreased for the SDS, EDI total (subscales: drive for thinness, bulimia, body dissatisfaction, ineffectiveness, intero- ceptive awareness, and maturity fears), and TAS-20 total (subscale: difficulty in identifying feelings), whereas the treatment showed a trend-level effect on perfectionism, interpersonal distrust, and difficulty describing feelings (DDF). The results revealed that, after treatment, AN patients approached the level of HCs in some psychological measurements: the post-AN group did not significantly differ from the HC group in SDS scores, EDI scores (subscales: drive for thinness, bulimia, body dissatisfaction, perfection- ism, interoceptive awareness, and maturity fears), and TAS-20 total scores (difficulty in identifying feelings). However, significant differences in EDI ineffectiveness and EDI total remained between the post-AN and HC groups, despite a significant treatment effect. Significant differences in EDI interpersonal distrust and TAS-20 DDF also remained between the post-AN and HC groups, and a significant treatment effect was not found. This indicates that post-treatment AN patients still exhibited AN-related pathological tendencies for some psychological features. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 8 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa Fig 1. Default mode network FC maps in HC and pre-AN. Spatial maps are plotted as t statistics thresholded at p < 0.05 and are family-wise error-corrected. MPFC: medial prefrontal cortex, PI: posterior insula, RSC: retrosplenial cortex, PCC: posterior cingulate cortex, Prec: precuneus. See S1 Fig for post-AN DMN and other RSNs. https://doi.org/10.1371/journal.pone.0283318.g001 Component identification and RSN FC maps by groups and treatment stages ICA extracted 17 independent components, among which three components were identified as RSNs of interest (correlation with the network template, component 12 and DMN: r = 0.473, component 16 and SN: r = 0.388, component 3 and FPN: r = 0.510). RSN FC maps of interest are shown for each group (Fig 1 and S1 Fig). The DMN included clusters in the medial prefrontal cortex, PCC/precuneus/RSC, posterior insula (PI)/transverse temporal gyrus and hippocampus. The PI is not generally considered a major region of the DMN. How- ever, in our study, especially in AN, clusters in the PI were detected as part of the DMN (Fig 1 and S1 Fig). Clusters in the ACC and AI were observed in the SN (S1 Fig). Clusters in the right DLPFC and left angular gyrus (AG) were observed in the FPN (S1 Fig). Exploratory analyses of group differences and treatment effects on FC in RSNs Comparison between the pre-AN and HC groups revealed AN-specific alterations in the main regions of DMN, SN, and FPN. The pre-AN group showed significantly decreased DMN FC compared with HCs in the RSC (Table 2 and Fig 2A). In the SN, the pre-AN group exhibited significantly decreased FC compared with HCs in the ventral AI (vAI) and rostral ACC (rACC) (Table 2 and Fig 2B, 2C). The pre-AN group showed significantly higher DMN FC than HCs in the PI (Table 2 and Fig 2D), although PI is generally not the main region of DMN. In the FPN, pre-AN exhibited significantly increased FC compared with HC in the AG (Table 2 and Fig 2E). We found a treatment effect in a few regions within the DMN and SN by comparison of FC maps of interest between the pre-AN and post-AN patients. The post-AN group showed sig- nificantly increased DMN FC in the hippocampus (Table 2 and Fig 3A) and SN FC in the dor- sal AI (dAI) (Table 2 and Fig 3B) compared with the pre-AN group. No treatment effect was observed in the FPN. We did not identify any regions with FC that was significantly decreased by the treatment. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 9 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa Table 2. Group difference and treatment effect on FC in resting-state networks. Comparison; Network Region Vol. Tmax Peak coordinates (x,y,z) HC > pre-AN pre-AN > HC Default mode network Retrosplenial cortex Salience network Ventral anterior insula Rostral anterior cingulate cortex Default mode network Posterior insula Frontal-parietal network Angular gyrus post-AN > pre-AN Default mode network Hippocampus Salience network Dorsal anterior insula pre-AN > post-AN No suprathreshold clusters Peak-level threshold p < 0.05, family-wise error-corrected HC: healthy control, pre-AN: anorexia nervosa patient in pre-treatment post-AN: anorexia nervosa patient in post-treatment https://doi.org/10.1371/journal.pone.0283318.t002 1 2 1 2 11 1 1 4.51 4.59 4.92 4.69 5.44 5.37 5.53 3, −55, 8 −30, 14, −13 0, 41, −7 45, −19, 2 −39, −73, 35 27, −19, −16 −27, 17, 11 Relationship of RSN FC in AN-specific ROIs with psychometric measurements SN FC in the rACC ROI was positively correlated with education (β = 0.441, p = 0.036), and negatively correlated with EDI interpersonal distrust (β = −0.489, p = 0.008). FPN FC in the AG ROI was positively correlated with TAS-20 DDF (β = 0.354, p = 0.034). In the other ROIs, resting-state FC was not significantly related to demographic indices or psychometric measurements. Pre- and post-treatment comparison of RSN FC in AN-specific ROIs In addition to the exploratory analysis of treatment effects on RSN FC, we examined the treat- ment effects on resting-state FC in the AN-specific ROIs in each of the DMN, SN, and FPN. The group comparison of post-AN versus pre-AN exhibited significantly increased FC in the RSC within the DMN (Fig 4A). Treatment effects on the FC were observed as increased con- nectivity also in the vAI and rACC within the SN, and in the PI within the DMN, although these effects did not remain statistically significant after multiple comparison correction (Fig 4B–4D). Thus, the treatment did not lead to significant improvement of FPN FC in the AG (Fig 4E). A change of RSN FC in any ROI was not significantly correlated with an increase of BMI. Discussion The current study revealed several main findings, as follows: 1) AN-specific alterations in brain regions within the DMN, SN, and FPN (e.g., the RSC within the DMN, vAI, and the rACC within the SN) that exhibited lower FC in the pre-AN group compared with the HC group. The PI within the DMN and AG within the FPN exhibited high FC in pre-AN in com- parison with HC. 2) The SN FC in the rACC ROI was negatively correlated with EDI interper- sonal distrust. FPN FC in the AG ROI was positively correlated with TAS-20 DDF. 3) Exploratory analyses for the treatment effect (post-AN versus pre-AN) exhibited increased FC in the hippocampus within DMN and dAI within the SN. When we focused on AN-specific PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 10 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa Fig 2. Group difference of FC in resting-state networks. The pre-AN group exhibited less connectivity than HCs between (A) the DMN and the retrosplenial cortex (RSC) (peak coordinate: 3, −55, 8), (B) the SN and the ventral anterior insula (vAI) (−30, 14, −13), and (C) the SN and the rostral anterior cingulate cortex (rACC) (0, 41, −7). The pre-AN group showed higher connectivity than HCs between (D) the DMN and the posterior insula (PI) (45, −19, 2), and (E) the FPN and the angular gyrus (AG) (−39, −73, 35), Two-sample t-test. Peak-level threshold p < 0.05 family-wise error-corrected. https://doi.org/10.1371/journal.pone.0283318.g002 Fig 3. Treatment effect on FC in resting-state networks. The post-AN group showed higher connectivity compared with the pre-AN group between (A) the DMN and the hippocampus (27, −19, −16), and (B) the SN and the dorsal anterior insula (dAI) (−27, 17, 11), Paired t-test. Peak-level threshold p < 0.05 family-wise error-corrected. https://doi.org/10.1371/journal.pone.0283318.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 11 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa Fig 4. Comparison of FC of AN-specific ROIs in the pre- and post-treatment images from AN patients (with HCs as a reference). (A) Default mode network, retrosplenial cortex. (B) Salience network, ventral anterior insula. (C) Salience network, rostral anterior cingulate cortex. (D) Default mode network, posterior insula. (E) Frontal-parietal network, angular gyrus. ROI analysis, Paired t-test using Marsbar. * q < 0.05, false discovery rate correction for multiple comparisons. https://doi.org/10.1371/journal.pone.0283318.g004 ROIs, the treatment effect was shown as increased FC in the RSC within DMN. Our main hypotheses were supported, as follows: (1) altered within-network connectivity was observed in pre-treatment AN patients. (2) There were relationships between these alterations and indi- vidual psychological outcomes. (3) We observed a treatment effect on RSNs. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 12 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa First, our findings indicated that DMN FC in the RSC showed AN-specific changes and treatment effects. The RSC is known to be structurally and functionally connected to the hip- pocampus and plays a role in self-reference processing while accessing memory information, such as retrieval of episodic memory, autobiographical memory, navigation with imagination, thinking about the future, introspection, and theory of mind [45–50]. Functional alterations in the RSC have been reported in psychiatric disorders involving impaired self-referential func- tion, such as schizophrenia [51], bipolar disorder [52], post-traumatic stress disorder [53, 54], social anhedonia [55, 56], individuals with high-trait-anxiety [57], and autism [58]. The cur- rent results are consistent with a previous study of AN reporting lower DMN FC than HCs in the region containing the RSC with no significant differences between HC and recovered patients [25]. As our behavioral results also revealed a significant recovery in self-cognition function after treatment, low DMN FC in the RSC suggests that impaired self-cognition is a specific feature in AN patients with symptoms. Thus, clinical improvement might be associ- ated with amelioration of DMN FC in the RSC. The significant recovery of DMN FC in the RSC with our integrated treatment might depend on either nourishment therapy or psychotherapy, or both. This finding could not be explained by a simple correlation with weight gain. Previous studies reported that DMN FC was strengthened by psychotherapy. For example, cognitive behavioral therapy for patients with chronic pain increased the amplitude of low-frequency fluctuation in the cerebellum and PCC (close to the RSC, which is part of the DMN) and FC between these areas [59]. A focused attention meditation increased connectivity from the striatum to the PCC and RSC [60]. Not only recovery of body weight owing to nourishment but also the psychotherapy in our pro- gram appeared to exert a therapeutic effect on DMN FC. With an exploratory analysis, we found that DMN FC in the hippocampus was increased following treatment, even though it was not detected as AN-specific alteration in DMN FC. Hippocampal functional improvement might be associated with progress of introspection in patients with AN. Generally, the hippocampus is coupled with the DMN during memory retrieval [61], and plays an important role in emotional regulation [62]. The hippocampus is known to be structurally and functionally connected to the RSC [46, 47], so that the treat- ment-induced increase of DMN FC in the hippocampus observed in our study is considered to be related to improvement in the RSC. Several studies have reported that psychotherapy, including cognitive behavioral therapy, increased hippocampal function in patients with other psychological disorders [63], such as major depression [64] and post-traumatic stress disorder [65]. The increased DMN FC in the hippocampus observed in the current study might be also associated with improvement of neural function by repeated introspection in psychotherapy, which is coupled with the improvement of DMN FC in the RSC. The exploratory analysis also revealed that the SN FC in the dAI was increased following the treatment, despite the absence of AN-specific changes at baseline. The activity or FC in dAI has been reported to change after treatment for other diseases, such as schizophrenia [66] and major depressive disorder [67], suggesting that the SN FC change in the dAI observed in our study might not be specific to AN. The dAI is involved both in processing somatosensory inputs and decision-making in the initiation of behavior. This area integrates internal and external sensory information to coordinate brain network dynamics to initiate switches of the DMN and FPN [68–70], leading to behavioral changes. Improved SN FC in the dAI might pro- mote appropriate behavior in response to perceived physical or psychological discomfort, such as perception of hunger in AN, which could be associated with improvement of the symptoms of AN. The right PI revealed higher FC to DMN in pre-AN patients than in HCs. The interaction between the DMN and PI might be associated with abnormal somatic sensation in patients PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 13 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa with AN. The PI would be associated with pain and somatosensory processes [69, 71]. Patients with chronic low back pain showed high FC between the DMN and PI [72], which is regarded as a diseased phenomenon. Patients with AN also present bodily complaints frequently, so their high FC in the PI to the DMN might indicate the psychological modification of somato- sensory information [73], which relates to their mind-induced somatic symptoms. The vAI showed lower SN FC in pre-AN patients compared with that in HCs. These phe- nomena might be associated with impaired socio-emotional processing, such as emotional cognition and empathy in patients with AN [74–77]. The vAI is connected to the rostral ACC [70], which plays an important role in socio-emotional processing in the SN [20, 69, 78, 79]. Abnormal function in the vAI has been reported in other diseases involving socio-emotional function, including bipolar disorder [80] and bronchial asthma with depression [81]. SN FC of the rACC was lower in pre-AN patients compared with that in HCs, and was negatively corre- lated with interpersonal distrust. These findings have important implications for social malad- justment in patients with AN. The rACC is involved in empathy [82–84], and dysfunction in this region is believed to be related to maladaptive emotional processing and interpersonal stress [85, 86]. Hypofunction of the rACC was reported in several previous resting-state neuro- imaging studies of patients with AN [21, 24, 87]. The current study showed that FPN FC in the AG was higher in pre-AN patients than in HCs and was positively correlated with TAS-20 DDF. The FPN is associated with executive cognition control [17–19], which is reported to be excessive in patients with AN from a clinical perspective [27, 88–90]. Patients with AN have been found to exhibit predominant FPN activ- ity during set shifting and cognitive flexibility [88, 91], which may reflect that top-down cogni- tive control is dominant [89, 90, 92]. This excessive functioning might be an impediment to expression of fluid feelings. Previous resting-state network studies in both patients with AN and participants who had recovered from AN reported higher FPN FC in the AG [23, 27]. In the current study, post-AN patients showed no significant improvement. This neural function was not changed by the treatment, potentially indicating a neural trait in patients with AN. Taken together, our findings indicated that DMN FC in the RSC, which is involved in self-ref- erence and coping, showed significant changes with treatment, suggesting that this element is more “improvable.” This may have occurred because, among the various elements of our psycho- therapeutic inpatient treatment, the promotion of introspection was reflected by improvements in neurological functioning. It is generally considered that treatments focused on improvable func- tion are more likely to be effective. This view suggests that currently used treatments (i.e., enhanced cognitive behavior therapy, Maudsley model therapy, and focal psychodynamic psycho- therapy) contain elements of introspection, compensate for weak functions, and have credible therapeutic effects [6, 93, 94], and should be noted when future treatment is revised. Limitations First, the small sample size is a potential limitation of the current study, which may affect the generalizability of the results. Second, because we adopted an integrated treatment regimen involving nourishment, psychotherapy, and medical treatment, we were unable to attribute the observed effects to a specific elemental therapy. Psychotherapy, as well as changes in body weight and medication use, may have affected the change in FC. Third, many patients were mildly relieved by treatment, but had not fully recovered at the time of discharge. Fourth, we were unable to analyze how the neural change with treatment influenced the clinical course after discharge. Future studies with a larger sample size may be needed to investigate the improvement in RSN FC with therapy, differences in RSN FC depending on the subtypes or clinical characteristics, and the influence of treatment on long-term clinical consequences. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 14 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa Conclusion We acquired rsfMRI from patients with AN before and after undergoing integrated inpatient treatment. We analyzed RSNs of interest using ICA. Post-AN patients exhibited higher DMN FCs in the RSC and the hippocampus than pre-AN patients. These results might be related to the improvement of self-referential function and progress of introspection induced by the inte- grated treatment. Post-AN patients exhibited higher SN FC in the dAI than pre-AN patients. The increase in FC following treatment might promote appropriate coping with discomfort in emotion and sensation. The RSN FCs in AN-specific ROIs (the right PI in DMN, vAI/rACC in SN and AG in FPN), except RSC in DMN, did not show significant changes between before and after treatment. These phenomena might reflect trait neural pathology regarding abnor- mal somatic perception, difficulty in socio-emotional processing, and excessive cognitive con- trol/difficulty in describing feelings in patients with AN. The current findings help to elucidate pathology and treatment effects in RSN in patients with AN, which may play a critical role in setting targets for future treatment. Supporting information S1 Table. An example of behavioral limitation. (PDF) S2 Table. Details of medication use. (PDF) S1 Fig. Resting-state network FC maps of interest in each group. (PDF) Acknowledgments We thank all of the participants, the nursing staff who engaged in treatment, and the radiologi- cal technician. Author Contributions Conceptualization: Motoharu Gondo, Yoshiya Moriguchi. Data curation: Motoharu Gondo. Formal analysis: Motoharu Gondo. Funding acquisition: Motoharu Gondo, Keisuke Kawai, Kazufumi Yoshihara. Investigation: Motoharu Gondo, Chihiro Morita, Makoto Yamashita, Sanami Eto. Methodology: Motoharu Gondo, Yoshiya Moriguchi, Akio Hiwatashi. Project administration: Keisuke Kawai, Nobuyuki Sudo. Resources: Motoharu Gondo, Keisuke Kawai, Akio Hiwatashi, Shu Takakura, Chihiro Morita, Makoto Yamashita, Sanami Eto. Software: Motoharu Gondo. Supervision: Yoshiya Moriguchi. Validation: Keisuke Kawai, Nobuyuki Sudo. Visualization: Motoharu Gondo. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 15 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa Writing – original draft: Motoharu Gondo. Writing – review & editing: Yoshiya Moriguchi, Akio Hiwatashi, Shu Takakura, Kazufumi Yoshihara. References 1. American Psychiatric Association APAD-, Force. T. Diagnostic and Statistical Manual of Mental Disor- ders: DSM-5, 5th edn2013:[xliv, 947pp.]. 2. Jenkins PE, O’Connor H. Discerning thoughts from feelings: the cognitive-affective division in eating disorders. Eat Disord. 2012; 20(2):144–58. Epub 2012/03/01. https://doi.org/10.1080/10640266.2012. 654058 PMID: 22364345. 3. Oldershaw A, Hambrook D, Stahl D, Tchanturia K, Treasure J, Schmidt U. The socio-emotional pro- cessing stream in Anorexia Nervosa. Neurosci Biobehav Rev. 2011; 35(3):970–88. Epub 2010/11/13. https://doi.org/10.1016/j.neubiorev.2010.11.001 PMID: 21070808. 4. Schmidt U, Treasure J. Anorexia nervosa: valued and visible. A cognitive-interpersonal maintenance model and its implications for research and practice. The British journal of clinical psychology. 2006; 45 (Pt 3):343–66. Epub 2006/12/07. https://doi.org/10.1348/014466505x53902 PMID: 17147101. 5. Rawal A, Park RJ, Williams JM. Rumination, experiential avoidance, and dysfunctional thinking in eat- ing disorders. Behaviour research and therapy. 2010; 48(9):851–9. Epub 2010/07/06. https://doi.org/ 10.1016/j.brat.2010.05.009 PMID: 20598670; PubMed PubMed Central PMCID: PMC2923742. 6. Fairburn CG. Cognitive Behavior Therapy and Eating Disorders. New York: The Guilford Press; 2008. 7. Biswal BB, Mennes M, Zuo XN, Gohel S, Kelly C, Smith SM, et al. Toward discovery science of human brain function. Proceedings of the National Academy of Sciences of the United States of America. 2010; 107(10):4734–9. Epub 2010/02/24. https://doi.org/10.1073/pnas.0911855107 PMID: 20176931; PubMed PubMed Central PMCID: PMC2842060. 8. Greicius MD, Supekar K, Menon V, Dougherty RF. Resting-state functional connectivity reflects struc- tural connectivity in the default mode network. Cereb Cortex. 2009; 19(1):72–8. Epub 2008/04/12. https://doi.org/10.1093/cercor/bhn059 PMID: 18403396; PubMed Central PMCID: PMC2605172. 9. Damoiseaux JS, Rombouts SA, Barkhof F, Scheltens P, Stam CJ, Smith SM, et al. Consistent resting- state networks across healthy subjects. Proceedings of the National Academy of Sciences of the United States of America. 2006; 103(37):13848–53. Epub 2006/09/02. https://doi.org/10.1073/pnas. 0601417103 PMID: 16945915; PubMed Central PMCID: PMC1564249. 10. Gusnard DA, Raichle ME. Searching for a baseline: functional imaging and the resting human brain. Nat Rev Neurosci. 2001; 2(10):685–94. Epub 2001/10/05. https://doi.org/10.1038/35094500 PMID: 11584306. 11. Buckner RL, Andrews-Hanna JR, Schacter DL. The brain’s default network: anatomy, function, and rel- evance to disease. Ann N Y Acad Sci. 2008; 1124:1–38. Epub 2008/04/11. https://doi.org/10.1196/ annals.1440.011 PMID: 18400922. 12. Kober H, Barrett LF, Joseph J, Bliss-Moreau E, Lindquist K, Wager TD. Functional grouping and corti- cal-subcortical interactions in emotion: a meta-analysis of neuroimaging studies. NeuroImage. 2008; 42 (2):998–1031. Epub 2008/06/27. https://doi.org/10.1016/j.neuroimage.2008.03.059 PMID: 18579414; PubMed Central PMCID: PMC2752702. 13. Andrews-Hanna JR, Reidler JS, Sepulcre J, Poulin R, Buckner RL. Functional-anatomic fractionation of the brain’s default network. Neuron. 2010; 65(4):550–62. Epub 2010/03/02. https://doi.org/10.1016/j. neuron.2010.02.005 PMID: 20188659; PubMed Central PMCID: PMC2848443. 14. Soto D, Theodoraki M, Paz-Alonso PM. How the human brain introspects about one’s own episodes of cognitive control. Cortex. 2018; 107:110–20. Epub 2017/12/05. https://doi.org/10.1016/j.cortex.2017. 10.016 PMID: 29198443. 15. Mason MF, Norton MI, Van Horn JD, Wegner DM, Grafton ST, Macrae CN. Wandering minds: the default network and stimulus-independent thought. Science. 2007; 315(5810):393–5. Epub 2007/01/ 20. https://doi.org/10.1126/science.1131295 PMID: 17234951; PubMed Central PMCID: PMC1821121. 16. Dosenbach NU, Fair DA, Miezin FM, Cohen AL, Wenger KK, Dosenbach RA, et al. Distinct brain net- works for adaptive and stable task control in humans. Proceedings of the National Academy of Sciences of the United States of America. 2007; 104(26):11073–8. Epub 2007/06/20. https://doi.org/10.1073/ pnas.0704320104 PMID: 17576922; PubMed Central PMCID: PMC1904171. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 16 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa 17. Beaty RE, Cortes RA, Zeitlen DC, Weinberger AB, Green AE. Functional Realignment of Frontoparietal Subnetworks during Divergent Creative Thinking. Cereb Cortex. 2021. Epub 2021/04/26. https://doi. org/10.1093/cercor/bhab100 PMID: 33895837. 18. Spreng RN, Stevens WD, Chamberlain JP, Gilmore AW, Schacter DL. Default network activity, coupled with the frontoparietal control network, supports goal-directed cognition. NeuroImage. 2010; 53(1):303– 17. Epub 2010/07/06. https://doi.org/10.1016/j.neuroimage.2010.06.016 PMID: 20600998; PubMed Central PMCID: PMC2914129. 19. Corbetta M, Shulman GL. Control of goal-directed and stimulus-driven attention in the brain. Nat Rev Neurosci. 2002; 3(3):201–15. Epub 2002/05/08. https://doi.org/10.1038/nrn755 PMID: 11994752. 20. Uddin LQ. Salience processing and insular cortical function and dysfunction. Nat Rev Neurosci. 2015; 16(1):55–61. Epub 2014/11/20. https://doi.org/10.1038/nrn3857 PMID: 25406711. 21. Gaudio S, Olivo G, Beomonte Zobel B, Schioth HB. Altered cerebellar-insular-parietal-cingular subnet- work in adolescents in the earliest stages of anorexia nervosa: a network-based statistic analysis. Transl Psychiatry. 2018; 8(1):127. Epub 2018/07/08. https://doi.org/10.1038/s41398-018-0173-z PMID: 29980676; PubMed Central PMCID: PMC6035187. 22. Phillipou A, Abel LA, Castle DJ, Hughes ME, Nibbs RG, Gurvich C, et al. Resting state functional con- nectivity in anorexia nervosa. Psychiatry Res Neuroimaging. 2016; 251:45–52. Epub 2016/04/26. https://doi.org/10.1016/j.pscychresns.2016.04.008 PMID: 27111812. 23. Boehm I, Geisler D, Tam F, King JA, Ritschel F, Seidel M, et al. Partially restored resting-state func- tional connectivity in women recovered from anorexia nervosa. J Psychiatry Neurosci. 2016; 41(6):377– 85. Epub 2016/10/22. https://doi.org/10.1503/jpn.150259 PMID: 27045551; PubMed Central PMCID: PMC5082508. 24. Gaudio S, Piervincenzi C, Beomonte Zobel B, Romana Montecchi F, Riva G, Carducci F, et al. Altered resting state functional connectivity of anterior cingulate cortex in drug naive adolescents at the earliest stages of anorexia nervosa. Sci Rep. 2015; 5:10818. Epub 2015/06/05. https://doi.org/10.1038/ srep10818 PMID: 26043139; PubMed Central PMCID: PMC4455287. 25. McFadden KL, Tregellas JR, Shott ME, Frank GK. Reduced salience and default mode network activity in women with anorexia nervosa. J Psychiatry Neurosci. 2014; 39(3):178–88. Epub 2013/11/28. https:// doi.org/10.1503/jpn.130046 PMID: 24280181; PubMed Central PMCID: PMC3997603. 26. Cowdrey FA, Filippini N, Park RJ, Smith SM, McCabe C. Increased resting state functional connectivity in the default mode network in recovered anorexia nervosa. Human brain mapping. 2014; 35(2):483– 91. Epub 2012/10/04. https://doi.org/10.1002/hbm.22202 PMID: 23033154. 27. Boehm I, Geisler D, King JA, Ritschel F, Seidel M, Deza Araujo Y, et al. Increased resting state func- tional connectivity in the fronto-parietal and default mode network in anorexia nervosa. Front Behav Neurosci. 2014; 8:346. Epub 2014/10/18. https://doi.org/10.3389/fnbeh.2014.00346 PMID: 25324749; PubMed Central PMCID: PMC4183185. 28. Uniacke B, Wang Y, Biezonski D, Sussman T, Lee S, Posner J, et al. Resting-state connectivity within and across neural circuits in anorexia nervosa. Brain Behav. 2019; 9(1):e01205. Epub 2018/12/28. https://doi.org/10.1002/brb3.1205 PMID: 30590873; PubMed Central PMCID: PMC6373651. 29. Cha J, Ide JS, Bowman FD, Simpson HB, Posner J, Steinglass JE. Abnormal reward circuitry in anorexia nervosa: A longitudinal, multimodal MRI study. Human brain mapping. 2016; 37(11):3835–46. Epub 2016/06/09. https://doi.org/10.1002/hbm.23279 PMID: 27273474; PubMed Central PMCID: PMC5448464. 30. Amemiya N, Takii M, Hata T, Morita C, Takakura S, Oshikiri K, et al. The outcome of Japanese anorexia nervosa patients treated with an inpatient therapy in an internal medicine unit. Eating and weight disor- ders: EWD. 2011. Epub 2011/10/15. https://doi.org/10.3275/8034 PMID: 21997338. 31. Zung WW. A Self-Rating Depression Scale. Arch Gen Psychiatry. 1965; 12:63–70. Epub 1965/01/01. https://doi.org/10.1001/archpsyc.1965.01720310065008 PMID: 14221692. 32. Garner DM, Olmstead MP, Polivy J. Development and validation of a multidimensional eating disorder inventory for anorexia nervosa and bulimia. International Journal of Eating Disorders. 1983; 2(2):15–34. https://doi.org/10.1002/1098-108x(198321)2:2<15::aid-eat2260020203>3.0.co;2-6. 33. Bagby RM, Parker JD, Taylor GJ. The twenty-item Toronto Alexithymia Scale—I. Item selection and cross-validation of the factor structure. J Psychosom Res. 1994; 38(1):23–32. Epub 1994/01/01. https:// doi.org/10.1016/0022-3999(94)90005-1 PMID: 8126686. 34. Bagby RM, Taylor GJ, Parker JD. The Twenty-item Toronto Alexithymia Scale—II. Convergent, dis- criminant, and concurrent validity. J Psychosom Res. 1994; 38(1):33–40. Epub 1994/01/01. https://doi. org/10.1016/0022-3999(94)90006-x PMID: 8126688. 35. Biswal B, Yetkin FZ, Haughton VM, Hyde JS. Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. Magn Reson Med. 1995; 34(4):537–41. Epub 1995/10/01. https:// doi.org/10.1002/mrm.1910340409 PMID: 8524021. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 17 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa 36. 37. Trojsi F, Di Nardo F, D’Alvano G, Caiazzo G, Passaniti C, Mangione A, et al. Resting state fMRI analysis of pseudobulbar affect in Amyotrophic Lateral Sclerosis (ALS): motor dysfunction of emotional expres- sion. Brain imaging and behavior. 2022. Epub 2022/11/13. https://doi.org/10.1007/s11682-022-00744- 4 PMID: 36370302. Tang F, Li L, Peng D, Yu J, Xin H, Tang X, et al. Abnormal static and dynamic functional network con- nectivity in stable chronic obstructive pulmonary disease. Front Aging Neurosci. 2022; 14:1009232. Epub 2022/11/04. https://doi.org/10.3389/fnagi.2022.1009232 PMID: 36325191; PubMed Central PMCID: PMC9618865. 38. Collins DL, Neelin P, Peters TM, Evans AC. Automatic 3D intersubject registration of MR volumetric data in standardized Talairach space. J Comput Assist Tomogr. 1994; 18(2):192–205. Epub 1994/03/ 01. PMID: 8126267. 39. Calhoun VD, Adali T, Pearlson GD, Pekar JJ. A method for making group inferences from functional MRI data using independent component analysis. Human brain mapping. 2001; 14(3):140–51. Epub 2001/09/18. https://doi.org/10.1002/hbm.1048 PMID: 11559959. 40. Li YO, Adali T, Calhoun VD. A feature-selective independent component analysis method for functional MRI. Int J Biomed Imaging. 2007; 2007:15635. Epub 2008/02/22. https://doi.org/10.1155/2007/15635 PMID: 18288254; PubMed Central PMCID: PMC2233814. 41. Bell AJ, Sejnowski TJ. An information-maximization approach to blind separation and blind deconvolu- tion. Neural Comput. 1995; 7(6):1129–59. Epub 1995/11/01. https://doi.org/10.1162/neco.1995.7.6. 1129 PMID: 7584893. 42. Shirer WR, Ryali S, Rykhlevskaia E, Menon V, Greicius MD. Decoding subject-driven cognitive states with whole-brain connectivity patterns. Cereb Cortex. 2012; 22(1):158–65. Epub 2011/05/28. https:// doi.org/10.1093/cercor/bhr099 PMID: 21616982; PubMed Central PMCID: PMC3236795. 43. Matthew Brett J-LA, Romain Valabregue, Jean-Baptiste Poline, editor Region of interest analysis using an SPM toolbox. the 8th International Conference on Functional Mapping of the Human Brain; 2002 June 2–6; Sendai, Japan. 44. Benjamini Y, Hochberg Y. Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society: Series B (Methodological). 1995; 57(1):289– 300. https://doi.org/10.1111/j.2517-6161.1995.tb02031.x. 45. Kaboodvand N, Backman L, Nyberg L, Salami A. The retrosplenial cortex: A memory gateway between the cortical default mode network and the medial temporal lobe. Human brain mapping. 2018; 39 (5):2020–34. Epub 2018/01/25. https://doi.org/10.1002/hbm.23983 PMID: 29363256. 46. Vann SD, Aggleton JP, Maguire EA. What does the retrosplenial cortex do? Nat Rev Neurosci. 2009; 10(11):792–802. Epub 2009/10/09. https://doi.org/10.1038/nrn2733 PMID: 19812579. 47. Corcoran KA, Yamawaki N, Leaderbrand K, Radulovic J. Role of retrosplenial cortex in processing stress-related context memories. Behav Neurosci. 2018; 132(5):388–95. Epub 2018/06/08. https://doi. org/10.1037/bne0000223 PMID: 29878804; PubMed Central PMCID: PMC6188831. 48. Cato MA, Crosson B, Gokcay D, Soltysik D, Wierenga C, Gopinath K, et al. Processing words with emo- tional connotation: an FMRI study of time course and laterality in rostral frontal and retrosplenial corti- ces. J Cogn Neurosci. 2004; 16(2):167–77. Epub 2004/04/08. https://doi.org/10.1162/ 089892904322984481 PMID: 15068589. 49. Piefke M, Weiss PH, Zilles K, Markowitsch HJ, Fink GR. Differential remoteness and emotional tone modulate the neural correlates of autobiographical memory. Brain. 2003;126(Pt 3):650–68. Epub 2003/ 02/05. https://doi.org/10.1093/brain/awg064 PMID: 12566286. 50. Oddo S, Lux S, Weiss PH, Schwab A, Welzer H, Markowitsch HJ, et al. Specific role of medial prefrontal cortex in retrieving recent autobiographical memories: an fMRI study of young female subjects. Cortex. 2010; 46(1):29–39. Epub 2008/12/17. https://doi.org/10.1016/j.cortex.2008.07.003 PMID: 19084220. 51. Mitelman SA, Shihabuddin L, Brickman AM, Hazlett EA, Buchsbaum MS. Volume of the cingulate and outcome in schizophrenia. Schizophr Res. 2005; 72(2–3):91–108. Epub 2004/11/25. https://doi.org/10. 1016/j.schres.2004.02.011 PMID: 15560955. 52. Nugent AC, Milham MP, Bain EE, Mah L, Cannon DM, Marrett S, et al. Cortical abnormalities in bipolar disorder investigated with MRI and voxel-based morphometry. NeuroImage. 2006; 30(2):485–97. Epub 2005/11/01. https://doi.org/10.1016/j.neuroimage.2005.09.029 PMID: 16256376. 53. Sartory G, Cwik J, Knuppertz H, Schurholt B, Lebens M, Seitz RJ, et al. In search of the trauma mem- ory: a meta-analysis of functional neuroimaging studies of symptom provocation in posttraumatic stress disorder (PTSD). PloS one. 2013; 8(3):e58150. Epub 2013/03/29. https://doi.org/10.1371/journal.pone. 0058150 PMID: 23536785; PubMed Central PMCID: PMC3607590. 54. Liberzon I, Taylor SF, Amdur R, Jung TD, Chamberlain KR, Minoshima S, et al. Brain activation in PTSD in response to trauma-related stimuli. Biological psychiatry. 1999; 45(7):817–26. Epub 1999/04/ 15. https://doi.org/10.1016/s0006-3223(98)00246-7 PMID: 10202568. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 18 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa 55. Zhang R-T, Yang Z-Y, Wang Y-M, Wang Y, Yang T-X, Cheung EFC, et al. Affective forecasting in indi- viduals with social anhedonia: The role of social components in anticipated emotion, prospection and neural activation. Schizophr Res. 2020; 215:322–9. Epub 2019/10/11. https://doi.org/10.1016/j.schres. 2019.10.006 PMID: 31611042. 56. Yang Z-Y, Zhang R-T, Li Y, Wang Y, Wang Y-M, Wang S-K, et al. Functional connectivity of the default mode network is associated with prospection in schizophrenia patients and individuals with social anhe- donia. Prog Neuropsychopharmacol Biol Psychiatry. 2019; 92:412–20. Epub 2019/02/26. https://doi. org/10.1016/j.pnpbp.2019.02.008 PMID: 30822447. 57. Imperatori C, Farina B, Adenzato M, Valenti EM, Murgia C, Marca GD, et al. Default mode network alter- ations in individuals with high-trait-anxiety: An EEG functional connectivity study. J Affect Disord. 2019; 246:611–8. Epub 2018/12/24. https://doi.org/10.1016/j.jad.2018.12.071 PMID: 30605880. 58. Hogeveen J, Krug MK, Elliott MV, Solomon M. Insula-Retrosplenial Cortex Overconnectivity Increases Internalizing via Reduced Insight in Autism. Biological psychiatry. 2018; 84(4):287–94. Epub 2018/01/ 31. https://doi.org/10.1016/j.biopsych.2018.01.015 PMID: 29523413. 59. Shpaner M, Kelly C, Lieberman G, Perelman H, Davis M, Keefe FJ, et al. Unlearning chronic pain: A randomized controlled trial to investigate changes in intrinsic brain connectivity following Cognitive Behavioral Therapy. Neuroimage Clin. 2014; 5:365–76. Epub 2014/01/01. https://doi.org/10.1016/j.nicl. 2014.07.008 PMID: 26958466; PubMed Central PMCID: PMC4749849. 60. Fujino M, Ueda Y, Mizuhara H, Saiki J, Nomura M. Open monitoring meditation reduces the involve- ment of brain regions related to memory function. Sci Rep. 2018; 8(1):9968. Epub 2018/07/04. https:// doi.org/10.1038/s41598-018-28274-4 PMID: 29967435; PubMed Central PMCID: PMC6028418. 61. Huijbers W, Pennartz CM, Cabeza R, Daselaar SM. The hippocampus is coupled with the default net- work during memory retrieval but not during memory encoding. PloS one. 2011; 6(4):e17463. Epub 2011/04/16. https://doi.org/10.1371/journal.pone.0017463 PMID: 21494597; PubMed Central PMCID: PMC3073934. 62. Fanselow MS, Dong HW. Are the dorsal and ventral hippocampus functionally distinct structures? Neu- ron. 2010; 65(1):7–19. Epub 2010/02/16. https://doi.org/10.1016/j.neuron.2009.11.031 PMID: 20152109; PubMed Central PMCID: PMC2822727. 63. Barsaglini A, Sartori G, Benetti S, Pettersson-Yeo W, Mechelli A. The effects of psychotherapy on brain function: a systematic and critical review. Prog Neurobiol. 2014; 114:1–14. Epub 2013/11/06. https:// doi.org/10.1016/j.pneurobio.2013.10.006 PMID: 24189360. 64. Goldapple K, Segal Z, Garson C, Lau M, Bieling P, Kennedy S, et al. Modulation of cortical-limbic path- ways in major depression: treatment-specific effects of cognitive behavior therapy. Arch Gen Psychia- try. 2004; 61(1):34–41. Epub 2004/01/07. https://doi.org/10.1001/archpsyc.61.1.34 PMID: 14706942. 65. Peres JF, Newberg AB, Mercante JP, Simao M, Albuquerque VE, Peres MJ, et al. Cerebral blood flow changes during retrieval of traumatic memories before and after psychotherapy: a SPECT study. Psy- chological medicine. 2007; 37(10):1481–91. Epub 2007/02/10. https://doi.org/10.1017/ S003329170700997X PMID: 17288648. 66. He H, Yang M, Duan M, Chen X, Lai Y, Xia Y, et al. Music Intervention Leads to Increased Insular Con- nectivity and Improved Clinical Symptoms in Schizophrenia. Front Neurosci. 2017; 11:744. Epub 2018/ 02/08. https://doi.org/10.3389/fnins.2017.00744 PMID: 29410607; PubMed Central PMCID: PMC5787137. 67. Delaveau P, Jabourian M, Lemogne C, Guionnet S, Bergouignan L, Fossati P. Brain effects of antide- pressants in major depression: a meta-analysis of emotional processing studies. J Affect Disord. 2011; 130(1–2):66–74. Epub 2010/10/30. https://doi.org/10.1016/j.jad.2010.09.032 PMID: 21030092. 68. Uddin LQ, Nomi JS, Hebert-Seropian B, Ghaziri J, Boucher O. Structure and Function of the Human Insula. J Clin Neurophysiol. 2017; 34(4):300–6. Epub 2017/06/24. https://doi.org/10.1097/WNP. 0000000000000377 PMID: 28644199; PubMed Central PMCID: PMC6032992. 69. Chang LJ, Yarkoni T, Khaw MW, Sanfey AG. Decoding the role of the insula in human cognition: func- tional parcellation and large-scale reverse inference. Cereb Cortex. 2013; 23(3):739–49. Epub 2012/ 03/23. https://doi.org/10.1093/cercor/bhs065 PMID: 22437053; PubMed Central PMCID: PMC3563343. 70. Deen B, Pitskel NB, Pelphrey KA. Three systems of insular functional connectivity identified with cluster analysis. Cereb Cortex. 2011; 21(7):1498–506. Epub 2010/11/26. https://doi.org/10.1093/cercor/ bhq186 PMID: 21097516; PubMed Central PMCID: PMC3116731. 71. Craig AD. How do you feel? Interoception: the sense of the physiological condition of the body. Nat Rev Neurosci. 2002; 3(8):655–66. Epub 2002/08/03. https://doi.org/10.1038/nrn894 PMID: 12154366. 72. Loggia ML, Kim J, Gollub RL, Vangel MG, Kirsch I, Kong J, et al. Default mode network connectivity encodes clinical pain: an arterial spin labeling study. Pain. 2013; 154(1):24–33. Epub 2012/11/01. https://doi.org/10.1016/j.pain.2012.07.029 PMID: 23111164; PubMed Central PMCID: PMC3534957. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 19 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa 73. Wentz E, Gillberg IC, Gillberg C, Rastam M. Ten-year follow-up of adolescent-onset anorexia nervosa: physical health and neurodevelopment. Dev Med Child Neurol. 2000; 42(5):328–33. Epub 2000/06/16. https://doi.org/10.1017/s0012162200000578 PMID: 10855653. 74. Nalbant K, Kalayci BM, Akdemir D, Akgul S, Kanbur N. Emotion regulation, emotion recognition, and empathy in adolescents with anorexia nervosa. Eating and weight disorders: EWD. 2019; 24(5):825– 34. Epub 2019/09/02. https://doi.org/10.1007/s40519-019-00768-8 PMID: 31473987. 75. Morris R, Bramham J, Smith E, Tchanturia K. Empathy and social functioning in anorexia nervosa before and after recovery. Cogn Neuropsychiatry. 2014; 19(1):47–57. Epub 2013/05/24. https://doi.org/ 10.1080/13546805.2013.794723 PMID: 23697879. 76. Cserjesi R, Vermeulen N, Lenard L, Luminet O. Reduced capacity in automatic processing of facial expression in restrictive anorexia nervosa and obesity. Psychiatry research. 2011; 188(2):253–7. Epub 2011/01/07. https://doi.org/10.1016/j.psychres.2010.12.008 PMID: 21208661. 77. Kucharska-Pietura K, Nikolaou V, Masiak M, Treasure J. The recognition of emotion in the faces and voice of anorexia nervosa. Int J Eat Disord. 2004; 35(1):42–7. Epub 2004/01/06. https://doi.org/10. 1002/eat.10219 PMID: 14705156. 78. Kelly C, Toro R, Di Martino A, Cox CL, Bellec P, Castellanos FX, et al. A convergent functional architec- ture of the insula emerges across imaging modalities. NeuroImage. 2012; 61(4):1129–42. Epub 2012/ 03/24. https://doi.org/10.1016/j.neuroimage.2012.03.021 PMID: 22440648; PubMed Central PMCID: PMC3376229. 79. Kurth F, Zilles K, Fox PT, Laird AR, Eickhoff SB. A link between the systems: functional differentiation and integration within the human insula revealed by meta-analysis. Brain Struct Funct. 2010; 214(5– 6):519–34. Epub 2010/06/01. https://doi.org/10.1007/s00429-010-0255-z PMID: 20512376; PubMed Central PMCID: PMC4801482. 80. Ellard KK, Gosai AK, Felicione JM, Peters AT, Shea CV, Sylvia LG, et al. Deficits in frontoparietal acti- vation and anterior insula functional connectivity during regulation of cognitive-affective interference in bipolar disorder. Bipolar Disord. 2019; 21(3):244–58. Epub 2018/12/20. https://doi.org/10.1111/bdi. 12709 PMID: 30565822; PubMed Central PMCID: PMC6504612. 81. 82. Zhang Y, Yang Y, Bian R, Yin Y, Hou Z, Yue Y, et al. Abnormal Functional Connectivity of Ventral Ante- rior Insula in Asthmatic Patients with Depression. Neural Plast. 2017; 2017:7838035. Epub 2017/07/07. https://doi.org/10.1155/2017/7838035 PMID: 28680706; PubMed Central PMCID: PMC5478859. Tolomeo S, Christmas D, Jentzsch I, Johnston B, Sprengelmeyer R, Matthews K, et al. A causal role for the anterior mid-cingulate cortex in negative affect and cognitive control. Brain. 2016; 139(Pt 6):1844– 54. Epub 2016/05/18. https://doi.org/10.1093/brain/aww069 PMID: 27190027. 83. Etkin A, Buchel C, Gross JJ. The neural bases of emotion regulation. Nat Rev Neurosci. 2015; 16 (11):693–700. Epub 2015/10/21. https://doi.org/10.1038/nrn4044 PMID: 26481098. 84. Moriguchi Y, Decety J, Ohnishi T, Maeda M, Mori T, Nemoto K, et al. Empathy and judging other’s pain: an fMRI study of alexithymia. Cereb Cortex. 2007; 17(9):2223–34. Epub 2006/12/08. https://doi.org/10. 1093/cercor/bhl130 PMID: 17150987. 85. Lane RD. Neural substrates of implicit and explicit emotional processes: a unifying framework for psy- chosomatic medicine. Psychosom Med. 2008; 70(2):214–31. Epub 2008/02/08. https://doi.org/10. 1097/PSY.0b013e3181647e44 PMID: 18256335. 86. Mulders PC, van Eijndhoven PF, Schene AH, Beckmann CF, Tendolkar I. Resting-state functional con- nectivity in major depressive disorder: A review. Neurosci Biobehav Rev. 2015; 56:330–44. Epub 2015/ 08/04. https://doi.org/10.1016/j.neubiorev.2015.07.014 PMID: 26234819. 87. Kojima S, Nagai N, Nakabeppu Y, Muranaga T, Deguchi D, Nakajo M, et al. Comparison of regional cerebral blood flow in patients with anorexia nervosa before and after weight gain. Psychiatry research. 2005; 140(3):251–8. Epub 2005/11/18. https://doi.org/10.1016/j.pscychresns.2005.08.002 PMID: 16288853. 88. Kaye WH, Wierenga CE, Bailer UF, Simmons AN, Bischoff-Grethe A. Nothing tastes as good as skinny feels: the neurobiology of anorexia nervosa. Trends Neurosci. 2013; 36(2):110–20. Epub 2013/01/22. https://doi.org/10.1016/j.tins.2013.01.003 PMID: 23333342; PubMed Central PMCID: PMC3880159. 89. Gaudio S, Wiemerslage L, Brooks SJ, Schioth HB. A systematic review of resting-state functional-MRI studies in anorexia nervosa: Evidence for functional connectivity impairment in cognitive control and visuospatial and body-signal integration. Neurosci Biobehav Rev. 2016; 71:578–89. Epub 2016/10/30. https://doi.org/10.1016/j.neubiorev.2016.09.032 PMID: 27725172. 90. Pignatti R, Bernasconi V. Personality, clinical features, and test instructions can affect executive func- tions in Eating Disorders. Eat Behav. 2013; 14(2):233–6. Epub 2013/04/06. https://doi.org/10.1016/j. eatbeh.2012.12.003 PMID: 23557828. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 20 / 21 PLOS ONEEffects of integrated hospital treatment on the resting-state brain networks in anorexia nervosa 91. Zastrow A, Kaiser S, Stippich C, Walther S, Herzog W, Tchanturia K, et al. Neural correlates of impaired cognitive-behavioral flexibility in anorexia nervosa. Am J Psychiatry. 2009; 166(5):608–16. Epub 2009/ 02/19. https://doi.org/10.1176/appi.ajp.2008.08050775 PMID: 19223435. 92. Ritschel F, Geisler D, King JA, Bernardoni F, Seidel M, Boehm I, et al. Neural correlates of altered feed- back learning in women recovered from anorexia nervosa. Sci Rep. 2017; 7(1):5421. Epub 2017/07/16. https://doi.org/10.1038/s41598-017-04761-y PMID: 28710363; PubMed Central PMCID: PMC5511172. 93. Schmidt U, Magill N, Renwick B, Keyes A, Kenyon M, Dejong H, et al. The Maudsley Outpatient Study of Treatments for Anorexia Nervosa and Related Conditions (MOSAIC): Comparison of the Maudsley Model of Anorexia Nervosa Treatment for Adults (MANTRA) with specialist supportive clinical manage- ment (SSCM) in outpatients with broadly defined anorexia nervosa: A randomized controlled trial. J Consult Clin Psychol. 2015; 83(4):796–807. Epub 2015/05/20. https://doi.org/10.1037/ccp0000019 PMID: 25984803. 94. Zipfel S, Wild B, Gross G, Friederich HC, Teufel M, Schellberg D, et al. Focal psychodynamic therapy, cognitive behaviour therapy, and optimised treatment as usual in outpatients with anorexia nervosa (ANTOP study): randomised controlled trial. Lancet. 2014; 383(9912):127–37. Epub 2013/10/18. https://doi.org/10.1016/S0140-6736(13)61746-8 PMID: 24131861. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283318 May 30, 2023 21 / 21 PLOS ONE
10.1371_journal.pone.0262053	RESEARCH ARTICLE Echocardiographic left ventricular stroke work index: An integrated noninvasive measure of shock severity Jacob C. JentzerID 1,2, Brandon M. Wiley1, Nandan S. Anavekar1 1 Department of Cardiovascular Medicine, Mayo Clinic, Rochester, Minnesota, United States of America, 2 Robert D. and Patricia E. Kern Center for the Science of Health Care Delivery, Mayo Clinic, Rochester, Minnesota, United States of America a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 jentzer.jacob@mayo.edu Abstract Background OPEN ACCESS Citation: Jentzer JC, Wiley BM, Anavekar NS (2022) Echocardiographic left ventricular stroke work index: An integrated noninvasive measure of shock severity. PLoS ONE 17(3): e0262053. https://doi.org/10.1371/journal.pone.0262053 Editor: Daniel A. Morris, Charite´ Universita¨tsmedizin Berlin - Campus Virchow- Klinikum: Charite Universitatsmedizin Berlin - Campus Virchow-Klinikum, GERMANY Received: August 19, 2021 Accepted: December 15, 2021 Published: March 9, 2022 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0262053 Copyright: © 2022 Jentzer et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Echocardiographic findings vary with shock severity, as defined by the Society for Cardio- vascular Angiography and Intervention (SCAI) shock stage. Left ventricular stroke work index (LVSWI) measured by transthoracic echocardiography (TTE) can predict mortality in the cardiac intensive care unit (CICU). We sought to determine whether LVSWI could refine mortality risk stratification by the SCAI shock classification in the CICU. Methods We included consecutive CICU patients from 2007 to 2015 with TTE data available to calcu- late the LVSWI, specifically the mean arterial pressure, stroke volume index and medial mitral E/e’ ratio. In-hospital mortality as a function of LVSWI was evaluated across the SCAI shock stages using logistic regression, before and after multivariable adjustment. Results We included 3635 unique CICU patients, with a mean age of 68.1 ± 14.5 years (36.5% females); 61.1% of patients had an acute coronary syndrome. The LVSWI progressively decreased with increasing shock severity, as defined by increasing SCAI shock stage. A total of 203 (5.6%) patients died during hospitalization, with higher in-hospital mortality among patients with lower LVSWI (adjusted OR 0.66 per 10 J/m2 higher) or higher SCAI shock stage (adjusted OR 1.24 per each higher stage). A LVSWI <33 J/m2 was associated with higher adjusted in- hospital mortality, particularly among patients with shock (SCAI stages C, D and E). Conclusions The LVSWI by TTE noninvasively characterizes the severity of shock, including both sys- tolic and diastolic parameters, and can identify low-risk and high-risk patients at each level of clinical shock severity. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 1 / 18 PLOS ONEFunding: The author(s) received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. LVSWI and SCAI shock stage Introduction Cardiogenic shock is a leading cause of morbidity and mortality in the cardiac intensive care unit (CICU) [1–3]. The presentation of cardiogenic shock (CS) varies across a continuum of severity as defined by the Society for Cardiovascular Angiography and Intervention (SCAI) shock stages classification [4,5]. Clinical studies have consistently demonstrated an association between higher shock severity, as represented by increasing SCAI shock stage A (At risk) to E (Extremis), and higher mortality in patients with CS, as well as CICU patients [6–15]. There- fore, accurate classification of CS severity is imperative in order to guide therapeutic interven- tions and optimize clinical outcomes [5]. Impaired cardiac hemodynamics characterize the pathophysiology of CS, making accurate evaluation of these parameters fundamental to the clinical assessment of patients presenting with CS [16]. Abnormal cardiac hemodynamic indices, measured either invasively with a pul- monary artery catheter or noninvasively using Doppler echocardiography, are associated with mortality risk in CS patients and CICU patients [10,14,15,17–21]. While several derived hemo- dynamic parameters such as cardiac power output (CPO) have been proposed to predict out- comes and guide therapy in patients with CS, no established hemodynamic marker exists for quantifying the severity of myocardial impairment across the spectrum of shock severity [10,16,18–22]. The left ventricular stroke work index (LVSWI) is a beat-by-beat assessment of myo- cardial systolic and diastolic function that integrates systemic hemodynamics to produce a comprehensive measure of cardiac performance [17,23]. The LVSWI can be calculated using Doppler echocardiography (ECHO-LVSWI) based on the stroke volume index (SVI) and ratio of mitral valve E velocity to medial mitral annulus e’ velocity (E/e’ ratio, used to estimate left ventricular filling pressures), and has been found to be strongly asso- ciated with mortality in CICU patients [17,23]. In a prior analysis, we demonstrated that patients with a low SVI or high E/e’ ratio had higher in-hospital mortality across the SCAI shock stages, making it likely that ECHO-LVSWI would be associated with mortality as well [10]. The ECHO-LVSWI appeared to decrease as the SCAI shock stage increased, sug- gesting a strong correlation with shock severity that might implicate ECHO-LVSWI as an integrated marker of hemodynamic compromise combining both systolic and diastolic left ventricular function [10]. Given the equipoise that exists regarding the ideal cardiac hemodynamics indices for defin- ing CS severity, we hypothesized that ECHO-LVSWI, as an integrated measurement reflecting overall myocardial performance, may better characterize CS severity when evaluated early dur- ing the clinical course. Therefore, we sought to evaluate the association of ECHO-LVSWI with in-hospital mortality across SCAI shock stages and to determine whether early assessment of this hemodynamic variable could augment risk-stratification. Methods Study population This study was approved by the Institutional Review Board of Mayo Clinic as posing minimal risk to patients and was performed under a waiver of informed consent. We retrospectively analyzed the index CICU admission of consecutive unique adult patients aged �18 years admitted to the CICU at Mayo Clinic Hospital St. Mary’s Campus between January 1, 2007 and December 31, 2015 who had a transthoracic echocardiogram (TTE) performed within 1 day before or after CICU admission [3,6–11,24–28]. We excluded patients who did not have available data to calculate the ECHO-LVSWI (i.e. MAP, SVI and E/e’ ratio). PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 2 / 18 PLOS ONELVSWI and SCAI shock stage Data sources We recorded demographic, vital sign, laboratory, clinical and outcome data, as well as proce- dures and therapies performed during the CICU and hospital stay [3,6–11,24–28]. The admis- sion value of all vital signs, clinical measurements and laboratory values was defined as either the first value recorded after CICU admission or the value recorded closest to CICU admis- sion. Admission diagnoses were defined as all International Classification of Diseases (ICD)-9 diagnostic codes within 1 day before or after CICU admission [3]. The Acute Physiology and Chronic Health Evaluation (APACHE)-III score, APACHE-IV predicted hospital mortality and Sequential Organ Failure Assessment score were automatically calculated with data from the first 24 hours of CICU admission using previously-validated electronic algorithms [3,24– 26]. The Charlson Comorbidity Index and individual comorbidities were extracted from the medical record using a previously-validated electronic algorithm [3,6–11,17,24–28]. Echocardiographic data The Mayo Clinic Echocardiography Database was queried and data extracted from the TTE performed closest to CICU admission, including vital signs at the time of TTE (S1 Table) [10,17]. One best LVEF value for each patient was determined using a hierarchical approach: volumetric LVEF calculated using Simpson’s biplane method was preferred, followed by monoplane volumetric approach, followed by linear methods and finally by visual estimation if these other methods were unavailable; the specific method used to measure the LVEF for each individual patient could not be determined [10,17]. We classified LVEF as mildly, moder- ately and severely reduced using gender-specific cut-offs as per current guidelines [29]. The mitral E/e’ ratio was used to estimate left ventricular end-diastolic pressure as LVEDP = 4.9 + 0.62 � mitral E/e’ ratio for calculation of ECHO-LVSWI using the formula ECHO-LVSWI = 0.0136 � stroke volume index (SVI) � (mean arterial pressure–LVEDP), as described by Choi, et al. (S2 Table) [17,23]. As per our prior study, we used the medial e’ veloc- ity to calculate ECHO-LVSWI (Fig 1); ECHO-LVSWI values were very similar when using either the lateral e’ velocity or mean e’ velocity (Pearson r correlation coefficients >0.99) [17]. Among patients with data for both medial and septal e’ velocities, there were no significant dif- ferences between the AUC values for discrimination of in-hospital mortality with ECHO-LVSWI calculated using the medial (0.756), lateral (0.751) or mean (0.754) e’ velocities Fig 1. Calculation of the ECHO-LVSWI (left) and an example using TTE data from a patient with cardiogenic shock (right). The stroke volume index (SVI) is calculated using the left ventricular outflow tract (LVOT) velocity-time integral (VTI) by spectral Doppler, indexed to the body surface area. The LVEDP is estimated using the ratio of the peak mitral early diastolic (E) wave velocity by spectral Doppler to the peak mitral early diastolic (e’) wave velocity by tissue Doppler via the formula LVEDP = 4.9 + 0.62 � mitral E/e’ ratio [17,23]. We used the medial/septal mitral e’ velocity for this analysis, although our data suggest that either the lateral or mean e’ velocity could be substituted. The mean arterial pressure (MAP) was determined either invasively or noninvasively and estimated as (systolic blood pressure + 2 � diastolic blood pressure) / 3. We used the formula ECHO-LVSWI = 0.0136 � SVI � (MAP–LVEDP) [17,23]. https://doi.org/10.1371/journal.pone.0262053.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 3 / 18 PLOS ONELVSWI and SCAI shock stage (all p >0.05 by De Long test; S1 Fig). Per Mayo Clinic Echocardiography Laboratory policy, mitral E velocities were not reported for patients with E-A fusion. Right atrial pressure was either recorded at the time of TTE (if measured invasively) or estimated based on inferior vena cava size and collapsibility [30]. Definition of SCAI shock stages We defined hemodynamic instability (including the need for inotropes), hypoperfusion (including the need for vasopressors), deterioration and refractory shock using data from CICU admission through the first 24 hours in the CICU (S3 Table) [6–11]. We mapped the five SCAI shock stages with increasing severity (A through E) using combinations of these var- iables, using an algorithm based on our prior analyses (S4 Table) [4,6–11]. Due to the small number of included patients in SCAI shock stage E, we grouped patients with SCAI shock stages D and E together for this analysis [7,10]. Statistical analysis In-hospital mortality was determined using electronic review of health records. Variables of interest were compared across the SCAI shock stages, and relevant analyses repeated in each SCAI shock stage. Categorical variables are reported as number (percentage) and the Pearson chi-squared test was used to compare groups; trends across the SCAI shock stages were deter- mined using logistic regression. Continuous variables are reported as mean ± standard devia- tion and Student’s t test was used to compare groups; trends across the SCAI shock stages were determined using linear regression. Classification and regression tree (CART) analysis was used to identify 4 risk groups using ECHO-LVSWI and SCAI shock stage. Discrimination of in-hospital mortality was assessed using area under the receiver-operator characteristic curve (AUC) values, which were compared using the De Long test. Logistic regression was used to determine odds ratio (OR) and 95% confidence interval (CI) values for prediction of in-hospi- tal mortality, before and after multivariable adjustment. Multivariable regression was per- formed using stepwise backward variable selection to minimize the value of the Akaike Information Criterion (AIC, a measure reflecting deviation from ideal model performance in the population). ECHO-LVSWI was analyzed as a continuous variable, dichotomized by the optimal cut-off, and according to prespecified categories. Candidate variables included demo- graphics, comorbidities, admission diagnoses, severity of illness scores (including SCAI shock stage), LVEF and procedures and therapies. Statistical analyses were performed using JMP Pro version 14.1.0 (SAS Institute, Cary, NC). Results Study population Out of a database of 10,004 unique CICU patients, we excluded 6,369 patients: 317 patients without an echocardiogram, 1,299 whose echocardiogram was not a TTE, 2,482 patients whose TTE was more than one day before or after CICU admission, and 2,271 patients whose TTE did not have data available to calculate the ECHO-LVSWI (Fig 2). The remaining 3,635 patients comprising the final study population had a mean age of 68.1 ± 14.5 years (36.5% females). Admission diagnoses included acute coronary syndrome in 61.1%, heart failure in 43.6%, cardiac arrest in 11.3%, cardiogenic shock in 9.4% and sepsis in 4.9%. The distribution of SCAI shock stages was: A, 50.8%; B, 27.9%; C, 15.6%; D, 5.3%; E, 0.4%. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 4 / 18 PLOS ONELVSWI and SCAI shock stage Fig 2. Flow diagram showing study inclusion and exclusion criteria. CICU, cardiac intensive care unit; LVSWI, left ventricular stroke work index; SCAI, Society for cardiovascular Angiography and Intervention; TTE, transthoracic echocardiogram. https://doi.org/10.1371/journal.pone.0262053.g002 Echocardiographic findings TTE was performed on the day of CICU admission in 42.9%. The mean LVEF was 48.3 ± 15.6%, and 51.8% patients had at least mildly reduced LVEF. The mean SVI was 40.7 ± 10.9 ml/m2, the mean E/e’ ratio was 15.9 ± 9.1, and the mean MAP at the time of the TTE was 82.9 ± 14.0 mmHg. The mean ECHO-LVSWI was 37.9 ± 13.8 J/m2, with a distribu- tion as follows (in J/m2): �50, 17.7%; 40–49, 23.2%; 30–39, 30.3%; 20–29, 20.1%; <20, 8.6%. Baseline characteristics varied substantially as a function of SCAI shock stage (Table 1), as did echocardiographic findings (Table 2). The ECHO-LVSWI decreased with increasing SCAI shock stage (Fig 3), with a distribution shifted toward lower values of ECHO-LVSWI (Fig 3); no patient in SCAI shock stage E had a LVSWI �40 J/m2. In-hospital mortality–unadjusted analyses A total of 203 (5.6%) patients died during hospitalization. The ECHO-LVSWI was lower among inpatient deaths compared to hospital survivors (27.0 versus 38.6 J/m2, p <0.001). ECHO-LVSWI was strongly and inversely associated with in-hospital mortality (unadjusted OR 0.453 per 10 J/m2 higher ECHO-LVSWI, 95% CI 0.396–0.518, p <0.001). The optimal ECHO-LVSWI cut-off for prediction of in-hospital death was 33.0 J/m2, and patients with ECHO-LVSWI <33.0 J/m2 had higher in-hospital mortality (10.9% versus 2.3%, unadjusted OR 5.112, 95% CI 3.709–7.046, p <0.001), accounting for 73.9% of in-hospital deaths. The association between ECHO-LVSWI and in-hospital mortality was weaker for patients with a heart rate >90 beats/minute (unadjusted OR 0.585 per 10 J/m2 higher ECHO-LVSWI, 95% CI 0.442–0.776, AUC 0.645) compared with patients who had a slower heart rate (unadjusted OR 0.461 per 10 J/m2 higher ECHO-LVSWI, 95% CI 0.392–0.543, AUC 0.740). LVSWI had a PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 5 / 18 PLOS ONETable 1. Baseline characteristics, comorbidities, admission diagnoses and therapies of patients according to SCAI shock stages. Data reported as mean ± standard deviation for continuous variables and number (percent) for categorical variables. P value is for linear regression (continuous variables) or logistic regression (categorical variables) across SCAI shock stages. Variable Stage A (n = 1845) Stage B (n = 1014) Stage C (n = 567) Stage D/E (n = 209) P value LVSWI and SCAI shock stage Demographics and outcomes Age Female gender White race CICU length of stay Hospital length of stay CICU mortality Hospital mortality Comorbidities Charlson Comorbidity Index History of MI History of HF History of CKD Prior dialysis Admission diagnoses ACS HF Cardiac arrest VF arrest Respiratory failure Sepsis Therapies and procedures Any vasoactive infusions Vasopressors Inotropes # vasoactives Peak VIS Peak NEE (mcg/kg/min) Invasive ventilator Noninvasive ventilator Dialysis in CICU CRRT IABP in CICU PAC in CICU Coronary angiogram PCI RBC transfusion In-hospital CPR Severity of illness APACHE-III score APACHE-IV predicted death (%) Day 1 SOFA score Non-cardiovascular SOFA Non-cardiovascular organ failure SIRS on admission Admission Braden score 67.6±14.2 579 (31.4%) 1722 (93.3%) 2.1±5.6 5.4±8.8 20 (1.1%) 41 (2.2%) 1.9±2.4 356 (19.3%) 207 (11.2%) 254 (13.8%) 41 (2.2%) 1202 (66.2%) 629 (34.6%) 132 (7.3%) 85 (4.7%) 191 (10.5%) 32 (1.8%) 164 (8.9%) 139 (7.5%) 52 (2.8%) 0.1±0.5 1.0±7.3 0.01±0.08 99 (5.4%) 187 (10.1%) 26 (1.4%) 1 (0.1%) 98 (5.3%) 38 (2.1%) 1333 (72.2%) 955 (51.8%) 102 (5.5%) 22 (1.2%) 50.4±17.7 9.1±10.5 2.1±1.8 1.1±1.6 201 (10.9%) 293 (15.9%) 18.6±2.8 67.3±15.0 418 (41.2%) 939 (92.6%) 2.5±2.4 7.0±7.5 27 (2.7%) 49 (4.8%) 2.2±2.5 197 (19.5%) 157 (15.5%) 166 (16.4%) 40 (3.9%) 564 (56.0%) 511 (50.7%) 110 (10.9) 69 (6.8%) 219 (21.8%) 63 (6.3%) 213 (21.0%) 192 (18.9%) 61 (6.0%) 0.3±0.8 3.5±12.9 0.03±0.13 138 (13.6%) 171 (16.9%) 32 (3.2%) 8 (0.8%) 116 (11.4%) 57 (5.6%) 632 (62.3%) 409 (40.3%) 110 (10.8%) 21 (2.1%) 59.8±20.2 15.0±15.8 3.1±2.6 1.9±2.3 212 (20.9%) 430 (42.4%) 17.6±3.2 71.2±14.4 234 (41.3%) 520 (91.7%) 2.5±2.3 6.9±7.6 30 (5.3%) 48 (8.5%) 2.6±2.7 120 (21.2%) 88 (15.5%) 118 (20.8%) 52 (9.2%) 325 (58.2%) 273 (48.9%) 80 (14.3%) 40 (7.2%) 114 (20.4%) 35 (6.3%) 55 (9.7%) 48 (8.5%) 13 (2.3%) 0.1±0.5 1.5±8.1 0.01±0.08 90 (15.9%) 85 (15.0%) 16 (2.8%) 5 (0.9%) 30 (5.3%) 12 (2.1%) 337 (59.4%) 194 (34.2%) 60 (10.6%) 12 (2.1%) 66.8±23.1 19.8±19.6 3.9±2.7 2.9±2.6 252 (44.4%) 231 (40.7%) 17.4±3.3 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 68.7±14.1 73 (34.9%) 192 (91.9%) 5.0±4.2 11.5±11.6 54 (25.8%) 65 (31.1%) 2.9±2.8 43 (20.6%) 55 (26.3%) 55 (26.3%) 29 (13.9%) 103 (49.8%) 153 (73.9%) 83 (40.1%) 41 (19.8%) 141 (68.1%) 47 (22.7%) 195 (93.3%) 181 (86.6%) 62 (29.7%) 2.0±1.2 34.9±58.8 0.33±0.59 135 (64.6%) 66 (31.6%) 33 (15.8%) 24 (11.5%) 57 (27.3%) 51 (24.4%) 117 (56.0%) 61 (29.2%) 70 (33.5%) 27 (12.9%) 96.0±31.0 46.5±28.9 9.3±3.9 6.5±3.3 168 (80.4%) 139 (66.5%) 14.3±3.5 <0.001 <0.001 0.16 <0.001 <0.001 <0.001 <0.001 <0.001 0.38 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 (Continued ) 6 / 18 PLOS ONELVSWI and SCAI shock stage Table 1. (Continued) Variable CardShock score Stage A (n = 1845) Stage B (n = 1014) Stage C (n = 567) Stage D/E (n = 209) 1.6±1.1 1.7±1.2 2.5±1.5 3.5±1.7 P value <0.001 Abbreviations: ACS, acute coronary syndrome; APACHE, Acute Physiology and Chronic Health Evaluation; CICU, cardiac intensive care unit; CKD, chronic kidney disease; CPR, cardiopulmonary resuscitation; HF, heart failure; IABP, intra-aortic balloon pump; MI, myocardial infarction; NEE, norepinephrine-equivalent dose; PAC, pulmonary artery catheter; PCI, percutaneous coronary intervention; RBC, red blood cell; SOFA, Sequential Organ Failure Assessment; VIS, Vasoactive-Inotropic Score. � Admission diagnoses are not mutually-exclusive and sum to greater than 100%. https://doi.org/10.1371/journal.pone.0262053.t001 higher AUC value (S2 Fig) than LVEF (AUC 0.662, p <0.001), cardiac index (AUC 0.604, p <0.001) and SVI (AUC 0.702, p = 0.06). When patients were grouped by quartiles of these var- iables, ECHO-LVSWI produced the greatest separation between high-risk and low-risk patients (S3 and S4 Figs and Table 3). At the optimal cut-off, LVSWI had the highest com- bined sensitivity and specificity (Table 4). In-hospital mortality increased with lower ECHO-LVSWI and higher SCAI shock stage (Fig 4), and patients with ECHO-LVSWI <33.0 J/m2 had higher in-hospital mortality in each SCAI shock stage (all p <0.05; Fig 4). Patients in each lower SCAI shock stage who had ECHO-LVSWI <33.0 J/m2 had similar in-hospital mortality as patients in the next higher SCAI shock stage with ECHO-LVSWI �33.0 J/m2 (all p >0.1). Patients in SCAI shock stage D/E with ECHO-LVSWI <33.0 J/m2 had the highest in-hospital mortality. ECHO-LVSWI was inversely associated with in-hospital mortality risk in each SCAI shock stage (Fig 5, all p <0.01). Echo-LVSWI alone had an AUC of 0.747 for discrimination of in-hospital mortality, which was equivalent to that for the SCAI shock stages (0.753, p = 0.78 by De Long test). The combination of ECHO-LVSWI and SCAI shock stage had an AUC of 0.803 for discrimination of in-hospital mortality, which was higher than either ECHO-LVSWI or SCAI shock stage alone (p <0.001). CART analysis separated patients into 4 risk groups based on ECHO-LVSWI (using the cut-off of 33.1 J/m2) and SCAI shock stage A/B/C versus D/E (Fig 6). In-hospital mortality increased from 2.0% among patients in SCAI shock stage A/B/C with ECHO-LVSWI �33.1 J/m2 to 35.0% among patients in SCAI shock stage D/E with ECHO-LVSWI <33.1 J/m2 (Fig 6). In-hospital mortality–multivariable analysis The final multivariable model had an AUC of 0.93 for discrimination of in-hospital mortality (Table 5). After adjustment, ECHO-LVSWI remained strongly and inversely associated with in-hospital mortality (adjusted OR 0.664 per 10 J/m2 higher ECHO-LVSWI, 95% CI 0.564– 0.782, p <0.001); ECHO-LVSWI had the second highest log worth in the model, after cardiac arrest. Patients with ECHO-LVSWI <33.0 J/m2 had higher adjusted in-hospital mortality (adjusted OR 2.232, 95% CI 1.507–3.305, p <0.001). When compared with patients who had ECHO-LVSWI <20 J/m2, patients in each higher ECHO-LVSWI group had lower adjusted in-hospital mortality (all p <0.05). When compared with patients who had ECHO-LVSWI �50 J/m2, patients in each lower ECHO-LVSWI group had higher adjusted in-hospital mor- tality (all p <0.05). When multivariable regression was repeated separately for each SCAI shock stage, a higher ECHO-LVSWI was associated with lower in-hospital mortality in each SCAI shock stage (all p <0.05 except SCAI shock stage B, p = 0.13; Fig 5). Patients with ECHO-LVSWI <33.0 J/m2 had higher adjusted in-hospital mortality in SCAI shock stage C (p <0.001) and D/E (p <0.05), but not in SCAI shock stage A (p = 0.17) or B (p = 0.81). PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 7 / 18 PLOS ONETable 2. Echocardiographic findings of patients according to SCAI shock stages. Data reported as mean ± standard deviation for continuous variables and number (percent) for categorical variables. P value is for linear regression (continuous variables) or logistic regression (categorical variables) across SCAI shock stages. Variable n with data Stage A (n = 1845) Stage B (n = 1014) Stage C (n = 567) Stage D/E (n = 209) P value LVSWI and SCAI shock stage Vital signs at TTE TTE on day of admission Systolic BP (mmHg) Diastolic BP (mmHg) Mean BP (mmHg) Pulse pressure (mmHg) Heart rate (BPM) Shock index Atrial fibrillation LV systolic function LVEDD LVESD Fractional shortening (%) LVEF (%) LVSD by ASE criteria Mild LVSD Moderate LVSD Severe LVSD Wall motion score index Lateral mitral s’ (cm/s) Systemic hemodynamics LVOT peak velocity (m/s) LVOT VTI (cm) SV (ml) SVI (ml/m2) LVSW (g�min) LVSWI (g�min/m2) LVSWI <33 g�min/m2 MCF CO (L/min) CI (L/min/m2) CPO (W) CPI (W/m2) SVR (dyne�s/cm5) SVR index (dyne�s/cm5�m2) LV diastolic function Mitral E velocity (cm/s) Mitral E/A ratio Mitral e’ velocity (cm/s) Mitral E/e’ ratio Mitral E DT (ms) RV function Estimated RAP (mmHg) Pressure-adjusted heart rate Peak TR velocity (m/s) Estimated RVSP (mmHg) 3635 3635 3635 3635 3635 3519 3519 3478 3492 3023 3019 3608 3608 2432 2676 3632 3635 3635 3635 3635 3635 3635 3201 3602 3602 3602 3602 3275 3275 3635 3031 3635 3635 3207 3302 3198 2896 2876 780 (42.3%) 122.5±20.3 67.1±12.8 85.6±13.0 55.4±18.8 68.8±12.9 0.58±0.15 114 (6.5%) 51.2±7.2 35.9±8.7 30.1±8.9 50.8±13.9 810 (45.2%) 356 (19.9%) 303 (16.9%) 151 (8.4%) 1.7±0.4 7.5±2.3 1.0±0.2 21.2±4.5 85.4±21.4 43.1±9.8 83.6±28.4 42.0±13.2 464 (25.2%) 0.45±0.15 5.7±1.4 2.9±0.7 1.1±0.3 0.5±0.2 1157±350 2260±657 0.8±0.3 1.1±0.6 6.1±2.2 14.9±8.3 202.1±54.8 7.8±4.2 6.5±4.2 2.7±0.5 38.5±13.8 440 (43.4%) 114.2±20.8 64.5±14.6 81.0±14.6 49.8±18.2 79.2±18.6 0.72±0.22 159 (17.8%) 51.7±8.0 38.3±10.6 26.9±10.5 46.4±16.4 531 (56.4%) 186 (19.8%) 174 (18.5%) 171 (18.2%) 1.8±0.5 7.3±2.6 1.0±0.2 19.3±5.1 76.0±23.3 38.9±11.5 68.4±26.7 34.8±12.8 517 (51.0%) 0.40±0.16 5.8±1.7 3.0±0.8 1.0±0.4 0.5±0.2 1072±568 2065±1122 0.9±0.3 1.3±0.7 6.2±2.4 16.1±9.2 182.2±51.4 9.7±5.0 10.0±6.7 2.7±0.5 50.9±13.7 241 (42.5%) 117.6±21.1 63.9±13.8 81.8±13.8 53.7±19.6 75 (16.6%) 0.66±0.21 83 (14.3%) 50.9±7.9 36.8±10.1 28.2±10.5 47.4±16.3 322 (54.0%) 112 (18.8%) 111 (18.6%) 99 (16.6%) 1.8±0.5 7.2±2.5 1.0±02 19. 6±5.0 76.3±23.0 39.7±11.1 69.3±27.6 35.9±13.3 229 (40.4%) 0.41±0.16 5.5±1.7 2.9±0.8 1.0±0.4 0.5±0.2 1143±421 2163±756 0.9±0.3 1.2±0.7 5.7±2.3 17.3±9.6 192.6±54.6 9.8±5.0 9.5±6.1 2.8±0.5 41.4±13.0 98 (46.9%) 105.0±19.6 59.0±13.4 74.3±13.5 46.0±17.2 79.9±19.0 0.79±0.26 48 (18.5%) 52.7±10.1 41.0±12.7 23.5±11.2 40.0±17.2 206 (74.1%) 54 (19.4%) 71 (25.5%) 81 (29.1%) 2.0±0.5 7.2±3.1 1.0±0.2 17.0±4.9 66.6±22.3 34.0±11.1 52.6±23.4 26.7±11.4 163 (78.0%) 0.34±0.14 5.1±1.8 2.6±0.9 0.8±0.3 0.4±0.2 1107±476 2158±940 0.8±0.3 1.3±0.8 5.1±2.1 18.8±11.0 177,9±51.9 13.1±5.3 14.3±7.2 2.7±0.6 43.7±14.2 0.34 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.09 <0.001 <0.001 <0.001 <0.001 <0.001 0.02 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.04 0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.52 <0.001 (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 8 / 18 PLOS ONELVSWI and SCAI shock stage Table 2. (Continued) Variable n with data Stage A (n = 1845) Stage B (n = 1014) Stage C (n = 567) Stage D/E (n = 209) Tricuspid s’ (cm/s) Global RV dysfunction Mild/mild-moderate Moderate/severe 2848 2073 11.9±3.2 401 (41.3%) 245 (25.2%) 156 (16.05) 11.4±3.5 305 (58.1%) 160 (30.5%) 145 (27.6%) 11.6±3.5 214 (59.4%) 110 (30.6%) 104 (28.9%) 10.2±4.2 169 (78.2%) 68 (31.5%) 101 (46.8%) P value <0.001 <0.001 Abbreviations: BP, blood pressure; CO, cardiac output; CI, cardiac index, CPO, cardiac power output; DT, deceleration time; LVEF, left ventricular ejection fraction; LVOT, left ventricular outflow tract; LVSW, left ventricular stroke work; LVSWI, left ventricular stroke work index; MCF, myocardial contraction fraction; RAP, right atrial pressure; RVSP, right ventricular systolic pressure; SV, stroke volume; SVI, stroke volume index; SVR, systemic vascular resistance; TR, tricuspid regurgitation; TTE, transthoracic echocardiogram; VTI, velocity-time integral. https://doi.org/10.1371/journal.pone.0262053.t002 Discussion The ECHO-LVSWI is a noninvasive measure of cardiac function that integrates relevant sys- tolic, diastolic and systemic parameters to quantify the degree of hemodynamic compromise in CICU patients with or at risk for CS. The ECHO-LVSWI assessed early in the clinical course is a powerful echocardiographic predictor of in-hospital mortality among CICU patients, even after adjusting for standard measures of shock severity and overall illness severity. The ECHO-LVSWI was inversely associated with in-hospital mortality risk, and patients at the extremes of ECHO-LVSWI (<20 J/m2 and �50 J/m2) had substantially different mortality than patients with intermediate values (with an optimal cut-off ~33 J/m2). When evaluated with respect to the SCAI stages of shock, ECHO-LVSWI demonstrated clear incremental value for risk stratification. As expected, ECHO-LVSWI valves progressively declined as the severity of shock worsened from SCAI shock stage C to D to E, reflecting wors- ening myocardial performance and systemic hemodynamics. Calculation of LVSWI using the mean invasive hemodynamic values reported by Thayer, et al. likewise demonstrates a drop in LVSWI as SCAI shock stage increases, supporting the validity of this finding [14]. What is more notable is that our analysis demonstrated that within each SCAI shock stage, decreasing ECHO-LVSWI values were associated with higher in-hospital mortality. Patients clinically classified into less severe SCAI shock stages who had paradoxically low ECHO-LVSWI had in- hospital mortality similar to patients in the next higher SCAI shock stage who had preserved ECHO-LVSWI. This finding suggests that ECHO-LVSWI can reclassify patients into higher- Fig 3. Mean ECHO-LVSWI (left) and distribution of ECHO-LVSWI (right) as a function of SCAI shock stage. https://doi.org/10.1371/journal.pone.0262053.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 9 / 18 PLOS ONELVSWI and SCAI shock stage Table 3. Unadjusted odds ratio (OR) and 95% confidence interval values for quartiles of selected echocardio- graphic for prediction of in-hospital mortality using univariable logistic regression, with Quartile 4 as referent. Median and interquartile range values defining the quartiles are as follows: ECHO LVSWI, 37.0 (21.0, 46.1) J/m2; CI, 2.8 (2.4, 3.3) L/min/m2; LVEF, 51 (36, 61) %; SVI, 41 (33, 47) ml/m2. Quartile LVEF CI SVI LVSWI 1 2 3 3.91 (2.52–6.05) 2.62 (1.74–3.95) 7.21 (4.31–12.05) 15.83 (7.68–32.64) 1.99 (1.23–3.20) 1.12 (0.70–1.80) 3.48 (2.04–5.93) 6.97 (3.29–14.74) 1.17 (0.69–1.98) 1.25 (0.79–1.98) 2.04 (1.13–3.68) 3.84 (1.75–8.41) 4 (referent) 1.0 1.0 1.0 1.0 https://doi.org/10.1371/journal.pone.0262053.t003 risk and lower-risk subgroups within the SCAI shock stage schema. ECHO-LVSWI permits the identification of clinically relevant myocardial dysfunction that might escape detection by the clinical exam or basic echocardiography. Performed early in the CICU stay, ECHO-LVSWI functioned as well as the SCAI shock stages schema for discrimination of in- hospital mortality, and incrementally improved mortality risk-stratification by the SCAI shock stages. The incremental risk-stratification provided by ECHO-LVSWI on top of the SCAI shock stages strengthens the argument that noninvasive hemodynamics should be routinely incorporated into shock severity assessment. The performance of a limited transthoracic echo- cardiogram or point-of-care ultrasound without the integration of Doppler derived hemody- namics may be inadequate for optimal risk stratification for CICU patients with or at risk of CS. This analysis must be contrasted with our recent studies to highlight its incremental value [10,17]. We previously examined several other echocardiographic variables in CICU patients across the SCAI shock stages, finding that a SVI <35 ml/m2 and a medial mitral E/e’ ratio >15 were independently associated with higher in-hospital mortality, while LVEF <40%, cardiac index <1.8 L/min/m2 and cardiac power output (CPO) were not [10]. Insofar as the ECHO-LVSWI is calculated using the SVI and mitral E/e’ ratio, the results of the present study demonstrating the predictive value of ECHO-LVSWI are not surprising. We have demon- strated that ECHO-LVSWI has higher discrimination for in-hospital mortality than LVEF, SVI and CI when analyzed as continuous variables, providing risk stratification by separating high-risk and low-risk CICU patients. While our prior study did demonstrate that ECHO-LVSWI decreased across the SCAI shock stages, the present study is the first to exam- ine the association between ECHO-LVSWI and outcomes as a function of shock severity [10]. In a larger analysis of unselected CICU patients, we previously reported the strong inverse association between ECHO-LVSWI and in-hospital mortality; however, this prior study did not account for SCAI shock stage [17]. By integrating the concepts demonstrated in these prior analyses, this study expands on the utility of echocardiographic ECHO-LVSWI for mortality risk-stratification in CICU patients across the spectrum of shock severity. Indeed, the ECHO-LVSWI has the highest Table 4. Optimal cut-off (maximum value of Youden’s J index = sensitivity + specificity– 1) with the associated sensitivity, specificity and overall accuracy for selected echocardiographic variables for prediction of in-hospital mortality using univariable logistic regression. Quartile Optimal cut-off Sensitivity Specificity Overall accuracy LVEF (%) CI (L/min/m2) SVI (ml/m2) LVSWI (J/m2) 45% 65.7% 60.6% 60.9% 2.50 L/min/m2 36 ml/m2 48.2% 71.4% 70.1% 66.5% 67.2% 67.2% 33.0 73.9% 64.2% 64.7% https://doi.org/10.1371/journal.pone.0262053.t004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 10 / 18 PLOS ONELVSWI and SCAI shock stage Fig 4. In-hospital mortality as a function of SCAI shock stage stratified by ECHO-LVSWI group (left) and for patients with ECHO-LVSWI < or �33.1 J/m2 (right). https://doi.org/10.1371/journal.pone.0262053.g004 discrimination value for in-hospital mortality of any of the echocardiographic variables we have examined, and was a more important predictor of in-hospital mortality than SCAI shock stage on multivariable regression and CART analysis [10,17]. In-hospital mortality was low among patients with preserved ECHO-LVSWI, even in the presence of severe (SCAI stage D/ E) shock. Likewise, patients with low ECHO-LVSWI (especially <20 J/m2) had high in-hospi- tal mortality, even in the absence of hemodynamic instability or shock during the first 24 hours after CICU admission. Therefore, the ECHO-LVSWI provided incremental refinement of prognosis at each SCAI shock stage, allowing identification of higher-risk and lower-risk subgroups that might require different approaches to management. The measurement of ECHO-LVSWI by TTE may be used to enhance prognostication and facilitate care in CICU patients by assessing the underlying degree of hemodynamic compromise beyond clinical assessment alone. The search for an optimal variable to define cardiac performance and hemodynamic com- promise in CS is ongoing, with several candidate variables identified. Several authors have pro- posed the CPO, derived from the MAP and cardiac output, as the preferred hemodynamic parameter for both prognosticating and guiding clinical care in CS patients [18–22]. This is logical as the MAP and cardiac output are relevant determinants of organ perfusion which can be manipulated with therapeutic interventions. However, when measured noninvasively using echocardiography, CPO is not as strongly associated with mortality as beat-by-beat parameters Fig 5. Forest plots showing unadjusted (left) and adjusted (right) odds ratio (OR) values for ECHO-LVSWI (per each 10 J/m2 higher) overall and in each SCAI shock stage. https://doi.org/10.1371/journal.pone.0262053.g005 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 11 / 18 PLOS ONELVSWI and SCAI shock stage Fig 6. Classification and regression tree (CART) analysis using ECHO-LVSWI and SCAI shock stage for stratification of in-hospital mortality risk. https://doi.org/10.1371/journal.pone.0262053.g006 such as SVI or ECHO-LVSWI; this could relate to the confounding effect of heart rate, which may compensate for low SV in some patients with circulatory failure and contribute to mea- surement error [10,17,31]. Recent multicenter analyses did not demonstrate substantial varia- tion in invasively-measured CI or CPO across SCAI shock stage, and neither CI nor CPO was associated with mortality [14,15]. This suggests that, while it remains a logical target for titrat- ing therapy in CS patients, CPO may not be the ideal hemodynamic parameter for predicting outcomes in this population. ECHO-LVSWI may prove to be a superior non-invasive parameter because it integrates diastolic function assessment, providing deeper insights into overall myocardial function that translates into improved mortality risk stratification. Another echocardiographic measure of cardiac performance is the myocardial contraction fraction (MCF), which indexes the SV to the myocardial volume to quantify how efficiently the myocardium is pumping [32]. MCF has theoretical advantages over the LVEF as a measure of LV systolic function, and like ECHO-LVSWI the MCF uses forward SV to quantify cardiac function. No prior studies have examined MCF in the context of shock severity, and expectedly we observed a decrease in the MCF as SCAI shock stage increased. ECHO-LVSWI was a stronger echocardiographic predic- tor of in-hospital mortality than MCF in this cohort, presumably resulting from its added prognostic value resulting from inclusion of diastolic function (represented by the mitral E/e’ ratio) [17]. Limitations of ECHO-LVSWI Controversy surrounds the ideal method to calculate the E/e’ ratio for estimation of left ven- tricular filling pressures, with different authors and guidelines using medial, lateral or mean e’ velocities in different settings [33–35]. Our institution has preferentially used the medial e’ PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 12 / 18 PLOS ONETable 5. Predictors of in-hospital mortality using multivariable logistic regression with stepwise backward vari- able selection to minimize the AIC value. LVSWI and SCAI shock stage Variable Demographics & comorbidities Age (per 10 years) Female sex Charlson Comorbidity Index (per point) History of diabetes mellitus Year of admission (per year) Admission diagnoses Respiratory failure Sepsis Severity of illness APACHE score (per 10 points) Day 1 SOFA score (per point) Admission Braden Skin Score (per point) Shock severity SCAI shock stage (per stage) Inodilators in first 24 hours Procedures and therapies Angiogram without PCI vs. no angiogram PCI vs. no angiogram PCI vs. angiogram without PCI pVAD or ECMO Dialysis CRRT Cardiac arrest VF CA vs. no CA Non-VF CA vs. no CA VF CA vs. non-VF CA IHCA LVSWI (per 10 J/m2) Adjusted OR 95% CI P value 1.300 0.713 1.127 0.652 0.904 2.525 2.106 1.223 0.890 0.900 1.243 1.864 0.726 0.424 0.583 6.021 2.005 3.272 3.036 6.878 2.266 3.372 0.664 1.116–1.514 0.492–1.034 1.056–1.202 0.436–0.975 0.843–0.970 1.661–3.836 1.294–3.429 1.096–1.364 0.813–0.974 0.844–0.960 1.024–1.508 1.054–3.296 0.473–1.112 0.267–0.673 0.348–0.980 1.621–22.365 0.826–4.866 1.001–10.698 1.795–5.135 4.112–11.505 1.246–4.121 1.740–6.534 0.564–0.782 0.0008 0.0741 0.0003 0.04 0.005 <0.0001 0.0027 0.0003 0.01 0.0013 0.03 0.03 0.14 0.0003 0.04 0.0073 0.12 0.05 <0.0001 <0.0001 0.007 0.02 <0.0001 Data are displayed as adjusted odds ratio (OR) and 95% confidence interval (CI) values. The final model C-statistic value was 0.927 for discrimination of in-hospital mortality. Candidate variables which were not selected for the model included APACHE-IV predicted hospital mortality; race; invasive and noninvasive ventilator use; history of myocardial infarction, heart failure, chronic kidney disease, dialysis and stroke; peak VIS and NEE in first 24 hours; use of vasopressors in first 24 hours; IABP and PAC use; blood transfusion; CardShock score; LVEF; and admission diagnosis of heart failure or acute coronary syndrome. https://doi.org/10.1371/journal.pone.0262053.t005 velocity based on data showing superiority of the E/e’ ratio using the medial e’ for estimation of left ventricular filling pressures, and this is reflected in the data availability within our cohort [36]. However, using either the medial or lateral e’ appears to have limitations under certain circumstances, and the use of the average or mean e’ has been advocated [33–35]. In our analy- sis, we found that ECHO-LVSWI was minimally affected by using the medial, lateral, or mean e’ velocity to estimate LVEDP, suggesting that this variable is minimally affected by the meth- odology used to calculate it and either lateral or mean e’ velocity could be substituted. Further- more, the presence of an elevated heart rate can produce fusion of the mitral E and A waves that poses a challenge to accurate assessment of the E/e’ ratio during stress, as is typical of PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 13 / 18 PLOS ONELVSWI and SCAI shock stage patients with shock; notably, the ECHO-LVSWI had lower discrimination for patients with an elevated heart rate or greater shock severity [33,34]. We calculated ECHO-LVSWI automatically using data extracted from formal TTE reports in the medical record, as opposed to manual review of the primary TTE images. Calculating the ECHO-LVSWI by hand at bedside is time-consuming and prone to errors, including the potential for intra- and intra-observer variability, measurement inaccuracy and arithmetic errors during hand calculation. Accordingly, the time demand necessary to manually calculate the ECHO-LVSWI and the extent to which the inter- and intra-observer variability of manu- ally calculated ECHO-LVSWI might influence its observed association with mortality remain uncertain. Therefore, it would be better to use an online calculator or ideally the ECHO-LVSWI could be calculated by the echocardiography imaging package or reporting software automatically. To facilitate use of ECHO-LVSWI, we have created an online calcula- tor that can be used at the bedside (S1 File). Until automatic calculation of ECHO-LVSWI is incorporated into Doppler hemodynamic packages of bedside ultrasound machines or echo- cardiographic reporting software, we suggest that ECHO-LVSWI be calculated primarily using data obtained from a formal TTE to mitigate against these issues. Further research is necessary to determine when the potentially laborious calculation of ECHO-LVSWI is neces- sary, as opposed to the less-predictive but simpler SVI itself. Limitations of this study This retrospective observational analysis cannot be used to determine cause-and-effect rela- tionships, and the presence of residual confounding cannot be excluded as a mediator of the association between ECHO-LVSWI and outcomes. Only a relative minority of the entire CICU population had a TTE within one day of CICU admission containing complete data to calculate the ECHO-LVSWI, leading to potential selection bias and a lower-risk cohort than in our prior studies [3,6–11,17,24–28]. While this is typical of retrospective echocardiographic studies in critical care settings, we could not determine whether patient characteristics could have influenced which echocardiographic images were obtained and the quality of the data leading to further bias [10,17]. Without invasive hemodynamic data, we cannot be assured that the TTE accurately estimated the ECHO-LVSWI; we could not exclude the presence of poor Doppler signal alignment or other issues that could have affected the accuracy of our TTE measurements. Finally, we could not determine was vasoactive drugs patients were receiving at the time of TTE, which could have influenced the observed ECHO-LVSWI and its association with outcomes. Conclusions ECHO-LVSWI was strongly and inversely associated with the risk of in-hospital mortality in CICU patients across the spectrum of cardiogenic shock severity, providing justification for its routine measurement in this population. At each SCAI shock stage, patients with a lower ECHO-LVSWI had higher in-hospital mortality, allowing ECHO-LVSWI to provide incre- mental risk stratification beyond the clinical assessment of shock severity. A lower ECHO-LVSWI identified high-risk patients in the group without overt shock, while a higher ECHO-LVSWI identified patients with hemodynamic instability or hypoperfusion who were at lower risk of adverse outcomes. Indeed, the presence of a low ECHO-LVSWI could reclas- sify patients at each SCAI shock into higher-risk subgroups with similar mortality to patients classified into a more severe SCAI shock stage. Our study emphasizes that hemodynamics (as assessed using Doppler TTE) can improve mortality risk stratification beyond the clinical defi- nition of the SCAI shock stages. Future prospective studies are needed to better understand PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 14 / 18 PLOS ONELVSWI and SCAI shock stage the association between ECHO-LVSWI and outcomes in CS patients, and to determine whether interventions designed to improve the ECHO-LVSWI will translate into lower mor- tality in CICU patients. Supporting information S1 Fig. Receiver-operator characteristic (ROC) curves demonstrating discrimination of in- hospital mortality by ECHO-LVSWI calculated using the medial (red), lateral (green) or mean (blue) e’ velocity to estimate LVDEP for patients (n = 2896) with available data for both medial and lateral e’ velocity. P values for comparison of AUC values were all >0.05 by De Long test. (TIF) S2 Fig. Receiver-operator characteristic (ROC) curves demonstrating discrimination of in- hospital mortality by ECHO-LVSWI (red), cardiac index (CI, green), LVEF (blue) and stroke volume index (SVI, orange). ECHO-LVSWI had a higher AUC value by the De Long test when compared with CI (p <0.0001), LVEF (p = 0.0002), or SVI (p = 0.06). (TIF) S3 Fig. Observed in-hospital mortality in patients grouped by quartiles of ECHO-LVSWI (red), CI (green), LVEF (blue) and SVI (orange). Median and interquartile range values defin- ing the quartiles are as follows: ECHO LVSWI, 37.0 (21.0, 46.1) J/m2; CI, 2.8 (2.4, 3.3) L/min/ m2; LVEF, 51 (36, 61) %; SVI, 41 (33, 47) ml/m2. (TIF) S1 Table. Measured and derived echocardiographic variables of interest. (DOCX) S2 Table. Formulas used to calculate echocardiographic hemodynamic parameters, using data from the time of the echocardiogram. (DOCX) S3 Table. Study definitions of hypotension, tachycardia, hypoperfusion, deterioration and refractory shock, as defined by Jentzer, et al. J Am Coll Cardiol 2019. (DOCX) S4 Table. Study definitions of Society for Cardiovascular Angiography and Intervention shock stages, as defined by Jentzer, et al. J Am Coll Cardiol 2019. (DOCX) S1 File. Automatic ECHO-LVSWI calculator based on 2-D and Doppler echocardiographic measurements. (XLSX) Author Contributions Conceptualization: Jacob C. Jentzer, Brandon M. Wiley, Nandan S. Anavekar. Data curation: Jacob C. Jentzer. Formal analysis: Jacob C. Jentzer. Investigation: Jacob C. Jentzer, Brandon M. Wiley, Nandan S. Anavekar. Methodology: Jacob C. Jentzer, Brandon M. Wiley, Nandan S. Anavekar. Resources: Jacob C. Jentzer. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 15 / 18 PLOS ONELVSWI and SCAI shock stage Validation: Jacob C. Jentzer. Visualization: Jacob C. Jentzer. Writing – original draft: Jacob C. Jentzer. Writing – review & editing: Jacob C. Jentzer, Brandon M. Wiley, Nandan S. Anavekar. References 1. Jentzer JC, Ahmed AM, Vallabhajosyula S, Burstein B, Tabi M, Barsness GW, et al. Shock in the car- diac intensive care unit: Changes in epidemiology and prognosis over time. Am Heart J. 2021; 232:94– 104. https://doi.org/10.1016/j.ahj.2020.10.054 PMID: 33257304 2. 3. Jentzer JC, van Diepen S, Barsness GW, Katz JN, Wiley BM, Bennett CE, et al. Changes in comorbidi- ties, diagnoses, therapies and outcomes in a contemporary cardiac intensive care unit population. Am Heart J. 2019; 215:12–19. https://doi.org/10.1016/j.ahj.2019.05.012 PMID: 31260901 Jentzer JC, van Diepen S, Murphree DH, Ismail AS, Keegan MT, Morrow DA, et al. Admission diagnosis and mortality risk prediction in a contemporary cardiac intensive care unit population. Am Heart J. 2020; 224:57–64. https://doi.org/10.1016/j.ahj.2020.02.018 PMID: 32305724 4. Baran DA, Grines CL, Bailey S, Burkhoff D, Hall SA, Henry TD, et al. SCAI clinical expert consensus statement on the classification of cardiogenic shock: This document was endorsed by the American College of Cardiology (ACC), the American Heart Association (AHA), the Society of Critical Care Medi- cine (SCCM), and the Society of Thoracic Surgeons (STS) in April 2019. Catheter Cardiovasc Interv. 2019; 94:29–37. https://doi.org/10.1002/ccd.28329 PMID: 31104355 5. 6. 7. 8. 9. Jentzer JC. Understanding Cardiogenic Shock Severity and Mortality Risk Assessment. Circ Heart Fail. 2020; 13:e007568. https://doi.org/10.1161/CIRCHEARTFAILURE.120.007568 PMID: 32900232 Jentzer JC, van Diepen S, Barsness GW, Henry TD, Menon V, Rihal CS, et al. Cardiogenic Shock Clas- sification to Predict Mortality in the Cardiac Intensive Care Unit. J Am Coll Cardiol. 2019; 74:2117– 2128. https://doi.org/10.1016/j.jacc.2019.07.077 PMID: 31548097 Jentzer JC, Baran DA, van Diepen S, Barsness GW, Henry TD, Naidu SS, et al. Admission Society for Cardiovascular Angiography and Intervention shock stage stratifies post-discharge mortality risk in car- diac intensive care unit patients. Am Heart J. 2020; 219:37–46. https://doi.org/10.1016/j.ahj.2019.10. 012 PMID: 31710843 Jentzer JC, Henry TD, Barsness GW, Menon V, Baran DA and Van Diepen S. Influence of cardiac arrest and SCAI shock stage on cardiac intensive care unit mortality. Catheter Cardiovasc Interv. 2020; 96:1350–1359. https://doi.org/10.1002/ccd.28854 PMID: 32180344 Jentzer JC, Lawler PR, van Diepen S, Henry TD, Menon V, Baran DA, et al. Systemic Inflammatory Response Syndrome Is Associated With Increased Mortality Across the Spectrum of Shock Severity in Cardiac Intensive Care Patients. Circ Cardiovasc Qual Outcomes. 2020; 13:e006956. https://doi.org/ 10.1161/CIRCOUTCOMES.120.006956 PMID: 33280435 10. Jentzer JC, Wiley BM, Anavekar NS, Pislaru SV, Mankad SV, Bennett CE, et al. Noninvasive Hemody- namic Assessment of Shock Severity and Mortality Risk Prediction in the Cardiac Intensive Care Unit. JACC Cardiovasc Imaging. 2020. https://doi.org/10.1016/j.jcmg.2020.05.038 PMID: 32828777 11. Padkins M, Breen T, Anavekar N, van Diepen S, Henry TD, Baran DA, et al. Age and shock severity pre- dict mortality in cardiac intensive care unit patients with and without heart failure. ESC Heart Fail. 2020. https://doi.org/10.1002/ehf2.12995 PMID: 32909377 12. Schrage B, Dabboura S, Yan I, Hilal R, Neumann JT, Sorensen NA, Gossling A, et al. Application of the SCAI classification in a cohort of patients with cardiogenic shock. Catheter Cardiovasc Interv. 2020. 13. Baran DA, Long A, Badiye AP and Stelling K. Prospective validation of the SCAI shock classification: Single center analysis. Catheter Cardiovasc Interv. 2020. https://doi.org/10.1002/ccd.29319 PMID: 33026155 14. Thayer KL, Zweck E, Ayouty M, Garan AR, Hernandez-Montfort J, Mahr C, et al. Invasive Hemody- namic Assessment and Classification of In-Hospital Mortality Risk Among Patients With Cardiogenic Shock. Circ Heart Fail. 2020; 13:e007099. https://doi.org/10.1161/CIRCHEARTFAILURE.120.007099 PMID: 32900234 15. Garan AR, Kanwar M, Thayer KL, Whitehead E, Zweck E, Hernandez-Montfort J, et al. Complete Hemodynamic Profiling With Pulmonary Artery Catheters in Cardiogenic Shock Is Associated With Lower In-Hospital Mortality. JACC Heart Fail. 2020; 8:903–913. https://doi.org/10.1016/j.jchf.2020.08. 012 PMID: 33121702 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 16 / 18 PLOS ONELVSWI and SCAI shock stage 16. 17. van Diepen S, Katz JN, Albert NM, Henry TD, Jacobs AK, Kapur NK, et al. Contemporary Management of Cardiogenic Shock: A Scientific Statement From the American Heart Association. Circulation. 2017; 136:e232–e268. https://doi.org/10.1161/CIR.0000000000000525 PMID: 28923988 Jentzer JC, Anavekar NS, Burstein BJ, Borlaug BA and Oh JK. Noninvasive Echocardiographic Left Ventricular Stroke Work Index Predicts Mortality in Cardiac Intensive Care Unit Patients. Circ Cardio- vasc Imaging. 2020; 13:e011642. https://doi.org/10.1161/CIRCIMAGING.120.011642 PMID: 33190537 18. Basir MB, Kapur NK, Patel K, Salam MA, Schreiber T, Kaki A, et al. Improved Outcomes Associated with the use of Shock Protocols: Updates from the National Cardiogenic Shock Initiative. Catheter Car- diovasc Interv. 2019; 93:1173–1183. https://doi.org/10.1002/ccd.28307 PMID: 31025538 19. Mendoza DD, Cooper HA and Panza JA. Cardiac power output predicts mortality across a broad spec- trum of patients with acute cardiac disease. Am Heart J. 2007; 153:366–70. https://doi.org/10.1016/j. ahj.2006.11.014 PMID: 17307413 20. Fincke R, Hochman JS, Lowe AM, Menon V, Slater JN, Webb JG, et al. Cardiac power is the strongest hemodynamic correlate of mortality in cardiogenic shock: a report from the SHOCK trial registry. J Am Coll Cardiol. 2004; 44:340–8. https://doi.org/10.1016/j.jacc.2004.03.060 PMID: 15261929 21. Sionis A, Rivas-Lasarte M, Mebazaa A, Tarvasmaki T, Sans-Rosello J, Tolppanen H, et al. Current Use and Impact on 30-Day Mortality of Pulmonary Artery Catheter in Cardiogenic Shock Patients: Results From the CardShock Study. J Intensive Care Med. 2020; 35:1426–1433. https://doi.org/10.1177/ 0885066619828959 PMID: 30732522 22. Tehrani BN, Truesdell AG, Sherwood MW, Desai S, Tran HA, Epps KC, et al. Standardized Team- Based Care for Cardiogenic Shock. J Am Coll Cardiol. 2019; 73:1659–1669. https://doi.org/10.1016/j. jacc.2018.12.084 PMID: 30947919 23. Choi JO, Lee SC, Choi SH, Kim SM, Choi JH, Park JR, et al. Noninvasive assessment of left ventricular stroke work index in patients with severe mitral regurgitation: correlation with invasive measurement and exercise capacity. Echocardiography. 2010; 27:1161–9. https://doi.org/10.1111/j.1540-8175.2010. 01222.x PMID: 20584058 24. 25. Jentzer JC, Bennett C, Wiley BM, Murphree DH, Keegan MT, Gajic O, et al. Predictive Value of the Sequential Organ Failure Assessment Score for Mortality in a Contemporary Cardiac Intensive Care Unit Population. J Am Heart Assoc. 2018;7. Jentzer JC, Murphree DH, Wiley B, Bennett C, Goldfarb M, Keegan MT, et al. Comparison of Mortality Risk Prediction Among Patients >/ = 70 Versus <70 Years of Age in a Cardiac Intensive Care Unit. Am J Cardiol. 2018; 122:1773–1778. https://doi.org/10.1016/j.amjcard.2018.08.011 PMID: 30227963 26. Bennett CE, Wright RS, Jentzer J, Gajic O, Murphree DH, Murphy JG, et al. Severity of illness assess- ment with application of the APACHE IV predicted mortality and outcome trends analysis in an aca- demic cardiac intensive care unit. J Crit Care. 2019; 50:242–246. https://doi.org/10.1016/j.jcrc.2018.12. 012 PMID: 30612068 27. 28. 29. Jentzer JC, Wiley B, Bennett C, Murphree DH, Keegan MT, Gajic O, et al. Early noncardiovascular organ failure and mortality in the cardiac intensive care unit. Clin Cardiol. 2020. https://doi.org/10.1002/ clc.23339 PMID: 31999370 Jentzer JC, Wiley B, Bennett C, Murphree DH, Keegan MT, Kashani KB, et al. Temporal Trends and Clinical Outcomes Associated with Vasopressor and Inotrope Use in The Cardiac Intensive Care Unit. Shock. 2020; 53:452–459. https://doi.org/10.1097/SHK.0000000000001390 PMID: 31169766 Lang RM, Badano LP, Mor-Avi V, Afilalo J, Armstrong A, Ernande L, et al. Recommendations for car- diac chamber quantification by echocardiography in adults: an update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging. J Am Soc Echocardiogr. 2015; 28:1–39 e14. https://doi.org/10.1016/j.echo.2014.10.003 PMID: 25559473 30. Rudski LG, Lai WW, Afilalo J, Hua L, Handschumacher MD, Chandrasekaran K, et al. Guidelines for the echocardiographic assessment of the right heart in adults: a report from the American Society of Echocardiography endorsed by the European Association of Echocardiography, a registered branch of the European Society of Cardiology, and the Canadian Society of Echocardiography. J Am Soc Echo- cardiogr. 2010; 23:685–713; quiz 786–8. https://doi.org/10.1016/j.echo.2010.05.010 PMID: 20620859 31. Mele D, Pestelli G, Molin DD, Trevisan F, Smarrazzo V, Luisi GA, et al. Echocardiographic Evaluation of Left Ventricular Output in Patients with Heart Failure: A Per-Beat or Per-Minute Approach? J Am Soc Echocardiogr. 2020; 33:135–147 e3. https://doi.org/10.1016/j.echo.2019.09.009 PMID: 31866323 32. Matthews SD, Rubin J, Cohen LP and Maurer MS. Myocardial Contraction Fraction: A Volumetric Mea- sure of Myocardial Shortening Analogous to Strain. J Am Coll Cardiol. 2018; 71:255–256. https://doi. org/10.1016/j.jacc.2017.09.1157 PMID: 29325647 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 17 / 18 PLOS ONELVSWI and SCAI shock stage 33. Mitter SS, Shah SJ and Thomas JD. A Test in Context: E/A and E/e’ to Assess Diastolic Dysfunction and LV Filling Pressure. J Am Coll Cardiol. 2017; 69:1451–1464. https://doi.org/10.1016/j.jacc.2016.12. 037 PMID: 28302294 34. Sharifov OF and Gupta H. What Is the Evidence That the Tissue Doppler Index E/e’ Reflects Left Ven- tricular Filling Pressure Changes After Exercise or Pharmacological Intervention for Evaluating Diastolic Function? A Systematic Review. J Am Heart Assoc. 2017;6. https://doi.org/10.1161/JAHA.116.004766 PMID: 28298372 35. Nauta JF, Hummel YM, van der Meer P, Lam CSP, Voors AA and van Melle JP. Correlation with inva- sive left ventricular filling pressures and prognostic relevance of the echocardiographic diastolic param- eters used in the 2016 ESC heart failure guidelines and in the 2016 ASE/EACVI recommendations: a systematic review in patients with heart failure with preserved ejection fraction. Eur J Heart Fail. 2018; 20:1303–1311. https://doi.org/10.1002/ejhf.1220 PMID: 29877602 36. Ommen SR, Nishimura RA, Appleton CP, Miller FA, Oh JK, Redfield MM et al. Clinical utility of Doppler echocardiography and tissue Doppler imaging in the estimation of left ventricular filling pressures: A comparative simultaneous Doppler-catheterization study. Circulation. 2000; 102:1788–94. https://doi. org/10.1161/01.cir.102.15.1788 PMID: 11023933 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262053 March 9, 2022 18 / 18 PLOS ONE
10.1371_journal.pone.0261952	RESEARCH ARTICLE Preterm birth and the risk of chronic disease multimorbidity in adolescence and early adulthood: A population-based cohort study Katriina Heikkila¨ ID Mika Gissler3,4,5, Sven SandinID 6,7,8, Eero Kajantie1,9,10 1, Anna Pulakka1, Johanna Metsa¨ la¨ 1, Suvi Alenius1,2, Petteri Hovi1, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Heikkila¨ K, Pulakka A, Metsa¨la¨ J, Alenius S, Hovi P, Gissler M, et al. (2021) Preterm birth and the risk of chronic disease multimorbidity in adolescence and early adulthood: A population- based cohort study. PLoS ONE 16(12): e0261952. https://doi.org/10.1371/journal.pone.0261952 Editor: Ramune Jacobsen, University of Copenhagen: Kobenhavns Universitet, DENMARK Received: October 7, 2021 Accepted: December 14, 2021 Published: December 31, 2021 Copyright: © 2021 Heikkila¨ et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Data cannot be shared publicly due to restrictions by law. Data can be requested through the Statistics Finland (info@stat.fi) and Finnish Social and Health Data Permit Authority Findata (info@findata.fi) after approval from relevant ethical committees. Funding: EK, SS and MG: European Union Horizon 2020 (grant no. 733280: RECAP: “Research on Children and Adults Born Preterm”), https://ec. europa.eu/programmes/horizon2020/en/home; Academy of Finland (grant no. 315690), https:// 1 Population Health Unit, Finnish Institute for Health and Welfare, Helsinki, Finland, 2 Children’s Hospital, University of Helsinki and Helsinki University Hospital, Helsinki, Finland, 3 Information Services Department, Finnish Institute for Health and Welfare, Helsinki, Finland, 4 Academic Primary Health Care Centre, Region Stockholm, Stockholm, Sweden, 5 Department of Molecular Medicine and Surgery, Karolinska Institutet, Stockholm, Sweden, 6 Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden, 7 Department of Psychiatry, Icahn School of Medicine at Mount Sinai, New York, NY, United States of America, 8 Seaver Autism Center for Research and Treatment at Mount Sinai, New York, NY, United States of America, 9 PEDEGO Research Unit, MRC Oulu, Oulu University Hospital and University of Oulu, Oulu, Finland, 10 Department of Clinical and Molecular Medicine, Norwegian University of Science and Technology, Trondheim, Norway katriina.heikkila@thl.fi Abstract Background People who were born prematurely have high risks of many individual diseases and condi- tions in the early part of the life course. However, our knowledge of the burden of multiple diseases (multimorbidity) among prematurely born individuals is limited. We aimed to inves- tigate the risk and patterns of chronic disease multimorbidity in adolescence and early adult- hood among individuals born across the spectrum of gestational ages, comparing preterm and full-term born individuals. Methods and findings We used individual-level data from linked nationwide registers to examine the associations of gestational age at birth with specialised healthcare records of �2 chronic diseases (multi- morbidity) in adolescence (age 10–17 years) and early adulthood (age 18–30 years). Our study population comprised 951,116 individuals (50.2% females) born alive in Finland between 1st January 1987 and 31st December 2006, inclusive. All individuals were followed from age 10 years to the onset of multimorbidity, emigration, death, or 31 December 2016 (up to age 30 years). We estimated hazard ratios (HRs) and 95% confidence intervals (CIs) for multimorbidity using flexible parametric survival models. During 6,417,903 person-years at risk (median follow-up: 7.9 years), 11,919 individuals (1.3%) had multimorbidity in adoles- cence (18.6 per 10,000 person-years). During 3,967,419 person-years at risk (median fol- low-up: 6.2 years), 15,664 individuals (1.7%) had multimorbidity in early adulthood (39.5 per 10,000 person-years). Adjusted HRs for adolescent multimorbidity, comparing preterm to full-term born individuals, were 1.29 (95% CI: 1.22 to 1.36) and 1.26 (95% CI: 1.18 to 1.35) PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 1 / 15 PLOS ONEPreterm birth and risk of multimorbidity www.aka.fi/en. EK: Foundation for Pediatric Research, https://www. lastentautientutkimussaatio.fi; and Sigrid Juse´lius Foundation https://www.sigridjuselius.fi/en/. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: Eero Kajantie reports research support from Novo Nordic Foundation for work other than the research described in this manuscript. There are no employment or consultancy relationships, patents, products in development or marketed products to declare in relation to this or any other funder. Other authors declare that no competing interests exist. This does not alter our adherence to PLOS ONE policies on sharing data and materials. in females and males, respectively. The associations of preterm birth with early adult multi- morbidity were less marked, with the adjusted HRs indicating 1.18-fold risk in females (95% CI: 1.12 to 1.24) and 1.10-fold risk in males (95% CI: 1.04 to 1.17). We observed a consis- tent dose-response relationship between earlier gestational age at birth and increasing risks of both multimorbidity outcomes. Compared to full-term born males, those born at 37–38 weeks (early term) had a 1.06-fold risk of multimorbidity in adolescence (95% CI: 0.98 to 1.14) and this risk increased in a graded manner up to 6.85-fold (95% CI: 5.39 to 8.71) in those born at 23–27 weeks (extremely premature), independently of covariates. Among females, the same risks ranged from 1.16-fold (95% CI: 1.09 to 1.23) among those born at 37–38 weeks to 5.65-fold (95% CI: 4.45 to 7.18) among those born at 23–27 weeks. The corresponding risks of early adult multimorbidity were similar in direction but less marked in magnitude, with little difference in risks between males and females born at 36–37 weeks but up to 3-fold risks observed among those born at 23–27 weeks. Conclusions Our findings indicate that an earlier gestational age at birth is associated with increased risks of chronic disease multimorbidity in the early part of the life course. There are currently no clinical guidelines for follow-up of prematurely born individuals beyond childhood, but these observations suggest that information on gestational age would be a useful character- istic to include in a medical history when assessing the risk of multiple chronic diseases in adolescent and young adult patients. Introduction Multimorbidity, the co-occurrence of two or more chronic, non-communicable diseases in one individual, is a growing health concern worldwide [1–5]. Multimorbidity is associated with many adverse health outcomes (including. increased mortality and numbers of hospital admissions, and decreased quality of life [6–8] and it incurs considerable human and health- care costs in all age groups. [9] Many risk factors for multimorbidity (e.g. age, tobacco smoking and socioeconomic deprivation) are now well-recognised, but other potential risk factors and high-risk groups are less well understood. [2, 3, 10–12] One potential risk factor for multimor- bidity is preterm birth (being born before 37 completed weeks of gestation) [13, 14]. Preterm birth constitutes approximately 6% of all births in Finland and the Nordic coun- tries, and 11% of births worldwide [13, 15, 16]. Children and adults who were born prema- turely have increased risks of many diseases and health conditions, including low cognitive abilities [17, 18] and cardiometabolic [19–25], respiratory [26, 27] and neuropsychiatric dis- eases [28–30]. Compared to term-born young adults, those who were born preterm have been estimated to have (depending on the degree of prematurity) 2.5- to 4.0-fold increased risk of metabolic syndrome [23, 24], 1.19- to 1.53-fold increased risk of ischaemic heart disease [19], 1.7- to 3.6-fold increased risk of asthma [27], 1.4- to 3.0-fold increased risk of depression [31], 1.35- to 2.31.fold increased risk of autism-spectrum disorders [29] and 1.23- to 1.44-fold increased risk of mortality [32]. However, the association of preterm birth with chronic disease multimorbidity (the co-occurrence of multiple chronic diseases in one individual) is unclear. PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 2 / 15 PLOS ONEPreterm birth and risk of multimorbidity Aims The primary aim of our study was to investigate the risks of chronic disease multimorbidity in adolescence (age 10–17 years) and young adults (age 18–30 years) across the spectrum of ges- tational ages, comparing preterm-born individuals and those born at full-term. The secondary aim was to examine the overlap of chronic disease multimorbidity in adolescence and early adulthood with intellectual disabilities and developmental disorders, comparing preterm and full-term born individuals. Methods Data sources Data on gestational age at birth, sex, date of birth, birth weight, multiple pregnancy and moth- er’s characteristics (parity, smoking during pregnancy, socioeconomic position) were ascer- tained from the Medical Birth Register. Congenital anomalies were ascertained from Register of Congenital Malformations and mother’s diabetes or hypertensive disorder from a combina- tion of data from the Medical Birth Register and Care Register for Health Care. Chronic dis- ease multimorbidity in adolescence and early adulthood, as well as intellectual disabilities and disorders of psychological development, behaviour or personality, were ascertained from the Care Register for Health Care. Information on dates of death and emigration from Finland was obtained from the nationwide Population Information System. All these data sources cover the whole population of Finland for the study period (between 1st January 1987 and 31st December 2016). Ethical approval In Finland, research based solely on analysing existing register data does not require ethical approval or consent from the individuals whose pseudonymised data will be used. The research reported here was approved by the institutional ethics review board of the Finnish Institute for Health and Welfare (THL/1984/6.02.01/2018) and the relevant register authori- ties. All register data were pseudonymised and linked prior to analysis by Statistics Finland and Information Services, Finnish Institute for Health and Welfare, and the researchers only had access to pseudonymised data. Patient and public involvement Our investigation was based on analysis of existing register data and no patients or members of the public were involved in the design, conduct or interpretation of the study. Due to data security and privacy constraints, it was not possible to involve members of the public in the investigation. Study population The study population comprised all individuals who were born alive in Finland between 1st January 1987 and 31st December 2006 (inclusive) and who remained alive and living in Fin- land at the age 10 years (n = 953,658). Individuals were followed up from age 10 years to the onset of multimorbidity (in adolescence, age 10–17 years, or in early adulthood, age 18–30 years), death, emigration or 31st December 2016. We excluded individuals with a gestational age earlier than 23 weeks (n = 411), those with implausible data on birth weight alone or birth weight in combination with gestational age (birth weight z-score smaller than -6, birth weight less than 300 grams, or gestational age less than 37 weeks and birth weight z-score larger than PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 3 / 15 PLOS ONEPreterm birth and risk of multimorbidity 4; n = 252). Further excluded were individuals of undetermined sex (n = 26) and individuals with incomplete data on birth weight and/or gestational age (n = 1,853, 0.2%). Exposure Gestational age in completed weeks was analysed as a categorical variable as follows: 23–27 weeks (extremely preterm), 28–31 weeks (very preterm), 32–33 weeks (moderately preterm), 34–36 weeks (late preterm), 37–38 weeks (early term), 39–41 weeks (full-term, reference cate- gory) and > = 42 weeks [13, 14, 33, 34]. Gestational age was determined by the best clinical estimate, based on ultrasound examination when available, or information on the last men- strual period [35]. Ultrasonography become standard practice in Finland during the late 1980s and early 1990s. Chronic disease multimorbidity Chronic disease multimorbidity was defined as one individual having records of two or more non-communicable chronic diseases (psychiatric or somatic) during the same or subsequent secondary care episode(s). The diseases constituting multimorbidity were selected based on evidence on commonly occurring diseases among preterm and term born adolescents and young adults [21, 24, 25, 36–44]. Multimorbidity was identified using data on both in-patient and out-patient hospital care (referred to in the Nordic context as specialised healthcare), based on diagnostic codes listed in S1 Table in S1 File. First, patient records with two or more of these diagnoses were identified and tagged as indicating chronic disease multimorbidity, with the admission or visit date as the date of onset of multimorbidity. Second, the admissions or visits including only one diagnostic code from S1 Table in S1 File were sorted according to admission date and the date of multimorbidity onset defined as the first time when an individ- ual has accumulated records of two or more chronic diseases. We examined two outcomes: chronic disease multimorbidity in adolescence (age 10–17 years) and in early adulthood (age 18–30 years). These were not mutually exclusive and an individual could have multimorbidity in adolescence as well as in early adulthood. Intellectual disabilities and disorders of psychological development, behaviour or personality Intellectual disabilities and disorders of psychological development, behaviour or personality (which are classified as disorders, rather than diseases, and include, for example, different lev- els of intellectual disability and autism-spectrum traits) were identified using diagnostic codes provided in S2 Table in S1 File. These were ascertained from each individual’s specialised care records at any time between age 1 year and the end of follow-up. The overlap of chronic dis- ease multimorbidity with these disorders was defined as having a specialised health care record of chronic disease multimorbidity, as well one or more of the diagnostic codes listed in S2 Table in S1 File. Covariates We adjusted the analyses for factors that previous research has shown to be associated with preterm birth [45–50] and which we hypothesised could also be associated with chronic dis- ease multimorbidity in adolescence and early adulthood. These potentially confounding fac- tors were birthweight z-score (calculated using data on birth weight and gestational age, based on Intergrowth Standards [51] as reference values, and modelled as a continuous variable), multiple pregnancy (yes vs. no), any (major or minor) congenital anomalies (yes vs. no), PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 4 / 15 PLOS ONEPreterm birth and risk of multimorbidity mother’s age (years, continuous), mother’s parity (0, 1, 2, 3, 4, 5+), mother’s smoking during pregnancy (yes vs. no), mother’s socioeconomic position (based on her occupation during pregnancy and categorised as follows: low: manual workers, pensioners, students and long- term unemployed; intermediate: non-manual workers; high: executive employees; self- employed and unknown), mother’s diabetes during pregnancy (yes vs. no), mother’s hyperten- sive disorder during pregnancy (yes vs. no), and year of birth (categories). Mother’s occupa- tion during pregnancy was unknown for 309,882 mothers (32.9%), who were likely to be stay- at-home mothers or students [52]; rather than exclude the children of these mothers from the analyses, we included them in a separate category for mother’s socioeconomic position. Statistical analyses Time at risk of adolescent multimorbidity was defined as time from age 10 to 17 years and time at risk of early adult multimorbidity as time from age 18 to 30 years, up to the first of the following: time of onset of multimorbidity, date of death, date of emigration or the end of fol- low-up (31st December 2016). Adolescent multimorbidity was modelled conditional on sur- vival to age 10 and early adult multimorbidity conditional on survival to age 18. The analyses were conducted separately for females and males, and for those with intellectual disabilities or disorders of psychological development, behaviour or personality. We estimated hazard ratios (HRs) and 95% confidence intervals (CIs) for both multimorbidity outcomes, across categories of gestational age, using flexible parametric survival models on the cumulative hazard scale, utilising Stata command stpm2 [53, 54]. With age as the timescale, these models fitted restricted cubic splines with 1 to 6 internal knots (depending on sex and covariates) to model the baseline hazard for each gestational age category. We ran five sets of models for all expo- sure-outcome pairs: unadjusted models, models adjusted for prenatal exposures (birthweight z-score, multiple pregnancy and congenital anomalies), models adjusted for mother’s charac- teristics (age, parity, smoking during pregnancy and socioeconomic position), models adjusted for pregnancy disorders (diabetes and hypertension) and multivariable-adjusted models (adjusted for all the above covariates). In addition, we undertook sensitivity analyses adjusted for the year of birth, as a marker for perinatal care. All analyses were conducted using Stata SE 16 (Stata Corporation, College Station, Texas, United States). Results Study population Our analyses were based on data from 951,116 individuals were born in Finland during the study period, remained alive and living in Finland at age 10 years and had data available on gestational age, birth weight and covariates (Table 1). During 6,417,903 person-years at risk of multimorbidity in adolescence (median follow-up: 7.9 years), 11,919 individuals (1.3%) had a specialised healthcare record of this outcome (18.6 per 10,000 person-years). During 3,967,419 person-years at risk of early adult multimorbidity (median follow-up: 6.2 years), 15,664 indi- viduals (1.7%) had a specialised healthcare record of this outcome (39.5 per 10,000 person- years). In all, 4,986 individuals (0.5%) died during the follow-up (between the ages of 10 and 30 years). Gestational age and chronic disease multimorbidity Individuals born preterm were more likely to have multimorbidity in adolescence than those born full-term: multivariable-adjusted HRs for adolescent multimorbidity in females and males, respectively, were 1.29 (95% CI: 1.22 to 1.36) and 1.26 (95% CI: 1.18 to 1.35). The PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 5 / 15 PLOS ONEAdolescence (age 10–17 years) Early adulthood (age 18–30 years) Covariates N (%) female Median (IQR) birth weight z-score1 N (%) multiple pregnancy N (%) with any congenital anomalies Table 1. Characteristics of study population. N (%) chronic disease multimorbidity Gestational age at birth (completed weeks) Preterm birth and risk of multimorbidity 23–27 140 (7.0) 28–31 207 (5.2) 32–33 160 (3.1) 34–36 610 (1.8) 37–38 39–41 42+ p 2 206 (1.3) 8 076 (1.2) 520 (1.3) <0.0001 86 (4.3) 162 (4.1) 141 (2.8) 694 (2.1) 2 856 (1.7) 11 000 (1.6) 725 (1.8) <0.0001 977 (48.9) 1 739 (43.8) 2267 (44.3) 15 718 (46.5) 81 957 (48.4) 353 811 (50.9) 20 768 (50.1) -0.03 (-1.02 to 0.59) 0.14 (-0.73 to 0.76) 0.14 (-0.54 to 0.72) 0.20 (-0.50 to 0.86) 0.55 (-0.15 to 1.24) 0.71 (-0.04 to 1.38) 0.69 (-0.01 to 1.36) 378 (18.9) 255 (12.8) 902 (22.7) 593 (14.9) 1 320 (25.8) 6 522 (19.3) 11 551 (6.8) 2 848 (0.4) 5 (0.01) 602 (11.8) 3 052 (9.0) 11 824 (7.0) 40 884 (5.9) 2 645 (6.4) Median (IQR) mother’s age 30 (26 to 34) 30 (25 to 34) 29 (25 to 33) 29 (25 to 33) 29 (26 to 33) 29 (25 to 32) 28 (25 to 32) N (%) multiparous mother 1 055 (52.8) 1 982 (49.9) 2 505 (48.9) 17 596 (52.0) 104 309 (61.5) 421 545 (60.6) 19 870 (47.9) N (%) mother smoked during pregnancy N %) mother had low SEP2 N (%) hypertension N (%) diabetes 579 (29.0) 1 053 (26.5) 1 220 (23.8) 6 941 (20.5) 31 057 (18.3) 118 483 (17.0) 7 699 (18.6) 409 (20.5) 251 (12.6) 44 (2.2) 838 (21.1) 817 (20.6) 195 (4.9) 1 100 (21.5) 7 029 (20.8) 34 458 (20.3) 143 489 (20.6) 9 461 (22.8) 1 020 (19.9) 4 944 (14.6) 16 481 (9.7) 35 838 (5.2) 329 (6.4) 2 703 (8.0) 13 035 (7.7) 33 021 (4.8) 1 495 (3.6) 1 438 (3.5) 1 Reference: Intergrowth Standards. 2 SEP: socioeconomic position. https://doi.org/10.1371/journal.pone.0261952.t001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 <0.0001 associations of preterm birth with early adult multimorbidity were less marked, with multivar- iable-adjusted HRs indicating 1.18-fold risk in females (95% CI: 1.12 to 1.24) and 1.10-fold risk in males (95% CI: 1.04 to 1.17). The associations of categories of gestational age (degrees of prematurity) with chronic dis- ease multimorbidity in adolescence and early adulthood are shown in Tables 2 and 3. We observed a dose-response relationship between increasing prematurity and increasing risks of both multimorbidity outcomes, and again the estimated increased risks were more prominent for adolescent multimorbidity than early adult multimorbidity. For example, the multivari- able-adjusted HRs indicate up to 7- and 6-fold risks of adolescent multimorbidity among males and females born at 23–27 weeks, respectively, whereas the corresponding risks for early adult multimorbidity in these groups were approximately 3-fold (Tables 2 and 3). Findings from the sensitivity analyses with additional adjustment for the year of birth were similar our main findings in direction and magnitude (S3 Table in S1 File). Overall, the absolute risks of chronic disease multimorbidity in adolescence and early adulthood in our study population were relatively low, but they were strongly accumulated in prematurely born individuals. The risks of adolescent multimorbidity ranged from 107.1/10,000 person-years in those born at 23–27 weeks to 17.2/10,000 person-years on those born at 39–41 weeks; the corresponding risks of early adult multimorbidity were 106.1/10,000 person years in the very earliest preterm group and 37.9/10,000 person-years among those born at full-term (S4 Table in S1 File). To examine combinations of chronic diseases that typically comprised multimorbidity in our study population, we examined the co-occurrence of individual chronic diseases and their associations with preterm birth (S5 Table in S1 File). Diseases that co-occurred with at least one other chronic disease more often among preterm than full-born individuals were urinary tract cancer, diabetes, schizophrenia, depression, multiple sclerosis, cardiovascular diseases, PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 6 / 15 PLOS ONEPreterm birth and risk of multimorbidity Table 2. Associations of gestational age with chronic disease multimorbidity in adolescence (age 10–17 years) in females and males. Gestational age (weeks) 23–27 28–31 32–33 34–36 37–38 39–41 42+ Gestational age (weeks) 23–27 28–31 32–33 34–36 37–38 39–41 42+ HR (95% CI) for chronic disease multimorbidity in adolescence (age 10–17 years) Females (n = 477,237) Unadjusted Adjusted for prenatal exposures1 6.27 (4.95 to 7.95) 5.88 (4.63 to 7.46) 4.44 (3.62 to 5.45) 4.18 (3.40 to 5.14) 2.38 (1.88 to 3.01) 2.31 (1.82 to 2.93) 1.58 (1.41 to 1.77) 1.56 (1.39 to 1.76) 1.20 (1.13 to 1.27) 1.19 (1.12 to 1.27) 1 (ref. cat.) 1 (ref. cat.) Adjusted for mother’s characteristics2 6.09 (4.81 to 7.72) 4.33 (3.53 to 5.32) 2.31 (1.82 to 2.92) 1.56 (1.40 to 1.75) 1.20 (1.13 to 1.27) 1 (ref. cat.) Adjusted for pregnancy disorders3 6.15 (4.85 to 7.79) 4.21 (3.44 to 5.18) 2.22 (1.75 to 2.81) 1.49 (1.33 to 1.67) 1.16 (1.09 to 1.24) 1 (ref. cat.) Multivariable- adjusted4 5.65 (4.45 to 7.18) 3.88 (3.15 to 4.77) 2.10 (1.65 to 2.67) 1.46 (1.30 to 1.65) 1.16 (1.09 to 1.23) 1 (ref. cat.) 1.03 (0.91 to 1.15) 1.02 (0.91 to 1.15) 1.01 (0.89 to 1.13) 1.03 (0.92 to 1.16) 1.01 (0.90 to 1.14) HR (95% CI) for chronic disease multimorbidity in adolescence (age 10–17 years) Males (n = 473,879) Unadjusted Adjusted for prenatal exposures1 7.74 (6.11 to 9.81) 6.94 (5.46 to 8.83) 5.46 (4.52 to 6.60) 4.79 (3.95 to 5.81) 3.30 (2.68 to 4.07) 3.02 (2.44 to 3.74) 1.61 (1.42 to 1.81) 1.51 (1.33 to 1.71) 1.10 (1.02 to 1.18) 1.08 (1.00 to 1.16) 1 (ref. cat.) 1 (ref. cat.) Adjusted for mother’s characteristics2 7.53 (5.94 to 9.55) 5.36 (4.44 to 6.48) 3.25 (2.63 to 4.01) 1.59 (1.41 to 1.79) 1.10 (1.02 to 1.18) 1 (ref. cat.) Adjusted for pregnancy disorders3 7.73 (6.09 to 9.79) 5.32 (4.40 to 6.43) 3.19 (2.58 to 3.93) 1.55 (1.37 to 1.75) 1.08 (1.00 to 1.16) 1 (ref. cat.) Multivariable- adjusted4 6.85 (5.39 to 8.71) 4.64 (3.82 to 5.64) 2.91 (2.35 to 3.60) 1.45 (1.28 to 1.65) 1.06 (0.98 to 1.14) 1 (ref. cat.) 1.13 (0.99 to 1.29) 1.12 80.98 to 1.28) 1.12 (0.98 to 1.27) 1.14 (1.00 to 1.30) 1.11 (0.98 to 1.27) 1 Birthweight z-score (continuous), multiple pregnancy (yes vs. no) and congenital anomalies (yes vs. no) 2 Mother’s age (years, continuous), parity (0, 1, 2, 3, 4, 5+), smoking during pregnancy (yes vs. no) and socioeconomic position (low, intermediate, high, not known) 3 Adjusted for mother’s diabetes (none, gestational diabetes, type 1/type 2 diabetes) and mother’s hypertension during pregnancy (yes vs. no) 4 Adjusted for all the above covariates. https://doi.org/10.1371/journal.pone.0261952.t002 chronic obstructive pulmonary disease, asthma, coeliac disease and/or dermatitis herpetifor- mis, rheumatoid diseases, renal disease, epilepsy and cerebral palsy. Compared to full-term born individuals, those born preterm were also more likely to have records of intellectual dis- abilities and disorders of psychological development, behaviour or personality co-occurring with other chronic diseases. Unadjusted HRs for the first chronic disease in adolescence or early adulthood are provided in S6 Table in S1 File. These indicate higher risks of any chronic disease in preterm born indi- viduals compared to those born at term, but the estimated risks for the first individual disease were notably smaller than the risks of chronic disease multimorbidity. Overlap of chronic disease multimorbidity with intellectual disabilities and disorders of psychological development, behaviour or personality The co-occurrence of multimorbidity in adolescence and early adulthood with intellectual dis- abilities and disorders of psychological development, behaviour or personality are shown in Table 4. Among those with intellectual disabilities, preterm birth was associated with consider- ably increased risks of adolescent multimorbidity (depending on gestational age), but there was no clear evidence for an association of gestational age with early adult multimorbidity in this group. The associations of gestational age with adolescent or early adult multimorbidity among individuals with disorders of psychological development, behavioural syndromes or PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 7 / 15 PLOS ONEPreterm birth and risk of multimorbidity Table 3. Associations of gestational age with chronic disease multimorbidity in early adulthood (age 18–30 years) in females and males. Gestational age (weeks) 23–27 28–31 32–33 34–36 37–38 39–41 42+ Gestational age (weeks) 23–27 28–31 32–33 34–36 37–38 39–41 42+ HR (95% CI) for chronic disease multimorbidity in early adulthood (age 18–30 years) Females (n = 324,094) Unadjusted Adjusted for prenatal exposures1 3.20 (2.44 to 4.19) 3.11 (2.37 to 4.08) 2.56 (2.06 to 3.18) 2.48 (1.99 to 3.08) 1.72 (1.37 to 2.15) 1.70 (1.36 to 2.13) 1.29 (1.16 to 1.42) 1.28 (1.15 to 1.43) 1.14 (1.08 to 1.20) 1.14 (1.08 to 1.20) 1 (ref. cat.) 1 (ref. cat.) Adjusted for mother’s characteristics2 3.12 (2.38 to 4.08) 2.52 (2.03 to 3.12) 1.68 (1.34 to 2.10) 1.27 (1.15 to 1.41) 1.14 (1.08 to 1.20) 1 (ref. cat.) Adjusted for pregnancy disorders3 3.20 (2.44 to 4.19) 2.53 (2.04 to 3.14) 1.69 (1.35 to 2.12) 1.26 (1.14 to 1.40) 1.12 (1.07 to 1.19) 1 (ref. cat.) Multivariable- adjusted4 3.04 (2.32 to 3.99) 2.40 (1.93 to 2.99) 1.62 (1.29 to 2.04) 1.24 (1.12 to 1.38) 1.12 (1.06 to 1.19) 1 (ref. cat.) 1.05 (0.94 to 1.16) 1.04 (0.94 to 1.15) 1.03 (0.93 to 1.14) 1.05 (0.95 to 1.16) 1.03 (0.93 to 1.14) HR (95% CI) for chronic disease multimorbidity in early adulthood (age 18–30 years) Males (n = 322,354) Unadjusted Adjusted for prenatal exposures1 2.79 (1.97 to 3.96) 2.61 (1.83 to 3.70) 2.88 (2.30 to 3.62) 2.61 (2.07 to 3.29) 1.76 (1.37 to 2.26) 1.65 (1.28 to 2.13) 1.29 (1.15 to 1.45) 1.23 (1.09 to 1.39) 1.06 (0.99 to 1.13) 1.04 (0.97 to 1.11) 1 (ref. cat.) 1 (ref. cat.) Adjusted for mother’s characteristics2 2.70 (1.90 to 3.83) 2.80 (2.23 to 3.52) 1.73 (1.35 to 2.22) 1.27 (1.13 to 1.43) 1.05 (0.99 to 1.12) 1 (ref. cat.) Adjusted for pregnancy disorders3 2.79 (1.97 to 3.96) 2.86 (2.28 to 2.60) 1.73 (1.35 to 2.22) 1.27 (1.13 to 1.43) 1.05 (0.98 to 1.12) 1 (ref. cat.) Multivariable- adjusted4 2.54 (1.78 to 3.60) 2.54 (2.01 to 3.20) 1.60 (1.24 to 2.06) 1.20 (1.06 to 1.35) 1.03 (0.96 to 1.10) 1 (ref. cat.) 1.23 (1.10 to 1.38) 1.23 (1.10 to 1.38) 1.22 (1.09 to 1.37) 1.24 (1.10 to 1.38) 1.22 (1.09 to 1.40) 1 Birthweight z-score (continuous), multiple pregnancy (yes vs. no) and congenital anomalies (yes vs. no) 2 2 Mother’s age (years, continuous), parity (0, 1, 2, 3, 4, 5+), smoking during pregnancy (yes vs. no) and socioeconomic position (low, intermediate, high, not known) 3 Adjusted for mother’s diabetes (none, gestational diabetes, type 1/type 2 diabetes) and mother’s hypertension during pregnancy (yes vs. no) 4 Adjusted for all above covariates. https://doi.org/10.1371/journal.pone.0261952.t003 personality disorders were similar to the main findings, with those born earlier having notably higher risks of these outcomes. Discussion Our findings, based on nationwide register data in Finland, suggest that compared to individu- als born full-term, preterm born individuals have ~1.5- to 7-fold increased risks of chronic dis- ease multimorbidity in adolescence and 1.2- to 3-fold increased risk of multimorbidity in early adulthood, with the estimated risks increasing in a dose-response manner with increasing pre- maturity. The risk patterns were broadly similar in males and females. There was also some indication that the association of gestational age with increased risk of multimorbidity was particularly prominent among individuals with intellectual disabilities or disorders of psycho- logical development, behaviour or personality. A growing body of evidence shows that prematurely born individuals have increased risks of individual chronic diseases (including metabolic syndrome [23, 24], ischaemic heart disease [19], asthma [27] and depression [31]) in the early part of the life course. The findings pre- sented here suggest that these risks also translate to increased risks of co-occurrence of multi- ple diseases among preterm born adolescents and young adults. Multimorbidity is a heterogeneous outcome and it is possible that the multimorbidity in our analyses represents concordant disease combinations (that share aetiology or risk factors) or discordant disease combinations (that co-occur for other reasons). We hypothesise that the associations of PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 8 / 15 PLOS ONEPreterm birth and risk of multimorbidity Table 4. Co-occurrence of developmental disorders and chronic disease multimorbidity. Intellectual disabilities Gestational age (weeks) N (%) with intellectual disabilities N (% of individuals with intellectual disabilities) HR1 (95% CI) Chronic disease multimorbidity in adolescence Chronic disease multimorbidity in early adulthood HR1 (95% CI) N (% of individuals with intellectual disabilities) 23–27 28–31 32–33 34–36 37–38 39–41 42+ All 151 (7.6) 256 (6.5) 201 (3.9) 780 (2.3) 2 588 (1.5) 8 589 (1.2) 583 (1.4) 13 148 (1.4) Disorders of psychological development Gestational age (weeks) N (%) with disorders(s) of psychological development 46 (30.5) 67 (26.2) 54 (26.9) 113 (14.5) 344 (13.3) 1 161 (13.5) 72 (12.4) 1 857 (14.1) 23.83 (6.53 to 87.03) 22 (3.5) 6.42 (4.41 to 9.35) 56 (21.9) 4.04 (2.96 to 5.51) 34 (16.9) 1.45 (1.18 to 1.78) 142 (18.2) 1.14 (1.01 to 1.30) 599 (23.2) 1 (ref. cat.) 2 380 (27.7) 0.78 (0.61 to 0.99) 150 (25.7) - 3 383 (25.7) 1.10 (0.68 to 1.80) 1.51 (1.10 to 2.05) 0.90 (0.62 to 1.29) 0.93 (0.77 to 1.12) 1.00 (0.91 to 1.10) 1 (ref. cat.) 0.92 (0.77 to 1.08) - Chronic disease multimorbidity in adolescence N (% of individuals with disorders(s) of psychological development) HR1 (95% CI) Chronic disease multimorbidity in early adulthood HR1 (95% CI) N (% of individuals with disorders(s) of psychological development) 23–27 28–31 32–33 34–36 37–38 39–41 42+ All 522 (26.1) 820 (20.7) 767 (15.0) 3 181 (9.4) 11 374 (6.7) 39 484 (5.7) 2 688 (6.5) 58 836 (6.2) Behavioural syndromes 79 (15.1) 89 (10.9) 69 (9.0) 175 (5.5) 493 (4.3) 1 644 (4.2) 99 (3.7) 2 648 (4.5) 7.95 (3.84 to 16.47) 36 (6.9) 4.41 (2.78 to 7.00) 3.10 (2.20 to 4.37) 64 (7.8) 38 (5.0) 1.60 (1.30 to 1.97) 121 (3.8) 1.14 (1.02 to 1.28) 448 (3.9) 1 (ref. cat.) 1 538 (3.9) 0.81 (0.66 to 1.00) 105 (3.9) - 2 350 (4.0) 6.86 (1.82 to 25.85) 5.52 (2.57 to 11.84) 2.44 (1.43 to 1.96) 1.45 (1.08 to 1.96) 1.23 (1.07 to 1.43) 1 (ref. cat.) 0.87 (0.70 to 1.08) - Gestational age (weeks) N (%) with behavioural syndrome(s) N (% of individuals with behavioural syndrome/s) HR1 (95% CI) Chronic disease multimorbidity in adolescence Chronic disease multimorbidity in early adulthood HR1 (95% CI) N (% of individuals with behavioural syndrome/s) 23–27 28–31 32–33 34–36 37–38 39–41 42+ All 61 (3.1) 158 (4.0) 210 (4.1) 958 (2.9) 4 151 (2.5) 16 171 (2.3) 923 (2.2) 22 642 (2.4) Personality disorders 13 (21.3) 20 (12.7) 23 (11.0) 90 (9.3) 317 (7.7) 1 278 (7.9) 68 (7.4) 1 809 (8.0) 4.29 (2.41 to 7.64) 12 (19.7) 2.12 (1.32 to 3.40) 30 (19.0) 1.63 (1.06 to 2.52) 24 (11.4) 1.27 (1.01 to 1.59) 120 (12.4) 1.00 (0.88 to 1.14) 512 (12.3) 1 (ref. cat.) 1 886 (11.7) 0.89 (0.69 to 1.13) 116 (12.6) - 2 700 (11.9) 4.82 (1.82 to 12.75) 3.52 (1.86 to 6.65) 1.54 (0.90 to 2.64) 1.43 (1.08 to 1.88) 1.21 (1.06 to 1.38) 1 (ref. cat.) 0.99 (0.81 to 1.21) - Gestational age (weeks) N (%) personality disorder (s) N (% of individuals with personality disorder7s) HR1 (95% CI) Chronic disease multimorbidity in adolescence Chronic disease multimorbidity in early adulthood HR1 (95% CI) N (% of individuals with personality disorder/s) 23–27 28–31 32–33 34–36 37–38 39–41 42+ All 128 (6.4) 306 (7.7) 382 (7.5) 2 348 (6.9) 10 103 (6.0) 38 923 (5.6) 2 343 (5.7) 54 533 (5.7) 27 (21.1) 31 (10.1) 26 (6.8) 160 (6.8) 575 (5.7) 2 010 (5.2) 114 (4.9) 2 943 (5.4) 8.18 (3.99 to 16.79) 25 (19.5) 3.09 (1.94 to 4.93) 36 (11.8) 1.68 (1.10 to 2.58) 42 (11.0) 1.54 (1.28 to 1.86) 209 (8.9) 1.18 (1.06 to 1.30) 833 (8.3) 1 (ref. cat.) 3 362 (8.6) 0.88 80.73 to 1.07) 197 (8.4) - 4 704 (8.6) 7.44 (3.15 to 17.59) 2.69 (1.54 to 1.70) 2.21 (1.47 to 3.33) 1.39 (1.13 to 1.70) 1.11 (1.01 to 1.23) 1 (ref. cat.) 0.92 (0.79 to 1.07) - 1 Adjusted for sex, birthweight z-score (continuous), multiple pregnancy (yes vs. no), congenital anomalies (yes vs. no), mother’s age (years, continuous), mother’s parity (0, 1, 2, 3, 4, 5+), mother’s smoking during pregnancy (yes vs. no), mother’s socioeconomic position (low, intermediate, high, unknown), mother’s diabetes (none, gestational diabetes, type 1/type 2 diabetes) and mother’s hypertension during pregnancy (yes vs. no). https://doi.org/10.1371/journal.pone.0261952.t004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 9 / 15 PLOS ONEPreterm birth and risk of multimorbidity preterm birth with multimorbidity in our study population may be driven by the diseases that co-occurred with at least one other chronic disease more often among preterm than full-born individuals: urinary tract cancer, diabetes, schizophrenia, depression, multiple sclerosis, car- diovascular diseases, chronic obstructive pulmonary disease, asthma, coeliac disease, rheuma- toid diseases, renal disease, epilepsy and cerebral palsy. Potential mechanisms One possible explanation for the increased risk of multimorbidity across the spectrum of pre- mature gestational ages is that preterm born individuals are exposed to in utero and perinatal events that can intervene in organ system development at critical stages (e.g. during neural or vascular site development, nephrogenesis and bone mineralisation) [39]. Examples of in utero exposures include mother’s diabetes or hypertensive disorder during pregnancy, which can influence glucose metabolism and insulin sensitivity in the prematurely born individual through shared genetic susceptibility or developmental programming. Perinatal events, e.g. prolonged ventilation, can cause injury and result in inflammation and potentially dysfunc- tional repair processes, leading to respiratory system disease later in life [39]. Other possibilities include some characteristics of preterm born individuals influencing their risk of multimorbidity. Longitudinal studies of people born very prematurely have reported on a “preterm behavioural phenotype”, characterised by symptoms and disorders associated with inattention, anxiety and social difficulties, as well as low cognitive abilities [28, 55]. Many preterm-born young adults attain lower education [56, 57] and are less likely than their term-born counterparts to have couple relationships, start a family and have children [58]. There is also evidence that preterm birth is associated with certain lifestyle factors, includ- ing sedentary behaviour [59, 60]. Multimorbidity could mediate the association of preterm birth with educational achievement or physical activity, or low socioeconomic position, living alone or sedentary lifestyle could be risk factors for multimorbidity among individuals with a preterm-born background. In this context, our findings of consistently increased risks of mul- timorbidity across the spectrum of premature gestational ages indicate directions for further research into the causal pathways between mental and physical health, and behavioural and socioeconomic characteristics of adolescents and adults born preterm. Strengths and limitations An important strength of our analyses is that we used a large set of nationwide register data, with up to 20 years’ follow-up. These whole-population data captured near-complete informa- tion on all births and specialised health care episodes in Finland during the study period, thus minimising the risks of sample selection or loss to follow-up introducing bias to our findings. Furthermore, a dataset of over 950,000 individuals and over 15,000 cases of chronic disease multimorbidity provided sufficient statistical power to precisely estimate the risks of multi- morbidity across a wide spectrum of gestational ages, including among extremely prematurely born individuals, who are often under-represented in other types of observational studies. Limitations of our study derive from using routinely collected register data. The multimor- bidity outcomes were ascertained from the nationwide Care Register for Healthcare, which overall has good coverage, with over 95% of individuals and care episodes identifiable by unique personal identity numbers [61]. Validation studies suggest that the accuracy of diag- nostic coding for cardiovascular diseases, cancer, psychiatric diseases and autism ranged from 88% to 96% [61]. Although it is thus possible that some multimorbid disease combinations have been incompletely recorded in Care Register for Healthcare, which might have intro- duced bias in our estimates, we would expect such bias to be unrelated to gestational age or PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 10 / 15 PLOS ONEPreterm birth and risk of multimorbidity covariates. Also, many chronic diseases (e.g. asthma or well-controlled diabetes) can be man- aged in primary care and for these diseases, the multimorbidity outcomes in our analyses rep- resent the severe end of the disease spectrum. Consequently, our findings are possibly not generalisable to multimorbidity consisting of less severe combinations of certain diseases. Unfortunately, good quality nationwide primary care data were not available for the time period covered in our investigation and we were thus unable to explore this further. Although our analyses were adjusted for a number of potential confounders, we cannot exclude the pos- sibility that residual confounding from imprecisely measured, unmeasured or unknown con- founders has influenced our estimates. For instance, consistently recorded data on mother’s pre-pregnancy weight and pregnancy weight gain were not available for the time period cov- ered in our analyses, and we were unable to adjust our analyses for these. Implications Our findings highlight the long-term health implications of preterm birth. Advances in perina- tal care in the past decades mean that increasing proportions of even the smallest prematurely born babies now survive, and as a consequence, increasing numbers of people in paediatric, school, student and occupational health care settings have a preterm-born background. There are currently no clinical guidelines for follow-up of prematurely born individuals beyond childhood. Whilst our observations on their own cannot provide a basis for clinical practice recommendations, they suggest that information on gestational age at birth would be a useful characteristic to include in a medical history when assessing the risk of chronic diseases in ado- lescent and adult patients. Preterm birth could be one early life exposure that could be used to inform targeted screening or preventive strategies. Conclusions Our findings indicate that earlier gestational age at birth is associated with an increased risk of chronic disease multimorbidity in adolescence and early adulthood. These findings suggest that information on gestational age would be a useful characteristic to include in a medical his- tory when assessing the risk of chronic diseases in adolescent and adult patients. Supporting information S1 File. (PDF) S1 Checklist. STROBE statement—checklist of items that should be included in reports of cohort studies. (PDF) Author Contributions Conceptualization: Katriina Heikkila¨. Data curation: Anna Pulakka, Mika Gissler. Formal analysis: Katriina Heikkila¨. Funding acquisition: Suvi Alenius, Petteri Hovi, Mika Gissler, Sven Sandin, Eero Kajantie. Investigation: Katriina Heikkila¨, Anna Pulakka, Johanna Metsa¨la¨, Suvi Alenius, Petteri Hovi, Eero Kajantie. PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 11 / 15 PLOS ONEPreterm birth and risk of multimorbidity Methodology: Katriina Heikkila¨, Anna Pulakka, Johanna Metsa¨la¨, Suvi Alenius, Petteri Hovi, Mika Gissler, Sven Sandin, Eero Kajantie. Project administration: Katriina Heikkila¨, Anna Pulakka, Johanna Metsa¨la¨, Suvi Alenius. Resources: Mika Gissler, Sven Sandin, Eero Kajantie. Writing – original draft: Katriina Heikkila¨. Writing – review & editing: Katriina Heikkila¨, Anna Pulakka, Johanna Metsa¨la¨, Suvi Alenius, Petteri Hovi, Mika Gissler, Sven Sandin, Eero Kajantie. References 1. Diederichs C, Berger K, Bartels DB. The Measurement of Multiple Chronic Diseases-A Systematic Review on Existing Multimorbidity Indices. Journals of Gerontology Series A Biological Sciences and Medical Sciences. 2011; 66(3):301–11. https://doi.org/10.1093/gerona/glq208 PMID: 21112963 2. Violan C, Foguet-Boreu Q, Flores-Mateo G, Salisbury C, Blom J, Freitag M, et al. Prevalence, determi- nants and patterns of multimorbidity in primary care: a systematic review of observational studies. PloS one. 2014; 9(7):e102149. https://doi.org/10.1371/journal.pone.0102149 PMID: 25048354 3. Xu X, Mishra GD, Jones M. Evidence on multimorbidity from definition to intervention: An overview of systematic reviews. Ageing Res Rev. 2017; 37:53–68. https://doi.org/10.1016/j.arr.2017.05.003 PMID: 28511964 4. Willadsen TG, Bebe A, Koster-Rasmussen R, Jarbol DE, Guassora AD, Waldorff FB, et al. The role of diseases, risk factors and symptoms in the definition of multimorbidity—a systematic review. Scand J Prim Health. 2016; 34(2):112–21. 5. Stirland LE, Gonzalez-Saavedra L, Mullin DS, Ritchie CW, Muniz-Terrera G, Russ TC. Measuring multi- morbidity beyond counting diseases: systematic review of community and population studies and guide to index choice. BMJ. 2020; 368:m160. https://doi.org/10.1136/bmj.m160 PMID: 32071114 6. Buurman BM, Frenkel WJ, Abu-Hanna A, Parlevliet JL, de Rooij SE. Acute and chronic diseases as part of multimorbidity in acutely hospitalized older patients. Eur J Intern Med. 2016; 27:68–75. https:// doi.org/10.1016/j.ejim.2015.09.021 PMID: 26477016 7. Fortin M, Lapointe L, Hudon C, Vanasse A, Ntetu AL, Maltais D. Multimorbidity and quality of life in pri- mary care: a systematic review. Health Qual Life Outcomes. 2004; 2:51. https://doi.org/10.1186/1477- 7525-2-51 PMID: 15380021 8. Ryan A, Wallace E, O’Hara P, Smith SM. Multimorbidity and functional decline in community-dwelling adults: a systematic review. Health Qual Life Outcomes. 2015; 13:168. https://doi.org/10.1186/s12955- 015-0355-9 PMID: 26467295 9. Sciences AoM. Multimorbidity: a Priority for Global Health Research. Academy of Medical Sciences; 2018. 10. Cornish RP, Boyd A, Van Staa T, Salisbury C, Macleod J. Socio-economic position and childhood multi- morbidity: a study using linkage between the Avon Longitudinal Study of Parents and Children and the General Practice Research Database. Int J Equity Health. 2013; 12:66. https://doi.org/10.1186/1475- 9276-12-66 PMID: 23962118 11. Russell J, Grant CC, Morton SMB. Multimorbidity in Early Childhood and Socioeconomic Disadvantage: Findings From a Large New Zealand Child Cohort. Acad Pediatr. 2020; 20(5):619–27. https://doi.org/ 10.1016/j.acap.2019.09.007 PMID: 31574311 12. Jani BD, Hanlon P, Nicholl BI, McQueenie R, Gallacher KI, Lee D, et al. Relationship between multimor- bidity, demographic factors and mortality: findings from the UK Biobank cohort. BMC Medicine. 2019;17. https://doi.org/10.1186/s12916-019-1250-8 PMID: 30764819 13. Chawanpaiboon S, Vogel JP, Moller AB, Lumbiganon P, Petzold M, Hogan D, et al. Global, regional, and national estimates of levels of preterm birth in 2014: a systematic review and modelling analysis. Lancet Glob Health. 2019; 7(1):e37–e46. https://doi.org/10.1016/S2214-109X(18)30451-0 PMID: 30389451 14. WHO WHO. WHO Fact Sheet Preterm Birth: WHO; 2018 [Available from: https://www.who.int/news- room/fact-sheets/detail/preterm-birth. 15. Blencowe H, Cousens S, Oestergaard MZ, Chou D, Moller AB, Narwal R, et al. National, regional, and worldwide estimates of preterm birth rates in the year 2010 with time trends since 1990 for selected countries: a systematic analysis and implications. Lancet. 2012; 379(9832):2162–72. https://doi.org/10. 1016/S0140-6736(12)60820-4 PMID: 22682464 PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 12 / 15 PLOS ONEPreterm birth and risk of multimorbidity 16. Howson C, Kinney M, Lawn J, editors. Born too soon: the global action report on preterm birth. Geneva: World Health Organisation; 2012. 17. Breeman LD, Jaekel J, Baumann N, Bartmann P, Wolke D. Preterm Cognitive Function Into Adulthood. Pediatrics. 2015; 136(3):415–23. https://doi.org/10.1542/peds.2015-0608 PMID: 26260714 18. Ekeus C, Lindstrom K, Lindblad F, Rasmussen F, Hjern A. Preterm birth, social disadvantage, and cog- nitive competence in Swedish 18- to 19-year-old men. Pediatrics. 2010; 125(1):e67–73. https://doi.org/ 10.1542/peds.2008-3329 PMID: 19969613 19. Crump C, Howell EA, Stroustrup A, McLaughlin MA, Sundquist J, Sundquist K. Association of Preterm Birth With Risk of Ischemic Heart Disease in Adulthood. JAMA Pediatrics. 2019; 173(8):736–43. https:// doi.org/10.1001/jamapediatrics.2019.1327 PMID: 31157896 20. Hovi P, Vohr B, Ment LR, Doyle LW, McGarvey L, Morrison KM, et al. Blood Pressure in Young Adults Born at Very Low Birth Weight: Adults Born Preterm International Collaboration. Hypertension. 2016; 68(4):880–7. https://doi.org/10.1161/HYPERTENSIONAHA.116.08167 PMID: 27572149 21. Markopoulou P, Papanikolaou E, Analytis A, Zoumakis E, Siahanidou T. Preterm Birth as a Risk Factor for Metabolic Syndrome and Cardiovascular Disease in Adult Life: A Systematic Review and Meta-Anal- ysis. J Pediatr. 2019; 210:69-80.e5. https://doi.org/10.1016/j.jpeds.2019.02.041 PMID: 30992219 22. Morrison KM, Ramsingh L, Gunn E, Streiner D, Van Lieshout R, Boyle M, et al. Cardiometabolic Health in Adults Born Premature With Extremely Low Birth Weight. Pediatrics. 2016; 138(4). https://doi.org/10. 1542/peds.2016-0515 PMID: 27590899 23. Sipola-Leppanen M, Vaarasmaki M, Tikanmaki M, Matinolli HM, Miettola S, Hovi P, et al. Cardiometa- bolic risk factors in young adults who were born preterm. Am J Epidemiol. 2015; 181(11):861–73. https://doi.org/10.1093/aje/kwu443 PMID: 25947956 24. 25. Liao L, Deng Y, Zhao D. Association of Low Birth Weight and Premature Birth With the Risk of Metabolic Syndrome: A Meta-Analysis. Front Pediatr. 2020; 8:405. https://doi.org/10.3389/fped.2020.00405 PMID: 32850529 de Mendonca E, de Lima Macena M, Bueno NB, de Oliveira ACM, Mello CS. Premature birth, low birth weight, small for gestational age and chronic non-communicable diseases in adult life: A systematic review with meta-analysis. Early Hum Dev. 2020; 149:105154. https://doi.org/10.1016/j.earlhumdev. 2020.105154 PMID: 32799034 26. Doyle LW, Andersson S, Bush A, Cheong JLY, Clemm H, Evensen KAI, et al. Expiratory airflow in late adolescence and early adulthood in individuals born very preterm or with very low birthweight compared with controls born at term or with normal birthweight: a meta-analysis of individual participant data. Lan- cet Respir Med. 2019; 7(8):677–86. https://doi.org/10.1016/S2213-2600(18)30530-7 PMID: 31078498 27. 28. Tronnes H, Wilcox AJ, Lie RT, Markestad T, Moster D. The association of preterm birth with severe asthma and atopic dermatitis: a national cohort study. Pediat Allerg Imm-Uk. 2013; 24(8):782–7. https:// doi.org/10.1111/pai.12170 PMID: 24298940 Johnson S, Marlow N. Preterm birth and childhood psychiatric disorders. Pediatr Res. 2011; 69(5 Pt 2):11R-8R. https://doi.org/10.1203/PDR.0b013e318212faa0 PMID: 21289534 29. Persson M, Opdahl S, Risnes K, Gross R, Kajantie E, Reichenberg A, et al. Gestational age and the risk of autism spectrum disorder in Sweden, Finland, and Norway: A cohort study. PLoS medicine. 2020; 17 (9):e1003207. https://doi.org/10.1371/journal.pmed.1003207 PMID: 32960896 30. Pierrat V, Marchand-Martin L, Marret S, Arnaud C, Benhammou V, Cambonie G, et al. Neurodevelop- mental outcomes at age 5 among children born preterm: EPIPAGE-2 cohort study. BMJ (Clinical research ed). 2021; 373:n741. https://doi.org/10.1136/bmj.n741 PMID: 33910920 31. Kajantie E, Strang-Karlsson S, Evensen KAI, Haaramo P. Adult outcomes of being born late preterm or early term—What do we know? Semin Fetal Neonat M. 2019; 24(1):66–83. 32. Risnes K, Bilsteen JF, Brown P, Pulakka A, Andersen A-MN, Opdahl S, et al. Mortality Among Young Adults Born Preterm and Early Term in 4 Nordic Nations. JAMA Netw Open. 2021; 4(1):e2032779. https://doi.org/10.1001/jamanetworkopen.2020.32779 PMID: 33416885 33. Gill JV, Boyle EM. Outcomes of infants born near term. Arch Dis Child. 2017; 102(2):194–8. https://doi. org/10.1136/archdischild-2015-309584 PMID: 27543506 34. Moutquin JM. Classification and heterogeneity of preterm birth. BJOG. 2003; 110 Suppl 20:30–3. https://doi.org/10.1016/s1470-0328(03)00021-1 PMID: 12763108 35. Gissler M, Teperi J, Hemminki E, Merilainen J. Data quality after restructuring a national medical regis- try. Scand J Soc Med. 1995; 23(1):75–80. https://doi.org/10.1177/140349489502300113 PMID: 7784857 36. Crump C, Winkleby MA, Sundquist J, Sundquist K. Prevalence of Survival Without Major Comorbidities Among Adults Born Prematurely. Jama-J Am Med Assoc. 2019; 322(16):1580–8. https://doi.org/10. 1001/jama.2019.15040 PMID: 31638681 PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 13 / 15 PLOS ONEPreterm birth and risk of multimorbidity 37. Heo JS, Lee JM. The Long-Term Effect of Preterm Birth on Renal Function: A Meta-Analysis. Int J Envi- ron Res Public Health. 2021; 18(6). https://doi.org/10.3390/ijerph18062951 PMID: 33805740 38. 39. Li W, Peng A, Deng S, Lai W, Qiu X, Zhang L, et al. Do premature and postterm birth increase the risk of epilepsy? An updated meta-analysis. Epilepsy Behav. 2019; 97:83–91. https://doi.org/10.1016/j. yebeh.2019.05.016 PMID: 31202097 Luu TM, Katz SL, Leeson P, Thebaud B, Nuyt A-M. Preterm birth: risk factor for early-onset chronic dis- eases. Cmaj. 2016; 188(10):736–46. https://doi.org/10.1503/cmaj.150450 PMID: 26644500 40. Paquette K, Coltin H, Boivin A, Amre D, Nuyt AM, Luu TM. Cancer risk in children and young adults born preterm: A systematic review and meta-analysis. PloS one. 2019; 14(1):e0210366. https://doi.org/ 10.1371/journal.pone.0210366 PMID: 30608983 41. Raju TNK, Buist AS, Blaisdell CJ, Moxey-Mims M, Saigal S. Adults born preterm: a review of general health and system-specific outcomes. Acta Paediatr. 2017; 106(9):1409–37. https://doi.org/10.1111/ apa.13880 PMID: 28419544 42. Robinson R, Lahti-Pulkkinen M, Schnitzlein D, Voit F, Girchenko P, Wolke D, et al. Mental health out- comes of adults born very preterm or with very low birth weight: A systematic review. Semin Fetal Neo- nat M. 2020; 25(3):101113. 43. Stahlberg T, Khanal P, Chudal R, Luntamo T, Kronstrom K, Sourander A. Prenatal and perinatal risk factors for anxiety disorders among children and adolescents: A systematic review. J Affect Disord. 2020; 277:85–93. https://doi.org/10.1016/j.jad.2020.08.004 PMID: 32799108 44. Sonnenschein-van der Voort AM, Arends LR, de Jongste JC, Annesi-Maesano I, Arshad SH, Barros H, et al. Preterm birth, infant weight gain, and childhood asthma risk: a meta-analysis of 147,000 European children. J Allergy Clin Immunol. 2014; 133(5):1317–29. https://doi.org/10.1016/j.jaci.2013.12.1082 PMID: 24529685 45. Philips EM, Santos S, Trasande L, Aurrekoetxea JJ, Barros H, von Berg A, et al. Changes in parental smoking during pregnancy and risks of adverse birth outcomes and childhood overweight in Europe and North America: An individual participant data meta-analysis of 229,000 singleton births. PLoS Med. 2020; 17(8):e1003182. https://doi.org/10.1371/journal.pmed.1003182 PMID: 32810184 46. Savran O, Ulrik CS. Early life insults as determinants of chronic obstructive pulmonary disease in adult life. Int J Chron Obstruct Pulmon Dis. 2018; 13:683–93. https://doi.org/10.2147/COPD.S153555 PMID: 29520136 47. Marleen S, Hettiarachchi J, Dandeniya R, Macgreggor R, Aquilina J, Khalil A, et al. Maternal clinical pre- dictors of preterm birth in twin pregnancies: A systematic review involving 2,930,958 twin pregnancies. Eur J Obstet Gynecol Reprod Biol. 2018; 230:159–71. https://doi.org/10.1016/j.ejogrb.2018.09.025 PMID: 30292096 48. Ruiz M, Goldblatt P, Morrison J, Kukla L, Sˇ vancara J, Riitta-Ja¨ rvelin M, et al. Mother’s education and the risk of preterm and small for gestational age birth: a DRIVERS meta-analysis of 12 European cohorts. J Epidemiol Community Health. 2015; 69(9):826–33. https://doi.org/10.1136/jech-2014- 205387 PMID: 25911693 49. Bramham K, Parnell B, Nelson-Piercy C, Seed PT, Poston L, Chappell LC. Chronic hypertension and pregnancy outcomes: systematic review and meta-analysis. BMJ. 2014; 348:g2301. https://doi.org/10. 1136/bmj.g2301 PMID: 24735917 50. Kozuki N, Lee AC, Silveira MF, Sania A, Vogel JP, Adair L, et al. The associations of parity and maternal age with small-for-gestational-age, preterm, and neonatal and infant mortality: a meta-analysis. BMC Public Health. 2013; 13 Suppl 3(Suppl 3):S2. 51. Papageorghiou AT, Ohuma EO, Altman DG, Todros T, Cheikh Ismail L, Lambert A, et al. International standards for fetal growth based on serial ultrasound measurements: the Fetal Growth Longitudinal Study of the INTERGROWTH-21st Project. Lancet. 2014; 384(9946):869–79. https://doi.org/10.1016/ S0140-6736(14)61490-2 PMID: 25209488 52. Gissler M, Merilainen J, Vuori E, Hemminki E. Register based monitoring shows decreasing socioeco- nomic differences in Finnish perinatal health. J Epidemiol Commun H. 2003; 57(6):433–9. https://doi. org/10.1136/jech.57.6.433 PMID: 12775790 53. Lambert PC, Royston P. Further development of flexible parametric models for survival analysis. Stata J. 2009; 9(2):265–90. 54. Royston P, Parmar MK. Flexible parametric proportional-hazards and proportional-odds models for censored survival data, with application to prognostic modelling and estimation of treatment effects. Stat Med. 2002; 21(15):2175–97. https://doi.org/10.1002/sim.1203 PMID: 12210632 55. Eves R, Mendonca M, Baumann N, Ni YY, Darlow BA, Horwood J, et al. Association of Very Preterm Birth or Very Low Birth Weight With Intelligence in Adulthood An Individual Participant Data Meta-analy- sis. Jama Pediatrics. 2021. https://doi.org/10.1001/jamapediatrics.2021.1058 PMID: 34047752 PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 14 / 15 PLOS ONEPreterm birth and risk of multimorbidity 56. Bilsteen JF, Taylor-Robinson D, Borch K, Strandberg-Larsen K, Nybo Andersen A-M. Gestational Age and Socioeconomic Achievements in Young Adulthood: A Danish Population-Based Study. JAMA Netw Open. 2018; 1(8):e186085. https://doi.org/10.1001/jamanetworkopen.2018.6085 PMID: 30646301 57. McBryde M, Fitzallen GC, Liley HG, Taylor HG, Bora S. Academic Outcomes of School-Aged Children Born Preterm: A Systematic Review and Meta-analysis. JAMA Netw Open. 2020; 3(4):e202027. https:// doi.org/10.1001/jamanetworkopen.2020.2027 PMID: 32242904 58. Mendonca M, Bilgin A, Wolke D. Association of Preterm Birth and Low Birth Weight With Romantic Part- nership, Sexual Intercourse, and Parenthood in Adulthood: A Systematic Review and Meta-analysis. JAMA Netw Open. 2019; 2(7):e196961. https://doi.org/10.1001/jamanetworkopen.2019.6961 PMID: 31298716 59. Lowe J, Watkins WJ, Kotecha SJ, Kotecha S. Physical Activity and Sedentary Behavior in Preterm- Born 7-Year Old Children. PloS one. 2016; 11(5):e0155229. https://doi.org/10.1371/journal.pone. 0155229 PMID: 27168339 60. Brylka A, Wolke D, Ludyga S, Bilgin A, Spiegler J, Trower H, et al. Physical Activity, Mental Health, and Well-Being in Very Pre-Term and Term Born Adolescents: An Individual Participant Data Meta-Analysis of Two Accelerometry Studies. Int J Env Res Pub He. 2021; 18(4). https://doi.org/10.3390/ ijerph18041735 PMID: 33579022 61. Sund R. Quality of the Finnish Hospital Discharge Register: a systematic review. Scand J Public Health. 2012; 40(6):505–15. https://doi.org/10.1177/1403494812456637 PMID: 22899561 PLOS ONE \| https://doi.org/10.1371/journal.pone.0261952 December 31, 2021 15 / 15 PLOS ONE
10.1371_journal.pone.0262642	RESEARCH ARTICLE Non-communicable diseases (NCDs) and vulnerability to COVID-19: The case of adult patients with hypertension or diabetes mellitus in Gamo, Gofa, and South Omo zones in Southern Ethiopia Fikre Bojola1, Wondimagegn Taye2, Habtamu Samuel2, Bahiru Mulatu2, Aknaw Kawza3, Aleme MekuriaID 2* 1 Department of Clinical nursing, Arba Minch College of Health Sciences, Arba Minch, Ethiopia, 2 Department of public health, Arba Minch College of Health Sciences, Arba Minch, Ethiopia, 3 Southern Region Health Office, Ethiopia * alemmekurishet@gmail.com Abstract a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Background Citation: Bojola F, Taye W, Samuel H, Mulatu B, Kawza A, Mekuria A (2022) Non-communicable diseases (NCDs) and vulnerability to COVID-19: The case of adult patients with hypertension or diabetes mellitus in Gamo, Gofa, and South Omo zones in Southern Ethiopia. PLoS ONE 17(1): e0262642. https://doi.org/10.1371/journal. pone.0262642 A growing body of evidence demonstrating that individuals with Non-Communicable Dis- ease (NCD) are more likely to have severe forms of COVID-19 and subsequent mortality. Hence, our study aimed to assess the knowledge of vulnerability and preventive practices towards COVID-19 among patients with hypertension or diabetes in Southern Ethiopia. Objective Editor: Tauqeer Hussain Mallhi, Jouf University, Kingdom of Saudi Arabia, SAUDI ARABIA To assess the knowledge and preventive practices towards COVID-19 among patients with hypertension or diabetes mellitus in three zones of Southern Ethiopia, 2020. Received: October 5, 2020 Accepted: December 31, 2021 Published: January 25, 2022 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0262642 Copyright: © 2022 Bojola et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Methods A community-based cross-sectional study design was used with a multi-stage random sam- pling technique to select 682 patients with hypertension or diabetes mellitus from 10th -17th July 2020 at the three zones of Southern Ethiopia. Logistic regression analysis with a 95% confidence interval was fitted to identify independent predictors of knowledge and preven- tive practices towards COVID-19. The adjusted odds ratio (AOR) was used to determine the magnitude of the association between the outcome and independent variables. P-value <0.05 is considered statistically significant. Results The Multi-dimensional knowledge (MDK) analysis of COVID-19 revealed that 63% of study subjects had good knowledge about COVID-19. The overall preventive practice towards COVID -19 was 26.4%. Monthly income (AOR = 1.42; 95% CI: 1.04, 1.94) significantly PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 1 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Funding: The authors received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. predicted knowledge towards COVID-19. Ninety-five percent of the study subjects knew that the COVID-19 virus spreads via respiratory droplets of infected individuals. One hun- dred and ten (16.2%) of study subjects correctly responded to the questions that state whether people with the COVID-19 virus who do not have a fever can infect the other. Knowledge about COVID-19 (AOR = 1.47; 95% CI: 1.03, 2.1) became the independent pre- dictor of preventive practice. Conclusions In this study, the knowledge of the respondents towards the COVID-19 pandemic was good. But the preventive practice was very low. There was a significant gap between knowledge and preventive practices towards the COVID-19 pandemic among the study subjects. Monthly income was significantly associated with knowledge of COVID-19. Knowledge of COVID-19 was found to be an independent predictor of preventive practice towards COVID- 19. Community mobilization and improving COVID-19- related knowledge and practice are urgently recommended for those patients with hypertension or diabetes mellitus. Introduction Non-communicable diseases (NCDs) are becoming a major health care challenge in the world. They are presently competing with traditionally leading killer diseases in the death toll. The topmost killer NCDs are; various types of cancers, chronic respiratory illnesses, stroke, and cardiovascular diseases [1]. Globally, ischemic heart disease, stroke, and Chronically Obstruc- tive Pulmonary Diseases (COPD) are the three leading causes of mortality. Non-communica- ble diseases (NCDs) constitute close to 54% of the overall disease burden, as measured in disability-adjusted life years (DALYs) [2]. Non-communicable diseases and associated risk factors (unhealthy diets, physical inactiv- ity, harmful use of alcohol) are on the rise in developing countries, posing a threat to the health and financial systems of emerging economies [2]. Because non-communicable diseases have not been received equal attention with communicable diseases in middle-income and low- income countries, people of these countries are disproportionately suffering from the conse- quences of these diseases [3, 4]. According to the World Health Organization (WHO) 2014 country profile, about 30% of total deaths in Ethiopia were associated with NCDs from which cardiovascular diseases, can- cers, chronic respiratory diseases, and diabetes are the leading causes of morbidity and mortal- ity. Similarly, the report revealed disproportionate age-specific death rates with a significant rise in deaths from non-communicable diseases between the ages 30 and 70 years [5]. Globally, the overall Case Fatality Rates of COVID-19 vary between countries. For instance, 4.1% in China, 4.6% in Spain, 8.3% in Italy,2.73% in Egypt, and 1.6% in Ethiopia [6]. The fatal- ity rate of patients with COVID-19 was highest in in-person aged 80, ranging from 13% to 16.7%, followed by 7.2%–8.9% among those aged 70–79 years [7]. However, patients with NCDs are more likely to have severe disease and subsequent mortality. The COVID-19 pan- demic has had widespread health impacts, revealing the particular vulnerability of those with underlying conditions [8]. The most commonly reported non-communicable diseases that complicate COVID-19 and lead to increased morbidity and mortality are: diabetes mellitus (DM), cardiovascular diseases PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 2 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 like hypertension, cerebrovascular disease, coronary artery disease (CAD), and respiratory dis- ease like chronic obstructive pulmonary disease (COPD) and tuberculosis [9]. Physical distancing or quarantine can lead to poor management of NCD behavioral risk factors, including unhealthy diet, physical inactivity, tobacco. Patients living with NCDs are at increased risk of the health impacts of emergencies such as COVID-19 [10]. An association between COVID-19 severity and NCDs has also been reported in China and the USA. However, many COVID-19 deaths also occur in older people who often have existing comorbidities [11]. Body-mass index (BMI) might also be associated with the severity of COVID-19; in China, patients with severe COVID-19 and non-survivors typically had a high BMI (>25 kg/m2). The impact of COVID-19 response measures on NCDs is multifaceted [12]. A study conducted on COVID-19 in Wuhan, China, showed that from 52 intensive care unit patients with novel coronavirus disease, 22% had cerebrovascular diseases and 22% had diabetes. The Same study in Wuhan revealed that out of 1099 patients with confirmed COVID-19, of whom the quarter had hypertension, 16�2% had diabetes mellitus [13]. In another study in the same place in China, out of 140 patients admitted to the hospital with COVID-19, 30% had hypertension, and 12% had diabetes [9]. Patients with NCDs are a highly affected group during the ongoing COVID -19 epidemic due to loss of jobs and wages coupled with disruptions in their usual sources of drug access. Moreover, non-adherent patients having NCDs have a manifold higher risk of complications resulting from uncontrolled disease [14]. NCD patients may, therefore, continue to be at per- sistent risk of COVID -19 while attending Primary Health Care (PHC)/Comprehensive Health Care (CHC) for meeting their health requirements. The failure to address and sufficiently resolve the barriers in attaining acceptable levels of care and management of patients having NCDs at the time of the COVID -19 pandemic represents a grave public health concern [15]. The survey report of WHO indicates that, due to the COVID -19 pandemic among 155 countries, 120 countries reported that NCD services are disrupted. The main factors related to service disruption are: a decrease in patient volume due to cancellation of elective care, closure of population-level screening programs, government or public transport lock-downs hinder- ing access to the health facilities, NCD related clinical staff deployed to provide COVID-19 relief, and closure of outpatient NCD services as per government directive. This disruption of routine health services and medical supplies risks increasing morbidity, disability, and avoid- able mortality over time in NCD patients [16]. Studies conducted regarding COVID-19 with NCDs in Ethiopia are very limited. There- fore, this study aimed to assess the knowledge and preventive practices towards COVID-19 among people with hypertension and diabetes mellitus in the Gamo, Gofa, and South Omo zones of Southern Ethiopia. Methods and materials Study setting The study was conducted in three Zones of Southern Ethiopia namely; Gamo, Gofa, and South Omo Zone. Arba Minch, the capital town of Gamo zone, is located 505 km south of Addis Ababa, the capital city of Ethiopia, and 275 km Southwest of Hawassa, capital of the South region. The total population of the town based on the 2007 census is 112,724. Of which the total number of women comprises 56,908. Sawla is a town of the newly established Gofa Zone, which is 500 km south of Addis Ababa and it has a population of 60,000. South Omo is a Zone of pastoralists in the region. Jinka is the capital with a total population of 37,000 [17]. [Fig 1]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 3 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 Fig 1. Map of the study area (Gamo, Gofa, and South Omo zones), Southern Ethiopia. https://doi.org/10.1371/journal.pone.0262642.g001 Study design and period A community-based cross-sectional study was conducted from 10th -17th July 2020 to assess the level of knowledge and preventive practice towards COVID-19 and assess predictors among patients with hypertension or diabetes mellitus. Population and sample Before sampling, chronic patients’ follow-up data records of each of the three zonal hospitals (Arba Minch, Sawla, and Jinka) were reviewed thoroughly for different types of non-commu- nicable diseases. However, the number of other NCD cases was insignificant to sampling. Therefore, we made the sampling frame only for patients with hypertension or diabetes PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 4 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 mellitus. Hence, this study was conducted among randomly selected people with diabetes mel- litus or hypertension patients. Patients who had a serious illness and who cannot communicate during data collection were excluded from the study. Sample size & sampling procedure We calculated the sample size for this study using a single population proportion formula by using a proportion of knowledgeable visitors about COVID-19 (72%) from a study conducted in Jimma University Medical College, Ethiopia [18]. Ninety-five percent certainty and 5% margin of error between populations and samples with a non-response rate of 5% and design effect of two. Therefore, the total calculated sample size was 704 hypertensive or diabetic patients. In Gamo Zone, there were four town administrations, among which Arba Minch town was selected randomly using the lottery method. In Gofa Zone, there were two town administra- tions and Sawla town was randomly selected. From the South Omo Zone, Jinka town was selected in the same manner. Data about the study participants were collected by reviewing hospitals’ follow-up clinic databases. A list of patients who were diagnosed with diabetes mellitus or hypertension was included in the sampling frame. Then, a simple random sampling technique was used to select the study participants until the calculated sample size was achieved. Finally, using the names of the selected patients, their addresses were sought, traced, and interviewed at the home level [Fig 2]. Instrument and measurement The knowledge and preventive practices towards the COVID-19 were measured using tools adapted from WHO and other resources [19, 20]. This study used a descriptive statistic to summarize the knowledge and preventive practices of hypertensive or diabetic patients towards the coronavirus pandemic. The data were collected using a pre-tested, structured interviewer-administered questionnaire. The questionnaire included socio-demographic char- acteristics, knowledge, and preventive practices towards COVID-19. The questions assessing knowledge (14 questions) were answered on a correct/not correct basis. A correct answer was assigned 1 point and an incorrect/unknown answer was assigned 0 points. The total knowl- edge score ranged from 0 to 14. Participants’ overall knowledge was categorized. Therefore, we used the cut-off point as ’poor’ if the score was less than 60% (< 8 of 14 points) and ‘good’ if the score was above 60% (> 8 of 14 points). Similarly, the questions assessing practice (9 ques- tions) were answered ‘yes’ or ’no’, the correct answer was assigned 1 point and an incorrect answer was assigned 0 points. The overall preventive practice score was categorized using the same parameter. If the overall score for preventive practice is less than 60% (<7 of 9 points) is ’poor’ and ’good’ if the score is more than 60% (> 7 of 9 points). The preventive practice com- ponent consists of questions about: frequently washed hands with water and soap, stopped shaking hands while giving a greeting, avoided proximity including while greeting (within 2 meters), have not been gone to a crowded place, using a face mask when leaving home, avoided touching eye, nose, mouth before washing hands, used cover/elbow during coughing/sneezing, used sanitizer (alcohol-rubbing, no contact with surfaces), and have stayed at home. Data collection, management, and analysis The data were collected using a structured, standardized, pre-tested interviewer-administered questionnaire. The questionnaire was developed in English and then translated into Amharic (the local language) then back-translated to the English language for its consistency by two PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 5 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 Fig 2. Schematic presentation of sampling procedure. https://doi.org/10.1371/journal.pone.0262642.g002 different language experts who speak both English and Amharic fluently. Pre-testing of the questionnaire was done on 5% of the calculated sample among hypertensive or diabetic patients who were not being included in the study. To ensure reliable data collection and attain standardization, the reliability of the knowledge and practice questions were checked. Cron- bach’s alpha was computed and its value was 0.76 and 0.83 for knowledge and preventive prac- tices, respectively. Twelve nurses (who had experience in data collection) were recruited and given training on data collection procedures. In addition, twelve home guides, who know the participants’ homes in each kebele (village), were used along with data collectors. Three Public Health experts were employed to supervise the daily data collection process. During the data collection process, data collectors and home guides were wearing medical masks and used sanitizer to rub their hands. To be safe for both the data collector and the respondent, a distance of 2 meters in between was kept. Data were entered into Epi Info ver- sion 7.00 software and then exported to SPSS version 25 statistical package for analysis. Descriptive statistics were done and summarized by the frequencies and proportions for cate- gorical predictors. The outcome variables were dichotomized as 0 = no and 1 = yes. Bivariate analysis was carried out to see the crude effect of each independent variable on the outcome variable. Asso- ciations with a p-value <0.05 were considered statistically significant. These variables with a p- PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 6 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 value < 0.25 in the bivariate analysis were candidates for multivariate logistic regression analysis. To control confounding effects and identify the independent predictors, multivariate logis- tic regression analysis was done using a backward stepwise variable selection method (back- ward LR). Hosmer and Lemeshow’s goodness of fitness test was used to check for model fitness. Multicollinearity among independent variables was checked using the Variance Infla- tion Factor (VIF>10). Adjusted odds ratio with its 95% confidence interval was used to iden- tify factors independently associated with the outcome variable and p-values < 0.05 were considered for statistical significance. Ethical approval and considerations Ethical approval was obtained from Arba Minch College of Health Sciences Institutional Review Board (IRB). Support letters were obtained from the three zonal health departments (Gamo, South Omo, and Gofa zones), town administrations, and health offices. The purpose of the study was explained to the study participants. Oral and written consent was secured before data collection. Confidentiality of the information was also ensured. Operational definition Poverty line. Is a level of personal or family income below which one is classified as poor according to governmental standards. The cut-off point for income rate is categorized based on the indicator of the proportion of the population below the international poverty line, which is defined as the percentage of the population living on less than $1.90 a day at interna- tional prices. The ’international poverty line ’ is currently set at $1.90 a day at international prices. We computed the income variable by multiplying $1.90 by the current exchange rate (Ethiopian Birr currency). Therefore, the cut-off point is 2500 ETB per month; meaning <2500 ETB (below poverty line) and � 2500 ETB (above poverty line). Results Socio-demographic characteristics of the respondents A total of 678 respondents willingly participated in this study, yielding a response rate of 96.3%. The majority of the respondents, 256 (37.8%) and 282 (41.6%) were within the age range of 31–50 and 51–64 years, respectively. The mean age was 54 ± 11.25 SD years. Among the total respondents, 436 (64.3%) were males and 513 (75.7%) were married. Orthodox Chris- tianity was the most frequent religion 463 (68.3%), followed by protestants 141 (20.8%). Three hundred and twenty-four respondents (47.8%) had a monthly family income of less than 2500 ETB with a median income of 2,900 ETB. Concerning educational status, two hundred and twenty-one (32.6%) achieved secondary school and above. Two hundred and thirteen (31.4%) respondents were government employees, and 48 (7.1%) were housewives. Three hundred and ninety respondents (57.5%) were from Arba Minch town [Table 1]. Knowledge of COVID-19 among patients with hypertension or diabetes mellitus The Multidimensional knowledge (MDK) analysis of COVID-19 revealed that 427(63%) of respondents had good knowledge of COVID-19 [Fig 3]. From the specific MDK questions, 412 (60.8%) of the respondents reported correctly that the main clinical symptoms of COVID-19 are fever, fatigue, dry cough, and myalgia. Regard- ing the knowledge of high risk about COVID-19, 468 (69%) patients correctly responded PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 7 / 19 PLOS ONETable 1. Socio-demographic characteristics of respondents, Southern Ethiopia, July 2020. Variables Age Sex Religion Educational level Residence Category 18–30 31–50 51–64 65–80 80+ Male Female Orthodox Protestant Muslim Others Cannot read and write Can read and write Primary school complete Secondary school complete Certificate and above Arba Minch Town Sawla Town Jinka Town Occupation Government Employee Pensioner Merchant NGO/Private Employee Housewife Others < 2500 � 2500 Income https://doi.org/10.1371/journal.pone.0262642.t001 Non-communicable diseases (NCDs) and vulnerability to COVID-19 Frequency(n) Percent (%) 11 256 282 127 2 436 242 463 141 60 14 101 63 90 203 221 390 171 117 213 119 118 82 48 98 324 354 1.6 37.8 41.6 18.7 0.3 64.3 35.7 68.3 20.8 8.8 2.1 14.9 9.3 13.3 29.9 32.6 57.5 25.2 17.3 31.4 17.6 17.4 12.1 7.1 14.5 47.8 52.2 Fig 3. Multidimensional knowledge (MDK) status of COVID-19 among patients with hypertension or diabetes mellitus, Southern Ethiopia, July 2020. https://doi.org/10.1371/journal.pone.0262642.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 8 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 whether all persons with COVID-19 will develop severe cases and those who are elderly, have chronic illnesses & are obese are more likely to develop severe cases. Ninety-five percent of the study subjects knew that the COVID-19 virus spreads via respiratory droplets of infected indi- viduals. For questions assessing knowledge about the mode of transmission and infectiousness, only 110 (16.2%) of patients correctly responded to the question that states whether people with the COVID-19 virus who do not have a fever can infect the other. From the components of knowledge about the ways of prevention, 640 (94.4%) responded that individuals should not go to crowded places or avoid taking public transports [Table 2]. The respondents’ preventive practices towards COVID-19 The majority, 499 (73.6%) respondents had poor practice towards prevention of COVID-19 [Fig 4]. Concerning the specific preventive practices, 379 (54.9%) were not frequently washing their hands,240 (35.4%) were not using sanitizer or alcohol rubs. Two hundred and forty-five (36.1%) respondents were still not using face masks [Table 3]. Awareness of vulnerability towards COVID-19 among patients with hypertension or diabetes mellitus Among the total respondents, 666 (98.2%) heard about the COVID-19 pandemic. Concerning vulnerability, 595 (87.8%) knew that people with non-communicable diseases are vulnerable Table 2. Knowledge of COVID-19 among people with hypertension or diabetes mellitus, Southern Ethiopia, July 2020. Knowledge variables Knowledge of symptoms Frequency Correct Not correct No. % No % The main clinical symptoms of COVID-19 are fever, fatigue, dry cough, and myalgia 412 60.8 266 39.2 Unlike the common cold, stuffy nose, runny nose, and sneezing are less common in persons infected with the COVID-19 virus 633 93.4 45 6.6 Knowledge of high risk Not all persons with COVID-2019 will develop severe cases. Only those who are elderly, have chronic illnesses & are obese are more likely to develop severe cases 468 69 210 31 There currently is no effective cure for COVID-2019, but early symptomatic and supportive treatment can help most patients recover from the infection 653 96.3 25 3.7 Knowledge about the mode of transmissions and infectiousness The COVID-19 virus spreads via respiratory droplets of infected individuals Eating or contacting wild animals would result in infection by the COVID-19 virus 646 95.3 609 89.8 32 69 4.7 10.2 Persons with COVID-19 cannot infect the virus to others when a fever is not present 568 83.8 110 16.2 Proper washing hand with soap and water is one method of preventing COVID-19 370 54.6 308 45.4 Knowledge about ways of prevention One way of prevention of COVID 19 is not touching the eye, nose by unwashed hand To prevent the infection by COVID-19, individuals should avoid going to crowded places such as train stations and avoid taking public transport 624 92 640 94.4 54 38 8 5.6 Ordinary residents can wear general medical masks to prevent the infection by the COVID- 19 virus 624 92 54 8 People who have contact with someone infected with the COVID-19 virus should immediately be isolated to a proper place Isolation and treatment of people who are infected with the COVID-19 virus are effective ways to reduce the spread of the virus Children and young adults don’t need to take measures to prevent the infection by the COVID-19 virus. 613 90.4 65 9.6 77 11.4 601 88.6 518 76.4 160 13.6 https://doi.org/10.1371/journal.pone.0262642.t002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 9 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 Fig 4. Status of preventive practices towards COVID-19 among patients with hypertension or diabetes mellitus, Southern Ethiopia, July 2020. https://doi.org/10.1371/journal.pone.0262642.g004 to COVID-19 infection. Health professionals were sources of information for the majority, 507 (55.2%) of respondents, followed by media 202 (22%). Five hundred and twenty-six 526 (77.6%) were diabetes patients. The majority, 648 (95.6%), had medical follow-up before the COVID-19 pandemic occurrence. Out of these, 509 (75.1%) had visited health facilities every month. However, during the COVID-19 pandemic, 192 (28.3%) had no follow-up at the health facilities. Out of those who discontinued follow-up, 144 (21.2%) were due to fear of acquiring COVID-19 infection from health facilities [Table 4]. Adherence to drug and control measures among patients with hypertension or diabetes Out of the 678 respondents, 550 (81.1%) had taken their drugs regularly. However, 128 (18.9%) respondents discontinued taking their drugs. The majority, 510 (75.2%), purchased their drugs once for three months. About 492 (72.6%) accessed it from public health facilities. Fifty-two (7.7%) conducted physical exercise as an alternative to the drug, and more than a half, 351(51.8%) understood taking of drugs properly and lifestyle modification as a means of controlling those specific diseases [Table 5]. Table 3. Preventive practices towards COVID-19 among patients with hypertension or diabetes, Southern Ethio- pia, July 2020. Practice variables Yes No Frequently wash hands with water and soap Stopped shaking hands while giving greeting Avoided proximity, including while greeting (within 2 m) have not gone to a crowded place Used face mask when leaving home Avoid touching eyes, nose, and mouth before washing hands Used cover /elbow during coughing/sneezing Others (alcohol-rubbing, no contact with surfaces) Have stayed at home https://doi.org/10.1371/journal.pone.0262642.t003 No. 299 582 624 505 433 557 632 468 440 % 44.1 85.8 92 74.5 63.9 82.2 93.2 64.6 64.9 No 379 96 54 173 245 121 46 240 238 % 54.9 14.2 8 25.5 36.1 17.8 6.8 35.4 35.1 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 10 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 Table 4. Vulnerability and NCD follow up related characteristics among patients with hypertension or diabetes mellitus, Southern Ethiopia, July 2020. Category Frequency Percent Variables Heard about COVID-19 Follow up before COVID-19 pandemic Yes No Yes No Reason not to follow up before COVID-19 pandemic Lack of money Frequency of follow up before COVID-19 pandemic Every month I feel better Follow up after COVID-19 pandemic Every three month Every two month Reason not to follow up after COVID-19 pandemic When feeling sick Yes No Fear of acquiring COVID-19 I feel better Lack of money 666 12 648 30 16 14 509 53 38 48 486 192 144 39 9 98.2 1.8 95.6 4.4 2.4 2.1 75.1 7.8 5.6 7.1 71.7 28.3 21.2 5.8 1.3 https://doi.org/10.1371/journal.pone.0262642.t004 Predictors of knowledge towards COVID-19 among study participants Both bivariate and multivariate logistic regression analyses were done. In the bivariate analysis, the association of socio-demographic and other variables with the knowledge status of the respondents were checked; Family monthly income and medical follow-up after the COVID- 19 pandemic were significantly associated at p <0.05. These variables which were significant in the bivariate analysis were entered into the multivariate logistic regression model. There- fore, family monthly income (AOR = 1.42; 95% CI: 1.04, 1.94) and medical follow-up after COVID-19 (AOR = 1.44; 95% CI: 1.02, 2.04) were found to be significant predictors of knowl- edge towards COVID-19 in this study. The odds of good knowledge among the study partici- pants who had family monthly income above or equal to 2500 ETB had about 1.42 times Table 5. Adherence to drug and control measures among patients with hypertension or diabetes mellitus, Southern Ethiopia, July 2020. Take drug regularly Variables Where do you get the drug from Reason for discontinuation of drugs Category Yes No Public health facilities Private health facilities Abscess of proper storage Lack of money I feel better Shifted to traditional medicines What measures you have taken alternative to drug Physical exercise Traditional medicines Spiritual remedy Nothing How do you control hypertension/diabetes Both proper taking of drugs & Lifestyle modifications https://doi.org/10.1371/journal.pone.0262642.t005 Proper taking of drugs alone Lifestyle modifications alone Frequency Percent 550 128 492 58 218 56 52 12 52 25 29 22 351 170 157 81.1 18.9 72.6 8.6 31.5 8.1 7.5 1.7 7.7 4.3 3.7 3.2 51.8 25.1 23.2 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 11 / 19 PLOS ONETable 6. Predictors of knowledge towards COVID-19 among patients with hypertension or diabetes mellitus, Southern Ethiopia, July 2020. Variables knowledge status COR (95%CI) AOR (95%CI) Good knowledge Poor knowledge Non-communicable diseases (NCDs) and vulnerability to COVID-19 Religion Orthodox Protestant Muslim Other Sex Male Female Marital status Single Married Family monthly income <2500 �2500 Educational status No formal education Can read and write Primary education Secondary education Certificate and above Follow up after COVID-19 No Yes �� significant at p<0.01 � significant at P<0.05. https://doi.org/10.1371/journal.pone.0262642.t006 300 (64.8%) 99 (70.2%) 43 (71.7%) 8 (57.1%) 280(64.2%) 170 (70.2%) 111 (67.3%) 339 (66.1%) 202 (62.3%) 248 (70.1%) 75 (74.3%) 42 (66.7%) 54 (60%) 137 (67.5%) 142 (64.3%) 136 (61%) 314 (69%) 163 (35.2%) 42 (29.8%) 17 (28.3%) 6 (42.9%) 156 (35.8%) 72 (29.8%) 54 (32.7%) 174 (33.9%) 122 (37.7%) 106 (29.9%) 26 (25.7%) 21 (33.7%) 36 (40%) 66 (32.5%) 79 (35.7%) 87(39%) 141 (31%) 1 0.78 (0.51,1.17) 0.72 (0.40, 1.31) 1.38 (0.47, 4.04) 1 0.82 (0.67,1.45) 0.65 (0.51, 1.47) 1.45 (0.53, 3.04) 1 1 1.31 (0.93,1.84) 0.76 (0.54,1.06) 1 1 1.05 (0.72, 1.53) 1.19 (0.81,1.71) 1 1.41 (1.02, 1.94) � 1 1.42 (1.04, 1.94) �� 0.62(0.36,1.05) 0.89 (0.49, 1.62) 1.19 (0.72,1.98) 0.86 (0.57,1.29) 1 0.58 (0.45, 1.35) 0.72 (0.81, 1.52) 1.21 (0.63,1.81) 0.8 (0.67,1.34) 1 1 1.43 (1.02, 1.99) � 1 1.44 (1.02, 2.04) �� higher knowledge about COVID-19 than their counterparts. Respondents who had medical follow-up after COVID-19 had 1.44 times higher knowledge as compared to those who had no medical follow-up [Table 6]. Predictors of preventive practices towards COVID-19 among patients with hypertension or diabetes mellitus The predictors of preventive practices towards COVID-19 were assessed. Both bivariate and multivariate logistic regression analysis were done and those variables with P-value < 0.25 in the bivariate analysis were candidates for the final model. Therefore, the selected variables were entered into the multivariate logistic regression analysis using the ’backward stepwise’ method of variable selection. In multivariate logistic regression, only variables such as knowl- edge status of the respondent (AOR = 1.47; 95% CI: 1.03, 2.12), follow up after COVID-19 (AOR = 2.21;95% CI:1.39, 3.52), and taking drugs regularly (AOR = 1.89; 95% CI:1.13, 3.17) were statistically significantly associated with preventive practices of COVID-19. The study participants who had good knowledge about COVID-19 were about 50% higher to have a good preventive practice as compared to those who had low knowledge. The likelihood of exercising preventive practices of COVID-19 was about two times higher among patients with hypertension/diabetes who had regular follow-up after COVID-19 than their counterparts [Table 7]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 12 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 Table 7. Predictors of preventive practices towards COVID-19 among patients with hypertension or diabetes mellitus, Southern Ethiopia, July 2020. Variables preventive practice status COR (95%CI) AOR (95%CI) Good practice Poor practice Religion Sex Orthodox Protestant Muslim Other Male Female Marital status Single Married Monthly income <2500 �2500 Educational status Cannot read and write Can read and write Primary education Secondary education Certificate and above Follow up after COVID-19 No Yes Knowledge status of the respondent Good knowledge Poor knowledge take drugs regularly No Yes � � Significant at P <0.01 � Significant at P <0.05. https://doi.org/10.1371/journal.pone.0262642.t007 127(27.4%) 31(22.0%) 15(25.0%) 6 (42.9%) 116(26.6%) 63 (26.0%) 36(21.8%) 143 (27.9%) 86 (25.8%) 93 (27.0%) 27 (26.7%) 17 (27.0%) 18 (20.0%) 61 (30.0%) 56 (25.3%) 84 (43.8%) 95 (19.5%) 113 (25.1%) 66 (28.9%) 61 (47.7%) 118 (21.8%) 336 (72.6.%) 110 (78.0%) 45 (75.0%) 8 (57.1%) 1 1.34 (0.85,2.09) 1. 13 (0.61, 2.10) 0.50 (0.17, 1.48) 1 1.52 (0.73,1.39) 1.34 (0.54, 2.46) 0.67 (0.42, 1.68) 320 (73.4%) 179 (74.0%) 129 (72.1%) 370 (72.1%) 247 (74.2%) 252 (73.0%) 74 (73.3%) 46 (73.0%) 72 (80.0%) 142 (70.0%) 162 (74.7%) 108 (56.3%) 391 (80.5%) 337 (74.9%) 162 (71.1%) 67 (52.3%) 432 (78.5%) 1 1 1.03 (0.72,1.47) 1.25 (0.54, 1.83) 1 1 0.72 (0.47, 1.09) 0.63 (0.25, 1.42) 1 1 0.81 (0.58, 1.15) 0.47 (0.76, 1.46) 0.93 (0.54, 1.58) 0.91 (0.48, 1.73) 1.35 (0.74,2.47) 0.79 (0.51,1.21) 1 0.58 (0.71, 1.43) 0.74 (0.61, 1.41) 1.25 (0.52, 2.51) 0.53 (0.43, 1.31) 1 1 2.02 (1.41, 2.89) � 1 2.21 (1.39, 3.52) �� 1.55 (1.09, 2.19) � 1.47 (1.03, 2.12) �� 1 1 1 3.3 (2.22, 4.98) � 1 1.89(1.13, 3.17) �� Discussion Pre-existing diabetes is a risk factor for poor outcomes and death after COVID-19. The associ- ation between COVID-19 and hyperglycemia in elderly patients with DM is likely to reflect metabolic inflammation and exaggerated cytokine release [21]. Recent data suggest that SARS-CoV2 infection can lead to a deterioration in glycemic control, involving both profound insulin resistance, and impaired insulin secretion, together with leading to diabetic ketoacido- sis, DKA [22, 23]. Moreover, the impairment at different levels of the innate and adaptive immune response is likely to be involved in the poorer ability to fight infection in these patients with diabetes and contribute to severe forms of morbidity and mortality [24–26]. Hypertension is also one of the risk factors for disease severity and death from SARS-Cov2 infection. The potential biological mechanism is that hypertensive patients can be more prone to Renin-Angiotensin System (RAS) imbalance, which in turn lead to vasoconstriction/ PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 13 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 inflammation due to unopposed Ang II effect [27]. This process is precipitated by increased Dipeptidyl Peptidase4 (DPP4) vascular activity/expression and by chronic low-grade inflam- mation [28]. The dysregulated response, allied with diminished physiologic cardiovascular reserve, induced by hypertension—arterial stiffening, left ventricular hypertrophy, and endo- thelial dysfunction creates the perfect milieu for both COVID-19 related tissue injury and worsening of cardiac, renal, and vascular function. This abnormal condition predisposes hypertensive patients to more complicated clinical outcomes [29]. Therefore, this study assessed the knowledge and preventive practices towards COVID-19 among patients with hypertension or diabetes mellitus in three large zones of Southern Ethio- pia. The study revealed that 63% of patients with hypertension or diabetes mellitus had good knowledge about COVID-19 (Based on the knowledge score of the participants). This finding is consistent with a study conducted in Northwest, Ethiopia [30], in which the level of good knowledge towards COVID-19 among NCD patients was 66%. However, the current study finding is lower than studies conducted in China and Iran [20, 31], in which the overall achieved knowledge towards COVID-19 in both studies was 90%. The difference could be due to the socio-economic and demographic differences of these countries and ours. This finding is higher than a study conducted in Thailand in which 73.4% of the study par- ticipants had poor knowledge of the pandemic [32]. This discrepancy may be, attributed to the time of study conducted. The current study was conducted when the number of cases alarm- ingly increased and the Federal Ministry of Health of Ethiopia has been engaged intensively in increasing the awareness of the general population. However, the study in Thailand was con- ducted during the time of early outbreak when the number of cases was very low. This finding is lower than the overall knowledge towards COVID-19 among health care workers (HCWs) in China and Pakistan. The knowledge of HCWs towards COVID 19 pre- vention in both studies was 89 and 92.3% respectively [33, 34]. This discrepancy could be due to the difference in the study populations. The previous two studies were conducted among health care workers who have exposure to the information about the COVID-19 pandemic whereas the current study was conducted among NCD patients from the general population. From specific knowledge assessing questions (knowledge about the mode of transmissions and infectiousness), 95% of the study subjects knew that the COVID-19 virus spreads via respi- ratory droplets of infected individuals. Concerning knowledge about the risk of vulnerability, 69% of respondents knew that not all persons with COVID-19 will develop severe cases, but only those who are elderly, have chronic illnesses & are obese are more likely to have severe forms of COVID-19 and subsequent mortality. This finding is in line with a study conducted in Jimma, Ethiopia [18]. The Ethiopian government has been working intensively on awareness creation towards this pandemic using different media, including social media, since March 2020, after a few new cases of COVID-19 reported in the capital city of the country. Henceforth, 98% of the study population heard about COVID-19 during the survey. This finding is similar to studies conducted in Ethiopia, a survey conducted in three countries in Africa, and Pakistan. In those studies, the number of study participants who heard about the disease accounts for 91.5, 94, and 90%, respectively [35–37]. In this study, only 26.4% of the respondents had good practice towards the prevention of COVID-19. Concerning the specific preventive practices, 55.1% of the respondents frequently washed their hands with soap and water, 64.6% used sanitizer or alcohol rubs. There was a sig- nificant gap between knowledge and preventive practices towards the pandemic among the study subjects. For instance, the knowledge of wearing medical masks to prevent infection was 92%, but the practice of wearing medical masks was only 63.9%. Knowledge of avoiding going to crowded places was 94.4%, whereas the practice of going to crowded places was 74.5%. This PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 14 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 finding is consistent with studies conducted in Ethiopia [18, 30]. However, this result is incon- sistent with studies conducted in India and Bangladesh, in which above 90% of the study sub- jects avoided crowded places and wore face masks when leaving home during the rapid rise period of the COVID-19 outbreak [38, 39]. This discrepancy may be due to socio-cultural and behavioral differences. Ethiopia is known for its diversity, social, cultural, religious ceremonies, and a high rate of overcrowded living conditions. Moreover, financial, cultural, and religious norms are the main barriers to preventive prac- tices of COVID-19 that affect the acceptability of public health measures [40, 41]. In this study, the high practice of going to crowded places during the pandemic is attributed to the strong cultural and religious norms. These norms are the restricting factors of preventive practices of COVID-19. In Ethiopian culture, practicing social gatherings during the occurrence of vital events is common. All rituals, including mourning, marriage, and other social and religious gatherings are long-lasting practices persisting in the era of COVID 19. This compromises the acceptance of the public health measures by the community which further challenges the effec- tive implementation of the measures against the COVID-19 pandemic. Family income significantly predicted knowledge of COVID-19. The odds of having good knowledge among the study participants who had family income of more than or equal to 2500 ETB per month was about 42 percent higher than their counterparts. This finding is sup- ported by other studies conducted in Ethiopia, China, Malaysia, & America [18, 37, 39, 42– 44], where high income was associated with good knowledge about COVID-19. The potential reason could be attributed to the economic status, which has a significant influence on the change of human health behavior. Moreover, the low economic status of people can hinder the ability to cover costs related to personal protective materials like face masks and others for daily consumption. Positive attitudes and preventive practices towards COVID-19 are modified by knowledge through successive works on increasing awareness and change of behavior [30]. The risk of infection with COVID-19 decreases by improving knowledge about the disease and patients’ preventive practices [45]. In this study, Knowledge about COVID-19 was significantly associ- ated with preventive practices of COVID-19. Respondents who had good knowledge about COVID-19 were more likely to exercise the preventive practices towards COVID-19 than their counterparts. This finding is consistent with studies conducted in Ethiopia and China [18, 30, 31]. A study conducted in Addis Ababa city administration of Ethiopia showed that 40–60% of NCD patients discontinued their regular medical follow-up during the first week after notifica- tion of the first positive COVID-19 case in the country [46]. In this study, 28.2% of patients discontinued their regular medical follow-up at health facilities and 18.9% of patients did not take their medications regularly during the COVID-19 pandemic. Fear of acquiring COVID- 19 infection was the most frequent reason for discontinuation of their medical follow-up. In the previous study, the discontinuation rate was higher than in the current study. the reason for this discrepancy is that the previous study was conducted in the earlier period of the pan- demic, the fear and frustration towards the disease was initially higher. Limitations This study is limited by its cross-section nature, whereby causal inferences may not be estab- lished; this limits the interpretation of the estimated associations. Moreover, this study should have assessed the attitudes and perceptions of patients through in-depth interviews and con- structed them as multi-dimensional measures. Thus, the findings of this study should be inter- preted within these limitations. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 15 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 Conclusions In this study, the knowledge of the respondents towards the COVID-19 pandemic was good. But the preventive practice was very low. There was a significant gap between knowledge and preventive practices towards the COVID-9 pandemic among the study subjects. Monthly income was significantly associated with knowledge of COVID-19. Knowledge of COVID-19 was found to be an independent predictor of preventive practices towards COVID-19. Health education programs aimed at mobilizing and improving COVID-19- related knowledge and practices are highly recommended for these patients with hypertension or diabetes mellitus. Concerning the preventive practices, great emphasis should be given to specific preventive practices such as frequent handwashing with soap and water, avoiding going to crowded places (social distancing), and using face masks while leaving home. These major preventive practices have to be adopted to prevent the contraction of the virus. Furthermore, for fear of acquiring the disease from health facilities, a significant number of patients with hypertension or diabe- tes interrupted their medical follow-up. Back tracing of those patients at a community level, continuing to follow up, a priority of testing, and vaccinations against COVID-19 at home level is highly recommended for patients with hypertension or diabetes mellitus. Supporting information S1 File. Dataset of NCDs and vulnerability to COVID-19: The case of adult patients with hypertension or diabetes mellitus in Gamo, Gofa, and South Omo zones in Southern Ethi- opia. (XLSX) S2 File. English and Amharic version questionnaires. (ZIP) Acknowledgments We would like to thank Arba Minch College of Health Sciences for taking the initiative and financial support to undertake this research. Our heartfelt thanks also go to the CEOs of Arba Minch, Sawla, and Jinka General Hospitals for their willingness and support during the survey. Last but not the least, we would like to pass our gratitude to the study participants for their willingness and giving time to complete the interview. Author Contributions Conceptualization: Fikre Bojola, Wondimagegn Taye, Bahiru Mulatu, Aleme Mekuria. Data curation: Habtamu Samuel, Bahiru Mulatu, Aknaw Kawza, Aleme Mekuria. Formal analysis: Fikre Bojola, Bahiru Mulatu, Aleme Mekuria. Funding acquisition: Wondimagegn Taye, Aknaw Kawza. Investigation: Fikre Bojola, Bahiru Mulatu, Aleme Mekuria. Methodology: Fikre Bojola, Bahiru Mulatu, Aleme Mekuria. Project administration: Wondimagegn Taye, Habtamu Samuel. Resources: Wondimagegn Taye. Software: Fikre Bojola, Aleme Mekuria. Supervision: Wondimagegn Taye, Habtamu Samuel, Aknaw Kawza, Aleme Mekuria. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 16 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 Validation: Fikre Bojola, Habtamu Samuel, Bahiru Mulatu, Aleme Mekuria. Visualization: Wondimagegn Taye, Habtamu Samuel, Bahiru Mulatu, Aknaw Kawza. Writing – original draft: Aleme Mekuria. Writing – review & editing: Fikre Bojola, Bahiru Mulatu, Aleme Mekuria. References 1. Kilpi F, Webber L, Musaigner A, Aitsi-Selmi A, Marsh T, Rtveladze K, et al. alarming predictions for obe- sity and non-communicable diseases in the Middle East. Public health nutrition. 2014; 17(5):1078–15 https://doi.org/10.1017/S1368980013000840 PMID: 23642403 2. Lozano R, Naghavi M, Foreman K, et al. Global and regional mortality from 235 causes of death for 20 age groups in 1990 and 2010: a systematic analysis for the Global Burden of Disease Study 2010. Lan- cet 2012; 380(9859): 2095–128. https://doi.org/10.1016/S0140-6736(12)61728-0 PMID: 23245604 3. Rahim HFA, Sibai A, Khader Y, Hwalla N, Fadhil I, Al Siyabi H, et al. Non- Communicable diseases in the Arab world. The Lancet. 2014; 383(9914):356–67. https://doi.org/10.1016/S0140-6736(13)62383-1 PMID: 24452044 4. Moran AE, Tzong KY, Forouzanfar MH, Roth GA, Mensah GA, Ezzati M, et al. Variations in ischemic heart disease burden by age, country, and income: The Global Burden of Diseases, Injuries, and Risk Factors 2010 study. Global heart. 2014; 9(1):91–9. https://doi.org/10.1016/j.gheart.2013.12.007 PMID: 24977114 5. WHO. Non-Communicable diseases country profiles 2014. Ethiopia: WHO; 2014. 6. World Health Organization. Estimating mortality from COVID-19: scientific brief, 4th August 2020. Geneva: World Health Organization; 2020. 7. Shigute Z, Mebratie AD, Alemu G, et al. Containing the spread of COVID-19 in Ethiopia. J Global Health. 2020; 10(1):010369. https://doi.org/10.7189/jogh.10.010369 PMID: 32566160 8. The Centre for Evidence-Based Medicine develops, promotes, and disseminates better evidence for healthcare (CEBM). Global Covid-19 Case Fatality Rates. Available from: https://www.cebm.net/global- covid-19-case-fatality-rates/. 9. Guan W-j, Ni Z-y, Hu Y, Liang W-h, Ou C-q, He J-x, et al. Clinical characteristics of coronavirus disease 2019 in China. New England Journal of medicine. 2020; 382(18):1708–20. 10. Yang J, Zheng Y, Gou X, Pu K, Chen Z, Guo Q, et al. Prevalence of comorbidities and its effects in patients infected with SARS-CoV-2: a systematic review and meta-analysis. International Journal of Infectious Diseases.2020; 94: 91–95 https://doi.org/10.1016/j.ijid.2020.03.017 PMID: 32173574 11. Epidemiology Working Group for NCIP Epidemic Response, Chinese Center for Disease Control and Prevention. [The epidemiological characteristics of an outbreak of 2019 novel coronavirus diseases (COVID-19) in China]. Zhonghua Liu Xing Bing Xue Za Zhi. 2020 Feb 10; 41(2):145–151. Chinese. https://doi.org/10.3760/cma.j.issn.0254-6450.2020.02.003 PMID: 32064853. 12. Richardson S, Hirsch JS, Narasimhan M, et al. Presenting Characteristics, Comorbidities, and Out- comes Among 5700 Patients Hospitalized With COVID-19 in the New York City Area. JAMA. 2020; 323 (20):2052–2059. https://doi.org/10.1001/jama.2020.6775 PMID: 32320003 13. Yang Xiaobo, et al. "Clinical Course and Outcomes of Critically Ill Patients With SARS-CoV-2 Pneumo- nia in Wuhan, China: a Single-centered, Retrospective, Observational Study." The Lancet. Respiratory Medicine, vol. 8, no. 5, 2020, pp. 475–481. https://doi.org/10.1016/S2213-2600(20)30079-5 PMID: 32105632 14. Zhang JJ, Dong X, Cao YY, Yuan YD, Yang YB, Yan YQ, et al. Clinical characteristics of 140 patients infected with SARS-CoV-2 in Wuhan, China. Allergy. 2020 Jul; 75(7):1730–1741. https://doi.org/10. 1111/all.14238 Epub 2020. Feb 27. PMID: 32077115 15. Basu S. Non-communicable disease management in vulnerable patients during Covid-19. Indian J Med Ethics. 2020 Apr-Jun; V (2):103–105. https://doi.org/10.20529/IJME.2020.041 PMID: 32393447. 16. WHO. Rapid assessment of service delivery for non-communicable disease (NCDs) during the COVID- 19 pandemic 18th May 2020. 17. Central Statistical Authority of Ethiopia (CSA). Population and Housing Census of Ethiopia: Statistical Abstract. Addis Ababa; Ethiopia: CSA; 2007. p. 7–8. 18. Kebede Y, Yitayih Y, Birhanu Z, Mekonen S, Ambelu A. Knowledge, perceptions and preventive prac- tices towards COVID-19 early in the outbreak among Jimma university medical center visitors, PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 17 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 Southwest Ethiopia. PLOS ONE. 2020; 15(5): e0233744 https://doi.org/10.1371/journal.pone.0233744 PMID: 32437432 19. World Health Organization (WHO). Risk Communication and Community Engagement (RCCE) Action Plan Guidance COVID-19 Preparedness and Response. 16th March 2020. 20. Zhong BL, Luo W, Li HM, Zhang QQ, Liu XG, Li WT, et al. Knowledge, attitudes, and practices towards COVID-19 among Chinese residents during the rapid rise period of the COVID-19 outbreak: a quick online cross-sectional survey. Int J Biol Sci. 2020 Mar 15; 16(10):1745–1752. https://doi.org/10.7150/ ijbs.45221 PMID: 32226294; PMCID: PMC7098034. 21. Xue T, Li Q, Zhang Q, Lin W, Wen J, Li L, et al. Blood glucose levels in elderly subjects with type 2 dia- betes during COVID-19 outbreak: a retrospective study in a single center. medRxiv (2020). https://doi. org/10.1101/2020.03.31.20048579 22. Rayman G, Lumb A, Kennon B, Cottrell C, Nagi D, Page E, et al. Guidance on the management of Dia- betic Ketoacidosis in the exceptional circumstances of the COVID-19 pandemic. Diabetes Med (2020) 37(7):1214–16. https://doi.org/10.1111/dme.14328 PMID: 32421882 23. ADA. How COVID-19 Impacts People with Diabetes. Available at: https://www.diabetes.org/ coronavirus-covid-19/how-coronavirus-impacts-peoplewith-diabetes 24. COVID-19 Infection in People with Diabetes. Touch ENDOCRINOLOGY. Available at: https://www. touchendocrinology.com/insight/covid-19-infections-in-people-with-diabetes/ (Accessed April 29, 2020). 25. Luzi L, Radaelli MG. Influenza and obesity: its odd relationship and the lessons for COVID-19 pan- demic. Acta Diabetol (2020) 57(6):759–64. https://doi.org/10.1007/s00592-020-01522-8 PMID: 32249357 26. Stefan N, Birkenfeld AL, Schulze MB, Ludwig DS. Obesity and impaired metabolic health in patients with COVID-19. Nat Rev Endocrinol (2020) 16:341–2. https://doi.org/10.1038/s41574-020-0364-6 PMID: 32327737 27. Carson P., Giles T., Higginbotham M., Hollenberg N., Kannel W., and Siragy H. M. (2001). Angiotensin receptor blockers: evidence for preserving target organs. Clin. Cardiol. 24, 183–190. https://doi.org/10. 1002/clc.4960240303 PMID: 11288962 28. Hoffmann M., Kleine-Weber H., Schroeder S., Kruger N., Herrler T., Erichsen S., et al. (2020). SARS- CoV-2 cell entry depends on ACE2 and TMPRSS2 29. Diaz J. H. (2020). Hypothesis: angiotensin-converting enzyme inhibitors and angiotensin receptor blockers may increase the risk of severe COVID-19. J. Travel Med. 27: taaa041. https://doi.org/10. 1093/jtm/taaa041 PMID: 32186711 30. Akalu Y, Ayelign B, Molla MD. Knowledge, Attitude, and Practice Towards COVID-19 Among Chronic Disease Patients at Addis Zemen Hospital, Northwest Ethiopia. Infect Drug Resist. 2020 Jun 24; 13:1949–1960. https://doi.org/10.2147/IDR.S258736 PMID: 32612371; PMCID: PMC7322118. 31. Erfani A, Shahriari Rad, Ranjbar K, Mirahmadizadeh A & Moghadami M. Knowledge, Attitude and Prac- tice toward the Novel Coronavirus (COVID-19) Outbreak: A Population-Based Survey in Iran. Bull World Health Organ. E-pub: 30th March 2020. http://dx.doi.org/10.2471/BLT.20.256651 32. Srichan P, Apidechkul T, Tamornpark R, Yeemard F, Khunthason S, Kitchen Paiboon S, et al. Knowl- edge, Attitude and Preparedness to Respond to the 2019 Novel Coronavirus (COVID-19) Among the Bordered Population of Northern Thailand in the Early Period of the Outbreak: https://ssrn.com/ abstract=3546046 or http://dx.doi.org/10.2139/ssrn.3546046 33. Zhang M, Zhou M, Tang F, Wang Y, Nie H, Zhang L, et al. Knowledge, attitude, and practice regarding COVID-19 among healthcare workers in Henan, China. J Hosp Infect. 2020 Jun; 105(2):183–187. https://doi.org/10.1016/j.jhin.2020.04.012 Epub 2020 Apr 9. PMID: 32278701; PMCID: PMC7194961. 34. Saqlain M, Munir MM, Rehman SU, Gulzar A, Naz S, Ahmed Z, et al. Knowledge, attitude, practice and perceived barriers among healthcare workers regarding COVID-19: a cross-sectional survey from Paki- stan. J Hosp Infect. 2020 Jul; 105(3):419–423. https://doi.org/10.1016/j.jhin.2020.05.007 Epub 2020 May 11. PMID: 32437822; PMCID: PMC7211584. 35. Bekele D, Tolossa T, Tsegaye R, Teshome W. The knowledge and practice towards COVID-19 pan- demic prevention among residents of Ethiopia. An online cross-sectional study. PLoS One. 2021 Jan 28; 16(1): e0234585. https://doi.org/10.1371/journal.pone.0234585 PMID: 33507906; PMCID: PMC7843015. 36. GeoPoll (2020). A study of the knowledge and perceptions of coronavirus (COVID-19) in South Africa, Kenya, and Nigeria 37. Ikhlaq A, Bint-E-Riaz H, Bashir I, Ijaz F. Awareness and Attitude of Undergraduate Medical Students towards 2019-novel Coronavirus. Pak J Med Sci. 2020 May; 36(COVID19-S4): S32–S36. https://doi. PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 18 / 19 PLOS ONENon-communicable diseases (NCDs) and vulnerability to COVID-19 org/10.12669/pjms.36.COVID19-S4.2636 COVID-19-S4.2636. PMID: 32582311; PMCID: PMC7306955. 38. Singh A.; Ahuja R. Knowledge, Attitude, and Practice of General Public Towards COVID-19 in India: An Online cross-sectional Study. Int. J. Innov. Res. Sci. Eng. Technol. 20. https://doi.org/10.15680/ IJIRSET.2020.0906009 39. Haque T, Hossain KM, Bhuiyan MMR, et al. Knowledge, attitude and practices (KAP) towards COVID- 19 and assessment of risks of infection by SARS-CoV-2 among the Bangladeshi population: An online cross-sectional survey. Research Square; 2020. https://doi.org/10.21203/rs.3.rs-24562/v1 40. Mantwill S, Monestel-Umaña S, Schulz PJ. The Relationship between Health Literacy and Health Dis- parities: A Systematic Review. PLoS One. 2015 Dec 23; 10(12): e0145455. https://doi.org/10.1371/ journal.pone.0145455 PMID: 26698310; PMCID: PMC4689381. 41. Asemahagn MA. Factors determining the knowledge and preventive practices of healthcare workers towards COVID-19 in Amhara region, Ethiopia: a cross-sectional survey. Trop Med Health. 2020; 48 (1):1–11. 42. Mersha A, Shibiru S, Girma M, et al. Perceived barriers to the practice of preventive measures for COVID-19 pandemic among health professionals in public health facilities of the Gamo zone, southern Ethiopia: a phenomenological study. BMC Public Health. 2021; 21(1):1–10. https://doi.org/10.1186/ s12889-020-10013-y PMID: 33388037 43. Azlan AA, Hamzah MR, Sern TJ, Ayub SH, Mohamad E. Public knowledge, attitudes and practices towards COVID-19: A cross-sectional study in Malaysia. PLoS One. 2020 May 21; 15(5): e0233668. https://doi.org/10.1371/journal.pone.0233668 PMID: 32437434; PMCID: PMC7241824. 44. Wolf SM, Serper M, Opsasnick L, Conor RMO, Curtis LM. Awareness, attitudes, and actions related to COVID-19 among adults with chronic conditions at the onset of the U. S. outbreak. Ann Intern Med. 2020; 9: 1–10. https://doi.org/10.7326/M20-1239 PMID: 32271861 45. Mceachan R, Hons BA, Taylor N, Gardner P, Conner M, Conner M Meta-analysis of the reasoned action approach (RAA) to understand ing health behaviors. Ann Behav Med. 2016; 50(4):592 612. https://doi.org/10.1007/s12160-016-9798-4 PMID: 27169555 46. Ketema Bezawit, Mirgessa K. and Adamu A. Challenges and coping strategies in non-communicable disease management during COVID-19 in Addis Ababa, Ethiopia: A qualitative study, 2020 PLOS ONE \| https://doi.org/10.1371/journal.pone.0262642 January 25, 2022 19 / 19 PLOS ONE
10.1371_journal.pone.0284609	RESEARCH ARTICLE California’s Public Safety Realignment Act and prisoner mortality Christopher L. RoweID 1, Alan Hubbard2, Jennifer Ahern1 1 Division of Epidemiology, School of Public Health, University of California, Berkeley, Berkeley, California, United States of America, 2 Division of Biostatistics, School of Public Health, University of California, Berkeley, Berkeley, California, United States of America chris.lloyd.rowe@gmail.com Abstract In 2011, a historic Supreme Court decision mandated that the state of California substantially reduce its prison population to alleviate overcrowding, which was deemed so severe as to preclude the provision of adequate healthcare. To comply, California passed the Public Safety Realignment Act (Assembly Bill [AB] 109), representing the largest ever court-ordered reduction of a prison population in U.S. history. AB109 was successful in reducing the state prison population; however, although the policy was precipitated by inadequate healthcare in state prisons, no studies have examined its effects on prisoner health. As other states grap- ple with overcrowded prisons and look to California’s experience with this landmark policy, understanding how it may have impacted prisoner health is critical. We sought to evaluate the effects of AB109 on prison mortality and assess the extent to which policy-induced changes in the age distribution of prisoners may have contributed to these effects. To do so, we used prison mortality data from the Bureau of Justice Statistics and the California Deaths in Custody reporting program and prison population data from the National Corrections Reporting Program to examine changes in overall prison mortality, the age distribution of pris- oners, and age-adjusted prison mortality in California relative to other states before and after the implementation of AB109. Following AB109, California prisons experienced an increase in overall mortality relative to other states that attenuated within three years. Over the same period, California experienced a greater upward shift in the age distribution of its prisoners relative to other states, suggesting that the state’s increase in overall mortality may have been driven by this change in age distribution. Indeed, when accounting for this differential change in age distribution, mortality among California prisoners exhibited a greater reduction relative to other states in the third year after implementation. As other states seek to reduce their prison populations to address overcrowding, assessments of California’s experience with AB109 should consider this potential improvement in age-adjusted mortality. Introduction The United States (US) has both the largest prison population, with over two million current prisoners, and the highest incarceration rate in the world [1]. Although the country has recently started shifting away from more punitive and towards more rehabilitative forms of a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Rowe CL, Hubbard A, Ahern J (2023) California’s Public Safety Realignment Act and prisoner mortality. PLoS ONE 18(4): e0284609. https://doi.org/10.1371/journal.pone.0284609 Editor: Andrea Cioffi, University of Foggia: Universita degli Studi di Foggia, ITALY Received: December 12, 2022 Accepted: April 4, 2023 Published: April 28, 2023 Copyright: © 2023 Rowe et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The study’s minimal underlying dataset can be found as follows: 1. Annual counts of deaths and crude mortality rates among state prisoners and yearend prison custody populations for all 50 states for years 2001-2015 obtained from the Bureau of Justice Statistics are available as Supplementary Information files. 2. De-identified individual-level data for all deaths among California state prisoners obtained from the from the California Deaths in Custody (DIC) reporting program are available as a Supplementary Information file. 3. Individual-level state prison inmate term records for California and several other states for years 2000-2015 are PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 1 / 18 PLOS ONEavailable as a restricted use dataset from the Inter- university Consortium for Political and Social Research (ICPSR). See: United States Department Of Justice. Office Of Justice Programs. Bureau Of Justice Statistics. National Corrections Reporting Program, 2000-2015: Version 1. 2017. doi:10. 3886/ICPSR36746.V1. This data source is available from ICPSR for researchers who meet the criteria for access via https://www.icpsr.umich.edu/web/ pages/ (ICPSR 36746). Funding: This work was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD Grant #1DP2HD080350-01 awarded to JA; funder URL: https://www.nichd.nih.gov/). In addition, JA is a Chan Zuckerberg Biohub Investigator. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. California’s Public Safety Realignment Act and prisoner mortality criminal justice and recent legislation has successfully decreased incarceration rates [2, 3], a substantial population remains incarcerated in prisons and jails in the US. California has the nation’s second largest prison population. Following substantial popula- tion growth in the 1980’s and 1990’s, California’s prisons were characterized by extreme over- crowding during the 2000’s. Although California’s prisons were not unique in this regard, the circumstances surrounding its efforts to resolve the problem of overcrowding were. Two class action lawsuits filed on behalf California state prisoners alleged that inadequate mental (Cole- man v. Brown, filed 1990) and physical (Brown v. Plata, filed 2001) healthcare stemming from overcrowded conditions violated prisoners’ constitutional rights. In connection with Brown v. Plata, federal courts established a receivership with full authority over the state’s prison health- care system in 2006. Components of both lawsuits were ultimately consolidated before a single three-judge court, which in 2009 declared overcrowding to be the primary reason for health- care deficiencies and mandated the state to reduce its prison population by nearly 50,000 indi- viduals, corresponding to approximately a 30% reduction. This order was upheld by the U.S. Supreme Court in 2011 and California was required to meet the mandated reduction target within three years of the ruling [4]. To comply with the court’s decision, California passed the Public Safety Realignment Act (Assembly Bill [AB] 109) in October 2011. AB109 sought to decrease the prison population by prospectively shifting the custodial responsibility of non-violent, non-serious, and non-sexual offenders from state prisons to county jails and probation departments, which was motivated by the idea that local agencies could do a better job of rehabilitating offenders [5, 6]. In addition, by limiting the funds pro- vided to counties for these additional inmates, the policy incentivized more cost-effective alter- natives to incarceration, such as the use of day reporting centers, shorter post-release community supervision, intensive probation, and home detention with GPS monitoring, with the aim of lowering the overall rate of incarceration across both prisons and jails [5, 6]. Empirical studies and policy analyses have examined the effects of AB109 on a variety of outcomes, including state prison and county jail populations [7–10], recidivism [7, 11–16], corrections spending [7], and crime [7, 8, 17–19]. Research suggests that AB109 significantly reduced the state prison population and extent of overcrowding, though not enough to reach the court-mandated target, while increasing the county jail population [7–9, 11]. The increase in the county jail population was smaller than the decrease in the state prison population, resulting in a lower rate of incarceration overall. However, one study found that AB109 may have exacerbated racial, ethnic, and gender disparities in incarceration in California’s prison system [10]. AB109 appears to have led to modest increases in recidivism statewide, but increases were smaller for counties that prioritized reentry programs [7, 11–15]. As of 2015, the policy appears to have had no effect on rates of violent crime, but may have increased rates of property crime, particularly auto theft [7, 8, 17–19]. Despite the fact that AB109 was precipitated by inadequate healthcare in California prisons, no independent studies have examined its effect on the health of prisoners. However, monthly reports tracking a wide variety of healthcare performance indicators did document improve- ments following the implementation of AB109 [20, 21]. Indeed, in 2013, Governor Jerry Brown of California declared that prison overcrowding no longer inhibited the delivery of timely and effective healthcare to prisoners. AB109 was implemented amidst an evolving prison healthcare system in California, includ- ing documented improvements under the federal receivership and the opening of a new medi- cal complex that was constructed in 2013 in response to decisions in both the Plata and Coleman cases [20]. As such, it is not possible to isolate the effects of AB109 from those due to other concurrent changes prompted by these lawsuits. However, given that overcrowding was identified as the primary reason for California’s inadequate prison healthcare [4], it is plausible PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 2 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality that any changes in health outcomes over this period are at least partially attributable to the de-crowding effects of AB109. Furthermore, the changes initiated by AB109 were unprece- dented in state-level criminal justice reform, representing the largest ever court-ordered reduc- tion of a prison population in the U.S. Indeed, AB109 has been characterized as “the biggest criminal justice experiment ever conducted in America” [22]. At the end of 2020, 10 states exceeded at least one measure of prisoner capacity [23]. As these states continue to grapple with overcrowded prisons and look to California’s experience with AB109, it is critical that we understand how the policy may have impacted all relevant outcomes, including prisoner health. Leveraging several complementary data sources, we aimed to evaluate the effects of AB109 on California state prisoner mortality. By shifting the responsibility of low-level offenders to county jails, AB109 impacted both the number of composition of prisoners in California state prisons. Because of these dynamics, we also sought to assess the extent to which effects of AB109 on prisoner mortality might be attributable contemporaneous changes in the age distri- bution of prisoners. Materials and methods Overview To evaluate the effects of AB109 on California state prisoner mortality and assess the extent to which changes in the age distribution may have contributed to these effects, we conducted four complementary analyses. First, we evaluated the effects of AB109 on crude mortality (i.e., not accounting for temporal changes in age distribution) using the synthetic control method. Second, we examined how the age distribution of California state prisoners changed following AB109 and compared these changes to those occurring in other states over the same time- frame. Third, we compared trends in crude and age-standardized mortality among California state prisoners before and after AB109. Lastly, we compared pre- to post-AB109 changes in mortality among California state prisoners to those occurring in other states while accounting for differential changes in prisoner age distributions between states. This study was approved by the Committee for Protection of Human Subjects at the University of California, Berkeley (Protocol Number 2018-09-11405) and granted a waiver of informed consent because it only involved coded administrative data with no personal identifiers. Data sources To comprehensively answer our research question, we would ideally leverage data on the uni- verse of state prisoners in the US, including their ages and dates of incarceration, as well as if and when they died in prison. However, such comprehensive data were not obtainable. Instead, we leveraged three complementary data sources that were either publicly available or practically obtainable in order to conduct the four analyses outlined above. First, we obtained the annual counts of deaths and crude mortality rates among state pris- oners for all 50 states for years 2001–2015 from publicly available reports published by the Bureau of Justice Statistics (BJS) [24, 25]. Next, we obtained individual-level state prison inmate term records for California and sev- eral other states for years 2000–2015, including month and year of birth and prison term dates, from the National Corrections Reporting Program (NCRP) [26]. Annually, the NCRP collects offender-level administrative data on prison admissions and releases and yearend custody populations from participating jurisdictions. The number of states submitting data to NCRP has varied over time, with at least 38 states providing some amount of data since 2000. These PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 3 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality data were available from the Inter-university Consortium for Political and Social Research under restricted conditions. Lastly, we obtained de-identified individual-level data for all deaths among California state prisoners, including age of the decedent and date of death, from the California Deaths in Cus- tody (DIC) reporting program [27]. The DIC Data includes all deaths that occur in law enforcement custody. To capture deaths among California state prisoners, we retained only deaths for which the California Department of Corrections and Rehabilitation had custody of the subject immediately preceding death. We excluded decedents held in out-of-state correc- tional facilities, those who died by execution, and those released on medical parole. These data were publicly available through California Department of Justice’s Open Justice program [28]. Collectively, these data sources provide us with or allow us to calculate crude annual mor- tality rates for California and comparison states (BJS Data), annual total mortality counts for California and comparison states (BJS Data), annual age-specific prisoner person-time for Cal- ifornia and comparison states (NCRP Data), and annual age-specific mortality counts for Cali- fornia only (DIC Data). Supplemental Table 1 in S1 Appendix provides an overview as to which components of these data sources were used in each of our four analyses. Crude mortality To evaluate the effect of AB109 on crude mortality, we used the synthetic control method with the annual crude mortality rates from the BJS as the outcome variable [29, 30]. Specifically, we considered California as the single treated unit, and included other states in the donor pool from which to construct the synthetic control. The synthetic control approach uses pre-policy covariates and outcome data to identify a weighted combination of control units whose weighted pre-policy covariates and outcomes best fit those of the treated unit. For our analysis, the pre-policy period included years 2001–2010 and the post-policy period included years 2012–2014. The year 2011 was excluded because AB109 was implemented in October 2011. Years after 2014 were excluded because California implemented Proposition 47, another major criminal justice law that reduced penalties for many low-level crimes, in November 2014. The effect of AB109 on prisoner mortality is estimated by comparing the observed Cali- fornia state prisoner mortality rate to that of the synthetic control unit in the post-policy period. Full methodological details of this analysis are provided in S1 Appendix. Changes in prisoner age distributions In order to investigate the extent to which the effects of AB109 on crude mortality may be attributable to changes in California’s prisoner age distribution, we first sought to assess how the age distribution of California state prisoners changed following AB109 relative to other states over the same timeframe. If California experienced a change in age-distribution follow- ing AB109 that was extreme relative to concurrent changes in other states, this would support the plausibility that effects of AB109 on mortality could be attributable to a shift in the Califor- nia’s prisoner age-distribution. To do this, we calculated the proportion of prisoner person- time corresponding to specific age groups (under 25, 25–34, 35–44, 45–54, 55–64, 65–74, over 74) for 2010 and each year 2012–2014 for California and each comparison state. Prisoner per- son-time was derived from the NCRP individual-level term records. Here, prisoner person- time for a given year refers to the total number of days incarcerated among all individuals incarcerated in that year; for example, if two prisoners were incarcerated in 2010, one for the entire year (i.e. 365 days) and one for only 30 days, these two prisoners would represent 365 + 30 = 395 days of prisoner person-time in 2010. For each state and year, calculating the pro- portion of prisoner person-time corresponding to specific age groups involved summing up PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 4 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality the incarcerated person-time among prisoners falling within each age group and dividing by the total incarcerated person-time to obtain proportions. We then calculated the relative change in each age group proportion from 2010 to each of the three post-policy years for each state. Comparing crude and age-standardized mortality among California state prisoners Next, we sought to investigate differences between the trend in crude California state prisoner mortality and the trend in California state prisoner mortality that accounts for changes in the age-distribution of prisoners over time and, particularly, following AB109. To do so, we com- pared the annual crude mortality rates among California state prisoners from 2008–2014 to annual age-standardized mortality rates over the same period, both calculated with death counts from the DIC data and person-time measures from the individual-level NCRP data. To calculate the annual age-standardize mortality rates, California’s annual age-specific mortality rates for each year from 2008–2014 were applied to the California state prisoner age-distribu- tion from 2008, which preceded the implementation of AB109. This procedure is known as direct standardization. The age-standardized mortality rates for each year were thus calculated as: XK lAS CA;j ¼ k¼1 XK lCA;j;kptCA;2008;k ptCA;2008;k k¼1 for years j ¼ 2008; 2009; . . . ; 2014 Where λAS CA,j is the California age-standardized rate for year j; λCA,j,k is the California mortal- ity rate for year j and age group k; and ptCA,2008,k is the California incarcerated person-time for 2008 and age group k. The age-standardized mortality trend illustrates how crude mortality would have changed if the California state prisoner age-distribution were static from 2008 to 2014. We were only able to standardize California’s age-specific prisoner mortality rates in this way because we had access to annual age-specific mortality rates for California but not for any comparison states. Comparing age-standardized mortality between California and other states Lastly, we sought to evaluate the effects of AB109 on mortality net of any effect due to resultant changes in the age distribution of California state prisoners and any secular trends approxi- mated by the experience of other states. To do so, we compared California’s change in state prisoner mortality from pre- to post-policy years to that experienced by other states, while con- trolling for differential changes in age distribution between California and other states over the same period. Standard approaches to adjust for differential changes in age distribution between California and other states would require that annual age specific mortality rates among California and other states be compared directly or standardized to and summed over the same age distribution, such as California’s age distribution in a given year (i.e., direct stan- dardization). However, as previously noted, the data available for this analysis precluded the ability to calculate age-specific rates for any state but California and thus we could not evaluate trends in age-adjusted mortality across states using the synthetic control method. We did, however, have access to the annual age-specific person-time from the NCRP data and the total number of deaths from the BJS data for comparison states. To compare mortality trends between California and other states while accounting for dif- ferential changes in age distribution, this combination of data is amenable to an extension of the method of indirect age standardization, in which a set of “standard” age-specific mortality PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 5 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality rates are standardized to and summed over a target age distribution to obtain the total number or rate of deaths expected in the target population had it experienced the “standard” age-spe- cific mortality rates. We calculated a single measure for each control state and post-policy year that captured the excess (or deficit) change in mortality that would have occurred in the con- trol state had it experienced California’s age-group specific mortality rates in the pre- and rele- vant post-policy year. Specifically, we used California’s annual age-specific mortality rates as the “standard” set of rates, which we standardized to and summed over the age distribution of each control state for each corresponding year. Thus, for each control state, we obtained four quantities that corresponded to the marginal mortality rate expected in each control state had it experienced California’s age-group specific mortality rates in 2010 and each year 2012–2014. Then, for each control state and post-policy year, we calculated the difference between the change in the expected marginal mortality rate from 2010 to a post-policy year and the change in the control state’s observed marginal mortality rate over the same period. Thus, the age- adjusted excess mortality was calculated as: 0 XK EMi;j ¼ @ k¼1 XK lCA;j;kpti;j;k (cid:0) pti;j;k k¼1 XK lCA;2010;kpti;2010;k k¼1 XK i¼1 pti;2010;k 1 A (cid:0) � � CRi;j (cid:0) CRi;2010 for comparison states i ¼ 1; . . . ; n and years j ¼ 2012; 2013; 2014 Where EMi,j is the age-adjusted excess mortality for comparison state i and post-policy year j; λCA,j,k is the mortality rate for California in post-policy year j for age group k; pti,j,k is the incar- cerated person-time for comparison state i, post-policy year j, and age group k; and CRi,j is the crude mortality rate for comparison state i and post-policy year j. A positive value for this quantity suggests that the change in the marginal mortality rate associated with California’s pre- and post-policy age-specific mortality rates was greater than the change associated with the control state’s pre- and post-policy age-specific mortality rates, a negative value suggests the opposite, and a value of zero suggests that there was no difference between the two. This quantity can also be framed as a difference-in-differences estimate, for which California’s age- specific mortality rates applied to a control state’s age distribution represent the treated unit and the control states observed mortality rates represent the control unit. For each post-policy year, we present the age-adjusted excess mortality rate estimates for each control state, the median value, and exact confidence intervals. Exact confidence intervals were obtained for the closest confidence levels above and below the 95% level via inversion of the one-sample sign test. The sign test is a non-parametric statistical test that makes no distri- butional assumptions about the underlying data distribution other than that data are drawn independently from a continuous distribution. We also conducted a placebo test using 2008 as the pre-policy year and 2010 as the post-pol- icy year and present the median and exact confidence intervals as described above. These years were selected so as to skip a single “implementation” year as in the main analysis and because 2010 is the last year prior to the actual implementation of AB109. Comparison states A full summary of comparison states included in the crude and age adjusted mortality analyses and rationale for any exclusions are provided in S1 Appendix. Sensitivity analysis for differences in mortality reporting As previously described, our comparison of changes in age-standardized mortality rates between California and other states leveraged multiple complementary datasets. There were PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 6 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality small differences between the annual number of deaths among California state prisoners in the DIC and BJS data, with DIC data consistently reporting fewer deaths than the BJS data (Sup- plemental Table 3 in S1 Appendix). For the main analysis, we used the deaths present in the DIC data. To account for the possibility that DIC data are not comprehensive and that the BJS data represent the truth, we conducted a sensitivity analysis in which we added deaths to the DIC data such that the annual number of deaths matched the number in the BJS data for 2010 and each post-policy year 2012–2014. Over 1000 iterations, we randomly assigned the added deaths to different age groups (with equal probability), calculated the age-adjusted excess mor- tality rate estimates for each state and post-policy year, and calculated the median value for each post policy year. We then present the 2.5% and 97.5% quantiles of the distributions of the median values for each post-policy year. Bias analyses for concurrent policy changes Two sources of potential bias in estimating the effects of AB109 on California state prisoner mortality are the nearly concurrent introduction of the medical parole program and the subse- quent introduction of the elderly parole program in the California state prison system. The medical parole program began in January 2011, granted its first parole in June 2011, and was expanded in July 2014. This program initiated a parole hearing process that allowed grantees who are permanently medically incapacitated and who require 24-hour care to be placed in a licensed healthcare facility in the community. The elderly parole program began in October 2014 and initiated a parole hearing process by which prisoners over 60 years old who have served at least 25 years of continuous incarceration could be assessed for parole. An elderly parole hearing differs from a standard parole hearing in that the panel gives special attention to the prisoner’s age, physical condition, and long-term confinement when determining a pris- oner’s suitability for parole. These two programs may have implications for our analysis because they were introduced after the end of the pre-policy period and could have impacted the mortality rate among Cali- fornia state prisoners if those granted parole through the programs were at systematically higher or lower risk of death than those who remained incarcerated. If this were the case, any changes in mortality in the post-policy period could be at least partially due to the initiation of these parole programs. We sought to assess the impact of these parole programs on the results of our comparison of age-standardized mortality rates between California and other states. It seems reasonable to assume that medical parolees would be at greater risk of death than other age-comparable pris- oners, and thus their release from custody may have resulted in a reduction in age-specific prisoner mortality rates. However, two points are worth noting. First, medical parole differs from compassionate release, which involves a recall of sentence for California state prisoners who are terminally ill and are estimated to have less than six months to live. Although medical parolees are permanently medically incapacitated, they are not necessarily terminally ill. Sec- ond, there were never more than seven medical parole deaths during any post-policy year (Supplemental Table 4 in S1 Appendix), which represent a small fraction of the annual number of deaths among California state prisoners. Regarding the elderly parole program, we have no specific reason to expect elderly parolees to be more or less at risk of death than age-compara- ble prisoners who remained incarcerated. To assess the sensitivity of our age-standardized analysis results to the introduction of the medical parole program, we calculated age-standardized excess mortality rates as described above but included deaths among California medical parolees and person time attributed among these deceased medical parolees in the calculation of the age-specific annual California PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 7 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality mortality rates. Deaths among medical parolees were present in the DIC data and were excluded from our main analysis. In addition, we re-ran the sensitivity analysis described above to correct for differences in death counts between DIC and BJS data, but for each itera- tion we added medical parolee deaths after updating the annual DIC death counts to match those reported by BJS. Because the medical parole program began after the end of the pre-policy period, including deaths among California medical parolees had no effect on the synthetic control fit for our crude mortality analysis. Their inclusion increased California’s observed mortality rate by <2% for each post-policy year and thus we do not present this sensitivity analysis for crude mortality. To explore the potential impact of the elderly parole program on the California prison mor- tality rate, we compared the trend in the mortality rate among California state prisoners aged 60 and older before and after the start of the elderly parole program to the trend in rate among prisoners younger than 60. Specifically, we calculated mortality rates separately among prison- ers above and below age 60 for each calendar quarter in 2013 and 2014. Because the variance of quarterly mortality rates among prisoners aged 60 and older were substantially larger than those among prisoners under 60 (60 and older variance = 100,463; under 60 variance = 530), we calculated the change in quarterly mortality rate since the rate in the first quarter (Q1) of 2013 standardized by the standard deviation of quarterly rates in 2013–2014 for each group. If the beginning of the elderly parole program led to a meaningful change in the mortality rate among prisoners aged 60 and older, we might expect to observe a standardized change in the mortality rate among this group in Q4 2014 that is not evident among prisoners under 60. Results Crude mortality The composition of the synthetic California is presented in Supplemental Table 5 in S1 Appen- dix, the crude mortality rates for California and the synthetic California are presented in Fig 1, and the difference between California’s annual mortality rate and those for the synthetic Cali- fornia are presented in Fig 2. Mortality among California state prisoners increased after the implementation of AB109 in 2011 relative to those of the synthetic California, but the two rates converged by 2014. These estimates suggest that, without accounting for age distribution changes, the mortality rate among California state prisoners was higher in 2012 and 2013 than would have been expected in the absence of AB109. Results of all permutation tests and robust- ness checks are provided in S1 Appendix. It is worth noting that the 2012 crude mortality rate for California reported by the BJS is and used in this analysis likely to over-estimate the true mortality rate (in deaths per person- time). The BJS uses the year-end custody population as the denominator for calculating annual rates, and AB109 led to a large and rapid reduction in the prison population that did not stabi- lize until the end of 2012. Specifically, the yearend custody populations reported by BJS for 2011 and 2012 were 147,051 and 132,624, respectively. As a result of this reduction from year- end 2011 to yearend 2012, the yearend population will under-estimate the true incarcerated person-time more in 2012 than in other years. Changes in prisoner age distributions After the implementation of AB109, the proportion of California state prisoners in older age groups increased and those in lower age groups decreased (Fig 3). From 2010 to 2012, Califor- nia had relative increases in the proportion of prisoners in older age groups that exceeded those of most control states with reliable person-time data (53% of control states for prisoners PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 8 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality Fig 1. Annual prisoner mortality rate for California and synthetic control, 2001–2014. https://doi.org/10.1371/journal.pone.0284609.g001 Fig 2. Annual prisoner mortality rate gap between California and synthetic control, 2001–2014. https://doi.org/10.1371/journal.pone.0284609.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 9 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality Fig 3. Relative change in proportion of person time by age group for 17 states with reliable National Corrections Reporting program data, 2010 to 2012. https://doi.org/10.1371/journal.pone.0284609.g003 aged 45–54 and 100% of control states for prisoners aged 55–64, 65–74, and over 74). Simi- larly, California had relative decreases in the proportion of prisoners for some younger age groups that exceeded those of most control states (80% of control states for prisoners aged 35– 44; 100% of control states for prisoners aged 25–34; 33% of control states for prisoners aged 18–24). Similar patterns were sustained through 2013 and 2014 (Supplemental Figs 12–13 in S1 Appendix). Comparing crude and age-standardized mortality among California state prisoners Both the crude and age-standardized mortality rates among California state prisoners from the period 2008–2014 are presented in Fig 4. The increase that was observed for crude mortality rates following the implementation of AB109 is absent from the age-standardized mortality rates. Whereas the crude mortality rate increased from 247.7 deaths per 100,000 person-years in 2010 to 260.5 in 2012 and 266.0 in 2013, the age-standardized mortality rate decreased from 223.9 in 2010 to 205.1 in 2012 and 202.9 in 2013. We also present the age-specific mortality counts and rates among California state prisoners for each year from 2008 to 2014 in Supple- mental Tables 6–7 in S1 Appendix. Age-specific mortality rates declined for all but two age groups (under 25 and 45–54) from 2010 to 2012 and declined for all but one age group (under 25) from 2010 to 2013 and 2014. Comparing age-standardized mortality between California and other states The age-standardized excess mortality rate estimates for each state and post-policy year are presented in Fig 5 and Supplemental Table 8 in S1 Appendix. Among the 13 control states included in the analysis, the median excess mortality rates were 7.9 (97.8% exact confidence interval: -72.8, 37.3; 90.8% exact confidence interval: -35.5, 35.8) in 2012, -17.4 (-77.2, 43.5; PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 10 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality Fig 4. Crude and age-adjusted California prisoner mortality rates, 2008–2014. https://doi.org/10.1371/journal.pone.0284609.g004 -69.9, 14.1) in 2013, and -47.1 (-99.2, -27.5; -49.0, -29.3) in 2014. These results suggest that if each control state’s prison population had experienced California’s age-specific mortality rates from 2010 to 2014, they would have tended to have greater decreases in mortality relative to what they actually experienced. By 2014, the median state would have experienced a nearly 50 fewer deaths per 100,000 prisoners, representing a nearly 20% reduction from the median mortality rate of 259 per 100,000 among the 13 control states in that year. When applied to the pre-policy period (2008 to 2010), the median excess mortality rate was 36.4 (-3.2, 108.6; 20.2, 84.4). This result suggests that if each control state’s prison population had experienced California’s age-specific mortality rates from 2008 to 2010, they would have tended to have greater increases in mortality relative to what they actually experienced. There was no evidence that the changes in crude mortality among states included in the analysis arise from a different distribution than excluded states from 2010 to 2012 (p = 0.402), 2013 (p = 0.806), or 2014 (p = 0.943), as assessed using Kolmogorov-Smirnov tests. Sensitivity analysis for differences in mortality reporting The results of our sensitivity analysis in which we randomly corrected for differences between the annual numbers of California state prisoner deaths included in the DIC data and those reported by BJS were consistent with our main age-standardized analysis results comparing changes in California to those in other states and are presented in Supplemental Table 9 in S1 Appendix. Bias analyses for concurrent policy changes The results of our sensitivity analysis that incorporated medical parole deaths were also similar to our main age-standardized analysis results comparing changes in California to those in other states (Supplemental Table 10 in S1 Appendix). As with our main analysis, randomly correcting for differences between DIC and BJS data had little effect on these estimates (Sup- plemental Table 11 in S1 Appendix). In regards to the elderly parole program, the standardized changes over time in quarterly mortality rates among prisoners above and below 60 years of age were very similar throughout PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 11 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality Fig 5. Age-adjusted excess mortality rates, 13 states, 2012–2014. https://doi.org/10.1371/journal.pone.0284609.g005 2013 and 2014, with the exception of a spike in deaths among prisoners 60 and older in Q1 2014 (Supplemental Figure 14 in S1 Appendix). Under the assumption that the relationship between the mortality trends among those above and below 60 years of age would have contin- ued through Q4 2014 in the absence of the elderly parole program, there is no indication that the elderly parole program impacted mortality rates among prisoners aged 60 and over. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 12 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality Discussion Following the implementation of AB109, California prisons experienced an increase in crude mortality that attenuated by 2014 and appears to be attributable to a concurrent shift in the age distribution of California state prisoners. When accounting for this change in the age distribu- tion, California’s change in mortality rate appeared consistent with those of other states in the first two post-policy years but exhibited a greater reduction relative to other states in the third year after implementation. This is the first study evaluating California state prisoner mortality following the implementation of California’s historic criminal justice initiative, and our analy- sis suggests improved mortality outcomes three years after implementation. The results of our synthetic control analysis suggest that the crude mortality among Califor- nia state prisoners increased relative to what would have occurred in the absence of AB109, with the largest increase occurring in 2012. This finding is supported by our permutation test when restricting to control states with good pre-policy fits and our robustness checks. How- ever, as previously noted, the crude mortality among California state prisoners that is reported by BJS and used in this analysis over-estimates the true mortality rate in 2012, thus the magni- tude of the effect in this year is likely exaggerated. Importantly, our age-focused analyses sug- gest that this increase is attributable to a substantial shift in the age distribution of California state prisoners as opposed to other mechanisms elevating prisoners’ mortality risk. We explored this possibility in three ways. First, we found that California experienced a shift in age distribution among prisoners that was more extreme than most comparison states, particularly in regards to increasing propor- tions of older prisoners, following the implementation of AB109. AB109 provisions were implemented prospectively such that that new low-level offenders would be incarcerated in county jails instead of state prisons but existing state prisoners would not be transferred to county jails or granted early release. Although both admissions and releases declined after AB109, the reduction in California’s prison population was driven by a decline in admissions that exceeded the corresponding decline in releases. The rapid change in the age distribution of California’s prisoner population following AB109 aligns with the fact that that older prison- ers are more likely to be serving longer sentences [31], such that prisoners released in a given year are disproportionately younger than the overall prisoner population. Prior to AB109, these releases would be approximately offset by new admissions, but new admissions were reduced under AB109, a dynamic that resulted in a smaller and older prisoner population. It is well established that mortality risk increases with age both in and out of prisons [24, 32], thus it is plausible that the age-shift precipitated by AB109 could increase the crude mortality rates. Second, we found that the California state prisoner age-specific mortality rates actually declined for nearly all age groups following the implementation of AB109. This is also reflected in our finding that while the crude mortality rate increased from 2010 to 2012 and subse- quently declined through 2014, the age-standardized rate declined over the entire period. Lastly, in our comparison of changes in age-standardized mortality rates between California and other states, we found no evidence that changes in California age-specific mortality rates were systematically different from those of 13 comparison states from 2010 to 2012 or 2013, but we did find that California’s age-specific rates decreased more than those of all but one comparison state in 2014. Furthermore, we presented evidence that if each comparison state had experienced Cali- fornia’s age-specific mortality rates before and after AB109, the median state would have experi- enced nearly 50 fewer deaths per 100,000 prisoners compared to what was actually observed, or approximately 20% relative to the median 2014 mortality rate among comparison states. Taken together, our findings suggest that crude mortality increased among California state prisoners following AB109 due to an evident change the age distribution and corresponding PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 13 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality risk profile of California’s incarcerated population, but that mortality within age strata declined both relative to the year before implementation and relative to contemporaneous changes in other states. Understanding the overall (i.e. crude) mortality risk of the incarcerated population is critical for surveillance and healthcare planning; however, to understand how the de-population of California’s state prisons under AB109 may have impacted the health of individual prisoners, it is essential to account for the concurrent change in age distribution and the corresponding mortality risk. The discrepant findings between our crude and age-adjusted analyses highlight an impor- tant methodological consideration when studying trends or impacts of policies on population measures. Specifically, researchers must take special care when applying methods and inter- preting results when the internal composition of an aggregate population being studied may change in a manner that impacts the outcome of intertest. In the present scenario, a careless interpretation of the results of our synthetic control analysis might conclude that AB109 increased the mortality risk among California state prisoners, which our age-adjusted analyses suggest is not the case. Importantly, any approach effectively accounting for such dynamics requires data at a level more granular than crude population measures (e.g., individual-level data or measures stratified by relevant covariates), which might not always be available. It is important to place our findings within the broader context of the California prison healthcare system leading up to and following the implementation of AB109. As noted in the Introduction, California’s prison healthcare system has a unique and complex history dating back decades from the implementation in AB109. Most notably, from 2006 through the end of the study period, California’s healthcare system was under full authority of a federal receiver specifically tasked with improving the quality of care [20]. From 2008 to 2010, both the crude and age-standardized mortality rates among California state prisoners increased; in fact, when calculating age-standardized excess mortality rates for this period, California appeared to have a greater increase in age-specific mortality rates than comparison states. Despite these increases in the early years of the receivership, a 2015 report by the receiver summarized sub- stantial improvements to the structure, processes, and outcomes related to medical care in Cal- ifornia prisons since 2008, though it also noted extensive variability in quality of care at the institutional level [20]. Also, from 2006 to 2014, a comprehensive death review process docu- mented a reduction in the rate of medically preventable deaths [33]. In addition to these changes that spanned the pre- and post-AB109 periods, the state opened the California Health Care Facility (CHCF) in July 2013. The CHCF is a 54-building and nearly 3000-prisoner capacity medical complex for prisoners with long-term medical or acute mental-health needs. It is thus clear that the de-population of California prisons under AB109 occurred in tandem with dedicated and seemingly effective efforts to improve the quality of medical care provided to prisoners. Indeed, although the original court decision and subsequent Supreme Court opinion for Coleman/Plata vs. Brown declared overcrowding as the primary reason for health- care deficiencies, they also made clear that resolving overcrowding alone would not be suffi- cient to improve the quality of care for prisoners. The changes that spanned or occurred during the post-policy period offer potential expla- nations for why we only observed the reduction in age-adjusted prisoner mortality in 2014, the third year after the implementation of AB109. Although the CHCF opened in July 2013, the receiver halted intake in January 2014 citing inadequate staffing and supply chain issues that precluded effective provision of healthcare at the facility [34]. Following a series of improve- ments, intake at CHCF resumed in July 2014 [35]. Thus, after a challenging first 6–12 months in operation, CHCF may have impacted system-wide mortality rates by admitting and provid- ing care for prisoners with the most complex and severe medical needs in 2014. Indeed, CHCF’s average end-of-month population was 366 across all 12 months in 2013 and 1,620 in PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 14 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality 2014 [36]. In addition, the state’s death review process determined that the rate of medically preventable deaths was relatively stable from 2009 to 2013 but decreased by one-third in 2014, which aligns with our findings [33]. On the other hand, the average number of identified lapses in care per prisoner death showed no meaningful trend prior to 2010 but decreased from 1.1 in 2010 to 0.8 in 2012–2013 and to 0.5 in 2014 [33], which aligns with the timing of AB109 implementation. It is worth noting that death is the most severe health outcome and that improvements in medical care are likely to impact intermediate health outcomes before ulti- mately impacting mortality. This is particularly true in the context of chronic conditions such as diabetes or hypertension, which require ongoing care and management and are more preva- lent among incarcerated populations [37]. In addition, the decline in California’s prison popu- lation resulting from AB109 mostly occurred in the first year after implementation, stabilizing around 133,000 inmates by September 2012 [38]. Accordingly, it seems plausible that de-popu- lation under AB109, not fully realized until late 2012, could have facilitated subsequent improvements in the quality of prisoner care that would correspond to the observed lag between the implementation of AB109 and changes in mortality. Our study is novel in that it evaluates the impacts of decarceration on individuals who remain incarcerated and that it focuses on a health outcome. Prior work has noted the poten- tial for decarceration efforts to improve the health and well-being of prisoners by reducing overcrowding and decreasing prisoner-to-staff ratios [39]; indeed, AB109 was predicated on this idea [4]. However, as illustrated by other research evaluating AB109 [7, 8, 11–19], out- comes related to public safety and recidivism tend to receive more attention among efforts to understand the impacts of decarceration. Our findings that AB109 may have decreased the risk of mortality among California state prisoners provides an important complement to this other work, and allows for a more comprehensive view of the impacts of AB109 and of decar- ceration more broadly. The present study has several limitations. First, our synthetic control analysis and our age- standardized excess mortality rate measures both rely on the assumption that the mortality experience of other states can be used to approximate what would have occurred in California in the absence of AB109, which is untestable. Second, we are unable to disentangle the mortal- ity effects of de-population under AB109 and concurrent improvements in the quality of medi- cal care in California’s prisons that were prompted by the same lawsuits as AB109; however, given that these were intended as concurrent and complementary strategies for improving prisoner healthcare, we are not certain that it would be meaningful to do so. Third, only 29 and 13 control states were included in the crude and age-adjusted analyses, respectively; how- ever, the majority of states were excluded due to small population sizes and numbers of pris- oner deaths or having integrated prison and jail systems, which may limit their comparability to California, which has the nation’s second largest state prison system. In addition, among states included in the crude mortality analysis, there was no difference in the distributions of crude mortality trends between those that were included and excluded from the age-adjusted analysis. Fourth, our age-adjusted analysis combines data from multiple sources, which involved population data of variable quality in NCRP data and discordant counts of annual deaths among California state prisoners in the BJS and DIC data. We sought to address the variable quality in the NCRP data by validating year-end custody counts against NPS data and only retaining states that exhibited reasonable quality over the entire study period. We sought to address the discordant death counts in our stochastic sensitivity analysis, which had trivial effects on our estimates and no impact on our conclusions. It is important to note that broader availability of reliable data on US prison populations, including health-related measures, would mitigate the majority of these limitations and improve the ability of researchers to rigor- ously evaluate the impacts of changes to the nation’s prison systems. Indeed, a study surveyed PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 15 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality 36 prominent publicly available health datasets and found the none could be used to assess the health of incarcerated individuals, highlighting the dearth of these data [40]. Despite these limitations, we leveraged multiple data sources to examine mortality among California state prisoners following a historic and unprecedented de-populating of a state prison system. Although the crude mortality rate increased initially, our findings suggest this was due to an expected shift in age distribution of California state prisoners. When accounting for this shift in age distribution, we found that mortality decreased more in California relative to comparison states three years after the implementation of AB109. As other states seek to reduce their prison populations to address overcrowding and corresponding healthcare defi- ciencies, any assessment of California’s experience with AB109 should consider this potential benefit. Supporting information S1 Appendix. Supplementary methods, results, tables, and figures. (DOCX) S1 Dataset. (ZIP) Acknowledgments The authors acknowledge and thank Steven Raphael for his contributions to the identification of data sources, interpretation of the results, and critical review of the manuscript. Author Contributions Conceptualization: Christopher L. Rowe, Jennifer Ahern. Data curation: Christopher L. Rowe. Formal analysis: Christopher L. Rowe. Funding acquisition: Jennifer Ahern. Methodology: Christopher L. Rowe, Alan Hubbard, Jennifer Ahern. Visualization: Christopher L. Rowe. Writing – original draft: Christopher L. Rowe. Writing – review & editing: Christopher L. Rowe, Alan Hubbard, Jennifer Ahern. References 1. R W. World Prison Population List, 12th Edition. London, UK: Institute for Criminal Policy Research (ICPR) at Birkbeck, University of London; 2018. 2. Bronson J, Carson EA. Prisoners in 2017. Bureau of Justice Statistics, Office of Justice Programs, U.S. Department of Justice; 2019. 3. Schrantz D, DeBor S, Mauer M. Decarceration strategies: How 5 states achieved substantial prison population reductions. Washington, D.C.: The Sentencing Project; 2018. 4. Brown v. Plata. 2011. 5. Misczynski D. Corrections Realignment: One Year Later. San Francisco, CA: Public Policy Institute of California; 2012. 6. Lofstrom M, Bird M, Martin B. California’s Historic Corrections Reforms. San Francisco, CA: Public Pol- icy Institute of California; 2016. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 16 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality 7. 8. 9. Lofstrom M, Martin B. Public Safety Realignment: Impacts So Far. San Francisco, CA: Public Policy Instute of California; 2015. Lofstrom M, Raphael S. Realignment, Incarceration, and Crime Trends in California. San Francisco, CA: Public Policy Institute of California; 2015. Lofstrom M, Raphael S. Impact of Realignment on County Jail Populations. San Francisco, CA: Public Policy Institute of California; 2013. 10. Gottlieb A, Charles P, McLeod B, Kjellstrand J, Bonsu J. Were California’s Decarceration Efforts Smart? A Quasi-Experimental Examination of Racial, Ethnic, and Gender Disparities. Criminal Justice and Behavior. 2020; 009385482092338. https://doi.org/10.1177/0093854820923384 11. Wootton A. AB 109 and its impact on prison overcrowding and recidivism: A policy analysis. Themis: Research Journal of Justice Studies and Forensic Science. 2016; 4. 12. Bird M, Grattet R, Nguyen V. Realignment and Recidivism in California. San Francisco, CA: Public Pol- icy Institute of California; 2017. 13. Bird M, Grattet R. Do Local Realignment Policies Affect Recidivism in California? San Francisco, CA: Public Policy Institute of California; 2014. 14. Lofstrom M, Raphael S, Grattet R. Is Public Safety Realignment Reducing Recidivism in California? San Francisco, CA: Public Policy Institute of California; 2014. 15. Bird M, Grattet R. Realignment and recidivism. The ANNALS of the American Academy of Political and Social Science. 2016; 664: 176–195. 16. Bird M, Grattet R. Policy change and recidivism: the effects of California realignment and local imple- mentation strategies on rearrest and reconviction. Criminal Justice Policy Review. 2015; 28: 601–623. 17. Lofstrom M, Raphael S. Public Safety Realignment and Crime Rates in California. San Francisco, CA: Public Policy Institute of California; 2013. 18. Sundt J, Salisbury EJ, Harmon MG. Is downsizing prisons dangerous? The effect of California’s realign- ment act on public safety. Criminology and Public Policy. 2016; 15: 315–341. 19. Lofstrom M, Raphael S. Incarceration and crime: evidence from California’s public safety realignment reform. The ANNALS of the American Academy of Political and Social Science. 2016; 664: 196–220. 20. Kelso J.C. Special Report: Improvements in the Quality of California’s Prison Medical Care System. 2015 Mar. Available: https://cchcs.ca.gov/wp-content/uploads/sites/60/2017/08/Kelso-Special-Report- Filed-031015.pdf 21. California Correctional Health Care Services. Health Care Services Dashboard. [cited 31 Aug 2020]. Available: https://cchcs.ca.gov/reports/#dashboard 22. Petersilia J. Looking back to see the future of prison downsizing in America. Keynote Address pre- sented at: National Institute of Justice (NIJ) Conference; 2012 Jun. Available: https://www.ojp.gov/ library/publications/looking-back-see-future-prison-downsizing-america-keynote-address-2012-nij 23. Carson EA. Prisoners in 2020—Statistical Tables. Bureau of Justice Statistics, Office of Justice Pro- grams, U.S. Department of Justice; 2021. 24. Carson EA, Cowhig MP. Mortality in State and Federal Prisons, 2001–2016—Statistical Tables. United States Department of Justice, Office of Justice Programs, Bureau of Justice Statistics; 2020 Feb. Report No.: NCJ 251920. 25. Noonan ME. Mortality in State Prisons, 2001–2014—Statistical Tables. United States Department of Justice, Office of Justice Programs, Bureau of Justice Statistics; 2016 Dec. Report No.: NCJ 250150. 26. United States Department Of Justice. Office Of Justice Programs. Bureau Of Justice Statistics. National Corrections Reporting Program, 2000–2015: Version 1. ICPSR—Interuniversity Consortium for Political and Social Research; 2017. https://doi.org/10.3886/ICPSR36746.V1 27. State of California, Department of Justice, Criminal Justice Statistics Center. Deaths in Custody Data, 2005–2018. 2019. 28. California Department of Justice. Open Justice. 2020 [cited 30 Sep 2020]. Available: https:// openjustice.doj.ca.gov/ 29. Abadie A, Diamond A, Hainmueller J. Synthetic control methods for comparative case studies: Estimat- ing the effect of California’s tobacco control program. Journal of the American Statistical Association. 2010; 105: 493–505. 30. Abadie A, Diamond A, Hainmueller J. Comparative Politics and the Synthetic Control Method. American Journal of Political Science. 2015; 59: 495–510. https://doi.org/10.1111/ajps.12116 31. Urban Institute. A Matter of Time: The Causes and Consequences of Rising Time Served in America’s Prisons. Washington, D.C.; 2017 Jul. Available: https://apps.urban.org/features/long-prison-terms/ intro.html PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 17 / 18 PLOS ONECalifornia’s Public Safety Realignment Act and prisoner mortality 32. Xu J, Murphy S, Kochanke K, Arias E. Mortality in the United States, 2018. Hyattsville, MD: National Center for Health Statistics; 2020. 33. Imai K. Analysis of 2014 Inmate Death Reviews in the California Correctional Healthcare System. 2015 Jul. Available: https://cchcs.ca.gov/wp-content/uploads/sites/60/2017/08/OTRES_ DeathReviewAnalysisYear2014_20150730.pdf 34. California Correctional Health Care Receivership. Achieving a Constitutional Level of Medical Care in California’s Prisons: Twenty-fifth Tri-Annual Report of the Federal Receiver’s Turnaround Plan of Action For September 1—December, 2013. 2014 Feb. 35. California Correctional Health Care Receivership. Achieving a Constitutional Level of Medical Care in California’s Prisons: Twenty-seventh Tri-Annual Report of the Federal Receiver’s Turnaround Plan of Action For May 1—August 31, 2014. 2014 Oct. 36. California Department of Corrections and Rehabilitation. Monthly Reports of Population, July 2013— December 2014. 37. Maruschak L, Berzofsky M, Unangst J. Medical Problems of State and Federal Prisoners and Jail Inmates, 2011–2012. U.S. Department of Justice. Office of Justice Programs. Bureau of Justice Statis- tics.; 2015. 38. California Department of Corrections and Rehabilitation. Monthly Reports of Population, September 2011— September 2012. 39. Garland B, Hogan N, Wodahl E, Hass A, Stohr MK, Lambert E. Decarceration and its possible effects on inmates, staff, and communities. Punishment & Society. 2014; 16: 448–473. https://doi.org/10.1177/ 1462474514539535 40. Ahalt C, Binswanger IA, Steinman M, Tulsky J, Williams BA. Confined to ignorance: the absence of pris- oner information from nationally representative health data sets. J Gen Intern Med. 2012; 27: 160–166. https://doi.org/10.1007/s11606-011-1858-7 PMID: 21922160 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284609 April 28, 2023 18 / 18 PLOS ONE
10.1371_journal.pone.0270038	RESEARCH ARTICLE Knowledge, attitudes, and practices towards COVID-19 among college students in China: A systematic review and meta-analysis Ling LiID 1, Fang Wang2, Xiaoling Shui1, Qian Liang1, Jingyi He1 1 School of Nursing, Chengdu University of Traditional Chinese Medicine, Chengdu, Sichuan Province, China, 2 Nursing Department, Affiliated Hospital of Chengdu University of Traditional Chinese Medicine, Chengdu, Sichuan Province, China a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 wangf7640@163.com Abstract Background OPEN ACCESS Citation: Li L, Wang F, Shui X, Liang Q, He J (2022) Knowledge, attitudes, and practices towards COVID-19 among college students in China: A systematic review and meta-analysis. PLoS ONE 17(6): e0270038. https://doi.org/ 10.1371/journal.pone.0270038 Editor: Hamidreza Karimi-Sari, Middle East Liver Diseases (MELD) Center, Tehran, Iran, ISLAMIC REPUBLIC OF IRAN Received: April 15, 2022 Accepted: June 2, 2022 Published: June 16, 2022 Copyright: © 2022 Li et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Since the outbreak of the respiratory infectious disease caused by the novel coronavirus in 2019, the COVID-19 epidemic has posed a serious threat to the life and safety of the public, and has also seriously affected the normal study and life of college students in China. Although a series of studies have been conducted on college students’ knowledge, attitudes and practices of COVID-19, the results vary widely. This study aimed to evaluate the pooled estimated level of knowledge, attitudes, and practices (KAP) about COVID-19 among col- lege students in China. Methods We conducted a comprehensive search on Scopus, ProQuest, PubMed, EMbase, Web of Science, the Cochrane Library, Chinese Biomedical Literature Database (CBM), China National Knowledge Infrastructure (CNKI), VIP Database and Wanfang Database up to 13 February 2022. We then assessed the quality of included studies using a checklist devel- oped by the Joanna Briggs Institute (JBI) for cross-sectional studies and analyzed using STATA.15 after two researchers independently extracted relevant data and entered them into Microsoft Excel. Funnel plots and Egger’s regression tests were used to check for publi- cation bias, and sensitivity analysis was performed to assess the robustness of the results. A random-effects model was used for the meta-analysis, on the basis of which subgroup analyses were performed by time of investigation (study period) and by gender and major of the subjects. Data Availability Statement: All relevant data are within the paper and its Supporting information files. Result Funding: The author(s) received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. A total of 42 studies including 174,440 subjects were included in this review, and the quality of the included studies was mainly intermediate and advanced. The findings of the meta- analysis showed that the overall levels of Chinese college students’ knowledge, positive and negative attitudes, and practice of preventive measures towards COVID-19 were 74% PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 1 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China (95%CI: 69%-79%), 84% (95%CI: 80%-88%), 31% (95%CI: 23%-38%) and 82% (95%CI: 77%-86%), respectively. The results of subgroup analysis showed that female and medical college students had higher levels of knowledge and practice on COVID-19. Conclusion The study findings showed that the overall KAP level of college students in China included in the study was relatively optimistic. Influenced by gender, major and time, there were dif- ferences in the KAP level of college students. In order to promote the effective prophylaxis and control of pandemic, we recommend implementing targeted measures to improve the awareness rate of COVID-19-related knowledge among this group and the implementa- tion rate of COVID-19 preventive measures among male and non-medical college students. Introduction In late December 2019, a cluster of severe pneumonia patients of unknown cause was reported in Wuhan, China. In the testing of lower respiratory tract samples from 4 patients, a novel strain of coronavirus belonging to the same family of viruses that cause Middle East Respira- tory Syndrome (MERS) and Severe Acute Respiratory Syndrome (SARS), as well as the four human coronaviruses related to the common cold, was isolated [1]. On January 30,2020, the World Health Organization (WHO) declared the novel coronavirus outbreak a public health emergency of international concern, and proposed to name the disease caused by the virus as COVID-19 on February 11 [2]. As of 10 March 2022, COVID-19 has spread to 225 countries and territories, with more than 6 million deaths [3]. According to existing studies, COVID-19 has the characteristics of strong infectivity, easy mutation, and general susceptibility of the population [4, 5]. Although targeted vaccines have been developed and widely used in the population, the duration of immunity obtained from vaccination is unclear [5]. Previous study [6] has shown that carrying out KAP surveys during the epidemic is conducive to understanding the public’s awareness of epidemic prevention and weak points of practice, so as to implement precise health education. Al Ahdab S [7] pointed out that good awareness, positive attitudes and qualified behaviors about COVID-19 are critical to the successful prophylaxis and control of the epidemic. While there were existing studies on the knowledge, attitudes and practices (KAP) towards COVID-19 among college students in China, they were not comprehensive and the results obtained vary. A review article [8] evaluated public awareness of COVID-19 in China, but no meta-analysis was conducted. Another review article [9] evaluating knowledge, attitudes, and practices about COVID-19 among healthcare workers and the general population also did not perform any quantitative analysis. In view of the importance of individual cognition and behavior in the effective prophylaxis and control of the epidemic and the particularity of this group, it is necessary to conduct a systematic review and meta-analysis of KAP of this popula- tion to provide a pooled estimated proportion of KAP for COVID-19 among college students in China, and provide reference for relevant departments to carry out scientific prophylaxis and control during the pandemic. To our knowledge, this is the first systematic review and meta-analysis of college students’ knowledge, attitudes, and practices about COVID-19. PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 2 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China Methods This systematic review and meta-analysis was conducted in accordance with the ‘Preferred Reporting Items for Systematic Reviews and Meta-analyses’ (PRISMA) [10] (S2 File). The sys- tematic review protocol has been registered on PROSPERO (CRD42022316375). Eligibility criteria Inclusion criteria. All observational studies on KAP towards COVID-19 among college students in China were considered for this study, with no restrictions on the gender, health sta- tus of the subjects, and language, time, quality or geographic location of the study. However, only studies that were available in full text and reported the sample size of the study and data related to any one component of the KAP, or provided data from which required outcomes could be calculated, were included. Exclusion criteria. We excluded certain studies that were conducted only on specific pop- ulations, such as medical students. In addition, review articles, studies with confused or poten- tially erroneous data, studies for which raw data could not be extracted or transformed, and studies reporting perception on coronaviruses other than COVID-19 would also be excluded. Information sources, search strategy and study selection The two researchers conducted systematic searches in Scopus, ProQuest, PubMed, EMbase, Web of Science, the Cochrane Library, Chinese Biomedical Literature Database (CBM, http:// www.sinomed.ac.cn/index.jsp), China National Knowledge Infrastructure (CNKI, https:// www.cnki.net/), VIP Database (http://lib.cqvip.com/) and Wanfang Database (https://new. wanfangdata.com.cn/index.html). In addition, other studies including grey literature were searched through Baidu Xueshu and Google Scholar. The main keywords for the search strat- egy included “Covid-19”, “2019-nCoV”, “SARS-CoV-2”, “Severe Acute Respiratory Syndrome Coronavirus 2”, “knowledge”, “perception”, “awareness”, “consciousness”, “attitude”, “prac- tice”, “action” and “KAP”. Search strategies for PubMed and CNKI databases are provided in S2 File. The preliminary search, which was carried out on 1 February 2022 and updated on 13 February. The retrieval results were stored and managed through Endnote software. After screening out duplicate articles, two researchers independently performed prelimi- nary screening of the remaining articles by reading the titles and abstracts according to pre- established criteria. When the title and abstract were insufficient to judge whether to exclude an article, the researcher would make the choice after reading the full text. Any disagreements would be resolved through discussions and consultations with a third researcher. Study outcomes This study involved three main outcomes of KAP associated with COVID-19 among college students in China. Knowledge. Main symptoms, routes of transmission, susceptible population, incubation period, knowledge or awareness of preventing COVID-19. Attitudes. Attitudes towards managing or controlling of COVID-19, and fears or worri- some about COVID-19. Practices. Measures such as wear masks, maintain social distance, avoid crowded places/ social activities and hand hygiene. PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 3 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China Data extraction The relevant data were independently extracted and entered into an Excel spreadsheet tem- plate by two researchers. Include: first author, year and month of study, study area, study design, sample size, demographics of study subjects, and study results related to any compo- nent of KAP. (The mastery and implementation of knowledge and practices by participants in each study was obtained by calculating the overall average proportion of each component of KAP.) In addition, the proportion of specific content in each component of KAP was also extracted from the included studies. Any disagreements during extraction were resolved through discussion and negotiation. When necessary, we would contact study authors for missing data. Quality assessment The quality of included studies was independently assessed by two researchers using the JBI Quality Assessment Checklist [11], which consists of eight questions, namely: (1) Were the inclusion criteria for study subjects clear?, (2) Were the research objects and places described in detail?, (3) Was the assessment method of exposure factors valid and credible?, (4) Was the assessment method of health problems objective and standard?, (5) Were the confounding fac- tors clearly defined?, (6) Were there any measures to control confounders?, (7) Were the out- comes measured in a valid and credible way?, and (8) Was the statistical analysis appropriate?. Each question has three options of “yes, no, not clear”, each of which accounts for 1 point, 0 point, and 0.5 point. After summarizing the scores obtained for each question, studies were considered to have a high risk of bias and low quality if the total score was less than 3 points; moderate risk of bias and moderate quality if the total score was 3 to <6 points; low risk of bias and high quality if the total score was 6–8 points [12]. Differences were resolved through discussions and consultations with a third researcher. Statistical analysis Descriptive analysis was used in most sections to report relevant content, and the statistical software package STATA.15 was used to analyze quantitative data extracted from each study. Funnel plots and Egger’s regression test were used to check for publication bias, with P <0.05 considered to exist potential publication bias [13]. The Cochrane Q statistic was used to test for heterogeneity between studies, and the heterogeneity was quantified by the I2 value, where 25%, 50%, and 75% indicated low, moderate, and high heterogeneity, respectively [14]. As het- erogeneity was found to be greater than 50%, we used the random effect model for the analysis. Subgroup analyses were also performed by study period (every 3 months as a subgroup), gen- der and major of subjects to explore sources of heterogeneity. To assess the robustness of the findings, we also performed a sensitivity analysis [15]. The final results were reported as pooled proportion and 95% confidence interval for COVID-19 KAP and were presented in forest plot. Results Search outcomes A total of 9,688 articles were retrieved from the selected database. After excluding 4,599 dupli- cate articles, another 5,047 articles were excluded according to the inclusion and exclusion cri- teria on the basis of reading the title, abstract and full text. Finally, 42 articles were included for qualitative and quantitative analysis, including 174,804 participants [16–57]. The PRISMA flow chart for the study selection process is presented in Fig 1. PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 4 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China Fig 1. PRISMA 2009 flow diagram. https://doi.org/10.1371/journal.pone.0270038.g001 Characteristics of included studies A total of 42 studies encompassing 174,440 participants were included, most of which were conducted in 2020. Except for the 4 studies that did not elucidate the gender characteristics of the participants, in the remaining 38 studies, 60.63% of the participants were female. All studies employed a cross-sectional design, and data were collected primarily through online questionnaires. Nine studies reported all three components of KAP, and the remaining 33 reported only two components or one component of KAP. The sample size of each study varied from 134 to 44446. For more information on included studies and participants charac- teristics, see S1 File. Quality assessment of included studies The overall average score of the included studies was 4.74 according to the JBI quality assess- ment checklist. Among them, 16 studies (38.1%) were rated as good quality (score �6) and 26 (61.9%) as moderate quality (score 3–5). All studies failed to score on questions 5 and 6, which were related to the failure to identify and deal with confounding factors in the research pro- cess. (See S2 File for details). PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 5 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China Synthesis of results The pooled proportion and 95% confidence interval of knowledge, attitudes, and practices towards COVID-19 among college students in China is presented in a forest plot (Figs 2–5). Among them, the pooled proportion of knowledge about COVID-19 was 74% (95%CI: 69%- 79%, I2 = 99.8%), the pooled proportions of positive and negative attitudes towards COVID- 19 were 84% (95%CI: 80–88%, I2 = 99.7%) versus 31% (95%CI: 23%-38%, I2 = 99.6%), and the pooled proportion for COVID-19 practice level was 82% (95%CI: 77%-86%, I2 = 99.7%). In addition, further analysis was carried out on the specific content of each component of KAP. The results showed that the pooled awareness proportions of Chinese college students about the main symptoms, routes of transmission, susceptible population, incubation period and preventive measures of COVID-19 were 80%, 73%, 43%, 87%, and 70%, respectively. The Fig 2. Forest plot showing the overall proportion of knowledge about COVID-19 among study subjects. https://doi.org/10.1371/journal.pone.0270038.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 6 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China Fig 3. Forest plot showing the pooled proportion of positive attitudes towards COVID-19 among study participants. https://doi.org/10.1371/journal.pone.0270038.g003 pooled proportions of practice for specific precautions against COVID-19 (wearing masks, washing hands, and social distancing) were 90%, 84%, and 87%, respectively. (See S2 File for details). Publication bias analysis Publication bias was evaluated by funnel plot and Egger’s regression test. The funnel plot and Egger’s regression test results indicated that: there was publication bias among the studies on positive attitudes towards COVID-19 among college students in China (P = 0.043 <0.05) (Fig 6). In addition, the research points of the other three parts were basically symmetrical in the funnel plot, and the results calculated by Egger linear regression (P = 0.333, P = 0.274, P = 0.867) also indicated that there was no significant publication bias. The result was PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 7 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China Fig 4. Forest plot showing the pooled proportion of negative attitudes towards COVID-19 among study participants. https://doi.org/10.1371/journal.pone.0270038.g004 corrected by the trim-and-fill method, and the result [83%, 95%CI: 77.6%-88.5%, P<0.001] was not significantly different from the value before correction [84%, 95%CI: 80%-88%, P<0.001], indicating that the pooled proportion of positive attitudes towards COVID-19 among college students in China was less affected by publication bias. Sensitivity analysis The robustness of the results was assessed using a leave-out-one sensitivity analysis, which showed that stepwise exclusion of each study did not result in any significant changes in the pooled proportion for knowledge, positive and negative attitudes, and level of practice. (See S2 File for details). PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 8 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China Fig 5. Forest plot showing the pooled proportion of using specific measures to fight COVID-19 among study subjects. https://doi.org/10.1371/journal.pone.0270038.g005 Subgroup analysis The results of subgroup analysis based on the gender and major of participants showed that women had relatively higher knowledge (75%, 95%CI: 63%-86% vs 72%, 95%CI: 61%-82%) and practice (79%, 95%CI: 72%-85% vs 76%, 95%CI: 67%-85%) compared with men, with no significant difference in attitude towards COVID-19 (79%, 95%CI: 70%-88% vs 79%, 95%CI: 68%-89%). Compared with non-medical college students, medical college students had a higher level of knowledge towards COVID-19 (75%, 95%CI: 57%-92% vs 66%, 95%CI: 55%- 78%), more positive attitude (76%, 95%CI: 65%-87% vs 73%, 95%CI: 60%-87%) and higher level of practice (79%, 95%CI: 66%-92% vs 75%, 95%CI: 66%-85%). PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 9 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China Fig 6. Begg’s funnel plot checking the publication bias of proportion of positive attitudes about COVID-19 among study subjects. https://doi.org/10.1371/journal.pone.0270038.g006 The results of subgroup analysis based on the study period showed that over time, the pooled proportions of the mastery and practice of COVID-19-related knowledge and preven- tive measures among college students in China showed an upward and downward trend, respectively. But there was no noticeable change in their attitudes towards COVID-19 (Table 1). Subgroup analyses for negative attitudes were not performed because of the rela- tively small number of studies reporting specific data on gender and major among all studies on negative attitudes, and the study period focused on the first three months. Discussion Although China has long entered the stage of normalizing the prophylaxis and control of the COVID-19 epidemic, clustered epidemic events on campus still occured from time to time. As a global pandemic that the most extensive to afflict humanity in a century [58], the prophylaxis and control of the COVID-19 is still a huge challenge of facing the world. As a special group with high educational level and high activity, college students’ epidemic prevention literacy level is not only related to their own health status, but also affects the cognition and behavior of the people around them to a certain extent. In this context, assessing the overall proportion of knowledge, attitudes, and practices towards COVID-19 among college students in China may help relevant departments to improve prophylaxis and control strategies for the COVID- 19 epidemic. PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 10 / 17 PLOS ONETable 1. Subgroup analysis based on gender and major. Knowledge, attitudes, and practices towards COVID-19 among college students in China Outcome Knowledge Positive attitude Practice Subgroup analysis based on gender and major Study characteristics No of studies Estimates score (%) (95%CI),p-value gender male female majors medical students non-medical students Study period January-March, 2020 April-June, 2020 July 2020 and beyond gender male female majors medical students non-medical students Study period January-March, 2020 April-June, 2020 gender male female majors medical students non-medical students Study period January-March, 2020 April-June, 2020 10 9 22 4 3 7 6 16 2 8 5 10 3 72 (61–82) <0.001 75 (63–86) <0.001 75 (57–92) <0.001 66 (55–78) <0.001 73 (67–80) <0.001 75 (64–86) <0.001 75 (66–84) <0.001 79 (70–88) <0.001 79 (68–89) <0.001 76 (65–87) <0.001 73 (60–87) <0.001 83 (78–88) <0.001 83 (77–89) 0.002 76 (67–85) <0.001 79 (72–85) <0.001 79 (66–92) <0.001 75 (66–85) <0.001 82 (76–87) <0.001 76 (61–91) <0.001 https://doi.org/10.1371/journal.pone.0270038.t001 I2 99.4% 99.8% 99.8% 99.6% 99.8% 97.8% 98.9% 99.1% 99.7% 99.8% 99.8% 99.8% 89.3% 99.6% 99.7% 99.8% 99.6% 99.7% 98.9% The findings of this meta-analysis showed that the pooled proportion of college students’ knowledge of COVID-19 was 74%, which is similar to the findings reported by Aynalem YA et al. [59]. The results of subgroup analysis showed that the pooled proportion of Chinese col- lege students’ mastery of COVID-19-related knowledge had gradually increased over time, and female and medical college students had a higher level of knowledge towards COVID-19. Similar findings were also reported in South Korea [60] and the United Arab Emirates [61]. These differences may be caused by the following reasons: in terms of time, less was known about the new disease at the beginning of the outbreak, while the gradual release of relevant information and guidelines later provided more knowledge about the disease. In terms of gen- der, on the one hand, the proportion of women in most studies is higher; on the other hand, it may be related to men’s risk-taking tendency and vulnerability to peer influence [62]. In terms of majors, on the one hand, medical students have more time and opportunities to contact health information and receive medical-related education, thereby developing a stronger sense of self-protection; on the other hand, medical students may be inclined to acquire and update their knowledge of COVID-19 from academic media and research articles. Although college students’ overall awareness of COVID-19-related knowledge was relatively optimistic, their PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 11 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China awareness of specific content such as “susceptible population, routes of transmission and pre- ventive measures” still needs to be improved. This also suggests that schools and relevant departments should carry out targeted publicity and education strategies in time. The pooled proportion of subjects with positive attitudes about COVID-19 in this study was 84%, which was higher than the reported results of two studies conducted in Ethiopia (67.2%) [63] and Japan (68.5%) [64]. This might be related to the effectiveness of China’s con- trol of COVID-19 as well as the concerted prevention and control efforts implemented across the country [65]. However, this study showed that the pooled proportion of participants who were worried or fearful about COVID-19 was 31%, which is much higher than the results reported by He LP et al. [66] during the SARS epidemic (12.5%). The reasons may be related to the long duration and wide spread of COVID-19, as well as the repeated closure measures taken by colleges and universities to avoid campus transmission. Relevant departments need to pay attention to the psychological status of this group and provide supportive measures to improve the psychological coping and resilience of college students during and following COVID-19 [67]. Furthermore, the pooled proportion for COVID-19 practice level was 82%, which was slightly lower than the findings of Yang K et al. [68] (84.4%). The results of subgroup analysis showed that female and medical college students had better practice level for COVID-19. However, the pooled proportion of Chinese college students’ practice of preventive measures against COVID-19 had a decreasing trend, which might be related to their reduced risk per- ception of the disease [69, 70]. As far as the pooled results were concerned, both in terms of overall level and specific measures, the implementation of COVID-19-related preventive mea- sures among college students in China was relatively optimistic. In this study, we observed a significant heterogeneity (I2>99%) in the pooled proportions of KAP components. To find the sources of heterogeneity, we used subgroup analysis and “a leave-out-one” sensitivity analysis were conducted to further assess relevant studies. The results showed that the pooled results of KAP components were robust and not influenced by a single study. Therefore, we considered that the high heterogeneity might be caused by differ- ences in the original study sample size, questionnaire content, definition of results, and mea- surement methods. Strength and limitation of this study About the advantages of this study: First, a systematic search was performed in multiple repre- sentative databases. Second: the quality of the included studies were mainly at the middle and high levels, and included most of the regions in China, which had a good representativeness. Third: to our knowledge, this is the first systematic review of KAP for Chinese college students on COVID-19, and the findings provided some meaningful data. Limitations of this study: First, the heterogeneity among studies was high. Despite a series of subgroup analyses, the sources of heterogeneity could not be determined. Second, the stud- ies included all used a cross-sectional design, which inevitably has inherent shortcomings of this design. Third, the original research lacks uniform measurement tools and operational definitions. Conclusion This study showed that the overall levels of components of KAP related to COVID-19 were higher among college students in China. In terms of knowledge and practices, female and medical college students had better mastery and implementation. Although the overall PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 12 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China situation is optimistic, in terms of knowledge, corresponding measures need to be imple- mented to increase the awareness rate of COVID-19 in this group. Supporting information S1 Checklist. PRISMA checklist. (DOC) S1 File. Relevant data extracted from reported studies. (XLSX) S2 File. (DOC) Author Contributions Conceptualization: Ling Li, Fang Wang. Data curation: Ling Li, Fang Wang, Xiaoling Shui. Formal analysis: Ling Li, Fang Wang, Qian Liang. Investigation: Ling Li, Jingyi He. Methodology: Ling Li, Fang Wang, Xiaoling Shui, Qian Liang, Jingyi He. Project administration: Fang Wang. Software: Ling Li, Qian Liang. Supervision: Fang Wang. Validation: Ling Li, Fang Wang, Jingyi He. Visualization: Ling Li, Xiaoling Shui. Writing – original draft: Ling Li, Xiaoling Shui. Writing – review & editing: Ling Li, Fang Wang. References 1. Cucinotta D, Vanelli M. WHO Declares COVID-19 a Pandemic. Acta bio-medica: Atenei Parmensis. 2020; 91(1):157–60. https://doi.org/10.23750/abm.v91i1.9397 PMID: 32191675. 2. World Health Organization. Timeline of WHO’s response to COVID-19. [cited 10 March 2022]. https:// www.who.int/emergencies/diseases/novel-coronavirus-2019/interactive-timeline#. 3. Worldometer.info. COVID-19 coronavirus pandemic. [cited 10 March 2022]. https://www.worldometers. info/coronavirus/. 4. Su S, Wong G, Shi W, Liu J, Lai AC, Zhou J, et al. Epidemiology, genetic recombination, and pathogen- esis of coronaviruses. Trends in microbiology. 2016; 24(6):490–502. https://doi.org/10.1016/j.tim.2016. 03.003 PMID: 27012512 5. China NHCotPsRo. Diagnosis and treatment plan for COVID-19 (trial version 8 revision). Chinese Jour- nal of Clinical Infectious Diseases. 2021; 14(2):81–8. 6. Smith HL. On the limited utility of KAP-style survey data in the practical epidemiology of AIDS, with ref- erence to the AIDS epidemic in Chile. Health Transit Rev. 1993 Apr; 3(1):1–16. PMID: 10148795. 7. Al Ahdab S. A cross-sectional survey of knowledge, attitude and practice (KAP) towards COVID-19 pandemic among the Syrian residents. BMC public health. 2021; 21(1):296. Epub 2021/02/07. https:// doi.org/10.1186/s12889-021-10353-3 PMID: 33546652. 8. Liu Y, Yu Y, Yi W, Wang Y, Wang J. Public Awareness Toward COVID-19 Among Chinese Residents During the Pandemic: A Systematic Review. Medical Information. 2020; 33(23):88–91. https://doi.org/ 10.3969/j.issn.1006-1959.2020.23.025 PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 13 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China 9. Bekele F, Sheleme T, Fekadu G, Bekele K. Patterns and associated factors of COVID-19 knowledge, attitude, and practice among general population and health care workers: A systematic review. SAGE open medicine. 2020; 8:2050312120970721-. https://doi.org/10.1177/2050312120970721 PMID: 33240497. 10. Moher D, Liberati A, Tetzlaff J, Altman DG, Group P. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS medicine. 2009; 6(7):e1000097–e. Epub 2009/07/ 21. https://doi.org/10.1371/journal.pmed.1000097 PMID: 19621072. 11. Joanna Briggs Institute. Critical Appraisal Tools Australia: The University of Adelaide; 2018. [cited 10 March 2022]. https://jbi.global/critical-appraisal-tools. 12. Arab-Zozani M, Pezeshki MZ, Khodayari-Zarnaq R, Janati A. Inappropriate Rate of Admission and Hos- pitalization in the Iranian Hospitals: A Systematic Review and Meta-Analysis. Value Health Reg Issues. 2020; 21:105–12. Epub 2019/11/11. https://doi.org/10.1016/j.vhri.2019.07.011 PMID: 31704488. 13. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ (Clinical research ed). 1997; 315(7109):629–34. https://doi.org/10.1136/bmj.315.7109.629 PMID: 9310563. 14. Higgins JPT, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ (Clinical research ed). 2003; 327(7414):557–60. https://doi.org/10.1136/bmj.327.7414.557 PMID: 12958120. 15. Yin G, Zhang Y, Chen C, Ren H, Guo B, Zhang M. Have you ever heard of Human Papillomavirus (HPV) vaccine? The awareness of HPV vaccine for college students in China based on meta-analysis. Human vaccines & immunotherapeutics. 2021; 17(8):2736–47. Epub 2021/04/01. https://doi.org/10. 1080/21645515.2021.1899731 PMID: 33787459. 16. Ding Y, Du X, Li Q, Zhang M, Zhang Q, Tan X, et al. Risk perception of coronavirus disease 2019 (COVID-19) and its related factors among college students in China during quarantine. Plos One. 2020; 15(8). https://doi.org/10.1371/journal.pone.0237626 PMID: 32790791 17. Gao Z, Ying S, Liu J, Zhang H, Li J, Ma C. A cross-sectional study: Comparing the attitude and knowl- edge of medical and non-medical students toward 2019 novel coronavirus. Journal of infection and pub- lic health. 2020; 13(10):1419–23. Epub 2020/07/01. https://doi.org/10.1016/j.jiph.2020.06.031 PMID: 32653479. 18. 19. Jia Y, Ma S, Bai L, Xiao Q, Wu Y, Gao Y, et al. Health Literacy and Disparities in Knowledge, Attitude and Practice Regarding COVID-19 Among College Students During the COVID-19 Outbreak in China: A Cross-Sectional Study. Risk Management and Healthcare Policy. 2021; 14:4477–88. https://doi.org/ 10.2147/RMHP.S319331 PMID: 34754254 Jiang R. Knowledge, attitudes and mental health of university students during the COVID-19 pandemic in China. Children and Youth Services Review. 2020;119. https://doi.org/10.1016/j.childyouth.2020. 105494 PMID: 33518860 20. Peng Y, Pei C, Zheng Y, Wang J, Zhang K, Zheng Z, et al. A cross-sectional survey of knowledge, atti- tude and practice associated with COVID-19 among undergraduate students in China. BMC public health. 2020; 20(1). https://doi.org/10.1186/s12889-020-09392-z PMID: 32847554 21. Shen D, Liu D, Cai M, Chen P, Wang Z, Zhang Y, et al. Association between supportive attitude and adoptive practice of control strategy against COVID-19 amosng college students in China: a cross-sec- tional study. BMC public health. 2021; 21(1). https://doi.org/10.1186/s12889-021-10752-6 PMID: 33902517 22. Tao X, Chen D, Liang R, Zhang X, Yu X, Chung S, et al. A comparative study on knowledge towards COVID-19 prevention among undergraduate students in Macao and Zhuhai, China. PeerJ. 2021;9. https://doi.org/10.7717/peerj.11833 PMID: 34414031 23. Xu H, Mendez MJG, Guo L, Chen Q, Zheng L, Chen P, et al. Knowledge, Awareness, and Attitudes Relating to the COVID-19 Pandemic Among Different Populations in Central China: Cross-Sectional Survey. Journal of Medical Internet Research. 2020; 22(10). https://doi.org/10.2196/22628 PMID: 32886623 24. Cheng YM, Cui GZ, Li WX, Liu Q, Ren Q, Xu L, et al. Investigation on cognitive and psychological status of prevention and control about COVID-19 among college students in Xi’an. J Community Med. 2021; 19(01):1–5. https://doi.org/10.19790/j.cnki.JCM.2021.01.01 25. Ding MJ, Hu CF. Study on the Influence of the Epidemic Situation of COVID-19 on the Psychological Behavior of College Students. Journal of Jiangsu Ocean University (Humanities & Social Sciences Edi- tion). 2020; 18(02):129–40. 26. Ge WJ, Xiao YY, Xing BY, Shen M, Liao QF. College students’ cognition of novel coronavirus pneumo- nia epidemic and its relationship with coping style. Chin J Sch Health. 2020; 41(05):670–2. https://doi. org/10.16835/j.cnki.1000-9817.2020.05.010 PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 14 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China 27. Guan CL, Wang Y, Zhao GP, Liu YN, Gao L, Qiao YF. Investigation and analysis of the COVID-19 pre- vention and control knowledge in the students of some university during the COVID-19 vaccination period. Journal of Logistics University of PAP (Medical Sciences). 2021; 30(05):100–3. https://doi.org/ 10.16548/j.2095-3720.2021.05.032 28. Guo L, Han S. Research on the cognitive and behavioral characteristics of college students under the COVID-19 pandemic—based on survey data of 7582 college students in China. Journal of Beijing City University. 2021;(02):44–50+65. https://doi.org/10.16132/j.cnki.cn11-5388/z.2021.02.008 29. Han JL, Qu SM, Fu ZY. Research on cognition and anxiety of college students under the new crown pneumonia epidemic. Youthful Years. 2020;(23):130–1. 30. Huang MY, Zhu LJ, Jin YL, Fang ZM, Wu N, DU MX, et al. Investigation of cognition and prevention and control behavior of COVID-19 among undergraduates in Anhui Province. Chin J Dis Control Prev. 2020; 24(05):529–33. https://doi.org/10.16462/j.cnki.zhjbkz.2020.05.007 31. Huang YJ, Hu QY, Zhang JY, Wang WJ. Investigation on the status quo of college students’ awareness of prevention and control of new coronary pneumonia and their ability to prevent and control. Journal of Hubei University of Science and Technology (Medical Sciences). 2020; 34(06):508–12. https://doi.org/ 10.16751/j.cnki.2095-4646.2020.06.0508 32. Huang YZ, Zou XH, Xiong SP, Lu JJ, Chen Y, Jiao Y. Investigation of college students’ cognition,atti- tude and related behaviors of COVID-19. Journal of Yichun University. 2020; 42(09):57–60+120. Avail- able from: https://www.cnki.com.cn/Article/CJFDTotal-YCSB202009015.htm. 33. 34. Jia LL, Ou YJ, Qiao YL, Li XQ, Bai SM, Zhou YH. Investigation on the Cognition Status of COVID-19 Knowledge Among Different Populations in Shanxi Province. Medicine and Society. 2021; 34(02):94–8. https://doi.org/10.13723/j.yxysh.2021.02.019 Jiang XQ, Duan LM. Investigation and analysis of college students’ prevention and control cognition and psychological behavior under the influence of the new coronavirus pneumonia epidemic-Taking Suzhou University as an example. Agriculture of Henan. 2021;(18):8–10. https://doi.org/10.15904/j. cnki.hnny.2021.18.003 35. Kuang ZL, Guo KW, Liu WK, Li JY, Kuang YF. Investigation of knowledge cognition status and mentality status of college students in Wuhan on epidemic prevention of COVID-19. J Trop Med. 2020; 20 (03):283–5+8. Available from: https://www.cnki.com.cn/Article/CJFDTotal-RDYZ202003002.htm. 36. 37. 38. 39. Li CX, Tian PY, Ni YJ, Zhou C. A study on the cognition of college students on COVID-19 prevention and control knowledge and health education in Panzhihua university. The Medical Forum. 2021; 25 (25):3684–5. https://doi.org/10.19435/j.1672-1721.2021.25.066 Li QJ, Luo MY, Wang ZP, Li XY, Zhang XT, Wang Y. The cognition and behavioral status of college stu- dents in China on COVID-19. Medical Diet and Health. 2020; 18(04):200–3. Li YX, Chen K. Cognition of COVID-19 among college students and its relationship with their physical and mental health. Journal of Henan University (Social Science). 2020; 60(03):20–5. https://doi.org/10. 15991/j.cnki.411028.2020.03.004 Liu M, Li Y, Liu L. Investigation and analysis of college students’ cognition, attitude and behavior of novel coronavirus pneumonia. Chinese General Practice Nursing. 2020; 18(08):938–40. https://doi.org/ 10.12104/j.issn.1674-4748.2020.08.064 40. Naijimai MMTAL, Pazilaiti ANW, Peierdun MJT. A cross-sectional survey on knowledge, attitude and practice towards coronavirus disease 2019 among university students in Urumqi,xinjiang. Chin Prev Med. 2021; 22(01):33–6. https://doi.org/10.16506/j.1009-6639.2021.01.007 41. Peng Y. Investigation on cognition, psychological reaction and behavior pattern of prevention of COVID-19 in college students. Journal of Xiangnan University. 2021; 42(03):111–6. https://doi.org/10. 3969/j.issn.1672-8173.2021.03.022 42. Tan J, Liu N. Study on the knowledge, attitude, and practice of COVID-19 in college students. Journal of Zunyi Medical University. 2020; 43(03):384–9. https://doi.org/10.14169/j.cnki.zunyixuebao.2020.0073 43. Wang S, Qin N, Zhang C, Hu C, Liu CY. An Investigation of the awareness of COVID-19 epidemic and psychological behavior among college students. Journal of Qingdao University (Medical Sciences). 2020; 56(06):738–41. 44. Wang LL,Wang JH, Ai CY, Yang YL, Zhang WZ. Risk perception and behavior analysis of Guizhou col- lege students on COVID-19 epidemic. Chin J Sch Health. 2021; 42(02):211–4. https://doi.org/10. 16835/j.cnki.1000-9817.2021.02.013 45. Wang S, Li H, Cai YX, Zhen Y, Zhang YF, Wang Q. Current status of knowledge, attitude and practice (KAP) of corona virus disease 2019 (COVID-19) prevention and control and influencing factors among 2842 students in Xuzhou City. Chin J School Doctor. 2021; 35(8). Available from: http://lib.cdutcm.edu. cn:29017/Qikan/Article/Detail?id=7105675803. PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 15 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China 46. Wang Y, Guo SH. The cognition, coping and exercise behavior of college students on COVID-19. Jour- nal of Chaohu University. 2020; 22(06):127–36. Available from: http://xb.chu.edu.cn/CN/Y2020/V22/I6/ 127. 47. Wang YY, Yang MT, Sun QR, Liu L, Ji WQ, Hou J, et al. Investigation on knowledge, attitude, practice and psychological reaction of Corona Virus Disease 2019 among students of a college in Henan. Henan Medical Research. 2020; 29(28):5185–9. https://doi.org/10.3969/j.issn.1004-437X.2020.28.001 48. Wu CJ, Xiong Y, Yan S, Zhou QW. A cross-sectional study on knowledge attitude and behavior towards COVID-19 prevention among college students in the phase of prevention and control in Yangzhou. Medical Higher Vocational Education and Modern Nursing. 2021; 4(6). https://doi.org/10.3969/j.issn. 2096-501X.2021.06.022 49. Wu P, Deng QD, Wang GP, Zhang L, Liu B, Gao JW, et al. Investigation on protective cognition and emotional response of college students during COVID-19 epidemic. China Journal of Health Psychol- ogy. 2020; 28(11). https://doi.org/10.13342/j.cnki.cjhp.2020.11.012 50. Xiao H, Wang LY, Xiao CC, Yan H. Analysis of psychological behavior of college students in Wuhan during the COVID-19 epidemic. J of Pub Health and Prev Med. 2020; 31(3). https://doi.org/10.3969/j. issn.1006-2483.2020.03.004 51. Xu LY, Zuo HJ, Wang YN, Qi LN. Investigation on knowledge, attitude and behavior of new coronary pneumonia epidemic among college students in a university in Hebei Province. Health Vocational Edu- cation. 2021; 39(05):109–11. Available from: http://qikan.cqvip.com/Qikan/Article/Detail?id= 7104225241. 52. Yang H, Bin P, He JW. Mental state and behavioral coping of college students in COVID-19 epidemic areas-A research report based on the online survey of students in four key colleges and universities in Wuhan. Governance Studies. 2020; 36(2):59–74. 53. 54. 55. 56. 57. 58. Zhang J, Zeng J, Luo J, Zou X, Gu J. Epidemiological survey of the COVID-19 epidemic-related knowl- edge, behaviors and psychology status among college students and their family members. Modern Pre- ventive Medicine. 2020; 47(20):3754–8. Available from: https://kns.cnki.net/KCMS/detail/detail.aspx? dbcode=CJFD&filename=XDYF202020029. Zhao CX, Yang S, Wang SW, Niu YL, Zhang ZL, Wang WX, et al. Study on comprehensive cognition, psychology and behavior of college students during COVID-19 outbreak. Prev Med Trib. 2020; 26 (08):566–70+83. Zhao N, Dai RJ, Zhang HL, Yue J, Pei LY, Wu R, et al. Knowledge, attitude, practice and demands after returning to school on COVID-19 in college students. South China J Prev Med. 2021; 47(3). Zhu JX, Xu RB, Ling HF, Yuan TM, Zhang QH. Cognition and psychological status of COVID-19 among college students in Zhejiang province. Occupational Health and Damage. 2021; 36(2). Available from: https://www.cnki.com.cn/Article/CJFDTotal-ZYWB202102002.htm. Zhu JX, Xu N, Pan YW, Ying PX, Ye JF, Liu CX, et al. Investigation research on cognition, prevention and control behavior and psychosomatic healthy status of different types of adults during the epidemic period of corona virus disease-19. China Medicine. 2020; 15(06):816–20. The State Council Information Office of the People’s Republic of China.《Fighting Covid-19 China in Action《White Paper. (2020-06-07) [cited 18 March 2022]. http://www.gov.cn/zhengce/2020-06/07/ content_5517737.htm. 59. Aynalem YA, Akalu TY, Gebresellassie Gebregiorgis B, Sharew NT, Assefa HK, Shiferaw WS. Assess- ment of undergraduate student knowledge, attitude, and practices towards COVID-19 in Debre Berhan University, Ethiopia. PloS one. 2021; 16(5):e0250444–e. https://doi.org/10.1371/journal.pone.0250444 PMID: 34003825. 60. Lee M, Kang B-A, You M. Knowledge, attitudes, and practices (KAP) toward COVID-19: a cross-sec- tional study in South Korea. BMC Public Health. 2021; 21(1):295-. https://doi.org/10.1186/s12889-021- 10285-y PMID: 33546644. 61. Hasan H, Raigangar V, Osaili T, Neinavaei NE, Olaimat AN, Aolymat I. A Cross-Sectional Study on Uni- versity Students’ Knowledge, Attitudes, and Practices Toward COVID-19 in the United Arab Emirates. Am J Trop Med Hyg. 2021; 104(1):75–84. https://doi.org/10.4269/ajtmh.20-0857 PMID: 33236710. 62. 63. Tang CS-k Wong C-y. Factors influencing the wearing of facemasks to prevent the severe acute respi- ratory syndrome among adult Chinese in Hong Kong. Prev Med. 2004; 39(6):1187–93. https://doi.org/ 10.1016/j.ypmed.2004.04.032 PMID: 15539054. Tadese M, Mihretie A. Attitude, preparedness, and perceived self-efficacy in controlling COVID-19 pan- demics and associated factors among university students during school reopening. PLoS One. 2021 Sep 2; 16(9):e0255121. https://doi.org/10.1371/journal.pone.0255121 PMID: 34473719. 64. Hatabu A, Mao X, Zhou Y, Kawashita N, Wen Z, Ueda M, et al. Knowledge, attitudes, and practices toward COVID-19 among university students in Japan and associated factors: An online cross-sectional PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 16 / 17 PLOS ONEKnowledge, attitudes, and practices towards COVID-19 among college students in China survey. PloS one. 2020; 15(12):e0244350–e. https://doi.org/10.1371/journal.pone.0244350 PMID: 33347488. 65. Yang Y, Peng F, Wang R, Yange M, Guan K, Jiang T, et al. The deadly coronaviruses: The 2003 SARS pandemic and the 2020 novel coronavirus epidemic in China. J Autoimmun. 2020; 109:102434-. Epub 2020/03/03. https://doi.org/10.1016/j.jaut.2020.102434 PMID: 32143990. 66. He LP, Guo Y, Feng Y, Li SY, Liu Q, Chen XY, et al. Knowledge, Attitude, Practice of college students to SARS in Wuhan. Chin J School Health. 2004;( 03):273–4. https://doi.org/http%3A//dx.doi.org/10. 3969/j.issn.1000-9817.2004.03.010 67. Sirrine EH, Kliner O, Gollery TJ. College Student Experiences of Grief and Loss Amid the COVID-19 Global Pandemic. Omega. 2021:302228211027461. Epub 2021/06/25. https://doi.org/10.1177/ 00302228211027461 PMID: 34162233. 68. Yang K, Liu H, Ma L, Wang S, Tian Y, Zhang F, et al. Knowledge, attitude and practice of residents in the prevention and control of COVID-19: An online questionnaire survey. Journal of advanced nursing. 2021; 77(4):1839–55. Epub 2020/12/02. https://doi.org/10.1111/jan.14718 PMID: 33259651. 69. Weinstein ND. The precaution adoption process. Health psychology. 1988; 7(4):355. https://doi.org/10. 1037//0278-6133.7.4.355 PMID: 3049068 70. Brewer NT, Chapman GB, Gibbons FX, Gerrard M, McCaul KD, Weinstein ND. Meta-analysis of the relationship between risk perception and health behavior: the example of vaccination. Health psychol- ogy. 2007; 26(2):136. doi: https://doi.org/10.1037/0278-6133.26.2.136 PMID: 17385964 PLOS ONE \| https://doi.org/10.1371/journal.pone.0270038 June 16, 2022 17 / 17 PLOS ONE
10.1371_journal.pone.0284364	RESEARCH ARTICLE L-type calcium channel blocker increases VEGF concentrations in retinal cells and human serum 1,2,3☯, Valma Harjutsalo1,2,3, Anmol KumarID Raija Lithovius1,2,3, Sinnakaruppan Mathavan4, Markku Lehto1,2,3, Timo P. Hiltunen2,5, Kimmo K. Kontula2,5, Per-Henrik Groop1,2,3,6* 1,2,3, Stefan Mutter1,2,3☯, Erika B. ParenteID 1 Folkha¨lsan Institute of Genetics, Folkha¨ lsan Research Center, Helsinki, Finland, 2 Research Program for Clinical and Molecular Metabolism, Faculty of Medicine, University of Helsinki, Helsinki, Finland, 3 Department of Nephrology, University of Helsinki and Helsinki University Hospital, Helsinki, Finland, 4 Vision Research Foundation, Sankara Nethralaya, Chennai, India, 5 Department of Medicine, University of Helsinki & Helsinki University Hospital, Helsinki, Finland, 6 Department of Diabetes, Central Clinical School, Monash University, Melbourne, Australia ☯ These authors contributed equally to this work. * per-henrik.groop@helsinki.fi Abstract Objective Vascular endothelial growth factor (VEGF) plays a key role in diabetic retinopathy (DR). Pre- viously, we have reported an association between mutations in a gene coding for the L-type calcium channel subunit, VEGF and DR. L-type calcium channel blockers (LTCCBs) have been widely used as antihypertensive medication (AHM), but their association with VEGF and DR is still unclear. Therefore, we explored the effect of LTCCBs compared to other AHMs on VEGF concentrations in retinal cells and human serum. Furthermore, we evalu- ated the association between the use of LTCCBs and the risk of severe diabetic eye disease (SDED). Research design and methods Mu¨ ller cells (MIO-M1) were cultured as per recommended protocol and treated with LTCCBs and other AHMs. VEGF secreted from cells were collected at 24 hours intervals. In an interventional study, 39 individuals received LTCCBs or other AHM for four weeks with a four-week wash-out placebo period between treatments. VEGF was measured during the medication and placebo periods. Finally, we evaluated the risk of SDED associated with LTCCB usage in 192 individuals from the FinnDiane Study in an observational setting. Results In the cell cultures, the medium VEGF concentration increased time-dependently after amlodipine (P<0.01) treatment, but not after losartan (P>0.01), or lisinopril (P>0.01). Amlo- dipine, but no other AHM, increased the serum VEGF concentration (P<0.05) during the a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Kumar A, Mutter S, Parente EB, Harjutsalo V, Lithovius R, Mathavan S, et al. (2023) L-type calcium channel blocker increases VEGF concentrations in retinal cells and human serum. PLoS ONE 18(4): e0284364. https://doi.org/ 10.1371/journal.pone.0284364 Editor: Satyajit Mohapatra, SRM Medical College Hospital and Research Centre, INDIA Received: May 18, 2022 Accepted: March 18, 2023 Published: April 13, 2023 Copyright: © 2023 Kumar et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: VEGF measurements from the GENRES study and all data for the participants of the observational FinnDiane study are not publicly available and available only under restricted access due to ethical and legal reasons and due to the consent provided by the participant at the time of data collection. The data access, which is subject to local regulations, can be obtained upon reasonable request by contacting: Maaria Puupponen (email: maaria. puupponen@helsinki.fi), Research Program Coordinator, Clinical and Molecular Metabolism PLOS ONE \| https://doi.org/10.1371/journal.pone.0284364 April 13, 2023 1 / 13 PLOS ONE(CAMM), University of Helsinki. Upon approval, analysis needs to be performed on a user-specific local server (with protected access) and requires the applicant to sign non-disclosure and secrecy agreements. Cell culture experimental data is available as supplementary material. Funding: The GENRES Study was supported by The Sigrid Juselius Foundation and The Finnish Foundation for Cardiovascular Research.The FinnDiane study was supported by grants from Folkha¨lsan Research Foundation; Wilhelm och Else Stockmanns Stiftelse; Academy of Finland (grants 299200, 275614, and 316664); Liv och Ha¨lsa Society; Novo Nordisk Foundation (grant NNFOC0013659); Medical Society of Finland (Finska La¨karesa¨llskapet); Dorothea Olivia, Karl Walter and Jarl Walter Perklen Foundation; Pa¨ivikki and Sakari Sohlberg Foundation; Sigrid Juselius Foundation; Finnish Foundation for Cardiovascular Research; and Helsinki University Hospital Research Funds (EVO). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: EBP reports receiving lecture honorariums from Eli Lilly, Abbott, Astra Zeneca, Sanofi, Boehringer Ingelheim and is an advisory board member of Sanofi. PHG reports receiving lecture honorariums from Astellas, Astra Zeneca, Boehringer Ingelheim, Eli Lilly, Elo Water, Medscape, MSD, Mundipharma, Novo Nordisk, PeerVoice, Sanofi, Sciarc, and being an advisory board member of Astellas, Astra Zeneca, Bayer, Boehringer Ingelheim, Eli Lilly, Janssen, Medscape, MSD, Mundipharma, Novo Nordisk, and Sanofi. This does not alter our adherence to PLOS ONE policies on sharing data and materials. AK, SM, TH, VH, RL, SiMa, CF, ML and KK declare no conflict of interest. L-type calcium channel blocker and VEGF concentrations interventional clinical study. The usage of LTCCB was not associated with the risk of SDED in the observational study. Conclusions LTCCB increases VEGF concentrations in retinal cells and human serum. However, the usage of LTCCBs does not appear to be associated with SDED in adults with type 1 diabetes. Introduction Diabetic retinopathy (DR) is a devastating complication among individuals with diabetes and may lead to permanent vision loss. Given the increasing number of individuals with diabetes worldwide, the prevalence of DR is expected to rise significantly in the coming years [1]. Hyperpermeability, hypoperfusion, and neo-angiogenesis of the intraocular microvasculature are key hallmarks of DR and will ultimately lead to anatomical and pathophysiological alter- ations [2]. Vascular endothelial growth factor (VEGF) is a well-known angiogenic factor in the eye, and is also a major pharmacological target for the treatment of severe and proliferative diabetic retinopathy (PDR) [3, 4]. Although, intraocular injections of anti-VEGF antibodies have emerged as a novel and effective pharmacological treatment, long-term efficacy and systemic safety data are lacking, stressing the need to explore alternative pathways and molecular targets for intervention [5]. Recently, we reported mutations in the CACNB2 gene encoding the L- type calcium channel’s β-subunit. These mutations contributed to the severity of PDR. Fur- thermore, after down-regulating the β-subunit coding CACNB2 gene by RNA interference, the concentration of VEGF was significantly reduced in retinal cells [6]. Notably, a network meta-analysis showed that renin-angiotensin-aldosterone system inhib- itors (RAASi) were associated with reduced risk of progression and increased the possibility of regression of DR [7]. However, although the calcium channel blockers seemed to worsen the outcome compared to placebo, it was not statistically significant [7]. L-type calcium channel blockers (LTCCB) such as amlodipine have been widely prescribed to treat arterial hyperten- sion as monotherapy or in combination with other classes of antihypertensive medications (AHM) including angiotensin-converting enzyme inhibitors (ACEi) such as lisinopril, angio- tensin II receptor blockers (ARB) such as losartan, β blockers such as bisoprolol and diuretics such as hydrochlorothiazide [8, 9]. However, the relationship between LTCCB and DR is still poorly understood. Therefore, in this work, we aimed to explore the effect of LTCCBs in comparison to other AHMs on the VEGF concentrations in retinal cell culture and serum of human subjects. We used human retina-derived Mu¨ller cells for in vitro experiments as these cells are a crucial source of VEGF in the pathogenesis of DR [10, 11]. In our experiments, we evaluated the effects of amlodipine, a commonly used LTCCB, alongside drugs of other subclasses of AHMs on the serum VEGF concentrations in in vitro experiments and in a clinical trial setting (The GENRES Study). These drugs were chosen as each of them represents a typical medication of the respective antihypertensive subclass, each has a favourable profile in terms of effectiveness versus side effects and each has been widely used as an antihypertensive agent in clinical and experimental settings. Finally, we evaluated the association between the use of LTCCB and the risk of severe diabetic eye disease (SDED) in adults with type 1 diabetes from the Finnish Dia- betic Nephropathy Study (FinnDiane) cohort. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284364 April 13, 2023 2 / 13 PLOS ONEL-type calcium channel blocker and VEGF concentrations Research design and methods Retinal cell culture study Mu¨ller glia or Mu¨ller cell lines, MIO-M1 were obtained from Prof. Astrid Limb [12]. MIO-MI cells were a gift from Prof. Astrid Limb, Ocular Repair and Regeneration Biology Unit, Depart- ments of Cell Biology and Pathology, Institute of Ophthalmology and Moorfields Eye Hospital, London, United Kingdom; Prof. Limb has ethics committee approvals from local health authority, eyes consented for research were obtained from Moorfields Hospital Eye Bank, Lon- don, United Kingdom to generate MIO-M1 cells in her laboratory. Cells were grown in Dul- becco’s Modified Eagle Medium (41965–039, Gibco/life technology), 10% Fetal Bovine Serum (10270106, Gibco/life technology), Penicillin-Streptomycin (15140122, Gibco/life technology) and Normocin™ (ant-nr-1, InvivoGen) in a humidified 5% CO2 cell culture incubator at +37˚C temperature. Cells were plated in 6-well plates (140675, Nunc™) with seeding density (~106 cells/well). Three medications—amlodipine besylate (A5605, Sigma-Aldrich), losartan potassium (phr1602-1g, Sigma-Aldrich) and lisinopril (phr1143-1g, Sigma-Aldrich) were dis- solved in solvents as per the manufacturer’s recommendation. Stock solution of 17.6 mM in DMSO (dimethyl sulfoxide) was prepared for amlodipine, 20 mM stock solution in 1XPBS for losartan and 25 mM stock solution for lisinopril in 1X PBS based on their solubility character- istics and chemical properties. The final working concentration (1–10 μM) of the drugs were achieved by further diluting the stock solution in DMEM supplemented with 0.02% bovine serum albumin (BSA) and antibiotics (Penicillin-Streptomycin- Normocin™). The medications and the dimethyl sulfoxide (DMSO)/1X Phosphate-buffered saline (PBS) control (the solvent used to dissolve the medication) were added in equal volume to the cell culture medium. The cell culture medium containing medications and control in equal volume were changed every 24 hours until the end of the experiment. Cell culture medium was collected and centrifuged at +8˚C temperature at 3000 rpm for 5 minutes to remove cell debris, and the resulting super- natants were collected and stored in a -80˚C freezer for further analysis. Interventional clinical study The GENRES Study is a randomized, double-blind, placebo-controlled, crossover study, which included Finnish men with moderate hypertension but without drug-treated diabetes. A total of 313 hypertensive Finnish men (aged 35 to 60 years) were initially recruited. The clinical study was completed by 228 subjects and blood serum samples (collected in the morning under non-fasting conditions and stored at -80˚C without any incidental thawing) were available for VEGF measure- ment assays from 39 subjects. The GENRES Study was approved by the Ethics Committee of Hel- sinki University Central Hospital. The clinical part of the study was carried out in accordance with the Declaration of Helsinki and Guidelines for Good Clinical Practice. All subjects gave a signed informed consent prior to study activities. The study is registered at ClinicalTrials.gov (NCT03276598). The detailed protocol can be found in a previous publication [13]. In brief, the GENRES study started with a four-week run-in placebo period, before which the participants dis- continued their previous antihypertensive medication, in case they were on medication. Then, participants were randomized into different sequences of four-week treatment periods taking one of the AHM (amlodipine, bisoprolol, hydrochlorothiazide, losartan) at a time. Thus, each partici- pant underwent four separate monotherapy periods, separated by four-week placebo periods [13]. Serum VEGF concentration was measured at eight different time points, at the end of the four placebo and the four drug treatment periods. Notably, the serum at the end of the first placebo period was not included in the analysis, since the blood sample was taken after an overnight fast as opposed to the other blood samples, which were taken in the non-fasting state. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284364 April 13, 2023 3 / 13 PLOS ONEL-type calcium channel blocker and VEGF concentrations VEGF measurements VEGF measurements were performed double-blinded in serum and cell culture media using the Human VEGF Quantikine ELISA Kit (DVE00 & SVE00, R&D systems/bio-techne) according to the manufacturer’s protocol. Observational cohort study The outcome was SDED, defined as a composite outcome including proliferative diabetic reti- nopathy (PDR), initiation of laser treatment or anti-VEGF therapy, diabetic maculopathy, vit- reous hemorrhage and vitrectomy identified from the Finnish Care Register for Health Care until the end of 2015. Information on purchases of AHMs was obtained from the Finnish Drug Prescription Reg- ister (maintained by the National Social Insurance Institution since 1994, which contains information on all prescribed, purchased, and reimbursed medications in outpatient care) until the end of 2015. Medications were coded according to the Anatomic Therapeutic Chemi- cal (ATC) classification system. The renin-angiotensin-aldosterone system inhibitors (RAASi) were identified by an ATC code starting with C09 including both ACEis and ARBs. To identify individuals on RAASi monotherapy, we excluded individuals with a prescription record of combination products that contained other AHMs in the same pill (coded with C09B and C09D) or with a prescription for any other AHMs prior to the baseline or during the follow- up. LTCCBs in the register included the following ATC codes: C07FB02, C07FB03, C08CA01, C08CA02, C08CA03, C08CA05, C08CA07, C08CA10, C08CA13, C09BB02, C09BB04, and C09BB05. Refill adherences were calculated using the so-called proportion of days covered method with a prescription refill interval of 90 days. For this longitudinal study, 192 individuals were selected from the FinnDiane study, which is an ongoing, nationwide, multicenter cohort of adults with type 1 diabetes in Finland [14, 15]. Type 1 diabetes was defined as age of diabetes onset below 40 years and requiring insulin within the first year after diagnosis. Currently, more than 5,400 individuals have undergone a detailed clinical examination at their first FinnDiane visit. We excluded 1,933 individuals due to SDED at baseline. These data (history of laser treatment) were registered based on medical records at baseline and obtained by the attending physician or nurse using a standardized form. The selection criteria for the remaining 3,468 individuals can be found in the S1 Fig. First, we required the study visit to take place at least half a year before the end of the prescrip- tion records at the end of 2015 and no SDED in the first half year of follow-up. As shown in S1 Fig, our cohort is made up of two groups: those on LTCCB plus RAASi and other AHM and a control group of those on RAASi as monotherapy. In both groups individuals had been on RAASi therapy at baseline, defined as either already being on RAASi prior to the study visit or starting RAASi therapy no longer than half a year after the visit. We further required at least 2 purchases of RAASi during the follow-up. The same selection criteria were applied for LTCCB in the group LTCCB plus RAASi and other AHM. In this latter group, individuals were allowed to use other AHMs, because, according to the hypertension guidelines, LTCCBs are usually considered as the second or third line drug after RAASi [16, 17]. Due to the strict adherence to the guidelines in Finland, we did not find enough individuals using LTCCB as monotherapy (n = 11) to be able to investigate them as a separate group. All subjects partici- pating in the study have been informed about the study and have given their informed consent before participation. The participants also consented to obtain and utilize the health registry data. We have permission Dnro THL/1519/5.05.00/2013 for the data from the Finnish Care Register for Health Care and Dnro Kela/22/522/2013 for the Finnish Drug Prescription Regis- ter data and the updated permission Dnro THL/207/14.02.00/2021 for both registers. The PLOS ONE \| https://doi.org/10.1371/journal.pone.0284364 April 13, 2023 4 / 13 PLOS ONEL-type calcium channel blocker and VEGF concentrations study follows the General Data Protection Regulation (EU GDPR). The study protocol was approved by the Ethical Committee of Helsinki and Uusimaa Hospital District (HUS/3318/ 2018). Statistical analysis Cell culture data were analyzed using nonparametric Mann-Whitney U test (S1 Table). A Bon- ferroni corrected p-value below 0.0125 was considered significant in the cell culture experi- ments. Fold changes in the VEGF concentration are calculated by dividing the concentration at a timepoint by the mean control concentration. For data collection for the statistical analy- sis, in vitro experiments were repeated under similar experimental settings with 2–3 technical repeats/cell culture wells for each treatments. In the analysis of the clinical study (The GENRES Study) results, the VEGF concentrations after the drug treatment periods were compared pairwise with the baseline VEGF concentra- tions (the mean values after the three placebo periods) for each subject. Since VEGF concen- trations at baseline and changes after the drug treatment periods were non-normally distributed, the nonparametric Wilcoxon signed ranks test was used to test the statistical sig- nificance of the drug-induced changes. The primary target variable in these analyses was the change caused by amlodipine treatment and the other drug treatments were analysed for their drug-specificity of the effect. Therefore, no correction for multiple comparisons was applied and a p-value below 0.05 was considered statistically significant. In the observational study, a multivariable logistic regression analysis was used to study the associations between the usage of LTCCB plus RAASi and AHM (n = 62) and the incidence of SDED as the outcome. Since RAASi usage has shown an association with reduced risk of pro- gression and increased possibility of regression of DR [7], first, we performed a logistic regres- sion using the adherence to RAASi usage as the independent variable and the incidence of SDED (n = 130) as the outcome. Since the adherence to RAASi usage was associated with the outcome, this variable was included as a covariate in the logistic regression model for the anal- ysis of LTCCB plus RAASi and AHM and incident SDED alongside the presence of any reti- nopathy other than SDED at baseline and clinical baseline variables that were significantly different between those that developed SDED and those that did not (Table 1). Therefore, the full set of covariates in the multivariable model was age, age of diabetes onset, presence of albu- minuria, HbA1c, adherence to RAASi treatment during follow-up, other antihypertensive medication on top of LTCCB plus RAASi during follow-up and the presence of any retinopa- thy other than SDED at baseline. For continuous variables, p-values were calculated using t- tests for normally distributed variables and Mann-Whitney U-tests for non-normally distrib- uted variables. For categorical variables, we used the χ2-test. P-value below 0.05 was consid- ered significant. All analyses on observational data were performed in R version 3.6 [18]. Results Retinal cell culture study After treating the MIO-M1 cells with 10 μM amlodipine at 24, 48, 72 and 96 hours, the treat- ment increased the secreted VEGF protein in the medium compared to each respective control treatment (Fig 1B). In contrast, the secreted VEGF protein in the medium did not change after lisinopril or losartan treatments at any time points compared to each respective control treatment (S2A and S2B Fig). Additionally, we tested lower concentrations of amlodipine (1 μM and 5 μM), but they did not affect the VEGF concentration in the medium of the cell culture at 24, 48, 72 and 96 hours PLOS ONE \| https://doi.org/10.1371/journal.pone.0284364 April 13, 2023 5 / 13 PLOS ONEL-type calcium channel blocker and VEGF concentrations Table 1. Characteristics of individuals with type 1 diabetes participating in the observational study. No SDED Incident SDED P-value N LTCCB (%) Other AHM (%) Women (%) Age (years) Age of diabetes onset (years) Diabetes duration (years) Systolic blood pressure (mmHg) Diastolic blood pressure (mmHg)))) ´)mmHg)mmHg(mmHg) Total cholesterol (mmol/L) Triglycerides (mmol/L) eGFR (mL/min/1.73m2) Albuminuria (%) HbA1c (%) HbA1c (mmol/mol) RAASi before baseline (years) Adherence to RAASi (%) Any retinopathy except SDED (%) 122 32.8 31.1 41.0 42.82 ± 10.65 18.34 ± 8.10 24.50 ± 11.75 141 ± 18 80 ± 10 4.00 ± 1.12 0.03 (0.69, 1.33) 93 ± 28 31.1 8.2 ± 1.3 66 ± 14 3.04 ± 2.77 74.94 ± 17.20 52.6 70 31.4 22.9 38.6 36.30 ± 11.93 12.92 ± 7.72 23.38 ± 9.71 137 ± 17 81 ± 10 4.89 ± 1.04 1.08 (0.81, 1.40) 96 ± 29 70.0 9.3 ± 2.0 78 ± 22 2.79 ± 2.31 70.05 ± 20.58 67.7 0.97 0.29 0.86 <0.001 <0.001 0.50 0.10 0.43 0.93 0.08 0.43 <0.001 <0.001 <0.001 0.53 0.08 0.07 Data are shown as mean ± SD or median (interquartile range). SDED, severe diabetic eye disease; LTCCB, L-type calcium channel blocker; RAASi, renin-angiotensin- aldosterone system inhibitors; AHM, antihypertensive medication other than LTCCB or RAASi, eGFR, estimated glomerular filtration rate; HbA1c, glycated haemoglobin A1c. https://doi.org/10.1371/journal.pone.0284364.t001 (S3 Fig). Finally, we performed cell viability tests at the end of our experiment (96 hours). Amlodipine (10 μM) reduced cell proliferation, whereas losartan (10 μM) and lisinopril (10 μM) did not affect the cell viability or proliferation compared to the control (DMSO/ 1XPBS) (S4A and S4B Fig). Interventional clinical study During the placebo periods, the median serum VEGF concentration was 277 pg/mL in 39 men with hypertension. During the LTCCB treatment period, there was a median increase of 18 pg/ml in the serum VEGF concentration (p-value = 0.03). Treatment with hydrochlorothiazide induced a nonsignificant increase (p-value = 0.09) while treatment with losartan or bisoprolol caused no significant changes in the VEGF concentration (Fig 2B). Observational study During a mean follow up of 8.6 ± 5.7 years, 70 events of SDED occurred. Individuals who developed SDED had a younger age at onset of diabetes and higher glycated hemoglobin (HbA1c) at baseline than those that did not develop SDED. The clinical characteristics of the individuals at baseline, according to the incidence of SDED, are shown in Table 1. The higher refill adherence to RAASi was associated with a lower risk of SDED (OR 0.97, 95% CI ☯0.95– 0.99], p-value = 0.005) in an unadjusted model. The use of LTCCB plus RAASi and other AHM was neither associated with increased odds of SDED (OR 0.94, 95% CI [0.50–1.77], p- value = 0.85) in the unadjusted model nor in the adjusted model (OR 6.12, 95% CI [0.88– 42.45], p-value = 0.07) (S2 Table). PLOS ONE \| https://doi.org/10.1371/journal.pone.0284364 April 13, 2023 6 / 13 PLOS ONEL-type calcium channel blocker and VEGF concentrations Discussion In the present study, we showed that amlodipine, an LTCCB, increases the secretion of VEGF in vitro by cells of human retinal origin, in a time-dependent manner. In addition, we showed that four weeks of amlodipine usage also increases the serum VEGF concentration in individu- als with arterial hypertension. However, in our observational study, the use of LTCCB was not associated with an increased risk of SDED in adults with type 1 diabetes. Fig 1. A: Schematic diagram showing the experimental setup and the sample collection timeline. B: Vertical bars represent the mean and the standard deviation of fold changes in the VEGF concentrations in the MIO-M1 cell culture medium at different time points (24 hours, 48 hours, 72 hours and 96 hours) after treatment with amlodipine (10 μM) vs control. Fold changes in the VEGF concentration are calculated by dividing the concentration at a timepoint by the mean control concentration of 125.8765 pg/ml. P-value = 0.007937. https://doi.org/10.1371/journal.pone.0284364.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284364 April 13, 2023 7 / 13 PLOS ONEL-type calcium channel blocker and VEGF concentrations Fig 2. A: Schematic diagram showing the interventional clinical study design and the sample collection timeline. B: Median and interquartile range changes in the serum VEGF concentration after 4 weeks of daily monotherapies with amlodipine (5 mg), bisoprolol (5 mg), hydrochlorothiazide (25 mg) or losartan (50 mg) compared to the wash-out phases. The nonparametric Wilcoxon signed ranks test was used to test the statistical significance of the drug-induced changes in serum VEGF. P-value < 0.05 denotes statistical significance. https://doi.org/10.1371/journal.pone.0284364.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284364 April 13, 2023 8 / 13 PLOS ONEL-type calcium channel blocker and VEGF concentrations Our observations are supported by a previous report showing that nifedipine, a first-generation LTCCB, induces VEGF secretion in human coronary artery smooth muscle cells (HCSMCs) [19]. We used 24–96 hours of treatment in contrast to 18–24 hours in previous in vitro studies [19, 20]. This prolonged in vitro treatment may better represent a chronic pharmacological intervention [19]. Additionally, we observed an anti-proliferative effect of amlodipine on MIO-M1 cells, similar to what has been shown with nifedipine on HCSMCs [19]. Another study has reported no effect of amlodipine on VEGF secretion by HCSMCs. However, the shorter treatment duration and lower medication dosage (5 μM) used in that study may explain the observed differences com- pared to our results [20]. To reinforce this hypothesis, we also tested 1 μM and 5 μM of amlodi- pine on MIO-M1 cells, and we did not find any changes in the VEGF concentration at any time of the treatment (24–96 hours). Importantly, DR disrupts the blood-retinal barrier increasing per- fusion of medications into the ocular compartment, thus altering intraocular pharmacokinetics and diffusion of medication molecules [21, 22]. This may increase the diffusion of L-type calcium channel blockers from the systemic circulation into the ocular compartments in individuals with DR and thereby elevating their intraocular concentrations. Our data from the interventional clini- cal study show that the serum VEGF concentration increases after four weeks of amlodipine treat- ment (5 mg/day). Although a previous study reported a positive association between serum and vitreous VEGF levels [23, 24], we are not able to conclude from our results whether an elevated serum VEGF concentration also translates into an increased VEGF concentration in the vitreous humour. Furthermore, it is not possible to ensure that the increased serum VEGF concentration after four weeks of amlodipine usage will persist after years of treatment. Further in vivo experi- ments using animal models are needed to clarify these questions, since collecting vitreous samples from human subjects is challenging. The observational study showed that long-term usage of LTCCB plus RAASi and other AHM was not associated with SDED. We are not sure whether the lack of association is due to the combination of LTCCB with RAASi, since RAASi have been linked to lower risk of DR [7, 9]. However, we included refill adherence to RAASi therapy as a covariate in our models. In our study, it was not possible to evaluate the risk of SDED related to the use of LTCCB as monotherapy, because the number of individuals using LTCCB without other medication in the FinnDiane cohort was too small. Indeed, this is aligned with the hypertension treatment guidelines in people with diabetes [16, 17] in which the calcium channel blockers are not the first-line choice and are commonly prescribed in association with RAASi as second or third- line therapy. Although we did not find an association between the use of LTCCB and SDED, we found that a higher adherence to RAASi was associated with lower odds of SDED, similar to what has been previously shown [7]. The major strength of the current study is the in vitro experiments with retina-derived Mu¨ller cells showing for the first time a time-dependent effect of LTCCB on VEGF concentrations. Addi- tionally, we validated these in vitro findings using a unique interventional clinical trial that showed an increase in serum VEGF after four weeks of exposure to LTCCB. The main drawback of our study is that we were not able to measure VEGF in the vitreous humour to evaluate the correlation between the serum and the vitreous VEGF concentrations. Furthermore, despite the inclusion of one of the largest cohorts of individuals with type 1 diabetes (the FinnDiane Study), there was not a large enough number of individuals using LTCCB as monotherapy in order to be able to explore a potential adverse association between the use of LTCCB and SDED. Conclusions Our results show that LTCCB increases VEGF concentrations in retinal origin cells and in human serum, but its usage in combination with RAASi and other AHM does not seem to be PLOS ONE \| https://doi.org/10.1371/journal.pone.0284364 April 13, 2023 9 / 13 PLOS ONEL-type calcium channel blocker and VEGF concentrations associated with SDED in adults with type 1 diabetes in an observational setting. Further studies including a larger sample of individuals with other types of diabetes using LTCCB as mono- therapy, as well as studies with animal models to measure the VEGF in the vitreous humour are warranted to evaluate the impact of chronic use of LTCCBs on the development and pro- gression of DR. Supporting information S1 Fig. The selection criteria for observational study cohort. Schematic diagram depicting selection criteria in the observational study with FinnDiane cohort. (TIF) S2 Fig. A. Effects on VEGF concentrations in the MIO-M1 cells after treatment with losartan. Vertical bars represent the mean and the standard deviation of fold changes in the VEGF con- centrations in the MIO-M1 cell culture medium at different time points (24 hours, 48 hours, 72 hours and 96 hours) after treatment with losartan (10 μM) vs control. Fold changes are cal- culate by dividing a concentration at a time point by the mean control concentration of 92.04108 pg/ml. NS = Not Significant (Mann-Whitney U test). B. Effects on VEGF concentra- tions in the MIO-M1 cells after treatment with lisinopril. Vertical bars represent the mean and the standard deviation of fold changes in the VEGF concentrations in the MIO-M1 cell culture medium at different time points (24 hours, 48 hours, 72 hours and 96 hours) after treatment with lisinopril (10 μM) vs control. Fold changes are calculate by dividing a concentration at a time point by the mean control concentration of 100.1139 pg/ml. NS = Not Significant (Mann-Whitney U test). (TIF) S3 Fig. Dose-response determination for amlodipine concentrations and time points to measure VEGF in cell culture media. Dose-response (1, 5 and 10 μM) pilot experiment for determination of effects of amlodipine on VEGF secretion in MIO-M1 cells and to find out an optimal experimental setting for amlodipine concentrations and time points to measure VEGF in cell culture media. Values for VEGF concentrations are calculated as mean of three technical repeats for every treatment from a single experiment. (TIF) S4 Fig. A. MIO-M1 cells viability determination. Survival of MIO-M1 cells at end of drug treatment (96 hours), based on intracellular ATP concentration. Survival mean percentages are calculated from two independent experiments. Vertical lines represent standard deviation. B. MIO-M1 cells viability, DMSO vs 1xPBS treatment. Luminescence [quantified as Relative Light Units (RLU)] based assay to measure intracellular ATP indicating general cell health. Our data show no statistically significant difference in the intracellular ATP concentration between DMSO and 1xPBS in MIO-M1 cells; 96 hours post treatment (End of experiment). Vertical lines represent standard deviation. (TIF) S1 Table. Detailed cell culture experimental data and statistics. Detailed information about statistical analysis and calculations with cell culture experimental data. (XLSX) S2 Table. Additional information from observational study. Detailed information about sta- tistical analysis and models used in observational study. (XLSX) PLOS ONE \| https://doi.org/10.1371/journal.pone.0284364 April 13, 2023 10 / 13 PLOS ONEL-type calcium channel blocker and VEGF concentrations S1 Appendix. FinnDiane study centers. List of FinnDiane study centers across Finland. (DOCX) Acknowledgments We acknowledge the important contributions of Carol Forsblom in the design and execution of this work. Unfortunately, he passed away before the work was completed. The skilled tech- nical assistance by Heli Krigsman and Susanna Saarinen, is gratefully acknowledged. The authors also acknowledge the participants, physicians and nurses at FinnDiane study centers (see S1 Appendix). Parts of this study have been presented in abstract form at the 56th Annual Meeting of the European Association for the Study of Diabetes. Author Contributions Conceptualization: Anmol Kumar. Data curation: Anmol Kumar, Stefan Mutter, Valma Harjutsalo, Timo P. Hiltunen. Formal analysis: Anmol Kumar, Stefan Mutter, Erika B. Parente, Timo P. Hiltunen. Funding acquisition: Kimmo K. Kontula, Per-Henrik Groop. Investigation: Anmol Kumar, Stefan Mutter, Erika B. Parente, Timo P. Hiltunen. Methodology: Anmol Kumar, Stefan Mutter, Erika B. Parente, Raija Lithovius, Timo P. Hiltunen. Project administration: Anmol Kumar. Resources: Valma Harjutsalo, Raija Lithovius, Sinnakaruppan Mathavan, Per-Henrik Groop. Software: Valma Harjutsalo. Supervision: Markku Lehto, Kimmo K. Kontula. Validation: Anmol Kumar. Visualization: Anmol Kumar. Writing – original draft: Anmol Kumar, Stefan Mutter, Erika B. Parente, Timo P. Hiltunen. Writing – review & editing: Anmol Kumar, Erika B. Parente, Sinnakaruppan Mathavan, Markku Lehto, Timo P. Hiltunen, Kimmo K. Kontula, Per-Henrik Groop. References 1. Yau JW, Rogers SL, Kawasaki R, Lamoureux EL, Kowalski JW, Bek T, et al. Global prevalence and major risk factors of diabetic retinopathy. Diabetes Care. 2012; 35(3):556–64. https://doi.org/10.2337/ dc11-1909 PMID: 22301125 2. Kusuhara S, Fukushima Y, Ogura S, Inoue N, Uemura A. Pathophysiology of Diabetic Retinopathy: The Old and the New. Diabetes Metab J. 2018; 42(5):364–76. https://doi.org/10.4093/dmj.2018.0182 PMID: 30362302 3. Mehanna CJ, Abdul Fattah M, Haddad S, Tamim H, Ghazi N, Salti H. Anti-VEGF Therapy for Persistent Neovascularization after Complete Panretinal Photocoagulation in Proliferative Diabetic Retinopathy. Ophthalmol Retina. 2019; 3(6):473–7. https://doi.org/10.1016/j.oret.2019.02.001 PMID: 31174668 4. Witmer AN, Vrensen GF, Van Noorden CJ, Schlingemann RO. Vascular endothelial growth factors and angiogenesis in eye disease. Prog Retin Eye Res. 2003; 22(1):1–29. https://doi.org/10.1016/s1350- 9462(02)00043-5 PMID: 12597922 5. Simo´ R, Sundstrom JM, Antonetti DA. Ocular Anti-VEGF therapy for diabetic retinopathy: the role of VEGF in the pathogenesis of diabetic retinopathy. Diabetes Care. 2014; 37(4):893–9. https://doi.org/ 10.2337/dc13-2002 PMID: 24652720 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284364 April 13, 2023 11 / 13 PLOS ONEL-type calcium channel blocker and VEGF concentrations 6. Vuori N, Sandholm N, Kumar A, Hietala K, Syreeni A, Forsblom C, et al. CACNB2 Is a Novel Suscepti- bility Gene for Diabetic Retinopathy in Type 1 Diabetes. Diabetes. 2019; 68(11):2165–74. https://doi. org/10.2337/db19-0130 PMID: 31439644 7. Wang B, Wang F, Zhang Y, Zhao SH, Zhao WJ, Yan SL, et al. Effects of RAS inhibitors on diabetic reti- nopathy: a systematic review and meta-analysis. Lancet Diabetes Endocrinol. 2015; 3(4):263–74. https://doi.org/10.1016/S2213-8587(14)70256-6 PMID: 25660574 8. Chamberlain JJ, Johnson EL, Leal S, Rhinehart AS, Shubrook JH, Peterson L. Cardiovascular Disease and Risk Management: Review of the American Diabetes Association Standards of Medical Care in Diabetes 2018. Ann Intern Med. 2018; 168(9):640–50. https://doi.org/10.7326/M18-0222 PMID: 29610837 9. Do DV, Wang X, Vedula SS, Marrone M, Sleilati G, Hawkins BS, et al. Blood pressure control for dia- betic retinopathy. Cochrane Database Syst Rev. 2015; 1:Cd006127. https://doi.org/10.1002/14651858. CD006127.pub2 PMID: 25637717 10. McDowell RE, Barabas P, Augustine J, Chevallier O, McCarron P, Chen M, et al. Mu¨ller glial dysfunc- tion during diabetic retinopathy in rats is reduced by the acrolein-scavenging drug, 2-hydrazino-4,6- dimethylpyrimidine. Diabetologia. 2018; 61(12):2654–67. 11. 12. Tien T, Zhang J, Muto T, Kim D, Sarthy VP, Roy S. High Glucose Induces Mitochondrial Dysfunction in Retinal Mu¨ ller Cells: Implications for Diabetic Retinopathy. Invest Ophthalmol Vis Sci. 2017; 58 (7):2915–21. Lawrence JM, Singhal S, Bhatia B, Keegan DJ, Reh TA, Luthert PJ, et al. MIO-M1 cells and similar muller glial cell lines derived from adult human retina exhibit neural stem cell characteristics. Stem Cells. 2007; 25(8):2033–43. https://doi.org/10.1634/stemcells.2006-0724 PMID: 17525239 13. Hiltunen TP, Suonsyrja¨ T, Hannila-Handelberg T, Paavonen KJ, Miettinen HE, Strandberg T, et al. Pre- dictors of antihypertensive drug responses: initial data from a placebo-controlled, randomized, cross- over study with four antihypertensive drugs (The GENRES Study). Am J Hypertens. 2007; 20(3):311–8. https://doi.org/10.1016/j.amjhyper.2006.09.006 PMID: 17324745 14. 15. Thorn LM, Forsblom C, Fagerudd J, Thomas MC, Pettersson-Fernholm K, Saraheimo M, et al. Meta- bolic syndrome in type 1 diabetes: association with diabetic nephropathy and glycemic control (the Finn- Diane study). Diabetes Care. 2005; 28(8):2019–24. https://doi.org/10.2337/diacare.28.8.2019 PMID: 16043748 Jansson FJ, Forsblom C, Harjutsalo V, Thorn LM, Wade´n J, Elonen N, et al. Regression of albuminuria and its association with incident cardiovascular outcomes and mortality in type 1 diabetes: the Finn- Diane Study. Diabetologia. 2018; 61(5):1203–11. https://doi.org/10.1007/s00125-018-4564-8 PMID: 29423580 16. Cosentino F, Grant PJ, Aboyans V, Bailey CJ, Ceriello A, Delgado V, et al. 2019 ESC Guidelines on dia- betes, pre-diabetes, and cardiovascular diseases developed in collaboration with the EASD: The Task Force for diabetes, pre-diabetes, and cardiovascular diseases of the European Society of Cardiology (ESC) and the European Association for the Study of Diabetes (EASD). European Heart Journal. 2019; 41(2):255–323. 17. Thomas Unger CB, Fadi Charchar, Nadia A Khan, Neil R. Poulter, Dorairaj Prabhakaran, Agustin Ramirez, et al. Schutte. 2020 International Society of Hypertension Global Hypertension Practice Guidelines. Hypertension 2020; 75:1334–57. 18. Team RC. R: A language and environment for statistical computing. 2020. 19. Miura S, Fujino M, Matsuo Y, Tanigawa H, Saku K. Nifedipine-induced vascular endothelial growth fac- tor secretion from coronary smooth muscle cells promotes endothelial tube formation via the kinase insert domain-containing receptor/fetal liver kinase-1/NO pathway. Hypertens Res. 2005; 28(2):147– 53. https://doi.org/10.1291/hypres.28.147 PMID: 16025742 20. Kawamura A, Miura SI, Matsuo Y, Tanigawa H, Saku K. Azelnidipine, Not Amlodipine, Induces Secre- tion of Vascular Endothelial Growth Factor From Smooth Muscle Cells and Promotes Endothelial Tube Formation. Cardiol Res. 2014; 5(5):145–50. https://doi.org/10.14740/cr352w PMID: 28348712 21. Jo DH, Yun JH, Cho CS, Kim JH, Kim JH, Cho CH. Interaction between microglia and retinal pigment epithelial cells determines the integrity of outer blood-retinal barrier in diabetic retinopathy. Glia. 2019; 67(2):321–31. https://doi.org/10.1002/glia.23542 PMID: 30444022 22. Durairaj C. Ocular Pharmacokinetics. Handb Exp Pharmacol. 2017; 242:31–55. https://doi.org/10. 1007/164_2016_32 PMID: 27787714 23. Baharivand N, Zarghami N, Panahi F, Dokht Ghafari MY, Mahdavi Fard A, Mohajeri A. Relationship between vitreous and serum vascular endothelial growth factor levels, control of diabetes and microal- buminuria in proliferative diabetic retinopathy. Clin Ophthalmol. 2012; 6:185–91. https://doi.org/10. 2147/OPTH.S27423 PMID: 22331976 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284364 April 13, 2023 12 / 13 PLOS ONEL-type calcium channel blocker and VEGF concentrations 24. Wang J, Chen S, Jiang F, You C, Mao C, Yu J, et al. Vitreous and plasma VEGF levels as predictive fac- tors in the progression of proliferative diabetic retinopathy after vitrectomy. PLoS One. 2014; 9(10): e110531. https://doi.org/10.1371/journal.pone.0110531 PMID: 25329921 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284364 April 13, 2023 13 / 13 PLOS ONE
10.1371_journal.pone.0277927	RESEARCH ARTICLE The oldest plans to scale of humanmade mega-structures Re´ my CrassardID Frank Preusser4, Hamida Seba5, Abd Errahmane Kiouche5, Emmanuelle Re´ gagnon1, Juan Antonio Sa´ nchez PriegoID 1, Thamer Almalki6, Mohammad Tarawneh7 1☯, Wael Abu-Azizeh1,2☯, Olivier BargeID 1☯, Jacques E´ lie Brochier3, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Crassard R, Abu-Azizeh W, Barge O, Brochier JE´, Preusser F, Seba H, et al. (2023) The oldest plans to scale of humanmade mega- structures. PLoS ONE 18(5): e0277927. https://doi. org/10.1371/journal.pone.0277927 Editor: Andrea Zerboni, Universita degli Studi di Milano, ITALY Received: June 28, 2022 Accepted: November 6, 2022 Published: May 17, 2023 Copyright: © 2023 Crassard et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting information files. In Jordan, the engraved slab from Jibal al- Khashabiyeh is stored at the French Institute for the Near East (Ifpo, Amman) and is expected to be soon housed in the Hussein Bin Talal University exhibition room of the South-Eastern Badia Archaeological Project in Wadi Musa. In Saudi Arabia, the engraved boulder at Jebel az-Zilliyat has not been removed from its original location. Funding: The South-Eastern Badia Archaeological Project (SEBAP; research at Jibal al-Khashabiyeh) 1 CNRS, Arche´ orient, UMR 5133, Maison de l’Orient et de la Me´diterrane´ e, Universite´ Lyon 2, Lyon, France, 2 MEAE, CNRS, USR 3135, Institut Franc¸ais du Proche-Orient (Ifpo), East-Jerusalem, Palestinian Territories, 3 CNRS, Minist Culture, LAMPEA, UMR 7269, MMSH, Aix Marseille Univ, Aix-en-Provence, France, 4 Institute of Earth and Environmental Sciences, University of Freiburg, Freiburg, Germany, 5 UCBL, CNRS, INSA Lyon, LIRIS, UMR5205, Universite´ de Lyon, Villeurbanne, France, 6 Ministry of Culture, Heritage Commission, Riyadh Department, Riyadh, Saudi Arabia, 7 Petra College for Tourism and Archaeology, Al-Hussein Bin Talal University, Wadi Musa, Jordan ☯ These authors contributed equally to this work. remy.crassard@cnrs.fr Abstract Data on how Stone Age communities conceived domestic and utilitarian structures are lim- ited to a few examples of schematic and non-accurate representations of various-sized built spaces. Here, we report the exceptional discovery of the up-to-now oldest realistic plans that have been engraved on stones. These engravings from Jordan and Saudi Arabia depict ‘desert kites’, humanmade archaeological mega-traps that are dated to at least 9,000 years ago for the oldest. The extreme precision of these engravings is remarkable, representing gigantic neighboring Neolithic stone structures, the whole design of which is impossible to grasp without seeing it from the air or without being their architect (or user, or builder). They reveal a widely underestimated mental mastery of space perception, hitherto never observed at this level of accuracy in such an early context. These representations shed new light on the evolution of human discernment of space, communication, and communal activi- ties in ancient times. Introduction Discovery context For at least 40,000 years, humans have reproduced mental images of their natural surround- ings by sculpting objects and by painting and engraving long-lasting physical surfaces. In particular, designing plans and maps, or creating rational two-dimensional images of three- dimensional spaces at a reduced scale, was a momentous cognitive development in abstract thinking and representation [1]. However, while human constructions have modified natural spaces and their surroundings for many millennia, few plans or maps of such humanmade structures predate the protohistoric period of the literate civilizations of Mesopotamia and Ancient Egypt. Indeed, archaeological and historical research have only documented a few PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 1 / 31 PLOS ONEis funded by grants from the French Ministry of Foreign Affairs, Al-Hussein Bin Talal University (project No. 164/2016) and the CNRS National Institute for Humanities and Social Sciences (WAA, MT). The GLOBALKITES project (research at Jebel az-Zilliyat) was funded by a French National Research Agency grant ANR-12-JSH3-0004-01 (RC). The Dumat al-Jandal archaeological project (research at Jebel az-Zilliyat) was funded by grants from the Saudi Heritage Commission, the French and Italian Ministries of Foreign Affairs, CNRS UMR-8167 Orient & Me´diterrane´e, the University L’Orientale of Naples (Guillaume Charloux and Romolo Loreto). Graph modeling studies were funded by IXXI, Institut rhoˆnalpin des systèmes complexes, Lyon, France (HS). Publication costs were funded by UMR 5133, Arche´orient (RC). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. The oldest plans to scale of humanmade mega-structures architectural plans and miniature models of buildings and large-sized objects from that time period (e.g.: domestic and ritual buildings, boats) [2, 3]. Before that, it is unknown how Stone Age communities conceived their buildings and use of their domestic or utilitarian structures, although a few examples show schematic, non-accurate, representations of various-sized built spaces [4, 5]. For the specific case of ‘desert kites’, which are prehistoric stone structures used as mega- sized traps to hunt wild animals, the existence of such representations is of the utmost impor- tance for understanding how they were conceived and perceived in the landscape, at a time when mapping was unknown [6, 7]. We report here the exceptional discovery of the up-to- now oldest realistic plans, engraved on stones, of some of these humanmade archaeological mega-traps, from south-eastern Jordan and northern Saudi Arabia, the oldest of which are dated to 9,000 years ago. Unlike other evidence for generic and rough representations of such structures, these engravings are extremely precise depictions of neighboring desert kite structures dated to the Neolithic [7, 8]. Such depictions were necessarily designed by the constructors and/or the users of the desert kites themselves, as the whole structure layout is impossible to grasp without seeing it from the air or without being their creator. They reveal a widely underestimated men- tal mastery of large structures, human landscapes and social spaces, hitherto never observed at this level of accuracy in such an early chrono-cultural context. Such plans may have been used for enhancing collective hunting strategies with these mega-traps. These representations shed new light on the evolution of human perception of space and communal activities in ancient times, and highlight proto-forms of inscribed communication, going beyond the mere depic- tion of a mental representation of a large-scaled structure. Desert kites: Definition and general issues Desert kites (or simply kites) are gigantic archaeological structures made of stone alignments and walls. Kites are composed of driving lines (from hundreds of meters to 5 km long) con- verging towards an enclosure (median size: 1 ha), which is surrounded by up to 4-meter-deep pits (called ‘pit-traps’, from 1 to more than 20 in number per enclosure) where animals were trapped by hunters [7] (Fig 1, and S1–S4 Figs). They represent some of the most impressive stone-built constructions erected by humans in recent prehistory. Desert kites are the earliest large-scale monuments known to date, dating back to as early as 9,000 years ago, during the Pre-Pottery Neolithic B (PPNB) period in Jordan [8]. Elsewhere, some of them were in use in more recent times [9]. Hunting in past societies has been widely studied by archaeologists for decades but hunt- ing using trapping techniques has rarely been comprehensively addressed. Our recent studies of kites in various regions propose to bridge this gap by applying a multi-proxy approach to the worldwide expansion of these spectacular types of traps, drawing on diverse disciplines such as archaeology and social anthropology, bioarchaeology and geoarchaeology, geomatics and statistics, and geochronology [6]. The whole dataset is based on a large body of archaeo- logical structures initially detected from satellite imagery, and then studied in the field. These massive structures visible from airplanes were first recognized in the 1920s and were quickly interpreted as hunting traps, which was confirmed by recent archaeological excavations [6, 7, 10–12]. Until recently, almost no in-depth studies had been carried out to enhance our understanding of their function, functioning, chronology or why they were so widespread in many regions ranging from Arabia to Uzbekistan. Kites are highly sophisticated structures in the landscape involving mass hunting strategies of wild animal herds, far from settled areas, in what are today arid environments. They reflect the emergence of innovative strategies of PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 2 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures Fig 1. Distribution and characterization of desert kites. (A) Desert kite plan from Kazakhstan (Ustyurt Plateau). (B) Desert kite plan from Armenia (Mount Aragats). (C) Desert kite plan from Jordan (Harrat al-Shaam). (D) Desert kite plan from Saudi Arabia (Khaybar). (E) Distribution area of desert kites from western Arabia to Uzbekistan. (F) Oblique aerial picture of a desert kite in Jordan (Harrat al-Shaam, photo OB, Globalkites Project). (G) Oblique aerial picture of a desert kite in Saudi Arabia (Khaybar, photo Khaybar Longue Dure´e Archaeological Project, RCU-AFALULA-CNRS). (H), (I), (J) Desert kite pit-traps during and after excavation (of half the pit) in the Harrat al-Shaam region of Jordan (photos RC, OB, WAA, Globalkites Project). https://doi.org/10.1371/journal.pone.0277927.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 3 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures animal resource procurement (essentially meat, but most likely other resources such as horns, hair and hide), extending beyond the sole purposes of subsistence. This new adaptive choice, inducing the artificialization of landscapes and the probable progressive degradation of biodiversity (materialized by possible extinctions of animal species) started at a time when agriculture and domesticate farming was established in neighboring regions of the Near East by well-settled societies. Regional setting and location of the engravings Currently, 6,255 desert kites have been recorded in the kite distribution area across the Middle East, Caucasia, and Central Asia [6, 7, 9, 13–17] (for permanently updated information, nota- bly on the total number of kites, see: www.globalkites.fr). The lava fields known as Harrat al- Shaam in southern Syria, eastern Jordan and northern Saudi Arabia, contain the most numer- ous and the highest density of kites (up to one kite/sq. km). Two specific zones of investigation at the edge of the lava fields, with different geographical and geological features, yielded two rock engravings depicting kites to scale. These are the Jibal al-Khashabiyeh plateau at the east- ern edge of the Al-Jafr Basin in south-eastern Jordan and the Jebel az-Zilliyat plateau at the northern edge of the Nefud Desert in northern Saudi Arabia, separated by 267 km. In Jordan, the Jibal al-Khashabiyeh area, located 80 km east of Al-Jafr city, is marked by a limestone plateau escarpment with a maximum elevation of 1,000 m above sea level, sloping to the east. A major hydrographic network dissects this escarpment towards the Al-Jafr basin. The limestone bedrock is rich in chert, particularly on the current surface of the desert rocky pavement and along the upper parts of the escarpment slope. The top of the escarpment yielded eight kites (S1 Table), built with the whitish limestone blocks and piles of chert slabs covering the surface of the desert, in order to visually mark the generally dark landscape [12] (Fig 2). The same number of archaeological complexes, interpreted as campsites, are associated with the kites. Three campsites have been partially excavated, including a rescue excavation at the looted site F15 (close to kite JKSH 08), where a stone engraved with a kite representation was found. Some 260 km to the east, Jebel az-Zilliyat in Saudi Arabia is part of a complex topography of the central part of the Al-Jawf province comprising numerous grabens and faults, including the Wadi As-Sirhan fault, which is one of the most significant geological features. A few kilo- meters to the north, the exposed bedrock mainly consists of sandstone, where the large and horizontal Jebel az-Zilliyat plateau is dissected by faults and drainage features. Two pairs of kites are visible (kites AB135/AB136 and AB547/AB549; S2 Table) on the edge of the plateau cliff, with a distance of 3.5 km separating the two distinct groups. A pedestrian survey in one of the drainage features of the plateau revealed a massive engraving to scale representing two kites, equidistant from these two pairs (Fig 3). Results The Jibal al-Khashabiyeh engraved stone At Jibal al-Khashabiyeh, the F15 site is a PPNB settlement disturbed by recent looting activities (Fig 4A). Architectural remains are nevertheless still visible on the current surface of the ground, including numerous carved stones scattered amongst the looters’ spoil. These com- prise elements comparable in shape to those known as ‘cigar-shaped stones’ in the architec- tural typology of the Mureybetian culture of the Near Eastern Pre-Pottery Neolithic A (PPNA) [18, 19]. On one of these limestone monoliths, 80 cm-long, 32 cm-wide and 18 cm-thick (exact dimensions are: length between 78.27 and 82.76 cm; width between 29.18 and 34.67 cm; thick- ness between 17.11 and 19.22 cm), found on the surface of the site in June 2015 (Fig 4B), a very PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 4 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures Fig 2. Location of fieldwork in Jordan. Location of the eight desert kites in the Jibal al-Khashabiyeh region in south-eastern Jordan (left), and detailed plans of these kites (right), with mentions (St.n) of excavated pit-traps. The red dot in the general map (left) shows the location of the engraving, found at the JKSH F15 site. The green dots in the kite plans indicate the excavated pit-traps and their numbering. https://doi.org/10.1371/journal.pone.0277927.g002 well-preserved engraving of a kite is clearly visible (Fig 4C and 4D). The weight of this heavy monolith is 92 kg. The edges of the limestone block were carved with a massive hard hammer. The engraving is made of a combination of multiple carving techniques, including fine incisions (notably to delineate the contours of the kite) and pecking (Fig 5 and S5 Fig). It was most probably pro- duced with a lithic tool on rather chalky, easy to incise limestone, such as a burin, a scraper, or even a raw flake or blade. High-quality natural and human-transformed chert literally covers the area and could have been used for the task. Technically, the kite is depicted in low relief, as the entire internal space of the kite has been carved out and smoothed to a depth of ca. 0.5–1 cm. This could have been achieved using a more massive lithic tool such as a thick flake or an adze-like tool. The engraving depicts a kite structure with its three essential components (Fig 4D). The driving lines are represented by two main converging curved lines of ca. 40 cm in length, lead- ing to a carved star-shaped enclosure of approximately 26 cm in diameter, at the circumfer- ence of which are eight circular cupules (cup-marks), ranging from 2.5 to 6 cm in diameter, PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 5 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures Fig 3. Location of the fieldwork in Saudi Arabia. Location of the four desert kites in the Jebel Az-Zilliyat region in northern Saudi Arabia (top), and detailed plans (bottom) of the two pairs of kites. The red dot in the general map (top) shows the location of the engraving, in the bed of Wadi az-Zilliyat. The green dots in the kite plans indicate the excavated pit-traps and their numbering. https://doi.org/10.1371/journal.pone.0277927.g003 representing the pit-traps of the kite. To the right of the entrance of the enclosure, part of the kite plan is missing, either due to weathering and exfoliation of the monolith, or because the engraving was unfinished. Each one of these cupules is located at the extremity of the star points. Given the overall symmetry in the drawing layout, an additional cupule was probably intended to be drawn at this specific location, bringing the total number of circular pit-traps on the engraving to nine. The driving lines run almost parallel to each other, forming a narrow and elongated corridor, before turning at an almost 90˚ angle to the right, where they finally converge towards the narrow entrance of the star-shaped enclosure. This overall layout and organization–the narrow corridor funnel shape, and the marked curve immediately preceding the final convergence point at the entrance of the enclosure–constitute a local structural char- acteristic of neighboring south-eastern Jordanian kites. This plan has not been documented elsewhere in the kite concentration in north-eastern Jordan, and only rarely in the northern Saudi Arabian group. A comparable shape is found in Saudi Arabia, at distances of 85 km and 270 km for the two closest examples (AB754: 29.989N; 37.881E and AB557: 29.24N;39.554E). Just before the curvature of the driving lines, a zigzag pattern of two to three incised parallel lines runs perpendicularly across the corridor width, forming a regular design of five chevrons. The exact meaning of this graphic pattern is unknown, but we postulate that it is of symbolic significance and could either indicate an unusual device–possibly in light materials and there- fore not preserved–related to the kite’s functioning (such as a beater-driven hunting spot or a PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 6 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures Fig 4. Discovery of the engraved stone in Jibal al-Khashabiyeh, Jordan. (A) Orthomosaic view of JKSH F15 site where the kite’s engraving was found on a monolith (in red is the location of the rescue excavation in the looter’s spoil). (B) Photograph of the engraved stone at the time of discovery at the JKSH F15 site (the monolith was found lying down and was set vertically for the photograph). (C) Photogrammetric 3D model of the engraved monolith showing the different faces, including the engraved upper face (top), while the hill-shaded model (bottom) shows the interpretative drawing of the engraved plan on the stone. (D) Drawing of a projected view of the kite representation engraved on the monolith from the JKSH F15 site. https://doi.org/10.1371/journal.pone.0277927.g004 net), or a representation of a topographic slope break feature. In view of our understanding of the overall functioning of kites gleaned from excavation results and spatial analysis [7], the lat- ter hypothesis seems more likely. Indeed, the presence of a specific device at this strategic loca- tion seems incongruous. On the other hand, the slope constitutes a conspicuous feature in the landscape and setting of kites, and is believed to be an active part of the successful functioning of kites [12]. The weathering and erosion of the harder upper layer of limestone evidences a PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 7 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures Fig 5. Detailed photographs of the engraved stone surface from Jibal al-Khashabiyeh, Jordan. The detailed views emphasize the various techniques used for the kite engraving found at the JKSH F15 site. The detailed photographs on the left are lettered, indicating their position on the engraved monolith, as shown on the right. https://doi.org/10.1371/journal.pone.0277927.g005 fragmentation pattern of an irregular line along the escarpment slope, and the graphic chevron pattern could be a seemly schematic representation of this (S6 Fig). The Jebel az-Zilliyat engraved rock The Jebel az-Zilliyat representation is much larger and was found during rock art surveys in March 2015 in Wadi az-Zilliyat cutting through Jebel az-Zilliyat (Fig 6). Two graphic units representing two kites are visible on a 382 cm-long and 235 cm-wide (maximum visible width) flat surface of a massive 105 cm-thick sandstone boulder, which fell to the wadi bed from the overhanging cliff edge (Fig 7). The kite representation on the eastern part of the boulder is almost entirely readable, while the one on the western part is extensively damaged by erosion. Unlike the Jibal al-Khashabiyeh example, the rock engravings at Jebel az-Zilliyat are exclusively made by pecking with unknown tools (Fig 8), possibly a lithic burin or a more massive hand- pick. Again, high-quality chert is widely available locally and may have been used for the task. The eastern engraving shows a kite with two short and widely spaced convergent driving lines of 80 and 85 cm in length leading to a star-shaped enclosed surface of about 50 to 60 cm in width. Six cupules (pecked cup-marks) between 3 and 5.5 cm in diameter are located at the extremities of point-shaped appendixes. The two southern cupules are located on another plane of the main rock surface, along the upper part of the southern side of the boulder. The eastern one of the latter shows a traced delimitation at the base of the point-shaped appendix (Fig 8B). This feature is known in the field on current kites as a locked-point pit-trap type. The western engraving is much more difficult to read, as the exfoliated surface of the sandstone prevents a clear interpretation of the design. The general shape seems to be quite similar to the eastern kite, with the remains of at least four cupules between 3.5 and 4.5 cm in diameter rep- resenting the pit-traps (Fig 7). Two 85- and 60-cm-long converging lines lead to a smaller enclosure of about 40 to 45 cm in width. The southern extremity of the enclosure is missing, PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 8 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures Fig 6. Location of the engraved rock in Wadi az-Zilliyat, Saudi Arabia. (A) General view of Wadi az-Zilliyat, from the northeast, the location of the engraved boulder is shown by the white circle. (B) General view of the collapsed boulders from the southeast, the white circle indicates the position of the engraving. (C) General view of the engraved rock (white circle) location among the collapsed boulders, from the east. (D) The engraved boulder as discovered during rock art survey, view from the north. https://doi.org/10.1371/journal.pone.0277927.g006 PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 9 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures Fig 7. The engraved boulder from Jebel az-Zilliyat, Saudi Arabia. (A) Drawing of a projected view of the kites’ representation showing picking tool traces, engraved and weathered zones. (B) Drawing of a projected view of the kites’ representation showing legible and unclear engravings, with a colored restitution of the microtopography of the boulder surface. https://doi.org/10.1371/journal.pone.0277927.g007 due to boulder fracturing. It is still unknown if the engraving was made on the cliff itself before the boulder collapsed, or if it was made on the collapsed boulder in the wadi bed. Description and dating of Jibal al-Khashabiyeh desert kites near the engraving The dating of the engraved rocks themselves or the sedimentary deposit from which the rocks were extracted was impossible. Therefore, we compared the designs of the representations to PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 10 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures Fig 8. Detailed photographs and hill-shaded surface models of the engraved stone surface from Jebel az-Zilliyat, Saudi Arabia. The detailed views emphasize the various techniques used for the kite engraving. The detailed photographs and hill-shaded surface models on the left are lettered, indicating their position on the engraved boulder, as shown on the right. A and B are showing examples from the eastern depiction of a kite, C and D from the western one. Hill-shaded surface models are used to better render the legibility of the engravings, otherwise less visible from the photographs made on site. https://doi.org/10.1371/journal.pone.0277927.g008 neighboring kites in both regions, in order to infer the age of the engraved rocks. Many simi- larities were observed. Additionally, in Jibal al-Khashabiyeh, a radiocarbon date was obtained from the F15 site itself (8016 ±29 BP, 6930 ± 80 cal. BC), where the engraving was found. The dated sample consists of charcoal collected in a disturbed context, during a rescue excavation in the looters’ spoil, but the material clearly originates from the original occupation. The extent of the latter is limited with no evidence for successive reuse throughout time, as evidenced by PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 11 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures Fig 9. Comparison of the kite layouts depicted on the engravings with the top-view plans of neighboring desert kites in Jibal al-Khashabiyeh, Jordan and Jebel az-Zilliyat, Saudi Arabia. (A) Comparison of the kite layout depicted on the engraved monolith (left) with the top-view plans of the four better preserved kites identified in Jibal al- Khashabiyeh (right). The red dotted line is the shape of the kite engraving, used for superimposition on the desert kite maps. (B) Comparison of the kite engraving found at Jebel az-Zilliyat (left) with top-view plans of the four neighboring desert kites (right). Gray zones are destroyed or reused areas, after the period of kite use. https://doi.org/10.1371/journal.pone.0277927.g009 the homogeneity of the material culture at the surface of the site. The date turned out moreover to be consistent with the dating of nearby kites. At Jibal al-Khashabiyeh, the general aspect of the eight kites is very similar to the engraving found at the F15 site. Among the eight kites, four display the characteristic shape of a narrow funnel formed by the driving lines before the marked curve preceding the final convergence point at the enclosure entrance (S7 Fig). Two of these kites, JKSH 07 and JKSH 08, are the nearest kites to site F15 where the kite engraving was discovered. These two kites show a spe- cific pattern, with the southern driving line marking a sudden change in direction opening the funnel widely, while the northern driving line remains straight (Fig 9). A similar layout is depicted in the kite engraving (Fig 4D). As kite enclosures suffered heavy erosion due to their location at the starting point of intermittent water drainage systems, the star-shaped polygonal enclosures are partially preserved and only clearly identifiable at three kites (Fig 9A). At kite JKSH 01, eight pit-traps are still visible, and their original total number can be presumed to be nine (based on the regularity of the shape and estimated distance between each pit-trap; S7 Fig). At kite JKSH 04, eight are visible, and three additional pit-traps are probably missing, while at kite JKSH 07, only three are still visible and six additional pit-traps are hypothesized. The average number of pit-traps is therefore similar to the number of cup-marks on the kite engraving. These three kites were excavated. At each one of them, half or quarter of two peripheral pit-traps were excavated, revealing hollow structures with carefully built internal PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 12 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures stone facing, and a depth ranging from 143 to 176 cm. Five radiocarbon datings on charcoal samples from the lower fill of the pit-traps at kite JKSH 01 provided an age of ca. 7,000 cal. BC (S3 Table) [8]. Description and dating of Jebel az-Zilliyat desert kites near the engraving At Jebel az-Zilliyat, the engraved rock is located in Wadi az-Zilliyat, between the western pair of kites AB547 and AB549 and the eastern pair AB135 and AB136 (Fig 9B). AB135 and AB136 are only 120 m apart, and their openings and funnel entrances are parallel to each other (Fig 3). As observed on the engravings, the orientation of the kites is similar, with the enclosure facing south (Fig 7). AB135 is a huge star-shaped example of a rather typical type of kite in the region. Three driving lines (3,379 m for the longest) lead to a 1.13 ha enclo- sure. The current state of preservation of the latter kite is very poor as many parts of the structure have been bulldozed by geophysical surveys for oil and gas exploration. The kite has also been partially reused by more recent (most probably later prehistoric) populations who built tombs (stone mounds, cairns) and circular structures (pastoral or dwelling enclo- sures) with stones from the kite. Nevertheless, field observations combined with aerial and satellite photo interpretation yield an accurate evaluation of the initial structural aspect of the AB135 kite. Standing tabular sandstone alignments and more rarely actual walls with two to three stone rows at most characterize the general construction technique. The AB135 kite presents four well-distinguished pit-traps on its north-western side, and two at its southern extremity (including the heavily damaged south-eastern one). These latter two are at the edge of the plateau’s cliff, and are intriguingly similar to the better-preserved kite engraving in the neighboring Wadi az-Zilliyat (S8 Fig). The eastern side of the kite is unfor- tunately too damaged and reused to gauge its original shape. Series of pit-traps are also visible at the northern part of the AB135 kite, along the driving lines, but construction tech- niques and desert varnish were slightly different from the main original kite star-shaped structure, and they are coarser in appearance. These field observations show that they were built at a later phase and are clearly add-ons. Two archaeological soundings were carried out in two of the peripheral pit-traps (AB135-L01 and AB135-L03). The sounding in the AB135-L01 pit-trap revealed very shallow infilling reduced to old red altered soil unsuitable for optically stimulated luminescence (OSL) or radiocarbon dating. The AB135-L03 pit- trap took advantage of the natural topography for the construction of its peripheral wall. The sounding in AB135-L03 yielded no artifacts but sediments yielded datable remains of five fragments of calcitic nutlets of Arnebia sp. (Boraginaceae) found about ten centimeters above the bottom of the pit-trap. This sample provided a conventional date of 6760 ± 40 BP (5670 ± 35 cal. BC). This age is confirmed by the OSL analysis dating of sediment sam- ples from the lower part of the same pit-trap, which yielded a date of 7690 ± 400 years (5675 ± 400 BC). These dates give an age for the initial filling of the AB135-L03 pit-trap after the abandonment of the kite, by this indicating a minimum age for its construction (S4 Table). The AB136 kite is less disturbed and damaged than its neighbor AB135 but consists of a right driving line 156 m in length, a left one of 54 m in length, an enclosure of 0.42 ha and six pit-traps. The driving lines might have worked together with those from AB135. The AB136 kite presented no pit-traps with enough sediment accumulation due to an unfavorable overall topographic context for good preservation. Thus, no excavations were carried out there. The western pair of kites AB547 and AB549, displaying the same construction techniques, is less disturbed and damaged. Two of their driving lines seem to have worked together (1,549 m and 2,623 m in length), while their enclosures respectively measure 0.55 and 0.63 ha, both with six PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 13 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures peripheral pit-traps (Fig 3). At kite AB547, no pit-trap was considered suitable for excavation. All were either on a slope, or contained very little accumulated sediment, jeopardizing a secure context for geoarchaeological analyses or OSL sampling. The kite was accurately mapped using Differential GPS, showing its direct relation with the neighboring AB549 kite, and rein- forcing field observations that seem to indicate that both kites functioned together simulta- neously. At kite AB549, one cell was excavated (AB549-L01), with a maximum depth of 60 cm between the present-day surface and the bedrock. Vertical slabs were aligned for the construc- tion of the inner pit wall, sometimes forming a double facing wall, and with four to five courses of flat horizontal stones. A few larger slabs were positioned over the top of this wall as a cor- belled construction. No artifacts were found, but sediment sampling yielded 11 small frag- ments of Arnebia sp. nutlets (from a depth of 49 cm), providing a radiocarbon date of 3300 ±40 BP (1570 ± 50 cal. BC), while the OSL dating of sediment samples from the lower parts of the same pit-trap provided two ages of 8040 ± 430 years (6025 ± 430 BC) and 7480 ± 460 years (5465 ± 460 BC). The disagreement between the estimate from radiocarbon dating and those from OSL underlines the need to use jointly, when possible, different dating methods. In the present case, the potential mobility of the small fragments of Arnebia nutlets in the stratigra- phy and the very small amount of carbon obtained after purification (0.2 mg) lead us to favor the results of OSL dating. These OSL dates give an age for the initial filling of the AB 549 LO1 pit-trap after the abandonment of the kite, by this indicating a minimum age for its construc- tion (S4 Table). Quantitative comparison of the engravings and neighboring kites In order to quantify the degree of similarity between the kite engravings and the archaeological kite structures, we undertook a computer-based verification. We compared each engraving with 69 archaeological kites from various areas (S5 and S6 Tables). To do so, we used a graph modelling of the kite structures from satellite images (Fig 10A–10C). Graphs, which are math- ematical structures composed of sets of vertices linked by sets of edges [20] are commonly used to model objects and shapes in several domains, including kite studies [21, 22]. We represented the lines by edges and the ending points of the lines by vertices. A kite graph is an attributed undirected weighted graph G = (V, E), where: • Each vertex v 2 V has two attributes x and y which represent the coordinates of the vertex in the Euclidean space. Note that x and y also represent the coordinates of the corresponding pixel in the satellite image. • Each edge e = (u, v) 2 E has a weight which corresponds to its length. Let (ux, uy) and (vx, vy) be the coordinates of vertices u and v respectively, the length ωe of edge e = (u, v) is com- puted as follows: r oe ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi �2 ux (cid:0) vx � Þ2 þ uy (cid:0) vy ð ð1Þ With this modelling, comparing kites becomes a problem of geometric graph comparison. Several similarity measures are proposed in the literature to compare geometric graphs [23], but their efficiency is closely related to the objects they were defined to compare. Therefore, we propose a new similarity measure adapted to kites. It is based on three main kite properties considered as discriminatory by archeologists, and which are in order of importance: 1) enclo- sure shape, 2) the external angles between the driving lines and the enclosure, and 3) the ratio between the length of the enclosure and the length of the driving lines as expressed by this PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 14 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures Fig 10. Computer-based verification using geometric graphs. (A) Desert kite satellite photo. (B) The kite extracted from the satellite photo. (C) The geometric graph extracted from the kite. (D) Polygon representation of the enclosure shape. (E) External angles between the driving lines and the enclosure. https://doi.org/10.1371/journal.pone.0277927.g010 equation: ð Sim K1; K2 Þ ¼ w1∗E K1; K2 ð Þ þ w2∗A K1; K2 ð Þ þ w3∗P K1; K2 ð Þ ð2Þ In this equation, E(K1, K2) measures the similarity between two enclosures, A(K1, K2) com- pares the angles between the driving lines and the enclosure and P(K1, K2) compares the driv- ing line and enclosure lengths for the two compared kites K1 and K2. w1, w2 and w3 represent the importance given to each criterion. They are numbers between 0 and 1 and their sum is equal to one. The enclosure is the most important part of the kite structure. The main properties distin- guishing two enclosures are the number of pit-traps around the enclosure and the enclosure shape. We capture the shape of the enclosure by a polygon defined by the vertices correspond- ing to the pit-traps and vertices half-way between two pit-traps (Fig 10D). To compare two enclosures, we focus on the shape of the corresponding polygons without emphasizing the exact dimensions. Then we represent each polygon by a string of angles and directions and we use string edit distance [23] to compare them. The angles between driving lines and the enclo- sure are also important properties of the kite structure (Fig 10E). To compare these angles for two kites, we use simply the difference between their values within (A(K1, K2). Finally, to take into account the lengths of the enclosure and the driving lines into our similarity measure, we PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 15 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures consider, within (P(K1, K2, the difference between their ratios, i.e. the ratio between the length of the enclosure and the length of the driving lines. S5 and S6 Tables present the similarity scores we obtained when comparing the engravings with the archaeological kites, using this similarity measure. The scores are given as percentage of similarity and the obtained values range from 26.83% to 81.43% for the Jebel az-Zilliyat engraving and from 11.46% to 75.90% for the Jibal al-Khashabiyeh engraving. The higher the score, the higher is the similarity between the kite and the engraving. According to this, the most similar kite to the engraving found in Jebel az-Zilliyat is the kite AB135, located 2.30 km from the site where this engraving was found, confirming that the engraving is most likely a reproduction of this kite. For the engraving found in Jibal al-Khashabiyeh, the most similar kite is JKSH 01 (with a similarity score of 75.90%), located 16.3 km away from the engraving, while the second most similar kite, JKSH 07 (with a similarity score of 73.52%), is only 1.4 km away from the engraving (S5 and S6 Tables). Discussion Similarity between neighboring kites and engravings of kites and plans to scale Comparisons between kite shape and the engraved representations of kites show clear simi- larities, easily observed with qualitative and quantitative approaches. The engraved designs represent neighboring kites both in Jordan and Saudi Arabia. The engravings are surpris- ingly realistic and accurate, and are moreover to scale, as observed by the geometric graph- based assessment of shape similarity. Kites are gigantic structures and are extremely diffi- cult to apprehend in the field without access to a comprehensive image of their layout (an aerial view is the most appropriate). Yet these engravings are obviously mental depictions that could only have been made by their users and/or builders. At Jebel az-Zilliyat, the number of pit-traps on the eastern representation is undoubtedly similar to the original state of the AB135 kite (Fig 9B). As mentioned above, the two southern pits are both at a very specific location along the edge of the escarpment. The better-preserved of the two engravings in the vicinity also shows the two southern cupules on another plan of the boul- der, giving the depiction of a realistic three-dimensional aspect. The orientation of the engravings on an unmovable boulder is also the same as that of the nearby kites, and finally, the engraving depicts a pair of kites, and two pairs of kites are the closest structures to it. The kites are depicted at a scale of approximately 1:175 and are in keeping with cardinal directions. At Jibal al-Khashabiyeh, the engravings are also similar to the kites, with a simi- lar number of pit-traps to kite JKSH 01 (S7 Fig). Overall, the shape, layout and proportions of the kite engraving are consistent with the actual remains of nearby kites (Fig 9A). The schematized shape of the drawing fits surprisingly well with the top-view drawings of the structures. In particular, the proportions between the corridor width delimited by the driv- ing lines and the diameter of the enclosure are similar to those of kites JKSH 01 and 04. Based on this comparison, the kite carving at JKSH was designed at a scale of approximately 1:425. The pit-traps in turn are significantly oversized on the drawing compared to their actual size (approximately from two to five times bigger). However, this is not surprising as these features would have been barely perceptible at the drawing scale, and certainly too small to be accurately carved in the stone. Moreover, our study and analysis showed that these features form the final trap and are thus the focal point of the hunting device where animals would accumulate [7]. The exaggeration of pit-trap size in the kite depictions could therefore also be interpreted through the prism of symbolism, in view of the significance of these features. PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 16 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures Other maps, plans and representations of mega-structures in human history Up until now, no cases of accurately depicted Neolithic architectural structures by those who designed them were known in the archaeological record. These examples of kite representa- tions are thus the oldest known architectural plans to scale in human history. The only older or contemporaneous examples of mental representations of space are schematic maps of the natural surroundings of human groups throughout recent prehistory. Back in earlier periods, some Upper Paleolithic engravings from Europe have been interpreted as maps, used for por- traying hunting strategies in a favorable topographic context of geographic bottlenecks and natural barriers, for example at Abauntz Cave in Spain, at Kiev-Kirillovskaya in Ukraine, or at Pavlov and Předmostı´ in the Czech Republic [24–29]. These are certainly examples of symbolic mental representation, but they are abstract representations, not scaled depictions of a land- scape. Recent work interpreted a Paleolithic depiction at Molı´ del Salt campsite in Spain [5] as a naturalistic depiction of a reality directly facing the artist. This cannot thus be considered as an actual map/plan, although it is convincing enough to be construed as a mental representa- tion of a social space. Some of the earliest documents described in Europe as actual map repre- sentations are the agricultural plot depictions from the Bronze and Iron Ages known from the Alps, at Mont Be´go or Val Camonica notably [30–33]. These are complex representations of agricultural landscapes including various field features, enclosures, and dwelling areas, as well as road networks. However, it is unclear as to whether they represent real landscapes in the vicinity of the engravings, or more distant, or even imaginary landscapes. In another region with rich or complex representations of probable maps, a probable slightly older depiction of a whole area was found on a massive schist slab from Saint-Be´lec in western France. The slab bears what appears to be a real cartography of a territory, with natural details such as moun- tains and rivers, and anthropogenic structures such as burial mounds, enclosures, or settle- ments [34]. The advent of the Neolithic at the beginning of the Holocene in the Near East saw the pro- gressive development of sophistication and specialization in built dwelling and utilitarian mega-structures and megalithic constructions, in parallel with major social and economic changes such as domestication, sedentarism, agriculture, societal hierarchism and the develop- ment of long-distance trade (e.g.: [35, 36]). Nevertheless, only a few depictions of social spaces and models of large-scale objects are known. A mural from level VII of C¸ atalho¨yu¨k (Turkey), dated to approximately 6,600 BC, is another example of a possible map, similar to the Upper Paleolithic specimens. It depicts a village with a possible reference to a volcanic eruption of Mount Hasan Dağı [4, 37–39]. The oldest three-dimensional model of a large-scale object is a sea-faring reed-bundle boat, known from the Ubaid-period H3 site in Kuwait, dated to around 5,000 BC [2]. Furthermore, the oldest models of built dwelling structures made of burnt clay are known from the Kodjadermen-Gumelniţa-Karanovo culture in southeast Europe around 4,500 BC [3]. All are fascinating examples of mental representations, in two or three dimen- sions, but are not comparable to the kite plans to scale. Comparable plans, at least intended to be to scale, can only be found much later during the third and second millennia BC in Meso- potamia, with for example the inscribed tablet of Yorgan Tepe dated to 2,300 BC [40], or the one from Nippur dated to 1,500 BC [41]. They both represent almost cadastral maps, some- times with rivers and mountains. Once again, they do not reach the degree of accuracy of the kite engravings. Other not-to-scale depictions of kites are however known elsewhere in Jordan [42–47], Syria [48] and Uzbekistan [49], from undated contexts. Nevertheless, they are schematic and do not reveal such realistic details as the Jibal al-Khashabiyeh and Jebel az-Zilliyat examples. PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 17 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures For instance, none of them depicts the pit-traps as hollow carved three-dimensional shapes. Instead, pit-traps are rather simply represented by circular two-dimensional shapes. But other crucial information is sometimes present on the not-to-scale examples, such as depictions of human and animal figures inside and outside the kite structure. Jordanian (Cairn of Hani and Wadi Hashad) and Syrian (Hemma) engraved rocks show herds of small horned mammals, in all likelihood gazelles, wild goats or ibexes. Only a few such engravings of animals have been found in the vicinity of Jibal al-Khashabiyeh, and much more abundantly around Jebel az-Zil- liyat, but they were never directly associated with specific kite representations. In S12 Fig, a series of previously known kite engravings is presented. In the top row (A and B), the two engravings show hunting scenes where animals and hunters are exaggeratedly depicted, they are also simplified and the kite is figuratively represented (especially in A, where for example the driving lines are symbolized by hatched lines). In the middle row (C, D, and E), only the distal part of the whole kite is represented, with almost no dedicated space for the driving lines. The pit-traps are very clearly exaggerated, probably to signify their functional impor- tance. In E, the pit-traps are contiguous all around the periphery of the enclosure, which is never observed in desert kites, at least in the region where the engraving was found. Finally, in all cases, it was not possible to clearly associate one kite in particular that could have been rep- resented by one engraving, contrary to what we found in the present study with the plans to scale of well-identified neighboring mega-structures. Why represent kites to scale? Construction planning, collective hunt organization or symbolic aspect? The construction of kites always has to deal with the generally complex local topography, and the different functional elements require careful arrangement by playing on convexities and breaks in slope [7, 9, 12, 14, 16, 49]. The trap incorporates the topography in which it is designed; it is both a complex object and a spatial object. The dual nature of this feature sug- gests the necessity for spatial information to explain the engraving of a plan. Two different functional uses of this plan can be investigated, in the frame of genuine hunting architecture [50]: a layout plan to plan the construction of the kite, or a descriptive plan to organize the col- lective hunt. Thirdly, a symbolic dimension may also come into play, as with other known rep- resentations of kites. In the case of a kite construction plan, detailed knowledge of the terrain and animal move- ment in the complex topography would have led to the formalization of the trap and its graphic representation before the construction phase. This would have ensured that the struc- ture was in accordance with the project during construction. However, in the cases studied here, the surface of the engraved rocks does not show, in reduction, the topographical details on which the construction was clearly based. It therefore seems unlikely that the physical mate- rialization of the trap would have endeavored to follow the engraved plan, especially as deter- mining topographic elements are not depicted. However, if this were the case, it would imply that these elements had been identified beforehand, but the fact that they are absent from the plan contradicts its very purpose. This is different from a modern development plan in which the topography is surveyed beforehand and then plotted. The second hypothesis of a plan for preparing hunting activities is more plausible. In such a case, the graphic representation of the various functional elements making up the trap would formalize the process, to determine, for example, the number and position of hunters, to coor- dinate their actions and to anticipate animal reactions. A map would most probably be used here as a means of communication (almost like an ancestral way of writing) and would enable the collective interaction required for the smooth running of hunting operations: designing PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 18 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures hunting/trapping strategies, understanding and explaining the architectural and topographical specificities of a structure that cannot be seen from the ground as a whole, and is only known to the builders and designers of the mega-trap. The use of the edge of the engraved rock to sig- nify that certain pits are hidden by the break in the slope (at Jebel az-Zilliyat, Fig 8B), or the chevron drawings (at Jibal al-Khashabiyeh, Fig 5C), are simplified but useful representations of the topography. Here, we perceive the principle of modern cartographic symbols: spatial features (here topographic details) are not drawn by scaling their shape but located and repre- sented by symbols. However, the meaning of these symbols is difficult to interpret, such as the zigzag pattern at Jibal al-Khashabiyeh and the locked-point pit-trap type at Jebel az-Zilliyat. These details may have been useful in explaining or planning the functioning of the trapping devices, but no clear-cut explanation can be advanced, as of yet. The depiction of two pit-traps along the cliff at Jebel az-Zilliyat also indicates the use of the natural topography in hunting strategies and kite constructions. The engraving of these features is clearly significant as the three-dimensional rendering on the kite plan indicates a specific purpose in the whole mega- trapping system. As a matter of fact, slope breaks and cliffs were frequently used to build pit- traps, almost everywhere in the kite global distribution zone. Such natural features served for using the different topographic levels to circumvent or reduce the effort of digging the pits too deep, and for dissimulating the presence of the pits [7]. This being said, a simpler diagram, where the elements are not necessarily to scale, but characterize and represent the different areas, could also be more effective than a scaled plan in the case of a purely practical use of this graphic representation. Thus, these engravings could also be interpreted as the consequence of a quest for a very advanced and sophisticated mastery of the knowledge of perceived space, which would be directly linked to the command of lived-in space. The latter ends up being overcontrolled by the construction of increasingly larger and more effective structures for trapping large numbers of wild animals. In short, these representations also bear symbolic significance, like an almost patrimonial need to demon- strate knowledge, to pass on knowledge of space (kites are built on and around precise topo- graphical criteria, for example), of what this space contains (the ability to hunt masses of animals as a result of excellent knowledge of animal behavior and migration seasons), and of how this knowledge can be harnessed (building mega-traps). However, these engravings should perhaps not be solely construed as symbolic depictions. A utilitarian function remains a credible hypothesis, either for planning kite construction or organizing collective hunting. These utilitarian and symbolic explanations for engraving kites to scale are not incompatible and could well be interconnected. Questioning the evolution of the perception of space and the conceptualization of gigantic shapes Three sometimes simultaneous driving forces have presided over the development of cartogra- phy throughout its history: the emergence of a need, the invention of a technique, and the emergence of a worldview. Firstly, the emergence of a need is a driving factor in the development of cartography. Such needs may arise in the case of territorial defense or tax collection, for example [51, 52]. In the case of kite engravings, as we have just seen, the need exists and could be sufficient motivation to develop a plan. This need is not necessarily exclusively practical and may be driven by the strong symbolic dimension intrinsic to kite structures. With the emergence of highly special- ized hunting techniques, generating innovative and unprecedented strategies for acquiring animal resources, it is likely that kites were a significant cultural feature of these hunting socie- ties, and that their economy and lifestyle were largely determined by these structures. This PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 19 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures socio-cultural or even identity-building dimension may have stimulated a symbolic need for a technically advanced representation of these remarkable structures, which were probably also territorial symbols. Secondly, the invention of a technique is also a driving force in the development of cartog- raphy, for example the invention of the compass which gave rise to portolans in the sixteenth century, or the invention of remote positioning in contemporary geomatic applications [53, 54]. The question of the techniques used to produce the engravings arises, as does that of the construction of kites at a scale far exceeding any architectural undertaking ever carried out by a human group. Fieldwork by present-day archaeologists shows that the morphology of a kite cannot be understood without recourse to aerial, satellite or planimetric support on account of its size and topographical complexity. Of the hundreds of kites studied in the field, it has never been possible to take a photo from the ground showing the entire enclosure, or the rest of the structure, such as the layout of the driving lines as a whole. A walker whose route would encounter a kite would not suspect the presence of a single, coherent layout. And without tools (tacheometer or GPS), a topographer would be hard pressed to produce a sketch of a kite as geometrically reliable as the engravings. It would therefore seem that kite-building hunters knew how to use a surveying technique, still unknown to us, involving notions of measure- ment and even calculation. The technical challenge of building a kite can only echo the appear- ance of these planimetric scaled representations. These two innovations—the construction of the largest structures ever made by human at the time, and cartographic representation to scale—are thus closely interdependent. The common denominator of these innovations is the need to master the three-dimensional perception of a territory or space (landscape). This pre- requisite is essential both for the planning of constructions as vast as kites in a complex topo- graphical environment, intended for the capture of animals whose behavior and movement patterns in the landscape must be known, and for the practice of hunting with these traps. We postulate that this major advance was also at the root of significant developments in the plani- metric representation of space and landscapes, leading to the production of the first plans to scale. Finally, the emergence of a worldview is a third vector of cartographic evolution, like the map of Eratosthenes or Ptolemy’s Geographia [55]. In the case of the kite engravings, there is no need for a worldview in terms of scale. The aim is not to represent a vast terri- tory, let alone the known world, but only the area of the layout. However, this requires decentering, a view from the sky, revealing an elaborate abstraction of space. The process of reducing a space, the elaboration of analogies between physical space and graphic represen- tation is an important development in abstract thought and symbolic representation [1]. In relation to the C¸ atalho¨yu¨k representations, which are somewhat more recent than the kite engravings, Meece [56] questions the ‘map’ status attributed to them. Meece assumes that the cognitive capacities required for such an imaginary representation are too great, requir- ing a shift from a view of individual points of view to a view from above, following the assumption of Gartner [57] that maps can only be made under conditions where aspects such as agriculture, private property, long-distance exchange, conflict, tribal relations and other aspects of redistributive economies are present. Indeed, work describing early formal maps (in Mesopotamia [41], the central Andes [57], China [58]) concerned complex literate societies with a tradition of written records or with trade relations with other literate socie- ties. These documented examples of formal cartography are relatively late, and no carto- graphic representations were previously known in the prehistory of the Near East or even in any Neolithic culture. It was also thought that small, landlocked societies had little need for spatial information [59] and therefore did not need maps for pathfinding or information storage. PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 20 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures However, in the same way as cases of empirical knowledge from Mesopotamia or the cen- tral Andes, a disjunction of human perception from ‘being in the world’ to ‘being above (or even beyond) the world’ [60] clearly underlies the kite engravings. This tends to call into question the idea that a complex literate society is a necessary condition for map-making. Indeed, we can consider that a need for complex spatial information, rather than literacy and a tradition of written production, is crucial for cartographic production. This need arose in societies that had to resolve complex management issues themselves, such as the social organization of agricultural land or urban planning. But it also occurred in the case of sim- pler, small-scale societies, in order to resolve other spatially and socially complex issues. There are many examples throughout the world of mental representations of space made explicit by their creators to outside observers ignorant of the spatial meanings of what may appear to be abstract artistic works (e.g.: Aboriginal lines and spirals to represent paths con- necting water points [61–63], Tuareg cartographic representations of the relative positions of several towns in an area of more than 2,000 km [64], or those of the Snake River by Native Americans representing a 600-km-long area [65]). This situation in traditional pre-literate societies does not contradict the fact that the emergence of this type of representation can only occur within specific social structures [60]. Devising a collective hunting strategy would thus present these characteristics of complexity, with the need to spatially formalize hunters’ actions on the one hand, and to communicate collectively around this formalization on the other. Conclusions The discovery of two prehistoric engravings, one in the northern reaches of Nefud in March 2015, the other in south-eastern Jordan in June 2015, provides some answers to these complex questions of the representation and conception of space, faced by human groups since ancient times. These examples of representations both show, in a similar style, images of kites that can be described as planimetric reproductions. They were found in the vicinity of studied and excavated kites dated to the Neolithic period. These two engravings are thus the oldest plani- metric representations to scale known to date. Moreover, desert kites appear to be the earliest stone-built constructions known up until now at such a large scale. On account of the size, organization, functionality, and effort required for their construction, these mega-structures should actually be considered as fixed infrastructures of the landscape. We believe that the technical challenges involved in the construction, use and maintenance of kites, and their strong sociocultural value in the context of Neolithic developments are at the essence of the earliest planimetric representations known to date. These engravings also teach us that the technical progress required for the construction of kites (large-scale construction requiring a developed perception of space and inscription in the topography), is at the origin of progress in techniques of planimetric representation, as a logical continuation of skills acquired in the perception of space in general. The community aspect of these large-scale hunts is another ele- ment supporting the hypothesis of a need to communicate, in particular to share spatial infor- mation, by means of a realistic representation intended for a human group participating in a common action. The precise history of kites is not yet known across the whole, very vast, kite distribution area. It is, however, now clear that with regard to knowledge of human/animal relationships— and more generally the evolution of subsistence modes in a region of the world where seden- tarism and the urban phenomenon emerged—the significance of these structures has been underestimated until now. The discovery of these very ancient representations highlights the question of the methods used by kite builders. As far as construction methods are concerned, PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 21 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures the walls were built with care and ingenious use was made of topographic features, but the techniques used are still simple and basically revolve around the assembly of unsquared stones of various sizes. In contrast, the layout of the entire system was based on relatively elaborate methods, hitherto unknown for the periods under consideration. Whereas kites could be built by marking out the visual range of a place, the crux of the matter here is the very design of the kite in its natural environment. Kites are large material structures that could not be designed without what we call today planning. Yet this is an intellectual construction, an abstraction calling upon an elaborate representation of space, which was very difficult to grasp up until now, due to the paucity of data on kites. The visual representation of lived-in space, perceived space, dreamed space, has existed for a long time, as far back as the first ‘artistic’ representa- tions (naming them in this way avoids giving them a utilitarian or clearly symbolic role) from the Middle Paleolithic or Middle Stone Age in Africa, and more figuratively from the begin- ning of the Upper Paleolithic in Europe or Australia. But the notion of an aerial view, now widely used since the western perception of the bird’s eye view of a territory, forms the basis of western cartography, and represents our exclusive conception of the territories that surround us. This notion did not make sense in traditional pre-literate societies that represented spatial information symbolically. These two major innovations, i.e., building what would become the largest structures in human history at that time and making cartographic representations to scale, are closely linked by a common point: mastering the three-dimensional perception of a space, and translating it into an inscribed form of communication. This continues to challenge our modern perception of the lived-in, perceived, and dreamed space, versus the represented one. The discovery of these examples of early accurate cartographic representations has far-reaching consequences for our understanding of the evolution of human cognition. The transposition of a mental con- ception of (a large, not possible to grasp as a whole) space onto a two-dimensional (small) surface inherent in these representations, is a milestone in intelligent behavior, progressively acquired from the early stages of human evolution. Materials and methods Site identification, survey and excavation At Jibal al-Khashabiyeh, kites were discovered in 2013 and recorded and excavated in 2015 and 2016 as part of the South-Eastern Badia Archaeological Project (SEBAP), in cooperation with the Department of Antiquities (DoA) of the Jordanian Ministry of Tourism and Antiqui- ties. They were recorded with two systems: the one used in the field by SEBAP (JKSH 01 to 08) and the one used on global satellite imagery interpretation by the Globalkites Project (JD1088 to 1095; for Globalkites methodology see [6, 7, 13]). At Jebel az-Zilliyat, kites were discovered in 2014 and excavated in 2015 as part of joint fieldwork by the Dumat al-Jandal archaeological project and the Saudi Heritage Commission. Kites were recorded in two systems: the Dumat al-Jandal database (DAJ137, DAJ138, DAJ139, DAJ140) and the one used by the Globalkites Project (respectively: AB135, AB136, AB547, AB549). In Jordan, excavation and survey per- mits for the 2015 and 2016 seasons were issued by the DoA (Num. 2015/31, 2016/48). In Saudi Arabia, the Saudi Heritage Commission issued the excavation and survey permits for the 2014 and 2015 seasons. The locations of each archaeological structure and feature were recorded in the field using high-precision Trimble Pathfinder Pro XRS Differential Global Positioning System (DGPS). Geomatic analyses of both fieldwork data and satellite imagery went through a Geographic Information System (GIS; using Esri ArcGIS software). A regularly updated online interactive map shows the current number of recorded kites (available at www.globalkites.fr). DGPS PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 22 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures mapping and kite aerial photography (KAP; using HD digital cameras triggered by interval- ometer or radio control system) led to Digital Surface Models (DSM) and photogrammetric rendering of topography and archaeology (with Agisoft PhotoScan software), documenting archaeological features visible on the surface, and archaeological excavations. Excavations were carried out by hand, usually in a half or a quarter of the pits surrounding the kite enclosure. All sediments were systematically dry-sieved (0.5 cm mesh; 80 to 100% of the total volume) to retrieve archaeological material, while geoarchaeological stratigraphic studies required the collection of small dust samples and large soil samples. Dust samples were analyzed under the petrographic microscope (Olympus BH2, x200 to x400) for mineral, anthropogenic and/or biogenic components. Larger soil samples were water-washed (0.5 mm mesh), observed and sorted under the stereomicroscope (x10) for biogenic material (radiocar- bon datable elements). Both engraved stones were discovered in 2015 during pedestrian surveys, by Juan Antonio Sa´nchez Priego and Wael Abu-Azizeh in Jordan and by Charly Poliakoff in Saudi Arabia. Analysis of the engravings, and photogrammetry The analyses of the macroscopic designs and of the technical parameters were based on various works (e.g.: [66–68]), directly by looking at the engravings, and also by carefully studying pho- togrammetric models, whose production is described hereafter. The Jibal al-Khashabiyeh engraved stone was handled post fieldwork, as the stone was mobile and removed as important artefactual evidence. It was therefore processed for photo- grammetry in a controlled-light environment, using spotlights with appropriate orientation. The stone was photographed from the front (with the engraving) and back (opposite of the engraving) from multiple view angles (total 132 photos). Ten markers (materialized by small white paper tapes with a dot mark inside) were positioned on the lateral sides of the stone, at regular intervals all around the stone (three markers on each long side, two markers on each of narrower top and bottom sides), allowing subsequent alignment and merging of the distinct 3D model chunks of the two sides of the stone. The 3D model was processed using PhotoScan (Metashape) Agisoft software v.1.6.1. (with the following settings: Align accuracy: High; Dense cloud quality generation: High; Mesh build: custom 4 000 000 faces; Texture size/count: 12288 x 1). In view of the quality of the photos, the 3D model underwent minimal standard cleaning processing during the various building steps. The model was set to scale through measure- ments taken on the marker points set on the stone prior to photography. The 3D model was then set into a defined orthonormal frame, with its origin set at the cen- ter of the stone, the x axis through the width of the stone, y axis through its length and z axis through its thickness. Orthomosaics were generated for six predefined views (front, back, left, right, bottom and top), producing a basic plate with all faces of the stone (Fig 4C). This docu- ment clearly illustrates the location of the drawing according to the relief of the stone, espe- cially on its curved long edges. We undertook additional processing steps to produce a projected view of the engraving. The engraving is made on a roughly parallelepipedal monolith, which was worked (carved) on the edges producing irregular curved sides. Important parts of the engraving (the corridor-like shape of the kite driving lines, as well as three of the cup marks materializing the pit-traps) were voluntarily drawn using the curved edges of the stone to emphasize their topographical situa- tion. The production of a projected view of the engraving in its entirety required therefore a specific technical approach. A total number of 260 successive orthomosaics of the stone surface were produced, starting from the front predefined view, and rotating the stone only a few degrees all around (using the x axis of the trackball in order to maintain a consistent y axis). PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 23 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures The 260 views produced over a rotation of 360˚ were made at a rotation angle of less than 1.5˚. All these orthomosaics were then processed with Adobe Photoshop. Only a central 3-cm-verti- cal strip was kept for each one (as the central part of the image is less subject to projection dis- tortions), and they were manually stitched together using the overlapping edges of each strip. This process generated a fully deployed projected view of the stone, revealing the whole kite engraving plan (S5 Fig). This document was used as a basis to draw the full layout of the kite engraving on a two-dimensional plan (Fig 4D), which in turn allowed for comparisons of the carved stone kite representation to actual top-views of the nearby kite structures (Fig 9A). Addi- tionally, a 3D model was produced, and is manipulable at: https://doi.org/10.34969/cnd3d/ 737849.o.2022. The stone was weighed on a scale (Health-o-meter Professional): its weight is 92 kg, which roughly corresponds to an estimation based on an average limestone density of 2.6 g/ cm3 and a volumetric measurement of the monolith on the 3D model of 36,768 cm3. The engraved boulder at Jebel az-Zilliyat has not been removed from its original location. Several photosets (total 170 photos) have been made in the field under different lights and angles, with one 50 cm photo scale for geometric point of reference. Two photosets were dedi- cated to the engraved details, and one to the general context; all have been merged into the general 3D model of the entire setting. The 3D model was processed using PhotoScan (Meta- shape) Agisoft software v.1.6.1. using two chunks, one for the details and one for the modeling of the whole block. The first chunk incorporated high-resolution alignments of all photosets. It was first exported without points of reference, as a digital terrain model (DTM) and an ortho- photo, planar view. These two documents were put to scale with ArcGIS into a three-dimen- sional Cartesian coordinate system, with axis lines x, y and z. Georeferencing information are then accurate in the sub-millimeter range. With PhotoScan, three points of reference were then transferred into the first chunk, in order to obtain a dense cloud and a high-resolution model to scale. From the dense cloud, a DTM and two orthomosaics were generated, one with the photos taken in sunlight, another without any cast shadows. From the DTM, in ArcGIS, we produced a shaded image, showing black and white visualizations with low-angled light. It was made through four lighting angles (45˚, 135˚, 225˚, 315˚), revealing all engraved designs, even the most tenuous ones (S9 and S10 Figs). For producing a projected view of the engraving, procedure was the same as the one of Jibal al-Khashabiyeh. As the boulder surface at Jebel az-Zilliyat was in this case almost flat, except for the two representations of pit-traps in the southeastern part of the rock, only 13 exports as DTM and orthophotos were generated in planar mode, with a rotation less than 1.5˚ on the x or y axis of the trackball. Obtained files were turned into shaded mosaics with Adobe Photo- shop, obtaining the most legible version of the engraving. To model the whole engraved stone into the same coordinate system, 16 points of reference were transferred into the second chunk with PhotoScan. In this local system, a dense cloud was generated and then cut according to the first chunk’s limits, to merge both dense clouds, and to incorporate to the contextual ensemble the high-resolution one of the engraving. This merged dense cloud is made of 157,655,331 points. A first 3D model was made from this cloud, including 30,461,629 faces. To produce a PDF file out of this 3D model, it was reduced to 182,321 faces. The model texture was generated from sunlight-exposed photoset as it offers more contrast for a lighter file (manipulable at: https://doi.org/10.34969/cnd3d/490124.o.2022. Petrographic analysis for Jebel az-Zilliyat boulder At Jebel az-Zilliyat the petrographic characterization of the block is based on the observation of fresh breaks with a stereomicroscope and on the mineralogical analysis of the grains with a petrographic Olympus BH2 microscope (see Supporting information). PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 24 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures Radiocarbon dating and optically stimulated luminescence dating At Jibal al-Khashabiyeh, charcoal remains were recovered from the deep stratigraphic fill of the kite pit-traps. They were either charcoal from small hearths, indicating a later reuse of the structures as ephemeral shelters, or fragments of residual charcoal trapped in aeolian sedi- ments. They were sampled at different depths of the sedimentary fill. Seven samples were pro- cessed for AMS radiocarbon dating at the AMS Laboratory of the University of Arizona. The chronometric dating results (S3 Table) are presented in detail elsewhere [8]. At Jebel az-Zil- liyat, Arnebia sp. calcitic nutlets (7 to 11 small fragments per sample) were used for radiocar- bon dating. They were sampled near the bottom of each stratigraphy. The two samples were processed at the Poznań Radiocarbon Laboratory (Adam Mickiewicz University). All dates were calibrated with Chronomodel v.2 [69] using IntCal20 atmospheric calibration curve [70]. OSL dating was carried out on sediment samples taken from the filling of trap pits. These ages will post-date the age of construction and final use. Small aliquots of quartz (1 mm) were employed for determination of Equivalent Dose (De) using a modified Single Aliquot Regener- ative Dose (SAR) protocol [71]. Mean De values were calculated using the Minimum Age Model [72] due to the presence of partial bleaching as observed in the De distributions. The concentration of dose rate relevant elements was determined using high-resolution gamma spectrometry (cf. [73]), annual dose rates and ages were calculated using ADELE v.2017 soft- ware [74] (see S11 Fig and S4 Table). Graph modeling studies We did not use an automated image segmentation algorithm to compute accurate similarities between the kites and the engravings. Kite shapes had to be extracted manually to deal with the numerous reliefs present in satellite images of kites and obtain accurate kite representa- tions. Thus, kite shapes were manually extracted from images by placing points all along the kite and linking them by drawing a line between two consecutive points (Fig 10A and 10B). The remaining tasks are fully automated by our algorithms. We modeled each extracted shape on a graph where edges are the lines, and vertices are the ending points of lines (Fig 10C). Each edge is weighted by its geometric length. To compute a similarity score between two kites, we also used an algorithm that extracts the shape of the enclosure and the external angles between the driving lines and the enclosure (Fig 10D and 10E). All the algorithms are imple- mented using Python 3.6. The source code of our algorithms as well as the data (vectorial images of the kites and their corresponding graph representations) are available on https:// gitlab.liris.cnrs.fr/aekiouche/kite-project. S5 and S6 Tables present the similarity scores we obtained when comparing the engravings with the archaeological kites. Inclusivity in global research Additional information regarding the ethical, cultural, and scientific considerations specific to inclusivity in global research is included in the Supporting Information. Supporting information S1 Appendix. Supplementary text. (PDF) S1 Fig. Kite JKSH 04 (JD1091). Top plan and different views of the excavation of pit trap St.02. (JPG) PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 25 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures S2 Fig. Kite DAJ137 (AB135). (A) Aerial view of pit-traps L03 (archaeological excavation), L04 and L05. (B) L03 before excavation. (C) L03 after excavation. (JPG) S3 Fig. Kite DAJ140 (AB549). Aerial oblique view of the enclosure. (TIF) S4 Fig. Kite DAJ140 (AB549). Aerial view of pit-traps L01 (archaeological excavation, on the left) and L02 (on the right). (JPG) S5 Fig. Composite ortho-mosaic illustrating the fully deployed projection view of the engraved monolith found at JKSH F15 site. An outlined drawing shows the interpretation of the whole engraving. (JPG) S6 Fig. Hypothetical interpretation of the chevron pattern engraved on the kite depiction on the monolith from JKSH F15 site, Jordan as a topographic symbol. The detail of the chevron pattern (bottom) is compared to the slope break seen in the topography (here at kite JKSH 04, top). (JPG) S7 Fig. Reconstruction of the layout of the enclosures at the three excavated kites in Jibal al-Khashabiyeh, Jordan. The reconstruction is based on the structural remains preserved in the field to estimate missing pit-traps around the enclosure’s perimeter. (JPG) S8 Fig. Comparison between the engraving and a desert kite from Jebel az-Zilliyat, Saudi Arabia. (A) Top-view and profiles of the southern part of desert kite AB135. (B) Top-view and profiles of the southern part of the engraving. (JPG) S9 Fig. Various angles of light to show details of the engraved eastern kite from Jebel az- Zilliyat, Saudi Arabia. From the DTM, in ArcGIS, we produced a shaded image, showing black and white visualizations with low-angled light. It was made through four lighting angles (45˚, 135˚, 225˚, 315˚), revealing all engraved designs, even the most tenuous ones. (JPG) S10 Fig. Various angles of light to show details of the engraved western kite from Jebel az- Zilliyat, Saudi Arabia. From the DTM, in ArcGIS, we produced a shaded image, showing black and white visualizations with low-angled light. It was made through four lighting angles (45˚, 135˚, 225˚, 315˚), revealing all engraved designs, even the most tenuous ones. (JPG) S11 Fig. Characterization of OSL signal properties and individual De distributions. (A) Natural and laboratory irradiated OSL signals show almost identical decay shape with a domi- nance of the fast component (exemplified for AJR1). Inset: De values are in the close-to-linear range of signal growth. (B), (C), and (D) De distributions plots with density probability func- tions for samples AJR1-3 reveal a distinct peak in De values but some values at the higher edge that are interpreted to represent incomplete resetting of the OSL signal in quartz grains of some of the aliquots. (JPG) PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 26 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures S12 Fig. Examples of previously known kite engravings, that are not represented to scale and less realistic. (A) From cairn of Hani’ site, Syria; redrawn from [42]:fig. 8. (B) From Khishaˆm-2 site, rock В 37, Syria; redrawn from [48]: fig. 3. (C) From Wisad Pools, Jordan; redrawn from [46]: fig. 44. (D) From Azraq Basin, Jordan; redrawn from [43]: fig. 3(6). (E) From Wisad Pools, Jordan; redrawn from [47]: fig. 8. (F) and (G) are the kite engravings described in the present study, respectively from Jibal al-Khashabiyeh and Jebel az-Zilliyat, represented in this figure to be directly compared with the previously known ones. (JPG) S1 Table. Corresponding kite numbers between the scientific teams and GPS location of the Jibal al-Khashabiyeh desert kites. (PDF) S2 Table. Corresponding kite numbers between the scientific teams and GPS location of the Jebel az-Zilliyat desert kites. (PDF) S3 Table. Radiocarbon analysis results from Jibal al-Khashabiyeh, Jordan, and Jebel az-Zil- liyat, Saudi Arabia. Dates were calibrated with Chronomodel v.2 [69] using IntCal20 atmo- spheric calibration curve [70]. Md, Q1 and Q3 refers to the median, first and third quartile of the distribution of calendar ages. Bold characters underline the earliest date obtained for each kite (i.e. the radiocarbon terminus ante quem of the construction). All calendar dates are rounded to the nearest ten. * Unreliable sample (0.24 mg of carbon only, very large standard error): not considered. (PDF) S4 Table. Summarized OSL dating data from Jebel az-Zilliyat, Saudi Arabia. AB135 stands for DAJ137 kite, and AB549 stands for DAJ140 kite. (PDF) S5 Table. Similarity scores obtained when comparing kites to the engraving found at Jebel az-Zilliyat, Saudi Arabia. (PDF) S6 Table. Similarity scores obtained when comparing kites to the engraving found at Jibal al-Khashabiyeh, Jordan. (PDF) Acknowledgments In Jordan, we thank the Department of Antiquities, as well as the French Institute for the Near East (Ifpo) and Al Hussein Bin Talal University for their support. In Saudi Arabia, we thank the Saudi Heritage Commission, Guillaume Charloux and Romolo Loreto codirectors of the Duˆmat al-Jandal archaeological project. We thank Charly Poliakoff who discovered the engraving in Saudi Arabia during a rock art survey, and for sharing his photographs and analy- ses of the kite representation from Jebel az-Zilliyat. We thank Sabine Sorin and the CND3D team for publishing the 3D models online. Author Contributions Conceptualization: Re´my Crassard, Wael Abu-Azizeh. PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 27 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures Formal analysis: Olivier Barge, Jacques E´lie Brochier, Frank Preusser, Hamida Seba, Abd Errahmane Kiouche. Funding acquisition: Re´my Crassard, Wael Abu-Azizeh, Hamida Seba, Mohammad Tarawneh. Investigation: Re´my Crassard, Wael Abu-Azizeh, Olivier Barge, Jacques E´lie Brochier, Frank Preusser, Hamida Seba, Abd Errahmane Kiouche, Emmanuelle Re´gagnon, Juan Antonio Sa´nchez Priego, Thamer Almalki, Mohammad Tarawneh. Methodology: Re´my Crassard, Wael Abu-Azizeh, Olivier Barge, Jacques E´lie Brochier, Frank Preusser. Project administration: Re´my Crassard, Wael Abu-Azizeh, Thamer Almalki, Mohammad Tarawneh. Software: Hamida Seba, Abd Errahmane Kiouche. Supervision: Hamida Seba, Mohammad Tarawneh. Visualization: Re´my Crassard, Wael Abu-Azizeh, Olivier Barge, Frank Preusser, Emmanuelle Re´gagnon. Writing – original draft: Re´my Crassard, Wael Abu-Azizeh, Olivier Barge. Writing – review & editing: Re´my Crassard, Wael Abu-Azizeh, Olivier Barge, Jacques E´lie Brochier, Frank Preusser, Hamida Seba. References 1. Robinson A. Early Thematic Mapping in the History of Cartography. Chicago: University of Chicago Press; 1982. 2. Carter R, Crawford H. Maritime Interactions in the Arabian Neolithic–Evidence from H3, As-Sabiyah, an Ubaid-Related Site in Kuwait. American School of Prehistoric Research Monographs Series 8. Leiden: Brill; 2010. 3. Lazăr C, Ignat T, Stan S, Moldoveanu K, Rădulescu F. Beyond the archaeological imagination. Obser- vations about Kodjadermen-Gumelniţa-Karanovo VI architecture based on a study of experimental archaeology. Mediterranean Archaeology and Archaeometry. 2012; 12(2): 55–79. 4. Mellaart J, Wheeler M. C¸ atal Hu¨ yu¨k: a Neolithic Town in Anatolia. London: Thames and Hudson; 1967. 5. Garcı´a-Diez M, Vaquero M. Looking at the camp: Paleolithic depiction of a hunter-gatherer campsite. PLoS One. 2015; 10(12): e0143002. https://doi.org/10.1371/journal.pone.0143002 PMID: 26629824 6. Crassard R, Barge O, Bichot CE, Brochier JE´ , Chahoud J, Chambrade ML, et al. Addressing the desert kites phenomenon and its global range through a multi-proxy approach. J. Archaeol. Method Theory. 2015; 22(4): 1093–1121. 7. Crassard R, Abu-Azizeh W, Barge O, Brochier JE´ , Chahoud J, Re´gagnon E. The use of desert kites as hunting mega-traps: functional evidence and potential impacts on socioeconomic and ecological spheres. J. of World Prehist. 2022; 35(1): 1–44. 8. Abu-Azizeh W, Tarawneh M, Crassard R, Sa´ nchez Priego JA. Discovery and excavation of desert kites in the southeastern badia of Jordan. In: Betts A, van Pelt P, editors. The Gazelle’s Dream. Game Drives of the Old and New Worlds. Sydney: University of Sydney Press; 2021. 9. Brochier JE´ , Barge O, Karakhanyan A, Kalantarian I, Chambrade ML, Magnin F. Kites on the margins: the Aragats kites in Armenia. Pale´orient. 2014; 40(1): 25–53. 10. Betts A. The Harra and the Hamad: Excavations and Surveys in Eastern Jordan. Sheffield: Continuum International Publishing Group; 1998. 11. Holzer A, Avner U, Porat N, Horwitz K. Desert kites in the Negev Desert and Northeast Sinai: their func- tion, chronology and ecology. J. Arid Environ. 2010; 74: 806–817. 12. Abu-Azizeh W, Tarawneh MB. Out of the harra: desert kites in south-eastern Jordan. New results from the South Eastern Badia Archaeological Project. Arabian Archaeology and Epigraphy. 2015; 26: 95– 119. PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 28 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures 13. Barge O, Brochier JE´ , Crassard R. Morphological diversity and regionalization of kites in the Middle East and Central Asia. Arabian Archaeology and Epigraphy. 2015; 26(2): 162–176. 14. Barge O, Brochier JE´ , Re´gagnon E, Chambrade ML, Crassard R. Unity and diversity of the kite phe- nomenon: a comparative study between Jordan, Armenia and Kazakhstan. Arabian Archaeology and Epigraphy. 2015; 26(2): 144–161. 15. Barge O, Brochier JE´ , Karakhanyan A. Northernmost kites? Quat. Int. 2016; 395: 104–112. 16. Barge O, Brochier JE´ , Deom JM, Sala R, Karakhanyan A, Avagyan A, et al. The “desert kites” of the Ustyurt plateau. Quat. Int. 2016; 395: 113–132. 17. Barge O, Abu-Azizeh W, Brochier JE´ , Crassard R, Re´ gagnon E, Nouˆ s C. Desert-kites et constructions apparente´ es: de´couvertes re´ centes et e´ volution de l’aire d’extension ge´ ographique. Pale´orient. 2020; 46(1–2): 179–200. 18. van Loon M. The Oriental Institute excavations at Mureybit, Syria: Preliminary report on the 1965 cam- paign: Part I: Architecture and general finds. J. Near East. Stud. 1968; 27(4): 265–282. 19. Brenet M, Sa´nchez Priego JA, Ibañez Estevez JJ. Les pierres de construction taille´ es en calcaire et les herminettes en silex du PPNA de Jerf al Ahmar (Syrie), analyses technologique et expe´ rimentale. In: Bourguignon L, Ortega I, Frère-Sautot M, editors. Pre´ histoire et approche expe´rimentale. Montagnac: Editions M. Mergoil; 2002. pp. 121–164. 20. Berge C. The theory of graphs. Dover books on mathematics series. New York: Dover Publications; 2013. 21. Madi K, Seba H, Kheddouci H, Bichot CE, Barge O, Chataigner C, et al. Kite recognition by means of graph matching. In: Cheng-Lin L, Bin L, Kropatsch WG, Cheng J, editors. Graph-Based Representa- tions in Pattern Recognition—10th International Workshop, GbRPR 2015, Beijing, China, May 13–15, 2015, Proceedings. Lecture Notes in Computer Science 9069. New York: Springer; 2015. pp.118-127. 22. Madi K, Seba H, Kheddouci H, Barge O. Graph-based approach for Kite recognition. Pattern Recognit. Lett. 2017; 87(1): 186–194. 23. Brun L, Foggia P, Vento M. Trends in graph-based representations for pattern recognition. Pattern Recognit. Lett. 2020; 134: 3–9. 24. Marshack A. Upper palaeolithic symbol systems of the Russian plain. Cognitive and comparative analy- sis. Curr. Anthropol. 1979; 20(2): 271–305. 25. Kozlowski J. L’art de la pre´histoire en Europe orientale. Paris: CNRS Editions; 1992. 26. Zu¨ chner C. The scaliform sign of Altamira and the origin of maps in prehistoric Europe. In: Moure A, edi- tor. El Hombre Fo´ sil 80 años despue´s. Santander: Universidad de Cantabria; 1996. pp. 325–343. 27. Svoboda J. Symbolisme gravettien en Moravie. Espace, temps et formes. Bulletin de la Socie´ te´ Pre´ his- torique de L’Ariège-Pyre´ ne´ es. 1997; 52: 87–104. 28. Svoboda J. On landscapes, maps and Upper Paleolithic lifestyles in the central European corridor: the images of Pavlov and Předmostı´. Velei. 2017; 34: 67–74. 29. Utrilla P, Mazo C, Sopena MC, Martı´nez-Bea M, Domingo R. A Palaeolithic map from 13,660 cal BP: engraved stone blocks from the Late Magdalenian in Abauntz Cave (Navarra, Spain). J. Hum. Evol. 2009; 57: 99–111. https://doi.org/10.1016/j.jhevol.2009.05.005 PMID: 19625071 30. Brocard M. Les gravures rupestres à parcellaire. E´ tudes Rurales. 2005; 175: 9–28. 31. Arcà A. The topographic engravings of Alpine rock-art: fields, settlements and agricultural landscapes. In: Chippindale C, Nash G, editors. Pictures in place: the figured landscapes of rock-art. Cambridge: Cambridge University Press; 2004. pp. 318–349. 32. Arcà A. Maps in alpine rock-art: cultivations and cultures, land plots and societies from Neolithic to Iron Age. Bollettino del Centro camuno di Studi preistorici. 2016; 41: 7–32. 33. Casti E. Bedolina: map or tridimensional model? Cartographica: the international Journal for geographic Information and Geovisualization. 2018; 53(1): 15–35. 34. Nicolas C, Pailler Y, Ste´phan P, Pierson J, Aubry L, Le Gall B, et al. La carte et le territoire: la dalle grave´ e du Bronze ancien de Saint-Be´ lec (Leuhan, Finistère). Bulletin de la Socie´ te´ Pre´ historique Fran- c¸aise. 2021; 118: 99–146. 35. Bar-Yosef O. Prehistory of the Levant. Annu. Rev. Anthropol. 1980; 9: 101–133. 36. Cauvin J., The Birth of the Gods and the Origins of Agriculture. Cambridge: Cambridge University Press; 2000. 37. Mellaart J. Excavations at C¸ atal Hu¨ yu¨ k, 1963, third preliminary report. Anatolian Studies. 1964; 14: 39– 193. 38. Hodder I. C¸ atalho¨yu¨ k in the context of the Middle Eastern Neolithic. Annu. Rev. Anthropol. 2007; 36: 105–120. PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 29 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures 39. Schmitt AK, Danisˇı´k M, Aydar E, Şen E, Ulusoy İ, Lovera OM. Identifying the volcanic eruption depicted in a Neolithic painting at C¸ atalho¨ yu¨k, Central Anatolia, Turkey. PLoS ONE. 2014; 9(1): e84711. 40. Meek TJ. Excavations at Nuzi, volume III, old Akkadian, Sumerian, and Cappadocian texts from Nuzi. Cambridge, MA: Harvard University Press; 1935. 41. Millard AR. Cartography in the Ancient Near East. In: Woodward D, Lewis GM, editors. The history of cartography: volume 2, book 3, Cartography in the traditional African, American, Arctic, Australian, and Pacific societies. Chicago: University of Chicago Press; 1987. pp. 107–116. 42. Harding GL. The cairn of Hani. Annual of the Department of Antiquities of Jordan. 1953; 2: 8–56. 43. Betts AVG, Helms S. Rock art in eastern Jordan: kite carvings? Pale´ orient. 1986; 12(1): 67–72. 44. Hershkovitz I, Ben-David Y, Arensburg B, Goren A, Pinchasov A. Rock engravings in southern Sinai. In: Gvirtzman G, Shmueli A, Gardos Y, Beit-Arieh I, Harel M, editors. Sinai Part 2, Human Geography. Givatayim: Eretz Ha-Tzvi; 1987. pp. 605–616. 45. MacDonald MC. Of Rock-art, ‘Desert Kites’ and Mesayid. In: Naumkina VV, Sedov AV, Smulyanskaya IM, editors. Arabia Vitalis: Arabskij Vostok, Islam, drevnyaya, Araviya: Sbornik Naychnykh statej, pos- vyashchennyj 60-letiyu. Moscow: Rossijskaya Akademiya Nauk; 2005. pp. 332–345. 46. Rollefson GO, Wasse A, Rowan Y. Images of the environment: Rock art and the Exploitation of the Jor- danian Badiah. Journal of Epigraphy and Rock Drawings. 2008; 2: 17–51. 47. Hill AC, Rowan YM, Wasse A, Rollefson GO. Inscribed landscapes in the Black Desert: Petroglyphs and kites at Wisad Pools, Jordan. Arabian Archaeology and Epigraphy. 2020; 31(2): 245–262. 48. Van Berg PL, Vander Linden M, Lemaitre S, Cauwe N, Picalause V. Desert-kites of the Hemma plateau (Hassake, Syria). Pale´orient. 2004; 30(1): 89–99. 49. Betts AVG, Yagodin V. Hunting traps on the Ustiurt plateau, Uzbekistan. In: Christian D, Benjamin C, editors. in Realms of the Silk Roads: Ancient and Modern Silk Road Studies, 4. Brepols: Turnhout; 2000. pp. 29–45. 50. Lemke A. Literal niche construction: Built environments of hunter-gatherers and hunting architecture. J. Anthropol. Archaeol. 2021; 62(82): 101276. 51. Harvey PDA. The History of Topographical Maps: Symbols, Pictures and Surveys. London: Thames and Hudson; 1980. 52. Talbert RJA, Unger RW. Cartography in Antiquity and the Middle Ages. Fresh Perspectives, New Meth- ods. Leiden: Brill; 2008. 53. Bagrow L. History of Cartography. London: C.A. Watts and Co; 1964. 54. Pickles J. A History of Spaces. Cartographic Reason, Mapping and the Geo-coded World. London: Routledge; 2003. 55. Shcheglov DA. Eratosthenes’ contribution to Ptolemy’s map of the world. Imago Mundi. 2017; 69(2): 159–175. 56. Meece S. A bird’s eye view–of a leopard’s spots. The Cataho¨ yu¨k ‘map’ and the development of carto- graphic representation in prehistory. Anatolian Studies. 2006; 56, 1–16. 57. Gartner WG. Mapmaking in the central Andes. In: Woodward D, Lewis GM, editors. The history of car- tography: volume 2, book 3, Cartography in the traditional African, American, Arctic, Australian, and Pacific societies. Chicago: University of Chicago Press; 1998. pp. 257–297. 58. Hsu ML. The Qin maps: A clue to later Chinese cartographic development. Imago Mundi. 1993; 45(1): 90–100 (1993). 59. Delano Smith C. Cartography in the Prehistoric Period in the Old World: Europe, the Middle East, and North Africa. In: Harley JB, Woodward D, editors. The history of cartography, volume 1, cartography in prehistoric, ancient, and medieval Europe and the Mediterranean. Chicago: University of Chicago Press; 1987. pp. 54–101. 60. Wood D. The fine line between mapping and map making. Cartographica. 1993; 30; 50–60. 61. McDonald J, Veth P. Rock art in arid landscapes: Pilbara and Western Desert petroglyphs. Australian Archaeology. 2013; 77: 66–81. 62. Flood J. Linkage between rock-art and landscape in Aboriginal Australia. In: Chippindale C, Nash G, editors. The figured landscape of rock-art: looking at pictures in place. Cambridge: Cambridge Univer- sity Press; 2004. pp. 182–200. 63. Sutton P. Icons of country: topographic representations in classical aboriginal traditions. In: Woodward D, Lewis GM, editors. The history of cartography: volume 2, book 3, Cartography in the traditional Afri- can, American, Arctic, Australian, and Pacific societies. Chicago: University of Chicago Press; 1998. pp. 353–386. PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 30 / 31 PLOS ONEThe oldest plans to scale of humanmade mega-structures 64. Bassett T. Indigenous mapmaking in intertropical Africa. In: Woodward D, Lewis GM, editors. The his- tory of cartography: volume 2, book 3, Cartography in the traditional African, American, Arctic, Austra- lian, and Pacific societies. Chicago: University of Chicago Press; 1998. pp. 24–48. 65. 66. 67. Lewis GM. Maps, mapmaking and map use by Native North Americans. In: Woodward D, Lewis GM, editors. The history of cartography: volume 2, book 3, Cartography in the traditional African, American, Arctic, Australian, and Pacific societies. Chicago: University of Chicago Press; 1998. pp. 51–172. d’Errico F. L’Art Grave´ Azilien. De la Technique à la Signification. Paris: CNRS; 1994. Fritz C. La Gravure dans l’Art Mobilier Magdale´ nien. Du Geste à la Repre´ sentation. Paris: Maison des Sciences de l’Homme; 1999. 68. Cassen S, Robin G. Recording art on Neolithic stelae and passage tombs from digital photographs. J. Archaeol. Method Theory. 2010; 17: 1–14. 69. Lanos P, Dufresne P. ChronoModel version 2.0: Software for Chronological Modelling of Archaeological Data using Bayesian Statistics. https://chronomodel.com; 2019. 70. Reimer PJ, Austin WEN, Bard E, Bayliss A, Blackwell PG, Bronk Ramsey C, et al. The IntCal20 North- ern Hemisphere Radiocarbon Age Calibration Curve (0–55 cal kBP). Radiocarbon. 2020; 62(4): 725– 757. 71. Murray AS, Wintle AG. Luminescence dating of quartz using an improved single-aliquot regenerative- dose protocol. Radiat. Meas. 2000; 32: 57–73. 72. Galbraith RF, Roberts RG. Statistical aspects of equivalent dose and error calculation and display in OSL dating: An overview and some recommendations. Quat. Geochronol. 2012; 11: 1–27. 73. Preusser F, Kasper HU. Comparison of dose rate determination using high-resolution gamma spec- trometry and inductively coupled plasma-mass spectrometry. Anc. TL. 2001; 19: 19–23. 74. Degering D, Degering A. Change is the only constant—time-dependent dose rates in luminescence dat- ing. Quat. Geochronol. 2020; 58: 101074. PLOS ONE \| https://doi.org/10.1371/journal.pone.0277927 May 17, 2023 31 / 31 PLOS ONE
10.1371_journal.pone.0283795	RESEARCH ARTICLE The Polish adaptation of the measurements of rule-governed behaviors: Generalized Pliance Questionnaire, Generalized Tracking Questionnaire and Generalized Self-Pliance Questionnaire Joanna DudekID 1☯, Maria Cyniak-Cieciura2☯, Paweł OstaszewskiID 3☯ a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 1 Faculty of Psychology in Warsaw, Center for Behavioral Research in Decision Making, SWPS University, Warsaw, Poland, 2 Advanced Clinical Studies and Therapy Excellence Center, Institute of Psychology, SWPS University, Warsaw, Poland, 3 Center for Behavioral Research in Decision Making, Institute of Psychology, SWPS University, Warsaw, Poland ☯ These authors contributed equally to this work. jdudek@swps.edu.pl OPEN ACCESS Abstract Citation: Dudek J, Cyniak-Cieciura M, Ostaszewski P (2023) The Polish adaptation of the measurements of rule-governed behaviors: Generalized Pliance Questionnaire, Generalized Tracking Questionnaire and Generalized Self- Pliance Questionnaire. PLoS ONE 18(4): e0283795. https://doi.org/10.1371/journal.pone.0283795 Editor: Marco Innamorati, Universita degli Studi Europea di Roma, ITALY Received: November 4, 2022 Accepted: March 19, 2023 Published: April 5, 2023 Copyright: © 2023 Dudek et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Data and codes are available at: DOI:10.17605/OSF.IO/PS4RM. Funding: The research was supported financially by the Faculty of Psychology in Warsaw, SWPS University. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. In some circumstances rule-governed behavior, a behavior that is governed by verbal rules instead of environmental consequences, may be beneficial for human beings. At the same time, rigid rule following is associated with psychopathology. Thus measurement of rule- governed behavior may be of special use in a clinical setting. The aim of this paper is to assess the psychometric properties of Polish adaptations of three questionnaires measuring generalized tendency to engage in various types of rule-governed behaviors: Generalized Pliance Questionnaire (GPQ), Generalized Self-Pliance Questionnaire (GSPQ), General- ized Tracking Questionnaire (GTQ). A forward-backward method was used for translation. Data was collected from two samples: general population (N = 669) and university students (N = 451). To measure the validity of the adapted scales the participants filled in a set of self-assessed questionnaires: Satisfaction with Life Scale (SWLS), Depression, Anxiety, and Stress Scale– 21 (DASS-21), General Self-Efficacy Scale (GSES), Acceptance and Action Questionnaire–II (AAQ-II), Cognitive Fusion Questionnaire (CFQ), Valuing Question- naire (VQ) and Rumination—Reflection Questionnaire (RRQ). The exploratory and confir- matory analyses confirmed the unidimensional structure of each of the adapted scales. All of those scales presented good reliability (internal consistency measured with Cronbach Alpha) and item-total correlations. The Polish versions of questionnaires presented signifi- cant correlations in the expected directions with relevant psychological variables in line with the original studies. The measurement occurred invariant across both samples as well as gender. The results provide evidence that Polish versions of GPQ, GSPQ and GTQ present sufficient validity and reliability to be used in the Polish-speaking population. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 1 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ Introduction To provide a behavior-analytic account of complex human behavior such as thinking and problem solving, in 1966 Skinner [1] introduced the term rule-governed behavior (RGB), defined as behavior that is under the control of rules and instructions, in contrast to contin- gency-shaped behaviors which are under control of direct contingencies in the environment. The functional behavioral analysis of the concept was first proposed almost 20 years later by Zettle and Hayes [2] and elaborated in detail within the framework of the Relational Frame Theory [3]. Beginning with Skinner, researchers emphasize that the ability to generate and follow verbal rules may be beneficial especially in contexts where learning through direct experience is dan- gerous (e.g. look both ways before crossing the street) or contingencies are delayed (e.g. attend classes and study to get the diploma). Thus, RGB helps people to achieve goals, learn from the experience of others, and cope with events before they occur [4]. However, under certain cir- cumstances, RGB can also produce undesired consequences, such as insensitivity to real envi- ronmental contingencies, rigidly following verbal rules despite their effectiveness or even harmful outcome of rule following, or persistent avoidance [5–7]. Therefore, the concept of RGB has become particularly important in the domain of clinical behavior analysis, as it provides both explanation of the development of a number of psycho- pathological symptoms [5–8], and helps to develop psychotherapeutic interventions, such as Acceptance and Commitment Therapy (ACT) with its focus on the psychological flexibility model [9]. To support both basic and clinical research on the RGB, reliable and valid methods of assessing the behaviors are required. The aim of the present paper is to present the valida- tion process of the three recently developed questionnaires measuring two functional types of RGB, generalized pliance (and self-pliance) and generalized tracking [10–12] in a Polish population. Different functional types of RGB were first introduced by Zettle and Hayes [2], who distin- guished pliance and tracking as two most fundamental functional classes of RGB, and a third type, augmenting, operating together with the two former classes by verbally changing the reinforcing or punishing strength of consequences included in the rules. Pliance is defined as a functional class of rule-governed behavior under the control of history of multiple interactions in which the speaker provides the listener with the reinforcement contin- gent on the correspondence between the rule (e.g. do not touch hot pot) and the relevant behav- ior (refraining from touching the pot). An example of reinforcement in such circumstances may be praising the individual (e.g. great that you did not touch the pot [2, 3, 13]). Taking into account that the listener and the speaker may be the same person [14], under some circum- stances those rules can also be generated by the individual and then called self-rules [10, 11]. Pliance, being the first type of rule-following developed [13], over-generalizes at some point in the child’s development. Yet, social interactions lead to contextualizing pliance (so that the child can distinguish when it is appropriate) and establishing tracking to help her to recognize natural consequences of her behavior [5, 7]. Lack of such learning history may lead to generalized pliance [11]. Generalized pliance can be problematic when it becomes the main source of impact on human behavior, as socially mediated consequences are less predictable and controllable than other types of consequences which may lead to lower contact with sources of positive reinforcement. Individuals displaying generalized pliance may be particularly insensitive to direct contingencies (e.g. a person believes that in order to obtain social approval she needs to start smoking, so she does that ignoring the negative consequences of smoking). As the child develops and gains fluency in relational framing, plys become more abstract (e.g. a person believes she needs to be a ‘good’ PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 2 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ in order to be loved [15]) which may increase the likelihood social approval to be the main source of reinforcement. As many of social rules treat aversive private events as events that need to be controlled or avoided, individuals displaying generalized pliance are more likely to engage in rigid patterns of behaviors and a tendency to engage in experiential avoidance—attempts to avoid, control or get rid of unwanted internal experiences even when doing so is harmful for the individual [5, 16]. Considering that generalized pliance may lead to losing the contact with the sources of posi- tive reinforcement and behaviors being controlled by negative reinforcement (avoidance) it is considered a risk factor for developing psychopathology (e.g. [5, 6, 8] and psychological inflex- ibility–difficulties in engaging in meaningful actions due to the presence of unpleasant internal experiences that the person wants to avoid, get rid of or control [4]. In child development, pliance is seen as a condition to develop tracking [13]. In contrast to pliance, tracking is sensitive to the direct environmental contingencies, so that if they change the individual is changing the behavior accordingly [2] and it may be regarded as a flexible rule-governed behavior [12]. Tracking is a functional class of behaviors under the control of the history of multiple exemplars in which following the rule leads to natural consequences derived from the way the world is arranged (e.g. following the rule “when it is cold, wear a warm coat” leads to feeling warm even when it is cold outside [2, 3, 13]). Generalized tracking is a pattern of behaviors that may be developed when an individual has been exposed to multi- ple interactions in which she has been encouraged to observe and describe functional relation- ships among events, e.g. recognize natural consequences of her behavior. The individual engaged in tracking, behaving both as speaker and listener, learns to establish functional rela- tionships among events and adjust her behavior accordingly [12]. Despite the interest in rule- governed behaviors especially in the area of contextual behavioral science, there are a number of limitations to experimental research investigating pliance and tracking [17]. One of the pro- posed explanations involves noting that both pliance and tracking are listener-oriented con- cepts [18] and therefore they cannot be produced by the speakers as the participants’ personal learning history may influence their performance more than the experimental rules [11, 12]. Thus, researchers proposed alternative strategy of measuring pliance and tracking by devel- oping self-report measures, which explore the perspective of the listener and investigate the individual’s learning history and personal experience with formulating and following rules [11, 12]. A measure of generalized pliance, Generalized Pliance Questionnaire (GPQ, [11]), self-pliance, Generalized Self-Pliance Questionnaire (GSPQ, [10]), and a measure of general- ized tracking, Generalized Tracking Questionnaire (GTQ, [12]), were created. Empirical evidence with the GPQ shows that generalized pliance is connected to various aspects of psychological inflexibility and measures of distress [11]. Higher scores in the GPQ were predictive of lower levels of mindfulness and sensitivity to changing contingencies [19]. Generalized pliance was positively correlated with repetitive negative thinking, dysfunctional attitudes, difficulties in valued living, and negatively correlated with life satisfaction [11]. Generalized tracking measured with GTQ was negatively correlated with generalized pliance, experiential avoidance, tendency to ruminate and emotional symptoms and positively correlated with valued living, life satisfaction and general self-efficacy and a wide range of executive functions [12]. Measuring self-reported patterns of rule-governed behaviors, such as the GPQ and GTQ, may lead to development of research on complex human behavior by broadening the knowl- edge on rule-governed behaviors, its impact and development (e.g. differences across age, gen- der, and cultures). These instruments may also be used to explain variability of results in experimental analyses, predict development of psychopathology and behavioral rigidity, and eventually, to analyze mediators and moderators of psychological interventions (e.g. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 3 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ Acceptance and Commitment Therapy). Thus, it seems important to provide researchers and practitioners in the area of clinical behavior analysis and contextual behavioral science with appropriate questionnaires in countries and cultures in which people speak other languages than English or Spanish, such as Polish. The strategy of adopting already existing question- naires with good psychometric qualities seems a good step, because it allows multilingual and multicultural comparisons of the same phenomena. We hope that the adaptation of GPQ. GSPQ, and GTQ into Polish language will aid in the development of contextual behavioral sci- ence in general, allow for multicultural analysis of the RGB, and most of all, provide a growing number of researchers and practitioners in Poland with valid questionnaires measuring RGB, widening the scope of both basic and applied research conducted. Materials and methods Scales’ adaptation The original items of the Generalized Pliance Questionnaire (GPQ18–18 items, and its shorter version—GPQ9), Generalized Tracking Questionnaire (GTQ– 11 items), and Generalized Self-Pliance Questionnaire (GSPQ– 12 items) were translated to Polish by three independent translators, who were practitioners in third wave cognitive-behavioral approaches and/or experts in contextual behavioral science. Then, three independent versions of the Polish items were presented to three independent judges–scientists and practitioners in the area of contex- tual and behavioral science as well as third wave cognitive-behavioral therapy. Final versions of items’ translations were chosen based on the judges’ opinions, back-translated to English and accepted by the author of the original questionnaires. The original instructions and response scales were kept (see [10–12], thus each item was rated on a scale from 1 to 7 (with 7 = always true and 1 = never true). These versions of the scales were used in the study aiming at verifying their psychometric properties. Participants The psychometric properties of Polish versions of GPQ18, GPQ9, GTQ, and GSPQ were checked in two independent studies and different samples described below. Sample A–students’ sample The participants were recruited via university SONA research panel. Students received credits for taking part in the study in accordance with the university policy. The study was closed after the data from at least 400 participants was collected. The only exclusion criteria were being less than 18 years old. A total of 451 people completed the study: 371 women (82.3%) and 80 men (17.7%) in the age of 18–64 (M = 27.61, SD = 8.27). Most of them had a secondary educational level (N = 245, 54.3%), 206 (40.7%) declared higher education. Most of the participants were living in the city with more than 500 000 residents (N = 252, 55.9%) and the least—in the rural area (N = 48, 10.6%). A total of 87 people (19.3%) were living in a city with less than 100 000 residents and 64 people (14.2%) in a city with 100–500 000 residents. Sample B–a general sample The participants were recruited by the Pollster research panel (https://pollster.pl/), the only exclusion criteria were being less than 18 years old. The data collection was stopped when data from at least 600 participants was collected. A total of 669 people completed the study: 333 women (49.8%) and 336 men (50.2%) in the age of 18–65 (M = 40.96, SD = 13.49). Most of them had a secondary educational level (N = 360, 53.8%), 272 (40.7%) declared higher PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 4 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ education and 37 (5.5%) finished only primary school. Most of the participants were living in the city with less than 100 000 residents (N = 254, 38%), 25.7% (N = 172)—in the city with 100–500 000 residents, 18.7% (N = 125) in the rural area, and 17.6% (N = 118) in a city with more than 500 000 residents. Procedure The study was conducted online between March—June 2021. The participants were asked to fill in a set of self-report questionnaires: a short demographic questionnaire, the Polish ver- sions of GPQ18, GTQ and GSPQ and a few other measures aiming at verifying GPQ18, GPQ9, GTQ and GSPQ’s validity (their short description is presented below). All of the partic- ipants signed an informed consent. The study was conducted following the Declaration of Hel- sinki and received a positive opinion from the local Ethics Committee (Nr 5/2021). Other measures Satisfaction with Life Scale. The Polish version of the Satisfaction with Life Scale (SWLS [20, 21]) was used to measure self-perceived well-being. It consists of five items. Participants rated each item on a 7-point scale, ranging from 7 = strongly agree to 1 = strongly disagree. Higher scores indicate a greater level of life satisfaction. Medium to large negative correlations were expected between SWLS and GPQ18, GPQ9, GSPQ, and positive—with GTQ. Depression, Anxiety, and Stress Scale– 21. The Polish version of Depression, Anxiety, and Stress Scale– 21 (DASS-21 [22, 23]) was used to measure the level of depression, anxiety and stress symptoms. It consists of 21 items and a 4-point Likert-type scale (3 = applied to me very much, or most of the time; 0 = did not apply to me at all). It contains three subscales: Depression, Anxiety, and Stress with higher scores indicating higher levels of symptoms. Medium to strong positive correlations were expected between DASS21 and GPQ18, GPQ9, and GSPQ, as well as negative with GTQ. General Self-Efficacy Scale. The Polish version of General Self-Efficacy Scale (GSES [24, 25]) was used to measure self-perceived self-efficacy. It comprises ten items assessed on a 4-point Likert scale (1 = not at all true, 4 = exactly true), which enable to calculate a general score (the higher the score, the higher the level of general self-efficacy). Medium negative cor- relations were expected between GSES and GPQ18, GPQ9, GSPQ, and positive with GTQ. Acceptance and Action Questionnaire–II. The Polish version of the Acceptance and Action Questionnaire-II (AAQ-II [26, 27]) was used to measure psychological inflexibility. It consists of seven statements. Participants rate each statement on a 7-point scale, ranging from 1 = never true to 7 = always true. Higher scores indicate higher psychological inflexibility. Medium positive correlations were expected between AAQ-II and GPQ18, GPQ9, GSPQ, and negative with GTQ. Cognitive Fusion Questionnaire. The Polish version of Cognitive Fusion Questionnaire (CFQ [28, 29]) was used to measure the level of cognitive fusion. The CFQ consists of seven items with a 7-point Likert-type scale (7 = always true; 1 = never true). Medium to strong posi- tive correlations were expected between the CFQ and GPQ18, GPQ9, GSPQ, and negative– with GTQ. Valuing Questionnaire. The Polish version of Valuing Questionnaire (VQ [28, 30]) was used to measure the general valued living during the past week. VQ consists of ten items and a 6-point Likert scale (6 = completely true; 0 = not at all true). It contains two subscales: Progress (defined as enactment of values, including clear awareness of what is personally important, and perseverance), as well as Obstruction (defined as disruption of valued living due to avoid- ance of unwanted experience and distraction from values). It was expected that the VQ PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 5 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ Progress scale will be negatively correlated with GPQ18, GPQ9, GSPQ, and positively with GTQ. Positive correlations were expected between VQ Obstruction scale and GPQ18, GPQ9, GSPQ, and negative with GTQ. Rumination—Reflection Questionnaire. The Polish version of Rumination—Reflection Questionnaire– 12 (RRQ12 [31, 32]) was used to measure the level of focus on one’s own expe- riences (rumination) motivated by fear, and the involvement in getting to know oneself (reflec- tion) motivated by curiosity. RRQ12 consists of twelve items assessed with a 5-point Likert scale (1 –I strongly disagree, 5 –I strongly agree) and contains two subscales–Rumination and Reflection. It was expected that RRQ Rumination scale would be positively correlated with GPQ18, GPQ9, GSPQ, and negatively with GTQ. Negative correlations were expected between RRQ Reflection scale and GPQ18, GPQ9, GSPQ, and positive with GTQ. The reliability coefficients (Cronbach Alphas) of the scales were satisfactory and are pre- sented in Table 3. Statistical analyses There were no missing values within GPQ18, GTQ and GSPQ items in both samples. A cross- validation procedure was applied with the analyses done on a data from a students’ sample A and then replicated in a general sample B. Exploratory factor analysis (EFA) was conducted only on sample A and confirmatory factor analysis (CFA) only on sample B. The measurement invariance across samples (A and B) and gender was checked. Then, the corrected item-total correlations and Cronbach Alpha coefficients were calculated in two samples respectively. Finally, to provide information about the validity of the scales, r-Pearson correlations between GPQ18, GPQ9, GTQ, GSPQ and other measures were calculated (in two samples respectively). All the analyses were conducted with the use of SPSS v. 25, FACTOR v.11.05.01 [33] and R lavaan package [34]. Results Exploratory factor analysis The EFA was conducted with the use of FACTOR v. 11.05.01 based on the data from sample A. Data included in the analyses was categorical and according to Mardia’s test it did not meet the assumptions of the multivariate normal distribution, due to the exceeded kurtosis values (The results of Mardia’s test for the GPQ18 items: b = 31.68, Z(1140) = 23.81, p = 1.00 for skewness and b = 445.22, Z = 33.72, p < .001 for kurtosis. For the GPQ9 items: b = 5.10, Z (165) = 23.81, p = 1.00 for skewness and b = 118.02, Z = 14.35, p < .001 for kurtosis. For the GTQ items: b = 13.48, Z(286) = 1013.46, p = 1.00 for skewness and b = 202.88, Z = 37.59, p < .001 for kurtosis. For the GSPQ items: b = 10.90, Z(364) = 818.99.46, p = 1.00 for skewness and b = 208.94, Z = 23.71, p < .001 for kurtosis). Data was analyzed with the use of robust diago- nally weighted least squares (RDWLS) extraction method with polychoric correlations and robust Promin rotation [33]. The number of dimensions was determined by means of the opti- mal implementation of parallel analysis (PA [35]). The Unidimensional Congruence (UniCo), Explained Common Variance (ECV), and Mean of Item Residual Absolute Loadings (MIR- EAL) indexes were used to assess the unidimensionality. Values larger than .95 and .85 in UniCo and ECV, respectively, as well as a value lower than .30 for the MIREAL suggest that data can be treated as essentially unidimensional [36]. In all the analyses the 95% bootstrap confidence intervals were estimated based on 500 samples. GPQ18. The Bartlett’s statistic was statistically significant (5102.1(153), p < .001), and the result of the Kaiser-Meyer-Olkin (KMO) test was satisfactory (.93, 95%CI [.90, .93]). The PA suggested extracting one factor accounting for 55.08% of variance (eigenvalue = 9.10). Table 1 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 6 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ shows that factor loadings were high for all the items: from .50 (item 2) to .86 (item 13). The UniCo, ECV and MIREAL values suggest that the data of the GPQ-18 can be treated as unidi- mensional (UniCo = .97 (95% CI [.96, .99], ECV = .90, (95%CI [.86, .90], MIREAL = .22, 95% CI [.19, .24]). GPQ9. The Bartlett’s statistic was statistically significant (2375.6(36), p < .001), and the result of the Kaiser-Meyer-Olkin (KMO) test was satisfactory (.89, 95%CI [.85, .90]). The PA suggested extracting one factor accounting for 59.96% of variance (eigenvalue = 4.99). Table 1 shows that factor loadings were high for all the items: from .49 (item 1) to .87 (item 4). The UniCo, ECV and MIREAL values suggest that the data of the GPQ-9 can be treated as unidi- mensional (UniCo = .95 (95% CI [.92, .98], ECV = .84, (95%CI [.81, .87], MIREAL = .28, 95% CI [.24, .32]). GTQ. The Bartlett’s statistic was statistically significant (2742.6(55), p < .001), and the result of the Kaiser-Meyer-Olkin (KMO) test was satisfactory (.92, 95%CI [.88, .92]). The PA suggested extracting one factor accounting for 60.71% of variance (eigenvalue = 6.06). Table 1 shows that factor loadings were high for all the items: from .65 (item 10) to .79 (item 7). The UniCo, ECV and MIREAL values suggest that the data of the GTQ can be treated as unidimen- sional (UniCo = .98 (95% CI [.97, .99], ECV = .88, (95%CI [.86, .91], MIREAL = .22, 95% CI [.17, .24]). GSPQ. The Bartlett’s statistic was statistically significant (2762.5(66), p < .001), and the result of the Kaiser-Meyer-Olkin (KMO) test was satisfactory (.91, 95%CI [.87, .92]). The PA suggested extracting one factor accounting for 59.69% of variance (eigenvalue = 6.07) 58.74% of variance (eigenvalue = 6.14). Table 1 shows that factor loadings were high for all the items: from .56 (item 7) to .82 (item 5). The UniCo, ECV and MIREAL values suggest that the data of the GSPQ can be treated as unidimensional (UniCo = .98 (95% CI [.98, .99], ECV = .89, (95% CI [.87, .91], MIREAL = .19, 95% CI [.14, .20]). Summarizing, the EFA results suggest that all the scales measure unidimensional latent con- structs and the one-factor solutions explain a significant portion of variance in each case. Confirmatory factor analysis The CFA was conducted with the use of the R lavaan package to analyze the fit of the one-fac- tor model of the GPQ18, GPQ9, GTQ and GSPQ in a general sample B. A weighted least squares–mean (WLSM) estimation method with polychoric correlations was utilized. Good- ness of fit was evaluated using the robust chi-square test, robust root-mean-square error of approximation (RMSEA), robust comparative fit index (CFI), the robust Tucker-Lewis index (TLI), and standardized root-mean-square residual (SRMR; the last one only in CFA). Hu and Bentler [37] proposed the following criteria of good model fit: RMSEA�.10; SRMR�.08; CFI�.90; TLI�.95, however recently the controversies to their application to categorical data have been raised [38]. Shi and Maydeu-Olivares [39] showed that SRMR is less sensitive to the choice of estimator, thus it is also reported. Standardized factor loading estimates are shown in Figs 1–4. GPQ18. The one-factor model exhibited a non-satisfactory fit due to the high value of robust RMSEA: robust χ2(135) = 2729.54, p < .001; robust RMSEA = .108 (95% CI: [.105, .112]), SRMR = .066, robust CFI = .983, robust TLI = .980. Therefore, we decided to analyse modification indices (MI) with a minimum value = 10. MI are univariate score tests that reflect the improvement of model fit after allowing some of the parameters to be free. After careful examination of MI we decided to modify the model allowing the error terms between items 16 and 17 (MI = 97.52), 10 and 11 (MI = 95.12), 4 and 5 (MI = 91.93), as well as 1 and 2 (MI = 71.56) to correlate. The decision was based on the high MI values as well as semantic PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 7 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ Table 1. The Polish version of GPQ18, GPQ9, GTQ and GSPQ items’ factor loadings and corrected item-total correlations. Items GPQ18 Factor 1 Corrected item-total correlation 1. Mo´j nastro´j zależy od tego, co myślą o mnie moi przyjaciele. 2. Bardzo przejmuję się tym, co myślą o mnie moi przyjaciele. 3. Czuję, że moja praca nie jest warta wysiłku, jeżeli inni ludzie jej nie doceniają. 4. To dla mnie bardzo ważne, aby czuć się akceptowanym przez innych. 5. Żeby czuć się szczęśliwym, potrzebuję być doceniany/a przez innych ludzi. 6. Moje poczucie własnej wartości zależy od tego, co inni ludzie o mnie myślą i mo´wią. 7. Moim gło´wnym celem w życiu jest bycie rozpoznawanym i poważanym przez otaczających mnie ludzi. 8. Duży wpływ na moje decyzje mają opinie innych oso´b. 9. Bardzo przejmuję się tym, aby prezentować idealny obraz siebie. 10. To co robię nic by nie znaczyło, gdyby inni nie mogli tego zobaczyć. 11. Ciężka praca ma wartość tylko wtedy, gdy inni ludzie ją dostrzegają. 12. Jest dla mnie ważne, aby inni ludzie mieli w głowie mo´j pozytywny obraz. 13. Potrzebuję aprobaty ze strony innych oso´b, aby czuć się dobrze ze sobą. 14. Nie mogę zawieść oczekiwań innych oso´b wobec mnie. 15. Przed podjęciem decyzji potrzebuję, aby inni ludzie rozumieli moje powody. 16. Przy podejmowaniu decyzji bardziej doceniam rady innych niż własne zdanie. 17. Przed zrobieniem czegoś ważnego proszę innych o poradę. 18. Obawa przed krytyką powstrzymuje mnie od robienia ro´żnych rzeczy. GPQ9 1. Bardzo przejmuję się tym, co myślą o mnie moi przyjaciele. 2. To dla mnie bardzo ważne, aby czuć się akceptowanym przez innych. 3. Żeby czuć się szczęśliwym, potrzebuję być doceniany/a przez innych ludzi. 4. Moje poczucie własnej wartości zależy od tego, co inni ludzie o mnie myślą i mo´wią. 5. Duży wpływ na moje decyzje mają opinie innych oso´b. 6. To co robię nic by nie znaczyło, gdyby inni nie mogli tego zobaczyć. 7. Ciężka praca ma wartość tylko wtedy, gdy inni ludzie ją dostrzegają. 8. Potrzebuję aprobaty ze strony innych oso´b, aby czuć się dobrze ze sobą. 9. Przy podejmowaniu decyzji bardziej doceniam rady innych niż własne zdanie. GTQ 1. Kiedy dostrzegam, że coś nie działa, pro´buję czegoś innego. 2. Lubię dowiadywać się, jak coś działa i wyciągać swoje własne wnioski. 3. Łatwo dostosowuję się do zmian. 4. Potrafię znaleźć nowe rozwiązania problemo´w. 5. Podejmuję decyzje w oparciu o swoje doświadczenie, a nie o to, co mo´wią inni. 6. Lubię pro´bować ro´żnych podejść, aby zobaczyć kto´re jest lepsze. 7. Jestem dobry/a w znajdywaniu bardziej skutecznych sposobo´w wykonywania zadań. 8. Kiedy zauważam, że coś nie działa, szybko zmieniam swo´j sposo´b postępowania. .681 .497 .615 .740 .782 .844 .695 .796 .607 .645 .627 .776 .857 .670 .700 .654 .588 .719 .493 .758 .812 .867 .745 .663 .652 .859 .578 .679 .690 .668 .778 .692 .723 .789 .745 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 .629 .444 .569 .661 .703 .764 .657 .757 .560 .586 .561 .727 .796 .631 .657 .591 .544 .674 .428 .659 .715 .758 .706 .582 .560 .779 .511 .614 .632 .596 .702 .634 .654 .711 .676 (Continued ) 8 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ Table 1. (Continued) Items Factor 1 Corrected item-total correlation 9. Z łatwością uczę się na konsekwencjach swoich działań. 10. Kiedy zdam sobie sprawę, że nie miałem/am racji, zmieniam swo´j sposo´b myślenia i działania. 11. Podejmuję decyzje na bazie uzyskanych wcześniej wyniko´w. GSPQ 1. Czuję, że tracę kontrolę nad życiem, jeżeli moje osobiste sprawy nie są w idealnej ro´wnowadze. 2. Potrzebuję kontrolować swoje lęki, żeby nie czuć się słabym/ą. 3. Szukam odpowiedzi na wszystko, aby nie czuć się głupim/ą. 4. Denerwuje mnie, że nie mogę robić wszystkiego w ten sam sposo´b. 5. Czuję się pogubiony/a, jeżeli nie jestem w stanie wykonać zaplanowanych działań. 6. Muszę dobrze traktować innych ludzi, aby nie czuć się złym człowiekiem. 7. Jeżeli nie trzymam się mocno w jednej pozycji, czuję się słaby/a. 8. Potrzebuję mieć w życiu porządek, aby nie czuć, że tracę kontrolę. 9. Kiedy muszę spędzić dużo czasu na jednej aktywności, zarzucam sobie, że nie poświęcam czasu innym rzeczom. 10. Czuję zdezorientowany/a, gdy nie mogę podążać za swoją rutyną. 11. Aby być usatysfakcjonowanym/ą, muszę wszystko zrobić perfekcyjnie, tak jak sobie tego życzę. 12. Czuję się tak, jakbym gubił/a się w swoim życiu, jeżeli nie spełniam swoich własnych oczekiwań. https://doi.org/10.1371/journal.pone.0283795.t001 .750 .645 .721 .749 .699 .678 .643 .816 .576 .562 .793 .625 .731 .673 .705 .682 .552 .621 .684 .638 .632 .586 .741 .542 .510 .706 .581 .650 .599 .642 equivalence of the paired items. The constraint was released one by one and resulted in the acceptable model fit: robust χ2(131) = 1937.53, p < .001; robust RMSEA = .090 (95% CI: [.087, .094]), SRMR = .056, robust CFI = .988, robust TLI = .986. The same procedure was followed in the case of GPQ9 and GSPQ, which is described below. GPQ9. The one-factor model exhibited a non-satisfactory fit due to the high value of robust RMSEA: robust χ2(27) = 810.56, p < .001; robust RMSEA = .129 (95% CI: [.122, .137]), SRMR = .064, robust CFI = .984, robust TLI = .979. Again, based on modification indices (MI), the error terms between items 10 and 11 (MI = 123.45), and 4 and 5 (MI = 75.76) were allowed to correlate. This improved the model fit: robust χ2(25) = 340.64, p < .001; robust RMSEA = .082 (95% CI: [.074, .90]), SRMR = .044, robust CFI = .994, robust TLI = .992. GTQ. The one-factor model exhibited a satisfactory fit: robust χ2(44) = 680.31, p < .001; robust RMSEA = .089 (95% CI: [.083, .095]), SRMR = .045, robust CFI = .992, robust TLI = .991. GSPQ. The one-factor model exhibited a non-satisfactory fit due to the high value of robust RMSEA: robust χ2(54) = 854.54, p < .001; robust RMSEA = .097 (95% CI: [.092, .103]), Fig 1. Standardized solution of the GPQ18 one-factor model in a Polish sample. https://doi.org/10.1371/journal.pone.0283795.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 9 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ Fig 2. Standardized solution of the GPQ9 one-factor model in a Polish sample. https://doi.org/10.1371/journal.pone.0283795.g002 SRMR = .055, robust CFI = .985, robust TLI = .982. one more time, based on modification indices (MI), the error terms between items 1 and 2 (MI = 75.93), 11 and 12 (MI = 60.53) were allowed to correlate. This improved the model fit: robust χ2(52) = 553.44, p < .001; robust RMSEA = .078 (95% CI: [.072, .084]), SRMR = .048, robust CFI = .991, robust TLI = .989. Summarizing, the results of CFA generally confirmed the one-factor solution obtained in original studies for each scale. In the case of three scales (GPQ18, GPQ9 and GSPQ) the origi- nal model presented a non-acceptable fit and some modifications were applied to obtain at least acceptable fit of the model. The final solutions present non-significant chi-square statis- tics (however, because of the big sample sizes this particular statistic is not a reliable source about the model fit, see [40]), satisfactory robust CFI, robust TLI and SRMR and at least acceptable values of robust RMSEA. Factor loadings were moderately or strongly related to their purported latent factor in the case of each scale. Measurement invariance across samples (A and B) and gender Metric, scalar and strict invariance across both samples (A and B) and gender were conducted. The relative fits of four increasingly restrictive models were compared: the multigroup baseline model (allowing factor loadings to vary across groups while the factor structure was identical across groups (i.e., configural invariance), the metric invariance model (placing equality con- straints on factor loadings across groups), the scalar invariance model (placing equality con- straints on factor loadings and item intercepts), and the strict invariance model (placing equality constraints on factor loadings, item intercepts and residuals). The models were com- pared taking into account the differences in robust RMSEA (ΔRMSEA), CFI (ΔCFI), and TLI (ΔTLI) indexes between nested models, with ΔCFI being regarded as least affected by the model complexity and sample size [41]. Although the chi-square statistics and their differences between the models are also presented, due to the big sample sizes they should not be treated as decisive; a ΔCFI, ΔTLI and ΔRMSEA less than .01 indicated invariance (see [40–43]). Fig 3. Standardized solution of the GTQ one-factor model in a Polish sample. https://doi.org/10.1371/journal.pone.0283795.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 10 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ The results of the analyses are presented in Table 2. Baseline models present well-fit in the case of each scale. When it comes to the invariance, it can be concluded that metric, scalar and strict invariance are supported across both samples A and B in the case of each scale except the GPQ9. At the same time, the measurement can be treated as invariant regardless of the gender in the case of each scale. Validity The bivariate r-Pearson correlations were calculated between GPQ18, GPQ9, GTQ and GSPQ themselves and between adapted scales and other tools measuring life satisfaction, the level of depression, anxiety and stress, the level of perceived general self-efficacy, psychological inflexi- bility, cognitive fusion, general valued living as well as rumination and reflection. The results are presented in Table 3. They generally support the convergent and divergent validity of adapted scales and are consistent across the samples (with differences not being tested directly). The correlation between GPQ18 and GPQ9 were very high in both samples, suggesting that both tools measure the same construct (accordingly with the expectations). Pliance (measured by both GPQ18 and GPQ9) was strongly related to self-pliance, which was also consistent with a priori hypotheses. Tracking was weakly (and negatively) or non-significantly related to pliance and self-pliance, showing that these constructs reflect different and strongly indepen- dent rule-governed behaviors. Pliance (as measured by GPQ18 and GPQ9) and self-pliance was moderately or strongly and positively related to the level of depressive, anxiety and stress symptoms, psychological inflexibility, cognitive fusion, obstruction of valued living and rumination. They showed weak positive or non-significant relationships with reflection, and progress in valued living. The relationship with general self-efficacy and life satisfaction was weak and negative or non-signif- icant. These least results seem less expected, suggesting that possibly life satisfaction and self- efficacy are less affected by the level of presented pliance and self-pliance (the hypothesis worth further testing). Tracking showed to be positively and moderately or strongly related to life satisfaction, gen- eral self-efficacy and progress in valued living and weakly with reflection. It was also negatively related (at least moderately) with depressive, anxiety and stress symptoms, psychological inflexibility, cognitive fusion, obstruction of valued living and rumination. The results are gen- erally consistent with set hypotheses and those obtained in a study of original scale [12]. Discussion The aim of the study was to evaluate the psychometric properties of Polish adaptations of the questionnaires measuring rule-governed behaviors such as pliance, self-pliance and tracking. The EFA and CFA analyses corroborated unidimensional structure with high factor load- ings found in the original validation studies for each of the measurements: GTQ [12], GSPQ Fig 4. Standardized solution of the GSPQ one-factor model in a Polish sample. https://doi.org/10.1371/journal.pone.0283795.g004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 11 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ Table 2. Measurement invariance across samples (A and B) and gender. Model χ2 (df) GPQ18 χ2 diff p diff RMSEA Measurement invariance across samples A and B ∆RMSEA CFI ∆CFI TLI ∆TLI GPQ9 GTQ GSPQ Baseline model Metric invariance Scalar invariance Strict invariance Baseline model Metric invariance Scalar invariance Strict invariance Baseline model Metric invariance Scalar invariance Strict invariance Baseline model Metric invariance Scalar invariance Strict invariance 293.83 (262) 375.51 (279) 539.35 (296) 554.75 (314) 310.61 (262) 447.51 (279) 523.78 (296) 545.62 (314) 77.72 (50) 90.02 (58) 208.73 (66) 213.70 (75) 84.76 (50) 117.92 (58) 169.63 (66) 180.73 (75) Baseline model 63.07 (82) Metric invariance 133.90 (92) Scalar invariance Strict invariance 169.08 (102) 181.70 (113) Baseline model 60.10 (82) Metric invariance 106.37 (92) Scalar invariance Strict invariance 116.34 (102) 133.10 (113) Baseline model Metric invariance Scalar invariance Strict invariance Baseline model Metric invariance Scalar invariance Strict invariance 130.56 (104) 185.09 (115) 235.92 (126) 240.29 (138) 129.82 (104) 158.13 (115) 186.00 (126) 193.24 (138) .060 < .001 .169 .044 .045 .055 .054 -.001 -.010 .001 Measurement invariance across gender < .001 < .001 .011 .046 .051 .054 .053 -.005 -.003 .001 Measurement invariance across samples A and B .372 < .001 .523 .054 .050 .077 .072 .004 -.027 .005 Measurement invariance across gender .003 < .001 .025 .057 .060 .068 .065 -.003 -.008 .003 Measurement invariance across samples A and B < .001 < .001 .097 .029 .042 .046 .045 -.013 -.004 .001 Measurement invariance across gender .007 .037 .012 .029 .035 .035 .035 -.006 .000 .000 Measurement invariance across samples A and B .006 < .001 .904 .041 .045 .050 .047 -.004 -.005 .003 Measurement invariance across gender .282 < .001 .600 .040 .039 .041 .039 .001 -.002 .002 26.88 328.71 23.60 46.85 157.13 34.55 8.66 248.85 8.11 23.18 114.62 19.01 35.98 66.56 17.37 24.44 19.24 24.14 26.38 90.02 6.23 13.18 48.03 10.18 .991 .990 .985 .985 .991 .988 .985 .985 .991 .990 .975 .975 .990 .987 .980 .979 .996 .991 .988 .987 .996 .994 .993 .992 .991 .987 .983 .984 .991 .990 .988 .988 -.001 -.005 .000 .003 .003 .000 .001 .015 .000 .003 .007 .001 -.005 -.003 -.001 -.002 -.001 -.001 -.004 -.004 .001 -.001 -.002 .000 .990 .989 .984 .985 .989 .987 .985 .985 .986 .988 .972 .976 .985 .983 .978 .980 .995 .989 .987 .988 .995 .992 .993 .992 .988 .986 .982 .984 .988 .989 .988 .989 .001 .005 -.001 .002 .002 .000 .016 .016 -.004 .002 .005 .005 -.002 .006 .002 -.001 .003 -.001 .001 .002 .004 -.002 -.001 .001 -.001 Note: GPQ18 –the 18-item version of Generalized Pliance Questionnaire, GPQ9 –the 9-item version of Generalized Pliance Questionnaire, GTQ–Generalized Tracking Questionnaire, GSPQ—Generalized Self-Pliance Questionnaire. https://doi.org/10.1371/journal.pone.0283795.t002 [10] and both versions of GPQ: GPQ18 and GPQ9 [11]. Model fit to data was acceptable. All of the adapted scales presented good reliability (internal consistency) and item-total correlations. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 12 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ Table 3. Descriptive statistics and correlations between GPQ18, GPQ9, GTQ, GSPQ and other measures’ scales in both samples. Scales SWLS DASS Depression DASS Anxiety DASS Stress GSES AAQ—II CFQ VQ Progress VQ Obstruction RRQ Rumination RRQ Reflection GPQ18 GPQ9 GTQ GSPQ SWLS DASS_D DASS_A DASS_S GSES AAQ CFQ VQP VQO RRQ_RU RRQ_RE GPQ18 GPQ9 GTQ GSPQ α M (SD) Skewness (SE) Kurtosis (SE) GPQ18 GPQ9 GTQ GSPQ Sample A .873 .913 .884 .888 .886 .933 .954 .830 .847 .825 .820 .934 .884 .903 .903 .896 .920 .874 .903 .927 .941 .961 .829 .858 .844 .694 .934 .890 .910 .899 22.42 (6.09) 6.51 (5.38) 5.40 (5.01) 8.29 (5.14) 29.92 (4.57) 24.09 (10.37) 27.62 (10.28) 16.89 (6.12) 12.51 (6.72) 20.47 (5.12) 22.80 (4.79) 71.12 (17.53) 35.45 (9.03) 53.47 (8.95) 47.90 (12.36) 19.50 (6.09) 6.25 (5.35) 4.43 (4.45) 7.04 (4.95) 29.40 (4.87) 23.27 (10.42) 23.82 (10.79) 15.99 (6.16) 11.82 (6.92) 18.61 (5.69) 19.31 (4.17) 67.47 (18.05) 33.82 (9.48) 53.31 (9.39) 46.72 (12.16) .23 (.12) .74 (.12) .84 (.12) .37 (.12) -.21 (.12) .23 (.12) -.07 (.12) -.27 (.12) .13 (.12) -.20 (.12) -.27 (.12) .02 (.12) -.01 (.12) .07 (.12) -.28 (.12) -.34 (.10) .76 (.10) 1.13 (.10) .48 (.10) -.70 (.10) .35 (.10) .25 (.10) -.22 (.10) .24 (.10) -.09 (.10) .37 (.10) .04 (.09) .03 (.09) .11 (.09) -.04 (.09) Sample B -.22 (.23) -.36 (.23) -.24 (.23) -.62 (.23) .95 (.23) -.67 (.23) -.57 (.23) -.38 (.23) -.68 (.23) -.27 (.23) -.72 (.23) .14 (.23) .20 (.23) .03 (.23) .27 (.23) .07 (.19) -.23 (.19) .80 (.19) -.44 (.19) 2.06 (.19) -.59 (.20) -.64 (.20) -.11 (.20) -.57 (.20) -.55 (.20) .47 (.20) .11 (.19) -.05 (.19) .08 (.19) .24 (.19) -.226 .312 .319 .342 -.267 .446 .437** -.119* .343 .422 .040 .008 .298 .279 .289 -.108 .429 .4398 .042 .406 .350 .121 -.206 .295 .293 .312 -.240 .410 .395 -.113* .320 .397 .029 .967 .000 .280 .261 .268 -.112 .405 .361 .022 .388 .321** .096* .974 .353 -.354 -.295 -.302 .620 -.338 -.263 .432 -.296 -.213 .238 -.199 -.176 .322 -.185 -.141 -.129 .581 -.234 -.184 .454 -.220 -.217 .130 .007 .001 -.231 .402 .430 .468 -.277 .543 .567 -.071 .436 .455 .035 .595 .534 -.174 -.104 .440 .401 .489 -.105 .523 .589 .077 .527 .508 .270 .594 .550** .067 Note: M–mean, SD–standard deviation, SE–standard error, α –Cronbach Alpha, SWLS–Satisfaction with Life Scale, DASS Depression–the Depression scale of Depression, Anxiety, and Stress Scale, DASS Anxiety–the Anxiety scale of Depression, Anxiety, and Stress Scale, DASS Stress–the Stress scale of Depression, Anxiety, and Stress Scale, GSES–General Self-efficacy Scale, AAQ–II–Acceptance and Action Questionnaire–II, CFQ–Cognitive Fusion Questionnaire, VQ Progress–the Progress scale of Valuing Questionnaire, VQ Obstruction–the Obstruction scale of Valuing Questionnaire, RRQ Rumination–the Rumination scale of Rumination– Reflection Questionnaire, RRQ Reflection–the Reflection scale of Rumination–Reflection Questionnaire, GPQ18 –the 18-item version of Generalized Pliance Questionnaire, GPQ9 –the 9-item version of Generalized Pliance Questionnaire, GTQ–Generalized Tracking Questionnaire, GSPQ—Generalized Self-Pliance Questionnaire;* p < .05, ** p < .01 https://doi.org/10.1371/journal.pone.0283795.t003 Polish versions of questionnaires presented significant correlations in the expected direc- tions with relevant psychological variables in line with the original studies [10–12]. Regarding ACT processes, pliance (measured by Polish versions of GPQ9 and GPQ18) and self-pliance (GSPQ) was positively related to, cognitive fusion and obstruction of valued living, and not related or negatively correlated with progress in valued living. This last score is slightly differ- ent from the results obtained by Ruiz and colleagues [11] and by Ruiz and Sua´rez-Falco´n and PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 13 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ colleagues [10], who showed consistent negative correlations between GPQ9, GPQ18 and GSPQ and progress in valued living. In our study, only in the student sample GPQ18 and GPQ9 were negatively correlated with progress in valued living, whereas in the general sample, as well as for GSPQ there were no significant correlations. Pliance and self-pliance were negatively related to psychological flexibility as measured by AAQ-II [26, 27] similarly to the results obtained by Ruiz and Sua´rez-Falco´n and colleagues [10] and Ruiz and colleagues [11] who also used AAQ-II in their studies. Despite AAQ-II have been questioned as a precise and adequate measure of psychological flexibility [44, 45], the result is in line with the results obtained by researchers that used different measurements with greater construct validity (e.g. CompACT [46]). Pliance and self-pliance were positively related with rumination and emotional symptoms in line with the original validation studies [10, 11] and showed weak and positive or non-significant relationship with reflection. Finally, they pre- sented weak and negative or non-significant correlations with life satisfaction and self-efficacy. In contrast to the original studies, GPQ18 and GPQ9 in the general sample were not related to life satisfaction. Finally pliance, self-pliance were positively correlated in both samples, yet pliance and self- pliance and tracking were negatively correlated only in the student sample, which is contrary to expectations and needs further replication. Summarizing, pliance and self-pliance seem negatively related to psychological flexibility. Although the results of original studies suggest pliance may be a process leading to lower psy- chological flexibility and the lower level of life satisfaction, the latter was not found in the Pol- ish sample. Tracking, as measured by the Polish version of GTQ, was positively related to progress in valued living and negatively related with psychological inflexibility, cognitive fusion, obstruc- tion to valued living. Moreover, it was positively correlated with life satisfaction, general self- efficacy and reflection and negatively with rumination and emotional symptoms in line with the original validation study [12]. The results support the idea that tracking, i.e. being in direct contact with contingencies and following the real consequences of behavior, is a process which may support living a valuable life, the feeling of greater self-efficacy, which may lead to greater satisfaction with life and, possibly, general health. Despite indicating the validity of the Polish adaptations of GPQ, GSPQ and GTQ, the pres- ent study should be complemented by further research. Although in general correlations with other measures are in line with original validation studies, there may be questions regarding lack or very weak correlation of pliance with life satisfaction. At the same time, the relationship between tracking and life satisfaction as well as self-efficacy was consistent in the original and Polish study. The explanation may be the cultural differences between Polish and South Amer- ican societies. The main difference between pliance and tracking is apparent source of rein- forcement for rule-following: in pliance it is social or arbitrary, in tracking–non arbitrary. For some reason following the behavior considered as socially approved in Poland seems to have little impact on life satisfaction while it may still have detrimental effects for individuals’ gen- eral functioning (because of negative relationship with psychological flexibility). Differences between Irish as well as Columbian adolescents in pliance and psychological inflexibility has been recently reported by Stapleton et al. [47]. Further longitudinal research allowing for cause-effect conclusions should determine the relationship between pliance, tracking, life satis- faction and self-efficacy in cross-cultural studies. The present study has some more limitations. First of all, the validity of the measurements should be tested in clinical samples due to its potential utility in research and practice in the area of psychopathology. Secondly, the validity of GTQ should also be confirmed in future studies including its relationship with executive functions. Further studies should include PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 14 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ more precise and adequate measurements of psychological flexibility. Furthermore, the stabil- ity of the results needs to be established in additional study. The study has been conducted online and we have not compared the mode of administration of measurements (e.g. online vs pen-and-paper methods might be regarded by participants as providing different levels of ano- nymity, and this may lead to differences in how socially biased responding would be [48]). Finally, the samples were not representative of the Polish general and student’s population. In conclusion, this study contributes with the GPQ, GSPQ and GTQ to the Polish language and they appear to be adequate to be used in Polish samples which may accelerate the develop- ment of research on pliance, self-pliance and tracking in Polish language and support clini- cians working with clients. Acknowledgments We would like to thank our colleagues: Krystyna Pomorska, Jan Topczewski, Lidia Baran and Monika Suchowierska-Stephany, who participated in the preparation of Polish translations of the questionnaires. Author Contributions Conceptualization: Joanna Dudek, Maria Cyniak-Cieciura, Paweł Ostaszewski. Data curation: Joanna Dudek, Maria Cyniak-Cieciura. Formal analysis: Maria Cyniak-Cieciura. Funding acquisition: Joanna Dudek. Investigation: Joanna Dudek, Maria Cyniak-Cieciura. Methodology: Joanna Dudek, Maria Cyniak-Cieciura, Paweł Ostaszewski. Project administration: Joanna Dudek, Maria Cyniak-Cieciura. Resources: Joanna Dudek, Maria Cyniak-Cieciura. Supervision: Paweł Ostaszewski. Visualization: Maria Cyniak-Cieciura. Writing – original draft: Joanna Dudek, Maria Cyniak-Cieciura, Paweł Ostaszewski. Writing – review & editing: Joanna Dudek, Maria Cyniak-Cieciura, Paweł Ostaszewski. References 1. Skinner BF. An operant analysis of problem solving. In Kleinmuntz B, editor. Problem solving: research, method and theory. New York: Wiley; 1966. p.225–57. 2. Zettle RD, Hayes SC. Rule-Governed Behavior: A Potential Theoretical Framework for Cognitive- Behavioral Therapy. In: Advances in cognitive–behavioral research and therapy Academic Press.1982, 73–118. 3. Barnes-Holmes D, O’;Hora D, Roche B, Hayes SC, Bissett RT, Lyddy F. Understanding and Verbal Regulation. In: Hayes SC, Barnes-Holmes D, Roche B, editors. Relational Frame Theory: A Post-Skin- nerian Account of Human Language and Cognition. Boston, MA: Springer US; 2001. p. 103–17. Avail- able from: https://doi.org/10.1007/0-306-47638-X_6 4. Hayes SC, Strosahl KD, Wilson KG. Acceptance and commitment therapy: An experiential approach to behavior change. New York, NY, US: Guilford Press; 1999. 5. Luciano C, Valdivia-Salas S, Ruiz F. The self as the context for rule-governed behavior. In In McHugh, Stewart (Eds.), The self and perspective taking: Research and applications, Context Press, Oakland, 2012. p. 143–60. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 15 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ 6. McAuliffe D, Hughes S, Barnes-Holmes D. The Dark-Side of Rule Governed Behavior: An Experimental Analysis of Problematic Rule-Following in an Adolescent Population With Depressive Symptomatology. Behav Modif. 2014 Jul 1; 38(4):587–613. https://doi.org/10.1177/0145445514521630 PMID: 24509717 7. To¨ rneke N, Luciano C, Salas SV. Rule-governed behavior and psychological problems. Int J Psychol Psychol Ther. 2008; 8:141–56. 8. Monestès JL, Villatte M, Stewart I, Loas G. Rule-Based Insensitivity and Delusion Maintenance in Schizophrenia. Psychol Rec. 2014 Jun 1; 64(2):329–38. 9. Hayes SC, Strosahl KD, Wilson KG. Acceptance and commitment therapy: The process and practice of mindful change, 2nd ed. New York NY, US: Guilford Press; 2012. 10. Ruiz FJ, Sua´rez-Falco´ n JC, Segura MA, Garcı´a-Martı´n MB, Henao AM, Bedoya-Valderrama L. Devel- opment of the Generalized Self-Pliance Questionnaire. Paper presented at: The ACBS World Confer- ence 18; 2020 Jul 14–16, online. 11. Ruiz FJ, Sua´rez-Falco´ n JC, Barbero-Rubio A, Flo´rez CL. Development and initial validation of the Gen- eralized Pliance Questionnaire. J Context Behav Sci. 2019 Apr 1; 12:189–98. 12. Ruiz FJ, Garcı´a-Martı´n MB, Sua´rez-Falco´n JC, Bedoya-Valderrama L, Segura-Vargas MA, Peña-Var- gas A, et al. Development and initial validation of the Generalized Tracking Questionnaire. PLOS ONE. 2020 Jun 11; 15(6):e0234393. https://doi.org/10.1371/journal.pone.0234393 PMID: 32525904 13. Hayes SC, Zettle RD, Rosenfarb I. Rule-Following. In: Hayes SC, editor. Rule-Governed Behavior: Cognition, Contingencies, and Instructional Control. Boston, MA: Springer US; 1989. p. 191–220. Available from: https://doi.org/10.1007/978-1-4757-0447-1_6 14. Skinner BF. Verbal behavior. New York: Appleton-Century-Crofts.1957. 15. Monestès JL, Greville WJ, Hooper N. Derived insensitivity: Rule-based insensitivity to contingencies propagates through equivalence. Learn Motiv. 2017; 59:55–63. 16. Hayes SC, Wilson KG, Gifford EV, Follette VM, Strosahl K. Experiential avoidance and behavioral disor- ders: A functional dimensional approach to diagnosis and treatment. J Consult Clin Psychol. 1996; 64:1152–68. 17. Kissi A, Hughes S, Mertens G, Barnes-Holmes D, De Houwer J, Crombez G. A Systematic Review of Pliance, Tracking, and Augmenting. Behav Modif. 2017 Sep 1; 41(5):683–707. https://doi.org/10.1177/ 0145445517693811 PMID: 28776429 18. Hayes SC, Hayes LJ. The Verbal Action of the Listener as a Basis for Rule-Governance. In: Hayes SC, editor. Rule-Governed Behavior: Cognition, Contingencies, and Instructional Control Boston, MA: Springer US; 1989 p. 153–90. Available from: https://doi.org/10.1007/978-1-4757-0447-1_5 19. O’Connor M, Byrne P, Ruiz FJ, McHugh L. Generalized Pliance in Relation to Contingency Insensitivity and Mindfulness. Mindfulness. 2019 May 1; 10(5):833–40. 20. Diener E, Emmons RA, Larsen RJ, Griffin S. The Satisfaction With Life Scale. J Pers Assess. 1985 Feb 1; 49(1):71–5. https://doi.org/10.1207/s15327752jpa4901_13 PMID: 16367493 21. 22. Jankowski KS. Is the shift in chronotype associated with an alteration in well-being? Biol Rhythm Res. 2015 Mar 4; 46(2):237–48. Lovibond PF, Lovibond SH. The structure of negative emotional states: comparison of the Depression Anxiety Stress Scales (DASS) with the Beck Depression and Anxiety Inventories. Behav Res Ther. 1995 Mar; 33(3):335–43. https://doi.org/10.1016/0005-7967(94)00075-u PMID: 7726811 23. Makara-Studzińska M, Tyburski E, Załuski M, Adamczyk K, Mesterhazy J, Mesterhazy A. Confirmatory Factor Analysis of Three Versions of the Depression Anxiety Stress Scale (DASS-42, DASS-21, and DASS-12) in Polish Adults. Front Psychiatry 2022 Jan 4; 12:770532. https://doi.org/10.3389/fpsyt.2021. 770532 PMID: 35058818 24. Schwarzer R, Jerusalem M. (1995). Generalized Self-Efficacy scale. In Weinman J, Wright S, Johnston M, Measures in health psychology: A user’s portfolio. Causal and control beliefs, Windsor, UK : NFER- NELSON. 1995, (pp. 35–37). 25. Juczyński Z. Narze¸dzia pomiaru w promocji i psychologii zdrowia. Pracownia Testo´ w Psychologicznych Polskiego Towarzystwa Psychologicznego. Warszawa. 2009. 26. Bond FW, Hayes SC, Baer RA, Carpenter KM, Guenole N, Orcutt HK, et al. Preliminary Psychometric Properties of the Acceptance and Action Questionnaire–II: A Revised Measure of Psychological Inflexi- bility and Experiential Avoidance. Behav Ther. 2011 Dec 1; 42(4):676–88. https://doi.org/10.1016/j. beth.2011.03.007 PMID: 22035996 27. Kleszcz B, Dudek JE, Białaszek P, Bond FW. The Psychometric Properties of the Polish Version of the Acceptance and Action Questionnaire-II (AAQ-II). Stud Psychol. 2018 Jun 5; 56(1):1–19. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 16 / 17 PLOS ONEThe Polish adaptation of the measurements of rule-governed behaviors: GPQ, GTQ and GSPQ 28. Baran L, Hyla M, Kleszcz B. Elastyczność psychologiczna. Polska adaptacja narzędzi pomiarowych dla praktyko´ w i badaczy Katowice: Wydawnictwo Uniwersytetu Śląskiego; 2019. Available from: https:// rebus.us.edu.pl/handle/20.500.12128/12945 29. Gillanders DT, Bolderston H, Bond FW, Dempster M, Flaxman PE, Campbell L, et al. The Development and Initial Validation of the Cognitive Fusion Questionnaire. Behav Ther. 2014 Jan 1; 45(1):83–101. https://doi.org/10.1016/j.beth.2013.09.001 PMID: 24411117 30. Smout M, Davies M, Burns N, Christie A. Development of the Valuing Questionnaire (VQ). J Context Behav Sci. 2014 Jul 1; 3(3):164–72. 31. Słowińska A, Zbieg A, Oleszkowicz A. Kwestionariusz Ruminacji-Refleksji (RRQ) Paula D. Trapnella i Jennifer D. Campbell—polska adaptacja metody. Paul D Trapnell and Jennifer D Campbell’s Rumina- tion-Reflection Questionnaire (RRQ)—polish adaptation of the measure. 2014; Available from: https:// repozytorium.ukw.edu.pl///handle/item/1371 32. 33. Trapnell PD, Campbell JD. Private self-consciousness and the five-factor model of personality: Distin- guishing rumination from reflection. J Pers Soc Psychol. 1999; 76:284–304. https://doi.org/10.1037// 0022-3514.76.2.284 PMID: 10074710 Lorenzo-Seva U, Ferrando PJ. FACTOR: a computer program to fit the exploratory factor analysis model. Behav Res Methods. 2006 Feb; 38(1):88–91. https://doi.org/10.3758/bf03192753 PMID: 16817517 34. Rosseel Y. lavaan: An R Package for Structural Equation Modeling. J Stat Softw. 2012 May 24; 48:1– 36. 35. 36. Timmerman ME, Lorenzo-Seva U. Dimensionality assessment of ordered polytomous items with paral- lel analysis. Psychol Methods. 2011 Jun; 16(2):209–20. https://doi.org/10.1037/a0023353 PMID: 21500916 Ferrando PJ, Lorenzo-Seva U. Assessing the Quality and Appropriateness of Factor Solutions and Fac- tor Score Estimates in Exploratory Item Factor Analysis. Educ Psychol Meas. 2018 Oct; 78(5):762–80. https://doi.org/10.1177/0013164417719308 PMID: 32655169 37. Hu L, Bentler PM. Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Struct Equ Model Multidiscip J. 1999 Jan 1; 6(1):1–55. 38. Xia Y, Yang Y. RMSEA CFI, and TLI in structural equation modeling with ordered categorical data: The story they tell depends on the estimation methods. Behav Res Methods. 2019 Feb 1; 51(1):409–28. 39. Shi D, Maydeu-Olivares A. The Effect of Estimation Methods on SEM Fit Indices. Educ Psychol Meas. 2020 Jun; 80(3):421–45. https://doi.org/10.1177/0013164419885164 PMID: 32425213 40. Cheung GW, Rensvold RB. Evaluating Goodness-of-Fit Indexes for Testing Measurement Invariance. Struct Equ Model Multidiscip J. 2002 Apr 1; 9(2):233–55. 41. Meade AW, Johnson EC, Braddy PW. Power and sensitivity of alternative fit indices in tests of measure- ment invariance. J Appl Psychol. 2008 May; 93(3):568–92. https://doi.org/10.1037/0021-9010.93.3.568 PMID: 18457487 42. Chen FF. Sensitivity of Goodness of Fit Indexes to Lack of Measurement Invariance. Struct Equ Model Multidiscip J. 2007 Jul 31; 14(3):464–504. 43. Xu H, Tracey TJG. Use of multi-group confirmatory factor analysis in examining measurement invari- ance in counseling psychology research. Eur J Couns Psychol [Internet]. 2017 Feb 14 [cited 2022 Sep 18];6(1). Available from: https://www.psycharchives.org/en/item/ebd7c42d-bfdb-4399-b6f4- b54708edb8d0 44. Doorley JD, Goodman FR, Kelso KC, Kashdan TB. Psychological flexibility: What we know, what we do not know, and what we think we know. Soc Personal Psychol Compass. 2020; 14(12):e12566. 45. Tyndall I, Waldeck D, Pancani L, Whelan R, Roche B, Dawson DL. The Acceptance and Action Ques- tionnaire-II (AAQ-II) as a measure of experiential avoidance: Concerns over discriminant validity. J Con- text Behav Sci. 2019; 12:278–84. 46. Waldeck D, Pancani L, Tyndall I. An examination of the construct validity of the Generalized Pliance Questionnaire. J Context Behav Sci. 2019; 11:50–4. 47. Stapleton A, Ruiz FJ, McHugh L. Comparative investigation of adolescents’ generalized pliance and psychological inflexibility across cultural contexts. Psychol Rec. 2022; 72:105–9. 48. Ward P, Clark T, Zabriskie R, Morris T. Paper/Pencil Versus Online Data Collection. J Leis Res. 2014 Mar 1; 46(1):84–105. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283795 April 5, 2023 17 / 17 PLOS ONE
10.1371_journal.pone.0282823	RESEARCH ARTICLE A systematic review of the burden of, access to services for and perceptions of patients with overweight and obesity, in humanitarian crisis settings Thomas ShortlandID Saran Shantikumar1, William Proto1, Bassit Malik1, Roger Yau1, Maddie CobbinID Ammar Sabouni3, Gavin Rudge4, Farah KidyID 2, Oyinlola Oyebode1, 1, 1, Majel McGranahanID 1, Daniel StewartID 1* a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 1 Warwick Medical School, University of Warwick, Coventry, United Kingdom, 2 National Public Health Specialty Training Programme, South West Training Scheme, Bristol, United Kingdom, 3 Syria Development Centre, London, United Kingdom, 4 Institute of Applied Health Research, College of Medical and Dental Sciences, University of Birmingham, Birmingham, United Kingdom OPEN ACCESS Citation: Shortland T, McGranahan M, Stewart D, Oyebode O, Shantikumar S, Proto W, et al. (2023) A systematic review of the burden of, access to services for and perceptions of patients with overweight and obesity, in humanitarian crisis settings. PLoS ONE 18(4): e0282823. https://doi. org/10.1371/journal.pone.0282823 Editor: Che Matthew Harris, Johns Hopkins Medicine, UNITED STATES Received: April 21, 2022 Accepted: February 23, 2023 Published: April 24, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0282823 Copyright: © 2023 Shortland et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. * f.kidy@warwick.ac.uk Abstract Introduction Excess body weight causes 4 million deaths annually across the world. The number of peo- ple affected by humanitarian crises stands at a record high level with 1 in 95 people being forcibly displaced. These epidemics overlap. Addressing obesity is a post-acute phase activity in non-communicable disease management in humanitarian settings. Information is needed to inform guidelines and timing of interventions. The objective of this review was to explore the prevalence of overweight and obesity in populations directly affected by humani- tarian crises; the cascade of care in these populations and perceptions of patients with over- weight and obesity. Methods Literature searches were carried out in five databases. Grey literature was identified. The population of interest was non-pregnant, civilian adults who had experience of humanitarian crises (armed conflict, complex emergencies and natural disasters). All study types pub- lished from January 1st, 2011, were included. Screening, data extraction and quality appraisal were carried out in duplicate. A narrative synthesis is presented. Results Fifty-six reports from forty-five studies were included. Prevalence estimates varied widely across the studies and by subgroups. Estimates of overweight and obesity combined ran- ged from 6.4% to 82.8%. Studies were heterogenous. Global distribution was skewed. Increasing adiposity was seen over time, in older adults and in women. Only six studies were at low risk of bias. Body mass index was the predominant measure used. There were no studies reporting cascade of care. No qualitative studies were identified. PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 1 / 29 PLOS ONEOverweight and obesity in humanitarian settings Funding: FK is supported by an NIHR Doctoral Research Fellowship (grant number 300688) and SS is supported by an NIHR Clinical Lectureship. The views expressed are those of the authors and not necessarily those of the NIHR or the Department of Health and Social Care. The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. https://www.nihr.ac.uk/. Competing interests: The authors have declared that no competing interests exist. Conclusion Overweight and obesity varied in crisis affected populations but were rarely absent. Improved reporting of existing data could provide more accurate estimates. Worsening obe- sity may be prevented by acting earlier in long-term crises and targeting risk groups. The use of waist circumference would provide useful additional information. Gaps remain in understanding the existing cascade of care. Cultural norms around diet and ideal body size vary. Introduction Costing 2.8% of the world’s gross domestic product, affecting over 2 billion people worldwide and causing 4 million deaths annually, excess body weight (including overweight and obesity) is a global health emergency [1]. The World Health Organisation (WHO) identifies excess body weight as a key risk factor for noncommunicable diseases (NCDs) [2]. Globally, more than 15 million die prematurely due to NCDs [2]. Reducing obesity can decrease premature mortality [3], thereby directly contributing to Sustainable Development Goal (SDG) 3.4 [4]. The potential negative impacts of obesity and overweight are not restricted to NCDs. As the global community has learnt over the course of the Covid-19 pandemic, increased adiposity is also a risk factor for morbidity and mortality caused by some infectious diseases [5]. The number of people affected by humanitarian crises including violence, persecution, nat- ural disasters and human rights violations has increased steadily since 2010 and now stands at record high level with 82.4 million people being forcibly displaced at the end of 2020 [6]. The majority of those displaced have remained in their own countries (internally displaced people [IDPs]) and following Colombia, the most affected countries are in Africa and the Middle East. As for those displaced across borders, approximately two thirds come from Syria, Vene- zuela, Afghanistan, South Sudan and Myanmar [6]. In many of these countries NCDs are now more significant causes of death and disability than communicable diseases and levels of obesity and overweight are increasing [7–9]. Whilst in the past the issue of NCDs in humanitarian crises was largely forgotten, the increasingly overlapping nature of these epidemics is now recognised. There have been calls from practitioners and patients to increase research [10] and to improve prioritisation, recog- nition, prevention and management of NCDs in these settings [11–14]. An informal working group chaired by the United Nations High Commissioner for Refugees (UNHCR) and with membership of academics, policy makers, WHO and key non-governmental organisations (NGOs) is leading the way on delineating operational considerations for NCD management in humanitarian settings [15]. Obesity features in these discussions as a risk factor for chronic diseases to be addressed after the acute phase of the crisis. To inform these developments, there are a suite of systematic reviews which bring together the evidence on diabetes [16, 17], substance misuse [18, 19], smoking [20], alcohol [21, 22], cardiovascular disease [23, 24], hypertension [25], mixed NCDs [26–29] and models of care [30, 31] in specific settings. However, to our knowledge, there has not been an attempt to col- late information focussing on obesity in the same way. The objective of this review is to explore the prevalence and incidence of overweight and obesity, and the changes in adiposity over time in populations directly affected by humanitar- ian crises; the cascade of care in these populations and perceptions of patients with overweight and obesity. PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 2 / 29 PLOS ONEOverweight and obesity in humanitarian settings Methods A systematic review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines [32] and applying the Synthesis With- out Meta-analysis (SWiM) extension [33]. A scoping exercise was carried out in August 2019. This informed decisions about eligibility, inclusion dates and synthesis. Eligibility criteria The PECO criteria described below form part of the eligibility criteria. Further, all study types, published in any language and carried out in any geographical location were considered eligi- ble. For the scoping exercise, studies published from January 1st, 1999, were included. Review- ing the returns showed that the data being presented in the earlier papers were out of date given the context of changing levels of obesity globally. Since we were interested in providing a description which could be used by service providers in the current time, we restricted this review to papers published from January 1st, 2011, onwards. Conference proceedings, letters, theses, clinical guidelines, opinion pieces and study proto- cols were excluded. Reports from NGOs are important in this field and were included as long as there was a description of the methods used to gather data. PECO criteria The population, exposure, comparator and outcome (PECO) criteria for the study are described below. Population. The population of interest was non-pregnant, civilian adults (aged 18 years or older) who had direct experience of humanitarian crises whether they were displaced or not. Economic migrants, Special Immigrant Visa entrants (those granted permanent residence in the USA for reasons including aiding US efforts in Afghanistan and Iraq [34]) and migrants unaffected by crises were not considered eligible. Service and military personnel, local combat- ants and prisoners of war were excluded. Service users attending general clinics were consid- ered eligible, unless selected on the basis of a specific disease, when they were excluded. Studies with a mixed population were included if the population of interest could be clearly differentiated. For qualitative studies, this meant that the views of participants with overweight or obesity had to be identifiable. The study authors’ definition of the type of migrant was applied. Exposure. The crises of interest were armed conflict, complex emergencies and natural disasters (including earthquakes, landslides, tidal waves, tsunamis, floods, cyclones, hurricane and drought). Study authors’ definitions of crises were applied. Exposures that began after or were ongoing in January 1999 were considered eligible. Exposures needed to be ongoing or previous to the time of data collection to be eligible. We did not impose other temporal restric- tions on the exposure- outcome relationship. We did not specifically search for COVID-19 related publications. We felt that the global nature of the pandemic meant that doing so would effectively result in a global prevalence esti- mate for overweight and obesity. Comparator. Comparators were not considered as an eligibility criterium. Outcome. Study authors’ definitions of overweight and obesity were applied regardless of the measure and cut-offs used. During risk of bias (ROB) assessment the decisions made by authors in this regard were evaluated. The primary outcomes of interest were: PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 3 / 29 PLOS ONEOverweight and obesity in humanitarian settings 1. The prevalence and incidence of overweight and / or obesity as defined by body mass index (BMI). 2. Change in adiposity over time in those diagnosed with overweight or obesity. 3. Cascade of care for overweight and / or obesity including recognition, seeking treatment or support and receiving treatment or support. 4. Patient knowledge and attitude to overweight and / or obesity. Secondary outcomes were: 1. Understanding of whether or not weight management is included as part of a wider pro- gramme of prevention or health promotion. 2. Barriers and facilitators to accessing treatment. 3. Evidence of use of other measures of adiposity. Information sources Medline, Embase, PsycINFO, Cumulative Index of Nursing and Allied Health Literature (CINAHL) and Web of Science were searched. Grey literature and newly published peer reviewed material was identified by searching Google, ReliefWEB, UN High Commissioner for Refugees, WHO Institutional Repository for Information Sharing, UNICEF, Me´decins Sans Frontières, International Rescue Committee, International Committee of the Red Cross, Centre for Disease Control and Prevention and Active Learning Network for Accountability and Performance (ALNAP). Search terms were adapted from our previous work [25] and can be seen in full in S1 Appendix. Searches were updated in January 2021 (databases) and May 2021 (Google searches). Rayyan was used to manage search returns [35]. Selection processes Two reviewers independently screened the titles and abstracts against the criteria described above. Conflicts were resolved by discussion. Papers included in the full text screening were also independently screened by two reviewers. Conflicts were again resolved by discussion. Reasons for exclusion were documented. Data collection Data collection was carried out by one reviewer and independently checked by a second. Data were extracted into a shared spreadsheet. For each report, details of the publication (authors, year, title), study type, geographical context, a description of the population and a description of the exposure were extracted. For quantitative studies, method(s) of measurement, number with overweight and / or obesity, prevalence, sample size, measure of spread, details of sub- groups and secondary outcomes were collected. For case control studies, we collected data from both cases and controls, but have presented data from controls only since cases may be systematically different from the general population due to the disease under study. For longi- tudinal studies data from each time point were extracted. In any study type, where subgroup data were available, these were extracted but only whole study level data are presented. For studies including adults and children, only data for those aged over 18 years was extracted. Where data for the whole study population were not presented, we used subgroup data to cal- culate these. In most cases this was done either by summing the numbers in mutually exclusive PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 4 / 29 PLOS ONEOverweight and obesity in humanitarian settings subgroups, or by applying rates reported in subgroups to the population of the subgroup to give the number in each subgroup. For qualitative studies (had any been identified) we planned to extract concepts, themes, barriers and facilitators as described by participants. Risk of bias assessment A tool for risk of bias (ROB) in prevalence studies proposed by Hoy et al [36] was adapted for our study. The original tool included a question about the study population in relation to the national population. This was not appropriate for our study since we were not seeking nation- ally representative prevalence estimates. External ROB was judged on choice of sampling frame, method of sample selection and extent of non-responsiveness. Internal ROB was judged on method of data collection, case definition, choice of measure of adiposity, use of standard- ised procedures and accuracy of reporting. To add further granularity to the discussion about ROB, a score of low, medium or high risk was given in the external, internal and overall domains. Had we identified any qualitative studies, we planned to use the Critical Appraisal Skills Programme (CASP) checklist [37]. ROB assessment was independently carried out by one reviewer and checked by a second. Discrepancies were resolved by discussion. Methods of synthesis Our scoping exercise, initial search results and previous work in this field demonstrated that included studies were heterogeneous [25]. As a result, a narrative synthesis was carried out. Definitions of overweight and obesity vary according to the anthropometric measure used and even with a single measure such as BMI, different cut-offs are proposed for different popu- lations [38]. For the purposes of this review, we focussed on BMI as our scoping exercise showed that this was the most commonly used measure of adiposity amongst the included studies. Findings of overweight and obesity are reported as defined by individual study authors, but details of the cut-offs used were extracted for the ROB assessment. Subgroups of interest for the synthesis included geographical setting, type of exposure, dis- placement status and ROB. Age and sex were considered important factors for the distribution of obesity and overweight [39]. Data are reported from all the included studies, but priority in interpretation is given to those with lower internal ROB since these studies are measuring the same phenomenon across the dataset. Heterogeneity was explored by describing the study type, population, exposure and setting of each study. Data are presented in separate tables for high income countries (HICs) and low and middle income countries (LMICs), grouped by exposure type and location of study and shaded to indicate ROB. Categorisation as HIC or LMIC was selected to allow comparison to other pub- lications in this field and to allow a rough assessment of resources available at a country level. The World Bank income-based classification system was used [40]. Results Overall, 20,376 non-duplicate search returns were identified and screened. Four hundred and eighty-one full-text reports were assessed for eligibility. Fifty-six reports from 45 studies were included in the review. Fifty reports were excluded because anthropometric measurements were presented in a way which did not allow categorisation and a further 13 were excluded as details were only given for underweight. The PRISMA flow diagram can be seen in Fig 1 and PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 5 / 29 PLOS ONEOverweight and obesity in humanitarian settings Fig 1. Showing the PRISMA flow diagram. From: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71. https://doi.org/10.1371/journal.pone.0282823.g001 the S1 Checklist is included in the supplementary material. Reports excluded after full-text screening and the reasons for exclusion are detailed in S2 Appendix. Description of the included studies Characteristics of the included studies are shown in Table 1. PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 6 / 29 PLOS ONEOverweight and obesity in humanitarian settings Table 1. Showing characteristics of included studies. First author and Year Title Yanni et al 2013 [42] Amr et al 2021 [55] Ratnayake et al 2020 [51] The health profile and chronic diseases comorbidities of US-bound Iraqi refugees screened by the International Organization for Migration in Jordan: 2007–2009 Mood and demographical factors as predictors of body mass index among Iraqi and Syrian refugees in Jordan Access to Care and Prevalence of Hypertension and Diabetes Among Syrian Refugees in Northern Jordan Population and location of study Description of crisis Study design Age included in data extraction Conflict Iraqi refugees in Jordan who are undergoing screening prior to entry to the USA Iraqi and Syrian refugees living in Jordan Syrian refugees, non-pregnant and aged 18 or over, living outside of camps and located in Northern Jordan. Biological measurements, including height and weight, were taken from adults aged 30 or over Not described, but most likely conflict Results of routine health screening 20 and older Iraqi and Syrian conflict Cross sectional 18 and older Syrian conflict Cross sectional 30 and older Mansour et al 2020 [50] Non-communicable diseases in Lebanon: results from World Health Organization STEPS survey 2017 Syrian refugees and host Lebanese population residing in 8 governates in Lebanon Syrian conflict Cross sectional 18–69 Al-Duais and Al Awthan 2019 [54] Balcilar 2016 [57] Eryurt and Menet 2020[52] Greene-Cramer et al 2020 [49] Association between qat chewing and dyslipidaemia among young males Young men attending Ibb University, Yemen War in Yemen Cross sectional 18–25 Health Status Survey of Syrian Refugees in Turkey Syrian refugee population residing in 10 provinces of Turkey Noncommunicable diseases among Syrian refugees in Turkey: An emerging problem for a vulnerable group Noncommunicable disease burden among conflict-affected adults in Ukraine: A cross-sectional study of prevalence, risk factors, and effect of conflict on severity of disease and access to care Syrian refugees aged 18–69 living in Turkey Adult IDPs throughout Ukraine (excluding Donetsk & Luhansk regions) and adults (aged 30 or older) living in government- controlled, conflict-affected regions in Donbas regions in eastern Ukraine Syrian conflict Cross sectional 18–69 Syrian conflict Cross sectional 18–69 Conflict in Ukraine starting in 2014 Cross sectional 30 and older Singh et al 2015 [56] Nutrition among men and household food security in an internally displaced persons camp in Kenya Chandra et al 2019 [41] Drummond et al 2011 [43] Renzaho et al 2014 [44] Maldari et al 2019 [53] Prevalence of chronic disease risk factors in 35- to 44-year-old humanitarian arrivals to New South Wales (NSW), Australia Knowledge of Cardiovascular Risk Factors in West African Refugee Women Living in Western Australia Obesity, Type 2 Diabetes and High Blood Pressure Amongst Recently Arrived Sudanese Refugees in Queensland, Australia The health status of newly arrived Syrian refugees at the Refugee Health Service, South Australia, 2016 Reznar et al 2020 [45] The Burden of Chronic Health Conditions among Iraqi Refugees in Michigan Male IDPs in a camp in Kenya Ethnic violence Cross sectional 18 and older Newly arrived refugees to Sydney, Australia Mixed exposures, majority from Iraq and Syria Cross sectional 35–44 Refugee women from Liberia or Sierra Leone living in Australia Not described, likely varied Cross sectional 20–67 Sudanese adult refugees living in Brisbane, Australia Conflict and other exposures in Sudan Cross sectional 18–70 Newly arrived Syrian adults and children who attended the Refugee Health Service in South Australia from 1 Jan to 31 Dec 2016 Iraqi refugees, Michigan, USA Syrian Conflict Cross sectional 18 and older Conflict since 2003, Iraqis resettled in the USA Cross sectional 18 and older (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 7 / 29 PLOS ONEOverweight and obesity in humanitarian settings Table 1. (Continued) First author and Year Title Population and location of study Description of crisis Study design Sastre et al 2020 [46] From the Democratic Republic of the Congo to North Carolina: An Examination of Chronic Disease Risk Refugees from Democratic Republic of Congo to the USA. Data collection October 2017 to March 2018 Mixed exposures, from Democratic Republic of Congo Cross sectional study Age included in data extraction 18 and older Jen et al 2015 [47] Pre- and Post-displacement Stressors and Body Weight Development in Iraqi Refugees in Michigan Iraqi refugees settled in Michigan, USA Jen et al 2018 [48] Sex differences and predictors of changes in body weight and on- communicable diseases in a random newly-arrived group of refugees followed for two years Iraqi refugees settled in Michigan, USA Cohort study 18–69 Cohort study 18–69 Mixed exposures, Iraqi refugees to Michigan. Data collected after arrival in Michigan and then re- collected one year later Mixed exposures, Iraqi refugees to Michigan. Data collected after arrival in Michigan and then re- collected one year later and at year 2 follow up Bhatta et al 2014 [60] Bhatta et al 2015 [59] Kumar et al 2014 [68] Bayyari et al 2013 [58] El Kishawi et al 2014 [61] Socio-demographic and dietary factors associated with excess body weight and abdominal obesity among resettled Bhutanese refugee women in Northeast Ohio, United States Chronic Disease Burden Among Bhutanese Refugee Women Aged 18–65 Years Resettled in Northeast Ohio, United States, 2008–2011 Noninfectious disease among the Bhutanese refugee population at a United States urban clinic Dieting behaviours, obesity and predictors of dieting among female college students at Palestinian universities Obesity and overweight: prevalence and associated socio demographic factors among mothers in three different areas in the Gaza Strip-Palestine: a cross- sectional study Long-standing Refugee Situation Bhutanese refugee women living in Northern Ohio, USA Political and ethnic persecution of Bhutanese refugees of Nepalese origin Cross sectional 18–65 Bhutanese refugee women living in Northern Ohio, USA Political and ethnic persecution of Bhutanese refugees of Nepalese origin Cross sectional 18–65 Adult Bhutanese refugees who attended Grady Refugee clinic, Atlanta, Georgia, USA Political and ethnic persecution of Bhutanese refugees of Nepalese origin Female Palestinian college students in Palestine Long term Palestinian situation Cross sectional 18 and older Cross sectional Young adults Mothers in the Gaza Strip, Palestine Long term Palestinian situation Cross sectional 18–50 Kory et al 2013 [73] Health ramifications of the Gush Katif evacuation Residents evacuated from Gush Katif, Gaza Strip, Palestine Long term Palestinian situation Cross sectional data from a cohort 21 and older Dhair and Abed 2020 [62] Damiri et al 2018 [70] Damiri et al 2019 [71] Abdollahi et al 2015 [67] The association of types, intensities and frequencies of physical activity with primary infertility among females in Gaza Strip, Palestine: A case-control study Metabolic syndrome among overweight and obese adults in Palestinian refugee camps Metabolic syndrome and related risk factors among adults in the northern West Bank, a cross-sectional study High occurrence of food insecurity among urban Afghan refugees in Pakdasht, Iran 2008: a cross-sectional study Couples of reproductive age living in the Gaza Strip, Palestine Long term Palestinian situation Case-control study 18–49 Palestinian refugees displaced due to conflict and who have lived in one of three camps in Nablus, West Bank, Palestine for at least 6 months Long term Palestinian situation Cross sectional 28–65 Palestinian adults aged living in the West Bank, Palestine Long term Palestinian situation Cross sectional 18–70 Afghan refugees living in Pakdasht, Tehran, Iran Conflict and other exposures in Afghanistan Cross sectional 24–60 (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 8 / 29 PLOS ONETable 1. (Continued) First author and Year Title Taherifard et al 2021 [69] Naigaga et al 2018 [72] Kim et al 2015 (NORNS 2) [64] Jung Kim et al 2016 (NORNS 3) [65] Kim et al 2018 (NORNS 1) [63] The prevalence of risk factors associated with non-communicable diseases in Afghan refugees in southern Iran: a cross-sectional study Body size perceptions and preferences favor overweight in adult Saharawi refugees Vitamin D status and associated metabolic risk factors among North Korean refugees in South Korea: a cross-sectional study Prevalence of metabolic syndrome and its related factors among North Korean refugees in South Korea: a cross- sectional study Prevalence of general and central obesity and associated factors among north Korean refugees in South Korea by duration after defection from North Korea: A cross-sectional study Jeong et al 2017 (NORNS 4) [66] Changes in body weight and food security of adult North Korean refugees living in South Korea Furusawa et al 2011 [74] Herrera- Fontana et al 2020 [75] Communicable and non-communicable diseases in the Solomon Islands villages during recovery from a massive earthquake in April 2007 Food insecurity and malnutrition in vulnerable households with children under 5 years on the Ecuadorian coast: a post-earthquake analysis Adrega et al 2018 [76] Prevalence of cardiovascular disease risk factors, health behaviours and atrial fibrillation in a Nepalese post-seismic population: A cross-sectional screening during a humanitarian medical mission Overweight and obesity in humanitarian settings Population and location of study Description of crisis Study design Age included in data extraction Afghan refugees in Southern Iran refugee camp Mixed exposure, refugees leaving Afghanistan since 1979 settling in Iran Cross sectional 25 and older Refugees from Western Sahara that have settled in the Algerian desert Western Sahara War Cross sectional 18–80 North Korean refugees resident in Seoul, South Korea. Part of the NORNS study Long standing refugee situation, North Korean refugees Cross sectional data from ongoing cohort study 30–81 30 and older 20–60 19 and older Natural disasters Communities affected by the earthquake in the Solomon Islands Earthquake leading to tsunami and landslides Cross sectional 18 and older Households located in La Punta, a rural community, located 40 minutes by road from the epicentre of the Ecuadorian earthquake on 16 April 2016. Data on overweight and obesity is reported for ’mothers or women responsible for the household’ Inhabitants of 14 villages in Sindhupalchok, a northern region of Nepal, located in the epicentre of the earthquake An earthquake of magnitude 7.8 on the Richter scale in the province of Manabi on the Ecuadorian coast Cross sectional 18–60 May 2015 earthquake in Nepal Cross sectional 18 and older Sakai et al 2020 (FHMS 1) [77] Relationship between the prevalence of polycythemia and factors observed in the mental health and lifestyle survey after the Great East Japan Earthquake Those forced to evacuate due to the Great East Japan Earthquake. Subset of participants from the Fukushima Health Management Survey Great East Japan Earthquake, 2011 Cross sectional 20–90 Ohira et al 2016 (FHMS 2) [78] Effect of evacuation on body weight after the Great East Japan Earthquake Ohira et al 2017 (FHMS 3) [79] Changes in Cardiovascular Risk Factors After the Great East Japan Earthquake: A Review of the Comprehensive Health Check in the Fukushima Health Management Survey Japanese men and women living in communities near the Fukushima Daiichi Nuclear Power Plant in the Fukushima prefecture, Japan. Subset of participants from the Fukushima Health Management Survey Residents living near the Fukushima Daiichi power plant, Japan. Subset of participants from the Fukushima Health Management Survey Great East Japan Earthquake, 2011 Cohort study 40–90 Great East Japan Earthquake, 2011 Cross sectional 40–90 PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 (Continued ) 9 / 29 PLOS ONETable 1. (Continued) First author and Year Title Satoh et al 2021 (FHMS 4) [80] Relationship between risk of hyper- low-density lipoprotein cholesterolemia and evacuation after the Great East Japan Earthquake Takahashi et al 2016 [81] Takahashi et al 2020 [82] Takahashi et al 2021 [83] Hikichi et al 2019 [86] Ebner et al 2016 [84] Weight Gain in Survivors Living in Temporary Housing in the Tsunami- Stricken Area during the Recovery Phase following the Great East Japan Earthquake and Tsunami Increased incidence of metabolic syndrome among older survivors relocated to temporary housing after the 2011 Great East Japan earthquake & tsunami Increase in Body Weight Following Residential Displacement: 5-year Follow-up After the 2011 Great East Japan Earthquake and Tsunami Residential relocation and obesity after a natural disaster: A natural experiment from the 2011 Japan Earthquake and Tsunami Lifestyle-related diseases following the evacuation after the Fukushima Daiichi nuclear power plant accident: a retrospective study of Kawauchi Village with long-term follow-up Nakamura et al [85] Psychological distress as a risk factor for dementia after the 2004 Niigata- Chuetsu earthquake in Japan Mulugeta et al 2018 [87] Mulugeta et al 2019 [88] Mulugeta et al 2019 [89] Rhodes et al 2016 [90] Longitudinal Changes and High-Risk Subgroups for Obesity and Overweight/ Obesity Among Refugees in Buffalo, NY, 2004–2014 Burden of Mental Illness and Non- communicable Diseases and Risk Factors for Mental Illness Among Refugees in Buffalo, NY, 2004–2014 Disease Burdens and Risk Factors for Diabetes, Hypertension, and Hyperlipidemia among Refugees in Buffalo, New York, 2004–2014 Development of Obesity and Related Diseases in African Refugees After Resettlement to United States Overweight and obesity in humanitarian settings Population and location of study Description of crisis Study design Great East Japan Earthquake, 2011 Prospective cohort Age included in data extraction 40–89 Great East Japan Earthquake, 2011 Prospective cohort study 18 and older Japanese adults with health insurance without a diagnosis of hyper-LDL cholesterolemia living near the Fukushima Daiichi nuclear power plant, Japan. Subset of participants from the Fukushima Health Management Survey Research project for prospective Investigation of health problems Among Survivors of the Great East Japan Earthquake and Tsunami Disaster—The survey was carried out between September 2011 and February 2012 in 3 municipalities in Iwate Prefecture located in the Tohoku area in the northern part of Honshu, Japan Survivors of 2011 earthquake and tsunami living in Iwanuma, Japan Great East Japan Earthquake, 2011 66 and older Natural experiment nested within cohort study Residents who attended National Health Screening programmes, Kawauchi village, Japan Great East Japan Earthquake, 2011 Cohort study 40 and older Residents living in Ojiya city, Japan, who did the annual health check examination after the 2004 earthquake Mixed exposures Adult and child refugees attending the Jericho Road Community Health Centre, Buffalo, New York, USA Adult refugees attending the Jericho Road Community Health Centre, Buffalo, New York, USA Adult refugees attending the Jericho Road Community Health Centre, Buffalo, New York, USA 2004 Niigata-Chuetsu earthquake Cohort study 40 and older Service open to all refugees Retrospective cohort 19 and older Service open to all refugees Cross sectional 18 and older Service open to all refugees Cross sectional 18 and older African refugees resettled to Rhode Island, USA, 2004–2007. Excluded children, pregnant women, and individuals without an electronic medical record or recorded height Mixed exposures, majority were African refugees (Eritrea, Ethiopia, Ghana, Liberia, Rwanda and Somalia) Cohort Study 18 and older Bardenheier et al 2019 (1) [91] Prevalence of tuberculosis disease among adult US-bound refugees with chronic kidney disease Medical examination of all adult refugees before arriving in the USA Various exposures as refugees from multiple locations Cross sectional 18 and older (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 10 / 29 PLOS ONEOverweight and obesity in humanitarian settings Population and location of study Description of crisis Study design Age included in data extraction Medical examination of all Burmeses refugees before arriving in the USA Violence, conflict and natural disasters Cross sectional 18 and older The three refugee populations with the greatest number of arrivals to California, USA, between 1995 and 2011 All non pregnant adults in the asylum process in Geneva, Switzerland, collected between June 2017 and March 2019 Mixed exposures, majority from Iran, Ukraine and Vietnam Cross sectional 18 and older Mixed exposures, majority from Afghanistan, Eritrea, Sri Lanka and Syria Cross sectional, mixed methods Adults Table 1. (Continued) First author and Year Title Bardenheier et al 2019 (2) [92] Nguyen and Rehkopf, 2016 [93] Amstutz et al 2020 [94] Modesti et al 2020 [95] Kortas et al 2017 [96] Trends in Chronic Diseases Reported by Refugees Originating from Burma Resettling to the United States from Camps Versus Urban Areas During 2009–2006 Prevalence of Chronic Disease and Their Risk Factors Among Iranian, Ukrainian, Vietnamese Refugees in California, 2002–2011 Nutritional Status and Obstacles to Healthy Eating Among Refugees in Geneva Blood pressure and fasting glucose changes in male migrants waiting for an asylum decision in Italy. A pilot study Male asylum seekers waiting for an asylum decision in Italy for between 0 and 30 months Mixed exposures, majority of asylum seekers from Eritrea, Ghana, Guinea and Nigeria Cross sectional 18 to 40 Screening for infectious diseases among asylum seekers newly arrived in Germany in 2015: a systematic single- centre analysis Asylum seekers at reception centre in Ausberg, Germany in 2015 Various exposures as refugees from multiple locations. Mainly Afghanistan, Albania, Eritrea, Syria and Nigeria Cross sectional 18 to 75 USA = United States of America STEPS = WHO STEPwise approach to surveillance of noncommunicable diseases IDPs = internally displaced people/ persons LDL = low density lipoprotein NY = New York NORNS = North Korean Refugee Health in South Korea study https://doi.org/10.1371/journal.pone.0282823.t001 In terms of the exposure, seventeen reports related to conflict situations [41–57], sixteen to long-standing refugee situations [58–73], thirteen to natural disasters [74–86] and ten included mixed exposures [87–96]. Several crises were the subject of multiple studies. The Great East Japan Earthquake was the exclusive exposure of nine reports [77–84, 86], the internal conflict in Syria of seven reports [41, 50–53, 55, 57] and the Palestinian situation of six reports [58, 61, 62, 70, 71, 73]. The map in Fig 2A shows the countries where exposures occurred, based on the number of reports mentioning those countries. The most frequently examined exposures were in Japan [77–86] (10 reports), followed by Syria [41, 50–53, 55, 57, 94, 96] (9 reports), then Pal- estine [58, 61, 62, 70, 71, 73] (6 reports) and Iraq [41, 42, 45, 47, 48, 55] (6 reports). With regards to setting, thirty five reports were from studies carried out in HICs, twenty were conducted in LMICs and one was conducted in both settings [91]. The map in Fig 2B shows the frequency of reports from different countries. The most common study countries were the United States [45–48, 59, 60, 68, 87–90, 93] (12 reports), then Japan [77–86] (10 reports) followed by Palestine [58, 61, 62, 70, 71, 73] (6 reports). Data reported in this paper were all collected after the exposure had begun. For those in long-standing refugee situations or going through asylum seeking or refugee resettlement pro- cesses, the exposure was considered to be ongoing. For those exposed to natural disasters, data collection took place between 4 months [75] and 4 years after the disaster [80, 82]. Forty reports considered displaced populations, while four considered non-displaced popu- lations [54, 58, 75, 84] and ten included both displaced and non-displaced participants [49, 61, PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 11 / 29 PLOS ONEOverweight and obesity in humanitarian settings Fig 2. A. Showing the frequency of reports by country of exposure. B Showing frequency of reports by study location. Maps were produced in ArcGIS from ESRI using base map data from the World Food Programme accessed via The National Archives (UK). Contains public sector information licensed under the Open Government Licence v3.0. https://doi.org/10.1371/journal.pone.0282823.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 12 / 29 PLOS ONEOverweight and obesity in humanitarian settings 62, 74, 78–83]. Displacement status was unclear for two reports [71, 85]. Of the HIC reports, twenty-six were related to populations who had been displaced from LMICs. The sample size also varied across the studies with some reports including as many as 444356 participants [91] whilst the smallest study only included 28 participants [75]. Only 8 studies reported measures of spread for either population level or subgroup estimates [44, 50, 51, 57, 62, 81, 82, 84]. Although studies may have included children, we have restricted data extraction to adults only and the age range of adults included is seen in Table 1. Most studies sampled a wide range of ages, but some were restricted to young adults [54, 58], middle aged adults [41] or older adults [86]. Most studies included men and women. However, Al-Duais and Al-Awthan [54], Modesti et al [95] and Singh et al [56] included men only. Bhatta et al [59, 60], Bayyari et al [58], El Kishawi et al [61], Herrera-Fontana et al [75], Drummond et al [43] and Dhair and Abed [62] included females only. Prevalence of overweight and obesity Tables 2 and 3 show the prevalence of overweight, obesity and overweight or obesity combined for LICs and HICs, respectively. Forty-seven reports used WHO recommended BMI cut-offs of 25 kg/m2 and 30 kg/m2 to define overweight and obesity, respectively [97] or regional varia- tions. Nine reports used non-standard definitions where a single cut-off was applied for both overweight and obesity, or no justification was given for the choice of cut-off used [43, 45, 65, 76, 77, 80, 81, 83, 85]. Prevalence was reported for the whole populations or by subgroups according to the aim of the study. As can be seen by entries in bold in Tables 2 and 3, where information was not avail- able for the whole population, it could be calculated. In whole populations, prevalence rates for overweight and obesity combined ranged from 6.4% [56] to 82.8% [51]. For overweight alone they ranged from 6.0% [56] to 53% [43] and for obesity alone from 0% [54] to 52.7% [51]. Prevalence of overweight and obesity in subgroups The prevalence ranges for our subgroups of interest are presented in Fig 3, derived from data presented in Appendix Three. The range of prevalence estimates remains wide across overweight, obesity and the two combined in most of the subgroups, however, it appears the widest for the estimates of over- weight and obesity combined. The upper and lower bounds of the obesity estimates are gener- ally lower than those of the overweight estimates. In terms of exposures, conflict generates the widest range of estimates. The mixed expo- sures group generates the narrowest range of estimates across all three metrics. This group includes studies which were carried out as part of routine pre-immigration or post-immigra- tion health checks. The lowest internal ROB and being displaced are associated with the widest range of preva- lence estimates for their subgroups. It is interesting to note that prevalence estimates are similar for studies carried out in both HICs and LMICs. In studies with low internal ROB, estimates for overweight and obesity com- bined were between 6.4% [56] and 82.8% [51] for LMICs and between 15.9% [95] and 65% [41] for HICs. Taking a closer look at geography, reports from the NORNS study [63–66] report relatively homogeneous results (overweight 22% [63] to 23.4% [64], obesity 18.9% [64] to 20.34% [65] and the two combined 42.1% [63] to 43% [66]), whereas those from earthquake afflicted populations in Japan are varied (overweight 23.2% [86] to 29.4% [78], obesity 3.4% PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 13 / 29 PLOS ONETable 2. Showing prevalence of overweight and obesity in studies carried out in low and middle income countries. First Author, Year Number (proportion %) with overweight Number (proportion %) with obesity Number (proportion %) with overweight or obesity Value of denominator Country BMI cut-offs applied Displaced Overweight and obesity in humanitarian settings 11898 1038 907 2134 77 5407 5492 1400 251 410 357 525 160 689 1082 1178 207 330 267 28 164 Jordan Jordan Jordan Lebanon Yemen Turkey Turkey Ukraine Standard Standard Standard Standard Standard Standard Standard Standard Yes Yes Yes Yes No Yes Yes Mixed Kenya Standard Yes Palestine Palestine Palestine Palestine Palestine Palestine Iran Iran Algeria Solomon Islands Standard Standard Standard Standard Standard Standard Standard Standard Standard No Mixed Yes Mixed Yes Unclear Yes Yes Yes Standard Mixed Ecuador Standard No Nepal non- standard Yes Yanni et al 2013 [42] 4495 (38) Amr et al 2019 [55] NR Ratnayake et al 2020 [51] Mansour et al 2020 [50] Al Duais and Al Awthan 2019 [54] ^ 273 (30.1) 661 (34.2) 17 (22.1) 3982 (34) NR 487 (52.7) 573 (28.6) 0 Balcilar 2016 [57] 1762 (32.6) 1498 (27.7) Eryurt and Menet 2020 [52] NR Greene-Cramer et al 2020 [49] Singh et al 2015 [56]** NR 15 (6) Bayyari et al 2013 [58] * El Kishawi et al 2014 [61] * Kory et al 2013 [73] Dhair and Abed 2020 [62] ^^ Damiri et al 2018 [70] Damiri et al 2019 [71] Abdollahi et al 2015 [67] Taherifard et al 2021 [69] Naigaga et al 2018 [72] 51 (12.4) 123 (34.5) 166 (31.6) 61 (38.1) 188 (27.3) 348 (32.2) 440 (37.3) 58 (28.0) 91 (27.6) Conflict 8477 (71) 628 (60.5) 751 (82.8) 1234 (57.8) 17 (22) 3260 (60.3) 3503 (64) 241 (17.2) NR NR 1 (0.4) 16 (6.4) Long-standing Refugee Situation 7 (1.7) 105 (29.7) 93 (17.7) 36 (22.5) 246 (35.7) 345 (33.1) 239 (20.3) 46 (22.2) 49 (14.8) 58 (14.1) 228 (63.9) 259 (49.3) 97 (60.6) 435 (63.1) 693 (64.1) 769 (57.6) 104 (50.2) 140 (42) Natural disasters Furusawa et al 2011 [74] 122 (45.7) 41 (15.4) 163 (61.0) Herrera-Fontana et al 2020 [75]* 10 (35.7) 6 (21.4) 16 (57.1) Adrega et al 2018 [76] + NR NR 41 (25) NR = not reported and can’t be calculated from available information Standard = WHO definition of overweight and obesity used. Regional = regional cut-offs as defined by national guidelines for overweight and obesity used Non-standard = neither WHO nor regionally defined cut-offs used. Green = low internal ROB Orange = moderate internal ROB No Fill = high internal ROB * = females only ** = males only ^ = males only. Data for non-Qat chewers presented ^^ = Females only. Data for controls presented + = reported as any BMI > = 25 https://doi.org/10.1371/journal.pone.0282823.t002 [78] to 32.7% [83] and the two combined 26.4% [85] to 81.6% [81]). Several studies examined populations from the Middle East region [42, 45, 47, 48, 50–55, 57, 58, 61, 62, 67, 69–71, 73]. The range of estimates for overweight (12.4% [58] to 38.3% [47]), obesity (0% [54] to 52.7% [51] and the two combined (14.1% [58] to 82.8% [51]) remain wide. Nguyen et al compared PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 14 / 29 PLOS ONETable 3. Showing prevalence of overweight and obesity in studies carried out in high income countries. First Author, Year Number (proportion %) with overweight Number (proportion %) with obesity Number (proportion %) with overweight or obesity Value of denominator Country of study BMI cut-offs applied Displaced Overweight and obesity in humanitarian settings Chandra et al 2019 [41] Drummond et al 2011 [43] * Renzaho et al 2014 [44] Maldari et al 2019 [53] Reznar et al 2020 [45] Sastre et al 2020 [46] 107 (45) 27 (53) 97 (30.9) 48 (25.8) NR 17 (37) Jen et al 2015 [47] Jen et al 2018 [48] + NR 111 (38.3) Bhatta et al 2014 [60] * NR Bhatta et al 2015 [5] * 38 (35.2) 66 (28) 14 (27) 63 (20.1) Conflict 173 (65) 41 (80) 160 (51) 82 (44.1) 130 (69.9) NR 12 (26) 56 (19.3) NR NR 32 (29.6) 403 (65.7) 29 (63) 167 (57.6) NR Long-standing Refugee Situation 70 (64.8) 70 (64.8) 34 (52) 237 51 314 186 613 48 298 282 108 108 66 381 708 Australia Standard Yes Australia non-standard Yes Australia Standard Australia Standard Yes Yes non-standard Yes USA USA USA USA USA USA USA Standard Standard Standard Regional Regional Standard Yes Yes Yes Yes Yes Yes Yes Kumar et al 2014 [68] Kim et al (NORNS 2) 2015 [64] Kim et al (NORNS 3) 2016 [65] Kim et al 2018 (NORNS 1) [63] Jeong et al (NORNS 4) 2017 [66] Sakai et al FHMS1 2020 [77] Ohira et al FHMS2 2016 [78] Ohira et al FHMS3 2017 [79] Satoh et al FHMS4 2021 [80] Takahashi et al (2) 2016 [81]++ Takahashi et al (3) 2020 [82] Takahashi et al (1) 2021 [83] Hikichi et al 2019 [86] Ebner et al 2016 [84] Nakamura et al 2019 [85]++ 28 (42) 6 (9) 89 (23.4) 72 (18.9) 161 (42.3) South Korea Regional NR 203 (22) 34 (22.8) 144 (20.34) NR South Korea non-standard Yes 183 (20) 30 (20.1) 386 (42.1) 917 (BMI) South Korea Regional 64 (43) 149 South Korea Regional Yes Yes (NR) 9718 (33.2) (NR) Natural disasters 8070 (29.4) 928 (3.4) 8998 (32.7) NR NR NR NR NR NR NR NR NR 5359 (28.7) 2122 (81.6) 1544 (21.1) 3238 (32.7) NR NR 827 (23.2) 967 (27.1) 1794 (50.3) NR NR 277 (35.3) NR NR 1496 (26.4) Mixed 29267 27486 27486 18670 6601 7318 9897 3567 784 5674 Japan Japan Japan Japan Japan Japan Japan Japan Japan Japan non-standard Yes Standard Mixed Standard Mixed non-standard Mixed non-standard Mixed Standard Mixed non-standard Mixed Standard Regional Yes No non-standard Unclear PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 (Continued ) 15 / 29 PLOS ONETable 3. (Continued) First Author, Year Number (proportion %) with overweight 345 (32.7) 345 (32.7) NR NR Mulugeta et al 2018 (1) [87] Mulugeta et al 2019 (2) [88] Mulugeta et al 2019 (3) [89] Rhodes et al 2016 [90]^ Bardenheier et al 2019 (1) [91] Bardenheier et al 2019 (2) [92] Overweight and obesity in humanitarian settings Number (proportion %) with obesity Number (proportion %) with overweight or obesity Value of denominator Country of study BMI cut-offs applied Displaced 158 (15) 158 (15) 361 (23) 17 (13.5) 503 (47.7) 503 (47.7) NR NR 1055 1055 1570 126 USA USA USA USA Standard Standard Standard Standard Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 73251 21968 352 195 1246 (USA) ^^^ Standard and regional USA Standard Switzerland Standard Italy Standard Germany Standard 120993 (27.2) 85231 (19.2) 206224 (46.4) 444356 (USA) ^^ Standard 22136 (30.2) 7342 (10) 29478 (40.2) Nguyen and Rehkopf, 2016 [93] NR 4457 (20.3) NR Amstutz et al 2020 [94] Modesti et al 2020 [95] ** Kortas et al 2017 [96] 110 (31.2) 38 (10.8) 148 (42) 30 (15.4) 295 (9.3) 1 (0.5) 31 (15.9) 116 (23.7) 411 (33) NR = not reported and can’t be calculated from available information Standard = WHO definition of overweight and obesity used. Regional = regional cut-offs as defined by national guidelines for overweight and obesity used Non-standard = neither WHO nor regionally defined cut-offs used. Green = low internal ROB Orange = moderate internal ROB No fill = high internal ROB * = Females only ** = Males only + = results presented graphically with insufficient information to extract data ++ = reported as any BMI > = 25 ^ = presenting data for refugees only ^^ = included US bound refugees from a wide range of countries ^^^ = included US bound Burmese refugees from a number of South East Asian countries https://doi.org/10.1371/journal.pone.0282823.t003 refugees who had relocated to California from Iraq, Vietnam and Ukraine. They report that those from Ukraine were more likely to be obese or severely obese than the other nationalities, (Adjusted Odds Ratio (AOR) 2.1; CI 1.9–2.3) and (AOR 2.5; CI 2.1–2.8) respectively [93]. Five reports had strikingly low estimates [54, 56, 58, 95, 96] with a combined prevalence of between 6.4% [56] and 23.7% [96].Three reports included male participants only [54, 56, 95], and one was majority (75.4%) male [96]. They also included younger participants either as part of their sampling strategy or due to attendees at the services. The report with the youngest mean age was Bayyari et al, at 20.1 (standard deviation (SD) 1.2) years [58] and the oldest was Singh et al at 37 (SD 16) years [56]. Three reports had strikingly high prevalence estimates [43, 51, 81]. They all sampled popu- lations originating from and living in different geographical settings and with different expo- sures. Drummond et al [43] examined West African women only with a mean age of 35 (SD 10.6) years and found a combined prevalence of 80%. Ratnayake et al [51] and Takahashi et al PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 16 / 29 PLOS ONEOverweight and obesity in humanitarian settings Fig 3. Showing the range of prevalence estimates in subgroups of interest for overweight, obesity and the two combined. https://doi.org/10.1371/journal.pone.0282823.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 17 / 29 PLOS ONEOverweight and obesity in humanitarian settings [81] sampled both sexes with an older mean age of 56 (SD 13,2) and 60 years for men, 64 years for women, respectively and found combined prevalences of 82.8% and 81.6% respectively. Some reports did formally compare overweight and / or obesity prevalence estimates between men and women. Several found that these measures were higher in women compared to men [51, 52, 69, 71, 72, 93, 94]. However, Mansour et al found higher rates of obesity in women, but no difference in rates of overweight between the sexes [50]. Damiri et al and Balci- lar also report that whilst obesity is more prevalent in women, overweight is more prevalent in men [57, 70]. All the studies that explored the relationship between age and adiposity found that preva- lence estimates increased with age [45, 51, 52, 57, 59, 61, 69, 72, 93, 94]. Changes over time and with displacement Findings related to displacement and longitudinal changes are difficult to tease out, so are reported here together. Whilst studies did not report on incidence specifically, ten reports mention change in prev- alence of obesity and overweight over time. Nine [47, 48, 66, 78, 79, 84, 86, 87, 90] of these reports considered populations in HICs and only one was in a LMIC [73]. All dealt with dis- placed populations. Of the ten reports, seven noted an increase [47, 48, 78, 79, 86, 87, 90] and one [73] reported no change over time. Two reports suggested an initial increase followed by a decrease, stabilisation or loss of adiposity [66, 84]. Of those that reported increases in overweight or obesity, Jen et al found that in the first year following relocation to the United States there was a significant increase in BMI and an upward shift in the prevalence of overweight and obesity amongst refugee populations [47]. Mulugeta et al found that for every additional year refugees lived in the USA, the risk of overweight or obesity increased by 23% among men (Odds Ratio (OR) = 1.23; 95% CI = 1.09–1.39) and 18% among women (OR = 1.18; 95% CI = 1.04–1.35) when adjusted for confounders [87]. Takahashi et al contribute further to the importance of place of displacement. They report significant increases in body weight in people relocated to temporary housing compared to those not relocated over a five year observation period [83]. Considering changes in BMI over time without categorising into overweight and obesity, three reports noted increases in BMI [65, 66, 94]. However, Modesti et al found no strong evi- dence for an association between time in an Italian immigration centre and increase in BMI over a 30 month period [95]. Four reports formally compared changes in adiposity before and after exposure to the Great East Japan Earthquake [78, 79, 84, 86]. Hikichi et al report that approximately 2.5 years after the disaster, the prevalence of obesity had increased amongst those displaced (25.0% to 35.1%) but decreased amongst those not displaced (26.9% to 26.6%) compared to 7 months before the disaster. [86] Ebner et al report that the OR of obesity was higher in the year after the disaster, but that this risk was no longer significant in the second year after the disaster (OR 1.31 (95% CI 1.06 to 1.38) and 1.07 (95% CI 0.93 to 1.24) respectively) [84]. Ohira et al report that BMI and obesity increased in earthquake affected populations. This increase was greater in those evacuated compared to those not evacuated and greater in males compared to females [78, 79]. The multivariable adjusted hazard ratio for overweight after the disaster was 1.61 (95% CI 1.47 to 1.77) [78]. Only one non-earthquake study compared BMI before and after exposure. No change was found [73]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 18 / 29 PLOS ONEOverweight and obesity in humanitarian settings Other outcomes The other outcomes of interest were considered less frequently. There were no papers reporting on the cascade of care for obesity. However, attempts have been made to gather information about risk factors for higher BMI and targets for primary prevention. Balcilar, 2016 reports that 14.1% of Syrian refugees in Turkey were advised to reduce their fat intake [57]. Several reports from countries in the Middle East show that there are poor levels of fruit and vegetable intake and low levels of physical activity in refugee populations in general [50, 52, 57, 69]. We did not identify any qualitative studies which met our selection criteria. However, in a cross-sectional study measuring both self-perceived body size and BMI in Saharawi refugees, Naigaga et al found that there was a preference for overweight applied to individuals of the opposite sex [72]. Comparing perceived body size to BMI indicated that obese men and women did not wish to gain weight and most obese or overweight women wanted to lose weight. In a study which included service providers and refugees living in Geneva and not selected by BMI, Amstutz et al found that fruit and vegetables were considered healthy and that lan- guage and financial hardship were the main barriers to a healthy diet [94]. Alternative measures of adiposity were infrequently used with only thirteen studies record- ing waist circumference (WC) [59, 60, 63–66, 69, 70, 76, 79, 82, 84, 94]. Risk of bias S4 Appendix gives details of ROB for each study and Fig 4 summarises this across all studies. There was evidence of good practice in this challenging field, but only six studies were at low ROB overall and nine studies at moderate ROB. The challenge of achieving low ROB in the external domain was largely around choice of sampling frame and methods of participant selection. With internal ROB, the use of self-reported measures, definition of overweight and obesity and some unclear reporting were the main problems noted. Discussion This review aimed to explore the prevalence and incidence of overweight and obesity, and the changes in adiposity over time in populations directly affected by humanitarian crises; the cas- cade of care in these populations and perceptions of patients with overweight and obesity. We included 56 reports derived from 45 studies. We found that prevalence estimates varied widely across the included studies and within subgroups based on study setting, internal ROB, expo- sure type and displacement status. Most studies report an increase in adiposity over time and compared to pre-exposure measures [47,48,63,78,79,87,90]. However, this relationship appears to be affected by displacement status. There were no reports detailing the cascade of care, but there is some evidence of limited physical exercise alongside a high calorie, low fruit and vege- table diet in refugee settings [50, 52, 57, 69]. We did not identify any studies in which the views of patients with obesity were sought qualitatively. However, a cross-sectional study did demon- strate cultural norms may differ in different settings [72]. Burden of disease Estimates of overweight range from 6.0% [56] to 53% [43]; for obesity from 0%[54] to 52.7% [51]; and for the two combined from 6.4% [56] to 82.8% [51]. These wide ranges persist in studies at low internal ROB. We did not identify any studies with no overweight and only one study with no obesity [54]. Whilst we were expecting to find overweight and obesity, we were surprised by the extent and ubiquitousness of the issue. PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 19 / 29 PLOS ONEOverweight and obesity in humanitarian settings Fig 4. Showing the risk of bias across the included reports, by domain. https://doi.org/10.1371/journal.pone.0282823.g004 Generating a global prevalence estimate for obesity is complex. WHO estimates suggest that in 2016, 39% of adults were overweight and 13% were obese [97]. In the Global Burden of Disease (GBD) Study [39], estimates of overweight and obesity were higher in developed coun- tries than developing countries in 2013. Our review would suggest that this pattern is not con- sistently seen in crisis-affected populations. Japanese and South Korean populations are the subject of nearly half of the HIC papers and these countries have some of the lowest levels of obesity and overweight for high income countries in the world [98]. For other HIC studies, populations came from LICs and were likely to be faced with poverty and other challenges in their new settings. The GBD study points out that BMI tends to reach a peak at around 55 years in men and 60 years in women and that more women than men have a BMI greater than 25 kg/m2 [39]. Sev- eral studies in this review formally tested the change in obesity and overweight estimates with increasing age and by sex. With regards to age, there was a consistent relationship between increasing age and increasing adiposity [45, 51, 52, 57, 59, 61, 69, 72, 93, 94]. With regards to sex, women were commonly found to have a higher rates of overweight and / or obesity than men in most reports [51, 52, 69, 71, 72, 93, 94]. However, it would appear that in some popula- tions sexes are differentially affected by overweight and obesity [57, 70]. All except two [70, 71] of these reports had low internal ROB. These findings suggest that service providers can expect to find more overweight and obesity in older adults and females within a crisis affected popula- tion. The heterogeneity of our studies and the moderate to high external ROB means that it is difficult to generalise the extent of these differences. Longitudinal changes in BMI are a function of age (as described above) and are part of the migration experience [99]. In migrant populations more generally, an initial health advantage PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 20 / 29 PLOS ONEOverweight and obesity in humanitarian settings is superseded by increased risk of overweight and obesity compared to the native population. These changes are dependent on where the migrant comes from and how long they remain in the host country [100–102] All the studies reporting on these changes involved displaced pop- ulations making it difficult to comment on the differential effects of exposure to crises, accul- turation and secular trends. Only five reports commenting on longitudinal changes were at low internal ROB [63, 66, 84, 94, 95]. Two reports from the NORNS study show that increasing duration in South Korea was associated with increase in weight, but that in some individuals weight loss is seen after the initial settling period [63, 66]. The Modesti report, which included only males, had a relatively short follow up period of 30 months which may explain why there was no significant change in adiposity seen [95]. And Ebner et al reported on individuals who had returned home after a relatively short displacement which may explain the change in tra- jectory of weight gain over time [84]. The Ohira et al papers have generated hazard ratios which show an increased risk of over- weight and / or obesity with exposure to earthquakes [78, 79]. It is tempting to interpret this as evidence of a causal link between exposure to earthquakes and weight gain. However, the stud- ies used observational data and a causal framework was not specified. These findings would suggest that service providers, particularly in protracted situations, need to be prepared for increasing levels of overweight and obesity and the cardiometabolic complications that come with this. Whilst we do not propose that overweight and obesity are addressed in the immediate aftermath of a crisis, this pattern of weight gain points to an opportunity to take preventative action early in the time frame of a crisis. As seen in other reviews of crisis affected populations [16, 24, 25], the geographical distribu- tion of the studies and the people being examined is skewed. There is a long history of measur- ing adiposity as part of monitoring the impact of food aid. At the full text screening stage, we found that 63 reports did carry out anthropometric measurements, but that results were either presented in a way which did not allow categorisation or only those who were underweight or malnourished were reported (See S2 Appendix). This does mean that our range of prevalence estimates may have lower minimum bounds than we have identified. It also suggests that changing monitoring and reporting requirements would provide more information about the true prevalence of overweight and obesity and would clarify targets where more research would be most beneficial. Cascade of care We could not identify evidence of information or interventions being directed specifically at those who were overweight or obese. It is likely that this reflects a genuine lack of interventions aimed at weight loss rather than NCD management more generally. However, there is evi- dence that, particularly those studies following WHO STEPS processes [103], were able to identify NCD risk factors. This provides a starting point for the discussion about targets for primary and secondary prevention. Namely, access to low calorie, high nutritional value food and promotion of active lifestyles [50, 52, 57, 69]. Looking at the cascade of care in NCD management more broadly, several recurring research and information gaps are noted. There is generally poor collection of standard data regarding disease states and recognised risk factors, there is a paucity of evidence to guide interventions, and there are infrastructure and supply problems even for those conditions in which treatments are available [16, 17, 31, 104]. Many of these factors are applicable to over- weight and obesity. With the additional challenge that overweight and obesity are considered much later in the crisis response [15], by which time resources are arguably too stretched to extend to further activities. PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 21 / 29 PLOS ONEOverweight and obesity in humanitarian settings Patient perceptions We did not identify qualitative studies in which we could differentiate the voices of those with overweight and obesity from other participants. However, cultural ideals and norms in rela- tions to body size and shape were noted [72]. This is echoed particularly in work examining the understanding of African refugees who described the pursuit of thinness as perplexing [105]. Language and financial barriers to seeking care for overweight and obesity and also part of the refugee and migrant experience of seeking health care in general [106, 107]. Crisis affected populations are largely city dwellers [6] and as such multi-pronged and multi-level interventions are needed for both prevention and treatment [108]. However, it is acknowledged that population level weight loss interventions are challenging to implement and sustain even in well-resourced settings [109]. Causal pathways in obesity are complex [109]. Qualitative work is key to understanding the causal relationships between perceptions, understanding and behaviour. We cannot expect to successfully influence disease trajectories without this information. Use of waist circumference We were surprised that only13 studies recorded WC [59, 60, 63–66, 69, 70, 76, 79, 82, 84, 94]. There is ongoing discussion about the most appropriate measure to determine increased adi- posity [110]. Waist circumference provides important additional information in assessing the risk of death and disease due to increased adiposity [111]. However, WC is no longer explicitly mentioned WHO’s Package of essential non-communicable disease interventions [112]. Risk of bias ROB poses a challenge in the interpretation of reported results. Identifying representative sam- ples in crisis settings is a challenge, particularly with the chaos associated with displacement. The majority of studies weighed and measured their participants directly or used health rec- ords where these measurements were recorded. Several of the papers, however, used self reported heights and weights. These were coded as having a high internal ROB given the potentially inaccurate measurements. In Bhatta et al’s 2015 report, 7 people reported them- selves to be overweight but the BMI data showed 70 people out of 120 to be overweight; the dif- ference being a factor of 10 [59]. Strengths and Limitations One of the main strengths of this review is the number of reports included in the final analysis. These were identified in a systematic manner across databases and online repositories. The reports cover different crises and different regions of the world. Though this gives our analysis breadth, this heterogeneity means that we were unable to perform a meta-analysis and that the findings are not generalisable across all settings. We used simple mathematics to derive missing prevalence estimates (marked in bold in Tables 2 and 3). In some cases, this involved calculations across multiple subgroups. This approach was taken as a pragmatic alternative to requesting access to individual patient data. We only included publications from Jan 2011 onwards. On one hand this is a strength as it allows for a contemporaneous picture to emerge. On the other hand, it could be viewed as a weakness, since we will be missing patterns in change over time. We did not identify reports discussing the cascade of care or the perception of patients with overweight or obesity. We believe that this genuinely reflects a paucity of data of this type. However, an alternative search strategy may have yielded different results. For example, PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 22 / 29 PLOS ONEOverweight and obesity in humanitarian settings searching for the study type in the settings of interest and then screening for the disease could unearth different information. Conclusion This study has shown that the prevalence of overweight and obesity vary in crisis affected pop- ulations but are rarely absent. Increases in adiposity over time, in older adults and in women are likely to be seen in most populations. Better quality descriptive information would help to identify precisely to who and when interventions should be offered in different settings. The lack of information about the cascade of care likely reflects limited efforts to address over- weight and obesity in these settings. The lack of qualitative research hampers our understand- ing of which interventions would be most likely to succeed. WC measures should be included as part of standard care. Supporting information S1 Checklist. PRISMA 2020 main checklist. (DOCX) S1 Appendix. Full search strategies. (DOCX) S2 Appendix. Studies excluded at full text screening. (DOCX) S3 Appendix. Prevalence ranges in subgroups. (DOCX) S4 Appendix. Details of risk of bias assessment. (DOCX) Author Contributions Conceptualization: Saran Shantikumar, Ammar Sabouni, Farah Kidy. Data curation: Majel McGranahan, Farah Kidy. Methodology: Thomas Shortland, Majel McGranahan, Daniel Stewart, Oyinlola Oyebode, William Proto, Ammar Sabouni, Farah Kidy. Project administration: Farah Kidy. Supervision: Oyinlola Oyebode, Saran Shantikumar, Farah Kidy. Visualization: Gavin Rudge. Writing – original draft: Thomas Shortland. Writing – review & editing: Majel McGranahan, Daniel Stewart, Oyinlola Oyebode, Saran Shantikumar, William Proto, Bassit Malik, Roger Yau, Maddie Cobbin, Ammar Sabouni, Farah Kidy. References 1. Swinburn BA, Kraak VI, Allender S, Atkins VJ, Baker PI, Bogard JR, et al. The Global Syndemic of Obesity, Undernutrition, and Climate Change: The Lancet Commission report. Lancet. 2019; 393: 791–846. https://doi.org/10.1016/S0140-6736(18)32822-8 PMID: 30700377 PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 23 / 29 PLOS ONEOverweight and obesity in humanitarian settings 2. Noncommunicable diseases. [cited 20 Jan 2022]. Available: https://www.who.int/news-room/fact- sheets/detail/noncommunicable-diseases 3. Ma C, Avenell A, Bolland M, Hudson J, Stewart F, Robertson C, et al. Effects of weight loss interven- tions for adults who are obese on mortality, cardiovascular disease, and cancer: systematic review and meta-analysis. BMJ. 2017; 359: j4849. https://doi.org/10.1136/bmj.j4849 PMID: 29138133 4. Transforming our world: the 2030 Agenda for Sustainable Development. [cited 20 Jan 2022]. Avail- able: https://sdgs.un.org/2030agenda 5. Cai Z, Yang Y, Zhang J. Obesity is associated with severe disease and mortality in patients with coro- navirus disease 2019 (COVID-19): a meta-analysis. BMC Public Health. 2021; 21: 1505. https://doi. org/10.1186/s12889-021-11546-6 PMID: 34348687 6. United Nations High Commissioner for Refugees. Global Trends in Forced Displacement– 2020. In: UNHCR [Internet]. [cited 20 Jan 2022]. Available: https://www.unhcr.org/statistics/unhcrstats/ 60b638e37/global-trends-forced-displacement-2020.html 7. Gouda HN, Charlson F, Sorsdahl K, Ahmadzada S, Ferrari AJ, Erskine H, et al. Burden of non-com- municable diseases in sub-Saharan Africa, 1990–2017: results from the Global Burden of Disease Study 2017. Lancet Glob Health. 2019; 7: e1375–e1387. https://doi.org/10.1016/S2214-109X(19) 30374-2 PMID: 31537368 8. GBD 2015 Eastern Mediterranean Region Obesity Collaborators. Burden of obesity in the Eastern Mediterranean Region: findings from the Global Burden of Disease 2015 study. Int J Public Health. 2018; 63: 165–176. https://doi.org/10.1007/s00038-017-1002-5 PMID: 28776243 9. Marinho FM, Soliz P, Gawryszewski V, Gerger A. Epidemiological transition in the Americas: changes and inequalities. The Lancet. 2013. p. S89. https://doi.org/10.1016/s0140-6736(13)61343-4 10. Kohrt BA, Mistry AS, Anand N, Beecroft B, Nuwayhid I. Health research in humanitarian crises: an urgent global imperative. BMJ Glob Health. 2019; 4: e001870. https://doi.org/10.1136/bmjgh-2019- 001870 PMID: 31798999 11. Cooper K. NCD care in humanitarian settings: a clarion call to civil society. Danish Red Cross; 2018. Available: https://www.rodekors.dk/sites/rodekors.dk/files/2018-10/NCD%20care%20in% 20humanitarian%20settings%20-%20a%20clarion%20call%20to%20civil%20society.pdf 12. Aebischer Perone S, Martinez E, du Mortier S, Rossi R, Pahud M, Urbaniak V, et al. Non-communica- ble diseases in humanitarian settings: ten essential questions. Confl Health. 2017; 11: 17. https://doi. org/10.1186/s13031-017-0119-8 PMID: 28932259 13. Slama S, Kim H-J, Roglic G, Boulle P, Hering H, Varghese C, et al. Care of non-communicable dis- eases in emergencies. Lancet. 2017; 389: 326–330. https://doi.org/10.1016/S0140-6736(16)31404-0 PMID: 27637675 14. Zablith N, Diaconu K, Naja F, El Koussa M, Loffreda G, Bou-Orm I, et al. Dynamics of non-communica- ble disease prevention, diagnosis and control in Lebanon, a fragile setting. Confl Health. 2021; 15: 4. https://doi.org/10.1186/s13031-020-00337-2 PMID: 33430916 15. Bausch FJ, Beran D, Hering H, Boulle P, Chappuis F, Dromer C, et al. Operational considerations for the management of non-communicable diseases in humanitarian emergencies. Confl Health. 2021; 15: 9. https://doi.org/10.1186/s13031-021-00345-w PMID: 33632275 16. Kehlenbrink S, Smith J, Ansbro E´ , Fuhr DC, Cheung A, Ratnayake R, et al. The burden of diabetes and use of diabetes care in humanitarian crises in low-income and middle-income countries. Lancet Diabetes Endocrinol. 2019; 7: 638–647. https://doi.org/10.1016/S2213-8587(19)30082-8 PMID: 30878268 17. Song K, Lee A. Factors influencing the effective management of diabetes during humanitarian crises in low- and middle-income countries: a systematic review. Public Health. 2021; 199: 110–117. https:// doi.org/10.1016/j.puhe.2021.08.020 PMID: 34626911 18. Greene MC, Kane JC, Khoshnood K, Ventevogel P, Tol WA. Challenges and opportunities for imple- mentation of substance misuse interventions in conflict-affected populations. Harm Reduct J. 2018; 15: 58. https://doi.org/10.1186/s12954-018-0267-1 PMID: 30486840 19. Ezard N. Substance use among populations displaced by conflict: a literature review. Disasters. 2012. pp. 533–557. https://doi.org/10.1111/j.1467-7717.2011.01261.x PMID: 22066703 20. 21. Lo J, Patel P, Roberts B. A systematic review on tobacco use among civilian populations affected by armed conflict. Tob Control. 2016; 25: 129–140. https://doi.org/10.1136/tobaccocontrol-2014-052054 PMID: 25770116 Lo J, Patel P, Shultz JM, Ezard N, Roberts B. A Systematic Review on Harmful Alcohol Use Among Civilian Populations Affected by Armed Conflict in Low- and Middle-Income Countries. Subst Use Mis- use. 2017; 52: 1494–1510. https://doi.org/10.1080/10826084.2017.1289411 PMID: 28471305 PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 24 / 29 PLOS ONEOverweight and obesity in humanitarian settings 22. Weaver H, Roberts B. Drinking and Displacement: A Systematic Review of the Influence of Forced Displacement on Harmful Alcohol Use. Substance Use & Misuse. 2010. pp. 2340–2355. https://doi. org/10.3109/10826081003793920 PMID: 20469970 23. Babaie J, Pashaei Asl Y, Naghipour B, Faridaalaee G. Cardiovascular Diseases in Natural Disasters; a Systematic Review. Arch Acad Emerg Med. 2021; 9: e36. https://doi.org/10.22037/aaem.v9i1.1208 PMID: 34027431 24. Jawad M, Vamos EP, Najim M, Roberts B, Millett C. Impact of armed conflict on cardiovascular dis- ease risk: a systematic review. Heart. 2019; 105: 1388–1394. https://doi.org/10.1136/heartjnl-2018- 314459 PMID: 31138670 25. Keasley J, Oyebode O, Shantikumar S, Proto W, McGranahan M, Sabouni A, et al. A systematic review of the burden of hypertension, access to services and patient views of hypertension in humani- tarian crisis settings. BMJ Glob Health. 2020; 5. https://doi.org/10.1136/bmjgh-2020-002440 PMID: 33168520 26. Massey E, Smith J, Roberts B. Health needs of older populations affected by humanitarian crises in low- and middle-income countries: a systematic review. Conflict and Health. 2017. https://doi.org/10. 1186/s13031-017-0133-x PMID: 29238401 27. Naja F, Shatila H, El Koussa M, Meho L, Ghandour L, Saleh S. Burden of non-communicable diseases among Syrian refugees: a scoping review. BMC Public Health. 2019; 19: 637. https://doi.org/10.1186/ s12889-019-6977-9 PMID: 31126261 28. Ngaruiya C, Bernstein R, Leff R, Wallace L, Agrawal P, Selvam A, et al. Systematic review on chronic non-communicable disease in disaster settings. bioRxiv. medRxiv; 2020. https://doi.org/10.1101/ 2020.10.15.20213025 29. De Rubeis V, Lee J, Anwer MS, Yoshida-Montezuma Y, Andreacchi AT, Stone E, et al. Impact of disasters, including pandemics, on cardiometabolic outcomes across the life-course: a systematic review. BMJ Open. 2021; 11: e047152. https://doi.org/10.1136/bmjopen-2020-047152 PMID: 33941635 30. Jaung MS, Willis R, Sharma P, Aebischer Perone S, Frederiksen S, Truppa C, et al. Models of care for patients with hypertension and diabetes in humanitarian crises: a systematic review. Health Policy Plan. 2021; 36: 509–532. https://doi.org/10.1093/heapol/czab007 PMID: 33693657 31. Shah S, Munyuzangabo M, Gaffey MF, Kamali M, Jain RP, Als D, et al. Delivering non-communicable disease interventions to women and children in conflict settings: a systematic review. BMJ Global Health. 2020; 5: e002047. https://doi.org/10.1136/bmjgh-2019-002047 PMID: 32341086 32. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021; 372: n71. https://doi. org/10.1136/bmj.n71 PMID: 33782057 33. Campbell M, McKenzie JE, Sowden A, Katikireddi SV, Brennan SE, Ellis S, et al. Synthesis without meta-analysis (SWiM) in systematic reviews: reporting guideline. BMJ. 2020; 368: l6890. https://doi. org/10.1136/bmj.l6890 PMID: 31948937 34. Employment-Based Immigrant Visas. [cited 22 Feb 2022]. Available: https://travel.state.gov/content/ travel/en/us-visas/immigrate/employment-based-immigrant-visas.html#fourth 35. Ouzzani M, Hammady H, Fedorowicz Z, Elmagarmid A. Rayyan-a web and mobile app for systematic reviews. Syst Rev. 2016; 5: 210. https://doi.org/10.1186/s13643-016-0384-4 PMID: 27919275 36. Hoy D, Brooks P, Woolf A, Blyth F, March L, Bain C, et al. Assessing risk of bias in prevalence studies: modification of an existing tool and evidence of interrater agreement. J Clin Epidemiol. 2012; 65: 934– 939. https://doi.org/10.1016/j.jclinepi.2011.11.014 PMID: 22742910 37. Brice R. CASP CHECKLISTS. In: CASP—Critical Appraisal Skills Programme [Internet]. 12 Oct 2020 [cited 24 Jan 2022]. Available: https://casp-uk.net/casp-tools-checklists/ 38. Caleyachetty R, Barber TM, Mohammed NI, Cappuccio FP, Hardy R, Mathur R, et al. Ethnicity-specific BMI cutoffs for obesity based on type 2 diabetes risk in England: a population-based cohort study. Lan- cet Diabetes Endocrinol. 2021; 9: 419–426. https://doi.org/10.1016/S2213-8587(21)00088-7 PMID: 33989535 39. Ng M, Fleming T, Robinson M, Thomson B, Graetz N, Margono C, et al. Global, regional, and national prevalence of overweight and obesity in children and adults during 1980–2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet. 2014; 384: 766–781. https://doi.org/10.1016/ S0140-6736(14)60460-8 PMID: 24880830 40. World Bank Country and Lending Groups–World Bank Data Help Desk. [cited 15 Aug 2022]. Available: https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-lending- groups PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 25 / 29 PLOS ONEOverweight and obesity in humanitarian settings 41. Chandra M, Duri A, Smith M. Prevalence of chronic disease risk factors in 35- to 44-year-old humani- tarian arrivals to New South Wales (NSW), Australia. Aust J Prim Health. 2019; 25: 19–23. https://doi. org/10.1071/PY18042 PMID: 30678748 42. Yanni EA, Naoum M, Odeh N, Han P, Coleman M, Burke H. The health profile and chronic diseases comorbidities of US-bound Iraqi refugees screened by the International Organization for Migration in Jordan: 2007–2009. J Immigr Minor Health. 2013; 15: 1–9. https://doi.org/10.1007/s10903-012-9578- 6 PMID: 22307545 43. Drummond PD, Mizan A, Burgoyne A, Wright B. Knowledge of cardiovascular risk factors in West Afri- can refugee women living in Western Australia. J Immigr Minor Health. 2011; 13: 140–148. https://doi. org/10.1007/s10903-010-9322-z PMID: 20140515 44. Renzaho AMN, Bilal P, Marks GC. Obesity, type 2 diabetes and high blood pressure amongst recently arrived Sudanese refugees in Queensland, Australia. J Immigr Minor Health. 2014; 16: 86–94. https:// doi.org/10.1007/s10903-013-9791-y PMID: 23576028 45. Reznar M, Dallo F, Kurecka AJ, Ruterbusch JJ, Kendra MS, and Tariq M. The Burden of Chronic Health Conditions among Iraqi Refugees in Michigan. Michigan Journal of Public Health. 2020; 10. Available: https://scholarworks.gvsu.edu/mjph/vol10/iss1/6 46. 47. 48. From the Democratic Republic of the Congo to North Carolina: An examination of chronic disease risk. J Refug Glob Health. 2020;3. https://doi.org/10.18297/rgh/vol3/iss1/3 Jen K-LC, Zhou K, Arnetz B, Jamil H. Pre- and Post-displacement Stressors and Body Weight Devel- opment in Iraqi Refugees in Michigan. J Immigr Minor Health. 2015; 17: 1468–1475. https://doi.org/10. 1007/s10903-014-0127-3 PMID: 25376128 Jen K-LC, Jamil H, Zhou K, Breejen K, Arnetz BB. Sex Differences and Predictors of Changes in Body Weight and Noncommunicable Diseases in a Random, Newly-Arrived Group of Refugees Followed for Two Years. Journal of Immigrant and Minority Health. 2018. pp. 283–294. https://doi.org/10.1007/ s10903-017-0565-9 PMID: 28343246 49. Greene-Cramer B, Summers A, Lopes-Cardozo B, Husain F, Couture A, Bilukha O. Noncommunic- able disease burden among conflict-affected adults in Ukraine: A cross-sectional study of prevalence, risk factors, and effect of conflict on severity of disease and access to care. PLoS One. 2020; 15: e0231899. https://doi.org/10.1371/journal.pone.0231899 PMID: 32315357 50. Mansour Z, Said R, Dbaibo H, Mrad P, Torossian L, Rady A, et al. Non-communicable diseases in Lebanon: results from World Health Organization STEPS survey 2017. Public Health. 2020; 187: 120– 126. https://doi.org/10.1016/j.puhe.2020.08.014 PMID: 32949882 51. Ratnayake R, Rawashdeh F, AbuAlRub R, Al-Ali N, Fawad M, Bani Hani M, et al. Access to Care and Prevalence of Hypertension and Diabetes Among Syrian Refugees in Northern Jordan. JAMA Netw Open. 2020; 3: e2021678. https://doi.org/10.1001/jamanetworkopen.2020.21678 PMID: 33052405 52. Eryurt MA, Menet MG. Noncommunicable Diseases Among Syrian Refugees in Turkey: An Emerging Problem for a Vulnerable Group. J Immigr Minor Health. 2020; 22: 44–49. https://doi.org/10.1007/ s10903-019-00900-x PMID: 31123913 53. Maldari T, Elsley N, Rahim RA. The health status of newly arrived Syrian refugees at the Refugee Health Service, South Australia, 2016. Aust J Gen Pract. 2019; 48: 480–486. https://doi.org/10.31128/ AJGP-09-18-4696 PMID: 31256505 54. Al-Duais MA, Al-Awthan YS. Association between qat chewing and dyslipidaemia among young males. J Taibah Univ Med Sci. 2019; 14: 538–546. https://doi.org/10.1016/j.jtumed.2019.09.008 PMID: 31908642 55. Amr RA, Al-Smadi AM, Akasheh RT, Deiranieh RA, Gammoh OS, Hammouh FG, et al. Mood and demographical factors as predictors of body mass index among Syrian and Iraqi refugees in Jordan. J Health Psychol. 2021; 26: 1988–1998. https://doi.org/10.1177/1359105319893108 PMID: 31830806 56. Singh KP, Bhoopathy SV, Worth H, Seale H, Richmond RL. Nutrition among men and household food security in an internally displaced persons camp in Kenya. Public Health Nutr. 2016; 19: 723–731. https://doi.org/10.1017/S1368980015001275 PMID: 25917287 57. Balcılar M. Health Status Survey of Syrian Refugees in Turkey. Non—Communicable Disease Risk Factors Surveillance Among Syrian Refugees Living in Turkey. 2016. 58. Bayyari WD, Henry LJ, Jones C. Dieting behaviours, obesity and predictors of dieting among female college students at Palestinian universities. Eastern Mediterranean Health Journal. 2013. pp. 30–36. https://doi.org/10.26719/2013.19.1.30 PMID: 23520903 59. Bhatta MP, Shakya S, Assad L, Zullo MD. Chronic disease burden among Bhutanese refugee women aged 18–65 years resettled in Northeast Ohio, United States, 2008–2011. J Immigr Minor Health. 2015. Available: https://idp.springer.com/authorize/casa?redirect_uri=https://link.springer.com/article/ 10.1007/s10903-014-0040-9&casa_token=gWF3pGCTXO0AAAAA:7v6NxrAIQ- PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 26 / 29 PLOS ONEOverweight and obesity in humanitarian settings 6pWanhchraNkRTKN9Vs4OXxHosaawLXMGFq9l5nLBdXgPpk8uUrLamCNsFcr7LVxIHAhad_Q https://doi.org/10.1007/s10903-014-0040-9 PMID: 24849870 60. Bhatta MP, Assad L, Shakya S. Socio-demographic and dietary factors associated with excess body weight and abdominal obesity among resettled Bhutanese refugee women in Northeast Ohio, United States. Int J Environ Res Public Health. 2014; 11: 6639–6652. https://doi.org/10.3390/ ijerph110706639 PMID: 24968209 61. El Kishawi RR, Soo KL, Abed YA, Muda WAMW. Obesity and overweight: prevalence and associated socio demographic factors among mothers in three different areas in the Gaza Strip-Palestine: a cross-sectional study. BMC Obes. 2014; 1: 7. https://doi.org/10.1186/2052-9538-1-7 PMID: 26217499 62. Dhair A, Abed Y. The association of types, intensities and frequencies of physical activity with primary infertility among females in Gaza Strip, Palestine: A case-control study. PLoS One. 2020; 15: e0241043. https://doi.org/10.1371/journal.pone.0241043 PMID: 33095804 63. Kim YJ, Kim SG, Lee YH. Prevalence of General and Central Obesity and Associated Factors among North Korean Refugees in South Korea by Duration after Defection from North Korea: A Cross-Sec- tional Study. Int J Environ Res Public Health. 2018; 15. https://doi.org/10.3390/ijerph15040811 PMID: 29677154 64. Kim KJ, Kim YJ, Kim SH, An JH, Yoo HJ, Kim HY, et al. Vitamin D status and associated metabolic risk factors among North Korean refugees in South Korea: a cross-sectional study. BMJ Open. 2015; 5: e009140. https://doi.org/10.1136/bmjopen-2015-009140 PMID: 26621518 65. Kim YJ, Lee YH, Lee YJ, Kim KJ, An JH, Kim NH, et al. Prevalence of metabolic syndrome and its related factors among North Korean refugees in South Korea: a cross-sectional study. BMJ Open. 2016; 6: e010849. https://doi.org/10.1136/bmjopen-2015-010849 PMID: 27251685 66. Jeong H, Lee S-K, Kim S-G. Changes in body weight and food security of adult North Korean refugees living in South Korea. Nutr Res Pract. 2017; 11: 307–318. https://doi.org/10.4162/nrp.2017.11.4.307 PMID: 28765777 67. Abdollahi M, Abdollahi Z, Sheikholeslam R, Kalantari N, Kavehi Z, Neyestani TR. High occurrence of food insecurity among urban Afghan refugees in Pakdasht, Iran 2008: a cross-sectional study. Ecol Food Nutr. 2015; 54: 187–199. https://doi.org/10.1080/03670244.2013.834819 PMID: 25574705 68. Kumar GS, Varma S, Saenger MS, Burleson M, Kohrt BA, Cantey P. Noninfectious Disease Among the Bhutanese Refugee Population at a United States Urban Clinic. Journal of Immigrant and Minority Health. 2014. pp. 922–925. https://doi.org/10.1007/s10903-013-9800-1 PMID: 23456726 69. Taherifard E, Moradian MJ, Taherifard E, Hemmati A, Rastegarfar B, Molavi Vardanjani H. The preva- lence of risk factors associated with non-communicable diseases in Afghan refugees in southern Iran: a cross-sectional study. BMC Public Health. 2021; 21: 442. https://doi.org/10.1186/s12889-021- 10482-9 PMID: 33663464 70. Damiri B, Abualsoud MS, Samara AM, Salameh SK. Metabolic syndrome among overweight and obese adults in Palestinian refugee camps. Diabetol Metab Syndr. 2018; 10: 34. https://doi.org/10. 1186/s13098-018-0337-2 PMID: 29713387 71. Damiri B, Badran L, Safadi D, Sawalha A, Yasin Y, Sawalha M, et al. Metabolic syndrome and related risk factors among adults in the northern West Bank, a cross-sectional study. Int Health. 2019. https:// doi.org/10.1093/inthealth/ihz093 PMID: 31679020 72. Naigaga DA, Jahanlu D, Claudius HM, Gjerlaug AK, Barikmo I, Henjum S. Body size perceptions and preferences favor overweight in adult Saharawi refugees. Nutr J. 2018; 17: 17. https://doi.org/10.1186/ s12937-018-0330-5 PMID: 29426331 73. Kory R, Carney A, Naimer S. Health ramifications of the Gush Katif evacuation. Isr Med Assoc J. 2013; 15: 137–142. 74. Furusawa T, Furusawa H, Eddie R, Tuni M, Pitakaka F, Aswani S. Communicable and non-communi- cable diseases in the Solomon Islands villages during recovery from a massive earthquake in April 2007. N Z Med J. 2011; 124: 17–28. PMID: 21750591 75. Herrera-Fontana ME, Chisaguano AM, Villagomez V, Pozo L, Villar M, Castro N, et al. Food insecurity and malnutrition in vulnerable households with children under 5 years on the Ecuadorian coast: a post- earthquake analysis. Rural Remote Health. 2020; 20: 5237. https://doi.org/10.22605/RRH5237 PMID: 31937106 76. Adrega T, Ribeiro J, Santos L, Santos JA. Prevalence of cardiovascular disease risk factors, health behaviours and atrial fibrillation in a Nepalese post-seismic population: a cross-sectional screening during a humanitarian medical mission. Nepal Hear J. 2018; 15: 9–14. 77. Sakai A, Nakano H, Ohira T, Maeda M, Okazaki K, Takahashi A, et al. Relationship between the prev- alence of polycythemia and factors observed in the mental health and lifestyle survey after the Great East Japan Earthquake. Medicine. 2020; 99: e18486. https://doi.org/10.1097/MD.0000000000018486 PMID: 31895781 PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 27 / 29 PLOS ONEOverweight and obesity in humanitarian settings 78. Ohira T, Hosoya M, Yasumura S, Satoh H, Suzuki H, Sakai A, et al. Effect of Evacuation on Body Weight After the Great East Japan Earthquake. Am J Prev Med. 2016; 50: 553–560. https://doi.org/10. 1016/j.amepre.2015.10.008 PMID: 26673483 79. Ohira T, Nakano H, Nagai M, Yumiya Y, Zhang W, Uemura M, et al. Changes in Cardiovascular Risk Factors After the Great East Japan Earthquake. Asia Pac J Public Health. 2017; 29: 47S–55S. https:// doi.org/10.1177/1010539517695436 PMID: 28330394 80. Satoh H, Okazaki K, Ohira T, Sakai A, Hosoya M, Yasumura S, et al. Relationship between risk of hyper- low-density lipoprotein cholesterolemia and evacuation after the Great East Japan Earthquake. J Epidemiol. 2021. https://doi.org/10.2188/jea.JE20200267 PMID: 33441506 81. 82. 83. Takahashi S, Yonekura Y, Sasaki R, Yokoyama Y, Tanno K, Sakata K, et al. Weight Gain in Survivors Living in Temporary Housing in the Tsunami-Stricken Area during the Recovery Phase following the Great East Japan Earthquake and Tsunami. PLoS One. 2016; 11: e0166817. https://doi.org/10.1371/ journal.pone.0166817 PMID: 27907015 Takahashi S, Yonekura Y, Tanno K, Shimoda H, Sakata K, Ogawa A, et al. Increased incidence of metabolic syndrome among older survivors relocated to temporary housing after the 2011 Great East Japan earthquake & tsunami. Metabol Open. 2020; 7: 100042. Takahashi S, Yonekura Y, Tanno K, Shimoda H, Sakata K, Ogawa A, et al. Increase in Body Weight Following Residential Displacement: 5-year Follow-up After the 2011 Great East Japan Earthquake and Tsunami. J Epidemiol. 2021; 31: 328–334. https://doi.org/10.2188/jea.JE20190333 PMID: 32536638 84. Ebner DK, Ohsawa M, Igari K, Harada KH, Koizumi A. Lifestyle-related diseases following the evacua- tion after the Fukushima Daiichi nuclear power plant accident: a retrospective study of Kawauchi Vil- lage with long-term follow-up. BMJ Open. 2016; 6: e011641. https://doi.org/10.1136/bmjopen-2016- 011641 PMID: 27401362 85. Nakamura K, Watanabe Y, Kitamura K, Kabasawa K, Someya T. Psychological distress as a risk fac- tor for dementia after the 2004 Niigata-Chuetsu earthquake in Japan. J Affect Disord. 2019; 259: 121– 127. https://doi.org/10.1016/j.jad.2019.08.041 PMID: 31445337 86. Hikichi H, Aida J, Kondo K, Tsuboya T, Kawachi I. Residential relocation and obesity after a natural disaster: A natural experiment from the 2011 Japan Earthquake and Tsunami. Sci Rep. 2019; 9: 374. https://doi.org/10.1038/s41598-018-36906-y PMID: 30675013 87. Mulugeta W, Glick M, Min J, Xue H, Noe MF, Wang Y. Longitudinal Changes and High-Risk Sub- groups for Obesity and Overweight/Obesity Among Refugees in Buffalo, NY, 2004–2014. Journal of Racial and Ethnic Health Disparities. 2018; 5: 187–194. https://doi.org/10.1007/s40615-017-0356-y PMID: 28364373 88. Mulugeta W, Xue H, Glick M, Min J, Noe MF, Wang Y. Burden of Mental Illness and Non-communica- ble Diseases and Risk Factors for Mental Illness Among Refugees in Buffalo, NY, 2004–2014. J Racial Ethn Health Disparities. 2019; 6: 56–63. https://doi.org/10.1007/s40615-018-0498-6 PMID: 29785707 89. Mulugeta W, Xue H, Glick M, Min J, Noe MF, Wang Y. Disease Burdens and Risk Factors for Diabe- tes, Hypertension, and Hyperlipidemia among Refugees in Buffalo, New York, 2004–2014. J Health Care Poor Underserved. 2019; 30: 1119–1131. https://doi.org/10.1353/hpu.2019.0077 PMID: 31422992 90. Rhodes CM, Chang Y, Percac-Lima S. Development of Obesity and Related Diseases in African Refu- gees After Resettlement to United States. J Immigr Minor Health. 2016; 18: 1386–1391. https://doi. org/10.1007/s10903-015-0278-x PMID: 26392397 91. Bardenheier BH, Pavkov ME, Winston CA, Klosovsky A, Yen C, Benoit S, et al. Prevalence of Tuber- culosis Disease Among Adult US-Bound Refugees with Chronic Kidney Disease. J Immigr Minor Health. 2019; 21: 1275–1281. https://doi.org/10.1007/s10903-018-00852-8 PMID: 30612283 92. Bardenheier BH, Phares CR, Simpson D, Gregg E, Cho P, Benoit S, et al. Trends in Chronic Diseases Reported by Refugees Originating from Burma Resettling to the United States from Camps Versus Urban Areas During 2009–2016. J Immigr Minor Health. 2019; 21: 246–256. https://doi.org/10.1007/ s10903-018-0749-y PMID: 29761353 93. Nguyen M-LT, Rehkopf DH. Prevalence of Chronic Disease and Their Risk Factors Among Iranian, Ukrainian, Vietnamese Refugees in California, 2002–2011. J Immigr Minor Health. 2016; 18: 1274– 1283. https://doi.org/10.1007/s10903-015-0327-5 PMID: 26691740 94. Amstutz D, Gonc¸alves D, Hudelson P, Stringhini S, Durieux-Paillard S, Rolet S. Nutritional Status and Obstacles to Healthy Eating Among Refugees in Geneva. J Immigr Minor Health. 2020; 22: 1126– 1134. https://doi.org/10.1007/s10903-020-01085-4 PMID: 32940816 95. Modesti PA, Scali E, Marzotti I, Ulivi N, Boddi M, Galanti G, et al. Blood pressure and fasting glucose changes in male migrants waiting for an asylum decision in Italy. A pilot study. Int J Cardiol. 2020; 309: 110–114. https://doi.org/10.1016/j.ijcard.2019.12.026 PMID: 31884005 PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 28 / 29 PLOS ONEOverweight and obesity in humanitarian settings 96. Kortas AZ, Polenz J, von Hayek J, Ru¨ diger S, Rottbauer W, Storr U, et al. Screening for infectious dis- eases among asylum seekers newly arrived in Germany in 2015: a systematic single-centre analysis. Public Health. 2017; 153: 1–8. 97. Obesity and overweight. [cited 27 Jan 2022]. Available: https://www.who.int/news-room/fact-sheets/ detail/obesity-and-overweight 98. Ritchie H, Roser M. Obesity. Our World in Data. 2017. 99. Murphy M, Robertson W, Oyebode O. Erratum to: Obesity in International Migrant Populations. Curr Obes Rep. 2017; 6: 438. https://doi.org/10.1007/s13679-017-0279-2 PMID: 28861866 100. Goulão B, Santos O, Carmo I do. The impact of migration on body weight: a review. Cad Saude Publica. 2015; 31: 229–245. https://doi.org/10.1590/0102-311x00211913 PMID: 25760158 101. Oza-Frank R, Cunningham SA. The weight of US residence among immigrants: a systematic review. Obes Rev. 2010; 11: 271–280. https://doi.org/10.1111/j.1467-789X.2009.00610.x PMID: 19538440 102. Murphy M, Robertson W, Oyebode O. Obesity in International Migrant Populations. Curr Obes Rep. 2017; 6: 314–323. https://doi.org/10.1007/s13679-017-0274-7 PMID: 28755178 103. STEPwise approach to NCD risk factor surveillance (STEPS). [cited 31 Jan 2022]. Available: https:// www.who.int/teams/noncommunicable-diseases/surveillance/systems-tools/steps 104. Kehlenbrink S, Mahboob O, Al-Zubi S, Boulle P, Aebischer Perone S, Alani AH, et al. An inter-humani- tarian agency study of diabetes care and surveillance in humanitarian settings. The lancet. Diabetes & endocrinology. 2022. pp. 159–162. https://doi.org/10.1016/S2213-8587(22)00036-5 PMID: 35143782 105. Wilson A, Renzaho AMN, McCabe M, Swinburn B. Towards understanding the new food environment for refugees from the Horn of Africa in Australia. Health Place. 2010; 16: 969–976. https://doi.org/10. 1016/j.healthplace.2010.06.001 PMID: 20609612 106. Hacker K, Anies M, Folb BL, Zallman L. Barriers to health care for undocumented immigrants: a litera- ture review. Risk Manag Healthc Policy. 2015; 8: 175–183. https://doi.org/10.2147/RMHP.S70173 PMID: 26586971 107. Hadgkiss EJ, Renzaho AMN. The physical health status, service utilisation and barriers to accessing care for asylum seekers residing in the community: a systematic review of the literature. Aust Health Rev. 2014; 38: 142–159. https://doi.org/10.1071/AH13113 PMID: 24679338 108. Danielli S, Coffey T, Ashrafian H, Darzi A. Systematic review into city interventions to address obesity. EClinicalMedicine. 2021; 32: 100710. https://doi.org/10.1016/j.eclinm.2020.100710 PMID: 33681735 109. Foresight (Program). Tackling Obesities: Future Choices—Project Report. 2007. 110. Reilly JJ, El-Hamdouchi A, Diouf A, Monyeki A, Somda SA. Determining the worldwide prevalence of obesity. The Lancet. 2018. pp. 1773–1774. https://doi.org/10.1016/S0140-6736(18)30794-3 PMID: 29739565 111. Ross R, Neeland IJ, Yamashita S, Shai I, Seidell J, Magni P, et al. Waist circumference as a vital sign in clinical practice: a Consensus Statement from the IAS and ICCR Working Group on Visceral Obe- sity. Nat Rev Endocrinol. 2020; 16: 177–189. https://doi.org/10.1038/s41574-019-0310-7 PMID: 32020062 112. NCD Management-Screening, Diagnosis, Treatment. WHO package of essential noncommunicable (PEN) disease interventions for primary health care. World Health Organization; 7 Sep 2020 [cited 24 Feb 2022]. Available: https://www.who.int/publications/i/item/who-package-of-essential- noncommunicable-(pen)-disease-interventions-for-primary-health-care PLOS ONE \| https://doi.org/10.1371/journal.pone.0282823 April 24, 2023 29 / 29 PLOS ONE
10.1371_journal.pone.0264346	RESEARCH ARTICLE A direct collocation framework for optimal control simulation of pedaling using OpenSim Sangsoo ParkID 1,2, Graham E. Caldwell1, Brian R. Umberger3* 1 Department of Kinesiology, University of Massachusetts Amherst, Amherst, Massachusetts, United States of America, 2 College of Medicine, Korea University, Seoul, South Korea, 3 School of Kinesiology, University of Michigan, Ann Arbor, Michigan, United States of America * umberger@umich.edu Abstract The direct collocation (DC) method has shown low computational costs in solving optimiza- tion problems in human movements, but it has rarely been used for solving optimal control pedaling problems. Thus, the aim of this study was to develop a DC framework for optimal control simulation of human pedaling within the OpenSim modeling environment. A planar bicycle-rider model was developed in OpenSim. The DC method was formulated in MATLAB to solve an optimal control pedaling problem using a data tracking approach. Using the developed DC framework, the optimal control pedaling problem was successfully solved in 24 minutes to ten hours with different objective function weightings and number of nodes from two different initial conditions. The optimal solutions for equal objective function weightings were successful in terms of tracking, with the model simulated pedal angles and pedal forces within ±1 standard deviation of the experimental data. With these weightings, muscle tendon unit (MTU) excitation patterns generally matched with burst timings and shapes observed in the experimental EMG data. Tracking quality and MTU excitation pat- terns were changed little by selection of node density above 31, and the optimal solution quality was not affected by initial guess used. The proposed DC framework could easily be turned into a predictive simulation with other objective functions such as fastest pedaling rate. This flexible and computationally efficient framework should facilitate the use of optimal control methods to study the biomechanics, energetics, and control of human pedaling. Introduction Optimal control theory paired with musculoskeletal modeling and simulation is a powerful approach for studying the biomechanics, energetics, and control of human movement [1–4]. In particular, musculoskeletal simulation has been used to study the neural control of pedaling as a representative locomotor task that has fewer stability requirements than walking [5–7]. The simulation approach has also been used to improve our understanding of the biomechan- ics and energetics of pedaling with applications that range from rehabilitation to athletic per- formance [8–11]. However, there are considerable challenges in applying optimal control theory to investigate the mechanics, energetics and control of human movement, such as the a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Park S, Caldwell GE, Umberger BR (2022) A direct collocation framework for optimal control simulation of pedaling using OpenSim. PLoS ONE 17(2): e0264346. https://doi.org/10.1371/journal. pone.0264346 Editor: Katherine Saul, North Carolina State University, UNITED STATES Received: July 12, 2021 Accepted: February 8, 2022 Published: February 22, 2022 Copyright: © 2022 Park et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The model and sample codes are freely available at the GitHub repository (http://github.com/sangsoopark1739/ pedaling_DC_OpenSim). Funding: This research was supported by the University of Massachusetts Amherst (SP - Graduate School Fellowship (https://www. umass.edu/graduate/), and School of Public Health and Health Sciences Dean’s PhD Summer Fellowship (https://www.umass.edu/ sphhs/about-us); GEC - Faculty Research Grant (https://www.umass.edu/)). The funders had PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 1 / 17 PLOS ONEno role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Direct collocation framework for optimal control pedaling problems substantial computational cost, the complexity of developing musculoskeletal models, and until recently the lack of user-friendly simulation environments. The computational cost of generating optimal control simulations of human movement has until recently been a major limiting factor to its widespread adoption. This limitation can largely be overcome by using the direct collocation (DC) optimal control framework, which exploits sparsity in the problem formulation to substantially lower the computational cost of simulating human movement compared with traditional methods [12–16]. Despite this advan- tage, the DC method has rarely been used to solve optimal control pedaling problems. Although Kaplan and Heegaard had already demonstrated the potential of DC for generating rapid simulations of human pedaling two decades ago [15], their complex symbolic formula- tion and lack of available models or code has likely been a factor limiting the adoption of their approach. The obstacles to implementing computational models of the musculoskeletal system can be substantially reduced using software packages such as OpenSim, a freely available musculo- skeletal modeling software that has seen increased use within the biomechanical modeling community [17,18]. OpenSim provides a graphical user interface in which users can create a musculoskeletal model and perform a variety of analyses including static optimization and for- ward simulation, without having to derive the dynamical equations of motion in symbolic form. Further, the models and analyses can be utilized in other programming environments via a robust application programming interface (API), allowing customized optimization problem formulation. One candidate for interfacing with OpenSim libraries is MATLAB (The MathWorks, Inc.), a common scripting language used in science and engineering that pro- vides both native optimizers (e.g. fmincon) and an interface (i.e., MEX) to access external opti- mization solvers such as IPOPT [19]. The OpenSim-MATLAB interface has been used to successfully solve DC optimal control problems of vertical jumping [12], walking [14,20], and running [14], but to our knowledge has not be used for simulating pedaling. Thus, the aim of this study was to develop a DC framework for optimal control simulation of human pedaling within the OpenSim modeling environment. First, we developed a planar musculoskeletal model of the lower limbs and bicycle crank arm in OpenSim. Using the Open- Sim API, the DC method was formulated in MATLAB (The MathWorks, Inc.) to solve optimal control pedaling problems that track a set of experimental human pedaling data. The quality of the optimal solution was evaluated by comparing simulation results with the experimental data. To facilitate other scientists being able to leverage our work in their own research, the model and sample codes are freely available at the GitHub repository (http://github.com/ sangsoopark1739/pedaling_DC_OpenSim). Method Bicycle-rider model A two-dimensional (2-D) bicycle-rider model was developed in OpenSim [17] to solve optimal control pedaling problems using the DC method [13,21]. The model consists of eight rigid seg- ments and nine muscle tendon units (MTUs) per leg (Fig 1), based on a recent model devel- oped to simulate movements with large leg flexion angles such as occur in pedaling [22]. The rigid segments represent the bicycle crank, pelvis and paired thighs, lower legs, and feet. The hip, knee and ankle joints were modeled by hinge joints with coupled 2-D translation for the knee joint center of rotation [23]. The rotational axis of the crank was also modeled as a hinge joint. The pelvis segment center and rotational axis of the crank were fixed in space, represent- ing seated pedaling on a stationary ergometer. Stiff springs (‘PointToPointSpring’ function in OpenSim, spring constant, k = 100,000) were used to connect the foot segments to the crank PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 2 / 17 PLOS ONEDirect collocation framework for optimal control pedaling problems Fig 1. The bicycle-rider model in OpenSim. The developed 2-D bicycle-rider model consists of eight rigid segments (bicycle crank, pelvis, paired thighs, lower legs, and feet) and nine muscle tendon unit per leg (iliopsoas, gluteus maximus, vasti, rectus femoris, hamstrings, biceps femoris short head, gastrocnemius, soleus, and tibialis anterior). Pelvic tilt angle (θ1) was set to 11.5˚ and pedal angle (θ2) was determined by angular deviation between the horizontal and line connecting from the toe to the heel. https://doi.org/10.1371/journal.pone.0264346.g001 segment at a location representing the distal end of the metatarsals, creating a closed kinemat- ics chain with ‘soft’ constraints between the feet and crank segments [24]. In the model, pedal angles were defined as the angular deviation from the horizontal of a straight line connecting the toe to the heel (Fig 1). Positive pedal angles in the model represent the toe being below a PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 3 / 17 PLOS ONEDirect collocation framework for optimal control pedaling problems horizontal line representing a pedal angle of zero (0˚). The mean pelvis angle across all partici- pants was used in this model (11.5˚; Fig 1), though there was a difference between males (9.3 ± 6.8˚) and females (17.8 ± 4.9˚), on average. Because the model and code are made avail- able, interested readers could test the effects of these differences in pelvis angle for themselves. Details for the experimental data are described in a later section. Each leg was actuated by nine muscle tendon units (MTUs): iliopsoas (IP), gluteus maximus (GMAX), vasti (VAS), rectus femoris (RF), hamstrings (HAM), biceps femoris short head (BFsh), gastrocnemius (GAS), soleus (SOL), and tibialis anterior (TA). A three-component equilibrium muscle model was used to represent each MTU, due to its computational effi- ciency and ability to match experimental data [25]. The force-producing potential of each MTU was determined by the force-velocity-length (F-V-L) relation of the contractile compo- nent (CC), and separate force-extension (F-Ext) relationships of series (SEC) and parallel (PEC) elastic components. Muscle-specific model parameters were selected to match net joint moments [26] and passive joint moments [27] in human participants. More details about the model development can be found in the Supplemental material 1 (S1 File). The resistive torque applied to the bicycle crank varied inversely with crank speed at a con- stant power output. We estimated this resistive crank torque profile (Tresistive) using the experi- mental crank torque and acceleration patterns. Tcrank (cid:0) Tresistive ¼ Ieff � acrank ð1Þ where Tcrank, αcrank, and Ieff are mean crank torque, crank angular acceleration, and moment of inertia [28]. The effective crank moment of inertia (Ieff) was 23.53 kg�m2 based on a gear ratio of 42/28 [28]. The resistive crank torque was smoothed and interpolated to 21 equally- spaced data points (‘PrescribedForce’ function in OpenSim), which was applied about the rota- tional axis of the crank segment. Experimental data Experimental pedaling data for the tracking optimizations were drawn from a previous study [29], including left pedal angles and pedal forces, and the activation patterns of ten left leg muscles. Data were recorded from seven consecutive crank cycles while fifteen recreational cyclists (11 males, 4 females; age: 25.8 ± 4.5 years; height: 1.7 ± 0.1 m; mass: 67.2 ± 9.8kg; mean ± one standard deviation) rode a stationary bicycle at 30 rpm with a power output at ~30 W, representing a typical application of ergometer cycling in a rehabilitation setting [30]. Participants had signed a prior written consent form approved by the Institutional review board at the University of Massachusetts Amherst (Protocol #: 2017–4085). Muscle activations were determined by electromyography (EMG) from tensor fasciae latae (TFL), rectus femoris (RF), biceps femoris long head (BFL), semitendinosus (ST), vastus lateralis (VL), vastus media- lis (VM), lateral gastrocnemius (LGA), medial gastrocnemius (MGA), soleus (SOL), and tibia- lis anterior (TA). For each muscle EMG amplitude was normalized by the average of its peak EMG values across the seven crank cycles. Crank torque was computed by multiplying the crank arm length (0.17m) by the tangential pedal force over the crank cycle. The central differ- ence method was used to compute crank angular velocity and acceleration using the filtered crank angle data [31]. Anterior and upward pedal forces and counterclockwise crank and pedal angles were defined as positive values. The pedal angle and pedal force data for the right leg were generated by assuming symmetry and shifting the left leg data by 180˚ of the pedal cycle. Mean pedal angles and pedal forces averaged across participants were used as data track- ing targets for the pedaling simulations. Mean crank angular velocity from the participants was 188˚/s (31 rpm), matching closely with the target cadence of 30 rpm. Further details on PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 4 / 17 PLOS ONEDirect collocation framework for optimal control pedaling problems the collection and processing of the experimental data are described in an earlier publication [29]. To scale the model segments, lower body segment lengths were measured in a standing pos- ture for each participant and averaged to compute mean anthropometric data. To determine the initial model posture, the pelvic tilt angle was measured by markers placed on the pelvis and recorded via still photos with the participants seated on the bicycle in a pedaling posture with the left pedal at the top dead center position. The mean pelvic angle was tilted 11.5˚ ante- riorly, while the model pedal angle was adjusted by 12.5˚ to align it with the experimental data. Dynamic tracking optimization Tracking optimizations were performed to simulate a two-legged pedaling motion over one complete crank cycle that matched kinematic and kinetic data from human participants using the DC method [13,21]. The freely available optimization solver IPOPT [19] was used to solve the optimization problem in MATLAB using the MEX function interface [13]. The objective function included both tracking and activation terms (Eq 1), similar to that used to generate simulations of human walking [14,32]. The first term (tracking) represents how well the model reproduces experimental pedal angles and forces [33]. The second term (activation) is the sum of cubed MTU activation integrals, known to produce distributed mus- cle activations that minimize muscle fatigue [16,34–36]. The objective function, J, is given by �2 1 0 � Pv J ¼ W1 B @ i¼1 Yij(cid:0) ^Y ij SDij Pn j¼1 v � n C A þ W2 ! R t PnMus o a3 m¼1 nMus � n mðtÞ ð2Þ where Yij are experimental pedal angle, and horizontal and vertical pedal force components, Ŷij are the same variables from the model simulated pedaling motion, SDij are inter-subject standard deviations, v is the number of variables being compared (6 variables), am is the activa- tion of the mth muscle at a given time, nMus is the number of muscles (18 muscles), n is the number of data points in a complete crank cycle, W1 and W2 are weighting values for the tracking and activation terms, respectively. Letters i, j, and m are indices for v, n, and nMus respectively. A sensitivity analysis was performed to determine how the weighting between tracking and activation terms affected the quality of the optimal solution (see following section). The states in the model were hip, knee, and ankle joint angles and angular velocities, crank angle and angular velocity, muscle activations, and muscle CC lengths, leading to a total of 50 states ([7 angles × 2] + [18 muscles × 2]). The controls were the 18 MTU excitation signals. For the DC method, the time series of states and controls were discretized into equally-spaced nodes (n), leading to a total of 68 × n unknowns. The system dynamical equations were discretized into n equally-spaced nodes, converting the system equations into algebraic equality constraints using the backward Euler method. xiþ1 (cid:0) xi Dt (cid:0) f iþ1 ¼ 0; i ¼ 1; 2; . . . ; n (cid:0) 1 ð3Þ where Δt is the difference between ti+1 and ti and fi+1 represents the time derivatives of the states x(t) at time i+1 (50 states × (n-1)). The states and controls at the first node were assumed to be the same as the sates and controls at the final node, except for the crank angle, which was fixed at 0˚ for the first node and 360˚ for the final node. OpenSim does not provide the dynamic equations of motion in symbolic form, so the sparsity pattern of the constraints Jaco- bian matrix, which is required for the IPOPT optimization algorithm, was generated manually. PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 5 / 17 PLOS ONE Direct collocation framework for optimal control pedaling problems The OpenSim function ‘computeStateVariableDerivatives’ was used to calculate the state vari- able time derivatives at each node. The time to complete one crank cycle (tFinal) was set at 1.914s, the mean duration observed in the experiment data. Thus, the total number of equality constraints varied from 569 and 9069 based on the selected value used for n ranging from 11 to 181 (number of constraints = 50 × (n−1) + 68 + 1, n = 11, 31, 51, 91 or 181). The DC approach generally converges better if the initial guess is already dynamically con- sistent [13]. Initial guesses for the states and controls were produced by generating multiple forward simulations using the OpenSim forward dynamics tool, with manual muscle excita- tion patterns based on published pedaling simulation and EMG studies [5,36–38]. The states and controls from this forward simulation will be referred to as the dynamic initial guess. We also used an initial guess where the states and controls were the same for all nodes (i.e., no movement, constant activation), referred to as the static initial guess. All optimizations were performed on the same laptop computer with a 2.7 GHz Intel i7-6820HQ CPU and 16 GB of RAM, running OpenSim 3.3, MATLAB R2018a, and IPOPT release 3.11.0. Sensitivity analyses We performed three sensitivity analyses to determine how different weightings, node densities, and initial conditions affected the optimal solutions. First, to examine which combination of objective function weights W1 and W2 could predict physiologically realistic MTU excitation patterns with low tracking errors, six combinations of W1 and W2 were used to solve the opti- mization problem at 31 equally-spaced nodes with the dynamic initial guess. The six pairs included ‘tracking only’ (W1 = 1, W2 = 0) and ‘activation only’ (W1 = 0, W2 = 1) solutions. Between these two extremes, the tracking term weight was held at W1 = 1, with the activation term weight varied from 0.5 to 10 (W2 = 0.5, 1, 5, 10). Secondly, to examine the effect of node spacing, tracking optimizations with equal weightings (W1 = W2 = 1) were solved at node den- sities from 11 to 181 (n = 11, 31, 51, 91, and 181) with the dynamic initial guess. The relation- ship between the node number and CPU time was evaluated by the correlation coefficient (‘corrcoef’, MATLAB). The number of nodes represent the discretization of the states and con- trols in the solution process, corresponding to Δt ranging between 0.1914 s and 0.0106 s. Finally, the effect of the dynamic versus static initial guess was evaluated by solving the track- ing optimization using each initial guess with the equal weightings (W1 = W2 = 1) at 31 equally-spaced nodes. The quality of optimal solutions was evaluated by comparing the objective function tracking term values, with smaller values indicating better matching of pedal angles and pedal forces between the simulated and experimental data. Simulated MTU excitation pat- terns were qualitatively evaluated by visual comparison with burst timings and shapes in corresponding experimental EMG patterns from six muscles (RF, BFL, VL, LGA, SOL, and TA). Further, the tracking quality was considered to be acceptable when simulated pedal angles and pedal forces were fell within one standard deviation of the experimental data, which is a commonly used criterion in evaluating musculoskeletal simulation results [6,8]. Results Effects of different objective term weightings The first sensitivity analysis examined the effects of different objective function term weight- ings on the quality of the optimal solutions at 31 nodes. The tracking term represents the abil- ity of the simulated pedal angles and pedal forces to match the corresponding experimental data, while the activation term is the sum of cubed MTU activation integrals. There were PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 6 / 17 PLOS ONEDirect collocation framework for optimal control pedaling problems Fig 2. The effects of changing relative weightings in the objective function tracking and activation terms. Objective function values for the optimizations with six combinations of tracking and activation term weightings. ‘TrackOnly’ weighting was 1 for tracking and 0 for activation (W1 = 1, W2 = 0); ‘ActOnly’ weighting was 0 for tracking and 1 for activation (W1 = 0, W2 = 1). In four other simulations the activation term weight was .5, 1 (‘Equal’), 5 or 10 while the tracking weight was fixed at 1. Smaller tracking term values indicate better match between the simulated and experimental data. Note that high tracking quality could be maintained with low muscle activation when the activation term weight was 0.5 or 1. https://doi.org/10.1371/journal.pone.0264346.g002 trade-offs associated with weighing one term much more heavily than the other (Fig 2). With tracking and activation weights set at 0 and 1 respectively (W1 = 0, W2 = 1, ActOnly), the objective function value due to activation was small (0.0001) but the value due to tracking had its highest value (5.3080), representing the greatest tracking error. In the opposite case (W1 = 1, W2 = 0, TrackOnly), the objective function value due to tracking was greatly reduced to 0.0105 representing excellent fit with the experimental data, but the activation term was increased to its largest value (0.0089). The optimal solutions for activation term weights of 0.5 and 1 were successful in terms of tracking, with the model simulated pedal angles and pedal forces tracking the experi- mental data closely, well within ±1 standard deviation (Fig 3A–3C). MTU excitation pat- terns were also well matched with experimental EMG data with these weightings (Fig 3F– 3H and 3J–3L), although the onset timing of HAM was slightly earlier than observed in the BFL EMG pattern. As expected the TrackOnly weightings also resulted in model pedal angles and pedal forces that fell within ±1 standard deviation (Fig 3A–3C). However, unrealistically high model MTU activations over much of the crank cycle were found (Fig 3D–3L), with the gastrocnemius and tibialis anterior MTU excitation patterns not phasic as observed in the EMG data (Fig 3J and 3L). On the contrary, for the ActOnly condition the MTU activation timing was generally matched with burst timings and shapes observed in the experimental EMG data; however, the model pedal angle and pedal force data devi- ated substantially from the experimental (Fig 3A–3C). When the activation term weight- ing was increased to 5 or 10, the model anterior-posterior pedal force became less accurate during the downstroke (around 90˚, Fig 3B), perhaps associated with less HAM activity compared to the simulations with term weights of 0.5 or 1. Thus, either the half (0.5) or equal (1) weighting for the activation term could attain both high tracking quality and physiologically realistic muscle activation patterns. PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 7 / 17 PLOS ONEDirect collocation framework for optimal control pedaling problems Fig 3. Simulated pedal angles, pedal forces, and MTU excitation patterns using different objective function term weightings. Simulated patterns of pedal angle (A), pedal forces (B and C), and MTU excitations (D-L) from simulations using the six combinations of objective term weightings compared with experimental data (dark gray solid line, with gray shaded area representing ±1 SD). Because peak timings throughout the seven crank cycles were varied, the maximum values in the averaged experimental EMG patterns was below one. The tracking term weight was held at W1 = 1, while the activation term weight (W2) was set to 0 (TrackOnly, black solid line), 0.5 (red solid line), 1 (blue dotted line), 5 (green solid line), and 10 (magenta dotted line). In ActOnly, the tracking term weight was W1 = 0 and the activation term weight was W2 = 1 (cyan solid line). Experimental EMG data for the gluteus maximus, iliopsoas, and biceps femoris short head were not available. Simulated data from the left leg are shown, those from the contralateral leg are almost identical. Either 0.5 or 1 weighting for the tracking term produced high tracking quality with physiologically realistic muscle activation patterns. https://doi.org/10.1371/journal.pone.0264346.g003 Effects of the number of nodes in the discretization When solving the tracking optimization with both objective function term weights set to 1, the computational cost increased almost linearly as the number of nodes increased from 11 to 181 (r = 0.998, p < 0.001). Although the optimization took only 24 minutes to solve with 11 nodes PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 8 / 17 PLOS ONEDirect collocation framework for optimal control pedaling problems Fig 4. Effects of changing the number of nodes in the discretization on the CPU time and objective function value. (A) The CPU time increased almost linearly as a function of the number of nodes. (B) The total objective function value (black dashed line) decreased 12.7% from 11 to 31 nodes, with little further change above 31 nodes. The slight changes in the total objective function value were reflected in both the activation term (dark gray solid line) and the tracking term (light gray solid line). Note that the tracking term can in principle go to zero (i.e. perfect tracking), while the activation term cannot. https://doi.org/10.1371/journal.pone.0264346.g004 (Fig 4A), the burst timings and shapes in MTU excitation patterns deviated from those found in the optimal solutions with the other higher node numbers due to insufficient node density (Fig 5D–5L). When the node number was changed from 11 to 31, the objective function value dropped significantly (12.7%), resulting in nearly identical simulated model pedal angles, pedal forces, and MTU excitation patterns to those found in the optimal solutions with the higher node numbers (Fig 5). The dynamic optimization took 1 hour (60.2 minutes) to solve with 31 nodes, increasing to 10.2 hours with 181 nodes (Fig 4A). However, the increased cost did not result in a substantially better solution, as the objective function decreased by only a small amount with the higher node density (Fig 4B). The total objective function did decrease slightly (1.9%) from 31 to 51 nodes, due to decreases in both tracking and activation term val- ues. However, the use of node densities above 31 had almost no effect on the simulated model pedal angles, pedal forces or MTU excitation patterns (Fig 5). Effect of different initial guesses Simulated pedal angles, pedal forces, and MTU excitation patterns estimated with the static initial guess were almost identical to those predicted with the dynamic initial guess (Fig 6). In the final objective function values, the activation term value was 0.0003 for both guesses, and the tracking term value was 0.0156 for the dynamic initial guess and 0.0158 for the static initial guess. However, the optimization started from the static initial guess took ~1.6 times longer to converge (97.3 minutes versus 60.2 minutes for the dynamic guess). Discussion We have presented an approach for generating optimal control simulations of seated pedaling using the DC method within the OpenSim and MATLAB environments. The optimal control pedaling problem was solved with different weights on the terms in the objective function, dif- ferent numbers of temporal nodes, and two different types of initial guesses to test how those factors affected tracking quality and predicted MTU excitation patterns. The relative weighting of the objective function terms had a substantial impact on the solution quality and must be PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 9 / 17 PLOS ONEDirect collocation framework for optimal control pedaling problems Fig 5. Simulated pedal angles, pedal forces, and MTU excitation patterns at each node. Simulated pedal angles (A), pedal forces (B-C), and MTU excitation patterns (D-L) obtained for simulations using 11 (black solid line), 31 (red solid line), 51 (green dotted line), 91 (blue solid line), and 181 (magenta dotted line) nodes compared with experimental pedal angles and forces (dark gray dotted line, with gray shaded area representing ±1 SD). The optimal solution quality was similar for temporal node densities of 31 or greater. https://doi.org/10.1371/journal.pone.0264346.g005 selected judiciously. On the other hand, the data tracking quality and MTU excitation patterns were relatively insensitive to the temporal node densities over the range considered despite wide variation in computational cost. The quality of the optimal solution was also not affected by the type of initial guess, but the computational cost was greater with the static initial guess compared with the dynamically consistent guess. It is common to use hybrid objective functions that combine data tracking and muscular effort terms, and in previous DC studies that approach has resulted in realistic kinematics, PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 10 / 17 PLOS ONEDirect collocation framework for optimal control pedaling problems Fig 6. Simulated pedal angles, pedal forces, and MTU excitation patterns using two different initial guesses. Simulated pedal angles (A), pedal forces (B-C), and MTU excitation patterns (D-L) found with dynamic (black solid line) and static (light gray dotted line) initial guesses. Initial states and controls were generated from OpenSim forward simulations with manually input MTU excitation patterns for the dynamic initial guess, but were the same for all nodes (i.e. no movement, constant activation) for the static initial guess. The simulated results for the two initial guesses were almost identical, but the optimized solution started from the dynamic initial guess was obtained in less time. https://doi.org/10.1371/journal.pone.0264346.g006 kinetics, and MTU excitation patterns [14,15,32]. However, the weighting between the track- ing and effort terms should be selected carefully. We identified a range of weights (Fig 2) that led to similarly good simulation results (Fig 3). However, outside of that range there were either discrepancies in the kinematic and kinetic data, or excessive and sometimes non-phasic MTU excitations, either of which would negatively impact interpretations based on the simula- tion results. The relative weights for the tracking and effort terms in prior studies were differ- ent from those found here [14,15,32], and likely cannot be generalized across studies. The PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 11 / 17 PLOS ONEDirect collocation framework for optimal control pedaling problems ideal weights will depend on several factors such as model, the task being simulated, and the specific quantities being tracked. Thus, the fact that equal weighting of the tracking and effort terms was found to yield good results is specific to the problem considered here and should be closely evaluated if the model is used for simulating other pedaling tasks or if different formu- lations of the objective terms are used. The current runtimes of 24 minutes to 10 hours on a commodity laptop computer across 11 to 181 temporal nodes were longer than the 20 minutes to 3 hours across 21 to 240 nodes reported in an earlier study where a similar pedaling optimization problem was solved [15]. In the current study, the optimal solution quality began to degrade on coarse grids such as 11 nodes (Δt = 0.191 s), which is consistent with the previous finding of degraded solution quality below 50 nodes in simulating half gait cycle (Δt = 0.013 s) [39]. While it is difficult to compare across modeling studies due to the varied time resolutions originating from the interaction between simulated time and node density [12,14,15], the greater computational costs in our study are likely due to overhead associated with frequent calls from MATLAB to external OpenSim functions (e.g., calculating state derivatives, spring forces), and the use of finite dif- ferences for derivatives. In the earlier DC pedaling study [15], analytical derivatives were used in solving the optimal control problem, which can provide a substantial computational perfor- mance advantage. Analytical derivatives are not currently possible within OpenSim, though the potential advantage of using algorithmic differentiation with OpenSim has recently been demonstrated [40]. While there is some overhead associated with using finite differences, OpenSim provides an easy to use environment for musculoskeletal modeling with a robust API that facilitates implementation of the DC method [13]. Moreover, the quality of solutions obtained with the OpenSim-MATLAB interface on relatively coarse temporal grids were simi- lar to the results on finer grids, and can be obtained in much less time than is required for the more common direct shooting methods that have been used for simulating pedaling [6,15,41]. The results obtained with the static and dynamic initial guesses were identical; however, it took ~1.6 times longer to solve the problem using the static initial guess compared with the dynamic initial guess, which is consistent with previous findings. In a prior study using the same algorithm, optimizations beginning from a static initial guess, as defined in the current study, took as much as 4 times longer compared with optimization solved as part of a grid refinement approach, where the initial guess is the optimal solution to the same problem obtained on a coarser temporal grid [13]. Likewise, a jumping optimization problem started from an initial guess produced by the computed muscle control algorithm solved approxi- mately 3 times faster than did same problem starting from an initial guess based on non-opti- mal bang-bang controls [12]. Better convergence starting from an initial guess closer to the optimal solution is not surprising, but another important consideration with the DC method is whether the initial guess is dynamically consistent. Using an initial guess that already satis- fies the system dynamical equations, even if it is far from the optimal solution, is not required for the DC method, but can lead to faster convergence. We initially modeled the interaction between the rider and bicycle as a closed kinematic loop, as commonly reported in past research (e.g., 6,41) using the “PointConstraint” feature in OpenSim. While kinematic constraints can be included in OpenSim models, in our pilot work we found that it was not possible to use models with kinematic constraints with our DC method implemented from within MATLAB. Therefore, we modeled the foot-pedal interac- tion with a stiff (100,000 N/m) spring using the “PointToPointSpring” force element in Open- Sim. This approach considers that the interaction between the foot and the pedal is not perfectly rigid and provides an easy means to extract pedal forces during simulations. The pedal forces obtained with this approach were in good agreement with experimental data (Fig 5). While the spring-based foot-pedal constraint worked well in practice, the newly released PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 12 / 17 PLOS ONEDirect collocation framework for optimal control pedaling problems OpenSim Moco optimal control framework can accommodate kinematic constraints and thus would allow the same pedaling problems to be solved using a closed kinematic loop if desired [42]. Aside from comparing simulated muscle activations with EMG data there are few opportu- nities to assess the other muscle-level variables in the model. The MTU CC lengths can be compared with ultrasound-based fascicle length measurements, though few such data are available for pedaling. The fascicle length of vastus lateralis was reported in an experimental study to change from 0.091 to 0.127 m in pedaling as the knee joint flexed [43], which was sim- ilar to the VAS MTU CC length change in our model (0.079 to 0.123 m). Gastrocnemius mus- cle fascicle length was found to change between 0.040 to 0.055 m in walking [44] while we found GAS MTU CC length to change from 0.028 to 0.0423 m in the model. These shorter lengths for the gastrocnemius CC make sense due to the much more flexed knee joint in pedal- ing than in walking [22,45,46]. More broadly, MTU CC length trajectories over the crank cycle across all muscles also remained within a physiologically reasonable range (S1 Fig). Specifi- cally, fascicle length changes of all the MTUs were between the lower and upper bounds of the force-length curves. Together, these results suggest that the observed MTU CC lengths during simulated pedaling were physiologically reasonable. Future studies using ultrasound measure- ment across a greater range of muscles during pedaling can aid to verify the interpretation [43,44]. Study limitations The computational performance results presented here are specific to the particular model used and pedaling problem that was solved and may not generalize to other scenarios. The bicycle-rider model developed in this study had eighteen MTUs with a constrained seated pos- ture and motion only permitted in the sagittal plane, combined with models of similar com- plexity having been used to successfully simulate pedaling in earlier studies (e.g., 5,6,8). The trade-off between the temporal discretization and computational cost may be different than those reported here if the model or task are changed, such as simulating non-seated cycling or including 3-D pelvis motion. Bilateral symmetry for the experimental pedal angle and pedal forces with a 180˚ phase off- set was assumed, which has been a common assumption in earlier pedaling simulation studies [6,8]. Pedaling in able-bodied people is generally symmetrical in leg muscle activity [47] though small asymmetries have been documented in specific variables [48–50], and larger asymmetries can be expected in certain clinical populations who are affected unilaterally, such as in stroke [51] or unilateral amputation [52]. While bilateral symmetry was assumed in this study it is not a requirement of the approach described here, which could also be used to study asymmetrical pedaling in future studies. The generation of the sparsity pattern for the constraints Jacobian matrix is challenging and may be prone to errors. We produced the constraints Jacobian matrix manually because the system equations of motion were not available in a symbolic form from OpenSim. Failing to identify all of the non-zero elements in the Jacobian matrix will prevent finding the optimal result, while using a dense Jacobian dramatically increases computational costs [13]. We have provided the constraints Jacobian sparsity patterns for the cases included in this study, but applying to approach to new problems will require generating the Jacobian patterns from scratch. A useful hybrid approach is adopted in the new OpenSim Moco software to automati- cally generate the sparsity pattern, at the expense of treating some sparse blocks as dense and therefore not achieving the maximal possible computational performance. PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 13 / 17 PLOS ONEDirect collocation framework for optimal control pedaling problems Summary The optimal control pedaling problem was successfully solved using the DC approach via an OpenSim-MATLAB interface. This flexible and computationally efficient framework should facilitate the use of optimal control methods to study the biomechanics, energetics, and control of human pedaling. Interested readers may freely access the model and example code to use and modify for their own research at the GitHub repository (http://github.com/ sangsoopark1739/pedaling_DC_OpenSim). Supporting information S1 Fig. Contractile component (CC) length changes over the crank cycle. Those CC length changes originated from the optimal solution with 31 nodes and equal weightings. All MTU CC length changes were fell between the lower and upper bound of their force-length curve, suggesting that the CC length changes estimated from the optimization are physiologically rel- evant. (TIF) S1 File. General overview: Development of a bicycle-rider model. This overview is about the description for the model development process. (DOCX) Author Contributions Conceptualization: Sangsoo Park, Graham E. Caldwell, Brian R. Umberger. Formal analysis: Sangsoo Park, Brian R. Umberger. Investigation: Sangsoo Park, Graham E. Caldwell, Brian R. Umberger. Methodology: Sangsoo Park, Graham E. Caldwell, Brian R. Umberger. Project administration: Graham E. Caldwell. Resources: Graham E. Caldwell. Software: Brian R. Umberger. Supervision: Graham E. Caldwell, Brian R. Umberger. Visualization: Sangsoo Park. Writing – original draft: Sangsoo Park. Writing – review & editing: Graham E. Caldwell, Brian R. Umberger. References 1. Pandy MG. Computer modeling and simulation of human movement. Annu Rev Biomed Eng. 2001; 3 (1):245–73. https://doi.org/10.1146/annurev.bioeng.3.1.245 PMID: 11447064 2. Umberger BR, Caldwell GE. Musculoskeletal Modealing. In: Robertson DGE, Caldwell GE, Hamill J, Kamen G& Whittlesey SN(Eds.), Research methods in biomechanics. 2nd ed. Champaign, IL: Human Kinetics; 2014. 247–276 p. 3. Umberger BR, Miller RH. Optimal Control Modeling of Human Movement. In: Mu¨ ller B. et al. (eds) Hand- book of Human Motion. Berlin: Springer; 2017. 1–22 p. 4. Valero-Cuevas FJ, Hoffmann H, Kurse MU, Kutch JJ, Theodorou EA. Computational models for neuro- muscular function. IEEE Rev Biomed Eng. 2009; 2:110–35. https://doi.org/10.1109/RBME.2009. 2034981 PMID: 21687779 PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 14 / 17 PLOS ONEDirect collocation framework for optimal control pedaling problems 5. Raasch CC, Zajac FE, Ma B, Levine WS. Muscle coordination of maximum-speed pedaling. J Biomech. 1997; 30(96):595–602. https://doi.org/10.1016/s0021-9290(96)00188-1 PMID: 9165393 6. Neptune RR, Hull ML. Evaluation of performance criteria for simulation of submaximal steady-state cycling using a forward dynamic model. J Biomech Eng. 1998; 120(3):334–41. https://doi.org/10.1115/ 1.2797999 PMID: 10412400 7. Ting LH, Raasch CC, Brown D a, Kautz SA, Zajac FE. Sensorimotor state of the contralateral leg affects ipsilateral muscle coordination of pedaling. J Neurophysiol. 1998; 80(3):1341–51. https://doi.org/10. 1152/jn.1998.80.3.1341 PMID: 9744943 8. Gidley AD, Marsh AP, Umberger BR. Performance criteria for generating predictive optimal control sim- ulations of bicycle pedaling. Comput Methods Biomech Biomed Engin. 2019; 22(1):11–20. https://doi. org/10.1080/10255842.2018.1522535 PMID: 30398070 9. Schutte LM, Rodgers MM, Zajac FE, Glaser RM. Improving the Efficacy of Electrical Stimulation- Induced Leg Cycle Ergometry: An Analysis Based on a Dynamic Musculoskeletal Model. IEEE Trans Rehabil Eng. 1993; 1(2):109–25. 10. 11. van Soest AJ, Casius LJ. Which factors determine the optimal pedaling rate in sprint cycling? Med Sci Sports Exerc. 2000; 32(11):1927–34. https://doi.org/10.1097/00005768-200011000-00017 PMID: 11079524 Zajac FE, Neptune RR, Kautz S a. Biomechanics and muscle coordination of human walking: Part I: Introduction to concepts, power transfer, dynamics and simulations. Gait Posture. 2002; 16:215–32. https://doi.org/10.1016/s0966-6362(02)00068-1 PMID: 12443946 12. Porsa S, Lin Y-C, Pandy MG. Direct Methods for Predicting Movement Biomechanics Based Upon Opti- mal Control Theory with Implementation in OpenSim. Ann Biomed Eng. 2015;. https://doi.org/10.1007/ s10439-015-1538-6 PMID: 26715209 13. 14. Lee L-F, Umberger BR. Generating optimal control simulations of musculoskeletal movement using OpenSim and MATLAB. PeerJ. 2016; 4:e1638. https://doi.org/10.7717/peerj.1638 PMID: 26835184 Lin Y-C, Pandy MG. Three-dimensional data-tracking dynamic optimization simulations of human loco- motion generated by direct collocation. J Biomech. 2017; 59:1–8. https://doi.org/10.1016/j.jbiomech. 2017.04.038 PMID: 28583674 15. Kaplan ML, Heegaard JH. Predictive algorithms for neuromuscular control of human locomotion. J Bio- mech. 2001; 34(8):1077–83. https://doi.org/10.1016/s0021-9290(01)00057-4 PMID: 11448699 16. Ackermann M, van den Bogert AJ. Optimality principles for model-based prediction of human gait. J Bio- mech. 2010; 43(6):1055–60. https://doi.org/10.1016/j.jbiomech.2009.12.012 PMID: 20074736 17. Delp SL, Anderson FC, Arnold AS, Loan P, Habib A, John CT, et al. OpenSim: open-source software to create and analyze dynamic simulations of movement. IEEE Trans Biomed Eng. 2007; 54(11):1940– 50. https://doi.org/10.1109/TBME.2007.901024 PMID: 18018689 18. Seth A, Hicks JL, Uchida TK, Habib A, Dembia CL, Dunne JJ, et al. OpenSim: Simulating musculoskele- tal dynamics and neuromuscular control to study human and animal movement. PLoS Comput Biol. 2018; 14(7). https://doi.org/10.1371/journal.pcbi.1006223 PMID: 30048444 19. Wa¨ chter A, Biegler LT. On the implementation of an interior-point filter line-search algorithm for large- scale nonlinear programming. Math Program. 2006; 106(1):25–57. 20. Nguyen VQ, Johnson RT, Sup FC, Umberger BR. Bilevel optimization for cost function determination in dynamic simulation of human gait. IEEE Trans Neural Syst Rehabil Eng. 2019; 27(7):1426–35. https:// doi.org/10.1109/TNSRE.2019.2922942 PMID: 31199264 21. Betts JT. Practical methods for optimal control and estimation using nonlinear programming. SIAM; 2010. 22. Lai AKM, Arnold AS, Wakeling JM. Why are Antagonist Muscles Co-activated in My Simulation? A Mus- culoskeletal Model for Analysing Human Locomotor Tasks. Ann Biomed Eng. 2017; 45(12):2762–74. https://doi.org/10.1007/s10439-017-1920-7 PMID: 28900782 23. Yamaguchi GT, Zajac FE. A planar model of the knee joint to characterize the knee extensor mecha- nism. J Biomech. 1989; 22(1):1–10. https://doi.org/10.1016/0021-9290(89)90179-6 PMID: 2914967 24. Yamaguchi GT. Dynamic modeling of musculoskeletal motion: a vectorized approach for biomechanical analysis in three dimensions. Springer Science & Business Media; 2001. 25. Millard M, Uchida T, Seth A, Delp SL. Flexing computational muscle: modeling and simulation of muscu- lotendon dynamics. J Biomech Eng. 2013; 135(2):21005. https://doi.org/10.1115/1.4023390 PMID: 23445050 26. Anderson DE, Madigan ML, Nussbaum MA. Maximum voluntary joint torque as a function of joint angle and angular velocity: model development and application to the lower limb. J Biomech. 2007; 40 (14):3105–13. https://doi.org/10.1016/j.jbiomech.2007.03.022 PMID: 17485097 PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 15 / 17 PLOS ONEDirect collocation framework for optimal control pedaling problems 27. Riener R, Edrich T. Identification of passive elastic joint moments in the lower extremities. J Biomech. 1999; 32(5):539–44. https://doi.org/10.1016/s0021-9290(99)00009-3 PMID: 10327008 28. Fregly BJ, Zajac FE, Dairaghi CA. Bicycle drive system dynamics: theory and experimental validation. J Biomech Eng. 2000; 122(4):446–52. https://doi.org/10.1115/1.1286678 PMID: 11036570 29. Park S, Caldwell GE. Muscular activity patterns in 1-legged vs. 2-legged pedaling. J Sport Heal Sci [Internet]. 2021; 10(1):99–106. Available from: https://www.sciencedirect.com/science/article/pii/ S2095254620300107. https://doi.org/10.1016/j.jshs.2020.01.003 PMID: 33518019 30. Ambrosini E, De Marchis C, Pedrocchi A, Ferrigno G, Monticone M, Schmid M, et al. Neuro-mechanics of recumbent leg cycling in post-acute stroke patients. Ann Biomed Eng. 2016; 44(11):3238–51. https:// doi.org/10.1007/s10439-016-1660-0 PMID: 27251336 31. Winter DA. Biomechanics and Motor Control of Human Movement. 4th ed. Hoboken, New Jersey: John Wiley & Sons; 2009. 32. van den Bogert AJ, Blana D, Heinrich D. Implicit methods for efficient musculoskeletal simulation and optimal control. Procedia IUTAM. 2011; 2:297–316. https://doi.org/10.1016/j.piutam.2011.04.027 PMID: 22102983 33. Neptune RR, Kautz SA, Zajac FE. Muscle contributions to specific biomechanical functions do not change in forward versus backward pedaling. J Biomech. 2000; 33(2):155–64. https://doi.org/10.1016/ s0021-9290(99)00150-5 PMID: 10653028 34. Anderson FC, Pandy MG. Static and dynamic optimization solutions for gait are practically equivalent. J Biomech. 2001; 34(2):153–61. https://doi.org/10.1016/s0021-9290(00)00155-x PMID: 11165278 35. Crowninshield RD, Brand RA. A physiologically based criterion of muscle force prediction in locomotion. J Biomech. 1981; 14(11):793–801. https://doi.org/10.1016/0021-9290(81)90035-x PMID: 7334039 36. Thelen DG, Anderson FC, Delp SL. Generating dynamic simulations of movement using computed muscle control. J Biomech. 2003; 36:321–8. https://doi.org/10.1016/s0021-9290(02)00432-3 PMID: 12594980 37. Hug F, Boumier F, Dorel S. Altered muscle coordination when pedaling with independent cranks. Front Physiol. 2013; 4:232. https://doi.org/10.3389/fphys.2013.00232 PMID: 24009587 38. Neptune RR, Herzog W. Adaptation of muscle coordination to altered task mechanics during steady- state cycling. J Biomech. 2000; 33(2):165–72. https://doi.org/10.1016/s0021-9290(99)00149-9 PMID: 10653029 39. 40. van den Bogert AJ, Hupperets M, Schlarb H, Krabbe B. Predictive musculoskeletal simulation using optimal control: effects of added limb mass on energy cost and kinematics of walking and running. Spec Issue Artic Proc IMechE Part P J Sport Eng Technol. 2012; 226(2):123–33. Falisse A, Serrancolı´ G, Dembia CL, Gillis J, De Groote F. Algorithmic differentiation improves the computational efficiency of OpenSim-based trajectory optimization of human movement. PLoS One. 2019; 14(10):e0217730. https://doi.org/10.1371/journal.pone.0217730 PMID: 31622352 41. Umberger BR, Gerritsen KGM, Martin PE. Muscle fiber type effects on energetically optimal cadences in cycling. J Biomech. 2006; 39(8):1472–9. https://doi.org/10.1016/j.jbiomech.2005.03.025 PMID: 15923008 42. Dembia CL, Bianco NA, Falisse A, Hicks JL, Delp SL. OpenSim Moco: musculoskeletal optimal control. PLOS Comput Biol. 2020; 16(12):e1008493. https://doi.org/10.1371/journal.pcbi.1008493 PMID: 33370252 43. Muraoka T, Kawakami Y, Tachi M, Fukunaga T. Muscle fiber and tendon length changes in the human vastus lateralis during slow pedaling. J Appl Physiol. 2001; 91(5):2035–40. https://doi.org/10.1152/ jappl.2001.91.5.2035 PMID: 11641341 44. Lichtwark GA, Bougoulias K, Wilson AM. Muscle fascicle and series elastic element length changes along the length of the human gastrocnemius during walking and running. J Biomech. 2007; 40(1):157– 64. https://doi.org/10.1016/j.jbiomech.2005.10.035 PMID: 16364330 45. Chapman A, Vicenzino B, Blanch P, Hodges P. Do differences in muscle recruitment between novice and elite cyclists reflect different movement patterns or less skilled muscle recruitment? J Sci Med Sport. 2009; 12(1):31–4. https://doi.org/10.1016/j.jsams.2007.08.012 PMID: 18077215 46. Winter DA. Kinematic and kinetic patterns in human gait: variability and compensating effects. Hum Mov Sci. 1984; 3(1–2):51–76. 47. Carpes FP, Diefenthaeler F, Bini RR, Stefanyshyn DJ, Faria IE, Mota CB. Influence of leg preference on bilateral muscle activation during cycling. J Sports Sci. 2011; 29(2):151–9. https://doi.org/10.1080/ 02640414.2010.526625 PMID: 21120741 48. Carpes FP. Bilateral pedaling asymmetry during a simulated 40. . . 2007;. PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 16 / 17 PLOS ONEDirect collocation framework for optimal control pedaling problems 49. Daly DJ, Cavanagh PR. Asymmetry in bicycle ergometer pedalling. Med Sci Sports. 1976; 8(3):204–8. https://doi.org/10.1249/00005768-197600830-00013 PMID: 979569 50. Hunt MA, Sanderson DJ, Moffet H, Inglis JT. Interlimb asymmetry in persons with and without an ante- rior cruciate ligament deficiency during stationary cycling. Arch Phys Med Rehabil. 2004; 85(9):1475–8. https://doi.org/10.1016/j.apmr.2003.10.017 PMID: 15375819 51. Chen H-Y, Chen S-C, Chen J-JJ, Fu L-L, Wang YL. Kinesiological and kinematical analysis for stroke subjects with asymmetrical cycling movement patterns. J Electromyogr Kinesiol. 2005; 15(6):587–95. https://doi.org/10.1016/j.jelekin.2005.06.001 PMID: 16051498 52. Childers WL, Kistenberg RS, Gregor RJ. Pedaling asymmetries in cyclists with unilateral transtibial amputation: Effect of prosthetic foot stiffness. J Appl Biomech. 2011; 27(4):314–21. https://doi.org/10. 1123/jab.27.4.314 PMID: 21896953 PLOS ONE \| https://doi.org/10.1371/journal.pone.0264346 February 22, 2022 17 / 17 PLOS ONE
10.1371_journal.pone.0263124	RESEARCH ARTICLE Influence of glucose on swarming and quorum sensing of Dickeya solani Roberta Gatta1, Andrzej Wiese1, Adam Iwanicki2, Michał ObuchowskiID 2* 1 Division of Medical Biotechnology, Intercollegiate Faculty of Biotechnology UG-MUG, University of Gdańsk, Gdańsk, Poland, 2 Division of Medical Biotechnology, Intercollegiate Faculty of Biotechnology UG-MUG, Medical University of Gdańsk, Gdańsk, Poland * michal.obuchowski@ug.edu.pl Abstract Dickeya solani is a pathogen most frequently responsible for infecting potato plants in Europe. As in the case of most plant pathogens, its ability to colonize and invade the host depends on chemotaxis and motility. The coordinated movement of Dickeya over solid sur- faces is governed by a quorum sensing mechanism. In D. solani motility is regulated by ExpI-ExpR proteins, homologous to luxI-luxR system from Vibrio fisheri, in which N-acyl- homoserine lactones (AHLs) serve as signaling molecules. Moreover, in many Gram-nega- tive bacteria motility is coupled with central metabolism via carbon catabolite repression. This enables them to reach more nutrient-efficient niches. The aim of this study was to ana- lyze the swarming motility of D. solani depending on the volume of the medium in the cultiva- tion plate and glucose content. We show that the ability of this bacterium to move is strictly dependent on both these factors. Moreover, we analyze the production of AHLs and show that the quorum sensing mechanism in D. solani is also influenced by the availability of glu- cose in the medium and that the distribution of these signaling molecules are different depending on the volume of the medium in the plate. Introduction Dickeya genus consists of pathogens targeting economically important plants. According to the current classification, the genus includes the following species: D. aquatica, D. chry- zanthemi, D. dadanti, D. dianthicola, D. fangzhongdai, D. paradisiaca, D. solani, and D. zeae [1–5]. Among them, Dickeya solani species are bacteria most commonly infecting potato plants in Europe causing black-leg and soft rot diseases [6]. In its natural environment, Dickeya must deal with limited nutrients availability and stress- ful conditions encountered in water and soil. Because of that, the pathogenesis of these bacteria must be strictly controlled to avoid waste of energy and unnecessary production of virulence factors. So far 10 major regulators of Dickeya virulence have been described: KdgR, PecS, PecT, CRP, H-NS, Fis, Fur, GacA, SlyA and MfbR [7, 8]. There are two phases of Dickeya path- ogenic cycle. In the first bacteria remain latent while penetrating the host through wounds or natural openings [9]. Upon colonization, bacteria multiply slowly while constantly monitoring the availability of plant soluble sugars which can be used in cellular metabolism. Along with a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Gatta R, Wiese A, Iwanicki A, Obuchowski M (2022) Influence of glucose on swarming and quorum sensing of Dickeya solani. PLoS ONE 17(2): e0263124. https://doi.org/10.1371/journal. pone.0263124 Editor: Abdelwahab Omri, Laurentian University, CANADA Received: June 17, 2021 Accepted: January 12, 2022 Published: February 22, 2022 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0263124 Copyright: © 2022 Gatta et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper. Funding: This work was supported by the National Science Center OPUS project no. 2018/29/B/NZ9/ 02339. The funders had no role in study design, PLOS ONE \| https://doi.org/10.1371/journal.pone.0263124 February 22, 2022 1 / 12 PLOS ONEdata collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Influence of glucose on swarming of D. solani the increasing availability of such sugars bacteria switch to the second phase of the pathogenic cycle in which they multiply massively and undergo profound metabolic changes. They start producing large amounts of plant cell wall-degrading enzymes (CWDEs), mainly pectate lyases, which provide bacteria with required nutrients and lead to the destruction of plant tis- sues [10]. An essential role in the colonization of plants by these pathogens is played by chemotaxis and motility. These two phenomena enable bacteria to sense compounds of plant origin and move towards their source which is most probably a wound or natural opening in the plant tis- sue. Bacterial movement over solid surface occurs by the mechanism of swarming [11] which, as a coordinated social behavior of bacteria, must be subjected to regulation. In D. solani and many other bacteria, this phenomenon is regulated by a quorum sensing mechanism which serves as a cell-to-cell communication system [12]. There are two quorum sensing systems found in members of the Dickeya genus. The first one is the classic system mediated by N-acyl-homoserine lactones (AHLs) and is homologous to the best-studied luxI-luxR system from Vibrio fisheri [13]. In Dickeya this system consists of ExpI-ExpR proteins [14] and is known to regulate protease production and motility of these bacteria [15]. Exp system relies mainly on N-(3-oxohexanolyl)-homoserine lactone with only minimal contribution of N-(hexanoyl)-homoserine lactone [16]. The second system is called Virulence Factor Modulating (VFM) and is encoded in 25 kb cluster of genes whose products are involved in the pathogenesis and production of CWDEs [17]. VFM was shown in Dickeya to be unrelated to motility [15]. Swarming was shown to be dependent on the presence of catabolite repression protein (Crc) in Pseudomonas syringae, another Gram-negative plant pathogen [18]. The canonical function of carbon catabolite repression is directing the choice of nutrients by favoring the uti- lization of the most efficiently metabolizable sugars [19]. Such coupling with the central metabolism suggests that the motility of bacteria is associated with the metabolic state of the cell and availability of nutrients in the environment helping bacteria in reaching more nutri- ent-efficient niches. In Dickeya no such direct link has been shown, nonetheless, the expres- sion of virulence genes is known to be activated when efficient carbon sources are exhausted. This regulation occurs mostly by the carbon repression protein (CRP) [20]. In this study, we investigated how the volume and composition of cultivation media influ- enced the swarming of D. solani. We also analyzed how the concentration of glucose in the medium impacts the production of quorum sensing molecules by these bacteria. Results Triggering swarming motility of bacteria in the laboratory requires precisely controlled condi- tions such as medium composition, percentage of agar, time of plate drying, temperature and humidity at incubation. Most publications concerning the swarming of Dickeya species pro- vide general information regarding media composition and percentage of agar which usually is 0.5%. We have used this concentration of agar and performed swarming assays with chang- ing volumes of the medium used in Petri plates. To our surprise, Dickeya was unable to swarm on plates containing more than 10 ml of solid medium (Fig 1). The most efficient swarming was observed in the case of plates containing 5 ml of medium, nevertheless, due to difficulties with the preparation of uniform plates (very fast solidification of agar), we decided to use plates containing 7.5 ml of medium in further experiments. The process of surface colonization and biofilm formation in Gram-negative plant patho- gens is coordinated via a quorum sensing system [12]. We wanted to check whether observed differences in swarming of Dickeya correlated with the presence of N-acyl homoserine lactones PLOS ONE \| https://doi.org/10.1371/journal.pone.0263124 February 22, 2022 2 / 12 PLOS ONEInfluence of glucose on swarming of D. solani Fig 1. Swarming and AHLs production by Dickeya solani IFB102 on plates containing following volumes of B-medium: (A, F) 7.5 ml, (B, G) 10 ml, (C, H) 15 ml, (D, I) 20 ml, (E, J) 25 ml. Swarming assays were performed for 24 and 48 h, as indicated. Grey circles represent spots at which AHLs were detected. Open circles represent spots at which no AHLs were detected. https://doi.org/10.1371/journal.pone.0263124.g001 (AHLs) in the medium. AHLs are common diffusible signaling molecules utilized by Gram- negative bacteria for this purpose [21]. We performed swarming assays on plates containing different volumes of the medium. Medium in the plates was then sampled at several spots and tested for the presence of AHL molecules using a biosensor strain of E. coli PSB401 [22]. After 24 hours of incubation, we were able to detect AHLs at all tested spots of 7.5 and 10 ml plates (Fig 1A and 1B), suggesting that these molecules were present in the entire plates. In the case of plates containing 15 and 20 ml of the medium we were able to detect these quorum sensing molecules at the inoculation spot and within 1 cm radius from it (Fig 1C and 1D). In the plates containing 25 ml of medium, we could detect AHLs only at the inoculation spot (Fig 1E). Pro- longation of the incubation time up to 48 hours changed the pattern of AHLs distribution in the medium. In the plates containing 7.5 ml of medium, where we observed the most efficient swarming of Dickeya, the radius of AHLs detection was reduced down to 2 cm from the inocu- lation spot (Fig 1F). In the plates with 10, 15, and 20 ml of the medium we detected AHLs at every tested spot (Fig 1G–1I). In the case of the plate containing 25 ml of medium AHLs were detected within 2 cm radius from the inoculation spot (Fig 1J). Swarming motility is dependent on the metabolic state of bacteria and the availability of nutrients in the environment. We started with verifying the swarming of Dickeya on a medium PLOS ONE \| https://doi.org/10.1371/journal.pone.0263124 February 22, 2022 3 / 12 PLOS ONEInfluence of glucose on swarming of D. solani Fig 2. Swarming and AHLs production by Dickeya solani IFB102 on plates containing 0.5% of agar and the changing concentrations of B- medium: (A, F) 0.5x, (B, G) 1x, (C, H) 1.5x, (D, I) 2x, (E, J) 2.5x. Swarming assays were performed for 24 and 48 h, as indicated. Grey circles represent spots at which AHLs were detected. Open circles represent spots at which no AHLs were detected. https://doi.org/10.1371/journal.pone.0263124.g002 prepared with increasing content of its components while keeping agar concentration at 0.5%. We observed that after 24 h of incubation Dickeya swarmed on 0.5x and 1x concentrated B- medium (Fig 2A and 2B). For higher concentrations of medium bacteria exhibited no swarming motility (Fig 2C– 2E). Prolongation of incubation time up to 48 h enabled Dickeya to additionally start swarming on plates containing 1.5x concentrated medium (Fig 2C). The pattern of AHLs distribution in the medium changed along with the increase of medium concentration and incubation time. For 24 hours incubation we detected AHL molecules all over the plates containing 0.5x, 1x, and 1.5x medium (Fig 2A–2C). In the case of plates with 2x concentrated medium, we could detect AHLs within 1 cm radius from the inoculation spot (Fig 2D). Sampling of plates con- taining 2.5x concentrated medium allowed us to detect AHLs only at the inoculation spot (Fig 2E). The 48-hour incubation changed the distribution of AHLs in the tested plates. In the plates containing the least concentrated medium (0.5x), we detected AHL molecules within 1 cm radius from the inoculation spot (Fig 2F). The radius of AHLs detection increased up to 2 cm in the case of plates with 1x medium (Fig 2G) and up to at least 3 cm from the inoculation spot as observed for the plates containing 1.5x medium (Fig 2H). Further increase in medium PLOS ONE \| https://doi.org/10.1371/journal.pone.0263124 February 22, 2022 4 / 12 PLOS ONEInfluence of glucose on swarming of D. solani Fig 3. Swarming and AHLs production by Dickeya solani IFB102 on plates containing 0.5% of glucose, 0.5% of agar and the following concentrations of remaining components of B-medium: (A, F) 0.5x, (B, G) 1x, (C, H) 1.5x, (D, I) 2x, (E, J) 2.5x. Swarming assays were performed for 24 and 48 h, as indicated. Grey circles represent spots at which AHLs were detected. Open circles represent spots at which no AHLs were detected. https://doi.org/10.1371/journal.pone.0263124.g003 concentration (2x, 2.5x) resulted in decreasing the radius of AHLs detection down to 1 cm from the inoculation spot (Fig 2I and 2J). The glucose content in the medium was shown to affect the motility of bacteria [23]. In our experiments with increasing concentration of medium components, one of them was this sugar. To assess whether observed changes in Dickeya swarming were due to the increased glu- cose concentration we performed a series of swarming assays on plates with a medium con- taining changing contents of its components while keeping glucose and agar concentrations at 0.5%. After both, 24 h and 48 h of incubation, we observed swarming of Dickeya only on plates containing 0.5x (Fig 3A and 3F) and 1x (Fig 3B and 3G) concentrated medium. The pattern of AHLs distribution in the medium also changed along with the increasing concentration of medium components and incubation time. In the case of the 24-hour incuba- tion, AHL molecules were detected in the entire plates containing 0.5x, 1x, and 1.5x concen- trated medium with constant content of 0.5% glucose (Fig 3A–3C). It is worth noticing that on the plate containing 1.5x medium bacteria were not swarming. In the case of 2x and 2.5x con- centrated medium with 0.5% glucose AHLs were detected within 1 cm radius from the PLOS ONE \| https://doi.org/10.1371/journal.pone.0263124 February 22, 2022 5 / 12 PLOS ONEInfluence of glucose on swarming of D. solani inoculation spot (Fig 3D and 3E). The 48-hour incubation enabled the detection of AHL mole- cules within 2cm radius from the inoculation spot in all plates apart from the plate containing 1.5x medium where AHLs detection radius increased up to 3 cm. Obtained results suggest that the swarming of Dickeya is affected not only by increasing glucose content but also other com- ponents of the B-medium. In the final experiment, we wanted to analyze the influence of increasing glucose content on the swarming motility of Dickeya. Therefore, we performed a series of 24-hour swarming assays on 0.5x concentrated B-medium with glucose content increasing from 0% to 5%. In the results we could distinguish four different patterns of Dickeya swarming: (i) a small central col- ony without visible swarming (0% to 0.1% of glucose, Fig 4A–4D), (ii) a central colony with the increasing ring of swarming bacteria (0.25% to 0.4% of glucose, Fig 4E–4G), (iii) a small colony at the inoculation spot with extending irregular dendrites (0.5% to 3% of glucose, Fig 4H–4M), (iv) a large uniform central colony without dendrites (4% and 5% of glucose, Fig 4N and 4O). Obtained results suggest that some minimal concentration of glucose (at least 0.25%) was required to trigger Dickeya swarming motility. High glucose content (4% and 5%) was also not optimal for Dickeya to develop an efficient swarming. The distribution of AHL mole- cules in the medium depended on glucose content. For the lowest glucose contents (0%, 0.01% and 0.02%) the radius of AHLs detection was 2 cm from the inoculation spot (Fig 4A–4C). For plates containing between 0.1% and 1% of glucose, AHLs were present within the radius of at least 3 cm from the inoculation spot (Fig 4D–4I). In the case of plates in which glucose content was between 1.5% and 3%, AHL molecules were detected in the area of medium covered with swarming bacteria (Fig 4J–4M). In the case of plates with the highest contents of glucose (4% and 5%), we were unable to detect any AHL molecules at all tested spots (Fig 4N and 4O). Discussion Plant pathogens, for most of their lives, reside outside of their host. The efficient colonization of the plant required the development of mechanisms enabling sensing and active movement of pathogen cells towards favorable sites of infection. Upon adhesion to the plant tissues, bacte- ria can penetrate them through wounds and natural openings, thus establishing a successful infection. These mechanisms are chemotaxis and motility. Virulence of different plant patho- gens, including Ralstonia solanacearum, Pseudomonas phaseolicola, Pseudomonas syringae, and Dickeya dadanti, have been demonstrated to be strictly dependent on the ability of bacte- ria to sense and actively move in the environment [24–26]. Genes involved in the motility regulation belong to the group exhibiting significant induc- tion in the plant tissue [27]. Variations in the motility of Dickeya solani have also been pro- posed to contribute to its aggressiveness variability [28]. These facts clearly suggest the importance of motility in the virulence of Dickeya and their capability of efficiently infecting the plant host. It also needs to be emphasized that the motility of D. solani is dependent on AHLs-based quorum sensing [15], and therefore should be regarded as the social behavior of these bacteria. Our observation that the swarming of D. solani on agar plates depends on the volume of used the medium (Fig 1) can be explained by two hypotheses. The first assumes that a small volume of medium in the plate results in a relatively fast depletion of available nutrients. In response to such conditions, bacteria could trigger swarming motility to enable the optimal dissemination of cells, as the local population was too large for a given niche [29]. An alterna- tive hypothesis assumes that a smaller volume of medium in the plate enables faster accumula- tion of factors secreted by bacteria. Among these factors, AHL quorum sensing molecules PLOS ONE \| https://doi.org/10.1371/journal.pone.0263124 February 22, 2022 6 / 12 PLOS ONEInfluence of glucose on swarming of D. solani Fig 4. Swarming and AHLs production by Dickeya solani IFB102 on plates with B-medium containing 0.5% of agar and the following concentrations glucose: (A) 0%, (B) 0.01%, (C) 0.02%, (D) 0.1%, (E) 0.25%, (F) 0.3%, (G) 0.4%, (H) 0.5%, (I) 1%, (J) 1.5%, (K) 2%, (L) 2.5%, (M) 3%, (N) 4%, (O) 5%. Swarming assays were performed for 24 h. Grey circles represent spots at which AHLs were detected. Open circles represent spots at which no AHLs were detected. https://doi.org/10.1371/journal.pone.0263124.g004 might be responsible for triggering the swarming motility once the threshold concentration is reached. Our results support this hypothesis because we could observe the increase in the area of the plate with detectable AHLs as the volume of the medium decreased (Fig 1). PLOS ONE \| https://doi.org/10.1371/journal.pone.0263124 February 22, 2022 7 / 12 PLOS ONEInfluence of glucose on swarming of D. solani The influence of glucose on the motility of bacteria has been demonstrated by different research groups. In the early study, Armitage et al. showed that 1% glucose present in the medium delayed swarming of Proteus mirabilis, an enterobacterium associated with infections of the urinary tract in humans [30]. The study performed on several species of enterobacteria suggested that glucose did not affect swimming motility [31]. Moreover, swarming but not swimming motility of Pseudomonas aeruginosa, is dependent on changing concentrations of glucose in the medium [32]. It is important to emphasize that swimming and swarming motili- ties are different phenomena. While swimming is a non-coordinated movement of bacteria through the water channels in the agar, swarming is a social phenomenon of bacterial move- ment over the solid surface, requiring specific conditions and precise regulation [12]. Jahid et al. showed that increasing concentration of glucose inhibited swarming of Aeromonas hydrophila, a pathogen of fish and amphibians. Moreover, these authors extended their research and analyzed the influence of glucose on biofilm formation and proteases production. They proposed a conceptual model in which observed effects resulted from glucose-dependent modulation of quorum sensing [23]. Our results are concordant with the above observations since we showed that the swarming of D. solani changed along with glucose concentration in the medium (Fig 4). A similar effect was observed in the experiment with a concentrated medium (Fig 2) and could result from a higher content of glucose. Along with glucose-depen- dent changes in the swarming pattern, we observed differences in the distribution of AHL quo- rum sensing molecules in the medium. This suggests a tight association between glucose concentration in the medium, swarming, and quorum sensing in D. solani. So far no direct link between these has been demonstrated for these bacteria, nevertheless the study of Potry- kus et al. indicated that quorum sensing systems in D. solani influenced the maceration of plant tissue, the production of plant cell wall-degrading enzymes, as well as swarming motility [15]. The major mechanism governing the utilization of different carbon sources in bacteria is carbon catabolite repression (CCR). This mechanism optimizes the uptake of the most effi- ciently metabolized sugars available at the moment in the environment, and glucose is the most preferred one [19]. The CCR was shown to regulate the virulence of different pathogenic bacteria. In Pseudomonas aeruginosa two quorum sensing systems: AHL-based (las and rhl) and non-AHL-based (qps and iqs), interact with one another to control the expression of genes according to the density of the bacterial population [33]. Quorum sensing in these bacteria was also shown to contribute to their virulence [34], moreover it was demonstrated to be cou- pled with the CCR via protein quality control (PQC) proteases Lon and ClpXP [35]. The AHL- based quorum sensing systems las and rhl in P. aeruginosa are involved in the production of rhamnolipids, biosurfactants required for efficient swarming of these bacteria [36]. These sys- tems were also shown to be controlled by the CCR [35]. In the case of plant pathogens, the expression of genes involved in the virulence of Dickeya dadanti was demonstrated to be regu- lated by the CCR [20]. To conclude, the regulation of D. solani swarming and AHLs production by glucose seems to be via the carbon catabolite repression. Increasing concentration of glucose in the medium ensures availability of the most efficiently metabolizable sugar and thus turns the swarming motility into the unnecessary expense of energy. This cessation of motility is associated with decreased production of AHLs, and one can hypothesize that this low abundance of quorum sensing signaling molecules in the medium prevents the triggering of swarming. Results pre- sented in our study broaden our knowledge about the physiology of the important plant patho- gen D. solani. Moreover, they might be useful from a practical point of view. The outcomes of swarming assays performed with this bacterium apparently can be dependent not only on such PLOS ONE \| https://doi.org/10.1371/journal.pone.0263124 February 22, 2022 8 / 12 PLOS ONEInfluence of glucose on swarming of D. solani conditions as medium composition, temperature, and humidity but also can change with the volume of medium in the plate used to cultivate bacteria. Materials and methods Bacterial strains and media Dickeya solani strain IFB102 [37] was used as a wild-type strain. Bacteria capability to produce and release acyl-HLs in the environment was assessed based on the reporter strain Escherichia coli [pSB401] (TcR; luxRI’::luxCDABE) developed by Winson et al. [22]. Bacteria were cultured in Luria broth (LB) medium (tryptose 10 g/l, yeast extract 5 g/l, NaCl 10 g/l) supplemented with antibiotic when required. The temperature of growth was set at 28˚C for D. solani and at 37˚C in the case of E. coli. Swarming motility was performed on synthetic B-medium [38] which contains 15 mM (NH4)2SO4, 8 mM MgSO4, 27 mM KCl, 7 mM sodium citrate, 50 mM Tris/HCl (pH 7.5) supplemented on the day of inoculation with 0.6 mM KH2PO4 2 mM CaCl2, 1 μM FeSO4, 10 μM MnSO4, 4.5 mM glutamic acid, 0.78 mM tryptophan, 0.8 mM Lysine and 0.5% (w/v) glucose. All plates were prepared by supplementing the medium with 0.5% (w/v) of Bacto agar. Swarming motility A single colony of D. solani IFB102 was inoculated in LB medium and incubated overnight with shaking at 28˚C. Two microliters of the overnight culture (OD600 � 0.8) were inoculated in the center of a B-medium plate (0.5% of agar) and incubated for 24h or 48h at 28˚C (relative humidity 55% saturation). Plates were prepared 1 h before the inoculation and dried open for 30 min in a laminar flow chamber. B-medium plates contained different volume of medium (7.5 ml, 10 ml, 15 ml, 20 ml and 25 ml) and increasing concentration of medium (0.5x, 1x, 1.5x, 2x and 2.5x). Glucose influence on swarming motility was determined with plates con- taining 7.5 ml of 0.5x concentrated B-medium (0.5% of agar) with increasing concentration of glucose (0, 0.01, 0.02, 0.1, 0.25, 0.3, 0.4, 0.5, 1, 1.5, 2, 2.5, 3, 4 and 5% (w/v)). Detection of AHLs released in B-medium agar plates In this study, we developed a fast-screening method for the detection of N-Acyl homoserine lactones (AHLs) in agar plates based on the method previously presented by Jafra et al. [39]. Escherichia coli [pSB401] was used as a biosensor for detection of AHLs due to its high level of bioluminescence mediated by the presence of N-3-(Oxohexanoyl)-L-homoserine lactones [22]. A single colony of E. coli [pSB401] was inoculated in 5 ml of LB supplemented with 20 μl/ ml of tetracycline and incubated overnight at 37˚C. The day after, the culture was diluted to OD600 � 0.1 and incubated for 5 hours. Swarming motility plates were examined to deter- mine the presence of AHLs. Thirteen holes of *9 mm diameter were cut from each plate by using a flamed cork borer. The circular agar pieces were collected at 1 cm away from each other, starting from the inoculation point and proceeding in the four directions (above, below, left, right) up to 3 cm away from the point of inoculation. The circular agar samples were transferred directly to a sterile 96 wells plate and each well was inoculated with 150 μl of diluted suspension of the indicator strain (OD600 � 0.2). Plates were incubated overnight at 37˚C, the growth temperature not permissive for D. solani. Emission of chemiluminescence was detected with the ChemiDoc XRS+ system (BIO-RAD). Author Contributions Conceptualization: Adam Iwanicki, Michał Obuchowski. PLOS ONE \| https://doi.org/10.1371/journal.pone.0263124 February 22, 2022 9 / 12 PLOS ONEInfluence of glucose on swarming of D. solani Data curation: Roberta Gatta, Adam Iwanicki. Formal analysis: Michał Obuchowski. Funding acquisition: Michał Obuchowski. Investigation: Roberta Gatta, Andrzej Wiese. Methodology: Roberta Gatta, Andrzej Wiese. Project administration: Michał Obuchowski. Supervision: Michał Obuchowski. Validation: Roberta Gatta, Adam Iwanicki. Visualization: Adam Iwanicki. Writing – original draft: Adam Iwanicki. Writing – review & editing: Roberta Gatta, Adam Iwanicki, Michał Obuchowski. References 1. Brady CL, Cleenwerck I, Denman S, Venter SN, Rodrı´guez-Palenzuela P, Coutinho TA, et al. Proposal to reclassify Brenneria quercina (Hildebrand and Schroth 1967) Hauben et al. 1999 into a new genus, Lonsdalea gen. nov., as Lonsdalea quercina comb. nov., descriptions of Lonsdalea quercina subsp. quercina comb. nov., Lonsdalea quercina subsp. iberica subsp. nov. and Lonsdalea quercina subsp. britannica subsp. nov., emendation of the description of the genus Brenneria, reclassification of Dickeya dieffenbachiae as Dickeya dadantii subsp. dieffenbachiae comb. nov., and emendation of the descrip- tion of Dickeya dadantii. Int J Syst Evol Microbiol. 2012; 62:1592–1602. https://doi.org/10.1099/ijs.0. 035055-0 PMID: 21890733 2. Parkinson N, DeVos P, Pirhonen M, Elphinstone J. Dickeya aquatica sp. nov., isolated from waterways. Int J Syst Evol Microbiol. 2014; 64:2264–2266. https://doi.org/10.1099/ijs.0.058693-0 PMID: 24719023 3. Samson R, Legendre JB, Christen R, Fischer-Le Saux M, Achouak W, Gardan L. Transfer of Pectobac- terium chrysanthemi (Burkholder et al. 1953) Brenner et al. 1973 and Brenneria paradisiaca to the genus Dickeya gen. nov. as Dickeya chrysanthemi comb. nov. and Dickeya paradisiaca comb. nov. and delineation of four novel species, Dickeya dadantii sp. nov., Dickeya dianthicola sp. nov., Dickeya dief- fenbachiae sp. nov. and Dickeya zeae sp. nov. Int J Syst Evol Microbiol. 2005; 55:1415–1427. https:// doi.org/10.1099/ijs.0.02791-0 PMID: 16014461 4. Tian Y, Zhao Y, Yuan X, Yi J, Fan J, Xu Z, et al. Dickeya fangzhongdai sp. nov., a plant-pathogenic bac- terium isolated from pear trees (Pyrus pyrifolia). Int J Syst Evol Microbiol. 2016; 66:2831–2835. https:// doi.org/10.1099/ijsem.0.001060 PMID: 27045848 5. Van der Wolf JM, Nijhuis EH, Kowalewska MJ, Saddler GS, Parkinson N, Elphinstone JG, et al. Dickeya solani sp. nov., a pectinolytic plant pathogenic bacterium isolated from potato (Solanum tuberosum). Int J Syst Evol Microbiol. 2014; 64:768–774. https://doi.org/10.1099/ijs.0.052944-0 PMID: 24225027 6. Toth IK, van der Wolf JM, Saddler G, Lojkowska E, Helias V, Pirhonen M, et al. Dickeya species: an emerging problem for potato production in Europe. Plant Pathol. 2011; 60:385–399. 7. Reverchon S, Nasser W. Dickeya ecology, environment sensing and regulation of virulence pro- gramme. Environ Microbiol Rep. 2013; 5:622–636. https://doi.org/10.1111/1758-2229.12073 PMID: 24115612 8. Reverchon S, Muskhelisvili G, Nasser W. Virulence program of a bacterial plant pathogen: the Dickeya model. Prog Mol Biol Transl Sci. 2016; 142:51–92. https://doi.org/10.1016/bs.pmbts.2016.05.005 PMID: 27571692 9. Rı´o-A´ lvarez I, Muñoz-Go´ mez C, Navas-Va´ squez M, Marta´nez-Garcıa PM, Antu´ nez-Lamas M, Rodru´ - guez-Palenzuela P, et al. Role of Dickeya dadantii 3937 chemoreceptors in the entry to Arabidopsis leaves through wounds. Mol Plant Pathol. 2015; 16:685–698. https://doi.org/10.1111/mpp.12227 PMID: 25487519 10. Metabolism Hugouvieux-Cotte-Pattat N. and virulence stratyegies in Dickeya-host interactions. Prog Mol Biol Transl Sci. 2016; 142:93–129. https://doi.org/10.1016/bs.pmbts.2016.05.006 PMID: 27571693 11. Henrichsen J. Bacterial surface translocation: a survey and a classification. Bacteriol Rev. 1972; 36:478–503. https://doi.org/10.1128/br.36.4.478-503.1972 PMID: 4631369 PLOS ONE \| https://doi.org/10.1371/journal.pone.0263124 February 22, 2022 10 / 12 PLOS ONEInfluence of glucose on swarming of D. solani 12. Daniels R, Vanderleyden J, Michiels J. Quorum sensing and swarming migration in bacteria. FEMS Microbiol Rev. 2004; 28:261–289. https://doi.org/10.1016/j.femsre.2003.09.004 PMID: 15449604 13. Fuqua C, Greenberg EP. Listening in on bacteria: acyl-homoserine lactone signalling. Nat Rev Mol Cell Biol. 2002; 3:685–695. https://doi.org/10.1038/nrm907 PMID: 12209128 14. Reverchon S, Bouillant ML, Salmond G, Nasser W. Integration of the quorum-sensing system in the regulatory networks controlling virulence factor synthesis in Erwinia chrysanthemi. Mol Microbiol. 1998; 29:1407–1418. https://doi.org/10.1046/j.1365-2958.1998.01023.x PMID: 9781878 15. Potrykus M, Hugouvieux-Cotte-Pattat N, Lojkowska E. Interplay of classic Exp and specific Vfm quorum sensing systems on the phenotypic features of Dickeya solani strains exhibiting different virulence lev- els. Mol Plant Pathol. 2018; 19:1238–1251. https://doi.org/10.1111/mpp.12614 PMID: 28921772 16. Cre´ pin A, Barbey C, Beury-Cirou A, He´ lias V, Taupin L, Reverchon S, et al. Quorum sensing signaling molecules produced by reference and emerging soft-rot bacteria (Dickeya and Pectobacterium spp.). PLoS One. 2012; 7:e35176. https://doi.org/10.1371/journal.pone.0035176 PMID: 22539957 17. Nasser W, Dorel C, Wawrzyniak J, Van Gijsegem F, Groleau M, Deziel E, et al. Vfm a new quorum sensing system controls the virulence of Dickeya dadantii. Environ Microbiol. 2013; 15:865–880. https:// doi.org/10.1111/1462-2920.12049 PMID: 23227918 18. Chakravarthy S, Butcher BG, Liu Y, D’Amico K, Coster M, Filiatrault MJ. Virulence of Pseudomonas syr- ingae pv. tomato DC3000 Is Influenced by the Catabolite Repression Control Protein Crc. Mol Plant Microbe Interact. 2017; 30:283–294. https://doi.org/10.1094/MPMI-09-16-0196-R PMID: 28384054 19. Go¨rke B, Stu¨ lke J. Carbon catabolite repression in bacteria: many ways to make the most out of nutri- ents. Nat Rev Microbiol. 2008; 6:613–624. https://doi.org/10.1038/nrmicro1932 PMID: 18628769 20. Reverchon S, Expert D, Robert-Baudouy J, Nasser W. The cyclic AMP receptor protein is the main acti- vator of pectinolysis genes in Erwinia chrysanthemi. J Bacteriol. 1997; 179:3500–3508. https://doi.org/ 10.1128/jb.179.11.3500-3508.1997 PMID: 9171393 21. Swift S, Throup JP, Salmond GPC, Williams P, Stewart GSAB. Quorum sencing: a population-density component in the determination of bacterial phenotype. Trends Biochem Sci. 1996; 21:214–219. PMID: 8744355 22. Winson MK, Swift S, Fish L, Throup JP, Jørgensen F, Chhabra SR, et al. Construction and analysis of luxCDABE-based plasmid sensors for investigating N-acyl homoserine lactone-mediated quorum sens- ing. FEMS Microbiol Lett. 1998; 163:185–192. https://doi.org/10.1111/j.1574-6968.1998.tb13044.x PMID: 9673021 23. Jahid IQ, Lee N-Y, Kim A, Ha S-D. Influence of glucose concentrations on biofilm formation, motility, exoprotease production, and quorum sensing in Aeromonas hydrophila. J Food Prot. 2013; 76:239– 247. https://doi.org/10.4315/0362-028X.JFP-12-321 PMID: 23433371 24. Antu´ nez-Lamas M, Cabrera-Ordo´ñez E, Lo´pez-Solanilla E, Raposo R, Trelles-Salazar O, Rodrı´guez- Moreno A, et al. Role of motility and chemotaxis in the pathogenesis of Dickeya dadantii 3937 (ex Erwi- nia chrysanthemi 3937). Microbiology. 2009; 155:434–442. https://doi.org/10.1099/mic.0.022244-0 PMID: 19202091 25. Ichinose Y, Shimizu R, Ikeda J, Taguchi F, Marutani M, Mukaihara T, et al. Need for flagella for com- plete virulence of Pseudomonas syringae pv. tabaci: genetic analysis with flagella-defective mutants ΔfliC and ΔfliD in host tobacco plants. J Gen Plant Pathol. 2003; 69:244–249. 26. Panopoulos N, Schroth MN. Role of flagellar motility in the invasion of bean leaves by Pseudomonas phaseolicola. Phytopathology. 1974; 64:1389–1397. 27. Czajkowski R, Fikowicz-Krosko J, Maciag T, Rabalski L, Czaplewska P, Jafra S, et al. Genome-Wide Identification of Dickeya solani Transcriptional Units Up-Regulated in Response to Plant Tissues From a Crop-Host Solanum tuberosum and a Weed-Host Solanum dulcamara. Front Plant Sci. 2020; 11:580330. https://doi.org/10.3389/fpls.2020.580330 PMID: 32983224 28. Khayi S, Blin P, Pe´ dron J, Chong TM, Chan KG, Moumni M, et al. Population genomics reveals additive and replacing horizontal gene transfers in the emerging pathogen Dickeya solani. BMC Genomics. 2015; 16:788. https://doi.org/10.1186/s12864-015-1997-z PMID: 26467299 29. Whitehead NA, Barnard AM, Slater H, Simpson NJ, Salmond GP. Quorum-sensing in Gram-negative bacteria. FEMS Microbiol Rev. 2001; 25:365–404. https://doi.org/10.1111/j.1574-6976.2001.tb00583.x PMID: 11524130 30. Armitage JP, Rowbury RJ, Smith DG. The role of cyclic adenosine monophosphate in the swarming phenomenon of Proteus mirabilis. Experientia. 1976; 32:1266–1267. https://doi.org/10.1007/ BF01953085 PMID: 185081 31. Lai HC, Shu JC, Ang S, Lai MJ, Fruta B, Lin S, et al. Effect of glucose concentration on swimming motil- ity in enterobacteria. Biochem Biophys Res Commun. 1997; 231:692–695. https://doi.org/10.1006/ bbrc.1997.6169 PMID: 9070873 PLOS ONE \| https://doi.org/10.1371/journal.pone.0263124 February 22, 2022 11 / 12 PLOS ONEInfluence of glucose on swarming of D. solani 32. Mattingly AE, Kamatkar NG, Morales-Soto N, Borlee BR, Shrout JD. Multiple Environmental Factors Influence the Importance of the Phosphodiesterase DipA upon Pseudomonas aeruginosa Swarming. Appl Environ Microbiol. 2018; 84:e02847–17. https://doi.org/10.1128/AEM.02847-17 PMID: 29427430 33. Lee J, Zhang L. The hierarchy quorum sensing network in Pseudomonas aeruginosa. Protein Cell. 2015; 6:26–41. https://doi.org/10.1007/s13238-014-0100-x PMID: 25249263 34. Rumbaugh KP, Griswold JA, Iglewski BH, Hamood AN. Contribution of Quorum Sensing to the Viru- lence of Pseudomonas aeruginosa in Burn Wound Infections. Infect Immun. 1999; 67:5854–5862. https://doi.org/10.1128/IAI.67.11.5854-5862.1999 PMID: 10531240 35. Yang N, Lai N. Pseudomonas aeruginosa Lon and ClpXP proteases: roles in linking carbon catabolite repression system with quorum-sensing system. Curr Genet. 2016; 62:1–6. https://doi.org/10.1007/ s00294-015-0499-5 PMID: 26045103 36. Pearson JP, Passador L, Iglewski BH, Greenberg EP. A second N-acylhomoserine lactone signal pro- duced by Pseudomonas aeruginosa. Proc Natl Acad Sci U S A.1995; 92:1490–1494. https://doi.org/10. 1073/pnas.92.5.1490 PMID: 7878006 37. Sławiak M, van Beckhoven JRCM, Speksnijder AGCL, Czajkowski R, Grabe G, van der Wolf JM. Bio- chemical and genetical analysis reveal a new clade of biovar 3 Dickeya spp. strains isolated from potato in Europe. Eur J Plant Pathol. 2009; 125:245–261. 38. Antelmann H, Engelmann S, Schmid R, Sorokin A, Lapidus A, Hecker M. Expression of a stress- and starvation-induced dps/pexB-homologous gene is controlled by the alternative sigma factor sigmaB in Bacillus subtilis. J Bacteriol. 1997; 179:7251–7256. https://doi.org/10.1128/jb.179.23.7251-7256.1997 PMID: 9393687 39. Jafra S, van der Wolf JM. Fast screening method for detection of acyl-HSL-degrading soil isolates. J Microbiol Methods. 2004; 57:415–420. https://doi.org/10.1016/j.mimet.2004.01.015 PMID: 15134889 PLOS ONE \| https://doi.org/10.1371/journal.pone.0263124 February 22, 2022 12 / 12 PLOS ONE
10.1371_journal.pone.0284104	RESEARCH ARTICLE Two edges of the screen: Unpacking positive and negative associations between phone use in everyday contexts and subjective well-being Teodora Sandra BudaID, Mohammed Khwaja, Roger Garriga, Aleksandar MaticID Research and Development, Koa Health, Barcelona, Spain teodorasandrabuda@gmail.com Abstract A plethora of past studies have highlighted a negative association between phone use and well-being. Recent studies claimed that there is a lack of strong evidence on the deleterious effects of smartphones on our health, and that previous systematic reviews overestimated the negative link between phone use and well-being. In a three-week long in-the-wild study with 352 participants, we captured 15,607 instances of smartphone use in tandem with rich contextual information (activity, location, company) as well as self-reported well-being mea- sures. We conducted an additional study to gather users’ perception of the impact of phone use on their well-being in different daily contexts. Our findings show that context and per- sonal characteristics greatly impact the association between screen time and subjective well-being. This study highlights the complexity of the relationship between phone use and well-being and it deepens our understanding of this problem. 1 Introduction a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Buda TS, Khwaja M, Garriga R, Matic A (2023) Two edges of the screen: Unpacking positive and negative associations between phone use in everyday contexts and subjective well-being. PLoS ONE 18(4): e0284104. https://doi.org/ 10.1371/journal.pone.0284104 Editor: Mathew Parackal, University of Otago, NEW ZEALAND Received: September 19, 2022 Accepted: March 23, 2023 Published: April 26, 2023 Copyright: © 2023 Buda et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: There are legal restrictions on sharing the data set used for this study, as the data are owned by Koa Health, the funder organisation, and was used under licence for the current study, and so are not publicly available. These restrictions are imposed by Koa Health. Data are however available from the authors upon reasonable request (with non- commercial intent) and with permission of Koa Health. A non-author contact information to which data requests may be sent is: Daniel Smartphones are the most ubiquitous device in human history and the phenomena related to its use are unsurprisingly discussed across a wide range of scientific domains. One particular theme that fuelled an ongoing debate is the relationship between smartphone use and our well-being. For several years, the predominant view was almost unipolar—studies highlighted the undesirable impact of phone use, including social isolation [1], depression [2], stress [3]. However, recent criticism of previous studies brought an alternative view highlighting a lack of solid evidence for the previous claims around the negative impact of smartphone use [4], that moderate use of digital technology is not intrinsically harmful and may be advantageous in a connected world [5], and others report even benefits with greater screen time [6, 7]. Nev- ertheless, there is a general consensus that more research is needed on this topic that has a spe- cific importance across multiple domains—from psychology and public health to human computer interaction. To deepen our understanding in this area, the importance of solving the methodological issues rooted in self-reported screen time measurements has been demonstrated in multiple records [8–10]. Self-reports generally suffer from subjectivity, memory dependence and recall PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 1 / 26 PLOS ONEClarke, SVP of CyberSecurity at Koa Health, email: d.clarke@koahealth.com. Funding: Koa Health (formerly Telefonica Innovation Alpha) provided financial resources to support this project’s realization. TSB, MK., R.G. and A.M. were employees of Telefonica Innovation Alpha (R.G. and A.M. are now employees of Koa Health) and received salary support. All authors worked on the analysis and writing of this manuscript. All authors received salary support from a funder of this study – Koa Health, a company offering breakthrough technologies to deliver mental health support. Competing interests: The authors have declared that no competing interests exist. Two edges of the screen: Associations between screen time and subjective well-being in everyday contexts bias, and specifically when it comes to the phone use it has been shown that the greater the phone use, the greater the under-reporting of the screen time [8]. Automatically (and objec- tively) capturing the phone use brought more solid findings in this domain. Moreover, scien- tists explored thoroughly what is under the surface of the findings that previously suggested a negative link between the total (i.e., accumulated) screen time and well-being. In this regard, several studies focused both on what the users are doing in the virtual world (i.e., which spe- cific apps such as music, games, productivity apps, etc) [7], a few studies focused on what the users were doing in the physical world while using the phone [11], and a recent study explored both aspects [12]. How the physical context (i.e., what the users where doing concurrently with the phone use, where, and with whom) as well as interpersonal differences (in terms of personal characteristics, typical daily routines, and habitual phone use patterns) impact the relationship between the phone use and well-being remain unanswered. Inspired by prior work, in this paper we unravel the unexplored detail around the ways that mobile phone users’ personalities and the real-world activities and context associated with phone use contribute to their subjective perception of well-being. We captured screen time from smartphone sensors, along with rich contextual information (activities, location and company) through Ecological Momentary Assessments delivered five times a day over a period of three weeks from 352 participants. We further extend and thoroughly analyse the associa- tions between screen time and well-being by considering various additional factors, as well as the interplay between them. These include contextual information such as detailed descrip- tions of activities, surrounding people (also referred to as company), location, and time of the day, as well as personal characteristics (personality, gender, age). Understanding the relationship between the use of smartphones and subjective well- being is of a particular importance to Human Computer Interaction (HCI) research. Design- ing interfaces to enhance user’s emotional well-being in a digital service lead to its increased use, therefore the designers of digital services have been lately focusing more on making ser- vices enjoyable rather than solely optimising for a usability [13]. A recent framework for designing digital experiences [13] highlighted “designing for well-being” as one of the three key designing principles (in addition to motivation and engagement). The lack of under- standing of the underpinning mechanisms in this relationship leaves the HCI researchers ill- equipped to leverage this knowledge for designing smartphone services that better promote well-being. Our results brings the attention for considering different physical contexts and interpersonal differences when designing services that aim to promote well-being and thus a higher engagement. 2 Related work Subjective well-being (SWB) refers to how people evaluate and experience their lives [14]. The literature also distinguishes two important facets of SWB, eudaemonia and hedonia [15, 16], colloquially known as feel-purpose vs. feel-good. Hedonistic well-being is related to positive and negative experiences such as happiness or sadness, whereas eudemonic well-being repre- sents a sense of purpose and meaning in life. Measuring eudaemonia and hedonia associated to activities in everyday life is based on asking people to report how happy vs how worthwhile they felt during an activity [15]. In our study, participants were asked to report both the happi- ness and worthwhileness in each self-report. We believe that distinction between the happiness and worthwhileness is relevant when studying well-being in the context of smartphone use given a multitude of smartphone applications that may have a different impact on these two facets. Although there is no clear-cut categorisation of different applications along the two well-being facets, one may argue that being involved in productivity, learning, or work related PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 2 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts activities may impact more “feel-purpose” whereas music, videos, games and similar may impact more “feel-good” constructs. Moreover, user’s ratings of the two constructs may be very different even with the same kind of activities e.g. using social media to talk to a colleague or to a friend can feel differently in terms of productivity / worthwhileness vs pleasure / happi- ness. Therefore, we considered both dimensions of subjective well-being relevant for this study, tapping into the associations between SWB and smartphone use. Smartphones are tightly embedded in our everyday life, and an ample interest in the effects of their use both by the scientific communities and the media does not come as a surprise. Pre- vious work highlighted a myriad of negative effects of smartphone usage both in the context of a typical as well as of atypical (usually referred to as excessive or problematic) phone use. Undesirable impact was reported in the relationship with well-being [17], cognitive develop- ment [18], and mental health [19–22], spanning to high correlations even with symptoms of mental health disorders including depression [2] and anxiety [23]. The negative effects have been confirmed also in longitudinal studies suggesting that increases in recreational screen time precede a lower psychological well-being [24–30]; refraining from using social media for even just one week was associated with an increase well-being [31], and decrease in anxiety and depression [32]. Moreover, several studies have shown that the presence of smartphones can negatively impact enjoyment during social interactions [33, 34]. Nevertheless, a vast num- ber of existing studies rely on self-reported smartphone usage metrics, which was one of the core methodological issues that called into question the previous findings and even opened a new research theme on exploring misreports of the smartphone use. In a study focusing on smartphone addiction [8], the authors demonstrated that self- reported duration of smartphone use is significant lower than the automatically quantified smartphone use captured via an application that logs every phone usage. The degree of underestimation was positively correlated with actual smartphone use (i.e., the more the par- ticipants used their smartphone, the greater the extent of underestimating its usage dura- tion). A potential reason behind the underestimation of the self-reports is the negative image associated to the overuse of smartphones [35], as suggested after studying this phenomenon from various disciplines (sociology [36], psychology [19], and medicine [37] as well as propa- gated in social media [38]). Either consciously or subconsciously, smartphone users tend to conceal an image of themselves overusing or in extreme cases being addicted to their smart- phone. Along this line, a few studies specifically analysed users’ perception of smartphone usage [9, 10] and report over/underestimation of self-declared usage as compared to actual usage. These findings challenge previous studies relying on self-reported measures of smart- phone usage and suggest that their findings need to be taken with reservation and the rela- tionship between the phone use and mental well-being should be explored further by relying on automatic quantification. This prompted a great deal of studies to delve deeper into the links between well-being and passively measured screen time. Mehrotra et al. [12] examined the relationship between smart- phone interaction and users’ emotional state. Similar to our study, they collected objective measures of smartphone use data (such as app use, notifications, interactions with the phone) as well as self-reported emotional states through experiential sampling. In addition, they included weather information, activity extracted from Google Activity Recognition API [39] (i.e. classifying the user’s activity into walking, bicycling, commuting on vehicle or still), and home and work locations. Among other findings, they reveal that people become more atten- tive and respond faster to notifications in stressful situations whereas an increase in users’ activeness level reduces the usage of music apps. In a similar line, Sarsenbayeva et al. [40] explored the relationship between user’s emotions and smartphone use based on objective measures of smartphone use data, collected together PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 3 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts with emotional states modelled from user’s facial expressions (passively captured using the smartphone’s front-facing camera). The study revealed that phone use typically drives certain emotions rather than the other way around, although they also identified cases in which appli- cation use drives emotions. The authors argued that the causal relationship between emotional state and smartphone use varies among people as emotions and behaviours are intrinsically personal. Moreover, the qualitative survey that they administered revealed various emotions when using the same app which depended on the content. Recently, scientists argued that there is a lack of strong evidence on the detrimental effects of smartphones use on our mental health [41, 42] and that previous studies and systematic reviews overestimated the negative associations [4]. In this regard, Orben et al. [43] found that digital-technology use has a small negative association with adolescent well-being. In particu- lar, the strength of the association between screen time and well-being was similar to neutral factors and the authors compared it to wearing glasses or regularly eating potatoes. Similar studies found null effects or even benefits with greater screen time [6, 7, 44–46]. Moreover, in [5], Przybylski et al. reveal that the moderate use of digital technology is not intrinsically harm- ful and may be advantageous in a connected world. We built on and expanded the previous work by decoupling the relationship between phone use and subjective well-being along two different factors, namely considering: contextual information (activity, location, company, and time of the day), personal characteristics (personality, age, gender). 3 Methods To thoroughly investigate the associations between smartphone usage in various daily contexts and users’ self-reported subjective well-being, we applied a mixed-methods approach by conducting: 1. A quantitative study in which we passively quantified smartphone usage which was coupled with users’ self-reports (Section 4) administered as Ecological Momentary Assessments (EMAs) [47], and 2. A qualitative survey, in which we probed the perceived subjective well-being associated to smartphone usage in different daily contexts (Section 6). The goal of the quantitative study was to understand associations between objectively mea- sured screen time and self-reported well-being measures, with respect to contextual informa- tion, personal characteristics, and individuals’ patterns of using the phone. To complement the quantitative analysis, we administered an extensive survey with an objective to understand which contexts users associate with more positive or more negative emotional responses. The survey was designed considering the quantitative study—we selected the activities, companies and locations that were most frequently reported or that were linked to the polarised emotional responses. Therefore, we delved into the associations between phone use and subjective well-being both from a subjective self-reflection standpoint as well as from the statistical analysis using longitudinal data captured in-situ. In addition, we also analysed the users’ declarative state- ments to acquire a descriptive layer of the same phenomenon in Section 6.3. 4 Quantitative study 4.1 Data collection The data for this study was obtained using a custom-built Android App, designed to collect passively phone usage events from smartphone sensors and well-being from self-reported PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 4 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts Fig 1. Data collection overview. (a) EMA data collection overview and screen time extraction example. (b) Example of EMA self-report and screen time features extraction. https://doi.org/10.1371/journal.pone.0284104.g001 questionnaires. Phone usage events included the timestamps when the phone was unlocked, and when the screen was turned on and off. Data was sampled every time one of these events occurred. Using Ecological Momentary Assessments (EMAs), the app prompted users to rate two dimensions of subjective well-being—happiness and worthwhileness (i.e., hedonic and eudemonic components). The momentary subjective well-being was rated on a 1 to 10 scale (1 being the lowest) at five instances each day during the study. Notifications were sent at a ran- dom time within five time windows: 8:00 AM–10:00 AM, 10:30 AM–12:30PM, 13:00 PM–3:00 PM, 3:30 PM–5:30 PM, 6:00 PM–8:00 PM. EMA notifications were not sent during night time in order not to interfere with sleep. If a user did not attend an EMA prompt, a reminder was sent after 15 minutes. A 30 min gap between each time slot prevented overlaps in subsequent EMA reports, and to provide users with enough time to answer the questions. Fig 1a shows an overview of the data collection categories in an illustrative scenario. In addition to well-being reports, EMAs included reports of the momentary context (activities, location and company). To obtain their context, we asked users: (a) what they were doing, (b) where they were, and (c) who they were with. Users could select multiple items from a list containing common activi- ties, locations and companies. The report had an option to report activity, location and com- pany items in a free-text format in case the items already in the list did not suffice. The language used for EMA prompts was English for English speaking countries and Spanish for Spanish speaking countries. Participants were asked to use the app for a period of 3 weeks. In compliance with the EU’s GDPR, participants were presented with a consent form describing details of the data collec- tion and purpose of the study upon installing the app. Thereafter, participants completed an onboarding process in which we captured basic demographic questions (gender, age, employ- ment status, etc.) and the Big Five personality traits using the 50 item-International Personality PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 5 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts Item Pool (IPIP) questionnaire [48]. No personal identifying information, such as name, email address etc., were collected during this step. 4.2 Data processing and features extraction 4.2.1 Contextual information extraction. As we outlined in Section 4.1, the user may report multiple activities, locations and company in each EMA. In all three context categories, the majority of EMAs reported only one instance. In particular 86.3% of the EMAs reported a single activity, 97.4% reported only one location and 92.2% a single company. For our analysis, we treated each of the activities reported in the EMA as different data instances and kept the same start and end of intervals. For simplicity, we assigned the first reported location and com- pany to all the data instances corresponding to the EMA. There were a total of 53,244 EMAs collected, which were used to generate 62,221 data instances. 4.2.2 Screen time features extraction. For the purpose of our study, we extracted the total screen time duration and percentage screen time duration in each of the data instances cor- responding to EMA reports. As first step, phone usage events were utilized to extract sessions of phone usage. We labelled the timestamp of each unlock event as the start of the session and the timestamp of the subsequent screen-off event as the end of this phone use session. Screen time duration of each session was directly computed as the number of minutes between the start and the end of the session. A total of 1,493,553 phone usage events were collected, which were transformed into 431,044 sessions. Similarly, we processed the data collected through EMAs to build the time interval that the reported contextual information belonged to. For instance, considering the exemplary EMA report from Fig 1a, the time interval would be from 12:00 PM to 1:00 PM as the recorded EMA timestamp (1pm) and the duration reported (1h). Due to the momentary nature of EMAs, the exact moment at which the activity, company or location changed was unknown (due to the fact that EMAs, by design, frequently arrive in the middle of an activity). For this reason, we only considered the time period until the user finished the EMA and we treated the logged timestamp as the end of the interval. Therefore the logged timestamp is considered the end of the activity and other reported contextual information. The start of the interval was computed by subtracting the reported duration from the EMAs timestamp. Next we proceed with extracting the screen time features per EMA report. Let us consider a person A commuting for an average of 120 minutes to get to work with a moderate total phone usage duration during this period of 60 minutes, equivalent to 50% of percentage dura- tion of screen time, against a person B commuting for an average of 10 minutes to get to work with a moderate total phone usage duration during this period of 5 minutes, equivalent to 50% of percentage duration of screen time. There is a clear difference between the two examples in terms of total duration of screen time, while the difference disappears when considering the percentage duration of screen time. This highlights the importance to consider both metrics. We computed the total screen time duration per EMA report. For each data instance, the total screen time duration was obtained by finding all the sessions of phone usage that overlapped with the EMA’s interval and by aggregating the amount of time they overlapped with that interval. This resulted in one of our target variables: the total screen time duration. The total screen time gives the absolute effect of the number of minutes spent using the phone while each specific context (e.g., activity, location, companion, time of day) is ongoing; with it, we aimed to discover the effect of spending more/less minutes using the phone while in the con- text. Secondly, we compute the proportion of time the user spent on the phone during an EMA interval by dividing the total screen time duration with the EMA reported duration. This resulted in the second target variable for our analysis: the percentage screen time duration. The PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 6 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts percentage screen time duration gives the relative effect of using the phone while in each con- text, which is important in this study as well because the self reports vary in length; with it, we aimed to discover the effect of spending a larger/smaller part of the time in the context using the phone. For instance, considering the exemplary timeline from Fig 1b, the user unlocked the phone at 11:40, locked it at 12:15, unlocked it at 12:35 and locked it again at 12:50. How- ever, since the self report was filled at 1pm and the activity duration was reported to be 1h, the time interval used to compute the screen time features is between 12pm and 1pm, which results in a total screen time duration of 30 minutes and a percentage screen time duration of 50%. 4.2.3 Context and personal characteristics categorization. From the self-reported activi- ties, companions and locations, we obtained the categories that were frequently reported to use as different contexts in our analyses. As users were allowed to input their own categories through a free text entry, we disregarded categories that have less than 100 entries in to analyse the statistical significance between groups. Moreover, we included in the categories of personal characteristics available from the onboarding questionnaire, such as gender, age and personal- ity. We expanded the contexts reported with the time-based categories in order to explore the association between phone use and well-being at certain time windows during the day, as well as during holidays. We categorised the time when the EMA was completed into: morning (from 6AM to 12PM), afternoon (from 12PM to 6PM), evening (from 6PM to 12AM), and the day as a holiday if it was a weekend or public holiday as well as if the user reported as being on holidays. Since we did not send notifications during sleeping hours, we only had 7 EMA reports filled at night (despite of not sending the reminders), hence this time interval was excluded from our analysis. With respect to personality, we utilized the answers from the Big Five questionnaire [48] and included all the five traits for our analysis (i.e., Extroversion, Agreeableness, Conscientiousness, Neuroticism and Intellect). The traits were normally dis- tributed (we tested normality with the Shapiro-Wilk normality test [49] to verify if the distri- butions were Gaussian). Further, we used the median value to split participants into two groups, e.g., into High Agreeableness versus Low Agreeableness. This methodology has been previously applied in the literature [50, 51] to analyse different clusters of individuals. The exhaustive categories considered for our analysis are shown in Table 1. Table 1. Contextual and personal characteristics categories. Group Categories Activity Browsing Internet, Commuting, Conversation, Eating, Emails, Exercise, Listening to Music, Reading, Shopping, Social Media, Studying, Taking a Nap, Taking a Shower, Waiting, Watching TV* Company Alone, Colleagues, Family, Friends, Kids, Partner, Not Alone, Unknown People Location At friends’ house, At parents’ house, Home, Restaurant, Sport facility, Street/Outdoors, Supermarket, Work*** Time Age Morning, Afternoon, Evening, Holiday 18–25y old, 26–34y old, 35–44y old Gender Male, Female Personality Extroverts, Introverts, Neurotics, Emotionally Stable, High and Low {Agreeableness, Conscientiousness, Intellect} * Instances corresponding to the Activity—Shopping category refer to both online and onsite shopping. Instances corresponding to the Activity—Working category were merged with Location—Work 1. Location—Public transport was merged with Activity—Commuting & 2. Location—University was merged with Activity—Studying, since they refer to the same contextual category and to avoid unnecessary sparsity across categories https://doi.org/10.1371/journal.pone.0284104.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 7 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts 4.3 Participants and inclusion criteria In total, the data from 921 users was collected for this study. The participants were recruited from five different countries (Chile, Colombia, Peru, Spain and the United Kingdom) by an external recruitment agency from February to August 2018. Participants were asked to use the application for a period of 3 weeks, and upon a successful completion of the study they were rewarded with a monetary compensation of EUR 40. Table 2 shows the number of individuals and data instances after each step of the inclusion criteria process. For the analysis, we first excluded participants that did not provide any EMA self-reports (i.e., who solely completed the onboarding questionnaire) and participants whose logs did not contain any sensor or unlock events for computing screen time. Furthermore, we removed the incomplete EMA records that did not contain a reported well-being score. After applying the inclusion criteria, the final dataset used for our analysis included in total 352 par- ticipants and 15,607 EMA records, out of which 344 participants provided their gender (129 Females, 215 Males) and age. In particular, over a half of the participants were aged between 26–34 (N = 237), while others belonged to various age ranges, including 35–44 (N = 70), 18–25 (N = 33), 45–54 (N = 1), 55–64 (N = 2) and 65+ (N = 1). 4.4 Methodology 4.4.1 Correlation analysis. As the first step in the analysis, we performed a correlation analysis using Spearman’s rank correlation coefficient between the well-being scores provided by the users in the EMA reports and the corresponding (passively captured) screen time mea- surements depending on the main categories defined in Table 1. 4.4.2 Multiple hypotheses correction. In order to control for the false discovery rate, we used the two-stage step-up method of Benjamini, Krieger and Yekutieli [52]. This technique is an adaptive method that controls the false discovery rate by incorporating an estimate of the number of true hypothesis based on the distribution of p-values. We applied this correction separately for the correlation p-values of Total duration and Percentage duration of screen time. We utilized a maximum threshold for the p-values computed of 0.05 for evaluating statis- tical significance for both before and after correction for multiple hypotheses (50 hypotheses evaluated, based on the categories specified in Table 1). 4.4.3 Regression analysis. Third, in order to leverage the fact that the activity, location and company contexts are features that capture the state of a user, we train linear mixed effects regression models for analysing the relationship between different aspects of phone usage and momentary subjective well-being. Mixed linear regression models with fixed effects were used to analyse the relationship between different aspects of phone usage and momentary or experi- enced happiness reports as dependent variable in the following configurations: (1) screen time features only, (2) screen time features with activities and their interaction, (3) screen time fea- tures with activities and company features and their interaction, and finally (4) screen time fea- tures with activity, company and location and their interaction. For all models, we utilised the user identifier as group to leverage that we have multiple EMA reports for each user. When Table 2. Breakdown of the inclusion criteria process. Inclusion criteria step # Users # Instances Original With EMAs With sensor events EMAs with well-being https://doi.org/10.1371/journal.pone.0284104.t002 921 873 479 352 62,221 62,221 33,610 15,607 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 8 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts using all the categories from Table 1 and their interactions, the model failed to converge, as coefficients and statistical significance were missing for less popular categories. To address this issue, we removed categories with less than 500 instances. In particular, the following catego- ries were included in the regression analysis: (a) Activity: ‘Social Media’, ‘Eating’, ‘Watching TV’, ‘Browsing Internet’, ‘Conversation’, ‘Waiting, ‘Listening to Music’, ‘Studying’, ‘Commut- ing’, ‘Taking a nap’ (b) Company: ‘Alone’, ‘Kids’, ‘Partner’, ‘Family’, ‘Friends’, ‘Colleagues’ (c) Location: ‘Home’, ‘Street/Outdoors’, ‘Work’. The variables included in the final model were the ‘Total phone usage (min)’, ‘Total phone look (min)’, all the categorical variables from (a), (b) and (c) and all interactions between ‘Total phone usage (min)’ with the categorical vari- ables from (a), (b) and (c). 5 Quantitative study: Results 5.1 General descriptive statistics The data collected included 53,244 EMAs and 1,493,553 phone usage events from a total of 921 participants. We analysed the phone usage patterns and EMA responses per hour of the day and visualized it in Fig 2. As shown in Fig 2a, our participants start using the phone between 6am and 7AM and their phone usage grows steadily until 10AM. The average number of minutes using the phone remains stable between 17.5 and 20 minutes per hour until 8PM and declines from that time until its minimum at 4AM. These patterns are in line with previ- ous findings [53–55]. Fig 2b shows a visualization on the percentage of EMA’s answered per hour of the day. Participants received notifications between 8AM and 8PM, as described in Fig 2. Patterns of phone usage and EMA reports. (a) Average minutes of phone usage per hour of the day. (b) Percentage of EMAs answered per hour of the day. (c) Distribution of reported duration in minutes in the EMAs. (d) Distribution of total phone usage. https://doi.org/10.1371/journal.pone.0284104.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 9 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts Section 4.1. The EMAs were answered quite evenly between 11AM and 8PM, with a peak at 5PM and a decline after 8PM similarly to the phone usage. Further, we analysed the duration of activities reported on EMA and the phone use while the activity was realised. As shown in Fig 2c, a large portion of reported activities lasted for 60 minutes or less, accounting for 65% of EMAs. Additionally, it shows a peak every 30 minutes that is more pronounced at each natural hour, and for a portion of activities (almost 6%), the users reported that the activities lasted 250 minutes, which was the maximum allowed to report. Fig 2d shows the distribution of total minutes of phone usage calculated based on the corresponding reported EMA intervals. The participants used their phone a median of 15 min- utes while performing the reported activity and for a period of less than 30 minutes in almost 70% of the cases. Finally, we analysed the subjective well-being of our participants, measured through reported happiness and worthwhileness scores, in the different contexts and their evolution during the week. Both happiness and worthwhileness measures have a similar distribution as shown in Fig 3a, and are highly correlated with a Spearman correlation value of 0.76 (p-value <0.001). The distribution is skewed towards larger values with a median value of 7 for both measures. During weekends, the reported happiness is significantly higher than during the rest of the week (p-value <0.01) and there is a negative trend on the happiness score from Monday to Wednesday, Wednesday being the day with the lowest score. These insights are inline with previous studies that show similar trends [56, 57]. Worthwhileness scores show a positive trend during the week until Saturday and are generally higher than happiness scores, except on Sundays, when the worthwhileness scores decrease. Previous studies have identi- fied that context plays a role in momentary subjective well-being [58, 59]. In our study, we also observed that participants reported higher subjective well-being scores on certain con- text categories with respect to others. In particular, users reported higher levels of happiness and worthwhilness while doing activities such as Exercise, Taking a shower and Having a conversation, and lower levels while Waiting, Browsing Internet/Social Media/Emails and Commuting. Sport facilities and Street/Outdoors were locations where participants reported higher subjective well-being, while they reported lower scores in Work, Home and Friends’ house. When participants were in company of their Kids or Friends their subjective well- being scores were higher and lower when Alone or with Unknown people. Please S1–S3 Figs for a complete breakdown of reported subjective well-being based on the context categories studied. Fig 3. Reported subjective well-being measures. a) Distribution of reported happiness and worthwhileness scores. (b) Average reported happiness and worthwhileness per day of the week. https://doi.org/10.1371/journal.pone.0284104.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 10 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts 5.2 Associations of phone use with subjective well-being 5.2.1 Correlation analysis. Fig 4 shows the significant Spearman correlation coefficients that are computed to assess the relationship between well-being self-reports and total duration of screen time and percentage duration of screen time. We discuss why we considered both Fig 4. Significant Spearman correlation results of total duration of screen time (a) and percentage duration of screen time (b) with reported happiness and worthwhileness. Correlations that remain significant after multiple hypothesis correction are illustrated with bold and italic font. https://doi.org/10.1371/journal.pone.0284104.g004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 11 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts metrics in Section 4.2.2. The figure is color-coded according to the valence of relationship and its effect size: darker colors indicate a stronger effect; blue indicates that a positive correlation between well-being and phone use, and orange indicates a negative correlation between well- being and phone use. Overall, we observed a significant low negative correlation between phone use and worth- whileness, yet only before correction for multiple hypotheses. Adding context as a dimension revealed several siginificant positive associations between phone use and well-being, after cor- rection for multiple hypotheses: (a) when performing certain activities such as studying, listen- ing to music, exercising, commuting, taking a nap, (b) when being in the presence of family, or partner, (c) when being outdoors, or at their friends house, (d) for introverts, or individuals with high agreeableness. Moreover, we observe negative associations between phone use and well-being: (a) when browsing internet, waiting, (b) when being on holidays, (d) for 35–44y olds, (e) across both female and male genders, and (f) being an extrovert, or neurotic, or indi- viduals with low agreeableness, or low conscientiousness. We observe that context plays a significant role when evaluating the associations between screen time and well-being. Overall, we observe more positive associations rather than nega- tive ones from correlations. Beyond activity, location, company, and time contexts, personality and age appear as well significant when considering the association between phone use and well-being. 5.2.2 Regression analysis. The results are presented in Table 3. We included only the interactions that were significant between screen time and categories. The total duration of screen time was significant (p < 0.01) in the first model only, and its effect disappeared when contexts and interactions were added. These results further suggest the importance of contexts in the association between screen time and subjective well-being. Yet, due to the design of this particular study, we refrain from making conclusions related to the causal link between screen time and the subjective well-being. The fixed effects models take into account the effect of each context and its interaction with the screen time, showing numerous significant correlations with the experienced happiness. The positive associations of fixed effects that were significant in all the models in which they were included are: the activities eating and taking a nap and being in the company of friends or kids. In contrast, the negative associations that were significant in all models are: waiting, being alone and being at work. There were some significant associations that disappeared as new context categories were added. For instance, watching TV, having a conversation or lis- tening to music were positively associated at a significant level only when company and loca- tion were not considered. These activities often involve the company of other people, which might be the reason why these effects disappear when company categories are included. Also, the company of colleagues had a significantly negative association (p < 0.01), but when loca- tion was incorporated it disappeared—possibly because the association was related to being at work where people share space with colleagues most of the time. As we are mainly interested in the interactions between screen time and their context, with the regression models we reveal five interactions between context and screen time that are sig- nificantly associated with experienced happiness. Among the significant interactions, three are negative (eating, waiting and browsing Internet) and two are positive (family, street/outdoors), illustrated in Fig 5. We discuss these interactions and their interpretation in Section 7.3. 6 Qualitative survey We conducted an online survey (N = 171) to better understand users’ perception towards the impact of smartphone usage on their well-being. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 12 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts Table 3. Regression coefficients, significance (: p< = 0.05, : p< = 0.01, : p< = 0.001) and clustered standard errors (in parentheses) for fixed effects models on experienced happiness for (1) basic model with screen time features only, (2) model with screen time and activities and their interaction, (3) model including screen time, activity, company and their interaction, and (4) model including screen time variables, all contexts (activity, company, location) and interactions. DV: Experienced happiness (1) Screen time (2) Screen time with activities and interactions (3) Screen time with activities, company and interactions (4): Screen time with activities, company location and interactions Screen time: Total phone usage (min) Total phone look (min) -0.001 (0.000)** -0.001 (0.001) -0.001 (0.001) -0.004 (0.003) 0.000 (0.001) 0.001 (0.001) Activity: Social Media Eating Watching TV Browsing Internet Conversation Waiting Listening to Music Studying Commuting Taking a nap Activity * Screen time (with interactions): Waiting * Total Duration Eating * Total Duration Browsing Internet * Total Duration Company: Alone Kids Partner Family Friends Colleagues Company * Screen time (with interactions): Family * Total duration Location: Home Street/Outdoors Work Location * Screen time (with interactions): Street/Outdoors * Total duration 0.037 (0.065) 0.317 (0.045)* 0.190 (0.054)* 0.104 (0.062) 0.396 (0.061)* -0.180 (0.066) 0.165 (0.074)* -0.104 (0.071) -0.099 (0.062) 0.325 (0.074)* 0.009 (0.067) 0.200 (0.048)* 0.086 (0.056) 0.071 (0.063) 0.127 (0.066) -0.290 (0.067)* 0.134 (0.075) -0.275 (0.073)* -0.114 (0.063) 0.280 (0.075)* -0.008 (0.002)* -0.007 (0.002)* -0.008 (0.002)* -0.003 (0.001)* -0.003 (0.001)* -0.004 (0.001)** -0.003 (0.001)* -0.347 (0.133)** 0.362 (0.154)* 0.055 (0.139) -0.207 (0.143) 0.307 (0.139)* -0.425 (0.138)** -0.004 (-0.001)* -0.003 (0.001)* -0.323 (0.133)* 0.365 (0.154)* 0.053 (0.139) -0.176 (0.143) 0.306 (0.138)* -0.263 (0.142) 0.009 (0.003) 0.008 (0.003) -0.003 (0.003) 0.000 (0.001) 0.001 (0.067) 0.185 (0.049)* 0.079 (0.058) 0.060 (0.064) 0.125 (0.066) -0.307 (0.067)* 0.139 (0.075) -0.294 (0.075)* -0.203 (0.067) 0.271 (0.076)*** -0.066 (0.045) 0.069 (0.071) -0.275 (0.069)* 0.006 (0.002)*** 13448 326 Number of observations 13981 Number of individuals 350 13981 350 https://doi.org/10.1371/journal.pone.0284104.t003 13652 350 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 13 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts Fig 5. Valence of associations for interactions between context and screen time based on the results of the most complete model (using all contextual categories and screen time variables) from the regression analysis. https://doi.org/10.1371/journal.pone.0284104.g005 6.1 Participants and inclusion criteria 171 Participants were recruited using various sources: a crowd-sourcing website, personal con- tacts, and social media. The full survey can be found in the S2 Appendix. To evaluate the qual- ity of data provided by each participant, we used three criteria: 1. The user took more than 5 minutes to answer the questionnaire, 2. The user owned a smartphone, 3. Their answer to a duplicated sanity check question. In particular, we presented the question: ’What is the highest degree or level of education you have completed or currently pursuing?’ twice in the survey; how- ever, the order of options was jumbled in the second question. If the user selected different options for the two variations of the same question, they were discarded. After filtering, we used the data from 143 participants (48 Female, 98 Males, 1 Other) for our analysis. Almost half of the participants were aged between 26–34 (N = 69), while others belonged to various age ranges, including 35–44 (N = 33), 18–25 (N = 20), 45–54 (N = 14), 55– 64 (N = 5) and 65+ (N = 2). Additionally, they lived in different continents around the world including North America/Central America (N = 55), Asia (N = 53), Europe (N = 33), Africa (N = 1) and South America (N = 1). 6.2 Perceived impact of phone use First, participants were asked to rate from 1–7 (1 being strongly negative, 4 neutral, and 7 being strongly positive) if they believed that using a smartphone had a negative or a positive impact on their well-being. Additionally, they were also asked to select from 1–7 (1 being strongly disagree) if they wanted to reduce their phone usage. On average, we observed that people believe that using smartphones has a slight positive impact on their well-being (μ = 4.8, σ = 1.35); however, they also slightly agree that would like to decrease their phone usage (μ = 4.37, σ = 1.60). This re-iterates that it is important to explore those situations in which phone usage is detrimental to well-being. Next, participants were presented with different contexts and asked if smartphone usage during these instances positively or negatively impacted their well-being. We selected the con- texts that matched the most common activities, locations and people that were obtained from the self-reports in the quantitative study. Participants were asked to rate their well-being PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 14 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts Fig 6. Box-plot of reported well-being ratings when smartphones are used in different contexts. In each box, the white circle represents the mean value, while the thick black line represents the median value of the reported well- being rating for that category. https://doi.org/10.1371/journal.pone.0284104.g006 between 1–7 (1 being very negative) when using their phone during those contexts. A box-plot of these ratings, grouped per context is presented in Fig 6. We observed that people believe that their well-being is impacted more negatively when using smartphones during eating and when they are with company (friends, family, partners, kids and colleagues), compared to other circumstances. Moreover we observe a variation in the median and mean ratings across contexts, reinforcing the key finding from the quantitative study, i.e., that context impacts the perceived effect that smartphone usage has on peoples’ subjective well-being. Some additional observations from this qualitative analysis include: 1. Most people find their phone to be a useful resource, as long as it does not interfere with their ongoing activity (83.9% voted positively, 10.5% voted neutrally and 6.3% voted negatively) 2. Most people think that their phones help them feel connected and better, when they are alone (81.2% voted positively, 13.9% voted neutrally and 4.9% voted negatively) 3. Most people think that they spend the majority of their time on their phones using social media (74.1% voted positively, 7.7% voted neutrally and 18.2% voted negatively) 4. Most people use their phones to take a break from work (66.43%) rather than for working (18.18%), while the remaining use their phones for both. 6.3 Testimonials In addition to asking users to rate their well-being during different contexts, we had two free- text questions to understand if there are certain experiences in participants’ everyday life that improve or worsen their well-being. These questions were: (Q1) ‘Are there certain experiences in your everyday life when using your smartphone improves your mood and makes you happy?’ PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 15 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts and (Q2) ‘Are there certain experiences in your everyday life when using your smartphone wors- ens your mood and makes you unhappy?’ Most users answered with short texts that were less than 20 words in length. For these qualitative responses, we performed a simple clustering exercise to explore if their answers could be grouped into overarching themes or observations. The following observations emerged from their answers: 1. Smartphones improve well-being when they are used for their original intended purpose— talking and texting with friends and family. Answers to Q1 included: “When I talk or text with a friend.”, “. . .talking to friends through social media like Facebook and WhatsApp. . .”, “Using my smartphone makes me happy when I communicate with my friends or family.”, and “Talking to my friends when I’m feeling lonely helps a lot. Especially in these tough times.”. Interestingly, many users specifically mention video calling—“Video chatting with my family members decreases my work pressure”, “Making a video call with my wife while away from home improves my mood.” and “Video calling friends/family while doing housework.”. 2. Listening to music with smartphones improves well-being: Numerous participants reported that listening to music or podcasts improves their well-being. Some answers to Q1 related to music included: “Listening to music and/or podcasts on my phone when I do exercise, sun- bathe, etc. makes me feel good.”, “Listening to music when I’m on the street or at home doing stuff around the house and want to block any surrounding sounds.”, “Being able to listen to music while on public transportation to block out noise improves my mood.” and “When I’m prepping dinner, or getting ready to do a really stressful workload I’ll turn on apple music playlist on my iPhone/smartphone” 3. Using smartphones to help with meditation, learning and exercise improves well-being: When smartphones are used as enablers to do these activities, users reported that it improves their well-being. Answers to Q1 included: “Listening to meditation, using it for exercise videos, using it to learn (i.e. a language), speaking to family and friends”, “I have several spiritual apps that send me notifications with positive messages throughout the day. They bring me good feelings and inspirations.”, “. . .Tracking my walking and cycling activities via GPS is also a great motivator (and the mapping is great for route finding)”. Some users also reported that listening to music on their smartphones encourages them to exercise. Answers included: “Listening to a favorite play list when going for a morning exercise has always made my mood awesome.” and “. . .I also listen to a lot of music, which encourages me to exercise and be active.”. 4. Social media usage both improves and worsens well-being. Answers to Q1 included: “I usu- ally use the mobile phone to see what’s coming up on social media. Although I am lonely when used, a feeling of being with others is prevalent.”, “Scrolling on social media improves my mood when I’m bored”, “When I have some alone time and I go on my phone to browse news and social media it improves my mood.” and “. . .usage of social media on my smartphone most especially twitter has always done great wonder to my mood.”. However, answers to Q2 included: “. . .Using Instagram also worsens my mood, I can’t help and use it but it makes you compare your life with others’ and that may not always be so positive.”, “When I scroll through Facebook for 30 minutes, and realize I’ve just wasted 30 minutes” and “When I see something on social media that I don’t like too much and it affects me.”. 5. Users reported that playing video games both improved and worsened their well-being. Q1 answers included: I play video games in my smartphone often and it helps me to be in a happy mood. and “Playing video games on my smartphone when I am less busy improves my PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 16 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts mood..”. Q2 answers included: “Sometimes by playing the games make me mood upset.”, “Playing longtime games makes me stressful” and “playing games”. 6. Many users reported that using phones when you’re physically with friends and family worsens their well-being. Answers to Q2 included: “Using phone in the evenings when my husband has finished work. We both use our phones so don’t talk. Also—WhatsApp messages can be so distracting. There is so much pressure to respond to all the messages you receive from various platforms”, “. . .Using it while out with friends, as then we won’t have really meaningful conversations.”, “When my wife used her mobile phone and she doesn’t respond to me. That time I get irritated.” and “when I have family time, and a really urgent text comes through on my iPhone, family members get upset with me because my time is not focused on them, and then I become upset with myself and my phone usage.”. 7. Users reported worse well-being when smartphones enable work life to invade personal and social life. Q2 answers included: “Non-stop messages and notifications from work chats and programs, work almost in each program we use (WhatsApp, Telegram, Facebook, Insta- gram, Notion, so on) -> no personal space and no silence. Even on holidays you are online and reachable. No respect to personal time (even at night is fine to write)”, “Using my phone to check work emails worsens my mood because my partner gets upset, but I can’t avoid it.” and “When I use it for work and it sometimes makes me feel as if I’m always on. . .”. 8. Users also don’t like to read bad news on their mobile devices, indicated by these Q2 answers: “Reading news articles on my phone about sad things going on in the world puts me in a bad mood”, “Seeing bad news”, “when there are same stories like Facebook news and other things nothing to cheer you up and you miss something real, that time it worsen the experience.”, “some times WhatsApp messages from groups about some political nature make me so uncomfortable, because i know the reality of that messages and they are created only from the frustration or agony against the opposite groups”. 7 Discussion 7.1 Effect size of screen time on subjective well-being Both the correlation and regression analyses in this study suggest a weak association between screen time and subjective well-being. This is inline with recent reports, such as [41, 42] which argued that there is a lack of strong evidence on the detrimental effects of smartphones use on our mental health [41, 42] and that previous studies and systematic reviews overestimated the negative associations [4]. Importantly, our study highlighted the role of daily contexts in the associations between the phone use and subjective well-being. In other words, previous research may be wrong in analysing the two in vacuum—in reality, it appears that it is about the interplay between screen time and contexts that drive positive and negative associations with well-being. 7.2 Accumulated effect is negative, yet not all screen time is deleterious Consistent with the previous literature, we found that the overall phone use association with subjective well-being is a negative one, regardless of context (see Fig 4, and regression coeffi- cient in (1) in Table 3). However, the effect disappeared in the regression analysis when con- sidering different contextual information. 7.3 Interactions between screen time and context: Regression analysis The regression results show that in the configurations where contextual features were added, the screen time features lost their significance. We revealed five interactions that remained PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 17 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts significant when adding both context and screen time features. The first is that while eating has a significantly positive correlation (p < 0.001) with experienced happiness, the association of its interaction with screen time and happiness is significantly negative (p < 0.05). This indi- cates that eating is positively associated with experienced happiness, but the use of the phone while eating has a negative component. Moreover, we’ve found that using the screen time in the presence of company produces a weak positive association with experienced happiness. We hypothesised that this may happen in the presence of people with whom the user is com- fortable with, such as partner and family (e.g., leisure time before bed time). We encourage further research to dive and dissect these results, as it would be interesting to discover for instance in which cultures this happens, as well as in the presence of which company (since partner may be labelled as family as well). While the previous results are not unexpected, interestingly, we hypothesised that while waiting, the user may find screen time as a positive activity (e.g., for communicating with oth- ers, for social media, for productive activities). Surprisingly, we’ve found a strong evidence (p < 0.001) that the using the phone during waiting is negatively associated with experienced happiness (both in correlation and regression analyses). One possible explanation for this is that not all idle time should be filled with screen time, and one should preserve this time for reflection. Another possible interpretation of this result is that when we have a lower experi- enced happiness while waiting, we use the phone more. Finally, we discovered the the use of phone in the outdoors or on the street was weakly but significantly positively associated with experienced happiness. This may be due to facilitator apps (e.g., offering directions, reviews of nearby places, etc.), yet we hypothesised that using the phone in outdoor activities would have the opposite effect. We encourage further research in this line to discover which outdoor activities were positively associated with experienced happiness. 7.4 Perceived impact from qualitative study against observed associations from quantitative study Our main finding is that indeed people report that context plays a big role into whether they perceive the impact of the phone use on well-being as positive or negative. Moreover, while we did not explore causal links in the quantitative study analysis, the qualitative study allowed us to explore suggestions of causal links. When diving deeper into people’s testimonials, it becomes evident that the majority of statements include at least one context that influences the association between phone use and well-being. For instance, users reported that when being on holidays or with family, receiving a text on their smartphone related to work causes a decrease in their well-being, shedding a light on the importance of considering the context when evaluating the impact of phone usage on well-being. Moreover, we observe that certain smartphone uses are perceived to be both positive and negative, such as using the phone for social media alone, versus scrolling for a long time, which are observable passively only when combined with available contextual information, in this case activity and company, and duration. We further compared the perceived effects of smartphone use on well-being from the quali- tative survey responses with the findings from the quantitative study. Some differences and similarities arise. For instance, we observed that people perceive that their well-being is nega- tively affected when they are with company (friends, family, partners, kids and colleagues), while it is positively impacted when they listen to music and exercise. While the quantitative study indicates that well-being does improve when people listen to music or exercise, it also shows that there is a positive association between well-being and phone use when people PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 18 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts report being in the company of family or their partner. Moreover, qualitative data does give us a layer of depth that is difficult to obtain from quantitative data. As an example, people pro- vided very detailed responses under which circumstances certain smartphone uses were posi- tive or negative. Using the free-text entries from our qualitative data, we noted that playing video games or using social media both improves and worsens mood, and people were very descriptive of the contexts when this happens. Bringing together both subjective views and objective measurements allowed us to further highlight the importance of considering context when evaluating smartphone usage impact on well-being. 7.5 Differences in associations between the two considered dimensions of well-being, happiness and worthwhileness We observed a Spearman correlation coefficient of 0.76 (p < 0.001) between happiness and worthwhileness. The high correlation between the two metrics suggests that the two dimen- sions of well-being do not differ largely from each other for the sample population under observation. Indeed, we did not find a single difference in direction of association between screen time and one dimension of well-being against the other, i.e. there are no associations that are positive with happiness and negative with worthwhileness or the opposite. In general, the insights observed for happiness and worthwhileness are in line and coincide when consid- ering the direction of association with screen time. Further research could dive deeper to explore whether there are any differences between the well-being dimensions when consider- ing individual differences in well-being. 7.6 Duration against relative percentage in various contexts In particular, we are interested in answering whether there are any differences in associations when looking at the total duration of screen time during an EMA episode compared to the per- centage of screen time during an episode. We observed a Pearson correlation coefficient of 0.51 (p < 0.001) between percentage duration and total duration of screen time. In general, the insights observed from the total duration of screen time and percentage duration of screen time are in line and coincide when considering the direction of association. The strength of the association and the significance may vary across the two depending on the context (e.g., Extroverts in Fig 4). 7.7 Implications The main contribution of our study stems from demonstrating how a set of factors influence the complex relationship between phone use and well-being beyond the mere screen time as per the previous literature. By combining qualitative and quantitative analysis, our study con- tributes to the HCI community by shedding light on the interrogations of recent reviews as well as on the conclusions in the popular press typically categorical about either the positive or negative impact of screen time (mostly the latter, which makes the headlines). This investiga- tion’s results show how context plays a big role into whether the associations with reported well-being are positive or negative. There is much potential for further innovation of designing for well-being as well as in the space of digital phenotyping and individual tracking apps that give insights into everyday screen time and device use. Church et al. [60] say that improved user interfaces and user expe- rience is likely to result in better data collection in device tracking studies. Given that most col- lection apps nowadays rely on users opening the app and engaging with it to avoid missing data, especially for Android phones, displaying insights into users’ contextualized screen time and well-being could engage users towards a more accurate passive data collection, as it would PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 19 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts offer information back to the user, complementary to collection only. Moreover, people report that context plays a big role on whether they perceived phone use as negative or positive for their well-being. Thus, we suggest that HCI’s interest in individual data collection tracking apps should not be limited only by passive data collection, but should engage users and provide insights such that they make sense of their screen time beyond device and application usage, to better reflect on how to organize their everyday lives. Moreover, designing mobile phones to raise awareness to users of how the context of phone use plays a role into their individual well-being could potentially lead to an improvement in their well-being. We envisage a future application, where depending on the valence of the asso- ciation, the user would be prompted to reflect on whether the phone use in that (passively detected or reported) context is necessary at that time and requires their full attention. Such design maintains users’ control on their phone use, but at the same time prompts the user on reflecting whether the potential negative impact on their well-being can be avoided. Such prompts could be particularly targeting those situational contexts that were shown to have a negative association with well-being in this study. Finally, further innovation for designing virtual interactions through mobile phones during various contextual situations could be explored, such as leveraging positive or negative rewards depending on the context. One such example is given in [61] where the context at play is hav- ing a meal. The authors suggest an interactive game between the individuals at the table shar- ing a meal where they are given a positive reward when not engaging with the mobile phone, (e.g., an ice cream sundae badge for not going online or using social media during a meal). Similarly, the insights from this study can be used to guide the design of such interactive games where rewards are given only in situational contexts that have been shown to play a sig- nificant role in the association of screen time with well-being, and we suggest that the rewards are tuned based on the valence and significance of the association. 7.8 Limitations One limitation of our study is that we rely on self-reported well-being measures. Recent studies exploit latest developments in automated emotions detection, such as the study by Sarsen- bayeva et al. [40] which uses the Affectiva Emotion SDK [62] to automatically detect emotions from facial expressions and map them to passive smartphone sensing measurements. The study also collects self-reports to triangulate against and check the accuracy of the Affectiva data. We envisage a future study where the collection of target well-being and emotional states can be done continuously and passively in time as the users use their smartphone to avoid rely- ing on self-reports for the target variables completely. Another limitation of this study is the fact that the screen time is measured only during the duration reported in the EMAs collected in order to obtain a mapping between context and smartphone use. Therefore, the findings can be affected by users’ memory, as the duration of an activity may be reported only approximately. This mainly affects the raw total duration screen time measurement, where we measure the amount of time the user spent on the phone during the context reported. We partly addressed this limitation by computing the percentage screen time measurement, as it is relative to the duration reported in the EMA. A more advanced potential solution to this problem is to explore acquiring contextual information in a passive manner as well. An additional limitation is that our study was not designed to explore causal effects on the EMA’s. To study the causal links, the study would have to have captured experienced happi- ness before and while performing the activities reported in the EMA’s to see how experienced happiness changes according to screen time usage within each context. Such study would be PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 20 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts very difficult to put in practice in the wild as it requires knowing when the participant is going to start an activity and prompting the participant for self-reports while the activity is ongoing. For that reason, the analyses performed in this paper present associations between experienced happiness and screen time together with context, and we refrain from making firm claims on the causal effects of screen time on experienced happiness. Finally, we explored users’ perception of screen time impact in various contexts using a different sample population, mainly due to the initial study being ran for a wider list of hypoth- eses and completed more than a year ago. We were still interested in how users perceive smart- phone use in different contexts and therefore ran a separate qualitative survey to enrich this study from this dimension. Another limitation of this survey is that the recruitment was done through social media and similar means, and thus the results may suffer from this selection bias. 7.9 Future research directions In our study participants belonged to a plethora of countries, both in the quantitative and qual- itative studies, as an extra consideration was given to ensure that a diverse cross-country sam- ple is collected. Yet our study was not designed to evaluate the cross-cultural aspect of the phone use relationship with well-being. We encourage future studies with a balanced sample across more cultures covering different continents to explore the role of culture in the relation- ship between screen time and well-being. Moreover, even though we attempted to remove the subjectivity by collecting passively the screen time, we still relied on active data for reporting activities duration and self-reported well-being measures. We encourage a future study to look at leveraging latest methods in emo- tional states detection and context (activities, companies and location are a few of the contexts that have been explored and promising methods emerged for their passive inference in [63– 67], to name a few). Subjective reports are still important as we can see from the testimonials from the qualita- tive study, which reveal that even the same context can trigger different responses and lead to both worsening or improving a user’s well-being based on other factors, such as how the user is currently feeling. Reports that social media both improves (when feeling lonely or bored) and worsens (when wasting time) well-being highlight how useful it is to consider a persons well-being in the first place, when diving into the valence of the associations between well- being and screen time. Finally, we encourage work in this area to dive deeper into deviant individuals and deviant screen time use associations with well-being. In particular, it would be interesting to discover whether happy or unhappy individuals (for instance as classified based on their individual SWB set-points) rate differently their SWB in different phone use contexts. Moreover, we encourage future work to explore the “threshold” for problematic screen time use, and the associations between SWB and screen time for deviant individuals (for instance as defined based on their screen time use). Lastly, we encourage future work to explore the associations between SWB and deviant screen time uses (for instance as defined per individual based on screen time distribution) in different contexts. 8 Conclusion Smartphones have become a central part of our daily lives, regardless of the controversy around its effects on our well-being. Prior studies have argued in plentiful numbers that using smartphones is associated with poorer well-being [17, 24] and mental health [2, 19]. Yet recent studies [4, 43] challenged these claims and argued that there is very little evidence on the PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 21 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts negative effects of smartphones on our well-being and that previous studies have overesti- mated the effects. Diving deeper into the associations of smartphone use with well-being, this study reveals other important factors to consider, such as surrounding context (such as activ- ity, location, company) and personal characteristics. We present detailed results under which factors phone use is associated with a positive increase in well-being or a negative decrease in well-being, leveraging passive smartphone sensing data coupled with active user reports on their context and well-being. Finally, we compare these results with the perceived effects of smartphone use on well-being, extracted from a subsequent qualitative study. Our results emphasize that context plays a big role into whether the associations with subjective well-being are positive or negative. Moreover, positive associations between screen time and well-being were discovered, along with negative ones, however all in all the strength of these associations was weak. Our paper highlights the complexity of this problem and reveals an obscured hidden part of the screen time iceberg. Supporting information S1 Fig. Box plot of subjective well-being in different contexts per activity. (PDF) S2 Fig. Box plot of subjective well-being in different contexts per company. (PDF) S3 Fig. Box plot of subjective well-being in different contexts per location. (PDF) S1 Appendix. Subjective well-being in different contexts. (PDF) S2 Appendix. Survey used for qualitative study. (PDF) Acknowledgments We thank Prof. Paul Dolan, who played a crucial role in designing and setting up Reflections —the larger study from which we derived data for this paper. We also extend our appreciation to Telefonica Alpha (now Koa Health) and the London School of Economics and Political Sci- ence for their generous funding and unwavering support throughout this study. Author Contributions Conceptualization: Teodora Sandra Buda, Aleksandar Matic. Data curation: Teodora Sandra Buda, Mohammed Khwaja, Roger Garriga. Formal analysis: Teodora Sandra Buda, Mohammed Khwaja, Roger Garriga. Funding acquisition: Aleksandar Matic. Investigation: Teodora Sandra Buda, Mohammed Khwaja, Aleksandar Matic. Methodology: Teodora Sandra Buda, Mohammed Khwaja, Roger Garriga, Aleksandar Matic. Project administration: Aleksandar Matic. Resources: Aleksandar Matic. Supervision: Aleksandar Matic. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 22 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts Validation: Teodora Sandra Buda, Aleksandar Matic. Visualization: Teodora Sandra Buda, Mohammed Khwaja, Roger Garriga. Writing – original draft: Teodora Sandra Buda, Mohammed Khwaja, Roger Garriga, Alek- sandar Matic. Writing – review & editing: Teodora Sandra Buda, Roger Garriga, Aleksandar Matic. References 1. Sa´ nchez-Martı´nez M, Otero A. Factors associated with cell phone use in adolescents in the community of Madrid (Spain). CyberPsychology & Behavior. 2009; 12(2):131–137. https://doi.org/10.1089/cpb. 2008.0164 PMID: 19072078 2. Rohani DA, Faurholt-Jepsen M, Kessing LV, Bardram JE. Correlations between objective behavioral features collected from mobile and wearable devices and depressive mood symptoms in patients with affective disorders: systematic review. JMIR mHealth and uHealth. 2018; 6(8):e165. https://doi.org/10. 2196/mhealth.9691 PMID: 30104184 3. Thome´e S, Ha¨renstam A, Hagberg M. Mobile phone use and stress, sleep disturbances, and symptoms of depression among young adults-a prospective cohort study. BMC public health. 2011; 11(1):66. https://doi.org/10.1186/1471-2458-11-66 PMID: 21281471 4. Orben A. Teenagers, screens and social media: a narrative review of reviews and key studies. Social psychiatry and psychiatric epidemiology. 2020; p. 1–8. PMID: 31925481 5. Przybylski AK, Weinstein N. A large-scale test of the goldilocks hypothesis: quantifying the relations between digital-screen use and the mental well-being of adolescents. Psychological Science. 2017; 28 (2):204–215. https://doi.org/10.1177/0956797616678438 PMID: 28085574 6. Odgers C. Smartphones are bad for some teens, not all; 2018. 7. Katevas K, Arapakis I, Pielot M. Typical phone use habits: Intense use does not predict negative well- being. In: Proceedings of the 20th International Conference on Human-Computer Interaction with Mobile Devices and Services; 2018. p. 1–13. 8. Lin YH, Lin YC, Lee YH, Lin PH, Lin SH, Chang LR, et al. Time distortion associated with smartphone addiction: Identifying smartphone addiction via a mobile application (App). Journal of psychiatric research. 2015; 65:139–145. https://doi.org/10.1016/j.jpsychires.2015.04.003 PMID: 25935253 9. Vanden Abeele M, Beullens K, Roe K. Measuring mobile phone use: Gender, age and real usage level in relation to the accuracy and validity of self-reported mobile phone use. Mobile Media & Communica- tion. 2013; 1(2):213–236. https://doi.org/10.1177/2050157913477095 10. Kobayashi T, Boase J. No such effect? The implications of measurement error in self-report measures of mobile communication use. Communication Methods and Measures. 2012; 6(2):126–143. https:// doi.org/10.1080/19312458.2012.679243 11. Hiniker A, Suh H, Cao S, Kientz JA. Screen time tantrums: how families manage screen media experi- ences for toddlers and preschoolers. In: Proceedings of the 2016 CHI conference on human factors in computing systems; 2016. p. 648–660. 12. Mehrotra A, Tsapeli F, Hendley R, Musolesi M. MyTraces: Investigating correlation and causation between users’ emotional states and mobile phone interaction. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies. 2017; 1(3):1–21. 13. Peters D, Calvo RA, Ryan RM. Designing for motivation, engagement and wellbeing in digital experi- ence. Frontiers in psychology. 2018; 9:797. https://doi.org/10.3389/fpsyg.2018.00797 PMID: 29892246 14. Steptoe A, Deaton A, Stone AA. Subjective wellbeing, health, and ageing. The Lancet. 2015; 385 (9968):640–648. https://doi.org/10.1016/S0140-6736(13)61489-0 15. Dolan P. Happiness by design: Finding pleasure and purpose in everyday life. Penguin UK; 2014. 16. Benson T, Sladen J, Liles A, Potts HW. Personal Wellbeing Score (PWS)—a short version of ONS4: development and validation in social prescribing. BMJ open quality. 2019; 8(2):e000394. https://doi.org/ 10.1136/bmjoq-2018-000394 PMID: 31206049 17. Volkmer SA, Lermer E. Unhappy and addicted to your phone?–Higher mobile phone use is associated with lower well-being. Computers in human behavior. 2019; 93:210–218. https://doi.org/10.1016/j.chb. 2018.12.015 18. Domingues-Montanari S. Clinical and psychological effects of excessive screen time on children. Jour- nal of paediatrics and child health. 2017; 53(4):333–338. https://doi.org/10.1111/jpc.13462 PMID: 28168778 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 23 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts 19. 20. Twenge JM, Campbell WK. Associations between screen time and lower psychological well-being among children and adolescents: Evidence from a population-based study. Preventive medicine reports. 2018; 12:271–283. https://doi.org/10.1016/j.pmedr.2018.10.003 PMID: 30406005 Twenge JM, Joiner TE, Rogers ML, Martin GN. Increases in depressive symptoms, suicide-related out- comes, and suicide rates among US adolescents after 2010 and links to increased new media screen time. Clinical Psychological Science. 2018; 6(1):3–17. https://doi.org/10.1177/2167702617723376 21. Hoare E, Milton K, Foster C, Allender S. The associations between sedentary behaviour and mental health among adolescents: a systematic review. International journal of behavioral nutrition and physi- cal activity. 2016; 13(1):108. https://doi.org/10.1186/s12966-016-0432-4 PMID: 27717387 22. Yang F, Helgason AR, Sigfusdottir ID, Kristjansson AL. Electronic screen use and mental well-being of 10–12-year-old children. The European Journal of Public Health. 2013; 23(3):492–498. https://doi.org/ 10.1093/eurpub/cks102 PMID: 22874734 23. Maras D, Flament MF, Murray M, Buchholz A, Henderson KA, Obeid N, et al. Screen time is associated with depression and anxiety in Canadian youth. Preventive medicine. 2015; 73:133–138. https://doi. org/10.1016/j.ypmed.2015.01.029 PMID: 25657166 24. Allen MS, Vella SA. Screen-based sedentary behaviour and psychosocial well-being in childhood: Cross-sectional and longitudinal associations. Mental Health and Physical Activity. 2015; p. 41–47. https://doi.org/10.1016/j.mhpa.2015.10.002 25. Babic MJ, Smith JJ, Morgan PJ, Eather N, Plotnikoff RC, Lubans DR. Longitudinal associations between changes in screen-time and mental health outcomes in adolescents. Mental Health and Physi- cal Activity. 2017; 12:124–131. https://doi.org/10.1016/j.mhpa.2017.04.001 26. Hinkley T, Verbestel V, Ahrens W, Lissner L, Molnar D, Moreno LA, et al. Early childhood electronic media use as a predictor of poorer well-being: a prospective cohort study. JAMA pediatrics. 2014; 168 (5):485–492. https://doi.org/10.1001/jamapediatrics.2014.94 PMID: 24639016 27. Kim HHs. The impact of online social networking on adolescent psychological well-being (WB): a popu- lation-level analysis of Korean school-aged children. International Journal of Adolescence and Youth. 2017; 22(3):364–376. https://doi.org/10.1080/02673843.2016.1197135 28. Kross E, Verduyn P, Demiralp E, Park J, Lee DS, Lin N, et al. Facebook use predicts declines in subjec- tive well-being in young adults. PloS one. 2013; 8(8):e69841. https://doi.org/10.1371/journal.pone. 0069841 PMID: 23967061 29. Schmiedeberg C, Schro¨der J. Leisure activities and life satisfaction: An analysis with German panel data. Applied Research in Quality of Life. 2017; 12(1):137–151. https://doi.org/10.1007/s11482-016- 9458-7 30. Shakya HB, Christakis NA. Association of Facebook use with compromised well-being: A longitudinal study. American journal of epidemiology. 2017; 185(3):203–211. PMID: 28093386 31. 32. Tromholt M. The Facebook experiment: Quitting Facebook leads to higher levels of well-being. Cyberp- sychology, behavior, and social networking. 2016; 19(11):661–666. https://doi.org/10.1089/cyber.2016. 0259 PMID: 27831756 Lambert J, Barnstable G, Minter E, Cooper J, McEwan D. Taking a one-week break from social media improves well-being, depression, and anxiety: a randomized controlled trial. Cyberpsychology, Behav- ior, and Social Networking. 2022;. https://doi.org/10.1089/cyber.2021.0324 PMID: 35512731 33. Dwyer RJ, Kushlev K, Dunn EW. Smartphone use undermines enjoyment of face-to-face social interac- tions. Journal of Experimental Social Psychology. 2018; 78:233–239. https://doi.org/10.1016/j.jesp. 2017.10.007 34. Kushlev K, Proulx JD, Dunn EW. Digitally connected, socially disconnected: The effects of relying on technology rather than other people. Computers in Human Behavior. 2017; 76:68–74. https://doi.org/ 10.1016/j.chb.2017.07.001 35. Rivron V, Khan MI, Charneau S, Chrisment I. Refining smartphone usage analysis by combining crowd- sensing and survey. In: 2015 IEEE International Conference on Pervasive Computing and Communica- tion Workshops (PerCom Workshops). IEEE; 2015. p. 366–371. 36. Larson LR, Szczytko R, Bowers EP, Stephens LE, Stevenson KT, Floyd MF. Outdoor time, screen time, and connection to nature: Troubling trends among rural youth? Environment and Behavior. 2019; 51(8):966–991. https://doi.org/10.1177/0013916518806686 37. Dumuid D, Olds T, Lewis LK, Martin-Ferna´ndez JA, Katzmarzyk PT, Barreira T, et al. Health-related quality of life and lifestyle behavior clusters in school-aged children from 12 countries. The Journal of pediatrics. 2017; 183:178–183. https://doi.org/10.1016/j.jpeds.2016.12.048 PMID: 28081885 38. Harmon E, Mazmanian M. Stories of the Smartphone in everyday discourse: conflict, tension & instabil- ity. In: Proceedings of the SIGCHI conference on human factors in computing systems; 2013. p. 1051– 1060. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 24 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts 39. Google’s Activity Recognition Application; 2017. Available from: http://developer.android.com/training/ location/activity-recognition.html. 40. Sarsenbayeva Z, Marini G, van Berkel N, Luo C, Jiang W, Yang K, et al. Does smartphone use drive our emotions or vice versa? A causal analysis. In: Proceedings of the 2020 CHI Conference on Human Fac- tors in Computing Systems; 2020. p. 1–15. 41. Orben A, Przybylski AK. Screens, teens, and psychological well-being: evidence from three time-use- diary studies. Psychological science. 2019; 30(5):682–696. https://doi.org/10.1177/0956797619830329 PMID: 30939250 42. Kaye LK, Orben A, Ellis DA, Hunter SC, Houghton S. The conceptual and methodological mayhem of “screen time”. International Journal of Environmental Research and Public Health. 2020; 17(10):3661. https://doi.org/10.3390/ijerph17103661 PMID: 32456054 43. Orben A, Przybylski AK. The association between adolescent well-being and digital technology use. Nature Human Behaviour. 2019; 3(2):173–182. https://doi.org/10.1038/s41562-018-0506-1 PMID: 30944443 44. Przybylski AK, Weinstein N. Digital Screen Time Limits and Young Children’s Psychological Well- Being: Evidence From a Population-Based Study. Child Development. 2019; 90(1):e56–e65. https:// doi.org/10.1111/cdev.13007 PMID: 29235663 45. Valkenburg PM, Peter J. Social consequences of the Internet for adolescents: A decade of research. Current directions in psychological science. 2009; 18(1):1–5. https://doi.org/10.1111/j.1467-8721.2009. 01595.x 46. Granic I, Lobel A, Engels RC. The benefits of playing video games. American psychologist. 2014; 69 (1):66. https://doi.org/10.1037/a0034857 PMID: 24295515 47. Shiffman S. Designing protocols for ecological momentary assessment. The science of real-time data capture: Self-reports in health research. 2007; p. 27–53. 48. Goldberg LR, Johnson JA, Eber HW, Hogan R, Ashton MC, Cloninger CR, et al. The international per- sonality item pool and the future of public-domain personality measures. Journal of Research in person- ality. 2006; 40(1):84–96. https://doi.org/10.1016/j.jrp.2005.08.007 49. Ghasemi A, Zahediasl S. Normality tests for statistical analysis: a guide for non-statisticians. Interna- tional journal of endocrinology and metabolism. 2012; 10(2):486. https://doi.org/10.5812/ijem.3505 PMID: 23843808 50. Vinciarelli A, Mohammadi G. A survey of personality computing. IEEE Transactions on Affective Com- puting. 2014; 5(3):273–291. https://doi.org/10.1109/TAFFC.2014.2341252 51. Khwaja M, Vaid SS, Zannone S, Harari GM, Faisal AA, Matic A. Modeling personality vs. modeling per- sonalidad: In-the-wild mobile data analysis in five countries suggests cultural impact on personality models. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies. 2019; 3(3):1–24. https://doi.org/10.1145/3351246 52. Benjamini Y, Krieger AM, Yekutieli D. Adaptive linear step-up procedures that control the false discov- ery rate. Biometrika. 2006; 93(3):491–507. https://doi.org/10.1093/biomet/93.3.491 53. Falaki H, Mahajan R, Kandula S, Lymberopoulos D, Govindan R, Estrin D. Diversity in Smartphone Usage. MobiSys ‘10. New York, NY, USA: Association for Computing Machinery; 2010. p. 179–194. Available from: https://doi.org/10.1145/1814433.1814453. 54. Bo¨ hmer M, Hecht B, Scho¨ ning J, Kru¨ger A, Bauer G. Falling asleep with Angry Birds, Facebook and Kindle: A large scale study on mobile application usage; 2011. p. 47–56. 55. Oliver E. The challenges in large-scale smartphone user studies. 2010; p. 5. 56. Stone AA, Schneider S, Harter JK. Day-of-week mood patterns in the United States: On the existence of ‘Blue Monday’, ‘Thank God it’s Friday’ and weekend effects. The Journal of Positive Psychology. 2012; 7(4):306–314. https://doi.org/10.1080/17439760.2012.691980 57. Helliwell JF, Wang S. Weekends and Subjective Well-Being. National Bureau of Economic Research; 2011. 17180. Available from: http://www.nber.org/papers/w17180. 58. Csikszentmihalyi M, Hunter J. Happiness in Everyday Life: The Uses of Experience Sampling. Journal of Happiness Studies. 2003; 4:185–199. https://doi.org/10.1023/A:1024409732742 59. Csikszentmihalyi M, Wong M. The Situational and Personal Correlates of Happiness: A Cross-National Comparison. Flow and the Foundations of Positive Psychology: The Collected Works of Mihaly Csiks- zentmihalyi. 2014; p. 69–88. 10.1007/978-94-017-9088-8 60. Church K, Ferreira D, Banovic N, Lyons K. Understanding the challenges of mobile phone usage data. In: Proceedings of the 17th International Conference on Human-Computer Interaction with Mobile Devices and Services; 2015. p. 504–514. 61. Moser C, Schoenebeck S, Reinecke K. Technology at the Table: Attitudes about Mobile Phone Use at Mealtimes. Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems. 2016;. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 25 / 26 PLOS ONETwo edges of the screen: Associations between screen time and subjective well-being in everyday contexts 62. Affectiva Emotion SDK; 2020. Available from: https://www.affectiva.com/what/products/. 63. Penha Natal Id, Avellar Campos Cordeiro Rd, Garcia ACB. Activity recognition model based on GPS data, points of interest and user profile. In: International Symposium on Methodologies for Intelligent Systems. Springer; 2017. p. 358–367. 64. Javed AR, Faheem R, Asim M, Baker T, Beg MO. A smartphone sensors-based personalized human activity recognition system for sustainable smart cities. Sustainable Cities and Society. 2021; 71:102970. https://doi.org/10.1016/j.scs.2021.102970 65. Meegahapola L, Ruiz-Correa S, Gatica-Perez D. Alone or With Others? Understanding Eating Epi- sodes of College Students with Mobile Sensing. In: 19th International Conference on Mobile and Ubiqui- tous Multimedia. MUM’20. New York, NY, USA: Association for Computing Machinery; 2020. p. 162– 166. Available from: https://doi.org/10.1145/3428361.3428463. 66. Lau BPL, Hasala MS, Kadaba VS, Thirunavukarasu B, Yuen C, Yuen B, et al. Extracting point of interest and classifying environment for low sampling crowd sensing smartphone sensor data. In: 2017 IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Work- shops). IEEE; 2017. p. 201–206. 67. Assem H, Xu L, Buda TS, O’Sullivan D. Spatio-Temporal Clustering Approach for Detecting Functional Regions in Cities. In: 2016 IEEE 28th International Conference on Tools with Artificial Intelligence (ICTAI); 2016. p. 370–377. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284104 April 26, 2023 26 / 26 PLOS ONE
10.1371_journal.pone.0279220	RESEARCH ARTICLE Do environmental, social, and governance scores improve green innovation? Empirical evidence from Chinese-listed companies Chunlian Zhang1,2‡, Danni ChenID 3‡* 1 School of Economics and Trade, Nanchang Institute of Technology, Nanchang, Jiangxi, China, 2 The Water Economy and Water Rights Research Center, Nanchang Institute of Technology, Nanchang, Jiangxi, China, 3 School of Finance, Jiangxi University of Finance and Economics, Nanchang, China ‡ DC and CZ have contributed equally to this work and share first authorship. * 2202121925@stu.jxufe.edu.cn Abstract Environmental, social, and governance (ESG) has become a buzzword in investment circles as ecological damage and climate warming occur. ESG assessment is one of the important institutions of the green financial system, which plays a significant part in boosting corporate green development. We use the number of green patent applications and green patent cita- tions to measure corporate green innovation and analyze the micro-green effects of the ESG score system using the panel fixed effects models, which means that we explore the impact of the ESG scores on corporate green innovation performance, the specific mecha- nism of this effect, and the asymmetry of this impact under different moderation effects by using Chinese listed A-shares in Shanghai and Shenzhen from 2010–2019 as our research sample. We find that ESG positively affects corporate green innovation; the higher the ESG evaluation, the more it improves firms’ green innovation performance. The promotion effect is reflected quantitatively and qualitatively and remains valid after several robustness tests. In addition, the contribution of ESG to corporate green innovation is achieved through two main paths improving corporate investment efficiency and government-enterprise relations. Corporate black attributes inhibit the contribution of ESG to green innovation, while green attributes strengthen the contribution of ESG to green innovation performance. Our study demonstrates the importance of corporate participation in environmental, social, and gover- nance practices for corporate green innovation, which is beneficial for achieving win-win environmental, social, and economic results. Furthermore, our research completes the research on the effects of corporate green performance and green finance. It can provide empirical references for promoting corporate green development and improving the ESG evaluation system. Introduction Green innovation mainly emphasizes the sustainability of innovation, which describes novel products, processes, and techniques that might minimize the dangers to the environment, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Zhang C, Chen D (2023) Do environmental, social, and governance scores improve green innovation? Empirical evidence from Chinese-listed companies. PLoS ONE 18(5): e0279220. https://doi.org/10.1371/journal. pone.0279220 Editor: Jose´ Antonio Clemente Almendros, Universidad Internacional de La Rioja, SPAIN Received: December 2, 2022 Accepted: May 3, 2023 Published: May 25, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0279220 Copyright: © 2023 Zhang, Chen. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. All files are available from PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 1 / 24 PLOS ONEthefigshare database (10.6084/m9.figshare. 22578787 10.6084/m9.figshare.22578808). Funding: The authors acknowledge the financial support from the project of the Water Economy and Water Rights Research Center, a school-level platform in Nanchang Institute of Technology: An empirical study on the Microeconomics of ESG performance under the ’Dual-carbon’ vision (22ZXZD01). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Do Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies pollution, and resource consumption throughout their life cycles [1–3]. Moreover, green inno- vation is an essential form for companies to practice the environmental, social, and governance concepts and an important tool to drive the green transformation of enterprises [4]. Green innovations for companies are low-carbon, energy efficient, and effective [5], but it also has characteristics such as long-term riskiness, public goods, and positive environmental externali- ties [6]. With the increasing economic globalization and industrialization, the natural world is subject to significant adverse impacts. Environmental pollution problems are becoming more prominent, severe climate problems are becoming more powerful, and green innovation may be essential to reconcile the contradictions between man and nature [7, 8]. Several elements drive corporate green innovation performance, as businesses need to maintain a competitive edge and increase corporate value. The literature has classified the fac- tors influencing green innovation into four categories. The first is market factors, including market pressure, green consumer demand, capital market opening, and environmental label- ing certification [9–14]. Corporate green innovation will be aided by the news media and pub- lic social supervision [15]. The second is environmental policy issues. Some studies have indicated that these rules can encourage corporate green innovation [16]. Some studies have discovered a link between environmental restrictions that first inhibit corporate green innova- tion and later promote it [17, 18]. Others have discovered pilot policies of emissions trading [19], low-carbon pilot policies [20, 21], emission permit systems [22], carbon emission trading systems [23, 24], green credit policies [25–27], clean production audit (CPA) program [28] and environmental information disclosure system [29–32] can stimulate enterprises to make green innovation. The third is the political-enterprise relationship, which manifests as political affiliation and government subsidies. Political affiliation inhibits firms’ green innovation, espe- cially when the market degree is low [33]. And subsidies have a driving influence on corporate green innovation performance. However, political affiliation encourages enterprises’ green innovation by raising R&D spending and organizational capital [34, 35], but some studies show no significant relationship between corporate subsidies and green innovation [36]. The fourth is internal corporate factors, CEO responsible leadership [3], executive academic expe- rience [37], sustainability goals [38], CSR performance [4], and internal controls of institu- tional investors all contribute to encouraging green business innovation. In a broad sense, environmental, social, and governance (ESG) can be seen as an extension of corporate social responsibility because it uses the three criteria of environmental, social, and internal governance to evaluate businesses [39], which reflects the degree of green transforma- tion, and environmental image of enterprises [40]. ESG is also an ESG investment concept pursued by investors and becomes an investment basket of ESG factors. Hence, the ESG con- cept gradually becomes the consensus of global enterprises, investors, and financial institutions [41, 42]. ESG is an essential indicator of corporate green development, which is gradually incorporated into corporate development strategies [43, 44]. Current research has focused on the link between ESG and company performance. Some argue that ESG is unrelated to corpo- rate profitability, cost of capital, or ESG deteriorates corporate performance [45]. Other researchers find that ESG scores can alleviate firms’ financing constraints and improve their business performance [46–49]. In addition, better business stock returns with correspondingly higher stock liquidity and a dampening effect on crash risk are linked to higher ESG ratings [50, 51]. ESG ratings can help firms improve innovation performance and corporate value [52, 53]. However, other research believes ESG has a detrimental effect on corporate value [54]. Furthermore, a study found ESG investors in the Asia-Pacific region and the US perform simi- larly to the market. ESG investments are more suitable for ’value-driven investors’ (VDI). It also found that European investors will pay the price for making ESG investments, which is not conducive to improving company performance [55]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 2 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies As market competition becomes increasingly fierce, green innovation capability is becom- ing increasingly widely concerned by society. Companies’ protection and attention to the envi- ronment have been strengthened by deepening their ESG practices. Although companies have gradually paid attention to carrying out ESG practices and focus on the sustainable develop- ment route of enterprises, there aren’t many studies about how the ESG performance of com- panies affects corporate green innovation. A portion of the literature has focused solely on the social responsibility component, contending that social responsibility institutions and perfor- mance favorably influence the quantity and caliber of corporate green innovation [4, 56]. The ESG performance of a firm is not well represented by the social responsibility perspective, which is simply one component of that performance. Some scholars have pointed out the posi- tive association of environmental, social, and governance practices on corporate green innova- tion from three ESG dimensions. However, the sample is heterogeneous and covers different research settings [57]. Even though there is research that specifically investigates the influence of ESG on green innovation using ESG rating data for Chinese business channels [58], the paper has the following shortcomings: On the one hand, their main regression using whether firms receive ratings as a quasi-natural test is not very plausible because the sample includes unrated firms and the financial and green performance of firms that receive ratings is naturally higher than those of non-rated firms. Green innovation output and quality are correspond- ingly higher. Hence their empirical models are highly endogenous and cannot be considered a quasi-natural experiment, and the grouping method is not clean. On the other hand, they also discuss the effect of ESG-specific ratings on firms’ green innovation, with a much smaller sam- ple size than the stated DID regression, and the small difference in ratings does not reflect the difference in the refinement of corporate ESG performance, which therefore does not support their conclusions. It is significant to recognize that ESG performance can influence corporate innovations, specifically how it affects business performance, share price, and corporate value. Therefore, to understand how ESG performance affects corporate innovation activities, business perfor- mance, share price, and enterprise value, we must first understand how it affects those activities. Understanding the impact of ESG scores on corporate green innovation activities, the specific mechanisms, and the asymmetry of the impact in different circumstances is of utmost practical importance in the context of environmental pollution and resource depletion to realize green economic development and corporate green transformation. To empirically analyze the rela- tionship between corporate ESG scores and corporate green innovation, as well as its role mech- anism and moderating effect, we use the number of green patent applications and the number of green patent citations to measure corporate green innovation, build an empirical model using ESG scores from 2010 to 2019 and data on the quantity and quality of green innovation from 2011 to 2020. We also make an empirical model with a sample of listed Chinese companies in Shanghai and Shenzhen. Our findings confirm our research hypothesis by demonstrating a favorable relationship between company ESG performance and green innovation. Meanwhile, corporate ESG scores promote corporate green innovation activities mainly through two paths: improving investment efficiency and improving political and business rela- tions. In addition, the stronger the green attributes of firms, the stronger the ESG’s contribu- tion to green innovation performance; the stronger the black attributes of firms, the weaker the positive impact of ESG on green innovation performance. We use Bloomberg ESG Disclo- sure Scores published by Bloomberg as a proxy variable for ESG for the following reasons. On the one hand, the scores data is published by a non-Chinese organization. Thus, it is more independent in evaluating the ESG of enterprises. On the other hand, the variable is score data, which overcomes the original rating problems that are the non-refined and non-accurate evaluation of the ESG of enterprises. PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 3 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies The main contributions of our study are as follows. Firstly, to overcome the lack of refine- ment and precision of ESG performance measurement in previous studies [58], we use the ESG score variables instead of the original ESG ratings. Secondly, we tested the effectiveness of the ESG evaluation system from two perspectives of green innovation quantity and quality. Existing studies on ESG only analyze from three perspectives of corporate performance, stock price, and value ignoring the environmental and green attributes of ESG. Studies on green innovation explore from a quantitative standpoint ignoring the quality of green innovation. The ESG concept is better reflected in our study’s green innovation elements, and the green innovation quality is better reflected in the quantitative indicators used to measure green inno- vation. Thirdly, we examine the possible mechanisms of ESG influence on green innovation and analyze the asymmetry of ESG influence on green innovation in terms of the green and black markets. The remainder of the paper is as follows. Section 2 introduces the relevant theoretical foun- dations and then presents the relevant research hypotheses. Section 3 outlines the data selec- tion and model design. Section 4 presents and analyzes the empirical results. Section 5 provides robustness tests. In Section 6, the final section, we conclude with a discussion. Theory and hypotheses development ESG and green innovation Enterprises pay more attention to the relationship with corporate stakeholders, whether it is the green innovation activities based on technology or market-oriented business models [59], and put more emphasis on the creation of multiple integrated values based on innovation-led economic, social, and environmental [9] which are all associated with the ESG performance. Green innovation is a type of innovation where businesses try to use resources more efficiently and use less energy, and employ cutting-edge techniques to accomplish the twin objectives of economic and environmental performance [1]. Through green process product innovation [2, 60], businesses can reduce emissions and save energy. The advantages of companies’ environ- mental, social, and governance practices favorably increase the intensity of green technology innovation [57], so the impact of ESG on corporate green innovation is mainly reflected in the following three aspects. First of all, the environmental responsibility of enterprises contributes to the promotion of green innovation activities. Businesses’ production and operation activities are under signifi- cant pressure from the legal limits of environmental rules and the informal constraints of pub- lic environmental expectations. Enterprises engage in environmental responsibility while compelled to take steps to enhance their environmental performance to preserve a positive environmental reputation. Consequently, for environmental performance, energy saving, and emission reduction, enterprises must use green innovation technology to improve production technology to achieve clean production. And financial institutions consider companies’ envi- ronmental compliance when making investment and financing decisions. Therefore good environmental performance can alleviate financing constraints by reducing the financing cost of enterprises [61]. Furthermore, the environment is an essential external stakeholder for com- panies, and active corporate environmental responsibility helps to promote environmental cooperation. Companies are more likely to gain ideas about environmental management from their partners, including suppliers, to drive responsible green innovation projects [62]. Second, corporate social responsibility will promote green innovation by improving the relationship with stakeholders and providing them with the resources and information needed for green innovation activities. According to the stakeholder theory, actively undertaking social responsibility can assist companies in establishing broader and stronger relationships PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 4 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies with multiple stakeholders, such as customers, investors, regulators, and the public. These stakeholders will support enterprises’ green innovation activities by increasing consumption and investment [63]. Based on the resource-based theory, corporate social responsibility is conducive to gaining the trust of stakeholders, including investors and consumers, and getting the market and financial resources needed for green innovation. Companies increase their green innovation investment, thus promoting green corporate innovation [30, 58]. According to signaling theory, on the one hand, CSR has the "information effect" of alleviating informa- tion asymmetry and principal-agent problems, thus providing information to help enterprises make long-term decisions on green innovation activities, which enables them to obtain more green patent outputs [64]. On the other hand, the active fulfillment of social responsibility can send positive signals to the market about the good business performance of the company, which indicates that the company is capable of participating in social responsibility activities with its resources and helps to attract positive media attention and improve their reputation and brand image [65, 66], and enterprises faceless media pressure to enhance their risk toler- ance for innovation failure and stimulate their innovation energy, which in turn drives them to conduct green innovation activities with high uncertainty [67, 68]. In addition, higher par- ticipation in socially responsible activities enhances firms’ product market recognition [69]. As green markets develop and consumer demand soars dramatically, firms are more willing to engage in environmentally friendly green innovation activities to increase corporate value. Third, the better the internal governance, the higher the performance of corporate green innovation [70]. As green innovation activities have the characteristics of higher risk and lon- ger cycle, enterprises do not tend to make innovation investment decisions, thus hindering green innovation, but good corporate governance alleviates principal-agent conflicts through incentive and constraint mechanisms, prompting corporate management to increase corpo- rate R&D and innovation investment to achieve long-term sustainable corporate development [71, 72]. In addition, better internal governance can improve corporate performance, thus pro- viding continuous and stable financial support for long-term corporate green innovation activ- ities by mobilizing internal and external resources. Furthermore, ESG can promote corporate green innovation by optimizing corporate governance structures [73]. Gender diversity in the board of directors and executive management promotes more ESG practices in firms [74]. Board diversity can help companies become green organizations by promoting corporate ESG practices to stimulate green creativity, which drives companies to engage in green innovation [75]. Otherwise, ESG can help businesses adopt the ideas of sustainable development and crea- tive growth [76]. On the one hand, these ideas encourage companies to pursue energy conser- vation, emission reduction, and clean production goals, as well as to increase their innovation spending and adopt technology that reduces energy consumption and protects the environ- ment [77] to apply in their production and operation processes. On the other hand, enterprises with the guidance of green concepts invest their funds in green projects through green financ- ing to achieve a pro-environmental and pro-climate transformation of internal capital flows or make investment funds further greener to provide strong support for green activities, includ- ing green innovation activities, which can encourage corporate innovation in green. Accordingly, we obtained the research hypothesis: H1: ESG is positively correlated with corporate green innovation. Mechanism of investment efficiency The ESG evaluation system provides information and resources to support corporate invest- ment and forms constraints on corporate investment. In general, the higher the ESG score, the PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 5 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies better performance of the company, the better the relationship between the firm and its stake- holders, and the more willing they are to provide information and resources to support the firm. At the same time, the ESG score also imposes constraints on enterprises, or to sustain the current score, enterprises have to make more green investments to maintain their corporate image, and the relevant investments are by the policy call and public expectation, so ESG enhances the efficiency of enterprises’ investment through both resource support and invest- ment constraints. In addition, it has been shown that corporate social responsibility can reduce agency costs and information asymmetry, so ESG firms have a low cost of equity [78]. The higher the ESG score of a firm, the lower the cost of equity, which is conducive to further enhancing the firm’s investment efficiency. The higher the investment efficiency, the lower the inefficient investment, the closer to the optimal investment level, the more the resource utiliza- tion rate is about sufficient, the more the innovation output, and the more the green innova- tion output, that is, the more the green innovation patents obtained by the enterprise. Therefore, the higher the ESG score, the more efficient the firm’s investment and green inno- vation output. Accordingly, we propose the research hypothesis: H2: ESG can promote corporate green innovation by promoting investment efficiency. Mechanism of government-business relations The higher the ESG score, the better the relationship between the company and its stakehold- ers. In Asia, many government-backed investment funds inject large amounts of money into ESG activities to reflect the importance of ESG practices for social development [79]. In China, companies pay particular attention to their relationship with the government because a good relationship with the government provides them with government support, such as govern- ment subsidies and tax breaks, and facilitates their financing, production, and management by obtaining government approval. In recent years, the local ecological environment has been related to the performance of the local government. Government regulation, technology push, and market pull are the three major influencing factors on carbon technology innovation activities. Government regulation is the only factor positively influencing carbon technology innovation activities [80]. The promotion of green technology innovation in China cannot be achieved without the power of the government, and the connection between the government and firms will impact enterprises’ green technology innovation activities. Therefore, the better the ESG performance of a company, the more the government will support it, and conversely, its development will be restricted by the government. Since green innovation is long-term and risky [6], this greatly constrains the willingness and confidence of firms to make green innova- tion decisions. However, firms that maintain a good relationship with the government can gain more government support to share innovation risks and losses [34], encouraging firms to engage in green innovation activities. In summary, ESG scores can improve the relationship between government and firms, provide more resources for green innovation, and thus pro- mote innovation. Therefore, we develop the following research hypothesis. H3: ESG scores promote corporate green innovation through improving government-busi- ness relations. Moderation effect of green and black attributes The ESG evaluation, one of the critical components of the green financial system, can contrib- ute to green finance by promoting the effectiveness of financial resource allocation through the green flow of funds, thereby addressing the issue of environmental externality. This system primarily affects the financing of small and medium-sized businesses. And the companies internal and external environmental variables impact the green micro effect of the ESG system. PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 6 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies The stronger the black attribute, the stronger the environmental information asymmetry, the greater the environmental risk, and the more inclined enterprises are to make green bleaching behavior to cover up the poor environmental performance, thus maintaining the false ESG score and the external regulation will identify the green bleaching behavior of enterprises and thus inhibit the role of ESG. Nevertheless, when green attributes are stronger, environmental information asymmetry is smaller, and the environmental risks faced by firms are reduced, prompting ESG scores to be more objective to the ESG performance of firms, thus making full utilization of the green micro effects of the ESG system. In summary, we obtain the following research hypotheses. As a green attribute of a com- pany, an increase in environmental disclosure is conducive to promoting corporate ESG prac- tices. Such environmental and ethical practices can promote the legitimization of corporate activities, improve corporate image and thus increase corporate financial performance [81]. Companies increase their investment in green technology innovation and enhance their inno- vation capabilities. H4: Black attributes can weaken the positive effect of ESG scores on the green innovation of firms, and green attributes can enhance the promotion effect of ESG scores on the green inno- vation of a company. Methodology Sample and data The sample of this study is a research sample of Chinese listed businesses in Shanghai and Shenzhen A-shares from 2010–2019 to analyze the impact of ESG on corporate green innova- tion performance. We conduct the following treatment for the sample: firstly, we remove the samples that were ST, PT, and ST; secondly, we remove listed companies in the financial sec- tor; thirdly, we remove companies listed before 2010; fourthly, we remove the samples with missing main variables. After processing, we finally obtained 8258 annual observations of 1090 listed companies. We use a 1% and 99% tail reduction (Winsorize) for the primary variables. The data green on patents is from the China Research Data Service Platform. The data (CNRDS) on corporate finance is from the CSMAR and Wind databases, data on environmen- tal disclosure from social responsibility reports published by Hexun.com, data on corporate ESG scores are from Bloomberg’s Corporate Social Responsibility Disclosure Index (Bloom- berg ESG Disclosure Scores), regional environmental data and economic data are from pro- vincial statistical yearbooks, and macroeconomic data are from CEINet. We declare that we have no human participants, human data, or human issues. We do not have any individual person’s data in any form. Variables Explained variable. The explanatory variables in this paper are corporate green innova- tion. We define firms’ green innovation performance as quantitative and qualitative to obtain two explanatory variables for the number of green innovations (GI) and green patent citations (GC). The green patent is the most widely selected indicator to measure the green innovation ability of enterprises. The number of green patents granted reflects an enterprise’s green inno- vation level more than the number of green patent applications, so we add one to the number of green patents granted and take the logarithm to measure the quantity of green innovation (GI) of enterprises. For the quality of green innovation, most existing scholars choose to mea- sure the number of green invention patents and the number of green patents cited, among which the number of patents cited is more convincing than the invention patents [58], so in PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 7 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies this paper, we choose the number of green patents cited plus one and take the natural loga- rithm to measure the quality of green innovation (GC) of enterprises. Explanatory variable. The core explanatory variable in this paper is the ESG score of firms. The ESG data is derived from the Bloomberg ESG Disclosure Scores, which consists of the ESG composite score and the ESG sub-scores with 122 sub-scores on 21 topics in three major categories. Intermediary variables. (1) Efficacy of investments comes first (IE). We utilize the absolute value of the residuals from the subsequent regression to measure inefficient investment as Model (1) [82]. The larger the indicator, the less efficient the firm’s investment. CIi;t ¼ b0 þ b1SGi;t(cid:0) 1 þ εi;t ð1Þ In Model (1), CIi,t represents the investment level of an enterprise, which is the proportion of fixed and intangible assets to total assets. ESGi,t represents the investment opportunity of an enterprise, which is the growth rate of sales revenue. The residual term represents the propor- tion of inefficient investment in the total investment, and the absolute value is taken to obtain the investment efficiency index IE. The larger the value, the less efficient the investment. (2) Government subsidies (Subsidy). We use the normalized government subsidy (Subsidy) as a proxy variable for the government-enterprise relationship, which reflects the characteristics of the sample. The larger value indicates that means, the more government subsidy a firm receives, the better the relationship between the firm and the government Control variables. By previous studies [4, 15, 56, 83], we take into account variables such as the firm’s age (year of foundation), gearing (leverage), return on total assets (ROA), and Tobin’s Q. (Q), net cash from investing activities (ICF), fixed assets (Fix), foreign ownership (QFII), dual employment (Dual), and audit opinion (Opinion). The key variables used in the empirical analysis are shown in Table 1. Model Baseline model. Our data are short panel data, so a baseline regression model can repre- sent the significant relationship between the independent variable ESG score and the depen- dent variable green innovation level. We use this model to control for year-fixed, industry- fixed, and province-fixed effects to control for the effect of unobservable factors at the industry and province levels overtime on the relationship between ESG score firms and green innova- tion level, and to city-level clustering. In addition, we can use the model to further examine the mechanisms and moderators of ESG scores affecting firms’ green innovation. Based on the prior analysis and variable definitions, we use Model (2) for testing hypothesis H1. GIi;tþ1 ¼ a0 þ a1ESGi;t þ gXi;t þ lt þ Zj þ εi;t ð2Þ Where GIi,t+1 repents the firm i’s level of green innovation in year t+1, ESGi,t denotes the firm i’s Bloomberg ESG score in that year, Xi,t suggests a series of control variables, λt denotes time fixed effects, ηj denotes industry fixed effects; and εi,t represents the random disturbance term. Intermediation model. To test H2 and H3, the mediating effects of investment efficiency (IE) and government-enterprise relationship (Subsidy), this paper further sets up the following mediation model and sets up the following testing steps [84, 85]. First, Model (1) shows the results of the regression model of corporate green innovation on ESG score. If β1 is significant, PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 8 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Table 1. Descriptive statistics of the variables. Variable classification Variable name Variable symbol Variable definition Explained variables Quantity of Green Innovation Quality of Green Innovation Core explanatory variables ESG Score Intermediate variables Investment efficiency Control variables Years of Establishment Government Grants Gearing Ratio Total Return on Assets Tobin’s Q Net cash from investing activities Fixed Assets Foreign equity holdings Two positions in one GIt+1 GCIt+1 ESG IE Subsidy Age Leverage ROA Q ICF Fix QFII Dual The logarithm of the number of green patents granted plus one to take the logarithm The logarithm of the number of green patent citations plus one to take the logarithm The logarithm of Bloomberg ESG Estimated from the Model (1) Normalized government grants Ln(year—year of establishment) Total liabilities/total assets Total profit/total assets Total market capitalization/total assets Net cash from investing activities/total assets Fixed Assets/Total Assets Foreign shareholding ratio The value is 1 if the chairman is also the general manager; otherwise, it is 0 https://doi.org/10.1371/journal.pone.0279220.t001 Audit opinion Opinion The standard unqualified opinion takes the value of 1; otherwise, it is 0 the second step is carried out. Second, the regression equation of ESG score and mediating variables (IE and Subsidy) on corporate green innovation is constructed. The mediating mech- anism exists if μ2 is significant and the signs of μ2 and β1 are the same. IEi;t=Subsidyi;t ¼ b0 þ b1ESGi;t þ gXi;t þ lt þ Zj þ εi;t GIi;tþ1 ¼ m0 þ m1ESGi;t þ m2IEi;t=Subsidyi;t þ gXi;t þ lt þ Zj þ εi;t ð3Þ ð4Þ Where IEi,t, and Subsidyi,t represent the investment efficiency and government subsidies, respectively, and the rest of the variables are consistent with the baseline model. Moderating effect model. To test H4, the moderating effect of the environmental attri- butes of firms, the following regression Model(5) was set up based on the baseline model. GIi;tþ1 ¼ a0 þ a1ESGi;t þ a2ESGi;t � Rit þ gXi;t þ lt þ Zj þ εi;t ð5Þ Where R consists of the black and green attributes of the company. Black attributes include regional, industry, and company pollution attributes. We use the high pollution region dummy variable HPP (The regional pollution index for the current year takes a value of 1 if it is higher than the average value, and 0 otherwise.), the high pollution industry dummy variable HPI (high pollution industry takes a value of 1 otherwise it takes a value of 0) and the high pol- lution company dummy variable HPC (If the enterprise is a key pollution monitoring unit take the value of 1, otherwise it takes the value of 0) separately to measure black attributes. Green attributes include provincial, city, and firm environmental attributes. We employ pro- vincial green finance DGF (normalized green finance index), city green innovation DGI (ratio of the total number of green patents in the city to the current year’s average), and corporate environmental disclosure (the number of quantitative disclosures of environmental liability items as a proportion of the total number of items) as green attributes. And the remaining var- iables are consistent with Model 2. PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 9 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Table 2. Descriptive statistics of the main variables. Variables GIt+1 GCt+1 ESG E S G Age Leverage ROA Q ICF Fix QFII Dual Opinion N 8258 8258 8258 6950 8035 8258 8258 8258 8258 8258 8258 8258 8258 8258 8258 Mean S. D. 0.25 0.41 2.97 2.16 3.07 3.80 2.87 0.47 7.29 1.90 -0.06 0.23 0.17 0.20 0.99 0.62 0.92 0.31 0.67 0.41 0.11 0.32 0.20 6.17 1.24 0.08 0.18 0.54 0.40 0.12 Max 2.48 4.56 3.77 3.72 4.03 4.05 3.53 0.89 36.44 8.78 0.17 0.70 2.79 1.00 1.00 Median 0.00 0.00 2.99 2.23 3.13 3.80 2.89 0.48 5.95 1.48 -0.05 0.19 0.00 0.00 1.00 Min 0.00 0.00 2.21 0.73 1.95 3.52 1.61 0.05 -8.26 0.88 -0.39 0.00 0.00 0.00 0.00 https://doi.org/10.1371/journal.pone.0279220.t002 Descriptive statistics. The results of the descriptive statistics for the primary variables are shown in Table 2, where the mean value of green patents (GI) is 0.25, the standard deviation is 0.62, the maximum value is 0.48, and the minimum value is 0. This data suggests that the sam- ple enterprises’ average level of green innovation is low and that there is significant enterprise- level variation in their capacity for green innovation. ESG scores (ESG) vary significantly among businesses; the mean value is 2.97, the standard deviation is 0.31, the maximum is 3.77, the minimum is 2.21, and the median value is 2.99. Correlation test. The Pearson correlation coefficient test matrix is displayed in Table 3. We can infer from Table 3 that there is a significant positive association between ESG score and corporate green innovation, which supports H1 preliminarily. Panel unit root test. The existence of unit roots in panel data can have serious conse- quences, such as pseudo-regression, so we use both the Im-Pesaran-Shin test and Levin-Lin- Chu test to perform unit root tests to ensure the smoothness of each variable. Table 4 shows the results of the panel unit root tests. It can be seen that all variables are stationary at the 1% level, which means no unit root exists in the series. The results strongly reject the null Table 3. Pearson correlation coefficient test. GIt+1 1.000 0.513 0.118* 0.110* 0.101* 0.028* GCt+1 1.000 0.200* 0.182* 0.154* 0.098* ESG E S G 1.000 0.833* 0.820* 0.514* 1.000 0.508* 0.260* 1.000 0.306* 1.000 GIt+1 GCt+1 ESG E S G Note p < 0.01 p < 0.05 p < 0.1. https://doi.org/10.1371/journal.pone.0279220.t003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 10 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Table 4. Panel unit root test. Variables GIt+1 GCt+1 ESG Age Leverage ROA Q ICF Fix QFII Dual Opinion Im-Pesaran-Shin test Levin-Lin-Chu test t-bar -1.814 -1.898 -1.657 -2.509 -1.850 -2.044 -1.838 -2.033 -1.984 -6.425 -4.737 -1.988 W[t-bar] -6.819 -8.581 -3.540 -21.349 -7.574 -11.631 -7.325 -11.404 -10.378 -103.261 -67.945 -10.455 P-value 0.000* 0.000* 0.000* 0.000* 0.000* 0.000* 0.000* 0.000* 0.000* 0.000* 0.000* 0.000* t-value -49.167 -47.758 -45.468 -71.791 -56.996 -57.075 -55.392 -50.308 -57.179 -161.793 -391.658 -283.569 t-star -9.414 -37.025 -24.492 -65.239 -44.343 -38.191 -33.445 -28.543 -41.668 -164.627 -415.238 -300.026 P-value 0.000* 0.000* 0.000* 0.000* 0.000* 0.000* 0.000* 0.000* 0.000* 0.000* 0.000* 0.000* https://doi.org/10.1371/journal.pone.0279220.t004 hypothesis of unit root, so we can argue that the data are stable and there is no biased informa- tion in the panel. Empirical results and analysis Baseline regression The ESG benchmark regression results are shown in Table 5. The explanatory variables in col- umns (1)-(4) are the quantity of green innovation. In column (1), the coefficient of ESG on the number of green innovation patents is 0.300, which is significant at the 1% level, indicating that ESG can increase the number of green innovation patents for companies. Based on the three sub-items of the ESG evaluation, we replace ESG with the natural logarithm of the corre- sponding scores for Environmental E, Social S, and Corporate Governance G. In column (2)- (4), the coefficient estimates of E and S are significantly positive at the 1% level, and the coeffi- cient estimates of G is significantly positive at the 10% level, indicating that E, S, and G scores all promote the level of green innovation in companies. The explanatory variables in columns (5)-(8) are the quality of green innovation. In column (5), the regression coefficient of ESG is 0.610, which is significant at the 1% level, which suggests that ESG encourages business cita- tion of green innovation patents. In columns (6)-(8), E, S, and G coefficient estimates are all significantly positive at the 1% level. The coefficient values are increasing in order, demon- strating that the positive effects of E, S, and G on the quality of green patents are in the order of G, S, E. The result above indicates that E, S, and G scores all promote the quality of green inno- vation in companies. The regression results show that the amount and quality of green innovation output increase with increasing ESG score, supporting H1. In addition, our regression results indicate that all three subcategories of ESG can promote the quantity and quality of green innovation in enterprises. For the subscores of corporate ESG scores, we find that the E score has the most significant impact on corporate green innovation, and the G score has the least significant impact on corporate green innovation. Still, overall, the subscores of ESG all drive the quantity and quality of corporate green innovation. The descriptive statistics of the remaining control variables are generally consistent with existing studies [35, 55, 58]. The results illustrate that ESG scores can increase the quantity and quality of green innova- tion and that ESG is a sustainable "substantive innovation" rather than a "masked innovation" PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 11 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Table 5. Baseline regression results. Variables ESG E S G Age Leverage ROA Q ICF Fix QFII Dual Opinion Constant Y/I/P FE Observations Adj R2 (1) GI_1 0.300* (7.04) -0.193 (-2.09) 0.183 (2.35) 0.002 (0.89) -0.028 (-2.30) -0.079 (-0.63) -0.035 (-0.26) 0.026 (0.83) 0.041 (1.16) 0.081* (1.80) -0.128 (-0.40) YES 8,258 0.114 (2) GI_1 (3) GI_1 (4) GI_1 0.123* (6.68) -0.194 (-2.00) 0.198 (1.98) 0.002 (0.95) -0.037* (-2.62) -0.125 (-0.78) -0.069 (-0.49) 0.034 (1.08) 0.047 (1.26) 0.087 (1.61) 0.498 (1.63) YES 6,950 0.120 0.167* (5.75) -0.195 (-2.11) 0.201* (2.66) 0.003 (1.22) -0.032* (-2.72) -0.063 (-0.48) 0.018 (0.13) 0.029 (0.91) 0.035 (1.03) 0.079* (1.76) 0.231 (0.73) YES 8,035 0.107 0.312* (1.75) -0.203 (-2.27) 0.209* (2.83) 0.003 (1.17) -0.034* (-2.86) -0.066 (-0.50) 0.035 (0.25) 0.029 (0.90) 0.029 (0.76) 0.102 (2.19) -0.446 (-0.55) YES 8,258 0.0975 Note: T-statistics calculated for city-level clusters in parentheses. https://doi.org/10.1371/journal.pone.0279220.t005 (5) GC_1 0.610* (7.32) -0.223 (-1.33) 0.334 (2.16) -0.006* (-1.78) 0.012 (0.68) -0.440* (-1.86) -0.387* (-2.74) 0.032 (0.75) 0.096 (1.27) -0.123 (-1.18) -0.712 (-1.28) YES 8,258 0.0993 (6) GC_1 (7) GC_1 (8) GC_1 0.267* (6.87) -0.217 (-1.12) 0.390** (2.01) -0.007* (-1.87) -0.006 (-0.28) -0.590 (-1.99) -0.478* (-2.91) 0.044 (0.98) 0.103 (1.25) -0.080 (-0.68) 0.373 (0.66) YES 6,950 0.0997 0.322* (6.39) -0.226 (-1.34) 0.379 (2.56) -0.005 (-1.54) 0.004 (0.24) -0.411* (-1.67) -0.290* (-1.92) 0.042 (0.95) 0.077 (1.04) -0.122 (-1.14) 0.017 (0.03) YES 8,035 0.0830 0.981* (3.11) -0.249 (-1.52) 0.367 (2.40) -0.006 (-1.55) 0.003 (0.16) -0.424* (-1.71) -0.260* (-1.74) 0.038 (0.86) 0.076 (0.96) -0.087 (-0.78) -2.645* (-1.80) YES 8,258 0.0754 to simply whitewash financial statements. It is worth mentioning that the G score affects the number of green innovations less significantly than the E and S scores, probably because green innovation projects crowd out the firm’s inherent resources and conflict with its short-term financial performance. We also find that when the explanatory variable is replaced with the number of green patents cited, all three aspects of ESG significantly improve the quality of green innovation at the 1% level. The coefficient of the G score is the largest. This result indi- cates that executives value the strategic perspective of the company’s long-term development and choose to make high-quality green innovations to improve the company’s competitive- ness, so companies with good green strategies significantly improve the quality of green innovation. Our results affirm the positive significance of ESG practices for green innovation, which positively affect companies’ green transformation. The results also demonstrate the critical PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 12 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies role of the ESG scores of companies in influencing their green innovation decisions and that favorable practices in environmental, social, and governance aspects of companies will jointly promote corporate green innovation, achieve a sustainable development path for enterprises, and promote the integration of environmental, social and economic effects of enterprises. Intermediary mechanism analysis Mechanism of investment efficiency. The regression results for the mediating influence of investment efficiency are shown in columns (1)-(3) of Table 5. The coefficient estimate of ESG in column (1) is -3.183 and significantly negative at the 1% level, which means businesses with higher ESG scores make better investors. The coefficient estimates of IEt+1 in columns (2) and (3) are -0.003 and -0.005, respectively, and are statistically negative at the 5% level, indicat- ing the existence of this mediating effect and the relationship between investment efficiency and green innovation performance, At the 1% level, both of the coefficient estimates of ESG in columns (2) and (3)—0.291 and 0.594, respectively—are significantly positive. Both columns (2) and (3) coefficient values of ESG 0.291 and 0.594, respectively—are statistically significant at the 1% level. The regression results suggest that ESG performance contributes to green inno- vation by improving firms’ investment efficiency. As a result, H3 should be accepted. The results suggest that the fulfillment of ESG responsibilities will drive companies to make green investments to cater to investors’ preference for environmentally friendly companies and that ESG practices are conducive to improving the efficiency of investments and the utili- zation of internal and external resources, which in turn will make companies willing to engage in more green innovation activities and improve their green technological innovation capabilities. Mechanism of government-business relations. Columns (4) to (6) of Table 6 show the regression results for the mediating effect of the government-firm relationship. The coefficient estimate of ESG in column (4) is 0.027 and significantly positive at the 1% level, indicating that the higher the ESG score, the more government subsidies the firm receives. In other words, the ESG score significantly improves the relationship between the government and the firm; the coefficient estimates of Subsidy in columns (5) and (6) are respectively 0.2289 and 5.407, and both are positive at the 1% level, which means that government subsidies significantly pro- mote green innovation, so the better the relationship with the government, the more govern- ment subsidies the enterprises receive, and the more funds they have to engage in green Table 6. Regression results for mediating mechanisms. (1) IEt+1 -3.184* (-4.65) Variables ESG IEt+1 Subsidy Constant 13.907* Controls Y/I/P FE Observations Adj R2 (3.50) YES YES 8,258 0.082 https://doi.org/10.1371/journal.pone.0279220.t006 (2) GIt+1 0.291* (7.16) -0.003 (-2.53) -0.109 (-0.34) YES YES 8,258 0.139 (3) GCt+1 0.594* (7.33) -0.005 (-2.38) -0.673 (-1.25) YES YES 8,258 0.163 (4) Subsidy 0.027*** (3.23) -0.036* (-1.69) YES YES 8,258 0.252 (5) GIt+1 0.239* (5.14) 2.289* (10.80) -0.065 (-0.21) YES YES 8,258 0.115 (6) GCt+1 0.465* (5.50) 5.407* (7.58) -0.541 (-1.04) YES YES 8,258 0.101 PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 13 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies innovation behavior. Thus this mediating effect exists. The coefficient value decreases com- pared to the baseline regression. Therefore the mediation effect is partial. The above regression results suggest that ESG performance promotes corporate green innovation by improving the relationship between government and business, which supports hypothesis H2. The results indicate the important role of government-business relationships in mediating the impact of ESG performance on corporate green innovation. The government encourages and supports ESG practice projects, so companies that participate in ESG practice projects can build a good corporate image, maintain good relations with the government, and gain political resources, including government subsidies and the economic resources they bring. These com- petitive resources can be regarded as a kind of external risk protection, which can reduce the cost of green innovation and reduce the risk of R&D, and enhance the motivation of enter- prises to invest in green innovation projects. Moderation effects analysis The regression results of Panel A in Table 7 show that the interaction coefficients of ESG with HPP, HPI, and HPC decrease in significance and coefficient values compared with the esti- mated values of the baseline regression ESG, which indicates that the stronger the black attri- butes of the firm, the weaker the promotion effect of ESG on green innovation. The regression Table 7. Regression results for moderating effects of black and green attributes. Variables ESG×HPP ESG×HPI ESG×HPC Constant Controls Y/I/P FE Observations Adj R2 ESG×DGF ESG×CGI ESG×EDG Constant Controls Y/I/P FE Observations Adj R2 (1) GIt+1 0.026 (1.40) 0.693 (2.40) YES YES 8,258 0.095 0.386* (3.99) -0.469* (-1.66) YES YES 8,258 0.104 https://doi.org/10.1371/journal.pone.0279220.t007 (2) GCt+1 0.101 (2.54) 0.977 (1.99) YES YES 8,258 0.068 0.826* (3.83) -1.521* (-2.64) YES YES 8,258 0.083 (3) GIt+1 Panel A: Black Features 0.015* (1.78) 0.677** (2.31) YES YES 8,258 0.096 Panel B: Green Features 0.572* (1.93) 0.820 (2.58) YES YES 5,439 0.127 (4) GCt+1 0.038 (2.24) 0.930* (1.86) YES YES 8,258 0.068 1.306 (2.48) 1.042 (2.13) YES YES 5,439 0.125 (5) GIt+1 (6) GCt+1 0.011 (1.43) 0.698 (2.38) YES YES 8,258 0.095 0.856* (4.80) 0.673 (2.34) YES YES 8,258 0.010 0.067* (4.14) 1.024 (2.05) YES YES 8,258 0.072 1.360* (4.31) 0.933* (1.91) YES YES 8,258 0.071 PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 14 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies results of Panel B show that the coefficient values of ESG and DGF, CGI, and EDG are all signif- icant at the 1% level and are greater than the baseline regression coefficient values, which sug- gests that the stronger the green attributes of firms, the greater the positive impact of ESG on green innovation. The results demonstrate the opposed effects of corporate green and black attributes on the relationship between ESG scores and corporate green innovation. At the black attribute level, the sample, whether a highly polluting industry or a highly polluting firm, exacerbates environ- mental information asymmetry and exposes firms to higher environmental risks. Firms will mask the inherent risks through green bleaching practices. Thus ESG scores are more biased towards a false reflection of ESG performance and will weaken the positive effect of ESG scores on green innovation. At the level of green attributes, whether at the province, city, or firm level, green attributes can reduce environmental information asymmetry, make ESG scores more realistic and reliable reflections of firms’ true ESG performance, and enhance the effec- tiveness of ESG scores in promoting corporate green innovation. The government should increase the punishment for polluting enterprises, increase the cost of polluting enterprises through environmental regulation pressure, and consciously promote the transformation of enterprises from black attributes to green attributes. And enterprises should increase the dis- closure of environmental information to reduce the uncertainty of environmental information and enhance their green attributes, and at the same time, reduce emissions and environmental pollution by improving production processes and greening production to reduce their black attributes, to better utilize the positive effect of ESG performance on green innovation. Robustness tests Replacing measures of core variables In the robustness test section, we use the number of green patent applications to measure the quantity of green innovation of the firm (GGI_1) and the number of green invention patents to measure the quality of green innovation of the firm (INNO_1). Table 8 reports the regres- sion results for replacing the core variable measures. The results are consistent with the bench- mark regression, where both the composite corporate ESG score and sub-scores contribute to the quantity and quality of corporate green innovation. Table 8. Regression results for replacing core variables. Variables ESG E S G Constant Controls Y/I/P FE Observations Adj R2 (1) GGI_1 0.389* (6.02) -1.058* (-3.25) YES YES 6,891 0.165 (2) GGI_1 (3) GGI_1 (4) GGI_1 0.186* (6.28) -0.345 (-1.06) YES YES 5,724 0.169 0.187* (4.16) -0.530 (-1.59) YES YES 6,675 0.153 0.326* (1.82) -1.179 (-1.45) YES YES 6,891 0.150 https://doi.org/10.1371/journal.pone.0279220.t008 (5) INNO_1 0.264* (5.48) -0.143 (-0.42) YES YES 8,258 0.0741 (6) (7) (8) INNO_1 INNO_1 INNO_1 0.122* (4.76) 0.368 (1.29) YES YES 6,950 0.0803 0.153* (4.40) 0.158 (0.49) YES YES 8,035 0.0682 0.472 (2.26) -1.154 (-1.19) YES YES 8,258 0.0616 PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 15 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Fig 1. Placebo test. https://doi.org/10.1371/journal.pone.0279220.g001 Placebo test We used a non-parametric permutation test to perform a placebo test on the baseline regres- sion. The placebo test is illustrated in Fig 1. We find that from the test results that the distribu- tion of the estimated coefficients for the 500 random samples is close to a normal distribution with mean zero and that the coefficients of the benchmark regressions for GIt+1 and GCt+1 green innovation indicated by the dashed lines in the figure, Table 5 columns (3) and (6) are dif- ferent from the correlation coefficients obtained from the non-parametric tests. Therefore, the test results exclude the possibility that the effect of ESG on green innovation performance is dependent on other unobservable factors. In other words, the interference of other events in the benchmark regression is excluded, and the obtained benchmark regression results are robust. Adding variables We next control provincial and national level economic variables that may affect firms’ green innovation based on the baseline regression column (1) to verify the robustness of the baseline regression results. We specifically introduce regional per capita gross product (PerGDP, the logarithm of regional per capita gross product), regional financial development level (FD, regional deposit and loan as a share of GNP), regional pollution level (DPG, industrial pollu- tion investment as a share of GNP), broad money growth rate (M2), and Shanghai interbank lending rate (Shibor, the annual 10-year Shanghai interbank lending average interest rate) to control for regional economic, environmental and macroeconomic effects on the benchmark regressions. Table 9 shows the results of the regressions with the addition of control variables. From the regression results, the coefficient estimates for ESG are all significantly positive at the 1% level. The regression results are generally consistent with the benchmark regression, which means the robustness of the benchmark regression. PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 16 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Table 9. Regression results for adding control variables. Variables ESG FD DPG M2 Shibor Constant Controls Y/I/P FE Observations Adj R2 (1) GIt+1 0.301* (7.10) 0.002 (1.07) -0.208 (-2.56) -5.300* (-2.73) YES YES 8,258 0.114 (2) GIt+1 0.301* (7.10) 0.002 (1.07) -0.208 (-2.56) 0.288* (5.05) 1.043* (4.90) -10.082* (-3.48) YES YES 8,258 0.114 (3) GCt+1 0.609* (7.33) -0.002 (-1.21) 0.089 (0.59) -0.341 (-0.13) YES YES 8,258 0.074 (4) GCt+1 0.609* (7.33) -0.002 (-1.21) 0.089 (0.59) -0.142* (-1.72) -0.476 (-1.55) 1.829 (0.46) YES YES 8,258 0.074 https://doi.org/10.1371/journal.pone.0279220.t009 Replacement regression models GIt+1 and GCt+1 are discrete variables suitable for Poisson, Tobin, and Negative Binomial regression models. Table 10 shows the results of the substitution regression model. From the regression results, the coefficient estimates of ESG are all significantly positive at the 1% level. The regression results are generally consistent with the baseline regression, which suggests the robustness of the baseline regression. Instrumental variables approach Using green innovation indicators for period t+1 avoids the problems associated with certain simultaneity biases while reducing the estimation error associated with reverse causality issues. However, the relationship between ESG and green innovation is still strongly endogenous, which means that firms with higher green innovation performance also have higher ESG scores. There may also be omitted variables that affect ESG scores. At the same time, there is Table 10. Regression results of the replacement model. Variables ESG Constant Controls Y/I/P FE Observations Loglikelihood Pseudo R2 (1) Poisson 1.181* (8.24) -4.124* (-3.71) YES YES 8,258 -4437 0.186 https://doi.org/10.1371/journal.pone.0279220.t010 (2) GIt+1 Tobit 0.300* (7.10) -0.148 (-0.46) YES YES 8,258 -7176 0.068 (3) NB 1.169* (7.35) -3.901* (-3.41) YES YES 8,258 -4338 0.147 (4) Poisson 1.376* (9.37) -4.029* (-3.72) YES YES 8,258 -6926 0.118 (5) Tobit GCt+1 0.609* (7.36) -0.737 (-1.33) YES YES 8,258 -10599 0.042 (6) NB 1.378* (8.08) -3.857* (-3.43) YES YES 8,258 -6345 0.067 PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 17 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Table 11. 2SLS and GMM results for instrumental variables. Variables ESGMeant-1 ESG Constant Observations Controls Y/I/P FE Adj R2 F statistics Kleibergen-Paaprk LM statistic Cragg-Donald Wald F-statistic Kleibergen-Paaprk Wald F-statistic https://doi.org/10.1371/journal.pone.0279220.t011 (1) ESG 0.679* (17.29) 2.766* (54.35) 8,258 YES YES 0.211 35.95 (2) 2SLS GIt+1 (3) GCt+1 (4) GIt+1 (5) GCt+1 GMM 0.757* (6.89) -1.419* (-4.45) 8,258 YES YES 0.071 19.64 235.653 262.190 299.097 1.156* (6.56) -2.260* (-4.45) 8,258 YES YES 0.072 12.26 235.653 262.190 299.097 0.755* (6.87) -1.391* (-4.37) 8,258 YES YES 0.071 1.159* (6.58) -2.238* (-4.40) 8,258 YES YES 0.072 simultaneously an impact on firms’ green innovation that makes the benchmark regressions biased and inconsistent. We use an instrumental variable to address this issue to eliminate the effect of potential endogeneity on the benchmark regression. This paper chooses the industry- level mean of ESG (ESGMeant-1) of the previous year as the instrumental variable [84]. The industry influences the ESG score, but the industry-level mean is not directly related to the green performance of individual firms, so ESGMeant-1 meets the requirements of an instru- mental variable. Before conducting the least squares regression of the instrumental variables, we first con- ducted a correlation coefficient test between ESGMeant-1 and ESG. The Pearson correlation coefficient test results showed that the correlation coefficient between the two was 0.194 and significant at the 1% level, so we can initially conclude that the higher the industry ESG means, the higher the ESG performance of the firm. The outcomes of the 2SLS and GMM results for the instrumental variables are shown in Table 11. The first three columns are the estimated results of 2SLS. According to the regression results, the first stage’s coefficient estimates of ESGMeant-1 is 0.679 and significant at the 1% level, suggesting that the industry in which a company operates impacts its ESG performance. The second stage regression shows that the predicted ESG coefficients are considerably positive at the 1% level, demonstrating that ESG improves business performance regarding green innovation. After conducting the main regression, we conduct a series of tests for instrumental variables such as homogeneity of instrumental variables, weak instrumental variables, and over-identification, whose results show that the Model passes all tests. The last two columns are the estimated results of GMM. The regression results also validate the baseline hypothesis of this paper. Propensity score matching To address the problem of sample selection bias, we choose the propensity score matching method (PSM), using a 1:1 nearest neighbor matching with a matching radius of 0.05, with whether it is a highly polluting industry as the grouping variable and all the control variables in column (1) as covariates, inducing age of establishment (Age), gearing (Leverage), return on total assets (ROA), Tobin’s Q (Q), net cash from investing activities (ICF), fixed assets (Fix), PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 18 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Table 12. PSM-benchmark regression results. Variables ESG E S G Constant Y/I/P FE Observations Adj R2 (1) GIt+1 0.260* (4.92) 0.016 (0.05) YES 3,379 0.109 (2) GIt+1 (3) GIt+1 (4) GIt+1 0.099* (4.06) 0.663* (2.69) YES 2,894 0.111 0.138* (3.85) 0.317 (1.16) YES 3,274 0.103 0.284 (1.57) -0.361 (-0.45) YES 3,379 0.098 https://doi.org/10.1371/journal.pone.0279220.t012 (5) GCt+1 0.504* (5.98) -0.129 (-0.23) YES 3,379 0.082 (6) GCt+1 (7) GCt+1 (8) GCt+1 0.236* (4.89) 0.867* (1.84) YES 2,894 0.080 0.260*** (6.10) 0.460 (0.96) YES 3,274 0.072 0.456 (1.05) -0.509 (-0.27) YES 3,379 0.069 foreign ownership (QFII), dual employment (Dual) and audit opinion (Opinion). After passing the common support hypothesis and parallel trend hypothesis tests, the benchmark regression was re-run, and the regression results are shown in Table 12. From the results, we find that the regression coefficients of ESG are all significantly positive at the 1% level. Meanwhile, the coef- ficient estimates of E and S are both significantly positive at the 1% level, but the coefficient estimate of G is not statistically significant after eliminating the problem of the sample, which indicates that the short-term corporate governance objectives of the company are contrary to the long-term green innovation activities, consistent with economic theory and experience. Conclusion and discussion Green innovation is a crucial manifestation of corporate applying the ESG concept, which reflects the micro-green effect of the ESG evaluation system. Using panel data and the sample of Chinese listed businesses from 2010 to 2019, we empirically explore the impact of ESG scores on corporate green innovation from corporate investment efficiency and government- enterprise relations perspectives. The results indicate both the composite and sub-scores of a company’s ESG contribute to the quantity and quality of its green innovation. And ESG sup- ports corporate green innovation by increasing businesses’ investment effectiveness and improving their government-business relationship. The results also show that corporate green attributes strengthen the promotion function of ESG on corporate green innovation. In con- trast, black attributes reduce the beneficial effects of ESG on corporate green innovation. According to our research, the following recommendations can be made for enhancing the ESG evaluation system and encouraging the sustainable growth of micro-enterprises. Firms need to implement the ESG concept, manage the various environmental risks they face, increase their level of pro-environmental preference, enhance the environmental disclosure mechanism, pay more attention to the non-financial performance of green performance, and promote business development and green development. The findings of this paper prove the importance of practicing environmental, social, and governance responsibilities and the posi- tive significance of ESG performance for enterprises’ green and sustainable development. The implementation of the ESG concept by enterprises is conducive to promoting the integration of environmental, social, and economic performance and achieving a win-win situation of environmental, social, and economic effects. Moreover, the findings of this paper also point PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 19 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies out the way and direction for enterprises to promote green innovation development. By actively fulfilling environmental and social responsibilities, enterprises can win the trust of stakeholders, including governments and investors, obtain key political and economic resources that are indispensable for green innovation, alleviate financing constraints, improve resource utilization, enhance the output and quality of green technology innovation, and embark on a sustainable green development transformation path. Moreover, the government should implement green development into practice, create fiscal policies for businesses based on the ESG evaluation system, subsidize green enterprises and restrict black enterprises, and encourage businesses to engage in green innovation activities that adhere to ESG standards. The conclusion of this paper proves the key role played by good political-business relations between ESG scores and corporate green innovation. Therefore, the government should focus on the role of important political and economic resources, including government subsidies and tax incentives, and strongly support enterprises to carry out ESG practice projects that are beneficial to social development and progress to attract enterprises to participate in green innovation activities consciously and actively, thus guiding more enterprises to take the green development path. Regulators should create distinct regulatory policies based on businesses’ environmental risks and enhance the mechanism for exchanging environmental information to encourage companies to engage in green innovation activities. Regulators should pay attention to the environmental information disclosure of enterprises, timely detect the possible "greenwashing" behavior of enterprises and punish these enterprises, to promote the ESG score to reflect the ESG performance of enterprises more truly and let the ESG performance promote the green innovation of enterprises in practice, that is, let the green attributes better promote the positive link between ESG score and green innovation of enterprises, and weaken the inhibiting effect of black attributes on the relationship between the two. Institutional investors need to pay attention to the ESG performance of enterprises and fur- ther incorporate ESG factors into their investment strategies to better identify enterprises’ internal and external environmental risks and provide enterprises with corresponding funds based on ESG evaluation. As an important external stakeholder of enterprises, enterprises will pay attention to the investment tendency of institutional investors to obtain more financing support. Therefore, institutional investors pay attention to ESG investment concepts, environ- mental protection of enterprises, and sustainable development strategies, which are conducive to guiding enterprises to pay attention to ESG practices, fulfilling environmental and social responsibilities, and enhancing their green innovation drive. The limitations of this paper lie in the following two aspects. On the one hand, we only explore the micro-green effect of the ESG evaluation system and do not analyze the role of the ESG evaluation system comprehensively. On the other hand, we ignored the motives of corpo- rate greenwashing and failed to eliminate the part of corporate greenwashing in green innova- tion. Future research can examine the relationship between ESG scores and green innovation from two aspects. First, the research can analyze the role of ESG in greenwashing behaviors such as environmental performance, production performance, and investment efficiency. Sec- ond, future research will have indicators to identify green innovation drifting green motives to better examine the effectiveness of the ESG evaluation system. Author Contributions Conceptualization: Danni Chen. Data curation: Chunlian Zhang, Danni Chen. PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 20 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies Formal analysis: Chunlian Zhang, Danni Chen. Methodology: Chunlian Zhang. Software: Danni Chen. Supervision: Danni Chen. Validation: Chunlian Zhang, Danni Chen. Visualization: Danni Chen. Writing – original draft: Chunlian Zhang, Danni Chen. Writing – review & editing: Chunlian Zhang. References 1. Xie X., Huo J., Zou H., "Green process innovation, green product innovation, and corporate financial performance: A content analysis method". Journal of Business Research, vol. 101, pp. 697–706, 2019. 2. Roh T., Lee K., Yang J. Y., "How do intellectual property rights and government support drive a firm’s green innovation? The mediating role of open innovation". Journal of Cleaner Production, vol. 317, Arti- cle ID 128422, 2021. 3. Wang Y., Shen T., Chen Y., et al., "CEO environmentally responsible leadership and firm environmental innovation: A socio-psychological perspective". Journal of Business Research, vol. 126, pp. 327–340, 2021. 4. Hao J., He F., "Corporate social responsibility (CSR) performance and green innovation: Evidence from China". Finance Research Letters, vol. 48, Article ID 102889, 2022. 5. Lin R.-J., Tan K.-H., Geng Y., "Market demand, green product innovation, and firm performance: evi- dence from Vietnam motorcycle industry". Journal of Cleaner Production, vol. 40, pp. 101–107, 2013. 6. Rennings K., "Redefining innovation—eco-innovation research and the contribution from ecological economics". Ecological Economics, vol. 32, no. 2, pp. 319–332, 2000. 7. Huang Y.-C., Chen C. T., "Exploring institutional pressures, firm green slack, green product innovation and green new product success: Evidence from Taiwan’s high-tech industries". Technological Fore- casting and Social Change, vol. 174, Article ID 121196, 2022. 8. Razzaq A., Wang Y., Chupradit S., et al., "Asymmetric inter-linkages between green technology innova- tion and consumption-based carbon emissions in BRICS countries using the quantile-on-quantile framework". Technology in Society, vol. 66, Article ID 101656, 2021. 9. 10. Triguero A., Moreno-Monde´ jar L., Davia M. A., "Drivers of different types of eco-innovation in European SMEs". Ecological Economics, vol. 92, pp. 25–33, 2013. Farza K., Ftiti Z., Hlioui Z., et al., "Does it pay to go green? The environmental innovation effect on cor- porate financial performance". Journal of Environmental Management, vol. 300, Article ID 113695, 2021. https://doi.org/10.1016/j.jenvman.2021.113695 PMID: 34649325 11. Cillo V., Petruzzelli A. M., Ardito L., et al., "Understanding sustainable innovation: A systematic literature review". Corporate Social Responsibility & Environmental Management, vol. 26, no. 5, pp. 1012–1025, 2019. 12. Islam T., Islam R., Pitafi A. H., et al., "The impact of corporate social responsibility on customer loyalty: The mediating role of corporate reputation, customer satisfaction, and trust". Sustainable Production and Consumption, vol. 25, pp. 123–135, 2021. 13. Sha Y., Zhang P., Wang Y., et al., "Capital market opening and green innovation——Evidence from Shanghai-Hong Kong stock connect and the Shenzhen-Hong Kong stock connect". Energy Economics, Article ID 106048, 2022. 14. Ren S., He D., Yan J., et al., "Environmental labeling certification and corporate environmental innova- tion: The moderating role of corporate ownership and local government intervention". Journal of Busi- ness Research, vol. 140, pp. 556–571, 2022. 15. Zhao L., Zhang L., Sun J., et al., "Can public participation constraints promote green technological inno- vation of Chinese enterprises? The moderating role of government environmental regulatory enforce- ment". Technological Forecasting and Social Change, vol. 174, Article ID 121198, 2022. 16. Wu B., Fang H., Jacoby G., et al., "Environmental regulations and innovation for sustainability? Moder- ating effect of political connections". Emerging Markets Review, vol. 50, Article ID 100835, 2022. PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 21 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies 17. Yang C.-H., Tseng Y.-H., Chen C.-P., "Environmental regulations, induced R&D, and productivity: Evi- dence from Taiwan’s manufacturing industries". Resource and Energy Economics, vol. 34, no. 4, pp. 514–532, 2012. 18. Zhong Z., Peng B., "Can environmental regulation promote green innovation in heavily polluting enter- prises? Empirical evidence from a quasi-natural experiment in China". Sustainable Production and Con- sumption, vol. 30, pp. 815–828, 2022. 19. Du G., Yu M., Sun C., et al., "Green innovation effect of emission trading policy on pilot areas and neigh- boring areas: An analysis based on the spatial econometric model". Energy Policy, vol. 156, Article ID 112431, 2021. 20. Hu G., Wang X., Wang Y., "Can the green credit policy stimulate green innovation in heavily polluting enterprises? Evidence from a quasi-natural experiment in China". Energy Economics, vol. 98, Article ID 105134, 2021. 21. Yang S., Wang W., Feng D., et al., "Impact of pilot environmental policy on urban eco-innovation". Jour- nal of Cleaner Production, vol. 341, Article ID 130858, 2022. 22. Ren K., Kong Y., Zhang T., et al., "The impact of the pollution permits system on green innovation: Evidence from the county-level data in China". Journal of Cleaner Production, vol. 344, Article ID 130896, 2022. 23. 24. Zhang W., Li G., Guo F., "Does carbon emissions trading promote green technology innovation in China?". Applied Energy, vol. 315, Article ID 119012, 2022. Zhou F., Wang X., "The carbon emissions trading scheme and green technology innovation in China: A new structural economics perspective". Economic Analysis and Policy, vol. 74, pp. 365–381, 2022. 25. Wang H., Qi S., Zhou C., et al., "Green credit policy, government behavior and green innovation quality of enterprises". Journal of Cleaner Production, vol. 331, Article ID 129834, 2022. 26. Wang Y., Li M., "Credit policy and its heterogeneous effects on green innovations". Journal of Financial Stability, vol. 58, Article ID 100961, 2022. 27. Zhang Y., Li X., Xing C., "How does China’s green credit policy affect the green innovation of high pollut- ing enterprises? The perspective of radical and incremental innovations". Journal of Cleaner Produc- tion, vol. 336, Article ID 130387, 2022. 28. Cui J., Dai J., Wang Z., et al., "Does Environmental Regulation Induce Green Innovation? A Panel Study of Chinese Listed Firms". Technological Forecasting and Social Change, vol. 176, Article ID 121492, 2022. 29. 30. Li H., Zhang X., Zhao Y., "ESG and Firm’s Default Risk". Finance Research Letters, vol., Article ID 102713, 2022. Luo Y., Xiong G., Mardani A., "Environmental information disclosure and corporate innovation: The "Inverted U-shaped" regulating effect of media attention". Journal of Business Research, vol. 146, pp. 453–463, 2022. 31. Wang S., Lin W., Zhang Z., et al., "Does the environment information announcement promote green innovation? A quasi-natural experimental evidence from the city-level of China". Ecological Indicators, vol. 136, Article ID 108720, 2022. 32. 33. 34. Zhang S., Zhang M., Qiao Y., et al., "Does improvement of environmental information transparency boost firms’ green innovation? Evidence from the air quality monitoring and disclosure program in China". Journal of Cleaner Production, vol. 357, no., pp. 131921, 2022. Zhang C., Zhou B., Tian X., "Political connections and green innovation: The role of a corporate entrepreneur- ship strategy in state-owned enterprises". Journal of Business Research, vol. 146, pp. 375–384, 2022. Liu J., Zhao M., Wang Y., "Impacts of government subsidies and environmental regulations on green process innovation: A nonlinear approach". Technology in Society, vol. 63, Article ID 101417, 2020. 35. Xiang X., Liu C., Yang M., "Who is financing corporate green innovation?". International Review of Eco- nomics & Finance, vol. 78, pp. 321–337, 2022. 36. Ren S., Sun H., Zhang T., "Do environmental subsidies spur environmental innovation? Empirical evi- dence from Chinese listed firms". Technological Forecasting and Social Change, vol. 173, Article ID 121123, 2021. 37. He K., Chen W., Zhang L., "Senior management’s academic experience and corporate green innova- tion". Technological Forecasting and Social Change, vol. 166, Article ID 120664, 2021. 38. Ullah S., Nasim A., "Do firm-level sustainability targets drive environmental innovation? Insights from BRICS Economies". Journal of Environmental Management, vol. 294, Article ID 112754, 2021. https:// doi.org/10.1016/j.jenvman.2021.112754 PMID: 34265739 39. Alda M., "The environmental, social, and governance (ESG) dimension of firms in which social responsi- ble investment (SRI) and conventional pension funds invest: The mainstream SRI and the ESG inclu- sion". Journal of Cleaner Production, vol. 298, Article ID 126812, 2021. PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 22 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies 40. Debbarma J., Choi Y., "A taxonomy of green governance: A qualitative and quantitative analysis towards sustainable development". Sustainable Cities and Society, vol. 79, Article ID 103693, 2022. 41. Kraus S., Rehman S. U., Garcı´a F. J. S., "Corporate social responsibility and environmental perfor- mance: The mediating role of environmental strategy and green innovation". Technological Forecasting and Social Change, vol. 160, Article ID 120262, 2020. 42. Khan S. A. R., Godil D. I., Jabbour C. J. C., et al., "Green data analytics, blockchain technology for sus- tainable development, and sustainable supply chain practices: evidence from small and medium enter- prises". Annals of Operations Research, pp. 1–25, 2021. 43. Tarmuji I., Maelah R., Tarmuji N., et al., "The Impact of Environmental, Social and Governance Prac- tices (ESG) on Economic Performance: Evidence from ESG Score ". International Conference on Eco- nomics & Finance Research; Vol. 37, no. 3, pp. 67–74,2016. 44. Nekhili M., Boukadhaba A., Nagati H., et al., "ESG performance and market value: the moderating role of employee board representation". International Journal of Human Resource Management, vol. 32, no. 14, pp. 3061–3087, 2021. 45. Atan R., Alam M. M., Said J., et al., "The impacts of environmental, social, and governance factors on firm performance". Management of Environmental Quality: An International Journal, vol. 29, no. 2, pp. 182–194, 2018. 46. Wong W. C., Batten J. A., Ahmad A. H., et al., "Does ESG certification add firm value?". Finance Research Letters, vol. 39, Article ID 101593, 2021. 47. Patel P. C., Pearce J. A., Oghazi P., "Not so myopic: Investors lowering short-term growth expectations under high industry ESG-sales-related dynamism and predictability". Journal of Business Research, vol. 128, pp. 551–563, 2021. 48. Alkaraan F., Albitar K., Hussainey K., et al., "Corporate transformation toward Industry 4.0 and financial performance: The influence of environmental, social, and governance (ESG)". Technological Forecast- ing and Social Change, vol. 175, Article ID 121423, 2022. 49. Becker M. G., Martin F., Walter A., "The power of ESG transparency: The effect of the new SFDR sus- tainability labels on mutual funds and individual investors". Finance Research Letters, Article ID 102708, 2022. 50. 51. Luo D., "ESG, liquidity, and stock returns". Journal of International Financial Markets, Institutions and Money, vol. 78, Article ID 101526, 2022. Lo¨ o¨ f H., Sahamkhadam M., Stephan A., "Is Corporate Social Responsibility investing a free lunch? The relationship between ESG, tail risk, and upside potential of stocks before and during the COVID-19 cri- sis". Finance Research Letters, vol. 46, Article ID 102499, 2022. 52. Giese G., Lee L., Melas D., et al., "Foundations of ESG Investing: How ESG Affects Equity Valuation, Risk, and Performance". Journal of Portfolio Management, vol. 45, no. 05, pp. 69–83, 2019. 53. Broadstock D. C., Matousek R., Meyer M., et al., "Does corporate social responsibility impact firms’ innovation capacity? The indirect link between environmental & social governance implementation and innovation performance". Journal of Business Research, vol. 119, pp. 99–110, 2020. 54. Fatemi A., Glaum M., Kaiser S., "ESG performance and firm value: The moderating role of disclosure". Global Finance Journal, vol. 38, pp. 45–64, 2018. 55. Auer B.R., Schuhmacher F., Do socially (ir)responsible investments pay? New evidence from interna- tional ESG data, (2016) Quarterly Review of Economics and Finance, 59, pp. 51–62 56. Mbanyele W., Huang H., Li Y., et al., "Corporate social responsibility and green innovation: Evidence from mandatory CSR disclosure laws". Economics Letters, vol. 212, Article ID 110322, 2022. 57. ESG and corporate financial performance: the mediating role of green innovation: UK common law ver- sus Germany civil law, Chouaibi S., Chouaibi J., Rossi M. EuroMed Journal of Business, 2022, 17 (1), pp. 46–71 58. Tan Y., Zhu Z., "The effect of ESG rating events on corporate green innovation in China: The mediating role of financial constraints and managers’ environmental awareness". Technology in Society, vol. 68, Article ID 101906, 2022. 59. Yoon B., Chung Y., "The effects of corporate social responsibility on firm performance: A stakeholder approach". Journal of Hospitality and Tourism Management, vol. 37, pp. 89–96, 2018. 60. Albort-Morant G., Leal-Milla´n A., Cepeda-Carrio´n G., "The antecedents of green innovation perfor- mance: A model of learning and capabilities". Journal of Business Research, vol. 69, no. 11, pp. 4912– 4917, 2016. 61. Shen H., Ma Z., "Local Economic Development Pressure, Firm Environmental Performance and Debt Financing". Journal of Financial Research, vol., no. 2, pp. 153–166, 2014. PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 23 / 24 PLOS ONEDo Corporate ESG Scores Improve Green Innovation?Empirical Evidence from Chinese Listed Companies 62. Adomako S, Tran MD. Environmental collaboration, responsible innovation, and firm performance: The moderating role of stakeholder pressure[J]. Business Strategy and the Environment, 2022, 31(4): 1695–1704. 63. Yuan B, Cao X. Do corporate social responsibility practices contribute to green innovation? The mediat- ing role of green dynamic capability[J]. Technology in Society, 2022, 68: 101868. 64. Yan J., "Corporate Social Responsibility Disclosure and Innovation Performance: An Empirical Study Based on Chinese Listed Firms during "the Post- mandatory Period"". Science of Science and Manage- ment of S.& T., vol. 42, no. 1, pp. 57–75, 2021. 65. Cowan K, Guzman F. How CSR reputation, sustainability signals, and country-of-origin sustainability reputation contribute to corporate brand performance: An exploratory study[J]. Journal of business research, 2020, 117: 683–693. 66. Cahan SF, Chen C, Chen L, et al. Corporate social responsibility and media coverage[J]. Journal of Banking & Finance, 2015, 59: 409–422. 67. Tian X, Wang TY. Tolerance for failure and corporate innovation[J]. The Review of Financial Studies, 2014, 27(1): 211–255. 68. He JJ, Tian X. The dark side of analyst coverage: The case of innovation[J]. Journal of Financial Eco- nomics, 2013, 109(3): 856–878. 69. Bardos KS, Ertugrul M, Gao LS. Corporate social responsibility, product market perception, and firm value[J]. Journal of Corporate Finance, 2020, 62: 101588. 70. Zhu D., Zhou X., "Equity Restriction, Managerial Ownership and Enterprise Innovation Efficiency". Nan- kai Business Review, vol. 19, no. 3, pp. 136–144, 2016. 71. Ross S. A., "The Determination of Financial Structure: The Incentive-Signaling Approach". Bell Journal of Economics, vol. 8, no. 1, pp. 23–40, 1977. 72. Cornell B., "ESG Preferences, Risk and Return". European Financial Management, vol. 27, no. 1, pp. 12–19, 2021. 73. Zhai Y, Cai Z, Lin H, et al. Does better environmental, social, and governance induce better corporate green innovation: The mediating role of financing constraints[J]. Corporate Social Responsibility and Environmental Management, 2022, 29(5): 1513–1526. 74. Buallay A., Hamdan R., Barone E., & Hamdan A. (2022). Increasing female participation on boards: Effects on sustainability reporting. International Journal of Finance and Economics, 27(1), 111–124. 75. Song W. H., & Yu H. Y. (2018). Green innovation strategy and green innovation: The roles of green cre- ativity and green organizational identity. Corporate Social Responsibility and Environmental Manage- ment, 25(2), 135–150. 76. Limkriangkrai M., Koh S., Durand R. B., "Environmental, Social, and Governance (ESG) Profiles, Stock Returns, and Financial Policy: Australian Evidence". International Review of Finance, vol. 17, no. 3, pp. 461–471, 2017. 77. Cheng B. T., Ioannou I., Serafeim G., "Corporate social responsibility and access to finance". Strategic management journal, vol. 35, no. 1, pp. 1–23, 2014. 78. 79. The effect of corporate social responsibility and the executive compensation on implicit cost of equity: Evidence from French ESG data, Chouaibi, Y., Rossi, M., Zouari, G., Sustainability (Switzerland), 2021, 13(20), 11510 Tolliver C, Fujii H, Keeley AR, et al. Green innovation and finance in Asia[J]. Asian Economic Policy Review, 2021, 16(1): 67–87. 80. Bi K., Huang P., Wang X. Innovation performance and influencing factors of low-carbon technological innovation under the global value chain: A case of Chinese manufacturing industry, (2016) Technologi- cal Forecasting and Social Change, 111, pp. 275–284. 81. Chouaibi S, Rossi M, Siggia D, et al. Exploring the moderating role of social and ethical practices in the relationship between environmental disclosure and financial performance: evidence from ESG compa- nies[J]. Sustainability, 2022, 14(1): 209. 82. Gomariz M. F. C., Ballesta J. P. S., "Financial reporting quality, debt maturity and investment efficiency". Journal of Banking & Finance, vol. 40, pp. 494–506, 2014. 83. Yuan B., Cao X., "Do corporate social responsibility practices contribute to green innovation? The medi- ating role of green dynamic capability". Technology in Society, vol. 68, Article ID 101868, 2022. 84. Baron R. M., Kenny D. A., "The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations". Journal of Personality and Social Psy- chology, vol. 51, pp. 1173–1182, 1986. https://doi.org/10.1037//0022-3514.51.6.1173 PMID: 3806354 85. Wu D., Zhao Q., Han J., "Corporate Social Responsibility and Technological Innovation: Evidence from China". Nankai Economic Studies, no. 3, pp. 140–160, 2020. PLOS ONE \| https://doi.org/10.1371/journal.pone.0279220 May 25, 2023 24 / 24 PLOS ONE
10.1371_journal.pone.0286311	RESEARCH ARTICLE Machine learning based canine posture estimation using inertial data Marinara MarcatoID, Salvatore TedescoID, Conor O’Mahony, Brendan O’Flynn, Paul Galvin Tyndall National Institute, University College Cork, Cork, Ireland marinara.marcato@tyndall.ie Abstract The aim of this study was to design a new canine posture estimation system specifically for working dogs. The system was composed of Inertial Measurement Units (IMUs) that are commercially available, and a supervised learning algorithm which was developed for differ- ent behaviours. Three IMUs, each containing a 3-axis accelerometer, gyroscope, and mag- netometer, were attached to the dogs’ chest, back, and neck. To build and test the model, data were collected during a video-recorded behaviour test where the trainee assistance dogs performed static postures (standing, sitting, lying down) and dynamic activities (walk- ing, body shake). Advanced feature extraction techniques were employed for the first time in this field, including statistical, temporal, and spectral methods. The most important features for posture prediction were chosen using Select K Best with ANOVA F-value. The individual contributions of each IMU, sensor, and feature type were analysed using Select K Best scores and Random Forest feature importance. Results showed that the back and chest IMUs were more important than the neck IMU, and the accelerometers were more important than the gyroscopes. The addition of IMUs to the chest and back of dog harnesses is recom- mended to improve performance. Additionally, statistical and temporal feature domains were more important than spectral feature domains. Three novel cascade arrangements of Random Forest and Isolation Forest were fitted to the dataset. The best classifier achieved an f1-macro of 0.83 and an f1-weighted of 0.90 for the prediction of the five postures, dem- onstrating a better performance than previous studies. These results were attributed to the data collection methodology (number of subjects and observations, multiple IMUs, use of common working dog breeds) and novel machine learning techniques (advanced feature extraction, feature selection and modelling arrangements) employed. The dataset and code used are publicly available on Mendeley Data and GitHub, respectively. 1 Introduction Animals express their feelings and emotions through behaviour, therefore behavioural moni- toring offers the opportunity to deepen our understanding of animal health and well-being. Ethogram markings have been traditionally used to record and qualify a set of behaviours shown in a particular setting for a determined duration, allowing the discovery of patterns, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Marcato M, Tedesco S, O’Mahony C, O’Flynn B, Galvin P (2023) Machine learning based canine posture estimation using inertial data. PLoS ONE 18(6): e0286311. https://doi.org/10.1371/ journal.pone.0286311 Editor: Mohamed Hammad, Menoufia University, EGYPT Received: November 13, 2022 Accepted: May 12, 2023 Published: June 21, 2023 Copyright: © 2023 Marcato et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The dataset is available from the Mendeley database (10.17632/ mpph6bmn7g.1.) and code is available on GitHub (https://github.com/mmarcato/dog_posture. Funding: This publication has emanated from research supported in part by a grant from The Ireland-Wales INTERREG Programme under the CALIN project [PG; grant number 80885; https:// irelandwales.eu/projects/calin], from Science Foundation Ireland (SFI) [PG, BO; grant number 12/RC/2289-P2; https://www.sfi.ie/]; from SFI and Department of Agriculture, Food and Marine [PG, PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 1 / 28 PLOS ONEBO; grant number 16/RC/3835; https://www.sfi.ie/, https://www.gov.ie/en/organisation/department-of- agriculture-food-and-the-marine/]. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: This research is related to the PhD of Marinara Marcato, which is part of collaborations with the Irish Guide Dogs for the Blind. This does not alter our adherence to PLOS ONE policies on sharing data and materials. Machine learning based canine posture estimation using inertial data similarities and differences between subjects [1, 2]. Accordingly, researchers and experts typi- cally use ethograms to characterise, quantify and monitor animal behaviour in order to evalu- ate well-being and temperament [3]. Ethograms are usually designed for each experimental protocol to include relevant behav- iours considering the subjects and research questions under investigation. Their implementa- tion normally involves manual annotation performed by experts during the observation session, or afterwards through video analysis of recorded sessions [1]. This method is time- consuming and may require expert knowledge, therefore ethogram markings are generally employed in short sessions with a selected group of subjects. In order to overcome the challenges related to obtaining individual-level behavioural data on dogs in the medium and long term, smart canine activity monitors have been developed. The recent evolution of motion sensing, processing power and artificial intelligence technol- ogy has enabled the development of automated systems for behavioural estimation and moni- toring [4–11]. They not only eliminate the need for direct observation and assessment by trained professionals but also remove the subjectivity and error associated with human mark- ings. Moreover, they allow continuous recording with an unprecedented level of detail. Machine learning techniques, such as supervised learning algorithms, have been employed to derive a classification model based on the data provided by inertial measurement units (IMUs) [5–9, 12]. The units are usually composed of a 3-axis accelerometer, gyroscope, and/or magnetometer, providing a reliable, accurate, cost-effective, and energy-efficient solution for motion analysis. Moreover, these sensors are small, lightweight, and low-cost. These have sig- nificant advantages when compared to motion recognition systems based on image analysis from camera systems [13–16] which pose constraints in terms of mobility relative to the cam- era and algorithm complexity. The aim of this study was to develop the first posture estimation system specifically for working dogs to create automatic ethograms considering five canine behaviours (walking, standing, sitting, lying down, and body shake). The data collection was designed to consider the specific application of this system to working dogs. Accordingly, only two common work- ing dog breeds were included and multiple IMUs were positioned to emulate the harness used by dogs. The data preprocessing technique utilised advanced feature extraction methods for the first time in this field. Three novel machine learning architectures were implemented and evaluated to improve the rate of correct detection of less frequent behaviour. The analysis of feature importance indicated that the back was the most influential sensor position and accelerome- ters were the most critical sensor type for posture estimation. The classification results achieved in this study demonstrate the superior performance of this model compared to previ- ous research. The potential applications of canine activity monitoring in the working dog industry which motivated the development of this work are presented in Section 1, while Section 2 examines the state-of-the-art in this area and identifies the research gaps. Finally, Section 3 elaborates on the novelty and highlights of this study. 1.1 Motivation The COVID-19 pandemic has profoundly changed people’s behaviour [17], and impacted dog-owner affective experiences and relationships [18], which in turn, correlate with dog’s physical activity [19]. Smart canine activity monitors have been developed to accurately and reliably monitor activity and health parameters to indicate overall well-being [9, 20, 21]. In order to achieve that, they need to first quantify activity levels and identify important PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 2 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data behaviours. Subsequently, such data could be fed to specific algorithms to extract meaningful information about the targeted application. This technological advancement provides an enormous opportunity not only for pet owners but also for working dog organisations including those involved in guide, assistance, police, search and rescue dogs, etc. These devices could be used on guide dogs to provide their organi- sation with well-being information while they are off-site working with their visually-impaired partners, as a means to address the current challenges in assessing and reporting well-being. They could also enable communication between assistance dogs and their partner’s support sys- tems, such as carers and healthcare professionals, as these dogs are trained to perform specific postures and behaviours to assist their partners. This also applies to search and rescue dogs, who are trained to communicate with their handlers through body postures while working remotely [4, 22, 23], and open-field guard or shepherd dogs [7]. In the case of all types of work- ing dogs, they could also assist in creating automatic scoring for the assessment of behavioural performance [24] and computer-canine training system [10, 25, 26]. Activities estimated by a posture recognition algorithm during a behaviour test could be used to predict ethogram items, rater scoring or ultimately the dog’s probability of success in their training programme [24]. This technology enables a range of applications including assessment of well-being and injury recovery. Canine activity monitors could be used for home activity monitoring of pet dogs with chronic conditions especially those affected by orthopaedic, neuromuscular, or neu- rological illnesses such as osteoarthritis, muscular dystrophy [27] or heart failure [28]. For example, they could provide an objective assessment method of evaluating dogs during an injury recovery process [9] and suffering from osteoarthritis [29, 30]. They could also be used to monitor skin conditions by identifying pruritic behaviours in allergic dogs and the occur- rence of seizures in epileptic dogs [31]. Another possible application of canine posture recognition systems regards the monitoring of well-being as some activity and health parameters have been associated with the occurrence of emotional states and stress in dogs [20, 32]. They could also be utilised for noise cancellation in the interpretation of physiological measurements [4, 33]. Furthermore, they could also be used to measure activity at night, as rest was positively associated with success in apprentice guide dogs [34], and static and inactive behaviours were associated with anxiety and hyper-vig- ilance [35]. 1.2 Related work The work presented in this manuscript builds upon the existing literature on posture recogni- tion systems for dogs. These systems can be classified according to the type of hardware and software technology deployed. The former relates to the method for gathering data for posture classification including inertial (accelerometer and gyroscope) or image (camera) based sys- tems [36]. The latter is responsible for estimating posture based on data from sensors. It can take place in either real-time or non-real-time and be embedded, computer, or cloud-based. Several classification algorithms featuring statistical models [10], supervised machine learn- ing [5, 6, 9] or knowledge engineering [11] techniques have been deployed to identify key canine postures and activities. In particular, this work features an inertial sensing, non-real- time, computer-based posture classification system. The possibility of estimating canine posture using IMUs was first investigated by Ribeiro et al. [22] and Brugarolas et al. [25], and subsequent attempts were introduced later [5, 6, 11]. Since then the design of several posture recognition systems has been reported in the literature using both custom-designed [4, 5, 25] and commercial inertial systems [8, 9, 12, 37, 38]. Some canine activity monitoring devices are currently commercially available [21, 35, 39, 40]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 3 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Table 1 summarises, in chronological order, nine published studies whose main goal was to develop or evaluate a canine posture estimation system based on inertial sensors. It includes information on subjects, sensors and placement, data collection methods, data pre-processing techniques, classification algorithms, as well as train-test split and results. Even though most of the studies present overall accuracy as the main performance measure, comparisons across studies must be carefully made. This metric is highly influenced by class imbalance. In other words, accuracy becomes unrepresentative of the algorithm’s performance as the number of observations in each category becomes dissimilar. Moreover, using data from the same subjects or same breeds in the training and test sets causes the performance metric to be higher compared to using different subjects [5, 37] or breeds [7]. Controlling for these fac- tors considering the final model’s intended application is vital in estimating its real-world per- formance. Several previously reported studies insufficiently reported the number of observations per class and subject, and the criteria used for splitting the original dataset into training and test sets, even though they significantly affect the model’s accuracy. Some research reports failed to include information regarding important methodological details such as the device sampling rate [11], window size [6] and overlap [5, 7, 8, 11], data col- lection methods [12], and feature extraction methods [9]. Insufficient information and incom- patible methodological designs pose great challenges when comparing the performance achieved in different studies and limit their contribution to state-of-the-art knowledge. More- over, Kumpulainen et al. [38] was the only study to publicly share the dataset used for model development. The number of subjects used for developing canine posture recognition algorithms was rather limited in early research; however, larger sample sizes were reported in more recent studies [7, 8, 37, 38]. Gerencser et al. [7] used two breeds of dogs and attained an inter-subject accuracy of 83%, and Kumpulainen et al. [8] used 20 breeds of dogs and achieved an accuracy of 76%. The lower figure in the latter study can be explained as the use of dogs from different breeds introduces more variability negatively impacting that metric. However, it is unclear whether the same subjects were used for training and testing the algorithm, which would posi- tively affect that metric [7, 37]. To address these issues, this study included 41 dogs belonging to common working dog breeds. Additionally, two studies were found whose aim was to evaluate commercially available canine activity monitors [35, 39]. These studies insufficiently describe the algorithm built as their primary objective was to assess and validate the classification model used in such com- mercially available devices, therefore they were not included in Table 1. The first one used Pet- Dialog+ powered by Oggi (Tel Aviv, Israel) and Zoetis (Dublin, Ireland) smart collars on 51 dogs classifying 8 different activities including walking, trotting, canter/galloping, sleeping, static/inactive, eating, drinking, and head shaking. It achieved a mean balanced accuracy of 89% and f1-score of 95% [35]. The second study used Whistle smart collars (MacLean, USA) in a large study involving a canine posture training database containing data from over 2,500 dogs. The system was capa- ble of identifying 14 activities, including nine behaviours: drinking, eating, licking object, lick- ing self, petting, rubbing, scratching, and sniffing; and five postures: lying down, sitting, standing, vigorous, and walking. It achieved a mean balanced accuracy of 85% and macro f1-score of 69% [39] using group cross-fold validation. Importantly, no study was found to develop a posture recognition system specifically for working dogs. This limitation in the state-of-the-art research was addressed in this research by collecting data on common working dog breeds and attaching multiple sensors to a harness similar to the one used by working dogs. None of the classification algorithms presented in the literature experimented with advanced feature extraction methods or employed anomaly PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 4 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Table 1. Canine posture recognition algorithms using IMUs reported in literature. Publication Subjects Sensors & Postures Data Collection Pre-processing Classification Algorithm Train-Test Split / Results [11] [9] [7] [5] [6] [12] [8] [37] Placement 5 Urban Search and Rescue dogs (breed not available) Custom-designed system 2-axis ACC Wither and rump 18 dogs (13 breeds) AX3, Axivity ACC @ 30Hz Neck collar 24 dogs (12 BM, 12 LR) Custom-designed system ACC and GYR @ 100Hz Back 7 dogs (6 LR, 1 KK) Custom-designed system ACC and GYR @ 10Hz Rump, Chest, Abdomen, and Withers 2 dogs (1 LR, 1 KK) Custom-designed system ACC and GYR @ 10 Hz Neck and rump 2 dogs (Pembroke Welsh Corgi, Toy Poodle) AX3, Axivity ACC @ 25Hz Neck Collar 24 dogs (20 breeds) GT9X Link, ActiGraph ACC and GYR @ 100Hz Neck collar 21 dogs (19 breeds) Apple Watch Series 1 ACC and GYR @ 50 Hz Neck collar 3 static: standing, sitting, lying down 1 dynamic: walking Postures repeated 5 times Window size of 2 s and overlap is not reported Calculation of angles from ACCs Algorithm: custom, uses angles of ACCs Training and test sets contain the same dogs from [22] Overall accuracy was 80% 17 behaviours: bark, chew, dig, drink, eat, excrete, jump, lie down, sit, pawing, run, shake, shiver, sniff, urinate, walk, unspecified Sessions lasted 20-40 minutes, freely moving dogs 3 static: sit, lay, stand 4 dynamic: run, trot, walk, bark, search Postures recorded on 2 days, sessions lasted 10min Window size of 1 s and overlap of 0.5 s PCA- based feature extraction (50 components, 95% of variance) Normalisation using ECDF Algorithm: K-nearest Neighbours (KNN) with k = 1 10-fold stratified cross-validation Overlap between training and test frames can be 48% Overall accuracy was 69% Window size of 1 s and overlap is not reported 126 Features: 69 from GYR, 45 from ACC Algorithm: Support Vector Machine (SVM) Hyperparameter optimisation: soft margin (C) and gamma 5-fold cross-validation, controlling for subjects Single dog accuracy: Intra- subject (BM 91%, LR 92%) Intra-breed Inter-subject (BM 70%, LR 73%) Inter-breed (74%) Multiple dog accuracy: Inter-subject 83% 5 static: sitting, standing, lying down, eating off ground, standing on two legs 3 dynamic: walking, climb stairs, down ramp Static postures repeated 5 times for 4s + climb stairs, down ramp Window size of 1 s and overlap is not reported Moving average filter Cascade machine learning algorithm C1. separate static behaviour and classify dynamic behaviour C2. posture classification for static behaviour 5-fold cross validation The test set was 30% of the data Intra-subject accuracy: C1. 92-100% C2. 78-99.5% Inter- subject accuracy: C1. 0-100% C2. NA 5 static: sit, stand, lie, eat off the ground, stand on two legs Static postures repeated 5 times for 4s on 2 days 10 behaviours: walking, eating, sitting, laying, sniffing, running, jumping, drinking, shaking, scratching No information available on data collection procedures 7 behaviours: stand, sit, lie down, walk, trot, gallop, sniff Each task lasted 3 minutes 8 behaviours: sit, lay, stand, walk, trot, run, eat, drink No information available on data collection procedures Window size and overlap are not reported Moving average filter Algorithm: Random Forest (RF), KNN, and Logistic Model Tree (LMT) Two-stage cascade learning technique used: C1. separated postures and transitions (ACC, GYR) C2. classification of 5 postures (ACC) Window size of 1 s and overlap 0.96 s Features from PCA and ECDF Algorithm: K-nearest Neighbours (k = 1) Distance metric: Euclidean Distance (ED), Dynamic Time Warping (DTW), DTW-D proposed by [41] Window size of 2 s and overlap is not reported 27 features created and Forward Feature Selection (FFS) Algorithm: Linear and Quadratic Discriminant Analysis classifier (LDA and QDA) Window size of 1.3 s and overlap of 0.8 s 252 features Models created for small, medium, and large- sized dogs (N = 7 in each group) Algorithm: SVM Hyperparameter optimisation: soft margin (C) and gamma 10-fold cross validation Inter-subject accuracy: C1. 81—92% C2. 92 -100% Training and test sets contain the same dogs Distance metric DTW-D outperformed ED and DTW Mean F1-measure 67% and accuracy is 78% 10-fold cross validation Train-test split criteria missing Accuracies All: (27 features) LDA 74%, QDA 73% FFS: (8 feat) LDA 73%, (5 feat) QDA 76% Leave one subject out Average accuracy 57% (results taken on comparable approach) (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 5 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Table 1. (Continued) Publication Subjects Sensors & Postures Data Collection Pre-processing Classification Algorithm Train-Test Split / Results [38] This study Placement 45 dogs (27 breeds) GT9X Link, ActiGraph ACC and GYR @ 100Hz Neck and back 42 dogs (41 LR, 1 GR) GT9X Link, ActiGraph ACC and GYR @ 100Hz Back, Chest, and Neck 3 static: sit, stand, lie down 4 dynamic: trotting, walk, play, treat-search Each task lasted 3 minutes Window size of 2 s and overlap of 1s 54 features and FFS Algorithm: LDA, QDA, SVM, and Tree 3 static: standing, sitting and lying down 2 dynamic: walking and body shake All postures for 1 minute, body shake spontaneously Window size of 1 s and overlap of 50% Cascade machine learning architectures combining: Random Forests and Isolation Forests Leave one subject out Accuracy using the best classifier (SVM): Back 91.4% and Neck 75.6% Development set 10-fold group cross validation Test set Hold-out Inter-breed Inter-subject f1-weighted 0.90 Breeds: LR = Labrador Retriever, BM = Belgian Malinois, KK = Kai Ken. Sensors: ACC = Accelerometer, GYR = Gyroscope. All sensors are 3-axis except stated otherwise. Features: PCA = Principal Component Analysis, ECDF = empirical cumulative density function. https://doi.org/10.1371/journal.pone.0286311.t001 detection classifiers to identify minority classes. Hence, these were explored in the present work which also addressed issues identified in previous research. In particular, it employed more robust cross-validation techniques and reported comprehensive performance metrics for a more complete evaluation of the model. 1.3 Novelty and contribution The main goal of this work was to build a novel system capable of reliably identifying key canine behaviours in working dogs. In addition to the most common dog postures, namely: walking, sitting, standing, and lying down, it was also of interest to predict body shake as it characterises a coping behaviour shown when dogs experience acute stress [42]. This study comprises four main methodological novelties, advancing the state-of-the-art by: 1. providing the largest open access dataset for dog activity recognition specific to predomi- nant working dog breeds, namely, Labrador Retrievers, Golden Retrievers, and their crosses [43–45]; 2. applying advanced feature extraction methods for the first time in this field; 3. evaluating three novel architectures of machine learning classifiers to address the natural class imbalance, including anomaly detection; 4. utilising data from multiple IMUs (i.e., located by the neck, back, and chest) and investigat- ing the effect of IMU placement in a large dataset. With the advent of FAIR principles [46], a significant improvement has been observed as authors adhere to best practices by sharing raw datasets [37, 38, 47]. Accordingly, this work contributes to the state-of-the-art by providing the annotated dataset on Mendeley Data (link) [48]. The code developed to implement the methodology which yielded the results presented in this study was also made publicly available on GitHub (link). This work proposes a frame- work for future studies to address the methodological issues discussed in the detailed analysis of past studies. We attempt to establish a model to promote the interoperability of research outputs, considering the particular technicalities in the field of dog activity recognition. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 6 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data 2 Methods 2.1 Subjects The subjects were 42 healthy apprentice dogs participating in the assistance dog programme at the Irish Guide Dogs for the Blind (IGDB)’s Training Centre in Cork, Ireland. The IGDB is a charitable organisation that provides guide and service dogs to help persons who are vision impaired and families of children with autism. The dogs were Golden Retrievers (N = 1, Mean Age = 14.6 months), Labrador Retrievers (N = 16, Mean Age = 17.3 month) and Crosses (N = 25, Mean Age = 11.6 months). 2.2 Ethical approval The Health Products Regulatory Authority (HPRA) is the competent authority in Ireland responsible for the implementation of EU legislation (Directive 2010/63/EU) for the protection of animals used for scientific purposes. Practices not likely to cause pain, suffering, distress or lasting harm equivalent to, or higher than, that caused by the introduction of a needle in accor- dance with good veterinary practice fall outside the scope of HPRA Scientific Animal Protec- tion Legislation. Consequently, no special permission from HPRA was required considering that the nature of the present work was non-invasive. The Animal Ethics Experimentation Committee (AEEC) and Social Research Ethics Com- mittee (SREC) at University College Cork (UCC) reviewed and approved this study’s data col- lection procedures as described below under request numbers 2019-007 and 2019-016, respectively. All participants were briefed on the experimental protocol and signed the written informed consent sheet. 2.3 Devices and data Inertial data were gathered during the data collection session by three IMUs GT9X Links (Actigraph, Pensacola, USA) containing a three-axial accelerometer, gyroscope, and magne- tometer. ActiLife software (Actigraph, Pensacola, USA) was used to initialise the devices with a 100Hz sampling rate. The IMUs were attached to fabric straps on the dogs through hook-and-loop fasteners glued to the devices and on the fabric straps (Xsens, Enschede, the Netherlands). The sensors were placed on the dogs’ neck, back, and chest following the placement shown in Fig 1. The IMU Raw datasets were concatenated producing 27 data streams, as follows: 3 IMUs (neck, chest, and back) containing 3 sensors each (accelerometer, gyroscope, magnetometer) with 3 axes each (X, Y, Z). The magnetometer features were removed from the dataset as initial experiments confirmed that they were not as predictive of posture. Therefore, the resulting IMU Raw dataset comprised accelerometer and gyroscope measurements which contained a total of 18 columns. 2.4 Data collection protocol Data collection sessions took place in a room with a test area of 11.5m x 8.5m at the IGDB Training Centre in Cork, Ireland. Firstly, the IMUs were initialised using the local time and placed on the dog to gather inertial data on canine postures. In order to synchronise the IMU data and video recording, the local time was shown at the start of the video recording. Then the behaviour test described in Table 2 was performed to gather information on the key activities. Walking, standing, sitting, and lying down were performed following commands, while body shakes occurred spontaneously. Some subjects were more cooperative than others in PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 7 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Fig 1. IMUs sensor placement on the dogs’ neck, back, and chest. https://doi.org/10.1371/journal.pone.0286311.g001 following the instructions to perform and hold postures for the predetermined period of time. Hence, positive reinforcement methods were administered by the handler using rewards in the form of verbal praise and food to increase adherence to the protocol. 2.5 Data annotation The postures performed by the dog in the video-recorded session were annotated and classi- fied into two types, and five postures, as described in Table 3. Posture Timestamps datasets were created for each video-recorded data collection session second by second. They comprised of the timestamp for the start and finish video recording time for each posture described. The IMU Raw dataset was combined with the Posture Timestamps dataset to form a unified annotated dataset entitled IMU Posture dataset which contains IMU Raw data, dog name and breed, data collection number, and the two labels, namely: posture and type. Transitions and miscellaneous body postures which did not fit any of the previous categories were manually excluded from the dataset. Table 4 shows the number of observations in the IMU Posture data- set per posture and type. Table 2. Behaviour Test Protocol for canine posture data monitoring and acquisition. Sub-test Description Familiarisation Handler walks in 2 steps, lets the dog off-lead, ignores the dog keeping arms closed for 1 min (wait). Walking Handler calls dog by name, praises dog, puts on lead, and walks the dog on a leash for 1 min. Standing Handler commands the dog to stand for about 1 min (stand/wait). Sitting Handler commands the dog to sit/wait for about 1 min (sit/wait). Lying down Handler commands the dog to lie down for about 1 min (down/stay). https://doi.org/10.1371/journal.pone.0286311.t002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 8 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Table 3. Categories of behaviour analysed. Type Dynamic Static Posture Walking, Body shake. Standing, Sitting, Lying down. https://doi.org/10.1371/journal.pone.0286311.t003 2.6 Feature extraction Features were calculated using the IMU Posture dataset considering windows of time where a unique posture was performed. Three parameters control the creation of the window as illus- trated in Fig 2 and described below: • Transition time (t_time): time between different postures: t_time = 0.25s. • Window size (w_size): rolling window size for calculating statistical measures: w_size = 1s. • Window offset (w_offset): sampling offsets in the rolling window: w_offset = 0.5s. The dataset was created by calculating diverse statistical measures on rolling windows con- trolled by the re-sampling hyper-parameters described above, following the standard 50% overlap between subsequent frames [9]. The feature set was obtained with tsfel package [49] which calculated 185 different variables for each of the 18 original raw inertial signals in the IMU Posture dataset, resulting in 3,330 features in the final set. These 185 features calculated by tsfel belonged to 60 types, including 26 spectral, 18 temporal, and 16 statistical features. 2.7 Classification models All classification models were developed using Python (v3.8.1), in particular, tsfel (v0.1.4) was used to extract features, and sklearn (v1.0.2) and imblearn (v0.9.0) libraries were used to implement machine learning techniques. The models were trained on ICHEC’s Kay super- computer nodes comprising of 2 x 20 core 2.4GHz Intel Xeon Gold 6148 processors [50]. The original dataset was split into three sets named development, golden, and test while controlling for the dogs in order to prevent overlap between datasets. The observation counts for each of these datasets after feature extraction using the rolling window are shown in Table 5. The development set contained 36 dogs and comprised 82% of the observations. The test set contained 5 dogs and consisted of 16% of the observations, and was created prioritising dogs who performed all the postures. The golden dataset contained data from the Golden Retriever dog only and represented the remaining 2% of the observations. The only dataset used for training models was the development set, the golden and test sets were used to analyse the intra-breed and inter-breed performance of the different models. Table 4. Type, posture, and the number of observations in the IMU Posture dataset. Type Dynamic Static Total Observations 475,802 1,199,200 Posture Walking Body shake Standing Sitting Lying down https://doi.org/10.1371/journal.pone.0286311.t004 Observations 466,502 9,300 640,700 323,200 235,300 1,675,002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 9 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Fig 2. Rolling window used for selecting data for feature extraction. Hyper-parameters shown include re-sampling transition time (t_time) between postures, window size (w_size) used to calculate the statistical measures, and window offset (w_offset) to create separate observations. https://doi.org/10.1371/journal.pone.0286311.g002 The development set was partitioned to create independent training and validation sets using 10-fold group cross-validation to ensure that those sets did not contain data from the same dogs. The resulting partitioned datasets were used to fit classification models in order to optimise the estimator hyper-parameters. Select K Best method was utilised to select the best features by calculating ANOVA F-value between the features and labels. Scores given were grouped by sensor position and type to esti- mate their importance. Random Forest classifier was deployed to fit the development set using all features selected by Select K Best. Feature importance was grouped by the position and type of the sensor, and also by feature domain and type. A grid search algorithm was utilised to evaluate all combinations of the hyper-parameters shown in Table 6. The grid search algorithm trained estimators individually on the relevant subset of the development set using each of the 1,000 hyper-parameter combinations. In par- ticular, the feature selection algorithm was set to use the top K = 10, 20, 35, 55, and 80 features out of a total of 3,300 initial features extracted by tsfel. It also optimised two hyper-parameters of the Random Forest classifier, namely maximum depth = 3, 5, 7, and 10 and the number of estimators = 25, 50, 100, 250, and 500. The performance metric chosen for optimising the estimators was f1-weighted, as it pro- vides a good evaluation of the model’s ability to predict the correct class in multi-class classifi- cation problems with a natural class imbalance. Accordingly, the grid search algorithm chose the hyper-parameter combination used to create the estimators, which yielded the best f1-weighed score when fitted to the development set. Three novel classifiers were built using the optimal estimators selected by the grid search algorithm. The next subsections outline different architectures used for combining the optimal estima- tors, in order to predict five postures, including three static postures (lying down, sitting, and standing) and two dynamic activities (walking and body shake). Classifier 1 comprised one esti- mator to predict the postures directly. Classifier 2 was a cascade classifier composed of 3 esti- mators, where the first one detected the type of posture (dynamic or static), and the other two further classified the posture. Classifier 3 had two estimators in sequence, where the first one detected body shakes as an anomaly, and the second one classified the remaining 4 postures. Table 5. Number of observations after feature extraction per dataset per posture. Posture Body shake Lying down Sitting Standing Walking Total Development 82 3,884 5,096 9,286 6,418 24,766 https://doi.org/10.1371/journal.pone.0286311.t005 Test 18 384 946 2,158 1,360 4,866 Golden 2 138 156 588 398 1,282 Total 102 4,406 6,198 12,032 8,176 30,914 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 10 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Table 6. Grid search hyper-parameter set for the classifiers. Method Select K Best Random Forest Hyper-parameters k = [10, 20, 35, 55, 80] max_depth = [3, 5, 7, 10] n_estimators = [25, 50, 100, 250, 500] https://doi.org/10.1371/journal.pone.0286311.t006 2.7.1 Classifier 1. Random Forest classifier was fitted to the entire development dataset with all the 5 classes in order to predict posture directly as indicated in Fig 3. 2.7.2 Classifier 2. This classifier comprised three estimators, namely type, static and dynamic, as shown in Fig 4. Each estimator was trained separately, as outlined below: • Type: Random Forest classifier was fitted to the entire development dataset to predict the label ‘type’ as described in Table 3 which comprised of two classes namely, static and dynamic. • Static: Random Forest classifier was fitted to a subset of the dataset containing static postures to predict the classes lying down, sitting, and standing. • Dynamic: Random Forest classifier was fitted to a subset of the dataset containing dynamic postures to predict the classes walking and body shake. At prediction time, they were arranged sequentially in two stages to estimate the final pos- ture as illustrated in the diagram in Fig 4. The class predicted by the type model in the first stage determined which estimator would be used in the second stage. 2.7.3 Classifier 3. This classifier comprised two estimators, namely anomaly and normal as indicated in Fig 5. In the first stage, it identified abnormal behaviour (body shake). Other instances went to the second stage where the normal estimator classified all the other four pos- tures in the second stage. • Anomaly: Isolation Forest, instead of Random Forest, was fitted to the entire dataset to iden- tify the minority class body shake while all other four postures were assigned to the majority class. The hyper-parameter space searched for comprised of the following: number of esti- mators = 25, 50, 100, 250, 500, and contamination = 0.005, 0.01, 0.05, 0.1. Fig 3. Direct estimator used in Classifier 1. https://doi.org/10.1371/journal.pone.0286311.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 11 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Fig 4. Type, static and dynamic estimators used in Classifier 2. https://doi.org/10.1371/journal.pone.0286311.g004 Fig 5. Anomaly and normal estimators used in Classifier 3. https://doi.org/10.1371/journal.pone.0286311.g005 • Normal: Random Forest classifier was fitted to a subset of the dataset containing static pos- tures to predict the classes lying down, sitting, standing, and walking. 2.8 Model performance As discussed, a grid search algorithm was used to find the optimal hyper-parameter combina- tion that would yield the model with the highest f1-weighted using 10-fold cross-validation. The optimal hyper-parameter set, and training and validation times were reported for the model that achieved the highest f1-weighted for each of the estimators composing the classifi- ers. The best estimators were built using the optimal hyper-parameter sets on the entire PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 12 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data development set, and then used for predicting the label on unseen data. Intra-breed and inter- breed classification performance was evaluated using the test set (Labrador Retriever and Crosses), and the results are presented in the golden set (Golden Retriever only), respectively. Besides the optimisation metric f1-weighted, other additional metrics were reported to facilitate the comparison between classifiers presented in this study and classifiers developed in other studies on canine posture recognition. In particular, the following metrics were reported for each of the postures: TPR (True Positive Rate, also known as Sensitivity or Recall), TNR (True Negative Rate, also known as Specificity), Accuracy, PPV (Positive Predictive Value, also known as precision), and f1 score. Hence, the following metrics were reported for each of the classifiers: f1-weighted, f1-macro, and confusion matrix. Confusion matrices were normalised considering the number of true instances in each class, i.e., the total count in the rows of the matrix. 3 Results Section 3.1 shows the grid search’s chosen best model hyper-parameter combination, cross- validation results on the development set and performance on the test set evaluated individu- ally and independently. Feature contribution was evaluated based on the f-classification score created by Select K Best and Random Forest classifier’s feature importance metric in Classifier 1. Section 3.2 presents the grouped figures as an estimate of the collective contribution of each sensor type and position. The best models were deployed as indicated in the classifier diagram in Figs 3–5. They were evaluated on the test set to estimate intra-breed performance in Section 3.3, and the golden set to estimate inter-breed performance in Section 3.4. The performance achieved using the best classifier to predict posture on the combined test and golden sets is shown in Section 3.5. 3.1 Classification models 3.1.1 Classifier 1. The best hyper-parameters chosen for the Direct estimator were K = 80 for Select K Best, and maximum depth = 10, and number of estimators = 250 for Random For- est. The time taken to build the estimator using the development set with 10-fold cross-valida- tion was (in seconds): training time (M = 340.90, SD = 3.62) and validation time (M = 0.19, SD = 0.01). This model achieved an f1-weighted mean of 0.89 ±0.04 on the validation set, and an f1-weighted 0.89 and f1-macro 0.82 on the test set. 3.1.2 Classifier 2. The best hyper-parameters chosen for the Type, Dynamic, and Static estimators are shown in Table 7 along with the training and validation times, and the f1-weighted performance metric for the validation and test sets. Table 7. Classifier 2 model’s hyper-parameter set selected by grid search, the time (in seconds) taken to train and validate the model using the development set, and f1-weighted performance on 10-fold validation sets (mean ± standard deviation) and test set. Estimator Type Dynamic Static Model Hyper-parameters k = 80 max_depth = 10 n_estimators = 100 k = 10 max_depth = 10 n_estimators = 25 k = 80 max_depth = 10 n_estimators = 500 https://doi.org/10.1371/journal.pone.0286311.t007 Training 138.17±1.47 Time Validation 0.12±0.00 F1-weighted Validation 0.95±0.02 1.38±0.05 0.03±0.00 0.98±0.01 471.11±7.92 0.20±0.00 0.93±0.03 Test 0.95 0.99 0.93 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 13 / 28 PLOS ONETable 8. Classifier 3 model’s hyper-parameter set selected by grid search, and f1-weighted performance on 10-fold validation sets (mean ± standard deviation) and test set. Machine learning based canine posture estimation using inertial data Estimator Anomaly Normal Model Hyper-parameters k = 10 n_estimators = 50 contamination = 0.005 k = 80 max_depth = 10 n_estimators = 500 https://doi.org/10.1371/journal.pone.0286311.t008 Training 4.63±0.09 Time Validation 0.18±0.01 F1-weighted Validation 0.99±0.03 685.07±11.61 0.27±0.01 0.90±0.03 Test 0.95 0.90 3.1.3 Classifier 3. The best hyper-parameters chosen by grid search for the Anomaly and Normal estimators are shown in Table 8 along with the training and validation times, and the f1-weighted performance metric for the validation and test set. 3.2 Feature importance Select K Best used f-classification to calculate the f-score and p-value for each feature taking the posture label into account. The score given was then normalised and grouped by sensor position (back, chest, and neck) and sensor type (accelerometer and gyroscope) to estimate their collective contribution to the model. These results are shown in Table 9 and indicate that the most important sensor position is on the back and the most important sensor type is the accelerometer for distinguishing between different postures. Table 10 shows the feature importance given by the Random Forest from Classifier 1 on the 80 features chosen by Select K Best. No features derived from the neck sensor were among the 80 highest-ranked ones. Once again, these results indicate that the most important sensor posi- tion was on the back and the most important sensor type was the accelerometer. Table 11 shows more information on the 80 highest-scoring features chosen by Select K best including their domain and collective importance. They belonged to 26 feature types out of the 60 domains calculated, specifically, there were 13 out of 26 spectral, 7 out of 18 temporal, Table 9. Feature scores calculated by f-classification in Select K Best by IMU position and sensor. Position Back Chest Neck SKB Score 0.45 0.39 0.16 https://doi.org/10.1371/journal.pone.0286311.t009 Type Acc Gyr Acc Gyr Acc Gyr SKB Score 0.26 0.19 0.22 0.17 0.09 0.07 Table 10. Feature importance calculated by Random Forest classifier considering the 80 features previously selected by Select K Best. Position Back Chest RF Importance 0.57 0.43 https://doi.org/10.1371/journal.pone.0286311.t010 Type Acc Gyr Acc Gyr RF Importance 0.37 0.20 0.24 0.19 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 14 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Table 11. Feature importance calculated by Random Forest classifier considering the 80 features selected by Select K Best by domain. Domain (N) Statistical (8) Temporal (5) Spectral (13) Importance Feature Type 0.47 0.32 0.21 Root mean square, median absolute deviation, empirical cumulative distribution function percentile, interquartile range, minimum, standard deviation, histogram, mean absolute deviation Area under the curve, peak to peak distance, negative turning points, entropy, positive turning points wavelet absolute mean, wavelet standard deviation, spectral distance, wavelet energy, linear prediction cepstral coefficients, spectral skewness, spectral decrease, spectral centroid, spectral slope, spectral kurtosis, maximum frequency, wavelet variance, spectral roll-off https://doi.org/10.1371/journal.pone.0286311.t011 and 5 out of 16 statistical features. In order of importance, the domains providing the best fea- tures were the statistical, temporal, and spectral domains. Even though the spectral domain had the biggest number of statistical feature types selected, it provided the least collective importance as calculated by Random Forest. 3.3 Intra-breed evaluation A summary of the classification metrics on the test set for Classifiers 1, 2, and 3 per class, and f1-weighted and f1-macro averages is shown in Tables 12–14, respectively. The three classifiers achieved very similar results. Classifier 3 achieved the best f1-macro as a result of its improved ability to correctly identify body shakes; however, its f1-weighted was slightly lower for stand- ing and lying down. The normalised confusion matrices with the predicted labels on the test set using Classifier 1, 2, and 3 models are shown in Figs 6–8, respectively. The main difference appears in the Table 12. Classification metrics per posture achieved using the best models selected by grid search in Classifier 1 on the test set. Posture Body shake Lying down Sitting Standing Walking f1-weighted f1-macro TPR 0.44 0.93 0.91 0.87 0.92 TNR 1.00 0.99 0.97 0.94 0.95 Accuracy 1.00 0.98 0.96 0.91 0.94 0.89 0.82 PPV 0.62 0.88 0.87 0.92 0.88 https://doi.org/10.1371/journal.pone.0286311.t012 Table 13. Classification metrics per posture achieved using the best models selected by grid search in Classifier 2 on the test set. Posture Body shake Lying down Sitting Standing Walking f1-weighted f1-macro TPR 0.39 0.95 0.91 0.87 0.92 TNR 1.00 0.99 0.97 0.94 0.95 Accuracy 1.00 0.99 0.96 0.91 0.94 0.90 0.81 PPV 0.54 0.88 0.87 0.92 0.88 https://doi.org/10.1371/journal.pone.0286311.t013 F1-score 0.52 0.90 0.89 0.89 0.90 F1-score 0.45 0.91 0.89 0.90 0.90 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 15 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Table 14. Classification metrics per posture achieved using the best models selected by grid search in Classifier 3 on the test set. Posture Body shake Lying down Sitting Standing Walking f1-weighted f1-macro TPR 0.61 0.93 0.91 0.87 0.91 TNR 1.00 0.99 0.97 0.94 0.96 Accuracy 1.00 0.98 0.96 0.91 0.94 0.89 0.83 PPV 0.50 0.87 0.87 0.92 0.89 F1-score 0.55 0.90 0.89 0.89 0.90 https://doi.org/10.1371/journal.pone.0286311.t014 body shake performance as this class was mainly correctly classified by Classifier 3 model only. This improved performance in the minority class did not negatively affect the model’s ability to identify other classes as they showed comparable figures to Classifiers 1 and 2. 3.4 Inter-breed evaluation The golden dataset was used to evaluate the performance of each experimental model trained on data from Labrador Retrievers (pure and Golden Retriever crosses) on dogs from another breed (Golden Retriever). A summary of the classification metrics on the golden set for Classi- fiers 1, 2, and 3 per class, and f1-weighted and f1-macro averages is shown in Tables 15–17, respectively. Classifier 3 again achieved the best f1-macro and f1-weighted metrics, it was the only model capable of correctly classifying body shakes. Fig 6. Confusion matrix for Classifier 1 on the test set. https://doi.org/10.1371/journal.pone.0286311.g006 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 16 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Fig 7. Confusion matrix for Classifier 2 on the test set. https://doi.org/10.1371/journal.pone.0286311.g007 Fig 8. Confusion matrix for Classifier 3 on the test set. https://doi.org/10.1371/journal.pone.0286311.g008 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 17 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Table 15. Classification metrics per posture achieved using the best models selected by grid search in Classifier 1 on the golden set. Posture Body shake Lying down Sitting Standing Walking f1-weighted f1-macro TPR 0.00 0.99 0.97 0.96 0.87 TNR 1.00 1.00 0.99 0.92 0.98 Accuracy 1.00 1.00 0.98 0.94 0.95 0.93 0.75 PPV 0.00 0.99 0.91 0.91 0.96 https://doi.org/10.1371/journal.pone.0286311.t015 Table 16. Classification metrics per posture achieved using the best models selected by grid search in Classifier 2 on the golden set. Posture Body shake Lying down Sitting Standing Walking f1-weighted f1-macro TPR 0.00 0.99 0.97 0.98 0.87 TNR 1.00 1.00 1.00 0.92 0.99 Accuracy 1.00 1.00 1.00 0.95 0.95 0.94 0.76 PPV 0.00 1.00 1.00 0.91 0.97 https://doi.org/10.1371/journal.pone.0286311.t016 Table 17. Classification metrics per posture achieved using the best models selected by grid search in Classifier 3 on the golden set. Posture Body shake Lying down Sitting Standing Walking f1-weighted f1-macro TPR 1.00 0.99 0.97 0.96 0.88 TNR 1.00 1.00 0.99 0.93 0.99 Accuracy 1.00 1.00 0.99 0.95 0.95 0.94 0.85 PPV 0.29 0.99 0.92 0.92 0.97 https://doi.org/10.1371/journal.pone.0286311.t017 F1-score 0.00 0.99 0.94 0.93 0.91 F1-score 0.00 1.00 0.99 0.94 0.92 F1-score 0.44 0.99 0.95 0.94 0.92 The normalised confusion matrices with the predicted labels for the golden set using Classi- fier 1, 2, and 3 models are shown in Figs 9–11, respectively. It can be visually seen that Classi- fier 3 achieves the overall best results in correctly classifying the classes. 3.5 Best classifier Intra-breed and inter-breed evaluation revealed that there were no significant performance differences between Golden Retrievers, Labrador Retrievers, and their crosses. Classifier 3 was selected as the best classifier considering the results from Section 3.1. In order to obtain unified performance metrics for the best model, Classifier 3 was used to calculate performance on the combined test and golden sets. A summary of the key classification metrics per class, and f1-weighted and f1-macro aver- ages is shown in Table 18; and a confusion matrix is shown in Fig 12. These are considered the final results of the present work. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 18 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Fig 9. Confusion matrix for Classifier 1 on the golden set. https://doi.org/10.1371/journal.pone.0286311.g009 Fig 10. Confusion matrix for Classifier 2 on the golden set. https://doi.org/10.1371/journal.pone.0286311.g010 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 19 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Fig 11. Confusion matrix for Classifier 3 on the golden set. https://doi.org/10.1371/journal.pone.0286311.g011 4 Discussion The advantages of using IMUs over camera systems in the field of behaviour monitoring have already been established [51]. In the present work, a novel posture estimation system was developed specifically for working dogs. Accordingly, the data collection protocol designed includes popular working dog breeds and positions multiple IMUs on the harness. The main advantage of the posture estimation system developed in this work is the superior prediction performance achieved as demonstrated through comparison with previous research. The novel machine learning algorithm developed contributes to the state-of-the-art by employing advanced feature extraction techniques and experimenting with three different cascade archi- tectures including anomaly detection models for the first time. This was the first time that advanced feature extraction methods were applied to a canine posture recognition system. Feature extraction was performed with the Python package tsfel Table 18. Classification metrics per posture achieved using the best models selected by grid search in Classifier 3 on the test and golden sets combined. Posture Body shake Lying down Sitting Standing Walking f1-weighted f1-macro TPR 0.65 0.95 0.92 0.89 0.90 TNR 1.00 0.99 0.97 0.94 0.96 Accuracy 1.00 0.99 0.96 0.91 0.95 0.90 0.83 PPV 0.45 0.90 0.88 0.92 0.91 F1-score 0.53 0.92 0.90 0.90 0.91 https://doi.org/10.1371/journal.pone.0286311.t018 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 20 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data Fig 12. Confusion matrix for Classifier 3 on the test and golden sets. https://doi.org/10.1371/journal.pone.0286311.g012 [49] using a rolling window, resulting in a dataset composed of 3,300 features and 30,224 observations. Features were chosen by the Select K Best algorithm based on the scores given by f-classification considering the label posture. The analysis of the collective contributions of the features deriving from each of the three IMUs placed on the dog’s neck, chest, and back revealed the importance of the back and chest sensors. Moreover, only features from the back and chest IMUs were selected by the algorithm, indicating that these are more informative of the targeted postures than the ones derived from the neck sensor. Brugarolas et al . [5] col- lected data from IMUs placed in similar positions on 7 dogs; however, only 2 of these dogs had IMUs placed on their chest and back sensors. Even though data were very limited, they sug- gested the significance of the rump, neck, and chest sensors which was confirmed in this research with a much bigger sample size (N = 42). Accelerometer data were more indicative of posture than gyroscope data as shown in Tables 9 and 10. This finding is in line with previous research [5, 7]. It is important to note that these results are influenced by the types of posture targeted in each study. A grid search was deployed to find the optimal combination of hyper-parameters for the feature selection and classifier as outlined in Section 2.7. In particular, it used a 10-fold group cross-validation technique to calculate the performance in 10 different subsets of the develop- ment set while controlling for dogs to select the hyper-parameters that resulted in the best f1-weighted performance. The main drawback of the grid search algorithm is that it can lead to overfitting as it selects the model that delivers the best mean fold performance and does not take into account the gap between training and test performance or the model complexity. Three classifier architectures were evaluated in order to improve the overall performance in correctly identifying postures despite the naturally unequal distribution of observations per label. Modifying the classifier architecture resulted in an improvement of the f1-score of the minority class (body shake) while maintaining a comparable performance on the majority PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 21 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data classes (lying down, sitting, standing, and walking). This can be particularly observed from Classifier 1 and 3 model results in Tables 12 and 14 on the test set, and Tables 15 and 17 on the golden set. In the latter case, Classifiers 1 and 2 did not appropriately identify the body shake at all, while Classifier 3 correctly labelled all body shakes. In terms of training and validation times, Classifier 1 was the fastest, while Classifiers 2 and 3 took a similar amount of time. One unexpected result was the superior performance of all classifiers on the golden set as compared to the test set. The only class performing less well was body shake as indicated by the f1-scores in Tables 12–14 on the test; and Tables 15–17 on the golden set. Interestingly, Gerencser et al. [7] also reported a similar result as the inter-breed model achieved higher per- formance than the intra-breed, from 70.3% to 73.6%, and 72.6% to 73.5% using only Malinois and Labradors data for training, respectively. This indicates that the variability deriving from the breed difference was less significant than the methodological limitations of the data collec- tion and annotation procedures. Hence, this result suggests that this model can be successfully utilised on not only Labrador Retrievers and crosses with Golden Retrievers but also pure Golden Retrievers as well. 4.1 Comparison with previous work Classifier 3 was chosen as the best classification model in the present work, and its perfor- mance on the test and golden sets combined as shown in Section 3.5 was compared to previous studies in Table 19. The inclusion criteria for such comparison considered only papers that (1) estimated canine posture using IMUs; (2) provided classification metrics per posture (body Table 19. Comparison between Classifier 3 performance and previous studies reporting inter-subject classification performance metrics per posture. The best geo- metric means between TPR and TNR (g-mean) are in bold. TNR 1.00 0.99 0.97 0.94 0.96 1.000 0.913 0.915 0.900 0.969 0.91 0.96 PPV 0.45 0.90 0.88 0.92 0.91 0.795 0.724 0.347 0.916 0.706 0.99 0.83 F1-score G-mean 0.53 0.92 0.90 0.90 0.91 0.851 0.772 0.375 0.850 0.792 0.95 0.89 0.81 0.97 0.95 0.91 0.93 0.957 0.868 0.612 0.845 0.935 0.91 0.95 Publication This study [39] [38] [35] [52] [7] [11] Posture Body shake Lying down Sitting Standing Walking Body shake Lying down Sitting Standing Walking Lying down Sitting Standing Walking Walking Walking Lying down Sitting Standing Walking Lying Down Sitting Standing Walking TPR 0.65 0.95 0.92 0.89 0.90 0.916 0.826 0.409 0.793 0.903 0.837 0.899 0.941 0.945 0.91 0.95 0.722 0.870 0.767 0.975 0.68 0.92 1.00 0.36 https://doi.org/10.1371/journal.pone.0286311.t019 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 22 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data shake, lying down, sitting, standing, and walking); (3) evaluated performance on different dogs due to significant performance difference between intra-subject and inter-subject metrics [5, 7]. Considering the nine research papers in Table 1, five papers were excluded because they did not meet the criteria [5, 6, 8, 9, 12]. However, studies that utilised commercial systems were included [35, 39, 52]. Chambers et al . [39] recommended the use of sensitivity (TPR) and specificity (TNR) to compare the performance of classifiers trained on different datasets in order to control for class imbalance effects. Hence, both metrics were combined in a geometric mean (g-mean) for ease of comparison. The highest g-mean value for each posture was highlighted in bold. Classifier 3 achieved the highest performance for the postures lying down, sitting, and standing on the combined test and golden sets. The best g-means for body shake and walking were reported by Chambers et al. [39] and Den Uijl et al. [52], respectively. Gerencser et al. [7] achieved the highest TPR for walking; however, there was insufficient information to calcu- late other performance metrics. It is important to note that, although Ribeiro et al . [11] reported a high TPR for sitting and standing, there were just a few observations from 5 dogs. Moreover, it was not clear whether the algorithm was built using data from the same dogs that were used in the test set. Classifier 3 achieved the top overall performance on the five postures considering that it attained an arithmetic mean of the f1-score, also known as f1-macro, of 0.83 compared to 0.73 in Chambers et al. [39]. Furthermore, Classifier 3 produced an average TPR of 0.92 compared to 0.905, 0.83, 0.74 in Kumpulainen et al. [38], Gerencser et al. [7], and Ribeiro et al . [11], respectively, taking into account the postures lying down, sitting, standing, and walking. Clas- sifier 3 has outperformed previous models published in the literature for predicting the five postures on dogs that did not belong to the original training set. Conclusively, these results indicate that the proposed system can outperform the state-of-the-art in a real environment. 4.2 Limitations and opportunities As discussed, inaccurate annotations could account for a significant number of errors. Two main methodological limitations were identified to affect the precision of labels. Firstly, data synchronisation was done by showing the standard time on camera at the beginning of the video recording while this could be improved by shaking the sensor in front of the camera. Secondly, data annotation was done second by second, while it could be improved by using some video annotation software to increase the precision of the timestamps. These limitations can result in labels being misaligned with the data. Other two causes of incorrect labels were identified: undefined transitions between two activities and rapid changes from one activity to another. The former case refers to the variabil- ity in the duration of transitions [11], for instance transitioning from walking to standing is nearly instantaneous; however, transitioning from standing to lying down can take a couple of seconds; such an observation could understandably be classified as sitting. The latter is observed in cases where the dog is walking but stands briefly only to resume walking again, such an observation could acceptably be classified as standing. This could explain the misclas- sification between such classes. In order to address the first issue, a fixed transition time between postures was used when extracting features; however, it could still not have been long enough for short postures. The second issue is harder to address as sometimes the posture itself may not be very well defined. It is advisable to utilise video commenting software for data labelling, allowing higher preci- sion in the annotations. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 23 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data The minority class body shake was sometimes incorrectly classified as walking. Such results can be attributed to two main causes. Firstly, it is reasonable to estimate that a significant num- ber of miss-classification cases could have resulted from inaccurate labels. This is because body shakes are very rapid movements typically happening between walking or standing postures. As annotations were made second by second, it is possible that some frames contain mixed postures. Secondly, these are the only dynamic postures while all others are static. It can be assumed that there exists a higher level of similarity between these postures when compared to all other postures, making it harder to distinguish them. Different window sizes could be employed as some classes like body shake are shorter in duration than others. Accordingly, other techniques could be explored including KNN with Dynamic Time Warping [12], and deep learning methods such as convolutional neural net- work (CNN) [53] and long short-term memory (LSTM) motifs [39]. 4.3 Recommendations We suggest that researchers use the framework employed in this research. In particular, they are requested to make their datasets publicly available, describe key methodological details and expand their performance reports. This will allow for easier comparison, reproducibility, and knowledge transfer for the advancement of this field. The following details are crucial and should be reported: data collection materials (sensor type and sampling rate); methods (number of dogs, breeds, context, types of behaviours, dura- tion, repetitions); data pre-processing (window size and overlap); feature extraction (calcula- tion, hyper-parameters); dataset statistics (size of training, development, and validation/test sets); dataset splitting criteria (e.g. breed, subject, number of observations); dataset details (number of observations per behaviour); prediction algorithm (type, hyper-parameters); opti- misation (hyper-parameters and search space); and evaluation (technique, hyper-parameters, metrics, training, and testing time). Future research should also test the performance of new dog activity recognition models on previously published datasets whenever possible to benchmark results. Researchers are encouraged to validate commercial canine activity monitoring devices, in terms of their per- formance in quantifying activity levels and identifying typical behaviours. 5 Conclusions The main goal of this study was to investigate whether the present system comprising neck, back, and chest sensors would surpass the performance reported in published work on canine posture classification systems, namely: body shake, lying down, sitting, standing, and walking. This work contributes to the state-of-the-art knowledge in using IMUs for posture prediction using machine learning, by expanding the current understanding of sensor position, feature extraction, and importance as well as model architecture. The importance of adding IMU sensors to the back and chest for more accurate posture prediction was demonstrated, encouraging the inclusion of IMU sensors on the dog harness for applications where the accuracy of posture predictions is critical, as in the case of some types of working dogs. Advanced feature extraction techniques for time series were been suc- cessfully employed and validated for the first time in the field of canine posture prediction. Feature selection was necessary to remove uninformative features, reducing the complexity of models and processing time. The best-performing model architecture was Classifier 3 which comprised an anomaly detection estimator to identify the minority class (body shake) followed by a classifier to detect the other four postures (lying down, sitting, standing, and walking). The noticeably higher PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 24 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data performance achieved by this novel classifier proves the advantage of combining different clas- sifiers to leverage their respective strengths. In particular, this result indicates that the cascade machine learning architecture including the anomaly detection model significantly improved the detection rate of less frequent behaviour (body shake). The novel canine posture estimator presented here achieved the best overall performance for predicting the five postures compared to previous research. The best performing canine posture estimator previously reported in the literature was presented in Chambers et al. [39]. This novel estimator attained a superior performance in terms of both f1-macro—0.83 versus 0.73—and mean g-mean—0.91 versus 0.84, respectively. Hence, this novel posture estimator system for working dogs provides the most correct predictions of five canine behaviours. Finally, the IMU Posture dataset built using 42 working dogs was made publicly available. This allows the application of other machine learning techniques and a fair performance com- parison between different methods. In particular, future research should investigate the use of other advanced feature selection techniques [54, 55], improve the explainability of models using techniques such as local interpretable model agnostic (LIME) and ELI5 [56, 57], and deep learning techniques (CNNs and LSTM networks) for the prediction of postures. Addi- tional data should be added to the dataset to include different postures and other working dog breeds to create more complete automatic ethograms. These would enable the development of a range of applications to assist working dog organisations during dogs’ training and working lives. Acknowledgments The authors wish to acknowledge the support received from the Irish Guide Dogs for the Blind, the computational resources provided by the Irish Centre for High-End Computing (ICHEC), and data collection support from students at UCC, especially MariaVittoria Lorenzi, Hazel Craven, Conghal Hewson, and Jennifer Kenny. Author Contributions Conceptualization: Marinara Marcato. Data curation: Marinara Marcato. Formal analysis: Marinara Marcato, Salvatore Tedesco. Funding acquisition: Brendan O’Flynn, Paul Galvin. Investigation: Marinara Marcato. Methodology: Marinara Marcato. Resources: Brendan O’Flynn, Paul Galvin. Software: Marinara Marcato. Supervision: Salvatore Tedesco, Conor O’Mahony, Brendan O’Flynn, Paul Galvin. Visualization: Marinara Marcato. Writing – original draft: Marinara Marcato. Writing – review & editing: Marinara Marcato, Salvatore Tedesco, Conor O’Mahony, Bren- dan O’Flynn, Paul Galvin. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 25 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data References 1. Jones AC, Gosling SD. Temperament and Personality in Dogs (Canis Familiaris): A Review and Evalu- ation of Past Research. Applied Animal Behaviour Science. 2005; 95(1-2):1–53. https://doi.org/10. 1016/j.applanim.2005.04.008 2. Gartner MC. Pet Personality: A Review. Personality and Individual Differences. 2015; 75:102–113. https://doi.org/10.1016/j.paid.2014.10.042 3. Bodnariu A. Indicators of Stress and Stress Assessment in Dogs. Lucrari Stiinfice: Medicina Veterinara Timisoara (Scientifical Papers: Veterinary Medicine). 2008. 4. Bozkurt A, Roberts DL, Sherman BL, Brugarolas R, Mealin S, Majikes J, et al. Toward Cyber-Enhanced Working Dogs for Search and Rescue. IEEE Intelligent Systems. 2014; 29(6):32–39. https://doi.org/10. 1109/MIS.2014.77 5. Brugarolas R, Loftin RT, Yang P, Roberts DL, Sherman B, Bozkurt A. Behavior Recognition Based on Machine Learning Algorithms for a Wireless Canine Machine Interface. In: IEEE International Confer- ence on Body Sensor Networks (BSN). Cambridge, MA, USA: IEEE; 2013. 6. Brugarolas R, Roberts D, Sherman B, Bozkurt A. Machine Learning Based Posture Estimation for a Wireless Canine Machine Interface. In: 2013 IEEE Topical Conference on Biomedical Wireless Tech- nologies, Networks, and Sensing Systems. IEEE; 2013. p. 10–12. 7. Gerencse´r L, Va´ sa´rhelyi G, Nagy M, Vicsek T, Miklo´si A. Identification of Behaviour in Freely Moving Dogs (Canis Familiaris) Using Inertial Sensors. PLoS ONE. 2013; 8(10):e77814. https://doi.org/10. 1371/journal.pone.0077814 PMID: 24250745 8. Kumpulainen P, Gizatdinova Y, Vehkaoja A, Valldeoriola A, Somppi S, To¨ rnqvist H, et al. Dog Activity Classification with Movement Sensor Placed on the Collar. In: Proceedings of the Fifth International Conference on Animal-Computer Interaction—ACI’18. Atlanta, Georgia: ACM Press; 2018. p. 1–6. 9. Ladha C, Hammerla N, Hughes E, Olivier P, Ploetz T. Dog’s Life: Wearable Activity Recognition for Dogs. In: Proceedings of the 2013 ACM International Joint Conference on Pervasive and Ubiquitous Computing—UbiComp’13. ACM Press; 2013. p. 415. 10. Majikes J, Brugarolas R, Winters M, Yuschak S, Mealin S. Balancing Noise Sensitivity, Response Latency, and Posture Accuracy for a Computer-Assisted Canine Posture Training System. International Journal of Human-Computer Studies. 2017. 11. Ribeiro C, Ferworn A, Denko M, Tran J. Canine Pose Estimation: A Computing for Public Safety Solu- tion. In: 2009 Canadian Conference on Computer and Robot Vision. IEEE; 2009. p. 37–44. 12. Kiyohara T, Orihara R, Sei Y, Tahara Y, Ohsuga A. Activity Recognition for Dogs Using Off-the-Shelf Accelerometer:. In: Proceedings of the International Conference on Agents and Artificial Intelligence. Lisbon, Portugal: SCITEPRESS—Science and and Technology Publications; 2015. p. 100–110. 13. Pons P, Jaen J, Catala A. Detecting Animals’ Body Postures Using Depth-Based Tracking Systems. In: Proceedings of Measuring Behavior. Dublin, Ireland; 2016. p. 5. 14. Mealin S, Domı´nguez IX, Roberts DL. Semi-Supervised Classification of Static Canine Postures Using the Microsoft Kinect. In: Proceedings of the Third International Conference on Animal-Computer Inter- action. Milton Keynes United Kingdom: ACM; 2016. p. 1–4. 15. Klishkovskaia T, Aksenov A, Sinitca A, Zamansky A, Markelov OA, Kaplun D. Development of Classifi- cation Algorithms for the Detection of Postures Using Non-Marker-Based Motion Capture Systems. Applied Sciences. 2020; 10(11):4028. https://doi.org/10.3390/app10114028 16. Tsai MF, Huang JY. Predicting Canine Posture With Smart Camera Networks Powered by the Artificial Intelligence of Things. IEEE Access. 2020; 8:220848–220857. https://doi.org/10.1109/ACCESS.2020. 3042539 17. Arora T, Grey I. Health Behaviour Changes during COVID-19 and the Potential Consequences: A Mini- Review. Journal of Health Psychology. 2020; 25(9):1155–1163. https://doi.org/10.1177/ 1359105320937053 PMID: 32551944 18. Morgan L, Protopopova A, Birkler RID, Itin-Shwartz B, Sutton GA, Gamliel A, et al. Human–Dog Rela- tionships during the COVID-19 Pandemic: Booming Dog Adoption during Social Isolation. Humanities and Social Sciences Communications. 2020; 7(1):155. https://doi.org/10.1057/s41599-020-00649-x 19. Va¨ a¨ ta¨ ja¨ H, Majaranta P, Cardo´ AV, Isokoski P, Somppi S, Vehkaoja A, et al. The Interplay Between Affect, Dog’s Physical Activity and Dog–Owner Relationship. Frontiers in Veterinary Science. 2021; 8:673407. https://doi.org/10.3389/fvets.2021.673407 PMID: 34957271 20. Jones S, Dowling-Guyer S, Patronek GJ, Marder AR, Segurson D’Arpino S, McCobb E. Use of Acceler- ometers to Measure Stress Levels in Shelter Dogs. Journal of Applied Animal Welfare Science. 2014; 17(1):18–28. https://doi.org/10.1080/10888705.2014.856241 PMID: 24484308 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 26 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data 21. Belda B, Enomoto M, Case BC, Lascelles BDX. Initial Evaluation of PetPace Activity Monitor. The Vet- erinary Journal. 2018; 237:63–68. https://doi.org/10.1016/j.tvjl.2018.05.011 PMID: 30089547 22. Ribeiro C, Ferworn A, Denko M, Tran J, Mawson C. Wireless Estimation of Canine Pose for Search and Rescue. In: 2008 IEEE International Conference on System of Systems Engineering. IEEE; 2008. p. 1–6. 23. Britt WR, Miller J, Waggoner P, Bevly DM, Hamilton JA. An Embedded System for Real-Time Naviga- tion and Remote Command of a Trained Canine. Personal and Ubiquitous Computing. 2011; 15(1):61– 74. https://doi.org/10.1007/s00779-010-0298-4 24. Mealin S, Foster M, Cleghern Z, Bozkurt A, Roberts DL. Using Inertial Measurement Unit Data for Objective Evaluations of Potential Guide Dogs. In: Proceedings of the Seventh International Confer- ence on Animal-Computer Interaction. Milton Keynes United Kingdom: ACM; 2020. p. 1–11. 25. Brugarolas R, Roberts D, Sherman B, Bozkurt A. Posture Estimation for a Canine Machine Interface Based Training System. In: 34th Annual International Conference of the IEEE EMBS. IEEE; 2012. p. 4489–4492. 26. Majikes JJ, Yuschak S, Walker K, Brugarolas R, Mealin S, Foster M, et al. Stimulus Control for Semi- autonomous Computer Canine-Training. In: Mangan M, Cutkosky M, Mura A, Verschure PFMJ, Pres- cott T, Lepora N, editors. Biomimetic and Biohybrid Systems. vol. 10384. Cham: Springer International Publishing; 2017. p. 279–290. 27. Barthe´ le´ my I, Barrey E, Thibaud JL, Uriarte A, Voit T, Blot S, et al. Gait Analysis Using Accelerometry in Dystrophin-Deficient Dogs. Neuromuscular Disorders. 2009; 19(11):788–796. https://doi.org/10.1016/j. nmd.2009.07.014 PMID: 19800232 28. Hansen BD, Lascelles BDX, Keene BW, Adams AK, Thomson AE. Evaluation of an Accelerometer for At-Home Monitoring of Spontaneous Activity in Dogs. American Journal of Veterinary Research. 2007; 68(5):468–475. https://doi.org/10.2460/ajvr.68.5.468 PMID: 17472445 29. Brown DC, Boston RC, Farrar JT. Use of an Activity Monitor to Detect Response to Treatment in Dogs with Osteoarthritis. Journal of the American Veterinary Medical Association. 2010; 237(1):66–70. https://doi.org/10.2460/javma.237.1.66 PMID: 20590496 30. Nuttall T, McEwan N. Objective Measurement of Pruritus in Dogs: A Preliminary Study Using Activity Monitors. Veterinary Dermatology. 2006; 17(5):348–351. https://doi.org/10.1111/j.1365-3164.2006. 00537.x PMID: 16961821 31. Muñana KR, Nettifee JA, Griffith EH, Early PJ, Yoder NC. Evaluation of a Collar-mounted Accelerome- ter for Detecting Seizure Activity in Dogs. Journal of Veterinary Internal Medicine. 2020; 34(3):1239– 1247. https://doi.org/10.1111/jvim.15760 PMID: 32293068 32. Beerda B, Schilder MBH, van Hooff JARAM, de Vries HW. Manifestations of Chronic and Acute Stress in Dogs. Applied Animal Behaviour Science. 1997; 52(3-4):307–319. https://doi.org/10.1016/S0168- 1591(96)01131-8 33. Part CE, Kiddie JL, Hayes WA, Mills DS, Neville RF, Morton DB, et al. Physiological, Physical and Behavioural Changes in Dogs (Canis Familiaris) When Kennelled: Testing the Validity of Stress Param- eters. Physiology & Behavior. 2014; 133:260–271. https://doi.org/10.1016/j.physbeh.2014.05.018 PMID: 24866912 34. 35. Tomkins LM, Thomson PC, McGreevy PD. Behavioral and Physiological Predictors of Guide Dog Suc- cess. Journal of Veterinary Behavior: Clinical Applications and Research. 2011; 6(3):178–187. https:// doi.org/10.1016/j.jveb.2010.12.002 den Uijl I, Go´mez A´ lvarez CB, Bartram D, Dror Y, Holland R, Cook A. External Validation of a Collar- Mounted Triaxial Accelerometer for Second-by-Second Monitoring of Eight Behavioural States in Dogs. PLOS ONE. 2017; 12(11):e0188481. https://doi.org/10.1371/journal.pone.0188481 PMID: 29186154 36. Mealin S, Howell S, Roberts DL. Towards Unsupervised Canine Posture Classification via Depth Shadow Detection and Infrared Reconstruction for Improved Image Segmentation Accuracy. In: Lepora NF, Mura A, Mangan M, Verschure PFMJ, Desmulliez M, Prescott TJ, editors. Biomimetic and Biohy- brid Systems. Lecture Notes in Computer Science. Cham: Springer International Publishing; 2016. p. 155–166. Ferdinandy B, Gerencse´ r L, Corrieri L, Perez P, U´ jva´ ry D, Csizmadia G, et al. Challenges of Machine Learning Model Validation Using Correlated Behaviour Data: Evaluation of Cross-Validation Strategies and Accuracy Measures. PLOS ONE. 2020; 15(7):e0236092. https://doi.org/10.1371/journal.pone. 0236092 PMID: 32687528 37. 38. Kumpulainen P, Cardo´ AV, Somppi S, To¨ rnqvist H, Va¨ a¨ ta¨ ja¨ H, Majaranta P, et al. Dog Behaviour Classi- fication with Movement Sensors Placed on the Harness and the Collar. Applied Animal Behaviour Sci- ence. 2021; 241:105393. https://doi.org/10.1016/j.applanim.2021.105393 39. Chambers RD, Yoder NC, Carson AB, Junge C, Allen DE, Prescott LM, et al. Deep Learning Classifica- tion of Canine Behavior Using a Single Collar-Mounted Accelerometer: Real-World Validation. Animals. 2021; 11(6):1549. https://doi.org/10.3390/ani11061549 PMID: 34070579 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 27 / 28 PLOS ONEMachine learning based canine posture estimation using inertial data 40. Marcato M, Kenny J, O’Riordan R, O’Mahony C, O’Flynn B, Galvin P. Assistance Dog Selection and Performance Assessment Methods Using Behavioural and Physiological Tools and Devices. Applied Animal Behaviour Science. 2022; 254:105691. https://doi.org/10.1016/j.applanim.2022.105691 41. Chen Y, Hu B, Keogh E, Batista GEAPA. DTW-D: Time Series Semi-Supervised Learning from a Single Example. In: Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. Chicago Illinois USA: ACM; 2013. p. 383–391. 42. Beerda B, Schilder MBH, van Hooff AM, de Vries HW. Behavioural and Hormonal Indicators of Enduring Environmental Stress in Dogs. Animal Welfare. 2000; p. 15. 43. Ennik I, Liinamo AE, Leighton E, van Arendonk J. Suitability for Field Service in 4 Breeds of Guide Dogs. Journal of Veterinary Behavior. 2006; 1(2):67–74. https://doi.org/10.1016/j.jveb.2006.06.004 44. Evans KM, Lewis TW, Asher L, Blythe S, Bottomley M, Tootill L, et al. Genetic Evaluation of Traits in a Standardized Behavioral Test for Potential Guide Dog Puppies Using Crossbreed Models. Journal of Veterinary Behavior. 2015; 10(6):459–464. https://doi.org/10.1016/j.jveb.2015.08.004 45. van Rooy D, Wade CM. Behavioral Genetics of Dog Breeds. In: Choe JC, editor. Encyclopedia of Ani- mal Behavior (Second Edition). Oxford: Academic Press; 2019. p. 312–322. 46. Wilkinson MD, Dumontier M, Aalbersberg IJ, Appleton G, Axton M, Baak A, et al. The FAIR Guiding Principles for Scientific Data Management and Stewardship. Scientific Data. 2016; 3(1):160018. https:// doi.org/10.1038/sdata.2016.18 PMID: 26978244 47. Vehkaoja A, Somppi S, To¨ rnqvist H, Valldeoriola Cardo´ A, Kumpulainen P, Va¨ a¨ ta¨ja¨ H, et al. Description of Movement Sensor Dataset for Dog Behavior Classification. Data in Brief. 2022; 40:107822. https:// doi.org/10.1016/j.dib.2022.107822 PMID: 35079615 48. Marcato M, Tedesco S, O’Mahony C, O’Flynn B, Galvin P. Inertial sensor dataset for Dog Posture Rec- ognition. Mendeley Data, V1. 49. Barandas M, Folgado D, Fernandes L, Santos S, Abreu M, Bota P, et al. TSFEL: Time Series Feature Extraction Library. SoftwareX. 2020; 11:100456. https://doi.org/10.1016/j.softx.2020.100456 50. ICHEC. Kay; 2022. https://www.ichec.ie/about/infrastructure/kay. 51. Stiles EK, Palestrini C, Beauchamp G, Frank D. Physiological and Behavioral Effects of Dextroamphet- amine on Beagle Dogs. Journal of Veterinary Behavior. 2011; 6(6):328–336. https://doi.org/10.1016/j. jveb.2011.03.001 52. den Uijl I, Gomez Alvarez C, Dror I, Manning N, Bartram D, Cook A. Validation of a Collar-Mounted Accelerometer That Identifies Eight Canine Behavioural States, Including Those with Dermatologic Sig- nificance. In: Validation of a Collar-Mounted Accelerometer That Identifies Eight Canine Behavioural States, Including Those with Dermatologic Significance. Weybridge, UK; 2016. p. 81–84. 53. Sahoo JP, Prakash AJ, Pławiak P, Samantray S. Real-Time Hand Gesture Recognition Using Fine- Tuned Convolutional Neural Network. Sensors. 2022; 22(3):706. https://doi.org/10.3390/s22030706 PMID: 35161453 54. Patro KK, Jaya Prakash A, Jayamanmadha Rao M, Rajesh Kumar P. An Efficient Optimized Feature Selection with Machine Learning Approach for ECG Biometric Recognition. IETE Journal of Research. 2022; 68(4):2743–2754. https://doi.org/10.1080/03772063.2020.1725663 55. Sinha VK, Patro KK, Pławiak P, Prakash AJ. Smartphone-Based Human Sitting Behaviors Recognition Using Inertial Sensor. Sensors. 2021; 21(19):6652. https://doi.org/10.3390/s21196652 PMID: 34640971 56. 57. Javed AR, Ahmed W, Pandya S, Maddikunta PKR, Alazab M, Gadekallu TR. A Survey of Explainable Artificial Intelligence for Smart Cities. Electronics. 2023; 12(4):1020. https://doi.org/10.3390/ electronics12041020 Javed AR, Khan HU, Alomari MKB, Sarwar MU, Asim M, Almadhor AS, et al. Toward Explainable AI- empowered Cognitive Health Assessment. Frontiers in Public Health. 2023; 11:1024195. https://doi. org/10.3389/fpubh.2023.1024195 PMID: 36969684 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286311 June 21, 2023 28 / 28 PLOS ONE
10.1371_journal.pone.0283491	RESEARCH ARTICLE Weekend effect on 30-day mortality for ischemic and hemorrhagic stroke analyzed using severity index and staffing level Seung Bin KimID 1‡, Bo Mi Lee2‡, Joo Won Park3, Mi Young KwakID 3, Won Mo JangID 4,5 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 1 Interdepartment of Critical Care Medicine, Seoul Metropolitan Government-Seoul National University Boramae Medical Center, Seoul, Republic of Korea, 2 HIRA Research Institute, Health Insurance Review & Assessment Service, Wonju, Republic of Korea, 3 Center for Public Healthcare, National Medical Center, Seoul, Republic of Korea, 4 Department of Public Health and Community Medicine, Seoul Metropolitan Government-Seoul National University Boramae Medical Center, Seoul, Republic of Korea, 5 Department of Health Policy and Management, Seoul National University College of Medicine, Seoul, Republic of Korea ‡ SBK and BML share first authorship on this work. * kmy805@gmail.com (MYK); thomasj@snu.ac.kr (WMJ) OPEN ACCESS Citation: Kim SB, Lee BM, Park JW, Kwak MY, Jang WM (2023) Weekend effect on 30-day mortality for ischemic and hemorrhagic stroke analyzed using severity index and staffing level. PLoS ONE 18(6): e0283491. https://doi.org/ 10.1371/journal.pone.0283491 Editor: Robert Jeenchen Chen, Stanford University School of Medicine, UNITED STATES Received: November 7, 2022 Accepted: March 11, 2023 Published: June 22, 2023 Copyright: © 2023 Kim et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Data cannot be shared publicly because we have used third-party data from National Health Insurance Service, and are not entitled to share the data. Data are available from the Review Board of National Health Insurance Service (contact via NHIS) for researchers who meet the criteria for access to confidential data. Anyone who conducts a joint study with a Korean researcher can access NHIS for customized health information data. Applications for data are available through National Health Insurance Data Sharing website (https:// Abstract Background and purpose Previous studies on the weekend effect—a phenomenon where stroke outcomes differ depending on whether the stroke occurred on a weekend—mostly targeted ischemic stroke and showed inconsistent results. Thus, we investigated the weekend effect on 30-day mor- tality in patients with ischemic or hemorrhagic stroke considering the confounding effect of stroke severity and staffing level. Methods We retrospectively analyzed data of patients hospitalized for ischemic or hemorrhagic stroke between January 1, 2015, and December 31, 2018, which were extracted from the claims database of the National Health Insurance System and the Medical Resource Report by the Health Insurance Review & Assessment Service. The primary outcome measure was 30-day all-cause mortality. Results In total, 278,632 patients were included, among whom 84,240 and 194,392 had a hemor- rhagic and ischemic stroke, respectively, with 25.8% and 25.1% of patients, respectively, being hospitalized during the weekend. Patients admitted on weekends had significantly higher 30-day mortality rates (hemorrhagic stroke 16.84%>15.55%, p<0.0001; ischemic stroke 5.06%>4.92%, p<0.0001). However, in the multi-level logistic regression analysis adjusted for case-mix, pre-hospital, and hospital level factors, the weekend effect remained consistent in patients with hemorrhagic stroke (odds ratio [OR] 1.05, 95% confidence inter- val [CI] 1.00–1.10), while the association was no longer evident in patients with ischemic stroke (OR 1.01, 95% CI 0.96–1.06). PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 1 / 15 PLOS ONEnhiss.nhis.or.kr/bd/ab/bdaba000eng.do), and additional information can be found at a customized health information data webpage (https://nhiss.nhis.or.kr/bd/ab/bdaba032eng.do). Funding: The author(s) received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. Weekend effect on 30-day stroke mortality Conclusions Weekend admission for hemorrhagic stroke was significantly associated with a higher mor- tality rate after adjusting for confounding factors. Further studies are required to understand factors contributing to mortality during weekend admission. Introduction Many studies have shown that the risk of poor clinical outcomes might be higher for patients admitted on weekends than for those admitted on weekdays, a phenomenon called the “weekend effect” [1–5]. In acute stroke management, onset-to-treatment time is critical for both ischemic [3, 6] and hemorrhagic [4, 7] stroke. Since acute stroke can occur at any time, efficient stroke care should always be provided; one system is the “24/7/365 (hours per day/days per week/days per year)” emergency system [8]. However, considerable variations exist in the availability of health- care resources for stroke treatment [9], affecting the clinical outcomes of patients. Several systematic reviews and meta-analyses have been recently performed in an attempt to summarize studies on the weekend effect [6, 10]. One study suggested factors related to ser- vice provision inside and outside the hospital and case-mix factors that may contribute to or modify the weekend effect [11]. In-hospital factors include lower staffing levels during week- ends [2], delayed assessment and management, fewer ward rounds [12], and disparities in resources and expertise [3]. Pre-hospital factors include the timeliness of patient referral and the availability of ambulance service [11]. Case-mix factors include patient characteristics and stroke severity. Most studies investigating weekend effects were conducted on either ischemic stroke [3, 6, 13] or both types of stroke [7, 14, 15]. However, the results varied. Some studies showed no association between weekend admissions and mortality after adjusting for case-mix factors [8, 14, 16, 17]. In contrast, one study found that hemorrhagic stroke patients admitted on week- ends had significantly higher in-hospital mortality rates after adjusting for patient characteris- tics, including comorbidities [7]. Various studies highlighted the unavailability of proper severity-of-illness measures to accurately adjust for the effect of case-mix factors as a major limitation [14]. This limitation, commonly observed in claims-based stroke studies, is particu- larly relevant, as it is a major determinant of stroke outcomes [18, 19]. Ideally, stroke severity should be evaluated using clinical neurological scales such as the National Institutes of Health Stroke Scale (NIHSS). However, the claims-based stroke severity index (SSI) can also be used as a proxy to measure stroke severity [20]. A weekend effect was not observed in a large cohort of patients with ischemic stroke treated at a stroke center, which was designated by the Brain Attack Coalition. This may be attributed to the 24/7/365 access to stroke specialists, nurses handling stroke cases, and the organized sys- tem for delivering care available at stroke centers [16]. However, previous studies did not iden- tify the timeliness of patient transfers to the hospital. In this study, we attempted to adjust for factors affecting healthcare delivery at a pre-hospital stage through variables involved in direct contact with and transferring to severe emergency centers. Several previous studies indicated that the weekend effect remains unclear even after adjust- ing for case-mix factors and that the findings are insufficient, as factors influencing emergency delivery systems were not considered. In this study, we attempted to examine the effect of weekend admission on 30-day mortality of patients with ischemic and hemorrhagic stroke after adjusting for explanatory factors, classified into case-mix, pre-hospital level, and hospital level factors. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 2 / 15 PLOS ONEWeekend effect on 30-day stroke mortality Methods Data source We retrospectively analyzed data extracted from the National Health Insurance System (NHIS) claims database from 2015 to 2018 and the Medical Resource Report by the Health Insurance Review & Assessment Service (HIRA) in 2018. Since almost all payments were based on a fee-for-service system, the NHIS claims database contains specific disease codes and all necessary data for reimbursement. These data include patient socio-demographic information, such as sex, age, health insurance type, residential area, comorbid diseases, diag- nostic tests, procedures, operations and prescriptions, and outcomes (including deaths). We constructed the dataset by adding hospital characteristics (e.g., staff and facility) from the Med- ical Resource Report (HIRA). This study was reviewed and approved by the Institutional Review Board of Seoul National University College of Medicine (IRB No. 07-2021-8). Informed consent was waived owing to the retrospective nature of the study. Study population We included all hospitalized ischemic and hemorrhagic stroke patients who had been admitted to the emergency department between January 1, 2015, and December 31, 2018. We excluded patients aged <20 years and those admitted to clinics using the same cutoffs as previously described [13]. Stroke was diagnosed using the International Classification of Disease 10th Revision (ICD-10) primary diagnosis codes: (1) hemorrhagic stroke (ICD-10 codes I60–I62) and (2) ischemic stroke (ICD-10 code I63) [21]. A single admission episode was defined as hospitalization and discharge during a single day in the same hospital, as multiple billing data could be claimed on a single day at the same hospital owing to separate monthly claims. A total of 286,606 cases were included, and cases with missing values were excluded. Ulti- mately, 278,632 cases were included for analysis. Variables The dependent variable was all-cause mortality within 30 days of each admission for ischemic/ hemorrhagic stroke [13, 17]. We defined the admission date of the first hospitalization for an episode as the index date, and if the date of death was included within 30 days of the index date, the case was classified to have mortality within 30 days. Weekend admission was investi- gated by determining whether patients with ischemic/hemorrhagic stroke were admitted to the emergency department on a Saturday or Sunday. Patient and hospital characteristics were classified as covariates. We classified case-mix and service provision factors as covariates at the in- and pre-hospital levels [11]. Case-mix factors included age (continuous), sex, health insurance type (national health insurance or medical aid), income level quartile (Q1–Q4), SSI, and interventions (medi- cation, procedures, operations) provided to the patient (yes/no) (S1 and S2 Tables) [22–24]. A severe emergency center included a tertiary hospital, regional or local emergency center (>500 hospital beds), or regional cardiocerebrovascular center. The type of contact with a severe emergency center was classified as direct or transferring. Direct contact indicates that a severe emergency patient who required treatment at a center-level institution was sent directly to a severe emergency center. In contrast, transferring means that a severe emergency patient first went to a local emergency medical agency or an unqualified institution. We classified the type of hospital based on the final medical treatment institution if the ini- tial assessment hospital differs from the final treatment hospital within a single day. Hospital level service provision included hospital type (tertiary hospital, general hospital, hospital); bed PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 3 / 15 PLOS ONEWeekend effect on 30-day stroke mortality size (<300 beds, 300–500 beds, 500–1,000 beds, >1,000 beds); ownership (public/private); stroke center (yes/no), designated by the Korea Stroke Society as an institution equipped with proper facilities, staffing, and tools to properly provide core treatment for patients with stroke; intervention volume indicating whether the hospital implemented more than the threshold of annual intervention (yes/no) (S3 Table) [20, 25–33]; number of physicians including neurolo- gists, neurosurgeons, emergency medical personnel, and radiologists (25th quartile: 25, 50th quartile: 32, 75th quartile: 42); and number of nurses (25th quartile: 426, 50th quartile: 724, 75th quartile: 966). We conducted the pre-hospital level analysis based on the 70 units of the catch- ment area classified by the Ministry of Health & Welfare (S1 Fig, S4 Table). Stroke severity index We used the SSI published in Taiwan [34, 35] as a proxy for the NIHSS in the model. The SSI was validated by demonstrating a close correlation between SSI results and actual stroke sever- ity assessed using the NIHSS. The SSI comprises seven claim items: airway suctioning, bacte- rial sensitivity test, general ward stay, intensive care unit stay, nasogastric intubation, osmotherapy, and urinary catheterization [34]. We used the criteria developed by customizing the coefficient values of each of the seven parameters to the Korean HIRA database (S5 Table). The SSI was obtained using the regression coefficients estimated from a multiple linear regres- sion equation in a previous study (S6 Table) [20]. The SSI was validated based on the NIHSS related to the ischemic stroke evaluation index; however, its validity was also confirmed for hemorrhagic stroke [36]. Statistical analysis Continuous variables are summarized as mean and standard deviation, and categorical variables are summarized as frequencies and percentages. Variables were compared between groups using Student’s t-test for continuous variables and the Chi-square tests for categorical variables. We performed hierarchical logistic regression analysis using multi-level models with the gener- alized linear mixed model (GLIMMIX) procedure at three levels, comprising case-mix (patient level), pre-hospital level (contact type), and hospital level variables. In this analysis, we examined their association with weekend admission and 30-day mortality after admission. All statistical analyses were performed using SAS statistical software version 9.3 (SAS Institute Inc., Cary, NC, USA). All p-values were two-sided and considered significant at <0.05. Results Characteristics of patients admitted for stroke In our study, 84,240 and 194,392 patients were diagnosed with hemorrhagic and ischemic stroke, respectively (Table 1). In the hemorrhagic stroke group, the mortality rate for weekend admission was higher than that for weekday admission (16.84%>15.55%; p<0.0001). The rate of female patients (50.90%>49.21%; p<0.0001) and the rate of patients who had received interventions (procedures 14.86%>14.13%; operations 16.75%>15.67%, p<0.0001) were higher in weekend admission than in weekday admission. Regarding the type of contact with the severe emergency center, the rate of direct type was higher in weekend admission than in weekday admission (81.32%>79.81%, p<0.0001). Additionally, the mean SSI score in patients admitted on weekends was higher than that in patients admitted on weekdays (11.13>10.77, p<0.0001). The characteristics of patients with 30-day mortality after admission are presented in Table 2. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 4 / 15 PLOS ONEWeekend effect on 30-day stroke mortality Case-mix variables in hemorrhagic stroke The rate of female patients who died was higher than that among those who survived (51.69%>49.26%, p<0.0001). The average age of those who died was higher than that of the survivors (69.18 years>63.19 years, p<0.0001). The average SSI score of patients who died was higher than that of the survivors (14.4>10.2, p<0.0001). The rate of patients who had not received intervention was higher among those who died than among the survivors (procedures 89.05%>85.05%, p<0.0001; operation 84.86%>83.90%, p = 0.0052). Case-mix variables in ischemic stroke The rate of female patients who died was higher than that among those who survived (53.88%>42.26%, p<0.0001). The average age of patients who died was higher than that of the survivors (78.81>70.65 years, p<0.0001). The average SSI score of patients who died was higher than that of the survivors (12.18>5.72, p<0.0001). The rate of patients who had not received medication was higher among those who died than among the survivors (16.48%> Table 1. Characteristics of the study population. Variables Hemorrhagic stroke (N = 84,240) Ischemic stroke (N = 194,392) Death within 30 days after admission Died Alive Case-mix Sex Male Female Age (mean, SD) Income level Health insurance 1st quartile 2nd quartile 3rd quartile 4th quartile Medical aid SSI score (m, SD) Intervention-Medication* Yes No Intervention-Procedure* Yes No Intervention-Operation* Yes No Pre-hospitallevel Weekend Weekday p-value Weekend Weekday p-value (N = 21,721) (N = 62,519) (N = 50,743) (N = 143,649) 3,658 (16.84) 9,721 (15.55) <0.0001 2,569 (5.06) 7,073 (4.92) <0.0001 18,063 (83.16) 52,798 (84.45) 48,174 (94.94) 136,576 (95.08) 10,666 (49.10) 31,751 (50.79) <0.0001 28,815 (56.79) 82,306 (57.30) <0.0001 11,055 (50.90) 30,768 (49.21) 21,928 (43.21) 61,343 (42.70) 69.18 (15.30) 63.19 (14.55) <0.0001 78.81 (11.11) 70.65 (12.86) <0.0001 4,186 (19.27) 4,101 (18.88) 4,891 (22.52) 7,047 (32.44) 1,496 (6.89) 11.13 (4.61) 11,847 (18.95) 14,155 (22.64) 20,259 (32.40) 4,359 (6.97) 10.77 (4.65) 11,899 (19.03) <0.0001 8,997 (17.73) 8,236 (16.23) 25,475 (17.73) 23,583 (16.42) <0.0001 10,814 (21.31) 30,462 (21.21) 18,680 (36.81) 52,094 (36.26) 4,016 (7.91) 12,035 (8.38) <0.0001 6.08 (4.00) 6.03 (3.96) <0.0001 - - - - 47,610 (93.83) 134,199 (93.42) <0.0001 3,133 (6.17) 9,450 (6.58) 3,227 (14.86) 8,835 (14.13) <0.0001 4,050 (7.98) 11,513 (8.01) <0.0001 18,494 (85.14) 53,684 (85.87) 46,693 (92.02) 132,136 (91.99) 3,639 (16.75) 9,797 (15.67) <0.0001 685 (1.35) 2,022 (1.41) <0.0001 18,082 (83.25) 52,722 (84.33) 50,058 (98.65) 141,627 (98.59) Type of contact with severe emergency center Direct Transferring Hospital level 17,663 (81.32) 49,896 (79.81) <0.0001 39,949 (78.73) 113,494 (79.01) 0.184 4,058 (18.68) 12,623 (20.19) 10,794 (21.27) 30,155 (20.99) PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 (Continued ) 5 / 15 PLOS ONETable 1. (Continued) Variables Type Tertiary hospital General hospital Hospital Bed volume �299 300–499 500–999 �1,000 Ownership Public Private Stroke center Yes No Intervention volume High Low Number of physicians 1st quartile 2nd quartile 3rd quartile 4th quartile Number of nurses 1st quartile 2nd quartile 3rd quartile 4th quartile Weekend effect on 30-day stroke mortality Hemorrhagic stroke (N = 84,240) Ischemic stroke (N = 194,392) Weekend Weekday p-value Weekend Weekday p-value (N = 21,721) (N = 62,519) (N = 50,743) (N = 143,649) 10,215 (47.03) 28,988 (46.37) 0.0304 23,434 (46.18) 67,508 (47.00) 0.0039 11,188 (51.51) 32,477 (51.95) 318 (1.46) 1,054 (1.69) 26,095 (51.43) 72,636 (50.56) 1,217 (2.40) 3,505 (2.44) 1,595 (7.34) 5,202 (8.32) <0.0001 4,891 (9.64) 14,239 (9.91) <0.0001 2,574 (11.85) 7,530 (12.04) 13,063 (60.14) 37,391 (59.81) 4,489 (20.67) 12,396 (19.83) 6,189 (12.20) 16,548 (11.52) 29,008 (57.17) 81,341 (56.62) 10,655 (21.00) 31,521 (21.94) 4,547 (20.93) 12,702 (20.32) 0.0638 11,532 (22.73) 34,084 (23.73) <0.0001 10,806 (79.09) 49,817 (79.68) 39,211 (77.27) 109,565 (76.27) 9,131 (42.04) 26,931 (43.08) 0.0077 22,395 (44.13) 63,480 (44.19) 0.8245 12,590 (57.96) 35,588 (56.92) 28,348 (55.87) 80,169 (55.81) 19,853 (91.40) 56,392 (90.20) <0.0001 44,580(87.85) 125,933(87.67) 0.2691 1,868 (8.60) 6,127 (9.80) 6,163(12.15) 17,716 (12.33) 5,032 (23.17) 5,200 (23.94) 5,732 (26.39) 5,757 (26.50) 4,887 (22.50) 5,707 (26.27) 5,630 (25.92) 5,497 (25.31) 15,171 (24.27) 0.0099 13,034 (25.69) 35,730 (24.87) 0.0001 14,599 (23.35) 16,298 (26.07) 16,451 (26.31) 12,056 (23.76) 33,985 (23.66) 12,570 (24.77) 35,642 (24.81) 13,083 (25.78) 38,292 (26.66) 14,774 (23.63) 0.0086 12,936 (25.49) 35,761 (24.89) <0.0001 16,248 (25.99) 15,910 (25.45) 15,587 (24.93) 12,485 (24.60) 34,311 (23.89) 12,709 (25.05) 36,400 (25.34) 12,613 (24.86) 37,177 (25.88) SD, standard deviation; SSI, stroke severity index. p<0.05 calculated using t-test and χ2 test. * S1 and S2 Tables present definitions of interventions (medication, procedures, and operations) in ischemic/hemorrhagic stroke. The results for ischemic stroke were similar to those for hemorrhagic stroke. In the ischemic stroke group, the mortality rate for weekend admission was higher than that for weekday admission (5.06%>4.92%; p<0.0001). However, regarding the type of contact with the severe emergency center, the rate of direct type was lower on weekend admission than in weekday admission, although not significantly different (78.73%<79.01%, p = 0.184). https://doi.org/10.1371/journal.pone.0283491.t001 5.95%, p<0.0001); however, the rate of patients who had not received intervention (procedures and operation) was lower among those who died than among the survivors (procedures 84.11%<92.41%, p<0.0001; operation 95.59%<98.76%, p<0.0001). Pre-hospital level variables Regarding the type of contact with the severe emergency center, the rate for the transferring type among those who died was higher than among those who survived (hemorrhagic stroke 22.45%>19.30%; ischemic stroke 25.24%>20.85%, p<0.0001). PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 6 / 15 PLOS ONEWeekend effect on 30-day stroke mortality Hospital level variables in hemorrhagic stroke The rate of tertiary hospitals was lower among those who died than among the survivors (43.34%<47.15%, p<0.0001). Moreover, the rate of hospitals with more than 1000 beds (16.13%<20.78%, p<0.0001) and the rate of hospitals with a stroke center (52.52%<58.07%, p<0.0001) were lower in those who died than in those who survived. The rate of hospitals with a high intervention volume was lower in those who died than in those who survived (87.80%< 91.02%, p<0.0001). Regarding staff numbers, the fewer the number of physicians, the higher the mortality rate (p<0.0001). This trend was also observed for the number of nurses (p<0.0001). Hospital level variables in ischemic stroke The rate of tertiary hospitals was lower among those who died than among the survivors (42.45%<47.01%, p<0.0001). Moreover, the rate of hospitals with more than 1000 beds (17.68%<21.91%, p<0.0001) and the rate of hospitals with a stroke center (50.88%<56.08%, Table 2. Characteristics of patients with 30-day mortality after admission. Variable Hemorrhagic stroke (N = 84,240) Ischemic stroke (N = 194,392) Hospitalization Weekday Weekend Case-mix Sex Male Female Age Income level Health insurance 1st quartile 2nd quartile 3rd quartile 4th quartile Medical aid SSI score Intervention-Medication* Yes No Intervention-Procedure* Yes No Intervention-Operation* Yes No Pre-hospital level Died Alive p-value Died Alive p-value (N = 13,379) (N = 70,861) (N = 9,642) (N = 184,750) 9,721 (72.66) 3,658 (27.34) 52,798 (74.51) <0.0001 18,063 (25.49) 7,073 (73.36) 2,569 (26.64) 136,576 (73.92) 0.2152 48,174 (26.08) 6,464 (48.31) 6,915 (51.69) 69.18 (15.3) 2,407 (17.99) 2,343 (17.51) 2,877 (21.50) 4,465 (33.37) 1,287 (9.62) 14.4 (3.55) 35,953 (50.74) <0.0001 34,908 (49.26) 63.19 (14.55) <0.0001 13,678 (19.30) <0.0001 13,605 (19.20) 16,169 (22.82) 22,841 (32.23) 4,568 (6.45) 10.2 (4.52) 4,447 (46.12) 5,195 (53.88) 78.81 (11.11) 1,712 (17.76) 1,417 (14.70) 1,873 (19.43) 3,631 (37.66) 1,009 (10.46) 106,674 (57.74) <0.0001 <0.0001 <0.0001 78,076 (42.26) 70.65 (12.86) 32,760 (17.73) 30,402 (16.46) 39,403 (21.33) 67,143 (36.34) 15,042 (8.14) <0.0001 12.18 (4.64) 5.72 (3.66) <0.0001 - - - - 1,465 (10.95) 10,597 (14.95) <0.0001 11,914 (89.05) 60,264 (85.05) 8,053 (83.52) 1,589 (16.48) 1,532 (15.89) 8,110 (84.11) 173,756 (94.05) <0.0001 10,994 (5.95) 14,031 (7.59) <0.0001 170,719 (92.41) 2,026 (15.14) 11,410 (16.10) 0.0052 425 (4.41) 2,282 (1.24) <0.0001 11,353 (84.86) 59,451 (83.90) 9,217 (95.59) 182,468 (98.76) Type of contact with severe emergency center Direct Transferring Hospital level 10,375 (77.55) 57,184 (80.70) <0.0001 3,004 (22.45) 13,677 (19.30) 7,208 (74.76) 2,434 (25.24) 146,235 (79.15) <0.0001 38,515 (20.85) PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 (Continued ) 7 / 15 PLOS ONETable 2. (Continued) Variable Type Tertiary hospital General hospital Hospital Bed volume �299 300–499 500–999 �1000 Ownership Public Private Stroke center Yes No Intervention volume High Low Number of physicians 1st quartile 2nd quartile 3rd quartile 4th quartile Number of nurses 1st quartile 2nd quartile 3rd quartile 4th quartile Weekend effect on 30-day stroke mortality Hemorrhagic stroke (N = 84,240) Ischemic stroke (N = 194,392) Died Alive p-value Died Alive p-value (N = 13,379) (N = 70,861) (N = 9,642) (N = 184,750) 5,794 (43.34) 7,256 (54.23) 329 (2.46) 1,276 (9.54) 1,787 (13.36) 8,158 (60.98) 2,158 (16.13) 33,409 (47.15) <0.0001 36,409 (51.38) 1,043 (1.47) 5,521 (7.79) 8,317 (11.74) 42,296 (59.69) 14,727 (20.78) <0.0001 2,573 (19.23) 14,670 (20.70) 0.1785 10,806 (80.77) 56,191 (79.30) 7,027 (52.52) 6,352 (47.48) 41,151 (58.07) <0.0001 29,710 (41.93) 11,747 (87.80) 64,498 (91.02) <0.0001 1,632 (12.20) 6,363 (8.98) 3,677 (27.48) 3,260 (24.37) 3,506 (26.21) 2,936 (21.94) 3,560 (26.61) 3,542 (26.47) 3,418 (25.55) 2,859 (21.37) 16,526 (23.32) <0.0001 16,539 (23.34) 18,524 (26.14) 19,272 (27.20) 16,101 (22.72) <0.0001 18,413 (25.98) 18,122 (25.57) 18,225 (25.72) 4,093 (42.45) 5,230 (54.24) 319 (3.31) 1,156 (11.99) 1,322 (13.71) 5,459 (56.62) 1,705 (17.68) 2,229 (23.12) 7,413 (76.88) 4,906 (50.88) 4,736 (49.12) 8,123 (84.25) 1,519 (15.75) 2,915 (30.23) 2,372 (24.60) 2,262 (23.46) 2,093 (21.71) 2,878 (29.85) 2,316 (24.02) 2,378 (24.66) 2,070 (21.47) <0.0001 <0.0001 86,846 (47.01) 93,501 (50.61) 4,403 (2.38) 17,974 (9.73) 21,415 (11.59) 104,890 (55.77) 40,471 (21.91) 43,387 (23.48) 0.4069 141,363 (76.52) 103,611 (56.08) <0.0001 81,139 (43.92) 162,390 (87.90) <0.0001 22,360 (12.10) 45,849 (24.82) 43,669 (23.64) 45,950 (24.87) 49,282 (26.67) 45,819 (24.80) 44,480 (24.08) 46,731 (25.29) 47,720 (25.83) <0.0001 <0.0001 SSI, stroke severity index. Data are presented as mean±standard deviation or n (%). p<0.05 calculated using t-test and χ2 test. * S1 and S2 Tables present definitions of interventions (medication, procedures, and operations) in ischemic/hemorrhagic stroke. https://doi.org/10.1371/journal.pone.0283491.t002 p<0.0001) were lower in those who died than among the survivors. The rate of hospitals with a high intervention volume was lower in those who died than among the survivors (84.25%< 87.90%, p<0.0001). Regarding staff numbers, the fewer the number of physicians and nurses, the higher the mortality rate (p<0.0001 for both professionals). Multi-level logistic regression analysis in hemorrhagic and ischemic stroke Mortality risk in patients with hemorrhagic stroke. Patients admitted on weekends had a significantly higher 30-day mortality risk than those admitted on weekdays (odds ratio [OR] 1.05, 95% confidence interval [CI] 1.00–1.10). Regarding the case-mix variables, older age (OR 1.02; 95% CI 1.02–1.02), medical aid (ref = quartile 4 of health insurance; OR 1.16; 95% CI 1.06–1.27), and a higher SSI score (OR 1.29; 95% CI 1.28–1.30) were associated with a higher mortality risk. The mortality risk of patients who did not receive intervention was higher than PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 8 / 15 PLOS ONEWeekend effect on 30-day stroke mortality that of patients who received the intervention (OR 1.94, 95% CI 1.80–2.09). Furthermore, patients who did not undergo an operation had a higher mortality risk (OR 2.08, 95% CI 1.94– 2.24) than did patients who underwent an operation (Table 3). Regarding pre-hospital level variables, the mortality risk for the transferring type of contact with a severe emergency center was higher than that for the direct type, although not signifi- cantly different (OR 1.06, 95% CI 0.84–1.33). Additionally, the subgroup analysis showed that patients who had the transferring type of contact with a severe emergency center had higher 30-day mortality for weekend admission (OR 1.36; 95% CI 1.06–1.75, S7 Table). Regarding hospital level variables, lower-level hospitals had a higher mortality risk than tertiary hospitals (OR 2.15; 95% CI 1.69–2.73). Hospitals with a bed volume of 500–1,000 had a higher mortality risk than hospitals with �1,000 beds (OR 1.18; 95% CI 1.08–1.35). Hospitals with low inter- vention volumes had a higher mortality risk than those with high intervention volumes (OR 1.45; 95% CI 1.28–1.64). Mortality risk in patients with ischemic stroke. The effect of weekend admission on mortality was not statistically significant in patients with ischemic stroke (OR 1.01, 95% CI Table 3. Multi-level logistic regression analysis of 30-day mortality. Variable Hospitalization Weekend Weekday Case-mix 1st quartile 2nd quartile 3rd quartile 4th quartile Sex Male Female Age Income level Health insurance Medical aid SSI score Intervention-Medication* Yes No Intervention-Procedure* Yes No Intervention-Operation* Yes No Pre-hospital level Type of contact with severe emergency center Direct Transferring Hospital level Hemorrhagic stroke OR 95% CI Ischemic stroke p-value OR 95% CI p-value 1.05 1.00 1.00 0.97 1.02 0.95 1.00 0.99 1.00 1.16 1.29 - - 1.00 1.94 1.00 2.08 1.00 1.06 1.00–1.10† 0.0357 0.92–1.02 1.02–1.02† 0.2424 <0.0001 0.90–1.00 0.93–1.06 0.93–1.04 0.0468 0.8905 0.6442 1.06–1.27† 1.28–1.30† 0.001 <0.0001 - - 1.80–2.09† <0.0001 1.94–2.24† <0.0001 0.84–1.33 0.6528 1.01 1.00 1.00 1.09 1.04 1.09 1.07 1.03 1.00 0.97 1.35 1.00 3.84 1.00 1.13 1.00 1.43 1.00 1.02 0.96–1.06 0.7315 0.99–1.10 1.04–1.04† 1.02–1.16† 0.99–1.15 0.97–1.10 0.90–1.05 1.34–1.36† 0.1058 <0.0001 0.0071 0.071 0.3916 0.4277 <0.0001 3.48–4.23† <0.0001 1.04–1.24† 0.0048 1.27–1.61† <0.0001 0.70–1.49 0.9334 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 (Continued ) 9 / 15 PLOS ONETable 3. (Continued) Variable Hemorrhagic stroke OR 95% CI Ischemic stroke p-value OR 95% CI p-value Weekend effect on 30-day stroke mortality Type Tertiary hospital General hospital Hospital Bed volume �299 300–499 500–999 �1000 Ownership Public Private Stroke center Yes No Intervention volume High Low Number of physicians 1st quartile 2nd quartile 3rd quartile 4th quartile Number of nurses 1st quartile 2nd quartile 3rd quartile 4th quartile 1.00 0.95 2.15 1.15 1.17 1.18 1.00 1.00 0.99 1.00 1.03 1.00 1.45 1.12 1.08 1.08 1.00 0.93 0.97 1.04 1.00 0.88–1.04 1.69–2.73† 0.85–1.55 0.89–1.54 1.03–1.35† 0.2775 <0.0001 0.3717 0.2603 0.0154 0.90–1.10 0.9162 0.97–1.09 0.3533 1.28–1.64† <0.0001 0.95–1.31 0.95–1.22 0.97–1.21 0.77–1.13 0.87–1.09 0.93–1.15 0.1738 0.2507 0.1707 0.4717 0.635 0.495 1.00 0.94 1.55 0.86 0.94 1.09 1.00 1.00 1.03 1.00 0.93 1.00 1.30 1.02 1.10 1.05 1.00 0.93 0.94 0.89 1.00 0.82–1.07 1.22–1.97† 0.54–1.36 0.60–1.48 0.93–1.27 0.3317 0.0004 0.5086 0.8000 0.3080 0.95–1.11 0.5139 0.87–1.00 0.0527 1.09–1.56† 0.0045 0.78–1.35 0.94–1.30 0.923–1.19 0.77–1.11 0.8–1.07 0.80–0.99† 0.8721 0.2467 0.4691 0.4089 0.3308 0.0294 †p<0.05 calculated using logistic regression analysis. OR, odds ratio; CI, confidence interval; SSI, stroke severity index. * S1 and S2 Tables present definitions of interventions (medication, procedures, and operations) in ischemic/hemorrhagic stroke. https://doi.org/10.1371/journal.pone.0283491.t003 0.96–1.06). Regarding case-mix variables, older age (OR 1.04; 95% CI 1.04–1.04), medical aid (ref = quartile 4 of health insurance; OR 1.09; 95% CI 1.02–1.16), and a higher SSI score (OR 1.35; 95% CI 1.34–1.36) had a higher mortality risk. Patients without interventions had a higher mortality risk (medication OR 3.84, 95% CI 3.48–4.23; procedure OR 1.13, 95% CI 1.04–1.24; operation OR 1.43, 95% CI 1.27–1.61) (Table 3). Regarding pre-hospital level variables, the mortality risk for the transferring type of contact with the severe emergency center was higher than that for the direct type, although not signifi- cantly different (OR 1.02; 95% CI 0.70–1.49). Regarding hospital level variables, lower-level hospitals had a higher mortality risk than tertiary hospitals (OR 1.55; 95% CI 1.22–1.97). Hos- pitals with low intervention volumes had a higher mortality risk than those with high interven- tion volumes (OR 1.30; 95% CI 1.09–1.56). The mortality risk regarding staffing numbers (physicians and nurses) was not statistically significant. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 10 / 15 PLOS ONEWeekend effect on 30-day stroke mortality Discussion Our study explored the effect of weekend admission on 30-day mortality in patients with ischemic or hemorrhagic stroke and confirmed explanatory factors influencing the weekend effect. We found higher mortality rates for weekend admission in patients with hemorrhagic stroke, even after adjusting for case-mix and service provision in-/pre-hospital variables. We found that hemorrhagic stroke patients admitted during weekends had a higher risk of death than those admitted on weekdays. Previous studies found that weekend admission for ischemic stroke was associated with higher 30-day all-cause mortality [3, 6, 13]. In contrast, other studies did not find significant associations between weekend admission for ischemic stroke and higher mortality [8, 16, 17, 37, 38]. When confirming the weekend effect for total stroke, the results showed that mortality was significantly higher in patients with hemorrhagic stroke admitted on weekends than in those admitted on weekdays but not in patients with ischemic stroke [7], which is consistent with our findings. However, previous studies failed to adjust for variables, including stroke severity and time from onset to arrival. Empirical evi- dence of the weekend effect on stroke mortality is mixed, with some studies indicating signifi- cantly higher mortality for weekend admissions and others finding no differences [14]. The presence or magnitude of the weekend effect varies based on the types of admission, case-mix factors and illness severity, geographic location, as well as contextual and methodological fac- tors [11]. Thus, we analyzed the weekend effect by classifying confounding factors into service provision hospital level, case-mix, and service provision pre-hospital level variables. Some studies showed that the weekend effect on mortality disappeared after adjusting for stroke severity scores, such as the NHISS and Charlson comorbidity index. The results also varied depending on the severity scale used [3, 7, 8, 13, 14, 16, 17]. This study used the SSI, a claims-based proxy for stroke severity. Patients admitted on weekends had a higher SSI score than those admitted on weekdays, whereas the number of admissions on weekends was lower than that on weekdays (S2 and S3 Figs). A negative correlation was found between the SSI score and the number of admissions by the day of the week, which is consistent with the results of previous studies [17]. This result is thought to be due to the fact that patients who had mild stroke during weekends postponed their hospital visits to a weekday, as described in a previous study [17]; this event is called the “Monday effect.” A review of previous studies [39] that ana- lyzed the number of patients with stroke onset and hospital presentation by the day of the week corroborates this finding; there was minimal difference in the number of patients with stroke onset and hospital presentation for moderate or severe stroke, but a significant differ- ence was observed for mild stroke. However, their results showed that the weekend effect dis- appeared after adjusting for SSI [17]. In our study, although there was a decrease in the effect, it remained statistically significant in the case of hemorrhagic stroke. This could be attributed to factors other than the difference in disease severity at hospitalization, which were deemed important in the weekend effect in hemorrhagic stroke. A subgroup analysis was performed on hemorrhagic stroke to analyze these additional factors, and the results are discussed below. In our study, hospital intervention volume significantly affected mortality, although we could not find an effect of staffing level. The problem of resource allocation between weekdays and weekends has been reported, with physician volume and experience level considered to be important factors [3, 6]. However, the effect of staffing numbers was not significant as shown by the multi-level logistic regression analysis (Table 3). Reduced availability of clinical person- nel on weekends may reduce the quality of care and influence outcomes following stroke [40]. Evidence from previous studies suggests that specialized stroke units, with around-the-clock availability of specialist stroke teams and rapid access to imaging and thrombolysis, reduce variations in the quality of care and outcomes throughout the week [16, 31, 37]. Our study PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 11 / 15 PLOS ONEWeekend effect on 30-day stroke mortality found that intervention volume significantly affected mortality, suggesting that the availability of treatments may be more important than staffing numbers for patient outcomes. Moreover, we found that the healthcare delivery system may influence the weekend effect on patients with hemorrhagic stroke. We conducted a subgroup analysis by dividing patients with hemorrhagic stroke who showed significant weekend effects into weekend and weekday groups after adjusting for various confounding factors (S7 Table). Among hemorrhagic stroke patients admitted on weekends, patients who had transferred with the severe emergency center had higher 30-day mortality after adjusting for the stroke severity. This suggests that patients with hemorrhagic stroke may have problems being directly admitted to the hospital on a week- end. Therefore, it is important to establish an extensive cooperative system with severe emer- gency centers and other medical facilities to ensure the availability of interventions and other resources, even on weekends. Reorganizing stroke care to provide 24/7 access to stroke special- ists, adequate staffing of nurses handling stroke cases on weekends, and an organized system for delivering care may alleviate the weekend effect and save lives [16, 17, 41, 42]. Our study has some limitations. First, we applied the SSI after validation in Korea [20], using a study methodology in Taiwan. Further investigations are required to determine the applicability of the SSI as a proxy for stroke severity in claims databases from other healthcare systems. Second, the diagnosis accuracy of stroke could not be guaranteed from the claims data. However, we expect that the diagnostic consistency in stroke could be generally reliable, and we extracted the population at primary diagnosis [43]. Third, we did not measure other indicators of care quality, such as onset-to-treatment time, time to assessment of rehabilitation, the intensity of rehabilitation therapy, and patient education level. Finally, we investigated the hospital level characteristics such as staffing level and bed number. However, we defined a weekend as Saturday and Sunday only, and we did not include holidays if they fell on week- days. Moreover, we could not examine variables that reflected differences in staffing level, including the number and experience of staff members during weekdays and weekends. Therefore, we could not identify the effect of differences in staffing levels on weekends. In summary, the significance of the weekend effect on mortality was maintained in hemor- rhagic stroke but not in ischemic stroke after adjusting for case-mix, hospital level, and pre- hospital level factors. Further research is required to analyze mortality in weekend and week- day holidays with more detailed variables for patient risk factors, hospital staffing level changes, and emergency delivery systems. Our results suggested that further endeavor is needed to find effective measures, including a systematic approach for mitigating the weekend effect on hemorrhagic stroke. More research is required to explore the proper outcome vari- able (e.g., disability occurrence) for ischemic stroke’s weekend effect. Supporting information S1 Fig. Rationale for using RI*CI for the consolidation of catchment areas. (TIF) S2 Fig. Thirty-day mortality after hospital admission and the mean stroke severity index score according to the day of the week. (TIF) S3 Fig. Number of admissions according to the day of the week. (TIF) S1 Table. Interventions for ischemic stroke. (DOCX) PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 12 / 15 PLOS ONEWeekend effect on 30-day stroke mortality S2 Table. Interventions for hemorrhagic stroke. (DOCX) S3 Table. Threshold for annual intervention. (DOCX) S4 Table. Information on 70 hospital service areas. (DOCX) S5 Table. Seven parameters comprising the stroke severity index and associated explana- tions. (DOCX) S6 Table. Multiple linear regression model for evaluating the stroke severity index. (DOCX) S7 Table. Subgroup analysis for 30-day mortality according to admission on weekends among patients with hemorrhagic stroke. (DOCX) Author Contributions Conceptualization: Won Mo Jang. Data curation: Bo Mi Lee, Joo Won Park, Mi Young Kwak, Won Mo Jang. Formal analysis: Seung Bin Kim, Joo Won Park, Mi Young Kwak, Won Mo Jang. Investigation: Seung Bin Kim, Bo Mi Lee, Mi Young Kwak, Won Mo Jang. Supervision: Mi Young Kwak, Won Mo Jang. Validation: Seung Bin Kim, Bo Mi Lee, Joo Won Park, Mi Young Kwak, Won Mo Jang. Writing – original draft: Seung Bin Kim, Bo Mi Lee. Writing – review & editing: Seung Bin Kim, Bo Mi Lee, Joo Won Park, Mi Young Kwak, Won Mo Jang. References 1. Rudd AG, Hoffman A, Down C, Pearson M, Lowe D. Access to stroke care in England, Wales and Northern Ireland: the effect of age, gender and weekend admission. Age Ageing. 2007; 36: 247–255. https://doi.org/10.1093/ageing/afm007 PMID: 17360793 2. Tung YC, Chang GM, Chen YH. Associations of physician volume and weekend admissions with ische- mic stroke outcome in Taiwan: a nationwide population-based study. Med Care. 2009; 47: 1018–1025. https://doi.org/10.1097/MLR.0b013e3181a81144 PMID: 19648828 3. Saposnik G, Baibergenova A, Bayer N, Hachinski V. Weekends: A dangerous time for having a stroke? Stroke. 2007; 38: 1211–1215. https://doi.org/10.1161/01.STR.0000259622.78616.ea PMID: 17347472 4. Janszky I, Ahnve S, Ljung R. Weekend versus weekday admission and stroke outcome in Sweden from 1968 to 2005. Stroke. 2007; 38: e94–e95. https://doi.org/10.1161/STROKEAHA.107.491449 PMID: 17656663 5. Hasegawa Y, Yoneda Y, Okuda S, Hamada R, Toyota A, Gotoh J, et al. The effect of weekends and hol- idays on stroke outcome in acute stroke units. Cerebrovasc Dis. 2005; 20: 325–331. https://doi.org/10. 1159/000087932 PMID: 16131801 6. Sorita A, Ahmed A, Starr SR, Thompson KM, Reed DA, Dabrh AMA, et al. Off-hour presentation and outcomes in patients with acute ischemic stroke: A systematic review and meta-analysis. Eur J Intern Med. 2014; 25: 394–400. https://doi.org/10.1016/j.ejim.2014.03.012 PMID: 24721584 7. Mekonnen B, Wang G, Rajbhandari-Thapa J, Shi L, Thapa K, Zhang Z, et al. Weekend effect on in-hos- pital mortality for ischemic and hemorrhagic stroke in US rural and urban hospitals. J Stroke PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 13 / 15 PLOS ONEWeekend effect on 30-day stroke mortality Cerebrovasc Dis. 2020; 29: 105106. https://doi.org/10.1016/j.jstrokecerebrovasdis.2020.105106 PMID: 32912515 8. Kim SC, Hong KS, Hwang SI, Kim JE, Kim AR, Cho JY, et al. Weekend admission in patients with acute ischemic stroke is not associated with poor functional outcome than weekday admission. J Clin Neurol. 2012; 8: 265. https://doi.org/10.3988/jcn.2012.8.4.265 PMID: 23323134 9. O’Brien EC, Wu J, Zhao X, Schulte PJ, Fonarow GC, Hernandez AF, et al. Healthcare resource avail- ability, quality of care, and acute ischemic stroke outcomes. J Am Heart Assoc. 2017; 6. https://doi.org/ 10.1161/JAHA.116.003813 PMID: 28159820 10. Hoshijima H, Takeuchi R, Mihara T, Kuratani N, Mizuta K, Wajima Z, et al. Weekend versus weekday admission and short-term mortality: A meta-analysis of 88 cohort studies including 56,934,649 partici- pants. Medicine (Baltimore). 2017; 96: e6685. https://doi.org/10.1097/MD.0000000000006685 PMID: 28445269 11. Chen YF, Armoiry X, Higenbottam C, Cowley N, Basra R, Watson SI, et al. Magnitude and modifiers of the weekend effect in hospital admissions: a systematic review and meta-analysis. BMJ Open. 2019; 9: e025764. https://doi.org/10.1136/bmjopen-2018-025764 PMID: 31164363 12. Mitchell WG, Pande R, Robinson TE, Jones GD, Hou I, Celi LA. The weekend effect for stroke patients admitted to intensive care: A retrospective cohort analysis. PLoS One. 2020; 15: e0234521. https://doi. org/10.1371/journal.pone.0234521 PMID: 32520977 13. Cho KH, Park EC, Nam CM, Choi Y, Shin J, Lee SG. Effect of weekend admission on in-hospital mortal- ity in patients with ischemic stroke: An analysis of Korean nationwide claims data from 2002 to 2013. J Stroke Cerebrovasc Dis. 2016; 25: 419–427. https://doi.org/10.1016/j.jstrokecerebrovasdis.2015.10. 014 PMID: 26654666 14. Angerer S, Buttinger K, Stummer H. The weekend effect revisited: evidence from the Upper Austrian stroke registry. Eur J Health Econ. 2019; 20: 729–737. https://doi.org/10.1007/s10198-019-01035-4 PMID: 30756194 15. Cavallazzi R, Marik PE, Hirani A, Pachinburavan M, Vasu TS, Leiby BE. Association between time of admission to the ICU and mortality: a systematic review and metaanalysis. Chest. 2010; 138: 68–75. https://doi.org/10.1378/chest.09-3018 PMID: 20418364 16. Albright KC, Savitz SI, Raman R, Martin-Schild S, Broderick J, Ernstrom K, et al. Comprehensive stroke centers and the ’weekend effect’: The SPOTRIAS experience. Cerebrovasc Dis. 2012; 34: 424–429. https://doi.org/10.1159/000345077 PMID: 23207423 17. Hsieh CY, Lin HJ, Chen CH, Li CY, Chiu MJ, Sung SF. “Weekend effect” on stroke mortality revisited: Application of a claims-based stroke severity index in a population-based cohort study. Medicine (Balti- more). 2016; 95: e4046. https://doi.org/10.1097/MD.0000000000004046 PMID: 27336904 18. Rost NS, Bottle A, Lee JM, Randall M, Middleton S, Shaw L, et al. Stroke severity is a crucial predictor of outcome: An international prospective validation study. J Am Heart Assoc. 2016; 5. https://doi.org/10. 1161/JAHA.115.002433 PMID: 26796252 19. Weimar C, Ko¨ nig IR, Kraywinkel K, Ziegler A, Diener HC. Age and National Institutes of Health Stroke Scale score within 6 hours after onset are accurate predictors of outcome after cerebral ischemia: Development and external validation of prognostic models. Stroke. 2004; 35: 158–162. https://doi.org/ 10.1161/01.STR.0000106761.94985.8B PMID: 14684776 20. Shim DH, Kim Y, Roh J, Kang J, Park KP, Cha JK, et al. Hospital volume threshold associated with higher survival after endovascular recanalization therapy for acute ischemic stroke. J Stroke. 2020; 22: 141–149. https://doi.org/10.5853/jos.2019.00955 PMID: 32027799 21. Kokotailo RA, Hill MD. Coding of stroke and stroke risk factors using International Classification of Dis- eases, revisions 9 and 10. Stroke. 2005; 36: 1776–1781. https://doi.org/10.1161/01.STR.0000174293. 17959.a1 PMID: 16020772 22. Kim JY, Lee KJ, Kang J, Kim BJ, Han MK, Kim SE, et al. Development of stroke identification algorithm for claims data using the multicenter stroke registry database. PLoS One. 2020; 15: e0228997. https:// doi.org/10.1371/journal.pone.0228997 PMID: 32059039 23. Lee WK, Oh CW, Lee H, Lee KS, Park H. Factors influencing the incidence and treatment of intracranial aneurysm and subarachnoid hemorrhage: time trends and socioeconomic disparities under an univer- sal healthcare system. J Neurointerv Surg. 2019; 11: 159–165. https://doi.org/10.1136/neurintsurg- 2018-013799 PMID: 29934441 24. Kim JY, Kang K, Kang J, Koo JS, Kim DH, Kim BJ, et al. Executive Summary of Stroke statistics in Korea 2018: a report from the Epidemiology Research Council of the Korean Stroke Society. Journal of Stroke. 2019; 21(1):42–59. https://doi.org/10.5853/jos.2018.03125 PMID: 30558400 25. Adamczyk P, Attenello F, Wen G, He S, Russin J, Sanossian N, et al. Mechanical thrombectomy in acute stroke: Utilization variances and impact of procedural volume on inpatient mortality. J Stroke PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 14 / 15 PLOS ONEWeekend effect on 30-day stroke mortality Cerebrovasc Dis. 2013; 22: 1263–1269. https://doi.org/10.1016/j.jstrokecerebrovasdis.2012.08.007 PMID: 23017430 26. Wang ZL, Gao BL, Li TX, Cai DY, Zhu LF, Xue JY, et al. Outcomes of middle cerebral artery angioplasty and stenting with Wingspan at a high-volume center. Neuroradiology. 2016; 58: 161–169. https://doi. org/10.1007/s00234-015-1611-8 PMID: 26515072 27. Kurogi R, Kada A, Ogasawara K, Kitazono T, Sakai N, Hashimoto Y, et al. Effects of case volume and comprehensive stroke center capabilities on patient outcomes of clipping and coiling for subarachnoid hemorrhage. J Neurosurg. 2020; 134: 929–939. https://doi.org/10.3171/2019.12.JNS192584 PMID: 32168489 28. Phillips P, Poku E, Essat M, Woods HB, Goka EA, Kaltenthaler EC, et al. Systematic review of carotid artery procedures and the volume–outcome relationship in Europe. Br J Surg. 2017; 104: 1273–1283. https://doi.org/10.1002/bjs.10593 PMID: 28632941 29. Solomon RA, Mayer SA, Tarmey JJ. Relationship between the volume of craniotomies for cerebral aneurysm performed at New York State hospitals and in-hospital mortality. Stroke. 1996; 27: 13–17. https://doi.org/10.1161/01.str.27.1.13 PMID: 8553389 30. Akbarian-Tefaghi H, Kalakoti P, Sun H, Sharma K, Thakur JD, Patra DP, et al. Impact of hospital case- load and elective admission on outcomes after extracranial-intracranial bypass surgery. World Neuro- surg. 2017; 108: 716–728. https://doi.org/10.1016/j.wneu.2017.09.082 PMID: 28943420 31. Rush B, Romano K, Ashkanani M, McDermid RC, Celi LA. Impact of hospital case-volume on subarach- noid hemorrhage outcomes: A nationwide analysis adjusting for hemorrhage severity. J Crit Care. 2017; 37: 240–243. https://doi.org/10.1016/j.jcrc.2016.09.009 PMID: 27663296 32. Johnston SC. Effect of endovascular services and hospital volume on cerebral aneurysm treatment out- comes. Stroke. 2000; 31: 111–117. https://doi.org/10.1161/01.str.31.1.111 PMID: 10625724 33. Boogaarts HD, van Amerongen MJ, de Vries J, Westert GP, Verbeek ALM, Grotenhuis JA, et al. Case- load as a factor for outcome in aneurysmal subarachnoid hemorrhage: a systematic review and meta- analysis: A systematic review. J Neurosurg. 2014; 120: 605–611. https://doi.org/10.3171/2013.9. jns13640 PMID: 24093633 34. Sung SF, Hsieh CY, Kao Yang YH, Lin HJ, Chen CH, Chen YW, et al. Developing a stroke severity index based on administrative data was feasible using data mining techniques. J Clin Epidemiol. 2015; 68: 1292–1300. https://doi.org/10.1016/j.jclinepi.2015.01.009 PMID: 25700940 35. Sung SF, Hsieh CY, Lin HJ, Chen YW, Chen CH, Kao Yang YH, et al. Validity of a stroke severity index for administrative claims data research: a retrospective cohort study. BMC Health Serv Res. 2016; 16: 509. https://doi.org/10.1186/s12913-016-1769-8 PMID: 27660046 36. Hung LC, Sung SF, Hsieh CY, Hu YH, Lin HJ, Chen YW, et al. Validation of a novel claims-based stroke severity index in patients with intracerebral hemorrhage. J Epidemiol. 2017; 27: 24–29. https://doi.org/ 10.1016/j.je.2016.08.003 PMID: 28135194 37. Albright KC, Raman R, Ernstrom K, Hallevi H, Martin-Schild S, Meyer BC, et al. Can comprehensive stroke centers erase the ’weekend effect’? Cerebrovasc Dis. 2009; 27: 107–113. https://doi.org/10. 1159/000177916 PMID: 19039213 38. Kazley AS, Hillman DG, Johnston KC, Simpson KN. Hospital care for patients experiencing weekend vs weekday stroke: A comparison of quality and aggressiveness of care. Arch Neurol. 2010; 67: 39–44. https://doi.org/10.1001/archneurol.2009.286 PMID: 20065127 39. Fang J, Saposnik G, Silver FL, Kapral MK. Investigators of the Registry of the Canadian Stroke Net- work. Association between weekend hospital presentation and stroke fatality. Neurology. 2010 Nov 2; 75(18):1589–96. https://doi.org/10.1212/WNL.0b013e3181fb84bc PMID: 21041782 40. Saad A, Adil MM, Patel V, Owada K, Winningham MJ, Nahab F. Clinical outcomes after thrombectomy for acute ischemic stroke on weekends versus weekdays. J Stroke Cerebrovasc Dis. 2014; 23: 2708– 2713. https://doi.org/10.1016/j.jstrokecerebrovasdis.2014.06.006 PMID: 25440362 41. McKinney JS, Deng Y, Kasner SE, Kostis JB. Myocardial Infarction Data Acquisition System (MIDAS 15) Study Group. Comprehensive stroke centers overcome the weekend versus weekday gap in stroke treatment and mortality. Stroke. 2011; 42: 2403–2409. https://doi.org/10.1161/STROKEAHA.110. 612317 PMID: 21868723 42. Bray BD, Ayis S, Campbell J, Cloud GC, James M, Hoffman A, et al. Associations between stroke mor- tality and weekend working by stroke specialist physicians and registered nurses: Prospective multicen- tre cohort study. PLoS Med. 2014; 11: e1001705. https://doi.org/10.1371/journal.pmed.1001705 PMID: 25137386 43. Kim RB, Kim BG, Kim YM, Seo JW, Lim YS, Kim HS, et al. Trends in the incidence of hospitalized acute myocardial infarction and stroke in Korea, 2006–2010. J Korean Med Sci. 2013; 28: 16–24. https://doi. org/10.3346/jkms.2013.28.1.16 PMID: 23341707 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283491 June 22, 2023 15 / 15 PLOS ONE
10.1371_journal.pone.0284380	RESEARCH ARTICLE Organic dry pea (Pisum sativum L.): A sustainable alternative pulse-based protein for human health Dil ThavarajahID Nathan JohnsonID Pushparajah Thavarajah1 1, Tristan LawrenceID 1, Lucas Boatwright1, Nathan Windsor1, 1, Joshua Kay1, Emerson Shipe1, Shiv Kumar2, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Thavarajah D, Lawrence T, Boatwright L, Windsor N, Johnson N, Kay J, et al. (2023) Organic dry pea (Pisum sativum L.): A sustainable alternative pulse-based protein for human health. PLoS ONE 18(4): e0284380. https://doi.org/ 10.1371/journal.pone.0284380 Editor: Aamir Raina, Aligarh Muslim University, INDIA Received: September 21, 2022 Accepted: March 29, 2023 Published: April 12, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0284380 Copyright: © 2023 Thavarajah et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting information files. 1 Plant and Environmental Sciences, Pulse Quality and Nutritional Breeding, Biosystems Research Complex, Clemson University, Clemson, South Carolina, United States of America, 2 Food Legumes Research Program, International Centre for Agricultural Research in the Dry Areas (ICARDA), Amlaha, India dthavar@clemson.edu Abstract Dry pea (Pisum sativum L.) is a cool-season food legume rich in protein (20–25%). With increasing health and ecosystem awareness, organic plant-based protein demand has increased; however, the protein quality of organic dry pea has not been well studied. This study determined the genetic variation of individual amino acids (AAs), total AAs (liberated), total protein, and in vitro protein digestibility of commercial dry pea cultivars grown in organic on-farm fields to inform the development of protein-biofortified cultivars. Twenty-five dry pea cultivars were grown in two USDA-certified organic on-farm locations in South Carolina (SC), USA, for two years (two locations in 2019 and one in 2020). The concentrations of most individual AAs (15 of 17) and the total AA concentration significantly varied with dry pea cultivar. In vitro protein digestibility was not affected by the cultivar. Seed total AA and protein for dry pea ranged from 11.8 to 22.2 and 12.6 to 27.6 g/100 g, respectively, with heritability estimates of 0.19 to 0.25. In vitro protein digestibility and protein digestibility cor- rected AA score (PDCAAS) ranged from 83 to 95% and 0.18 to 0.64, respectively. Heritabil- ity estimates for individual AAs ranged from 0.08 to 0.42; principal component (PCA) analysis showed five significant AA clusters. Cultivar Fiddle had significantly higher total AA (19.6 g/100 g) and digestibility (88.5%) than all other cultivars. CDC Amarillo and Jetset were significantly higher in cystine (Cys), and CDC Inca and CDC Striker were significantly higher in methionine (Met) than other cultivars; CDC Spectrum was the best option in terms of high levels of both Cys and Met. Lysine (Lys) concentration did not vary with cultivar. A 100 g serving of organic dry pea provides a significant portion of the recommended daily allowance of six essential AAs (14–189%) and daily protein (22–48%) for an average adult weighing 72 kg. Overall, this study shows organic dry pea has excellent protein quality, sig- nificant amounts of sulfur-containing AAs and Lys, and good protein digestibility, and thus has good potential for future plant-based food production. Further genetic studies are war- ranted with genetically diverse panels to identify candidate genes and target parents to develop nutritionally superior cultivars for organic protein production. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284380 April 12, 2023 1 / 14 PLOS ONEFunding: Founding Agencies: 1. National Institute of Food and Agriculture, Organic Agriculture Research and Extension Initiative (OREI) (award no. 2018-51300-28431/proposal no. 2018-02799) and the United States Department of Agriculture, 2. National Institute of Food and Agriculture, Organic Agriculture Research and Extension Initiative (OREI) (award no. 2021-51300-34805/proposal no. 2021-02927) (DT, LB) 3. USDA National Institute of Food and Agriculture, [Hatch] project [1022664] (DT); 4. Good Food Institute (DT) 5. FoodShot Global (DT) The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Organic dry pea protein quality Introduction The plant-based protein market has been steadily growing globally. Globally, the plant-based protein market will continue to increase to a $9.5B industry by 2025. American retail sales increased by 6% in 2021, bringing the plant-based protein market’s total value to $7.4B in 2021 [1]. About 39% of Americans consume plant-based protein alternative foods due to various health and ecosystem concerns [2,3]. To meet the global demand for plant-based protein, ingredient suppliers have expanded the need for novel, pesticide-free, and gluten-free plant proteins from organic pulse crops, including dry pea (Pisum sativum L.). Dry peas are the most in-demand ingredient for this segment of the food industry due to their high protein (20–25%) and low fat (<1%) levels. Global dry pea production is approximately 14.6 million metric tons, and the United States of America (USA) accounted for 0.9 million metric tons in 2020 [4]. Dry pea is a vital pulse crop to provide global nutritional security, especially for pro- tein and low digestible carbohydrates at a low cost. Canada, Russian Federation, France, China, and India have become the world’s most extensive dry pea producers [4]. Certified organic dry pea production has increased globally and in US regions, but the exact statistics for the current production and harvested area for regional states are unavailable. However, US regions that have not been historically used to grow these pulse crops, including South Caro- lina (SC), will increase pulse production to meet the demand for plant-based protein [5,6]. Current food choices, including the “Beyond Burger,” use dry pea as their primary protein ingredient. However, concerns related to plant-based proteins focus on amino acid (AA) bal- ance, i.e., legume-based protein is low in sulfur-containing AAs (SAAs: cystine, Cys, and methionine, Met) and poor digestibility. The development and selection of nutritionally supe- rior organic dry pea cultivars will bring significant economic benefits to organic growers and nutritional value to consumers. Organic agriculture is the fastest-growing segment of US agriculture, with total sales of $9.9B in 2019 [6]. Organic grains, including corn, wheat, and soybean, accounted for $1.18B of this total, an increase of 55% from 2016 [6]. Pulse crops are an integral part of the global food system and can provide protein, low-digestible carbohydrates, and micronutrient-rich foods at a lower price than systems centered on animal proteins [5,7–9]. Certified organic dry pea pro- duction in the USA is small, with 16,666 ac in 2019 and a $5.9M value of sales from 104 USDA-certified organic farms [6]. Legume-inclusive cropping systems bring multiple benefits to organic agriculture: (1) at the food system level, pulses provide a significant amount of pro- tein, low-digestible carbohydrates, and a range of micronutrients with low phytate for both humans and animals; (2) at the production system level, legumes fix atmospheric N and improve soil phosphorus (P), which provides economic value to organic producers, making them more suitable for low-input cropping systems and mitigating greenhouse gas emissions as a result of reduced rates of N and P fertilizer inputs; and (3) at the cropping system level, legumes can be used as diversification crops in agroecosystems (e.g., in rotation with cereals), resulting in increased cereal crop yields due to disrupted pest (disease, insect, and weed) cycles, conserved soil moisture, and improved soil health via soil microbial activity [5,10–12]. Plant-based protein represents about 60% of the total global protein consumed, with the remainder from animal sources [13]. Developing countries mainly depend on plant-based proteins, and the global population is shifting toward alternative proteins. However, the quan- tity of protein in plant-based foods is not the only indicator of their ability to meet the nutri- tional demands of growing populations; protein quality, i.e., AA balance, must be considered [14]. Animal and plant-based proteins have different AA profiles and digestibility. Humans cannot synthesize essential AAs; therefore, these must be obtained from dietary sources [15]. Cereals have low lysine (Lys) concentrations and moderate-to-high concentrations of the PLOS ONE \| https://doi.org/10.1371/journal.pone.0284380 April 12, 2023 2 / 14 PLOS ONEOrganic dry pea protein quality SAAs, namely methionine (Met) and cysteine (Cys); however, pulses have high Lys and low- to-moderate levels of Met and Cys. Therefore, plant-based diets require cereals supplemented with Lys-rich ingredients such as dry pea [9,16]. With the increasing global population, depen- dence on animal sources for daily human protein requirements is not viable (e.g., higher energy and labor requirements, antibiotic resistance, and greenhouse gas emissions). There- fore, breeding traditional pulse crops for protein biofortification is essential to provide clean, allergen- and gluten-free, and highly nutritious plant-based protein to meet the world’s protein demands by 2050 [9,16,17]. Biofortification is a sustainable approach to increasing nutritional quality using conventional plant breeding and genomic tools to develop staple food crops with bioavailable micronutrients [18–20]. Dry pea micronutrient enrichment has been successful over the years, and breeding programs worldwide use available tools to develop mineral- and vitamin-rich cultivars with low phytic acid [21–23]. These biofortified cultivars perform well under non-organic growing sys- tems but have low yields and protein content when grown under an organic system without synthetic fertilizers and pesticides [5,24]. Therefore, organic nutritional breeding of pulse crops for increased protein quality is vital to overcome the issues related to growers, the food industry, and the nutrition community to meet growing consumer demand. This study aimed to evaluate if the current dry pea cultivars in production vary in protein quality (AA composition, total AA, total protein, and in vitro protein digestibility) in response to organic cropping systems. This study did not include the non-organic system for comparison. The objectives of this study were to assess 25 dry pea cultivars grown in two organic on-farm locations for two years to determine the genetic variation of AA profiles, total AA, protein, and in vitro digestibility to identify suitable cultivars for organic production with increased nutrition quality. Materials and methods Experimental details The experimental field design was a randomized complete block design (RCDB) with 25 culti- vars with two replications at two locations in 2019 and three replications at one location in 2020 (n = 175; [5]. The seeds were purchased from Pulse USA (Bismark, ND, USA), Meridian Seeds (Mapleton, ND, USA), and the Washington State Crop Improvement Association (Pullman, WA, USA). Material transfer agreements (MTAs) were signed with the seed companies before planting these cultivars in SC, USA. Detailed experimental design, agronomic details, and results (grain yield and nutritional quality) have already been published [5]. In brief, the experi- mental field design was a randomized complete block design (RCDB) with 25 cultivars with two replications at two locations in 2019 and three replications at one location in 2020 (n = 175). USDA-certified organic on-farm locations were WP Rawl and Sons (Pelion, SC, USA) and Cal- houn Fields Laboratory (Clemson University, SC, USA). Before planting, fields were tilled using a disc harrow and smoothly leveled. The plot size was 1.4×6 m (8.4 m2) with seven rows spaced 20 cm apart, a seeding depth of 5–7 cm, and a seeding rate of 90 seeds m-2. USDA-certified organic inoculant (Peaceful Valley Farm Supply, Inc., USA) was added at 3.1 g kg-1 seeds. At physiological maturity (110–115 days after planting), the plots were harvested, and 500 g of seeds were hand cleaned, finely ground using a UDY grinder, and stored at −10˚C until protein quality analysis. All protein quality data are reported on a dry mass (15% moisture). Protein analysis Total seed N concentration was measured using N combustion at the Soil Testing Laboratory, Clemson University, SC, and then values were converted to total protein content by multiply- ing by 6.25. Protein data are reported in our previous publication [5]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284380 April 12, 2023 3 / 14 PLOS ONEOrganic dry pea protein quality Amino acid (AA) analysis Reagents, solvents, and high-purity standards for AA analysis were purchased from Sigma Aldrich Co. (St. Louis, MO), Fisher Scientific (Waltham, MA), and VWR International (Rad- nor, PA). Ultrapure water and deionized water (ddH2O) to a resistance of �18.2 MO×cm (PURELAB flex 2 system, ELGA LabWater North America, Woodridge, IL) were used. The AA analysis method is reported elsewhere [25] with modifications from the literature [26,27]. Samples (40 mg) of dry pea powder (particle size � 0.5 mm) were weighed into glass culture tubes (16×125 mm, PTFE lined cap). Performic acid was synthesized from formic acid and hydrogen peroxide (9:1 ratio). Once chilled in an ice bath, 5 mL of performic acid was added to each tube, which was then gently swirled on a vortex mixer before being capped and refrig- erated for 16 h to convert Cys and Met to their derivatives, methionine sulfone and cysteic acid, which are more stable under acid hydrolysis. A 1/8 in. × tube length PTFE boiling rod was inserted into each tube before evaporating to dryness in a vacuum oil bath (3 gal. resin trap, BACOENG, Suzhou, China) at ~70–80˚C and ~610 mmHg. Once cooled, tubes were removed, and 4.9 mL of 6 M HCl and 0.1 mL of the standard internal mix (25 mM norvaline, 25 mM sarcosine) were added to each tube, which was then capped and gently swirled. Tubes were then placed in a gravity convection oven at 110˚C for 24 h to hydrolyze peptide bonds. Samples were cooled to room temperature, vortex mixed, and filtered through a 0.22 μm poly- propylene syringe filter. As before, one mL of the sample was added to a clean culture tube and evaporated to dryness. Samples were rehydrated with 1 mL of HPLC mobile phase A and pipetted into HPLC vials for analysis. AA analysis was performed via high-performance reverse phase chromatography on an 1100 series Agilent system (Agilent Technologies, Santa Clara, CA, USA) [28,29] with a diode array detector at two wavelengths (338 nm, 10 nm bandwidth, reference 390 nm, 20 nm band- width; and 262 nm, 10 nm bandwidth, reference 390 nm, 20 nm bandwidth). An aqueous and an organic solvent were used for mobile phases A and B, respectively. Mobile phase A con- tained 10 mM sodium phosphate, 10 mM sodium tetraborate decahydrate, and 5 mM sodium azide with a pH adjusted to 8.2 with 12 M HCl. The solution was then filtered through 0.2 μm regenerated cellulose. Mobile phase B consisted of 45% methanol, 45% acetonitrile, and 10% water (v/v/v). A lab reference dry pea sample was included in every digestion batch to monitor batch-to-batch variation; an AA standard mix was run on the high-performance liquid chro- matography (HPLC) before analyzing each batch of samples. Calibration standards (9–900 pmol/μL) with internal standards norvaline and sarcosine (500 pmol/μL) were run, and linear calibration models were generated based on peak areas for calculating sample AA concentra- tions, which were converted into percent of dry pea flour. The total AA concentration was cal- culated by summing all AA concentrations for each sample. In vitro protein digestibility analysis Protein digestibility was measured using the Megazyme Protein Digestibility Amino Acid Score assay kit with the modified protocol for a 100 mg sample size (Megazyme K-PDKAAS kit, Lancing, MI). Ground samples (100 mg) were weighed into 50 mL plastic falcon tubes, to which 3.8 mL of 0.06 N hydrochloric acid was added, and the mixture vortexed. The tubes were then placed into a tabletop heated air shaker at 37˚C for 30 min at 300 rpm. After shak- ing, 0.2 mL of pepsin solution was added to the tube, and the mixture vortexed. The tubes were then placed back into the shaker at 37˚C for 60 min at 300 rpm, and after the pepsin incu- bation, 0.4 mL of TRIS buffer was added. The tubes were then vortexed, and 40 μL of trypsin/ chymotrypsin solution was added to the tubes and then placed back in the air shaker for 4 h. After the trypsin/chymotrypsin incubation, the tubes were placed in a 100˚C water bath for 10 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284380 April 12, 2023 4 / 14 PLOS ONEOrganic dry pea protein quality min and then vortexed and brought to room temperature on the counter for a minimum of 20 min. After the overnight cold incubation, the tubes were centrifuged for 10 min. Ninety-six well plates were utilized for the colorimetric analysis. The Megazyme Excel calculator was modified to change the approximate sample mass from 0.5 to 0.1 g. In addition to the controls in the assay kit, a lab reference lentil sample was included in every batch to monitor batch-to- batch variation. The protein digestibility corrected amino acid score (PDCAAS) was calculated based on the Megazye Excel calculator, determined by comparing the AA profile of the dry pea against a standard AA profile, with one as the highest possible score. Statistical analysis Replicates, years, and cultivars were used as class variables. Data from both years were com- bined (after testing for heterogeneity) and analyzed using a general linear model procedure (PROC GLM) mixed model. Fisher’s least significant difference (LSD) at � 0.05 was per- formed for mean separation. Correlations (Pearson correlation coefficients) among traits were determined. A statistical model was developed to estimate broad-sense heritability (H2) with the class variables and genotype as random effects. The model was fit using restricted maxi- mum likelihood (REML). H2 was estimated as the proportion of variance due to cultivar, and analyses were performed using JMP 14.0.0 and SAS 9.4 [30]. Percent recommended dietary allowance estimates (%RDA) were calculated for the essential AAs [Cys, histidine (His), isoleu- cine (Iso), leucine (Leu), Lys, Met, phenylalanine (Phe), threonine (Thr), valine (Val)] and total AA concentration. Estimates were based on a 72 kg adult consuming 100 g of dry pea (15% moisture content) per day: 8–12 mg/kg His, 10 mg/kg Iso, 14 mg/kg Leu, 12 mg/kg Lys, 13 mg/kg Met + Cys, 14 mg/kg Phe + Tyr, 10 mg/kg Val, and 0.8 g/kg protein [31]. Results Analysis of variance Cultivars showed significant variation at P<0.05 and P<0.1 for most traits except for His, hydroxyproline (Hpr), Lys, and in vitro protein digestibility (Table 1). Location was significant for most cases except for serine (Ser) and total AAs. Similarly, the year effect was significant at P<0.05 and P<0.1 for 12 of 17 AAs, total AAs, total protein, and in vitro digestibility. Signifi- cant interactions of either cultivar × location or cultivar × year varied with the traits. The in vitro protein digestibility showed a significant effect only with the location and year; no effect was evident with cultivar × location or cultivar × year (Table 1). Broad-sense heritability esti- mates were very low to moderate (0.06–0.42), with the highest for arginine (Arg; 0.42) and total protein (0.25). Broad-sense heritability estimates were very low for SAAs (Met and Cys) and Lys (Table 1). Protein quality Organic dry pea cultivars had values of 11.8 to 22.2 g/100 g for total AAs (liberated), 12.6 to 27.6 g/100 g for total protein, 0.18 to 0.64 for PDCAAS value, and 83 to 95% for in vitro protein digestibility (Table 2). Dry pea contained a range of individual AAs, including nine essential AAs with a mean of 0.22 g/100 g for SAAs and 0.88 g/100 g for Lys (Table 2). These organic dry pea cultivars provide a significant amount of the recommended daily allowance (%RDA) of several AAs (14–66% His, 79–138% Iso, 76–169% Leu, 57–147% Lys, 15–85% Met + Cys, 76–189% Phe + Tyr, 94–169% Val) as well as protein (22–48%) (Table 2). Pearson’s correlation analysis revealed that most correlations were significantly positive except for Hpr vs. His and in vitro protein digestibility vs. Lys (Table 3). Total protein showed a significant positive PLOS ONE \| https://doi.org/10.1371/journal.pone.0284380 April 12, 2023 5 / 14 PLOS ONETable 1. Analysis of variance and broad-sense heritability estimates of protein quality traits evaluated for dry pea tested in SC, USA. Organic dry pea protein quality Component Alanine Arginine Asparagine Cystine Glutamine Glycine Histidine Hydroxyproline Isoleucine Leucine Lysine Methionine Phenylalanine Proline Serine Threonine Valine Total AA Total Protein In-vitro Digestibility Cultivar Location Year * Cultivar × Location Cultivar × Year * NS NS NS ** * NS * * * NS * NS NS NS NS NS NS NS NS NS NS NS * NS NS * NS NS NS NS * ** * NS H2 0.11 0.42 0.08 - 0.24 0.19 0.14 - 0.23 0.18 0.17 0.12 0.23 0.18 0.13 0.06 0.13 0.19 0.25 0.09 ** significant at P<0.05; * significant at P<0.1; Not significant (NS); H2 broad-sense heritability estimate. https://doi.org/10.1371/journal.pone.0284380.t001 correlation with all AAs except Hpr; Lys and Cys were also not correlated; and Hpr showed non-significant correlations in several cases (Table 3). The first two principal components (PCA) of the principal component analysis (PCA) accounted for 12.46, and 1.83 for the eigen- values. Cluster summary showed components of the total variance: (1) component 1 (62.3%): total AAs and 13 of 17 AAs; (2) component 2 (9.17%): Hpr and His; (3) component 3 (8.07%): in vitro protein digestibility; (4) component 4 (5.34%); protein and Arg; and (5) component 5 (2.93%): Cys (Fig 1). Most of the variation was captured by the first component (62.3%), which is highly correlated with the values of most AAs excluding Hpr and His. Cultivar responses Dry pea cultivars showed a normal distribution pattern for Cys, Met, total AAs, and in vitro protein digestibility (Fig 2). Out of 175 observations, 6.4% were high in Cys and Met, 8.8% were high in total AAs, and 5.6% were high for in vitro protein digestibility (Fig 2). Among the 25 cultivars tested, 10 cultivars showed more than 18 g/100 g of total AAs, with Fiddle being the highest and AAC Carver and AC Earlystar the lowest (Fig 3). For in vitro protein digestibil- ity, 17 of 25 cultivars showed a digestibility of 87% or better, with Fiddle having the highest value and AAC Carver the lowest (Fig 3). CDC Saffron, CDC Spectrum, and CDC Striker showed significantly higher concentrations of SAAs than AAC Carver and AC Earlystar (Fig 4). AAC Comfort showed higher Lys concentrations than other cultivars, but the effects were not significant (Fig 4). PLOS ONE \| https://doi.org/10.1371/journal.pone.0284380 April 12, 2023 6 / 14 PLOS ONEOrganic dry pea protein quality Table 2. Range and mean amino acid concentrations of organic dry pea grown in SC. Composition (g/100 g) Alanine Arginine Asparagine Cystine Glutamine Glycine Histidine Hydroxyproline Isoleucine Leucine Lysine Methionine Phenylalanine Proline Serine Threonine Valine Total AA (liberated) Total Protein± PDCAAS value In vitro digestibility (%) Range 0.61–1.01 0.95–2.22 1.59–3.07 0.02–0.10 1.82–3.56 0.60–1.08 0.08–0.38 0.48–2.00 0.57–0.99 0.77–1.70 0.49–1.27 0.12–0.26 0.38–1.16 0.42–1.32 0.58–1.09 0.39–0.74 0.68–1.22 11.8–22.2 12.6–27.6 18–64 83–95 Mean 0.86 1.5 2.36 0.05 2.86 0.88 0.26 1.16 0.8 1.33 0.88 0.17 0.89 1.04 0.89 0.59 0.97 17.5 20.9 54 87 Genotype Effect ** * NS NS NS ** * ND NS %RDA 15–85ǂ 14–66 79–138 76–169 57–147 76–189 22–48 ** significant at P<0.05; * significant at P<0.1; Not significant (NS); ND: Not detected; PDCAAS: Protein digestibility corrected amino acid score. ± Protein values are from [4]. Values are based on the combined statistical analysis of 175 data points for the current study (dry weight basis). Percent recommended dietary allowance estimates were calculated for the essential amino acids cystine (Cys), histidine (His), isoleucine (Iso), leucine (Leu), lysine (Lys), methionine (Met), phenylalanine (Phe), threonine (Tyr), and valine (Val), as well as for total AA concentration. Estimates were for a 72 kg adult consuming 100 g of dry pea (15% moisture content) per day given the following dietary requirements: 8–12 mg/kg His, 10 mg/kg Iso, 14 mg/kg Leu, 12 mg/kg Lys, 13 mg/kg Met + Cys, 14 mg/kg Phe + Tyr, 10 mg/kg Val, and 0.8 g/kg protein [31]. ǂ%RDA was calculated for both Cys+Met. https://doi.org/10.1371/journal.pone.0284380.t002 Discussion Our results demonstrate that current dry pea cultivars bred for conventional systems vary in terms of seed AA profile, total AAs, total protein, and in vitro protein digestibility when grown under organic cropping systems. Organic dry pea is a rich source of essential AAs, as a 100 g serving of organic dry pea provides 0.02–3.07 g/100 g of nine essential AAs (14–180% of RDA), 11.8–22.2 g of total AAs, and 22–48% of the daily protein requirement, with an in vitro protein digestibility of 83–95% (Table 2). In contrast to previous literature that states pulses are generally low in SAAs, our results demonstrate organic dry pea is a good source of SAAs (Met and Cys), with a 100 g serving providing 220 mg of total SAAs (Met+Cys) and 0.88 g of Lys (Table 2; Fig 4) [32,33]. According to our knowledge, this study is the first report on the detailed protein quality of commercial dry pea cultivars grown in an organic system to adapt these cultivars for organic produciton. The dry pea cultivars tested under organic field conditions in this study had mean protein and total AA (liberated) concentrations of 20.9 g/100 g and 17.5 g/100 g, respectively (Table 2). PLOS ONE \| https://doi.org/10.1371/journal.pone.0284380 April 12, 2023 7 / 14 PLOS ONETable 3. Pearson’s correlation analysis of nutritional traits among dry pea cultivars grown in the organic system. Cys Asp Glu Ser His Gly Thr Met Arg Ala Val Phe Iso Leu Lys Hpr Pro AA Pr Dig Organic dry pea protein quality Cys Asp Glu Ser His Gly Thr Met Arg Ala Val Phe Iso Leu Lys Hpr Pro Total AA Total Protein± Digestibility - NS NS - - ** * - - - * - significant at P<0.05; Not significant (NS); ± Protein values are from [4]. https://doi.org/10.1371/journal.pone.0284380.t003 - - - - NS NS - - - NS NS - - NS - - NS - - - - Several dry pea cultivars had high total AAs (>18 g/100 g) and >87% in vitro protein digest- ibility (Fig 3), demonstrating they are suitable for organic plant-based protein production. Among the cultivars tested, Fiddle had the highest total AA concentrations (19.6 g/100 g), and AAC Carver (15.5 g/100 g) and AC Earlystar (16.1 g/100 g) the lowest. Our previous study on the agronomic adaptability of dry pea [5] indicated AAC Carver, Jetset, and Mystique as the highest yielding cultivars (>2000 kg/ha) and most suitable for organic production without a yield penalty compared to conventional growing systems. However, the current study indicates Fig 1. Principal components of individual amino acids (g/100 g), total amino acids (g/100 g), protein (g/100 g), and in vitro digestibility of organic dry pea: (A) scatter plots and (B) biplots of components 1 and 2. https://doi.org/10.1371/journal.pone.0284380.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284380 April 12, 2023 8 / 14 PLOS ONEOrganic dry pea protein quality Fig 2. Dry pea cultivar distribution for cystine, methionine, and total amino acid (liberated) concentration as well as in vitro digestibility. https://doi.org/10.1371/journal.pone.0284380.g002 Fig 3. Total amino acids (liberated, g/100 g) and in vitro protein digestibility (%) of dry pea cultivars grown in the organic system. The bars indicate the total amino acids (liberated) g/100g and the line indicates in-vitro digesterbility (%). The bars/lines with different letters indicate significant difference at P<0.05. https://doi.org/10.1371/journal.pone.0284380.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284380 April 12, 2023 9 / 14 PLOS ONEOrganic dry pea protein quality Fig 4. Organic dry pea cultivar genetic variation for seed cystine, methionine, and lysine concentrations. The bars/lines with different letters indicate a significant difference at P<0.05. https://doi.org/10.1371/journal.pone.0284380.g004 these three cultivars have low total AAs and in vitro protein digestibility (Fig 3). A negative correlation between protein quality and crop adaptability suggests further testing is needed with diverse dry pea germplasm to develop biofortified organic cultivars with better grain yield, agronomic adaptability, and protein quality for organic systems [5,9,34]. Earlier litera- ture [32,35] indicates the AA composition of dry pea varies with cultivar and growing environ- ment, similar to the current study’s results. Further, one of these earlier studies shows dry pea has high concentrations of Arg, Leu, Lys, aspartic acid, and glutamic acid and low concentra- tions of His, Met, Thr, and Cys [35]. Another study compared several plant-based protein iso- lates for essential and non-essential AAs and found dry pea protein isolates contained only 5.9% Lys and low concentrations of Met [36]. In contrast, our study results show most modern cultivars have higher Cys, Met, and total AA concentrations and good in vitro protein digest- ibility (Fig 3). The best options to use for better protein quality are CDC Spectrum for Met and Cys, CDC Inca and CDC Striker for Met, and CDC Amarillo and Jetset for Cys (Fig 4). These cultivars have AA values within the range of the AAs reported in the literature for conventional cropping systems [33,36]. Incorporating these cultivars into dry pea breeding programs would benefit the development of better protein quality cultivars; however, more field testing is required to understand the genetic, environmental, and management interactions. Organic agriculture management varies with respect to on-farm practices for weeds, diseases, pests, and fertilizer; therefore, breeding dry pea cultivars best suited for organic management with increased nutritional quality is challenging [37]. AAs are critical for all forms of life. Humans cannot synthesize all 20 AAs needed for pro- tein synthesis for good health. Nine essential AAs must be obtained from the diet: Lys, Met, and Thr of the aspartate (Asp) family pathway; phenylalanine (Phe) and tryptophan (Trp) of the aromatic AAs; Val, Ile, and Leu of the branched-chain Aas (BCAAs); and His [38]. Lys, Met, Thr, and Trp levels limit the nutritional quality of plant-based foods because levels of these four AAs in plants are very low compared with those required for optimal human PLOS ONE \| https://doi.org/10.1371/journal.pone.0284380 April 12, 2023 10 / 14 PLOS ONEOrganic dry pea protein quality nutrition [9,38]. PCA analysis in the current study revealed seven essential AAs (Val, Iso, Thr, Leu, Met, Lys, and Phe) of organic dry pea in component 1, and one essential AA (His) in component 2 (Fig 1). These essential AAs are also positively correlated with total AA, protein, and in vitro digestibility (Table 3), indicating biosynthesis of these AAs could be upregulated using available genomic and biotechnology tools for early prediction of protein quality traits in breeding programs [9,38,39]. Plant-soluble Met and Lys levels might represent limiting fac- tors for synthesizing Met- or Lys-rich proteins [37]. Expressing genes that increase Lys and Met biosynthesis in combination with genes encoding proteins rich in Lys and Met codons appears to increase the levels of Lys in transgenic corn [39]. However, these transgenic approaches are not approved in USDA-certified organic agriculture systems. Conventional breeding approaches for selecting genetic material with higher levels of AAs and protein qual- ity using association mapping and genomic prediction tools are the only recommended meth- ods for organic pulse breeding. Dietary protein quality has two components: AA composition and availability. Availability is “the proportion of the dietary amino acids that are digested and absorbed in a form suitable for body protein synthesis” [40]. PDCAAS is the most common method used to determine protein availability [41]. We determined in vitro protein digestibility using an enzyme assay and then calculated PDCAAS based on the AA scores. This method is inexpensive and high- throughput and can be used to screen a larger number of seed samples for breeding programs than available in vivo methods [42]. The PDCAAS values for organic dry pea cultivars tested in this study ranged from 0.18 to 0.64 with 83–95% in vitro protein digestibility. Most organic dry pea cultivars have high protein digestibility (>87%), and these values are similar to those from the literature [43]. Plant-based proteins are an inexpensive, healthy choice for many peo- ple and a vital source of daily essential AAs. These proteins have several limitations in terms of human nutrition: they often lack one or more essential AAs, they are often not fully digestible, and toxins and pesticides are concentrated during protein extraction and drying procedures [44]. Therefore, pursuing nutritional breeding or biofortification of dry pea using an organic system approach is vital to overcome these nutritional and production issues for pulse growers and consumers. Organic nutritional breeding of pulses is challenging and demands better phe- notyping and genetic resources for cultivar development. With the increasing availability of genomic resources, expanding organic pulse breeding targets to produce better-quality pro- teins with higher digestibility will be possible. Supporting information S1 Data. (XLSX) Acknowledgments We thank the Crop Improvement Team (Clemson University), David Robb (Organic Farm, Clemson University), and Charles Wingard and Ashely Rawls from WP Rawl & Sons, Inc. for field operations; Martin Hochhalter (Meridian Seeds), Tyler Kres (Pulse USA), and the Wash- ington State Crop Improvement Association for providing dry pea seeds; and Elizabeth Bean, Jacob Johnson, Lindsey Moroney and Richard Baker for the technical assistant for the pulse breeding program. Author Contributions Conceptualization: Dil Thavarajah, Emerson Shipe, Shiv Kumar. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284380 April 12, 2023 11 / 14 PLOS ONEOrganic dry pea protein quality Data curation: Dil Thavarajah, Tristan Lawrence, Lucas Boatwright, Nathan Windsor, Nathan Johnson, Joshua Kay, Emerson Shipe, Shiv Kumar, Pushparajah Thavarajah. Formal analysis: Dil Thavarajah, Nathan Johnson, Shiv Kumar. Funding acquisition: Dil Thavarajah, Lucas Boatwright, Emerson Shipe, Shiv Kumar. Investigation: Dil Thavarajah, Shiv Kumar. Methodology: Dil Thavarajah, Tristan Lawrence, Nathan Windsor, Nathan Johnson, Joshua Kay, Pushparajah Thavarajah. Project administration: Dil Thavarajah. Resources: Dil Thavarajah. Software: Dil Thavarajah, Nathan Windsor, Nathan Johnson, Joshua Kay. Supervision: Dil Thavarajah. Validation: Dil Thavarajah, Nathan Johnson, Shiv Kumar, Pushparajah Thavarajah. Visualization: Dil Thavarajah. Writing – original draft: Dil Thavarajah. Writing – review & editing: Dil Thavarajah, Tristan Lawrence, Lucas Boatwright, Emerson Shipe, Shiv Kumar, Pushparajah Thavarajah. References 1. The Good Food Institute. 2021 State of the industry report. 2021. 2. Kerry. Winning with Plant-Based. Unlocking the Keys to Success for a Growing Market, Proprietary con- sumer research. 2020. https://www.fooddive.com/press-release/20200309-kerry-releases-report-on- plant-based-food-beverage-1/. 3. 4. 5. Linda Milo Ohr. Plant-Based Protein Market Grows Stronger. Food Technology Magazine. 2020: 74 (10). https://www.ift.org/news-and-publications/food-technology-magazine/issues/2020/october/ columns/nutraceuticals-plant-based-protein-market-grows-stronger. FAO. FAOSTAT. In: FAOSTATS- Crops [Internet]. 2020 [cited 9 Dec 2020]. http://www.fao.org/faostat/ en/#data/QC. Thavarajah D, Lawrence TJ, Powers SE, Kay J, Thavarajah P, Shipe E, et al. Organic dry pea (Pisum sativum L.) biofortification for better human health. PLoS One. 2022; 17: e0261109. Available: https:// doi.org/10.1371/journal.pone.0261109 PMID: 35025919 6. USDA -NASS. National Agriculture Statistical Services. In: 2020 [Internet]. 2017. https://quickstats. nass.usda.gov/results/FA8BCA73-7E79-3DA8-96A8-68D53707599D. 7. 8. Thavarajah D, Thavarajah P, Wejesuriya A, Rutzke M, Glahn RP, Combs GF Jr., et al. The potential of lentil (Lens culinaris L.) as a whole food for increased selenium, iron, and zinc intake: Preliminary results from a 3 year study. Euphytica. 2011; 180. https://doi.org/10.1007/s10681-011-0365-6 Johnson N, Johnson CR, Thavarajah P, Kumar S, Thavarajah D. The roles and potential of lentil prebi- otic carbohydrates in human and plant health. PLANTS, PEOPLE, PLANET. 2020;n/a. https://doi.org/ 10.1002/ppp3.10103 9. Salaria S, Boatwright JL, Thavarajah P, Kumar S, Thavarajah D. Protein Biofortification in Lentils (Lens culinaris Medik.) Toward Human Health. Front Plant Sci. 2022; 13. https://doi.org/10.3389/fpls.2022. 869713 PMID: 35449893 10. Stagnari F, Maggio A, Galieni A, Pisante M. Multiple benefits of legumes for agriculture sustainability: an overview. Chemical and Biological Technologies in Agriculture. 2017. https://doi.org/10.1186/ s40538-016-0085-1 11. Powers SE, Thavarajah D. Checking Agriculture’s Pulse: Field Pea (Pisum Sativum L.), Sustainability, and Phosphorus Use Efficiency. Front Plant Sci. 2019; 10. https://doi.org/10.3389/fpls.2019.01489 PMID: 31803218 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284380 April 12, 2023 12 / 14 PLOS ONEOrganic dry pea protein quality 12. Powers S, Mirsky E, Bandaranayake A, Thavarajah P, Shipe E, Bridges W, et al. Field pea (Pisum sativum L.) shows genetic variation in phosphorus use efficiency in different P environments. Sci Rep. 2020; 10. https://doi.org/10.1038/s41598-020-75804-0 PMID: 33144592 13. Nijdam D, Rood T, Westhoek H. The price of protein: Review of land use and carbon footprints from life cycle assessments of animal food products and their substitutes. Food Policy. 2012; 37: 760–770. https://doi.org/10.1016/j.foodpol.2012.08.002 14. Ritchie H, Reay DS, Higgins P. Beyond Calories: A Holistic Assessment of the Global Food System. Frontiers in Sustainable Food Systems. 2018. p. 57. Available: https://www.frontiersin.org/article/10. 3389/fsufs.2018.00057. 15. 16. Lizarazo CI, Lampi A-M, Liu J, Sontag-Strohm T, Piironen V, Stoddard FL. Nutritive quality and protein production from grain legumes in a boreal climate. J Sci Food Agric. 2015; 95: 2053–2064. https://doi. org/10.1002/jsfa.6920 PMID: 25242296 Leinonen I, Iannetta PPM, Rees RM, Russell W, Watson C, Barnes AP. Lysine Supply Is a Critical Fac- tor in Achieving Sustainable Global Protein Economy. Frontiers in Sustainable Food Systems. 2019. p. 27. Available: https://www.frontiersin.org/article/10.3389/fsufs.2019.00027. 17. Khazaei H, Subedi M, Nickerson M, Martı´nez-Villaluenga C, Frias J, Vandenberg A. Seed Protein of Lentils: Current Status, Progress, and Food Applications. Foods. 2019; 8. https://doi.org/10.3390/ foods8090391 PMID: 31487958 18. Bouis HE, Welch RM. Biofortification—A Sustainable Agricultural Strategy for Reducing Micronutrient Malnutrition in the Global South. Crop Sci. 2010; 50: S-20–S-32. https://doi.org/10.2135/cropsci2009. 09.0531 19. 20. 21. Thavarajah D, Johnson CR, McGee R, Thavarajah P. Phenotyping nutritional and antinutritional traits. Phenomics in Crop Plants: Trends, Options and Limitations. 2015. https://doi.org/10.1007/978-81-322- 2226-2_15 Jha AB, Warkentin TD. Biofortification of Pulse Crops: Status and Future Perspectives. Plants. 2020; 9. https://doi.org/10.3390/plants9010073 PMID: 31935879 Thavarajah D, Warkentin T, Vandenberg A. Natural enrichment of selenium in Saskatchewan field peas (Pisum sativum L.). Can J Plant Sci. 2010; 90. https://doi.org/10.4141/CJPS09154 22. Ray H, Bett K, Tar’an B, Vandenberg A, Thavarajah D, Warkentin T. Mineral micronutrient content of cultivars of field pea, chickpea, common bean, and lentil grown in Saskatchewan, Canada. Crop Sci. 2014; 54. https://doi.org/10.2135/cropsci2013.08.0568 23. Shunmugam AS, Liu X, Stonehouse R, Tar’An B, Bett KE, Sharpe AG, et al. Mapping seed phytic acid concentration and iron bioavailability in a pea recombinant inbred line population. Crop Sci. 2015; 55: 828–836. https://doi.org/10.2135/cropsci2014.08.0544 24. De Santis MA, Rinaldi M, Menga V, Codianni P, Giuzio L, Fares C, et al. Influence of Organic and Con- ventional Farming on Grain Yield and Protein Composition of Chickpea Genotypes. Agronomy. 2021. https://doi.org/10.3390/agronomy11020191 25. Madurapperumage A, Johnson N, Thavarajah P, Tang L, Thavarajah D. Fourier-transform infrared spectroscopy (FTIR) as a high-throughput phenotyping tool for quantifying protein quality in pulse crops. Plant Phenome J. 2022; 5: e20047. https://doi.org/10.1002/ppj2.20047 26. Gehrke CW, Wall LL Sr, Absheer JS, Kaiser FE, Zumwalt RW. Sample Preparation for Chromatography of Amino Acids: Acid Hydrolysis of Proteins. J Assoc Off Anal Chem. 1985; 68: 811–821. https://doi.org/ 10.1093/jaoac/68.5.811 27. Manneberg M, Lahm HW, Fountoulakis M. Quantification of cysteine residues following oxidation to cysteic acid in the presence of sodium azide. Anal Biochem. 1995; 231: 349–353. https://doi.org/10. 1006/abio.1995.9988 PMID: 8594984 28. Agilent Application Note. Separation of two sulfurated amino acids with other seventeen amino acids by HPLC with pre-column derivatization. Agil Technol. 2010. 29. Long W. Automated Amino Acid Analysis Using an Agilent Poroshell HPH-C18 Column. Agil Technol. 2015. 30. SAS Institute Inc. SAS Institute Inc. Cary; NC; 2013: User’s guide: Statistics SAS Institute; 2013. 31. National Research Council (US) Subcommittee on the Tenth Edition of the Recommended Dietary Allowances. In: Recommended Dietary Allowances: 10th Edition. Washington (DC): National Acade- mies Press (US) [Internet]. 1989 [cited 26 Aug 2022] p. 6. https://www.ncbi.nlm.nih.gov/books/ NBK234922/. 32. Khan MA, Jacobsen I, Eggum BO. Nutritive value of some improved varieties of legumes. J Sci Food Agric. 1979; 30: 395–400. https://doi.org/10.1002/jsfa.2740300409 PMID: 459444 33. Bonke A, Sieuwerts S, Petersen IL. Amino Acid Composition of Novel Plant Drinks from Oat, Lentil and Pea. Foods. 2020; 9: 429. https://doi.org/10.3390/foods9040429 PMID: 32260282 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284380 April 12, 2023 13 / 14 PLOS ONEOrganic dry pea protein quality 34. Daba SD, Morris CF. Pea proteins: Variation, composition, genetics, and functional properties. Cereal Chem. 2022; 99: 8–20. https://doi.org/10.1002/cche.10439 35. Holt NW, Sosulski FW. Amino Acid composiiton and Protein quality of field peas. Can J Plant Sci. 1979; 59: 653–660. https://doi.org/10.4141/cjps79-103 36. Gorissen SHM, Crombag JJR, Senden JMG, Waterval WAH, Bierau J, Verdijk LB, et al. Protein content and amino acid composition of commercially available plant-based protein isolates. Amino Acids. 2018; 50: 1685–1695. https://doi.org/10.1007/s00726-018-2640-5 PMID: 30167963 37. Boyhan GE, Stone SP. Breeding for Organic and Sustainable Production BT—Organic Farming for Sustainable Agriculture. In: Nandwani D, editor. Cham: Springer International Publishing; 2016. pp. 123–136. 38. Galili G, Amir R, Fernie AR. The Regulation of Essential Amino Acid Synthesis and Accumulation in Plants. Annu Rev Plant Biol. 2016; 67: 153–178. https://doi.org/10.1146/annurev-arplant-043015- 112213 PMID: 26735064 39. Galili G, Amir R. Fortifying plants with the essential amino acids lysine and methionine to improve nutri- tional quality. Plant Biotechnol J. 2013; 11: 211–222. https://doi.org/10.1111/pbi.12025 PMID: 23279001 40. Rutherfurd SM, Moughan PJ. Available versus digestible dietary amino acids. Br J Nutr. 2012; 108: S298–S305. https://doi.org/10.1017/S0007114512002528 PMID: 23107541 41. FAO/WHO. Dietary protein quality evaluation in human nutrition. Report of an FAO expert consultation. Food and Nutrition Paper No 9. 2013. 42. Cordero-Clavijo LM, Serna-Saldı´var SO, Lazo-Ve´lez MA, Gonza´ lez JFA, Panata-Saquicilı´ D, Briones- Garcia M. Characterization, functional and biological value of protein-enriched defatted meals from sacha inchi (Plukenetia volubilis) and chocho (Lupinus mutabilis). J Food Meas Charact. 2021; 15: 5071–5077. https://doi.org/10.1007/s11694-021-01084-5 43. Nosworthy MG, Franczyk AJ, Medina G, Neufeld J, Appah P, Utioh A, et al. Effect of Processing on the in Vitro and in Vivo Protein Quality of Yellow and Green Split Peas (Pisum sativum). J Agric Food Chem. 2017; 65: 7790–7796. https://doi.org/10.1021/acs.jafc.7b03597 PMID: 28796503 44. Gavahian M, Khaneghah AM. Cold plasma as a tool for the elimination of food contaminants: Recent advances and future trends. Crit Rev Food Sci Nutr. 2020; 60: 1581–1592. https://doi.org/10.1080/ 10408398.2019.1584600 PMID: 30924347 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284380 April 12, 2023 14 / 14 PLOS ONE
10.1371_journal.pone.0284587	RESEARCH ARTICLE Population genomics of fall armyworm by genotyping-by-sequencing: Implications for pest management Tamylin Kaori IshizukaID Carlos Eduardo de Arau´ jo Batista4☯, Marı´a Gabriela Muru´ a5‡, Jose´ Baldin Pinheiro4, Amit Sethi6, Rodney N. Nagoshi7☯, Josemar Foresti2‡, Maria Imaculada Zucchi1,8* 1,2, Erick Mauricio Goes Cordeiro3☯, Alessandro Alves-Pereira1☯, 1 Institute of Biology, State University of Campinas, Campinas, Brazil, 2 Corteva Agriscience, Mogi Mirim, São Paulo, Brazil, 3 Department of Entomology, University of São Paulo–ESALQ/USP, Piracicaba, Brazil, 4 Department of Genetics, University of São Paulo–ESALQ/USP, Piracicaba, Brazil, 5 Instituto de Bioprospeccio´ n y Fisiologı´a Vegetal (INBIOFIV), CONICET-UNT, Tucuma´n, Argentina, 6 Corteva Agriscience, Johnston, Iowa, United States of America, 7 United States Department of Agriculture- Agricultural Research Service, Gainesville, Florida, United States of America, 8 Agência Paulista de Tecnologia dos Agronego´ cios, Po´ lo Regional Centro-Sul, Piracicaba, Brazil ☯ These authors contributed equally to this work. ‡ MGM and JF also contributed equally to this work. * tamylin@gmail.com a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Abstract Citation: Ishizuka TK, Cordeiro EMG, Alves-Pereira A, de Arau´jo Batista CE, Muru´a MG, Pinheiro JB, et al. (2023) Population genomics of fall armyworm by genotyping-by-sequencing: Implications for pest management. PLoS ONE 18(4): e0284587. https://doi.org/10.1371/journal.pone.0284587 Editor: Umakanta Ngangkham, ICAR Research Complex for North Eastern Hill Region Manipur Centre, INDIA Received: December 11, 2022 Accepted: April 3, 2023 Published: April 18, 2023 Copyright: This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication. Data Availability Statement: Illumina reads are available as FASTQ files in the NCBI Sequence Read Archive (SRA) repository, accession PRJNA847933 (https://dataview.ncbi.nlm.nih.gov/ object/PRJNA847933?reviewer= nac1b0jr7e3tb4r1hgtsbcej7s). The DNA sequences used for host strain identification are deposited into GenBank (accession numbers ON704174 - ON704629). The fall armyworm (FAW), Spodoptera frugiperda, is a significant pest of many crops in the world and it is native to the Americas, where the species has shown the ability to rapidly evolve resistance to insecticides and transgenic plants. Despite the importance of this species, there is a gap in the knowledge regarding the genetic structure of FAW in South America. Here, we examined the genetic diversity of FAW populations across a wide agri- cultural area of Brazil and Argentina using a Genotyping-by-Sequencing (GBS) approach. We also characterized samples by their host strain based on mitochondrial and Z-linked genetic markers. The GBS methodology enabled us to discover 3309 SNPs, including neu- tral and outlier markers. Data showed significant genetic structure between Brazil and Argentina populations, and also among the Argentinian ecoregions. Populations inside Bra- zil showed little genetic differentiation indicating high gene flow among locations and con- firming that structure is related to the presence of corn and rice strains. Outlier analysis indicated 456 loci putatively under selection, including genes possibly related to resistance evolution. This study provides clarification of the population genetic structure of FAW in South America and highlights the importance of genomic research to understand the risks of spread of resistance genes. Introduction The fall armyworm (FAW), Spodoptera frugiperda (J. E. Smith) (Lepidoptera: Noctuidae), is a major agricultural pest that can feed on several different hosts [1]. FAW have the ability to PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 1 / 16 PLOS ONEFunding: The authors T.K.I.; A.S.; and J.F. received support came from Corteva Agriscience in the form of salaries and for collections of field samples in Brazil. The author R.N.N. received support came from the Agricultural Research Service of the United States Department of Agriculture. The author M.G.M. received support came from Consejo Nacional de Investigaciones Cientı´ficas y Te´cnicas (CONICET) and the grant PIP N˚ 206 (2021-2023 GI) (CONICET). The author A.A.-P. received support from São Paulo Research Foundation (grant FAPESP 2018/00036-9) in the form of a post-doctoral scholarship. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Population genomics of fall armyworm in Brazil and Argentina evolve rapid resistance to insecticides and transgenic crops, which can impact the effectiveness of control strategies [2]. The spread of such resistance traits is dependent on the migratory behavior of FAW and it is therefore important for effective pest management to delineate the pattern and magnitude of population movement across national boundaries. In North America, FAW populations from Texas and Florida make annual migrations northwards to recolonize areas in the north of USA and Canada [3]. This behavior reflects the inability of FAW to survive freezing temperatures, which limits winter populations to the most southern locations in the states of Texas and Florida [4]. Less is known about migration pat- terns in South America, which experiences a significantly different climate than North Amer- ica. Climate suitability modeling shows that South America features suitable conditions for persistent FAW populations in this continent, except for most of Argentina [5]. This suggests an annual migration towards Argentina from more suitable locations, with the neighboring countries being the prime candidates for the migratory sources [6]. A complicating factor for FAW is that the species can be subdivided into two groups called host strains that differ in their distribution on plant hosts in the field, with the C-strain prefer- entially found in corn and the R-strain in pasture grasses and, to a lesser extent, rice [7, 8]. The host strains are morphologically indistinguishable and so can only be identified through the use of molecular markers, with the most commonly used being polymorphisms in the mito- chondrial cytochrome oxidase subunit I (COI) and the Z-chromosome linked triosephosphate isomerase (Tpi) genes [9, 10]. At this time, the most consistent evidence of population structure in South America is that indicative of the two host strains, although some level of genetic structure has been observed for populations within Paraguay and within Brazil [11–13]. Overall, it remains unclear whether and to what extent geographically separated FAW in South America exhibit genetic differentiation. To better address this issue, we used Genotyping-by-Sequencing (GBS), a vari- ation of the commonly used restriction-site-associated DNA sequencing methodology to iden- tify single nucleotide polymorphisms (SNPs). GBS data allows us to resolve patterns of genetic diversity and spatial structure at very fine scale [14], and it was utilized in this study to evaluate patterns of gene flow and genetic structure of FAW populations across Brazil and Argentina. We additionally examined the information of host strains to discover informative loci puta- tively under selection. According to the results, we discuss some practical implications of our findings to the FAW management in South America. Material and methods Sampling and DNA extraction Fall armyworm caterpillars were manually collected in fields from 12 locations in Brazil under the SISBIO license number 58435, and from 3 locations of Argentina, between June of 2018 and January of 2021 (Fig 1, Table 1). Larvae were transferred to trays containing artificial diet and reared in laboratory conditions until become moths. Moths were placed in 1.5 mL poly- propylene microcentrifuge tubes with 98% ethanol and stored at -20ºC. Species identification was confirmed by morphology and sequencing of the COI gene (see below). We extracted DNA from the legs of random adults using CTAB-based method [15]. DNA quality and quan- tification were assayed by agarose electrophoresis gel (1% w/v) stained with SYBR Safe DNA Gel Stain (Invitrogen, Carlsbad, CA, USA). DNA concentrations were adjusted to approxi- mately 30 ng/μL. Distribution map of the populations was drawn using the software QGIS v3.28.3-Firenze. (Open Access Geographic Information System, https://qgis.org/en/site/. Accessed on February 26th, 2023). Publicly available shapefile of South American country boundaries was downloaded PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 2 / 16 PLOS ONEPopulation genomics of fall armyworm in Brazil and Argentina Fig 1. Fall armyworm sampling locations in Brazil and Argentina according to ecoregions. Shapefile of limits of countries in South America: IBGE-Mapas (IBGE—Brazilian Institute of Geography and Statistics; Available at: <https://geoftp.ibge.gov.br/ cartas_e_mapas/bases_cartograficas_continuas/bc250/versao2021/shapefile/bc250_shapefile_2021_11_18.zip>; Acessed on February 26, 2023) QGIS v3.28.3 (Available at: <https://qgis.org/en/site/>; Acessed on: February 26, 2023). DATUM: SIRGAS 2000. https://doi.org/10.1371/journal.pone.0284587.g001 from IBGE-Mapas (IBGE–Brazilian Institute of Geography and Statistics, https://geoftp.ibge. gov.br/cartas_e_mapas/bases_cartograficas_continuas/bc250/versao2021/shapefile/bc250_ shapefile_2021_11_18.zip. Accessed on February 26th, 2023). Host strains identification For the 228 specimens whose reads were successfully retained after GBS processing, we did polymerase chain reaction (PCR) amplifications of the COI and TpiEI4 regions to characterize host strains based on diagnostic polymorphisms at mCOI1164, mCOI1287 and gTpi183 posi- tions using the same primer sequences and procedures as described elsewhere [4]. Samples featuring both C and T nucleotides at the gTpi183 position were identified as hybrids (TpiH). The COI sequences were used to confirm species identification as well. Sequences were aligned PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 3 / 16 PLOS ONEPopulation genomics of fall armyworm in Brazil and Argentina Table 1. Fall armyworm (Spodoptera frugiperda) sampling information. N refers to the number of samples successfully sequenced after GBS processing, totaling 228 individuals. Location Code Country AR01 AR02 AR03 BA02 BA03 DF GO MA01 MA02 MT01 MT02 PR RS SC SP Argentina Argentina Argentina Brazil Brazil Brazil Brazil Brazil Brazil Brazil Brazil Brazil Brazil Brazil Brazil Locality (City, State) Meta´n—Salta America—Buenos Aires San Justo—Santa Fe São Deside´rio—Bahia Barreiras—Bahia Planaltina—Distrito Federal Rio Verde—Goia´s São Luı´s—Maranhão São Luı´s—Maranhão Campo Novo do Parecis—Mato Grosso Campo Novo do Parecis—Mato Grosso Toledo—Parana´ Alegrete—Rio Grande do Sul Chapeco´—Santa Catarina Taquarituba—São Paulo https://doi.org/10.1371/journal.pone.0284587.t001 Collection Date Latitude Longitude Host N 22-Feb-18 24-Jan-18 27-Feb-18 26-Jun-18 19-Jul-18 8-Jun-18 22-Jun-18 12-Dec-18 19-Dec-18 25-Jun-18 1-Jul-18 20-Jun-18 13-Jan-21 14-Dec-20 14-Jul-18 -25.5 -35.48 -30.78 -12.44 -11.78 -15.87 -17.75 -2.59 -2.58 -13.97 -14.27 -24.67 -30.02 -27.09 -23.54 -64.97 -62.97 -60.58 -45.64 -45.78 -47.4 -51.04 -44.21 -44.21 -57.98 -57.76 -53.76 -55.71 -52.64 -49.22 Corn Corn Corn Corn Corn Corn Corn Corn Corn Corn Corn Corn Rice Corn Corn 11 3 15 19 18 20 24 10 13 24 21 9 3 15 23 with Clustal W [16] using MEGA 7 [17]. The genetic data presented in this study are publicly available on GenBank (BioProject PRJNA847933, and accession numbers ON704174— ON704629). GBS library sequencing and data processing The steps of the GBS library preparation were done according to methodology described else- where [18] with the following modifications. Genomic DNA was digested with restriction enzymes NsiI-MseI (New England Biolabs—NEB, Ipswich, MA, USA). The barcoded NsiI adapters and a common MseI adapter were ligated to the digested DNA of each sample. Bar- coded DNA fragments from all samples were pooled in a single tube and amplified by PCR. The libraries were single-end sequenced to 150 nucleotides on a single lane using the Illumina NextSeq500/550 sequencing kit v2 (Illumina, Inc. San Diego, CA, USA) at the Genome Inves- tigation and Analysis Laboratory of University of São Paulo. Sequencing quality of GBS libraries was evaluated using FastQC [19]. The 3’end of raw reads were trimmed to 90 bp and were inspected for adaptor sequences removal. We per- formed demultiplex using process-radtags in STACKS v.1.42 [20]. Reads could be assigned to each individual based on the sequence of the barcodes. Samples with less than 200,000 sequences and/or unexpected GC contents were removed for further analysis. Reads were aligned to the Spodoptera frugiperda genome ZJU_Sfru_1.0, under Bioproject PRJNA590312 [21], using the algorithm BWA-MEM of the software BWA 0.7.17 [22]. Alignment files were processed with SAMtools [23] and Picard (http://broadinstitute.github.io/picard). We identi- fied SNPs using freebayes 1.3.4 [24] with - -standard_filters option. Filtering was performed using VCFtools v0.1.12a [25] and BCFtools 0.1.12 [22]. We retained bi-allelic SNPs that passed the following criteria: minor allele frequency � 0.01, read depth � 5X, mapping quality � 20, maximal amount of missing data per locus = 10%. Variants were separated by a minimum dis- tance of 150 bp and r2 threshold of 0.6. Results were stored in variant call format (VCF) after an additional filter to remove six samples with more than 25% of missing data, and SNPs iden- tified at sexual chromosome W present only in females, mitochondrial genome or in unan- chored contigs of the reference genome. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 4 / 16 PLOS ONEPopulation genomics of fall armyworm in Brazil and Argentina Genetic differentiation analysis Genetic diversity was estimated by calculating the observed heterozygosity HO, expected heterozygosity HE, nucleotide diversity π, the inbreeding coefficients (FIS), and the pairwise Fixation Index (FST) using hierfstat package [26, 27]. We tested for significant differences in heterozygosities and FIS using confidence intervals calculated based on 1000 bootstrap resam- ples. FST relations were illustrated by heatmaps generated by the RColorBrewer R package. Genetic structure was additionally explored through the principal component analysis (PCA) and the discriminant analysis of principal components (DAPC) using ade4 [28] and adegenet [29, 30] for R software [31]. Specimens were grouped by the ADMIXTURE v1.3.0 software, and the best value of inferred genetic groups (K) was implemented by the cross-validation method [32]. Outlier SNPs detection and annotation We searched for loci putatively under selection in the set of 15 populations and in the set of strains, where samples were identified as R- or C- strains using the Tpi marker as previously described [4]. TpiH individuals were not included in this analysis. Outlier SNPs were identified using three methods. The fsthet package [33] generates smoothed quantiles for the FST–hetero- zygosity relationship from the empirical distribution, and outliers were selected with a 95% confidence level threshold. The pcadapt package [34] considers population structure deter- mined by PCA, and outliers were identified using a false discovery rate (FDR) of 0.05. The pro- gram FLK [35] assumes that SNPs were polymorphic in an ancestral population, and uses a population tree to build a null model of the behavior of FST. We retained candidate markers identified by at least two methods and the remaining loci were considered as neutral. We compared the sequences containing outliers with the annotation files of the reference genome to identify loci present in encoding genes. We generated a fasta file containing the protein sequences to run Blast2GO [36] software using blastp with Insecta Taxa, InterPro Scan, GO mapping and functional annotation. The GO terms for each gene were submitted to Web Gene Ontology Annotation Plot (WEGO) for summarization [37]. Results Fall armyworm host strains Utilizing diagnostic polymorphisms in the COI and Tpi regions, we were able to determine the strain composition of the fall armyworm populations. A higher percentage of the samples collected from corn fields were identified as C-strain using the Tpi marker (95%) than COI (86%), consistent with other studies suggesting that Tpi is a better strain marker than COI [10]. The AR03 population in Argentina (Santa Fe) had the lowest agreement between markers (7%). Tpi-hybrids (TpiH) were found only in Brazilian locations, mainly in MA02 population, which also had many R-strain samples (Fig 2). SNPs discovery The GBS library generated a total of 187,842,557 reads that were retained after demultiplexing and quality checking. A total of 3309 SNP loci were retained in 228 individuals from 15 loca- tions across Brazil and Argentina. Mean sequencing depth per SNP was 23.3 x (min = 8.7x, max = 45.1x). Estimates of genetic diversity were similar across locations and are summarized in S1 Table. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 5 / 16 PLOS ONEPopulation genomics of fall armyworm in Brazil and Argentina Fig 2. Haplotype frequencies in fall armyworm populations of Brazil and Argentina using COI and Tpi markers. A) COI strains COI-RS (R-strain) and COI-CS (C-strain) B) Tpi marker identifies R-strain (TpiR), C-strain (TpiC), and hybrids (TpiH) that feature both diagnostic polymorphisms at site gTpi183. https://doi.org/10.1371/journal.pone.0284587.g002 Gene flow and population genetic structure Pairwise FST calculations (S1 Fig, Fig 3A) and PCA analysis (Fig 3B) revealed high levels of genetic differentiation among the three populations from Argentina, ranging from 0.3742 to 0.4305, which is a strong indicative of reduced gene flow among these locations. On the other hand, Brazilian populations belonging to C-strain had very low FST values (ranging from 0.000 to 0.006). Likewise, Brazilian R-strain populations MA02 and RS had pairwise FST = 0. Lower pairwise FST distances (S3 Fig) can be a result of lower divergence time in addition to the pres- ence of gene flow across Brazil. Aditionally, the two first components of PCA grouped Brazil- ian populations by their host strains, and showed that Argentinean population AR02 was more related to the Brazilian C-strain populations than to the Argentinean AR01 and AR03. Outlier detection When considering the 15 FAW populations, we identified 209 outlier SNPs in common by at least two of the three tests performed, whereas by comparing R- and C-strains we found 293 outliers in common by at least two of the three tests performed (S4 Fig). Altogether, we identified a total of 456 outlier loci putatively under selection (13.8%), and the remaining 2853 SNPs were considered to be neutral. By analyzing only neutral markers, the PCA plot indicated PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 6 / 16 PLOS ONEPopulation genomics of fall armyworm in Brazil and Argentina Fig 3. Genetic structure of FAW populations from Brazil and Argentina based on 3309 SNP loci. (A) Heatmap and dendrogram based on FST pairwise distances among 15 locations. Red color represents a greater degree of differentiation. (B) Principal Component Analysis (PCA) showing the first two components. Geographic locations are represented by different colors, and dots represent different individuals. https://doi.org/10.1371/journal.pone.0284587.g003 three clusters, grouping the AR02 (Buenos Aires) closest to the Brazilian populations (Fig 4A). Analysis of populations using Admixture showed that four individuals from MA02 featured similar admixture patterns as FAW individuals from AR02 and AR03 (Fig 5A). The outlier loci in turn resulted in four clusters: C-strain-Argentina, AR03, C-strain-Brazil, and R-strain- Brazil (Fig 4B). When considering loci under positive selection, an increased number of Fig 4. PCA analysis generated with (A) 2,853 neutral SNPs, (B) 456 outliers. Dots represent different individuals of each location. https://doi.org/10.1371/journal.pone.0284587.g004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 7 / 16 PLOS ONEPopulation genomics of fall armyworm in Brazil and Argentina Fig 5. Admixture analysis of fall armyworm populations based on (A) 2,853 neutral SNP loci and (B) 456 outlier SNP loci. Bar plot colors indicate estimated proportions of ancestry for each individual shown as a vertical line. https://doi.org/10.1371/journal.pone.0284587.g005 distinct genetic pools was obtained in Admixture analysis (K = 6), and more individuals appeared to feature non-admixed pattern (Fig 5B). Annotation of outlier loci For outlier analyses, 456 SNP loci putatively under selection were compared to the annotations of the genome of a specimen collected in China (ZJU_Sfru_1.0) and 306 of these were within predicted protein coding genes of FAW. After Blast2GO analysis, 220 proteins were success- fully mapped, and 94 were annotated (S2 Table). We found many outlier loci within genes possibly involved in binding functions and associated with the cell membrane (Fig 6A). Among the 94 loci with GO IDs, SNP_1055 was annotated as an ABC transporter C subfamily member 13, and the mutation was putatively under selection when comparing different loca- tions, rather than related to the presence of two host strains. Regarding detoxifying activity, the locus SNP_3077 was in a gene similar to cytochrome P450 CYP314A1 (this locus was detected as outlier when comparing host strains), and the locus SNP_1559 was in a gene simi- lar to an annotated esterase FE4-like. Another mechanism of insect defense involves chitin processing, and here we had five loci related to cuticle proteins or chitin binding (SNP_574, SNP_1209, SNP_1853, SNP_2799, SNP_3139). Analysis of outliers comparing host strains resulted in 68 outlier SNPs (23% of 293) in the sex chromosome Z, which concentrated over twice the number of outliers compared to the autosomal chromosomes (Fig 6B). The locus SNP_421 located at chromosome 3 was within a gene similar to an odorant receptor and could be related to sensorial stimulus to either host perception or male mating preferences. The locus SNP_3298 was in a gene that had 100% PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 8 / 16 PLOS ONEPopulation genomics of fall armyworm in Brazil and Argentina Fig 6. Outlier SNPs under positive selection using Pcadapt, FstHet, and FLK. (A) GO categories of putative loci under selection generated from WEGO. The results are presented in three main categories: biological process, cellular component, and molecular function. The left y-axis indicates the percentage of a specific category of genes in the main category. The right y-axis indicates the number of genes in each category. (B) Manhattan plot of FST analysis comparing C-strain and R-strain against the position of SNPs on each of the 31 chromosomes. Chromosome 1 corresponds to the sex chromosome Z. Each SNP is represented by a dot. Chromosomes are represented by blue (odd) and orange (even). https://doi.org/10.1371/journal.pone.0284587.g006 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 9 / 16 PLOS ONEPopulation genomics of fall armyworm in Brazil and Argentina identity to a GPI-anchored glycoprotein. Besides this locus, another 188 loci were successfully blasted with more than 90% similarity, but had no GO ID associated with them (S3 Table), including one locus identified as a cytochrome P450 (SNP_423), one cadherin (SNP_1010), one zinc carboxypeptidase (SNP_1352), one GABA receptor (SNP_719), and one UDP-gluco- syltransferase (SNP_2719). The frequencies of each polymorphism for some candidate genes under selection were also calculated for the Brazil locations (S5 Fig). Discussion GBS has been used successfully in insect pests to reveal insights about gene flow and coancestry [38], spatial and temporal genetic structure [39, 40], and incursions of invasive pests [41, 42]. Here we associated the informative data set provided by GBS with molecular markers used for host strain identification to better explain the patterns of FAW population structure in Brazil and Argentina and to identify candidate genes putatively under selection. By assessing a large number of samples in Brazil, we confirmed that FAW collected in corn fields were predominantly C-strain, with less than 4% of samples featuring both the COI-RS and TpiR diagnostic markers, and the few specimens featuring discordant genotypes likely represent vestiges of interbreeding events that occurred in the past. Based solely on diagnostic polymorphisms in COI and Tpi regions, MA02 and AR03 populations showed increased levels of strain hybridization, and we were able to describe the level of gene flow of these locations based on GBS data set. GBS data revealed high levels of gene flow and low genetic differentiation between MA02 and RS population, which was composed by pure R-strain samples. Since these two popula- tions were the most geographically distant locations sampled in Brazil, apart over 3200 km, we presumed that MA02 was composed by R-strain specimens with recent events of hybridiza- tion. Altogether, the genetic analysis based on pairwise FST distances and PCA plots confirmed that the Brazilian populations are structured by host strains, rather than by geographical ecoregions. Long distance migration enables FAW populations to travel from Southern Texas and Flor- ida up to Canada, a distance of nearly 2500 km, in less than three months [3, 43]. Therefore given its strong flight performance, we can hypothesize that FAW is also performing long dis- tance migration within Brazil sufficient to keep populations homogeneous within each host strain. Several studies comparing populations from Brazil and Argentina in the past showed strong similarities between the two countries [4, 44–46], which would be expected if the great major- ity of Argentina FAW are derived from seasonal migrations from Brazil. To some extent, Bra- zilian and Argentinian populations featured common ancestry, and the population from Buenos Aires (AR02) had the lowest genetic differentiation with Brazilian populations. Nevertheless, FST analysis indicated significant genetic structure between countries and among provinces inside Argentina. This result is consistent with observations of mating incompatibility between populations collected in northern Argentina compared to those from the Pampas region, which indicated pre-reproductive isolating barriers between geographically separated populations [6]. GBS data for another important Noctuidae pest in the Americas, the sugarcane borer (SCB), showed a similar pattern of genetic structure between Argentina and Brazil populations, and among populations within Argentina [47, 48]. We hypothesize that Argentina likely contains one or more endemic FAW populations that exhibit significant geographical isolation and that additional studies are required to better investigate the fine scale genetic structure of FAW as well as identify locations capable of supporting permanent FAW populations in this country. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 10 / 16 PLOS ONEPopulation genomics of fall armyworm in Brazil and Argentina In order to explain how selection pressures might be affecting FAW populations in South America, we examined the putative annotations of genes containing outlier SNPs. Host strain outliers were mostly concentrated on the sex chromosome Z, suggesting that the selection pressures are acting upon specific regions of the FAW genome. This result corroborates with previous study where the preponderance of strain specific SNPs were Z-linked [49], and is consistent with the proposal that strain divergence is being driven primarily by Z-chromosome functions [50]. We believe that by reducing complexity of the genome, the GBS method was able to capture a fairly large number of polymorphisms in the Z-chromosome, and thereby dis- criminate between the R- and C-strains. It is possible that previous research based on nuclear SSR markers [12] that did not differentiate the host strains lacked sufficient coverage of the sex chromosome. Other functional annotations revealed proteins that were likely involved in binding activi- ties and that were present in or related to the cell membrane. Mutations in Cry receptor genes have been reported in numerous lepidopteran species to be the most common mechanism of resistance against Bt toxins [51], and here we found outlier SNPs in genes likely coding for Cry receptors such as GPI-anchored glycoprotein, cadherin and zinc carboxypeptidase. We also found outlier SNPs in genes possibly coding many important enzymes such as cytochrome P450 CYP314A1 [52], esterase FE4-like [53], JHAMT [54], and also proteins related to cuticle and chitin, which may be an indication of response to management with insecticides [55]. Two noteworthy outlier SNPs associated with host strains were in genes possibly encoding an odorant receptor and a UDP-glucuronosyltransferase. Odorant receptors function in insects olfaction process, which is indispensable for host selection for feeding and oviposition [54]. UDP-glucuronosyltransferase, in turn, appears to be associated with C-strain ability to detox- ify DIMBOA [56], a toxic compound produced by corn plants but not rice. In conclusion, our work strongly suggest that positive selection is affecting allele frequencies at the level of popu- lations and host strains. From the Insect Resistance Management (IRM) perspective, resistance evolution is one of the most challenging problems in the sustainable control of FAW [57]. Therefore understand- ing patterns of gene flow and consequent risks for spread of field-evolved resistance alleles are crucial for effective management. Our GBS data set poses a challenging scenario in Brazil, where locations presented high levels of gene flow across all ecoregions and low genetic struc- ture within host strains. Moreover, pairwise FST distances showed genetic structure between FAW populations of Brazil and Argentina, which has important IRM implications if resistant populations are reported in either country. In conclusion, by combining classic molecular markers for FAW host strain identification, and genome-wide SNPs identified with GBS, we obtained more resolution of population struc- ture than previously reported. The genetic structure and pattern of FAW in Argentina and Brazil reinforces the importance of phytosanitary barriers between countries for effective FAW management in each location. In agreement with this issue, outlier analysis suggested that positive selection is associated with field management and host strain divergences. Taking all this into consideration, current GBS data proved to be useful for population genomics research in South America and it may be applied to other geographies where the species has been introduced. Supporting information S1 Table. Genetic diversity estimates of fall armyworm (Spodoptera frugiperda) popula- tions from 15 locations of Brazil and Argentina estimated from 3309 SNP loci. Most populations were not found at equilibrium. Brazilian populations featured lower observed PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 11 / 16 PLOS ONEPopulation genomics of fall armyworm in Brazil and Argentina heterozygosity than the expected at p < 0.05. The coefficient FIS also indicated that the Brazil- ian populations collected in corn fields had significant inbreeding. On the other hand, the fall armyworm populations from Argentina featured more outbreeding and private alleles. (DOCX) S2 Table. Gene Ontology (GO) biological process description for outlier loci under positive selection. associated with populations. associated with host strains and populations. Loci with no mark are associated with host strains. (DOCX) S3 Table. Outlier loci under positive selection with blast hits in NCBI with > 90% similar- ity. associated with populations. **associated with host strains and populations. Loci with no mark are associated with host strains. (DOCX) S4 Table. Geographic distance matrix showing the straight-line distances (Km) between locations. Darker orange indicates longer distances. (DOCX) S1 Fig. Pairwise FST for fall armyworm sampling locations. (A) FST calculated with all vari- ant loci (3309 SNPs). (B) FST calculated using neutral markers (2853 SNPs). (C) FST was cal- culated using candidates putatively under positive selection (456 SNPs) obtained by three methods (FLK, PCAdapt, FstHet). Darker green color represents a higher degree of differentia- tion. (TIF) S2 Fig. Detection of outlier SNPs under positive selection using Pcadapt, FstHet, and FLK. The Venn diagrams shows the number of outlier SNPs associated to (A) FAW host strains and (B) populations. (TIF) S3 Fig. Cross-validation method implemented in ADMIXTURE v1.3.0 to estimate the number of ancestral populations (k) inferred from (A) neutral markers and (B) outlier markers. (TIF) S4 Fig. Discriminant Analysis of Principal Components (DAPC) scatterplot showing the first two linear discriminants. Geographic locations are represented by different colors, and dots represent different individuals. The inset shows BIC values for different number of k clus- ters. Analysis performed with (A) All SNP loci, (B) 2,853 neutral SNPs, (C) 456 outliers. Plots generated using adegenet package for R software. Sampling locations were considered as priori groupings. (TIF) S5 Fig. Polymorphisms frequencies in candidate genes under selection in Brazilian loca- tions. Locations were represented mostly by C-strain moths, except for MA02 and RS loca- tions where most samples were identified as R-strain. (TIF) Acknowledgments We are truly thankful to Janaı´na Marques Mondego for helping with collections in Maranhão locations. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 12 / 16 PLOS ONEPopulation genomics of fall armyworm in Brazil and Argentina Author Contributions Conceptualization: Josemar Foresti, Maria Imaculada Zucchi. Data curation: Tamylin Kaori Ishizuka, Erick Mauricio Goes Cordeiro, Alessandro Alves- Pereira. Formal analysis: Tamylin Kaori Ishizuka, Erick Mauricio Goes Cordeiro, Alessandro Alves- Pereira. Funding acquisition: Josemar Foresti. Investigation: Tamylin Kaori Ishizuka. Methodology: Tamylin Kaori Ishizuka, Alessandro Alves-Pereira, Carlos Eduardo de Arau´jo Batista. Resources: Marı´a Gabriela Muru´a, Jose´ Baldin Pinheiro, Josemar Foresti, Maria Imaculada Zucchi. Supervision: Maria Imaculada Zucchi. Writing – original draft: Tamylin Kaori Ishizuka, Rodney N. Nagoshi, Maria Imaculada Zucchi. Writing – review & editing: Tamylin Kaori Ishizuka, Erick Mauricio Goes Cordeiro, Alessan- dro Alves-Pereira, Carlos Eduardo de Arau´jo Batista, Marı´a Gabriela Muru´a, Jose´ Baldin Pinheiro, Amit Sethi, Rodney N. Nagoshi, Josemar Foresti, Maria Imaculada Zucchi. References 1. Montezano DG, Specht A, Sosa-Go´ mez DR, Roque-Specht VF, Sousa-Silva JC, Paula-Moraes SV, et al. Host plants of Spodoptera frugiperda (Lepidoptera: Noctuidae) in the Americas. Afr Entomol. 2018; 26(2):286–300. https://doi.org/10.4001/003.026.0286 2. Gutie´ rrez-Moreno R, Mota-Sanchez D, Blanco CA, Whalon ME, Tera´ n-Santofimio H, Rodriguez-Maciel JC, et al. Field-Evolved Resistance of the Fall Armyworm (Lepidoptera: Noctuidae) to Synthetic Insecti- cides in Puerto Rico and Mexico. J Econ Entomol. 2019; 112: 792–802. https://doi.org/10.1093/jee/ toy372 PMID: 30535077 3. Westbrook JK, Nagoshi RN, Meagher RL, Fleischer SJ, Jairam S. Modeling seasonal migration of fall armyworm moths. Int J Biometeorol. 2016; 60(2):255–67. https://doi.org/10.1007/s00484-015-1022-x PMID: 26045330 4. Nagoshi RN, Vizuete JLA, Muru´a MG, Garce´ s-Carrera S. Comparisons of fall armyworm haplotypes between the Gala´ pagos Islands and mainland Ecuador indicate limited migration to and between islands. Sci Rep. 2021; 11: 3457. https://doi.org/10.1038/s41598-021-83111-5 PMID: 33568766 5. Timilsena BP, Niassy S, Kimathi E, Abdel-Rahman EM, Seidl-Adams I, Wamalwa M, et al. Potential dis- tribution of fall armyworm in Africa and beyond, considering climate change and irrigation patterns. Sci Rep. 2022; 12: 539. https://doi.org/10.1038/s41598-021-04369-3 PMID: 35017586 6. Muru´ a MG, Vera MT, Abraham S, Juare´ z ML, Prieto S, Head GP, et al. Fitness and mating compatibility of Spodoptera frugiperda (Lepidoptera: Noctuidae) populations from different host plant species and regions in Argentina. Ann Entomol Soc Am. 2008; 101: 639–649. https://doi.org/10.1603/0013-8746 (2008)101[639:FAMCOS]2.0.CO;2 7. Pashley DP. Host-associated Genetic Differentiation in Fall Armyworm (Lepidoptera: Noctuidae): A Sib- ling Species Complex? Ann Entomol Soc Am. 1986; 79: 898–904. https://doi.org/10.1093/aesa/79.6. 898 8. Pashley DP, Martin JA. Reproductive Incompatibility Between Host Strains of the Fall Armyworm (Lepi- doptera: Noctuidae). Ann Entomol Soc Am. 1987; 80: 731–733. https://doi.org/10.1093/aesa/80.6.731 9. Levy HC, Garcia-Maruniak A, Maruniak JE. Strain identification of Spodoptera frugiperda (Lepidoptera: Noctuidae) insects and cell line: PCR-RFLP of cytochrome oxidase C subunit I gene. Florida Entomol. 2003; 85: 186–190. https://doi.org/10.1653/0015-4040(2002)085[0186:SIOSFL]2.0.CO;2 10. Nagoshi RN. The fall armyworm triose phosphate isomerase (Tpi) gene as a marker of strain identity and interstrain mating. Ann Entomol Soc Am. 2010; 103: 283–292. https://doi.org/10.1603/An09046 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 13 / 16 PLOS ONEPopulation genomics of fall armyworm in Brazil and Argentina 11. Silva-Brandão KL, Peruchi A, Seraphim N, Murad NF, Carvalho RA, Farias JR, et al. Loci under selec- tion and markers associated with host plant and host-related strains shape the genetic structure of Bra- zilian populations of Spodoptera frugiperda (Lepidoptera, Noctuidae). PLoS One. 2018; 13: 1–28. https://doi.org/10.1371/journal.pone.0197378 PMID: 29787608. 12. Arias O, Cordeiro EMG, Corrêa AS, Domingues FA, Guidolin AS, Omoto C. Population genetic struc- ture and demographic history of Spodoptera frugiperda (Lepidoptera: Noctuidae): implications for insect resistance management programs. Pest Manag Sci. 2019; 75: 2948–2957. https://doi.org/10.1002/ps. 5407 PMID: 30868715 13. Martinelli S, Barata RM, Zucchi MI, De Castro Silva-Filho M, Omoto C. Molecular variability of Spodop- tera frugiperda (Lepidoptera: Noctuidae) populations associated to maize and cotton crops in Brazil. J Econ Entomol. 2006; 99: 519–526. https://doi.org/10.1093/jee/99.2.519 14. Cheng J, Kao H, Dong S. Population genetic structure and gene flow of rare and endangered Tetraena mongolica Maxim. revealed by reduced representation sequencing. BMC Plant Biol. 2020; 20: 391. https://doi.org/10.1186/s12870-020-02594-y PMID: 32842966. 15. Doyle JJ, Doyle JL. A rapid DNA isolation procedure for small quantities of fresh leaf tissue. Phytochem- ical Bulletin. 1987. pp. 11–15. 16. Thompson JD, Higgins DG, Gibson TJ. CLUSTAL W: improving the sensitivity of progressive multiple sequence alignment through sequence weighting, position-specific gap penalties and weight matrix choice. Nucleic Acids Res. 1994; 22: 4673–4680. https://doi.org/10.1093/nar/22.22.4673 PMID: 7984417. 17. Kumar S, Stecher G, Tamura K. MEGA7: Molecular Evolutionary Genetics Analysis Version 7.0 for Big- ger Datasets. Mol Biol Evol. 2016; 33: 1870–1874. https://doi.org/10.1093/molbev/msw054 PMID: 27004904. 18. Poland JA, Brown PJ, Sorrells ME, Jannink JL. Development of high-density genetic maps for barley and wheat using a novel two-enzyme genotyping-by-sequencing approach. PLoS One. 2012; 7. https:// doi.org/10.1371/journal.pone.0032253 PMID: 22389690. 19. Andrews S. FASTQC. A quality control tool for high throughput sequence data. 2010. https://www. bioinformatics.babraham.ac.uk/projects/fastqc/ 20. Catchen JM, Amores A, Hohenlohe P, Cresko W, Postlethwait JH. Stacks: Building and Genotyping Loci De Novo From Short-Read Sequences. G3 Genes\|Genomes\|Genetics. 2011; 1: 171 LP– 182. https://doi.org/10.1534/g3.111.000240 PMID: 22384329. 21. Xiao H, Ye X, Xu H, Mei Y, Yang Y, Chen X, et al. The genetic adaptations of fall armyworm Spodoptera frugiperda facilitated its rapid global dispersal and invasion. Mol Ecol Resour. 2020; 20: 1050–1068. https://doi.org/10.1111/1755-0998.13182 PMID: 32359007 22. 23. Li H. Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM. arXiv Prepr arXiv. 2013; 0(0):3. Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, et al. The Sequence Alignment/Map format and SAMtools. Bioinformatics. 2009; 25: 2078–2079. https://doi.org/10.1093/bioinformatics/btp352 PMID: 19505943. 24. Garrison E, Marth G. Haplotype-based variant detection from short-read sequencing. arXiv Prepr arXiv12073907. 2012; 9. 25. Danecek P, Auton A, Abecasis G, Albers CA, Banks E, DePristo MA, et al. The variant call format and VCFtools. Bioinformatics. 2011; 27: 2156–2158. https://doi.org/10.1093/bioinformatics/btr330 PMID: 21653522. 26. Goudet J. hierfstat, a package for r to compute and test hierarchical F-statistics. Mol Ecol Notes. 2005; 5: 184–186. https://doi.org/10.1111/j.1471-8286.2004.00828.x 27. Weir B, Cockerham C. Estimating F-Statistics for the Analysis of Population-Structure. Evolution 38: 1358–1370. Evolution (N Y). 1984;38: 1358–1370. https://doi.org/10.1111/j.1558-5646.1984.tb05657.x PMID: 28563791 28. Dray S, Dufour A-B. The ade4 Package: Implementing the Duality Diagram for Ecologists. J Stat Softw. 2007; 22: 1–20. https://doi.org/10.18637/jss.v022.i04 29. 30. Jombart T, Devillard S, Balloux F. Discriminant analysis of principal components: a new method for the analysis of genetically structured populations. BMC Genet. 2010; 11: 94. https://doi.org/10.1186/1471- 2156-11-94 PMID: 20950446. Jombart T, Ahmed I. adegenet 1.3–1: new tools for the analysis of genome-wide SNP data. Bioinformat- ics. 2011; 27: 3070–3071. https://doi.org/10.1093/bioinformatics/btr521 PMID: 21926124. 31. R Core Team. A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. 2016. http://www.R-project.org/ PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 14 / 16 PLOS ONEPopulation genomics of fall armyworm in Brazil and Argentina 32. Alexander DH, Novembre J, Lange K. Fast model-based estimation of ancestry in unrelated individuals. Genome Res. 2009; 19: 1655–1664. https://doi.org/10.1101/gr.094052.109 PMID: 19648217. 33. 34. Flanagan SP, Jones AG. Constraints on the FST-Heterozygosity Outlier Approach. J Hered. 2017; 108: 561–573. https://doi.org/10.1093/jhered/esx048 PMID: 28486592 Luu K, Bazin E, Blum MGB. pcadapt: an R package to perform genome scans for selection based on principal component analysis. Mol Ecol Resour. 2017; 17: 67–77. https://doi.org/10.1111/1755-0998. 12592 PMID: 27601374 35. Bonhomme M, Chevalet C, Servin B, Boitard S, Abdallah J, Blott S, et al. Detecting selection in popula- tion trees: the Lewontin and Krakauer test extended. Genetics. 2010; 186: 241–262. https://doi.org/10. 1534/genetics.104.117275 PMID: 20855576. 36. Go¨tz S, Garcı´a-Go´ mez JM, Terol J, Williams TD, Nagaraj SH, Nueda MJ, et al. High-throughput func- tional annotation and data mining with the Blast2GO suite. Nucleic Acids Res. 2008; 36: 3420–3435. https://doi.org/10.1093/nar/gkn176 PMID: 18445632. 37. Ye J, Zhang Y, Cui H, Liu J, Wu Y, Cheng Y, et al. WEGO 2.0: a web tool for analyzing and plotting GO annotations, 2018 update. Nucleic Acids Res. 2018; 46: W71–W75. https://doi.org/10.1093/nar/gky400 PMID: 29788377. 38. 39. Zucchi MI, Cordeiro E, Allen C, Novello M, Viana JPG, Brown PJ, et al. Patterns of Genome-Wide Varia- tion, Population Differentiation and SNP Discovery of the Red Banded Stink Bug (Piezodorus guildinii). Sci Rep. 2019; 9: 14480. https://doi.org/10.1038/s41598-019-50999-z PMID: 31597944. Zucchi MI, Cordeiro EMG, Wu X, Lamana LM, Brown PJ, Manjunatha S, et al. Population genomics of the neotropical brown stink bug, Euschistus heros: The most important emerging insect pest to soybean in Brazil. Front Genet. 2019; 10: 1035. https://doi.org/10.3389/fgene.2019.01035 PMID: 31749834. 40. Bergamo LW, Silva-Brandão KL, Vicentini R, Fresia P, Azeredo-Espin AML. Genetic Differentiation of a New World Screwworm Fly Population from Uruguay Detected by SNPs, Mitochondrial DNA and Micro- satellites in Two Consecutive Years. Insects. 2020; 11: 539. https://doi.org/10.3390/insects11080539 PMID: 32824385. 41. Anderson CJ, Tay WT, McGaughran A, Gordon K, Walsh TK. Population structure and gene flow in the global pest, Helicoverpa armigera. Mol Ecol. 2016; 25: 5296–5311. https://doi.org/10.1111/mec.13841 PMID: 27661785. 42. Cordeiro EMG, Pantoja-Gomez LM, de Paiva JB, Nascimento ARB, Omoto C, Michel AP, et al. Hybrid- ization and introgression between Helicoverpa armigera and H. zea: an adaptational bridge. BMC Evol Biol. 2020; 20: 61. https://doi.org/10.1186/s12862-020-01621-8 PMID: 32450817. 43. Nagoshi RN, Goergen G, Koffi D, Agboka K, Adjevi AKM, Du Plessis H, et al. Genetic studies of fall armyworm indicate a new introduction into Africa and identify limits to its migratory behavior. Sci Rep. 2022; 12: 1941. Epub 20220204. https://doi.org/10.1038/s41598-022-05781-z PMID: 35121788. 44. Clark PL, Molina-Ochoa J, Martinelli S, Skoda SR, Isenhour DJ, Lee DJ, et al. Population variation of the fall armyworm, Spodoptera frugiperda, in the Western Hemisphere. J Insect Sci. 2007; 7: 1–10. https://doi.org/10.1673/031.007.0501 PMID: 20334595. 45. Nagoshi RN, Rosas-Garcı´a NM, Meagher RL, Fleischer SJ, Westbrook J, Sappington TW, et al. Haplo- type profile comparisons between Spodoptera frugiperda (Lepidoptera: Noctuidae) populations from mexico with those from Puerto Rico, South America, and the United States and their implications to migratory behavior. J Econ Entomol. 2015; 108: 135–144. https://doi.org/10.1093/jee/tou044 PMID: 26470113 46. Schlum KA, Lamour K, de Bortoli CP, Banerjee R, Meagher R, Pereira E, et al. Whole genome compari- sons reveal panmixia among fall armyworm (Spodoptera frugiperda) from diverse locations. BMC Genomics. 2021; 22: 179. Epub 20210312. https://doi.org/10.1186/s12864-021-07492-7 PMID: 33711916. 47. 48. 49. Fogliata S V, Perera MF, Alves-Pereira A, Zucchi MI, Muru´ a MG. Unraveling the population structure of the sugarcane borer, Diatraea saccharalis, in Argentina. Entomol Exp Appl. 2022; 170: 530–545. https://doi.org/10.1111/eea.13174 Francischini FJB, Cordeiro EMG, de Campos JB, Alves-Pereira A, Viana JPG, Wu X, et al. Diatraea saccharalis history of colonization in the Americas. The case for human-mediated dispersal. PLoS One. 2019; 14: e0220031. https://doi.org/10.1371/journal.pone.0220031 PMID: 31339922 Tessnow AE, Raszick TJ, Porter P, Sword GA. Patterns of genomic and allochronic strain divergence in the fall armyworm, Spodoptera frugiperda (J.E. Smith). Ecol Evol. 2022; 12: e8706. Epub 20220321. https://doi.org/10.1002/ece3.8706 PMID: 35356552. 50. Prowell DP. Sex linkage and speciation in Lepidoptera. In: Howard D, Berlocher S, editors. Endless forms: Species and speciation. New York: Oxford Press; 1998. p. 309–19. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 15 / 16 PLOS ONEPopulation genomics of fall armyworm in Brazil and Argentina 51. Peterson B, Bezuidenhout CC, Van den Berg J. An Overview of Mechanisms of Cry Toxin Resistance in Lepidopteran Insects. J Econ Entomol. 2017; 110: 362–377. https://doi.org/10.1093/jee/tow310 PMID: 28334065 52. Shu B, Yu H, Li Y, Zhong H, Li X, Cao L, et al. Identification of azadirachtin responsive genes in Spodop- tera frugiperda larvae based on RNA-seq. Pestic Biochem Physiol. 2021; 172: 104745. https://doi.org/ 10.1016/j.pestbp.2020.104745 PMID: 33518039 53. Boaventura D, Buer B, Hamaekers N, Maiwald F, Nauen R. Toxicological and molecular profiling of insecticide resistance in a Brazilian strain of fall armyworm resistant to Bt Cry1 proteins. Pest Manag Sci. 2020/09/10. 2021; 77: 3713–3726. https://doi.org/10.1002/ps.6061 PMID: 32841530 54. Guo P, Zhang Y, Zhang L, Xu H, Zhang H, Wang Z, et al. Structural basis for juvenile hormone biosyn- thesis by the juvenile hormone acid methyltransferase. J Biol Chem. 2021; 297: 101234. https://doi.org/ 10.1016/j.jbc.2021.101234 PMID: 34562453 55. Zhu J-Y, Li L, Xiao K-R, He S-Q, Gui F-R. Genomic and Transcriptomic Analysis Reveals Cuticular Pro- tein Genes Responding to Different Insecticides in Fall Armyworm Spodoptera frugiperda. Insects. 2021; 12. https://doi.org/10.3390/insects12110997 PMID: 34821798 56. Silva-Brandão KL, Murad NF, Peruchi A, Martins CHZ, Omoto C, Figueira A, et al. Transcriptome differ- ential co-expression reveals distinct molecular response of fall-armyworm strains to DIMBOA. Pest Manag Sci. 2021; 77: 518–526. https://doi.org/10.1002/ps.6051 PMID: 32815313 57. Zhang L, Liu B, Zheng W, Liu C, Zhang D, Zhao S, et al. Genetic structure and insecticide resistance characteristics of fall armyworm populations invading China. Mol Ecol Resour. 2020; 20: 1682–1696. https://doi.org/10.1111/1755-0998.13219 PMID: 32619331 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284587 April 18, 2023 16 / 16 PLOS ONE
10.1371_journal.pone.0285894	RESEARCH ARTICLE Toward improvement of knowledge of financial conflicts of interest in a large medical school in France J. F. AlexandraID D. Dreyfuss2,3,7* 1,2, D. Roux3,4☯, H. Maisonneuve5☯, B. Chousterman2,5, P. Ruszniewski6, 1 APHP, Hoˆ pital Bichat Claude Bernard, Service de Me´ decine Interne DMU Victoire Paris, Paris, France, 2 Universite´ Paris-Cite´ , Paris, France, 3 AP-HP, Hoˆ pital Louis Mourier, DMU ESPRIT, Service de Me´ decine Intensive Re´ animation, Colombes, France, 4 Universite´ Paris Cite´ , INSERM UMR-S1151, CNRS UMR- S8253, Institut Necker Enfants Malades, Paris, France, 5 Herve´ Maisonneuve, MD, Scientific Editor & Consultant, Paris, France, 6 Doyen de l’UFR de Me´decine, Universite´ Paris Cite´ , Pairs, France, 7 French National Institute of Health and Medical Research (INSERM), UMR_S1155, Common and Rare Kidney Diseases (CORAKID), Hoˆ pital Tenon, Sorbonne Universite´, Paris, France ☯ These authors contributed equally to this work. * didier.dreyfuss@aphp.fr Abstract Introduction Conflict of interests (COIs) adversely affect the integrity of science and public health. The role of medical schools in the teaching and management of COIs has been highlighted by the publication of an annual evaluation of American medical schools based on their COIs policies by the American Medical Student Association (AMSA). A deontological charter was adopted by French medical schools in 2018 but its impact on COI comprehension by stu- dents and its effects on COI prevention were not evaluated. Methods A 10-item direct survey was conducted among about 1000 students in Paris-Cite´ University in order to investigate the respect of the charter regarding COIs both in the medical school and in affiliated teaching hospitals. Results Cumulative results show a satisfying respect of prevention policies regarding COIs in the medical school and hospitals despite the fact that the existence of the charter and its major aspects were insufficiently known. Disclosure of COIs by teachers was insufficient. Conclusion This first direct study among students shows better results than expected according to cur- rent non-academic surveys. Moreover, this study demonstrates the feasibility of this kind of survey whose repetition should be an appropriate tool to improve the implementation of the a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Alexandra JF, Roux D, Maisonneuve H, Chousterman B, Ruszniewski P, Dreyfuss D (2023) Toward improvement of knowledge of financial conflicts of interest in a large medical school in France. PLoS ONE 18(5): e0285894. https://doi. org/10.1371/journal.pone.0285894 Editor: Alberto Molina Pe´rez, Consejo Superior de Investigaciones Cientificas (CSIC), SPAIN Received: December 21, 2022 Accepted: May 4, 2023 Published: May 22, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0285894 Copyright: © 2023 Alexandra et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285894 May 22, 2023 1 / 8 PLOS ONEToward improvement of knowledge of financial conflicts of interest in a large medical school in France Funding: the author(s) received no specific funding for this work. charter within medical schools and teaching hospitals, in particular mandatory disclosure of COIs by teachers. Competing interests: the authors have declared that no competing interests exist. Introduction The impact of conflict of interests (COIs) on public health has become a major concern over the past fifteen years [1–3]. Starting from 2007, the American Medical Student Association (AMSA) published an annual evaluation of American medical schools based on their COIs policies. Such initiative induced major shifts in policies in the recent years [4]. In contrast with the strong policies of COIs management in many North American medical schools, a recent review showed that COIs prevention policies are ranked poorly [5] in European medical schools or teaching hospitals. In Belgium, for instance, a recent study reported little transpar- ency regarding potential COIs and only isolated initiatives to protect student from pharmaceu- tical promotion [6]. Research works have highlighted the deleterious effects of conflicts of interest on practice, research and training in medicine in France [7]. A web-based study conducted in 2012 [8] reported that medical students (preclinical, clinical and residents) are insufficiently aware of potential bias that COIs may pose with respect to drug prescriptions and want to be informed about the COIs of their lecturers. A 2017 national survey of French medical schools based on AMSA criteria [9] was con- ducted by a non-academic association named FORMINDEP (which stands for FORmation Medicale INDEPendante: Independent Medical Training). This association aims at both pro- motion of evidence-based medicine and prevention of the influence of industry’s economic interests on patients, physicians, medical students and public policy makers [10]. The same year, the conference of Deans of Schools of Medicine and Dentistry (a national institution that brings together the Deans of the 34 French medical schools and of the 16 Schools of Dentistry) [11] adopted an ethical charter. This charter meets an ethical requirement, particularly with regard to scientific and professional integrity and to COI avoidance and promotes the inde- pendence of medical training through an institutional and academic framework. A summary of the charter is available in S1 Appendix. Formindep survey was reiterated in 2018 [12] and 2021 [13]. In each instance, the conclu- sions were sobering as they indicated a quasi-complete lack of academic policy to prevent COIs, in particular those related to drugs and devices despite previous adoption of the charter. Moreover, we had no information on medical student’s awareness of the charter. In order to promote the knowledge and application of the charter, several measures have been taken by the university since 2017: 1. A 2-hour mandatory course dealing with the relationships with industry and with the man- agement and prevention of COIs was organized for all 4th year students, starting in 2018 and given by one of us (DD). Definition and examples of financial COIs and the way to dis- close and avoid it are presented. Available literature on the influence of COIs on scientific integrity is discussed. The main aspects of the charter are detailed and the link to the charter and its summary is recalled. 2. The charter was sent to all academics of medical school of Paris Cite´ University in June 2019. This university, one of the largest in France has more than 25 000 students, 1400 teachers and 34 affiliated hospitals or research centers. They were asked to read and sign it. In addition, for the sake of simplicity, a summary of the charter was drafted by the PLOS ONE \| https://doi.org/10.1371/journal.pone.0285894 May 22, 2023 2 / 8 PLOS ONEToward improvement of knowledge of financial conflicts of interest in a large medical school in France Commission of Ethics of our Medical School. This summary underlined major points of the charter and in particular the prevention and management of COIs. It was also sent to all academics. In addition, both the complete charter and its summary were posted on our Medical School website (available at . . .). The administration of the medical school and the Heads of all medical departments of the University-affiliated hospitals were requested to display the summary of the charter in all premises where teaching was provided. 3. Every year from 2019, all newly recruited assistant professors attend a session of formation. The course is mainly devoted to explain pedagogic tools and policy of our Medical School. It is also the occasion to sensitize young teachers to the importance of COI understanding and management during their teaching activity as well as for their own behaviour. Four years after the adoption of these measures, we wanted to evaluate their effects. We subsequently conducted a survey of medical students in their 4th and 5th year of medical cur- sus in our medical school which is affiliated to Universite´ Paris Cite´ between December 2021 and January 2022. The aim of this survey was to evaluate student awareness of the charter and to what extent it was applied. In particular, we aimed at investigating how COIs prevention policies were implemented in both university premises and teaching hospitals of our medical school Methods 1. The present survey was prepared in 2021. The Commission on Ethics of our medical school developed a 10-item questionnaire with two objectives: evaluation of student knowledge of the charter and evaluation of the frequency of its display in all teaching facilities (including medical school premises and university-affiliated teaching hospitals. The questionnaire aimed at evaluating every point of the charter that deals with the problem of COIs. To allow a nuanced analysis of the questionnaire all questions were on a 4-level graduation scale from “absolutely yes” to “absolutely not” (the last question only could be answered by yes or no) 2. All students in their 4th and 5th year of graduation received an e-mail informing them that they were asked to participate in an anonymized survey of the charter at the time of their examinations. They were free to answer or not. 3. At the time of the digital exams, within the university premises, students were asked to complete the questionnaire. Answers were anonymized for analysis. 4. Answers where collected by the software of the university and subsequently analysed with- out the need for any specific statistical methodology. Patient and public involvement: No patient involved in our study. French law on research involving human subjects applies only to research on patients and to surveys on health data. It therefore does not apply to an anonymized survey of medical stu- dent feeling about their medical school policy. No personal data was involved. In a study like ours, no IRB approval is required by French law. Obviously students were informed of the pur- pose of the survey and were perfectly free to refuse it. Results A total of 710 out of 1000 registered students (71.0%) in their 4th year of graduation provided at least one answer. A total of 552out of 852 (64.7%) in their 5th year of graduation provided at least one answer. Overall, there were between 1179 and 1186 answers per question for 4th and PLOS ONE \| https://doi.org/10.1371/journal.pone.0285894 May 22, 2023 3 / 8 PLOS ONEToward improvement of knowledge of financial conflicts of interest in a large medical school in France Table 1. Do your teachers indicate their COIs before the course Do the courses given to you name the molecules in INN or in therapeutic class? Absolutely yes 246 (21%) Rather yes 445 (37%) 274 (23%) 730 (63%) Do your teaching materials (books, handouts) from university contain advertising for the drug industries? 18 (2%) 43 (4%) Are you aware of the university’s Charter of ethics? Is the ethics charter clearly posted on university premises? Is the ethics charter clearly posted on the premises of your teaching hospital? Do you have meetings with drug manufacturers within your teaching hospital? 172 (15%) 353 (30%) 62 (5%) 245 (21%) 51 (4.5%) 156 (13.5%) 137 (12%) 301 (25%) Do you benefit from services in kind (breakfast, buffet, gifts, etc.) from manufacturers within your teaching hospital? 44 (4%) 102 (9%) Do you benefit from services in kind (breakfast, buffet, gifts, etc.) from manufacturers within the university premises? 33 (3%) 0 Yes Absolutely not 149 (13%) 21 (2%) 644 (54%) 305 (25%) 251 (22%) 447 (38%) 401 (34%) 807 (68%) 939 (81%) Rather not 342 (29%) 143 (12%) 472 (40%) 353 (30%) 607 (52%) 515 (44%) 344 (29%) 229 (19%) 192 (16%) No Do you know the procedure for reporting cases of breaches of the principles of the charter in your medical school (discrimination, harassment, conflict of interest)? 187 (16%) 994 (84%) https://doi.org/10.1371/journal.pone.0285894.t001 5th year students, respectively (63,7% response rate). The number of questionnaires entirely filled was 1179 (64%). Results are shown in Table 1. Eighty four (84%) percent of students indicated that their books and handbooks provided by the medical school teachers denominated products by therapeutic class or International Non proprietary Names (INN). Moreover, 94% declare that these teaching materials (books, handouts) did not contain any advertising. However, students were only 58% to report that teachers declare their COIs when they begin their course. Ninety percent (90%) declared receiving no gift nor benefit of any kind although 37% still had contact with drug manufacturers in the hospital units. The charter was poorly known and not widely displayed (63% of students declared they did not know it and between 71% and 80% did not know where it was posted in medical school premises nor in their teaching hospi- tals). Last, 80% of students did not know the procedure for reporting breaches of the charter. Discussion Our study, the first taking directly in account students’ perception in France, revealed con- trasting but encouraging elements: the spirit of the charter is widely respected although the let- ter is somewhat poorly known in our university and related teaching hospitals. The results of the first survey conducted by FORMINDEP [10] were published in 2017 under a self-explicative title: “Conflict of Interest Policies at French Medical Schools: Starting from the bottom”. The methodology allowed only indirect evaluation of schools’ performance on COI policy: some student organizations reported their appreciation of the implementation of the charter in each of the medical schools via a questionnaire. The Conference of Deans did not answer to this questionnaire despite solicitation [14]. At this time, 9/37 only of French medical schools had either introduced a related curriculum or implemented a COIs-related policy. Of these, only 1 had restrictive policies for any category. FORMINDEP conducted 2 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285894 May 22, 2023 4 / 8 PLOS ONEToward improvement of knowledge of financial conflicts of interest in a large medical school in France other studies in 2018 and 2021, in the aftermath of their initial survey in order to evaluate the national implementation of the charter [12, 13]. Results were still considered unsatisfactory: of the 36 Medical Schools, only one obtained a score above the average with 18 points out of a possible 34 (based on a 17 criteria scale). Next came 11 schools with a highest score ranging between 10 and 14 points for the highest scores. Similar findings were reported in an indepen- dent study conducted in the 32 French teaching hospitals in 2017 [15]. Only 17 had rules and regulations in link with a limited number of items of the charter. Four of them considered implementing a policy and only 2 actually started the implementation. Fifteen had no evidence of COIs policies. The accuracy of results of previous studies is beyond doubt but may have been affected by some kind of bias: first, as mentioned above, only partial or complete lack of responses was observed for 6 out of 17 criteria. Second, the survey consisted in questionnaires sent to Deans only and on observational data gathered by student organizations. There was no individual student survey. In addition, FORMINDEP studies mainly focused on declarative items, regardless of the actual content of the teachings within university or within hospitals. As a result, some items may have been underestimated. Nevertheless, this must not obscure the fact that teaching of COIs is seldom in French Medical Schools and that the vast majority of stu- dents is not aware of the charter. Thus, the persistence of poor results observed in the most recent FORMINDEP survey are unlikely to be explained by methodological bias only. Several measures were taken over recent years in order to better manage financial COIs. First, a Sales Visit charter was signed in 2014 by the health minister and the drug industry union [16]. Second, a transparency digital register directly inspired by the 2011 sunshine act was implemented in 2018 [17]. This register collects every amounts of money received by doctors from industry whatever the reason (consultancy, travel aid, meals and research contracts).Third, as already alluded to, the Deans’ charter explicitly requests that teachers display their COIs at the beginning of every lecture and prohibits the industry representatives from meeting with students, either on the teaching premises or on hospital placement sites. Last, the internal rules of our teaching hospi- tals (Assistance-Publique-Hoˆpitaux de Paris) have been amended to be consistent with the provi- sions of the charter concerning the prohibition of contact between industry representatives and students. This amendment occurred after our survey was conducted. Our survey, the first one among students in France (contrarily to other countries [18]) was conducted 3 years after implementation of these measures and focused on items reflecting the application and knowledge of the charter based directly on student statements. We chose to focus on 4th and 5th year students since these years are both those where teaching of drugs and devices is predominant and where the students are on internship at the hospital. Three points deserve mention. First, the rather good response rate shows the feasibility of this kind of survey. Thus, repeti- tive studies may be an appropriate tool to measure the implementation of the charter. Second, our results confirm in part those of previous FORMINDEP surveys but do not sustain their poor rating. In almost 90% of courses, drugs are denominated by therapeutic class and/or INN. Students hardly ever receive gifts or benefits in kind. Last, pedagogic tools provided by our medical school are free from advertising. However, a still noticeable number (about one third) of students meet drug sale representatives in teaching hospitals. Upon our request, the Assistance Publique-Hoˆpitaux de Paris (the hospital network that includes all academic hospi- tals in the Paris area) has modified its own rule in order to align it with the charter. More recently, French Health regulator (HAS) published guidelines to prevent COI within medical school [19] We can therefore reasonably expect that meetings of students with sale representative will decrease. On the other hand, poor knowledge of the existence of the charter is prevalent as PLOS ONE \| https://doi.org/10.1371/journal.pone.0285894 May 22, 2023 5 / 8 PLOS ONEToward improvement of knowledge of financial conflicts of interest in a large medical school in France more than half students are not aware of its existence (despite the fact that all were supposed to receive it via their e-mail address). A further effort should be made in order to increase the dis- play of the charter in our teaching hospitals and in the premises of our Medical School. Similar efforts should be made, to increase awareness of procedures to declare the cases of charter breach and the obligation to disclose the COIs before the course. Hence, we can assume than despite a poor knowledge of the charter itself, its principles are at least partly respected in our Medical School and its affiliated hospitals. There are potential pitfalls in our study. First, our study focused only on students’ percep- tion of the charter implementation. Thus we did not take the potential influence of partnership between teachers and industry (personal honoraria for counselling or presentations during congresses and sponsoring for research) into account. These ties may affect the integrity of medical education [20] and declaring COIs may not be sufficient to guarantee the indepen- dence of professionals [21]. Indeed, teachers are important role models for students, exerting influence on their future prescribing habits and on their future relationship with the pharma- ceutical industry [22]. Although transparency is not sufficient to guarantee independence, it is a first mandatory step. Moreover, it has been shown that students in medical schools with strict COIs policies are less likely to experience influence of drug companies [23]. Last, we did not investigate the behavior of the teachers, but they probably have a part of responsibility for the poor results observed in our study. Some teachers, perhaps the seniors, have been accustomed to extensive relationships with industry. They have had less awareness of COIs and have little motivation to change a behavior they have engaged in during their careers. An effort should be made in order to increase awareness of senior academics and promote their observance of the charter. This study seems to be both feasible and reliable and regular repetition could be an adequate tool to measure the evolution of practices. Based on these first results, the areas for improvement are well identified and will focus primarily on knowledge and display of the charter. We are undertaking an information campaign to raise awareness of the charter and will evaluate the results of this campaign in the next academic year. The next analysis will allow us to evaluate the achievement of these objectives. At the same time, an important point would be to investigate how students are sensitised to COI by their teachers. This is the purpose of a future study. We assume in accordance with an abundant literature that early interventions such as those implemented in our medical school will foster improved student knowledge of this issue [22–25]. Conclusions and perspectives Our study aimed at reinforcing the effects of studies carried out by the FORMINDEP and of recently adopted laws on COIs management. Indeed, a direct and standardized measurement of ethical and deontological charter application may improve its implementation. This study provides clear evidence of feasibility of a wide survey among a large population of medical stu- dents. Moreover, the results show that policies preventing COIs are better applied than expected despite imperfect knowledge of the charter. To improve this knowledge, we plan to reiterate the survey in our institution. In addition, we aim at implementing this survey at the national level in all French Medical Schools, under the auspices of the Deans Conference. We believe that promotion of an official national academic policy is an attainable goal. Supporting information S1 Appendix. Summary of the charter. (DOCX) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285894 May 22, 2023 6 / 8 PLOS ONEToward improvement of knowledge of financial conflicts of interest in a large medical school in France Author Contributions Conceptualization: J. F. Alexandra, D. Roux, B. Chousterman, D. Dreyfuss. Formal analysis: J. F. Alexandra, D. Roux. Investigation: J. F. Alexandra, D. Dreyfuss. Methodology: J. F. Alexandra, D. Roux, D. Dreyfuss. Supervision: H. Maisonneuve, P. Ruszniewski. Validation: P. Ruszniewski. Writing – original draft: J. F. Alexandra, D. Dreyfuss. Writing – review & editing: H. Maisonneuve, D. Dreyfuss. References 1. Kassirer JP, Angell M. Financials conflicts of interest in biomedical research N Engl J Med 1993 3; 329:570–1 2. Rodwin MA. Conflicts of interest and the future of medicine: The United States, France and Japan. Oxford University Press New York 2011; pp 374. 3. Austad KE, Kesselheim AS. Conflict of interest disclosure in early education of medical students. JAMA. 2011; 306:991–2 https://doi.org/10.1001/jama.2011.1233 PMID: 21900140 4. AMSA Just medicine scorecard 2015 available at: https://ifmsa.org/wp-content/uploads/2019/09/ AM19-Integrity-and-transparency-in-medical-education.pdf [Accessed on September 2, 2022] 5. Fabbri A, Hone KR, Hro´bjartsson A et al. Conflict of Interest Policies at Medical Schools and Teaching Hospitals: A Systematic Review of Cross-sectional Studies. Int J Health Policy Manag. 2022, 11(8), 1274–1285 6. Bechoux L, De Vleeschouwer O, Vanheuverzwijn C et al. Conflict of interest policies at Belgian medical faculties: Cross-sectional study indicates little oversight. PLoS ONE 16(2): e0245736. [Accessed on September 2, 2022] https://doi.org/10.1371/journal.pone.0245736 PMID: 33566836 7. Hauray B, Boullier H, Gaudillière JP, Michel H, Editors. Conflict of interest and medicine. Knowledge, practices, and mobilizations. Routledge, Oxon UK 2022; pp 251. 8. Etain B, Guittet L, Weiss N et al. Attitudes of medical students towards conflict of interest: a national sur- vey in France. PLoS One. 2014 Mar 26; 9(3):e92858. https://doi.org/10.1371/journal.pone.0092858 eCollection 2014 [Accessed on September 2, 2022] PMID: 24671179 9. Scheffer P, Guy-Coichard C, Outh-Gauer D et al. Conflict of Interest Policies at French Medical Schools: Starting from the Bottom. PLoS One. 2017 Jan 9; 12(1):e0168258. https://doi.org/10.1371/ journal.pone.0168258 eCollection 2017 [Accessed on September 2, 2022] PMID: 28068362 10. https://formindep.fr 11. Available at https://conferencedesdoyensdemedecine.org/wpcontent/uploads/2020/10/V3_Charte_ faculte%CC%81s_me%CC%81decine_odontologie_2017.pdf [Accessed on September 2, 2022] 12. Avaible at https://facs.formindep.fr/2018/index.html 13. Avaible at https://facs.formindep.fr/index.html 14. ULaval. Politique en matière de conflits d’inte´ rêts dans les activite´ s de formation, d’e´valuation et de recherche [Internet]. 2015. Available from https://www.fmed.ulaval.ca/fileadmin/documents/faculte- reseau/documents-officiels-et-promotionnels/documents/politique-conflits-interets-activites-formation- recherche.pdf [Accessed on September 2, 2022] 15. Guy-Coichard C, Perraud G, Chailleu A et al. Inadequate conflict of interest policies at most French teaching hospitals: A survey and website analysis. PLoS One. 2019 Nov 1; 14(11):e0224193. https:// doi.org/10.1371/journal.pone.0224193 eCollection 2019 [Accessed on September 2, 2022] PMID: 31675383 16. Les entreprises du me´dicament, Comite´ e´ conomique des produits de sante´. Charte de l’information ou prospection visant la promotion des me´ dicaments. October 15, 2014 available at: https://solidarites- sante.gouv.fr/IMG/pdf/CHARTE_CEPS_LEEM_promotion_vm_signee.pdf [Accessed on September 2, 2022] PLOS ONE \| https://doi.org/10.1371/journal.pone.0285894 May 22, 2023 7 / 8 PLOS ONEToward improvement of knowledge of financial conflicts of interest in a large medical school in France 17. Ministère des solidarite´s et de la sante´. Transparence sante´ public. available at: https://www. transparence.sante.gouv.fr [Accessed on September 2, 2022] 18. Saito S., Maeno T., Miyata Y. et al. Medical students’ attitudes toward interactions with the pharmaceuti- cal industry: a national survey in Japan. BMC Med Educ 18, 286 (2018). https://doi.org/10.1186/ s12909-018-1394-9 PMID: 30509273 19. Ehringhaus SH, Weissman JS, Sears JL. Responses of medical schools to institutional conflicts of inter- est. JAMA. 2008 Feb 13; 299(6):665–71 https://doi.org/10.1001/jama.299.6.665 PMID: 18270355 20. Avaible at https://www.has-sante.fr/upload/docs/application/pdf/2022-12/has_090_synthese_revue_ info_promo_cd_20221208_vf.pdf 21. Angell M, Kassirer JP. Editorials and conflicts of interest. N Engl J Med. 1996 Oct 3; 335(14):1055–6 https://doi.org/10.1056/NEJM199610033351410 PMID: 8793932 22. Austad KE, Avorn J, Kesselheim AS. Medical Students’ Exposure to and Attitudes about the Pharma- ceutical Industry: A Systematic Review. Lexchin J, editor. Med PLoS. 2011. May 24; 8(5):e1001037 https://doi.org/10.1371/journal.pmed.1001037 [Accessed on September 2; 2022] PMID: 21629685 23. King M, Essick C, Bearman P, Ross JS. Medical school gift restriction policies and physician prescribing of newly marketed psychotropic medications: difference-in-differences analysis. BMJ. 2013 Jan 30; 346:f264 https://doi.org/10.1136/bmj.f264 PMID: 23372175 24. Kao AC, Braddock C, Clay M, et al. Effect of Educational Interventions and Medical School Policies on Medical Students’ Attitudes Toward Pharmaceutical Marketing Practices: A Multi-Institutional Study: Acad Med. 2011. November; 86(11):1454–62 https://doi.org/10.1097/ACM.0b013e3182303895 PMID: 21952057 25. Holloway K. Uneasy subjects: Medical students’ conflicts over the pharmaceutical industry. Soc Sci Med. 2014. August; 114:113–20. https://doi.org/10.1016/j.socscimed.2014.05.052 [Accessed on Sep- tember 2, 2022] PMID: 24922608 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285894 May 22, 2023 8 / 8 PLOS ONE
10.1371_journal.pone.0283546	RESEARCH ARTICLE Maternal effect in salinity tolerance of Daphnia–One species, various patterns? Andrzej MikulskiID, Danuta Mazurczak Department of Hydrobiology, Institute of Functional Biology and Ecology, Faculty of Biology, University of Warsaw, Warsaw, Poland a.mikulski@uw.edu.pl Abstract We experimentally tested the hypothesis that individuals from a single species but geneti- cally different exposed to the same chemical stress factor are able to realize opposite life history strategies–they can invest more resources in current reproduction and release neo- nates well-prepared to harmful condition or they can invest in their own safety as well as future reproductions and release neonates of poor quality condition. In order to do this, we used the Daphnia-salinity model: we exposed Daphnia magna females originating from vari- ous ponds to two concentrations of sodium chloride, and then observed the key life histories parameters of their offspring exposed or not exposed to salinity stress. Our results con- firmed the hypothesis. In a clone from one pond, Daphnia exposed to salinity stress pro- duced neonates which were worse-prepared to the local conditions than those released by non-stressed females. In clones from the two other ponds, Daphnia released newborns sim- ilarly or better-prepared to cope with the salinity stress, depending on the concentration of salt and the duration of their exposure to salinity. Our results suggest that both longer (two- generational) and stronger (higher salt concentration) impacts of selective factors may be perceived by individuals as information indicating reduced chances of successful reproduc- tion in the future and, thus, they may drive mothers to produce better-prepared descendants. Introduction Among the forces determining the phenotype of individual organisms and, in consequence, their fitness [1], population dynamics [2] and microevolutionary processes [3, 4], the role of maternal effect remains one of the key unsolved questions. Each form of influence of maternal phenotype or genotype on the phenotype of descendants is referred to as the maternal effect [5]. In this context, maternal effect is usually interpreted as an intergenerational phenotypic plastic- ity. Most studies of phenotypic plasticity are dedicated to investigate its adaptive role, enabling the interpretation of results in a broad evolutionary context. In consequence, adaptive maternal effect is, in fact, the most commonly studied of these. Additionally, researchers studying mater- nal effect have usually tried to find simple cases where mothers who were exposed to adverse environmental factors would produce offspring better prepared to deal with them [6–12]. Cur- rent insight into maternal effect requires analysing the intergenerational phenotypic plasticity a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Mikulski A, Mazurczak D (2023) Maternal effect in salinity tolerance of Daphnia–One species, various patterns? PLoS ONE 18(4): e0283546. https://doi.org/10.1371/journal.pone.0283546 Editor: Elena Gorokhova, Stockholm University, SWEDEN Received: September 19, 2022 Accepted: March 11, 2023 Published: April 4, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0283546 Copyright: © 2023 Mikulski, Mazurczak. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: We have uploaded the dataset from our experiment to a public repository RepOD - https://doi.org/10.18150/ KFMBPG. Funding: AM Grant No NN304138940 Polish Ministry of Science and Higher Education. The PLOS ONE \| https://doi.org/10.1371/journal.pone.0283546 April 4, 2023 1 / 13 PLOS ONEfunders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. within the context of maternal fitness, considering not only the fitness of particular descendants but also the number of descendants in subsequent broods or the whole-life reproductive success of mother [13]. Many offspring traits which were previously treated as maladaptive, may in this context be considered to be examples of strategy-increasing maternal fitness. Generally, in the studies of maternal effect, two types of scenarios have usually been identi- Maternal effect in salinity tolerance of Daphnia fied, depending on the degree of provisioning of offspring. In the first one, stressed mothers produce poor quality neonates, shifting the cost of living in adverse conditions to offspring [14] and/or they direct resources to the subsequent reproduction [15, 16]. The second sce- nario, so-called anticipatory maternal effect [17], i takes place when mothers exposed to an environmental threat produce neonates better-prepared to face such conditions than those released by females living in benign conditions. There is also another, until now overlooked, scenario. Stressed females can produce progenies of similar quality as non-stressed mothers (equally prepared to face stressful conditions), change the trade-off between the number and size of the neonates or compensate costs of living under unfavourable environmental condi- tions by impairing other life history traits (e.g. by delaying maturity). The most intriguing question is what drives the maternal selection of a particular strategy. The likelihood of breeding again in the future is crucial for this ‘decision’–the existing strate- gies of semelparous and iteroparous organisms significantly differ in this aspect [18]. The chance of reproducing under better conditions in the future is another frequently-discussed issue [17]. It is interesting if the strategy is determined at the species level or it varies between genotypes within a species. The ideal organismal model to be used in investigating such a problem would be an iteropa- ric, clonal animal able to adapt to local environmental conditions. This would allow to observe various maternal effect strategies in multiple clones within one species at controlled strength and duration of stress. The biology of the water flea Daphnia fully satisfies these demands. This organism is commonly used as a model in studies on phenotypic plasticity, including its intergenerational mode [19, 20]. The role of maternal effect in the expression of the phenotypic reaction of Daphnia to environmental factors has been extensively discussed [5, 21] and described in the context of predator-induced shifts in morphology [7], resting egg production induced by deteriorating environmental conditions [22–24] and enhanced tolerance to toxic cyanobacteria [25] and parasitic diseases [26]. Maternal contributions to predator-induced changes in Daphnia life history were also demonstrated by Mikulski and Pijanowska [24, 27]. The ideal environmental threat should be one that is common in nature and it should act as an effective selective factor acting proportionally to its concentration, from a small impairment of life history to a lethal effect. Salt seems to be a good choice for such studies. An excess of salinity leads to dehydration of tissues in aquatic organisms, which disturbs many life func- tions. This effect, as well as the complex mechanisms responsible for tolerance to salinity is well known in Daphnia [28]. LC50 (the median of a lethal concentration) of salinity for this cladoceran is about 5.5 g NaCl L-1 [29, 30]. Increased Daphnia mortality caused by salinity was observed by many authors [29–32]. Under salinity stress, Daphnia growth rate [33] and size at first reproduction decrease [34–36], age at first reproduction increases [35, 37] and number of neonates significantly decreases [34, 35]. Daphnia demonstrate local adaptations to salinity [38] and salinity can strongly modify the effects of other adverse biotic [39, 40] and abiotic [41, 42] factors affecting Daphnia life history. The main aim of the study was to test (using the Daphnia-salinity model) the hypothesis that individuals from one species but different genotypes exposed to the same chemical stress factor are able to realize opposite life history strategies–they can invest mostly in current repro- duction and release neonates well-prepared to harmful condition or they can invest in their own safety as well as future reproductions and release neonates of poor quality. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283546 April 4, 2023 2 / 13 PLOS ONEMaternal effect in salinity tolerance of Daphnia Methods In order to test our hypothesis, we aimed at assaying a broad spectrum of Daphnia strategies to deal with excessive salinity, and thus we used three clones of Daphnia magna from three habitats representing opposite extremes of crucial environmental gradients. All studied clones were established by hatching resting eggs isolated from the natural envi- ronment. The first clone (C1) originated form Binnesee (north-west Germany), which is a large, brackish lake (area 47 790 000 m2, max. depth 3 m) inhabited by fish and occasionally salted by inflow from the Baltic sea (salinity reached 2.5‰). The second clone (TO) was iso- lated from a small, freshwater, astatic Topiel pond located in Warsaw (area 4785 m2, max. depth 0.45 m). The third clone (KS) originated from the Książęca pond, a larger, freshwater, astatic concreted pond located in Warsaw (area 808 m2, max. depth 0.65 m). Prior to the experiments, to eliminate the interclonal phenotypic differences that could be caused by a directional maternal effect, animals from all clones were cultured for three genera- tions under constant conditions, the same as in the experiments–individually in 200 ml glass under constant dim light and at a temperature of 20˚C; they were fed green algae Scenedesmus obliquus concentrated at 1 mg Corg. � L-1 (the medium with food was changed daily). The base of the medium was lake water with a low salt content (conductivity below 400 μS/cm, chlorides bellow 60 mg/L), aerated for several weeks and filtered before use (filter size 0.2 μm). The use of natural salt-containing water in the experiment was intended to compare animals from par- ticular treatments to control animals reared in comfortable salinity conditions. Neonates from the second clutch were used to establish the next generation. Neonates released by females from the second clutch of the third generation were split into three groups of 10 individuals each (Fig 1), and placed individually into 200 ml of the appropriate medium. Individuals from the first group were cultured in the control medium. Individuals from the second were cul- tured in a medium with an addition of low concentration of sodium chloride (3.5 g NaCl � L- 1). Those from the third group were cultured in a medium with an addition of low concentra- tion of sodium chloride (4.5 g NaCl � L-1). All females were transferred to the control medium shortly before releasing their eggs to the brood chambers. Next, neonates produced by three randomly selected females from each maternal treatment were randomly split into the same groups as the mothers (cultured without salt addition, with 3.5 and 4.5 g NaCl � L-1). Conse- quently, nine groups of 10 individuals were obtained which differed in their combinations of maternal and daughter environments. Additionally, ten neonates from each maternal group were used to determine the initial weight of individuals from this generation (W0). Crucial life history parameters were measured: size at birth to a 1 μm accuracy (using NIS-elements Nikon software), age at first reproduction as the age at the moment of releasing first-clutch neonates from the brood chamber (to a 1 hour accuracy), total dry mass at first reproduction of a single female, including mass of the first-clutch neonates (with 0.1 μg accuracy using Orion Cahn C- 35 Ultra-Microbalance, Thermo Electron Corporation, USA), number of first-clutch neonates, dry mass of a single neonate in the first clutch (to a 0.1 μg accuracy), and total growth-rate including reproductive investment (i.e. mass of whole first clutch). Before weighing, Daphnia were individually placed into aluminium ‘boats’ and dried for 24 hours at 60˚C. Growth rate Gj was calculated using the formula: Gj ¼ lnW1 (cid:0) t1 (cid:0) lnW0 t0 Where: W0 –weight of newborn Daphnia W1 –weight of adult Daphnia (with first-clutch neonates—max. one hour after they have been released) PLOS ONE \| https://doi.org/10.1371/journal.pone.0283546 April 4, 2023 3 / 13 PLOS ONEMaternal effect in salinity tolerance of Daphnia Fig 1. Experimental design. https://doi.org/10.1371/journal.pone.0283546.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283546 April 4, 2023 4 / 13 PLOS ONEMaternal effect in salinity tolerance of Daphnia t1 –t0 time from birth to reach maturity Juvenile growth rate Gj is a parameter used as a good proximation of fitness in Daphnia [43]. Unfortunately, mortality (often associated with toxicity) renders the connection between Gj and fitness less credible (fitness such expressed is overestimated when mortality occurs). In Daphnia, Gj may be still better measure of their coping with toxicity than r, because it is de facto the measure of net assimilation rate, therefore, it takes into account the direct costs aris- ing from the toxicity. Parameter r is not adequate because it does not consider impairing of assimilation rate under toxicity and costs associated with detoxification. It is very important given that energetic reserves seem to be crucial for fitness under such conditions. We limited our analysis to the first reproduction of offspring generation because (1) under unpredictable environmental impact, the first reproduction seems to be subjected to the pre- cise optimization (it may be the only opportunity to reproduction), (2) we tried to connect analysis of life history parameters related to reproduction (e.g. age at first reproduction, num- ber of neonates etc.) with analysis of Gj. Culturing experimental animals till the second repro- duction would make Gj analysis more difficult, and (3) we tried to analyze the same reproduction episode in both generations. The effects of clone, maternal experience of salinity and direct effect of salinity on Daphnia life history parameters were tested using a MANOVA model. Next, maternal effect and direct effect of salinity on single Daphnia traits were tested using ANOVA with a T-Tukey test (for different N–e.g. Spjotvoll-Stoline test) as a post-hoc. According to European low, for isolation of resting eggs from public lakes and for inverte- brate experiments permits are not required. Results We observed variability in Daphnia survivorship during the experiment. Mortality seemed to be stochastic and did not obstruct the analysis of the results. The exception was the group of individuals from TO clone, exposed to the highest salt concentration for two generations, where mortality exceeded 50%. It makes the conclusions less credible in this particular case. The clonal origin (MANOVA, Λwilks (10,412) = 0.03, P < 0.00001) as well as direct (via own experience) (MANOVA, Λwilks (10,412) = 0.26, P < 0.00001) and indirect (via mother’s experi- ence) (MANOVA, Λwilks (10,412) = 0.66, P < 0.00001) effect of the salt concentration signifi- cantly affected key parameters of the Daphnia life history. We also found a very strong effect for the interactions between all combinations of these factors (MANOVA, Λwilks (40,900) = 0.57, P < 0.00001). C1 clone Maternal experience of salinity did not affect size at birth of Daphnia from clone C1 (ANOVA, F(2,25) = 1.33, P = 0.28173, Fig 2A), but the direct exposure to salinity had a significant impact on most of the life history parameters. Both: indirect (ANOVA, F(2,71) = 2.58, P = 0.00003) and direct (ANOVA, F(2,71) = 46.32, P < 0.00001) effects of salinity influenced Daphnia age at first reproduction in this clone. There was also a strong interaction between these two factors (ANOVA, F(4,71) = 7.87, P = 0.00003). Experiencing salinity significantly delayed maturity only in animals released by females exposed to a high concentration of salt (Fig 2B). Indirect (ANOVA, F(2,71) = 21.49, P = 0.00002) and direct (ANOVA, F(2,71) = 12.49, P < 0.00001) impacts of salinity affected total mass of Daphnia from the C1 clone during first reproduction. There was no interaction between these two factors (ANOVA, F(4,71) = 2.04, P = 0.09785). Females exposed to high concentrations of salt and released by females exposed to salt were smaller than the others (Fig 2C). PLOS ONE \| https://doi.org/10.1371/journal.pone.0283546 April 4, 2023 5 / 13 PLOS ONEMaternal effect in salinity tolerance of Daphnia Fig 2. Direct and indirect (via maternal effect) effects of NaCl on life history parameters of Daphnia magna from three clones (averages ± 1 SD); large arrows and their directions indicate significant differences compared to animals born from females not exposed to increased salinity, post-hoc Tukey test calculated only for maternal effect (—p<0.05, —p<0.01, **—p<0.001); small letters indicate homogenous groups, post-hoc Tukey test for different N (Spjotvoll-Stolone test), p<0.05. https://doi.org/10.1371/journal.pone.0283546.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283546 April 4, 2023 6 / 13 PLOS ONEMaternal effect in salinity tolerance of Daphnia Although maternal (ANOVA, F(2,71) = 13.93, P < 0.00001) and direct (ANOVA, F(2,71) = 12.62, P = 0.00001) experience of salinity affected the number of neonates released by Daphnia from the C1 clone, there was no interaction between these factors (ANOVA, F(4,71) = 0.34, P = 0.84936). Both maternal and direct effect of salt similarly decreased the number of newborns released by females (Fig 2D). None of the factors under study had an influence on the mass of single neonates (maternal effect: ANOVA, F(2,71) = 0.21, P = 0.80937, direct effect of salinity: ANOVA, F(2,71) = 1.28, P = 0.28409). There was, however, a significant interaction between the examined factors (ANOVA, F(4,71) = 4.08, P = 0.00492, Fig 2E). Maternal (ANOVA, F(2,71) = 26.68, P < 0.00001) and direct (ANOVA, F(2,71) = 48.96, P < 0.00001) effects of salt significantly influenced Daphnia growth rate from the C1 clone, but these two factors did not interact (ANOVA, F(4,71) = 1.47, P = 0.21921). Likewise, in the case of the number of first-clutch neonates, both maternal and direct effect of salt similarly decreased the growth rate of females from this clone (Fig 2G). TO clone Maternal experience of salt had a significant impact on size at birth of females from the TO clone (ANOVA, F(2,20) = 11.29, P = 0.00052). Daphnia released by females exposed to high salt concentrations were significantly larger at birth than others (Fig 2G). In relation to other life history traits, maternal effect seems to be much less visible in the TO clone than in the C1 clone. In the TO clone, maternal experience of salinity did not affect Daphnia age at first repro- duction (ANOVA, F(2,65) = 1.68, P = 0.19485), and did not interact with the direct effect of this factor (ANOVA, F(4,65) = 1.69, P = 0.16288). Direct effect was significant (ANOVA, F(2,65) = 23.50, P < 0.00001) in delaying the reproduction of animals exposed to high concentrations of salt (it was significant only in the case of Daphnia born from non-stressed mothers–Fig 2H). Similarly, maternal experience of salinity did not affect Daphnia mass at first reproduction (ANOVA, F(2,65) = 3.06, P = 0.05360) and did not interact with direct effect of this factor (ANOVA, F(4,65) = 0.25, P = 0.90637). However, direct effect was highly significant (ANOVA, F(2,65) = 41.47, P < 0.00001). Daphnia exposed to high concentrations of salt were smaller than others (Fig 2I). There was significant maternal effect (ANOVA, F(2,65) = 3.78, P = 0.02793) and direct effect of salt concentration (ANOVA, F(2,65) = 61.65, P < 0.00001) on number of neonates but there was no interaction between these factors (ANOVA, F(4,65) = 0.71, P = 0.58763). High concen- trations of salt caused nearly a twofold reduction in the number of neonates. Maternal experi- ence of high salinity had a similar, albeit weaker, effect (Fig 2J). There was a significant effect of maternal (ANOVA, F(2,65) = 22.41, P < 0.00001) and direct (ANOVA, F(2,65) = 56.52, P < 0.00001) effect of salinity on the mass of single neonates from the TO clone. There was also a significant interaction between these factors (ANOVA, F(4,65) = 15.87, P < 0.00001, Fig 2K). The direct effect of salt influenced the growth rate of Daphnia from TO clones (ANOVA, F(2,65) = 60.64, P < 0.00001). Neonates released by females exposed to high concentrations of salt had a lower growth rate as compared to neonates from the control and low salt treatments (Fig 2). Although there was a weak effect of maternal experience of salinity on growth rate of neonates (ANOVA, F(2,65) = 3.75, P = 0.02864), this was not reflected in any of the results of the post-hoc test (Fig 2L). KS clone Maternal experience of salinity strongly impacted size at birth of Daphnia from the KS clone (ANOVA, F(2,25) = 21.50, P < 0.00001). Daphnia released by females exposed to high PLOS ONE \| https://doi.org/10.1371/journal.pone.0283546 April 4, 2023 7 / 13 PLOS ONEMaternal effect in salinity tolerance of Daphnia concentrations of salt were significantly larger at birth than others (Fig 2M). The maternal effect on other life history traits of Daphnia from this clone was more complex and strongly dependent on the concentration of salt to which the mothers were exposed. Maternal (ANOVA, F(2,74) = 11.89, P = 0.00003) and direct (ANOVA, F(2,74) = 30.42, P < 0.00001) experience of salinity significantly influenced the age at first reproduction of Daphnia from the KS clone. There was also a significant interaction between maternal and direct effect (ANOVA, F(4,74) = 6.14, P = 0.00025). Salinity delayed maturation in Daphnia exposed to low and high concentrations of salt, but this was significant only among those neonates released by mothers exposed to low salinity stress (Fig 2N). Maternal (ANOVA, F(2,74) = 19.26, P < 0.00001) and direct effect of salinity (ANOVA, F(2,74) = 9.17, P = 0.00028) significantly influenced the weight of females at first reproduction, but there was no significant interaction between these factors (ANOVA, F(4,74) = 1.63, P = 0.13578). The presence of salt in low concentrations caused the increase of the weight of repro- ducing females, but it was significant only among Daphnia released by females exposed to high concentrations of salt (Fig 2O). Maternal effect did not influence the number of first-clutch neonates released by females from the KS clone (ANOVA, F(2,74) = 2.03, P = 0.13805), but there was a significant interaction between the maternal and direct effects of salinity (ANOVA, F(4,74) = 3.11, P = 0.02025). The direct effect of salinity was also significant (ANOVA, F(2,74) = 20.80, P < 0.00001). Maternal experience of the presence of salt increased the number of neonates in non-stressed Daphnia (Fig 2P). The weight of a single first-clutch neonate depended on both maternal (ANOVA, F(2,74) = 20.76, P < 0.00001) and direct (ANOVA, F(2,74) = 36.29, P < 0.00001) effects of salinity. There was also significant interaction between these factors (ANOVA, F(4,74) = 9.55, P < 0.00001). Females exposed to high concentrations of salt released larger neonates. Larger neonates were released by both: highly stressed females and those confronted with low salinity, but born by highly stressed mothers (Fig 2R). Maternal effect (ANOVA, F(2,74) = 14.41, P < 0.00001), direct effect of salinity(ANOVA, F(2,74) = 23.74, P < 0.00001) and their interaction between these factors (ANOVA, F(4,74) = 5.09, P = 0.00112) influenced the total growth rate of Daphnia from the KS clone. Increased levels of salinity caused the reduction of the growth rate which was most visible among the Daphnia neonates released by females exposed to low concentrations of salt (Fig 2S), and were not visible in female neonates released by highly-stressed mothers (in these cases, there was no effect of salinity). Discussion The observed reaction of Daphnia to salinity was consistent with earlier published data. The mortality observed during the experiment is comparable with that described before under sim- ilar salt concentrations [29, 31]. Animals exposed to this stress factor mature later [35, 37] and are smaller at first reproduction [34–36]. They also release fewer neonates [34, 35], but the neonates are similar or larger than neonates released by not stressed females. Size of neonates is the only life history feature not impaired by salinity in our experiment (Fig 2E, 2K, 2R). This result is new and shows a peculiar ‘reluctance’ of Daphnia to transferring to offspring the costs of living under adverse conditions As our results show, Daphnia has no universal mechanism of reacting to increased salinity. Each investigated clone reacted in its own, individual way. In clone C1, from Binnensee (Fig 2, left panel), individuals from the first generation which were exposed to salinity showed no change in age at first reproduction nor in mass of first PLOS ONE \| https://doi.org/10.1371/journal.pone.0283546 April 4, 2023 8 / 13 PLOS ONEMaternal effect in salinity tolerance of Daphnia neonates, but their other characteristics were impaired: weight at maturity, number of first- clutch neonates and total growth rate decreased, most often in proportion to the strength of the stress factor. The reaction of animals from the next generation was stronger, and the impairment of their life history traits depended on the level of salinity experienced by the mothers. This was clearly visible in age at first reproduction (Fig 2B)–only neonates released by females exposed to salt delayed maturity when confronted with increased salinity them- selves. Thus, in this clone, mothers presumably ‘transferred’ to offspring the costs of living in an uncomfortable environment. A different scheme was employed by Daphnia from the Topiel pond (TO) (Fig 2 –central panel). Low concentrations of salt had no influence on their life history traits. In a similar manner to the C1 clone, high concentrations of salt caused a decrease in weight at the age of first reproduction, as well as a decrease in the number of first-clutch neonates and the growth rate. However, the contrary phenomenon was also observed–females exposed to high levels of salinity released larger neonates than others (Fig 2K). Generally, individuals released by stressed mothers were not impaired by salinity compared to those released by non-stressed animals. Their reaction to salinity was similar, with one exception: though Daphnia exposed for two generations to high salinity modified their reproductive strategy to ‘more K’–they lim- ited the number of neonates they released, and produced much larger (almost twice as large) newborns. Apart from the last example, it can be concluded that mothers of TO clone exposed to salinity produce offspring of similar ‘quality’ to those of non-stressed females (they do not transfer the costs of living in a harmful environment to their offspring). Females giving birth to exceptionally large neonates when exposed to high levels of salinity stress for two genera- tions sanctions the hypothesis that Daphnia are capable of releasing offspring better prepared to cope with high levels of salinity than those released by non-stressed mothers. High mortality (up to 60%) observed in this clone (among individuals exposed to the highest salt concentra- tion for two generations) undermines this conclusion. However, the range of values of all mea- sured life history parameters among surviving individuals consistently differed from other treatments which, again, validates this conlusion. Reaction to salinity in the clone from Książęca pond (KS) (Fig 2 –right panel) seems to be the most complex. In the first generation exposed to high levels of salinity, the cost of living in a salty environment was visible among Daphnia exposed to low concentrations of salt, delaying the age at first reproduction which, in consequence, resulted in decreased total growth rate. Daphnia exposed to high salinity tended to decrease the number and increase the size of their first-clutch neonates, so they react as indi- viduals from clone TO exposed to salinity for one generation longer. Maternal effect also depends heavily on salt concentration. Daphnia released by females exposed to low concentra- tions of salt seemed to bear the costs of the maternal environment, in a similar way to the maternal generation, resulting in delayed maturation and thus a decrease in total growth rate. Daphnia released by females exposed to high levels of salinity did not bear such costs, although they prepared their offspring to face harmful conditions (individuals exposed to salinity pro- duced larger neonates than the others, even those which were exposed to low concentrations of salt). Most importantly, salinity did not impair their growth rate (Fig 2S), so they seemed to be much better prepared for living in high salinity conditions than neonates released by females exposed to low concentrations of salt or not exposed to salt at all. Three different maternal strategies of Daphnia living under conditions of high salinity were observed in our study. In the first strategy, which was shown by Daphnia from the C1 clone, offspring bore the costs that their mothers incurred in harmful environments, being worse-prepared to current conditions than neonates released by non-stressed females. The second strategy was seen in the TO clone, and in those females from KS clone which were released by mothers exposed to low salinity. In this case, females did not transfer the costs of living in harmful environments PLOS ONE \| https://doi.org/10.1371/journal.pone.0283546 April 4, 2023 9 / 13 PLOS ONEMaternal effect in salinity tolerance of Daphnia to the next generation and produced offspring similarly prepared to the current conditions as those released by females living in salt-free conditions. In the third scenario, observed in the case of Daphnia from the KS clone exposed to high levels of salinity, mothers prepare offspring to adverse conditions and produce more fit descendants (which reproduce when achieving larger body sizes and then release larger neonates) than those released by non-stressed females. Choice of maternal strategy does not depend solely on genotype. In a single clone, different strategies may be realized depending on the degree of harmful threat encountered in the envi- ronment or the number of generations the organisms are exposed to this threat. The change of maternal strategy under two-generational exposure to high levels of salinity (environmental stress) is so far a poorly-described ecological phenomenon. Longer exposure to salinity stress drives a mother Daphnia to better prepare offspring—to living in adverse conditions, an effect similar to that shown by animals reacting to the highest concentrations of salt. Longer expo- sure to salinity stress and the intergenerational transfer of its costs may be a physiological equivalent of short term exposure to higher levels of a stress factor. On the other hand, long- term exposure can carry information about the persistence of environmental stress and may be perceived by individuals as a forecast of reduced chances for future reproduction (see [17]). As a consequence, longer exposure to abiotic stress may promote increased investment in cur- rent reproduction. Increased investment in current reproduction in maternal strategy is also connected with increasing level of stress factors. Experiencing the sub-lethal salt concentration decreases the probability of survival until the next opportunity to reproduce and promotes greater invest- ment in the quality of neonates from the current reproduction. This observation supports ear- lier identification of factors determining the adoption of particular strategies by mothers [17, 18]. A weak relationship was found between the size / weight and ‘quality’ of the offspring. Stressed females should release significantly larger (better equipped) newborns to ensure off- spring resistance to stress similar to the resistance of offspring born by non-stressed females (the TO clone). Neonates of sizes similar to those neonates released by non-stressed females turned out to be unable to cope efficiently with increased levels of salinity. Size at birth is not always a good predictor of Daphnia fitness and maternal effect is not limited to determining the quantity of resources uploaded to eggs, but also in the transfer of information in the form of hormones and other transcription factors (see e.g. [5]) which can reprogram individual development and adjust it to current conditions. The adaptive maternal effect was usually interpreted as a kind of “immunization” against specific selection factors which the females predict [6–12]. However, it is a mechanism imple- mented by mothers, which is hard to analyse properly abstracting from maternal fitness. The adaptive maternal strategy may result in no change in the fitness of a single offspring individ- ual or even a decrease in its fitness. This creates a continuum of strategies involving energy allocation and using various mechanisms to influence the ontogenesis of offspring. Our results show that the choice of such a strategy is not necessarily determined genetically, but may itself be an element of phenotypic plasticity. It also shows the great complexity of optimization mechanisms related to reproduction and the weakness of fitness analyses that neglect the fate of the offspring. Conclusions 1. Opposite maternal strategies–investing mostly in current reproduction and release neo- nates well-prepared to harmful condition or investing in own safety and future reproduc- tions but releasing neonates of poor quality–may be realized in individuals exposed to the PLOS ONE \| https://doi.org/10.1371/journal.pone.0283546 April 4, 2023 10 / 13 PLOS ONEMaternal effect in salinity tolerance of Daphnia same stress factor and belonging to the same species, but genetically different or belonging to one genotype but exposed to different intensity of the stress factor. 2. Both longer (two-generational) and stronger (higher concentration) impacts of selective factors may be perceived by individuals as information indicating reduced chances of suc- cessful reproduction in the future and, thus, they may drive maternal strategies to produce first clutch descendants better-prepared to adverse conditions. Acknowledgments We are grateful to Joanna Pijanowska for her help at each stage of work on the manuscript and to Cleve Hicks for improving the language of the manuscript and valuable substantive com- ments. We also thank the anonymous reviewers for their comments that led to an improved manuscript. Author Contributions Conceptualization: Andrzej Mikulski. Data curation: Andrzej Mikulski. Formal analysis: Andrzej Mikulski, Danuta Mazurczak. Funding acquisition: Andrzej Mikulski. Investigation: Andrzej Mikulski, Danuta Mazurczak. Methodology: Andrzej Mikulski. Project administration: Andrzej Mikulski. Resources: Andrzej Mikulski. Software: Andrzej Mikulski. Supervision: Andrzej Mikulski. Validation: Andrzej Mikulski. Visualization: Andrzej Mikulski, Danuta Mazurczak. Writing – original draft: Andrzej Mikulski. Writing – review & editing: Andrzej Mikulski. References 1. Prizak R, Ezard THG, Hoyle RB. Fitness consequences of maternal and grandmaternal effects. Ecol Evol 2014; 4: 3139–3145. https://doi.org/10.1002/ece3.1150 PMID: 25247070 2. Inchausti P, Ginzburg LR. Maternal effects mechanism of population cycling: a formidable competitor to the traditional predator-prey view. Phil Trans Roy Soc B 2009; 364: 1117–1124. 3. Uller T. Developmental plasticity and the evolution of parental effects. Trends Ecol Evol 2008; 23: 432– 438. https://doi.org/10.1016/j.tree.2008.04.005 PMID: 18586350 4. Badyaev AV, Uller T. Parental effects in ecology and evolution: mechanisms, processes and implica- tions. Phil Trans Roy Soc B 2009; 364: 1169–1177. https://doi.org/10.1098/rstb.2008.0302 PMID: 19324619 5. Mosseau TA, Fox CW. Maternal effect as adaptation. Oxford University Press, New York, Oxford 1998. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283546 April 4, 2023 11 / 13 PLOS ONEMaternal effect in salinity tolerance of Daphnia 6. Fox CW, Thakar MS, Mousseau TA. Egg size plasticity in a seed beetle: An adaptive maternal effect. Am Nat 1997; 149: 149–163. 7. Agrawal AA, Laforsch C, Tollrian R. Transgenerational induction of defences in animals and plants. Nature 1999; 401: 60–63. 8. Agrawal AA. Transgenerational consequences of plant responses to herbivory: an adaptive maternal effect? Am Nat 2001; 157: 555–569. https://doi.org/10.1086/319932 PMID: 18707262 9. Agrawal AA. Herbivory and maternal effects: mechanisms and consequences of transgenerational induced plant resistance. Ecology 2002; 83: 3408–3415. 10. Galloway LF. Parental environmental effects on life history in the herbaceous plant Campanula ameri- cana. Ecology 2001; 82: 2781–2789. 11. Plaistow SJ, St Clair JJ, Grant J, Benton TG. How to put all your eggs in one basket, empirical patterns of offspring provisioning throughout mother’s lifetime. Am Nat 2007; 170: 520–529. 12. Keiser CN, Mondor EB. Transgenerational behavioral plasticity in a parthenogenetic insect in response to increased predation risk. J Insect Behav 2013; 26: 603–613. 13. Sheriff MJ, Love OP. Determining the adaptive potential of maternal stress. Ecol Lett 2013; 16: 271– 280. https://doi.org/10.1111/ele.12042 PMID: 23205937 14. Eriksson-Wiklund AK, Sundelin B. Impaired reproduction in the amphipods Monoporeia affinis and Pon- toporeia femorata as a result of moderate hypoxia and increased temperature MEPS 2001; 222: 131– 141. 15. Cunningham EJA, Russell AF. Egg investment is influenced by male attractiveness in the mallard. Nature 2000; 404: 74–77. https://doi.org/10.1038/35003565 PMID: 10716444 16. Uller T, Eklo¨f J, Andersson S. Female egg investment in relation to male sexual traits and the potential for transgenerational effects in sexual selection. Behav Ecol Sociobiol 2005; 57: 584–590. 17. Marshall DJ, Uller T. When is a maternal effect adaptive? Oikos 2007; 116: 1957–1963. 18. Sopinka NM, Hinch SG, Middleton CT, Hills JA, Patterson DA, Stibor H. Mother knows best, even when stressed? Effects of maternal exposure to a stressor on offspring performance at different life stages in a wild semelparous fish. Oecologia 2014; 175: 493–500. https://doi.org/10.1007/s00442-014-2915-9 PMID: 24619199 19. Lampert W. Daphnia: model herbivore, predator and prey. Pol J Ecol 2006; 54: 607–620. 20. Seda J, Petrusek A. Daphnia as a model organism in limnology and aquatic biology: introductory remarks. J Limnol 2011; 70: 37–344. 21. Rossiter MC. Incidence and consequences if inherited environmental effects. Ann Rev Ecol Evol Syst 1996; 27: 451–476. 22. Alekseev V, Lampert W Maternal control of resting-egg production in Daphnia. Nature 2001; 414: 899– 901. 23. La Montagne JM, Mc Cauley E. Maternal effects in Daphnia: what mothers are telling their offspring and do they listen? Ecol Lett 2001; 4: 64–71. 24. Mikulski A, Pijanowska J. Maternal experience can enhance production of resting eggs in Daphnia exposed to the risk of fish predation. Fund Appl Limnol 2009; 174: 301–305. 25. Xiaodong J, Wei Y, Shiye Z, Huishuang L, Yunlong Z, Liqiao C, et al. Maternal effects of inducible toler- ance against the toxic cyanobacterium Microcystis aeruginosa in the grazer Daphnia carinata. Envir- onm Poll 2013; 178: 142–146. 26. Garbutt JS, Scholefield JA, Vale PF, Little TJ. Elevated maternal temperature enhances offspring dis- ease resistance in Daphnia magna. Funct Ecol 2014; 28: 424–431. 27. Mikulski A, Pijanowska J. When and how can Daphnia prepare their offspring for the threat of predation? Hydrobiologia 2010; 643: 21–26. 28. Thabet R, Ayadi H, Koken M, Leignel V. Homeostatic responses of crustaceans to salinity changes. Hydrobiologia 2017; 799: 1–20. 29. Martınez-Jeronimo F, Martınez-Jeronimo L. Chronic effect of NaCl salinity on a freshwater strain of Daphnia magna Straus, Crustacea: Cladocera: A demographic study. Ecotox Environm Saf 2007; 67: 411–416. 30. Santos MAPF, Vincensotti J, Monteiro RTR. Sensitivity of Four Test Organisms, Chironomus xanthus, Daphnia magna, Hydra attenuata and Pseudokirchneriella subcapitata. to NaCl: an Alternative Refer- ence Toxicant. J Braz Soc Ecotoxicol 2007; 2: 229–236. 31. Baillieut M, Selens M, Blust R. Scope for growth and fitness of Daphnia magna in salinity—stressed conditions. Funct Ecol 1996; 10: 227–233. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283546 April 4, 2023 12 / 13 PLOS ONEMaternal effect in salinity tolerance of Daphnia 32. Casey R, Scrimgeour G., Kendall S Effects of water temperature and treated pulp mill effluent on sur- vival and growth of Daphnia magna, Cladocera: Daphnidae. and Taenionema, Plecoptera Taenioptery- gidae. Alberta Environment Sustainable Forest Management Research Program, Final report. Pub no: T/678 2000. 33. Gonc¸alves AMM, Castro BB, Pardal MA Gonc¸alves F. Salinity effects on survival and life history of two freshwater cladocerans, Daphnia magna and Daphnia longispina. Ann Limnol Int J Limnol 2007; 43: 13–20. 34. Arner M, Koivisto S. Effects of salinity on metabolism and life history characteristics of Daphnia magna. Hydrobiologia 259:69–77. 35. Teschner M (1995) Effects of salinity on the life history and fitness of Daphnia magna variability within and between populations. Hydrobiologia 1993; 307: 33–41. 36. Hall CJ, Burns CW. Mortality and growth response of Daphnia carinata to increases in temperature and salinity. Freshwat Biol 2002; 47: 451–458. 37. Grzesiuk M, Mikulski A. The effect of salinity on freshwater crustaceans. Pol J Ecol 2006; 54: 669–674. 38. Latta LC, Weider LJ, Colbourne JK, Pfrender ME, Hessen D. The evolution of salinity tolerance in Daph- nia: a functional genomics approach. Ecol Lett 2012; 5: 794–802. 39. Bezirci G, Akkas SB, Rinke K, Yildirim F, Kalaylioglu Z, Severcan F et al. Impacts of salinity and fish- exuded kairomone on the survival and macromolecular profile of Daphnia pulex. Ecotoxicology 2012; 21: 601–614. 40. Hall MD, Vettiger A, Ebert D. Interactions between environmental stressors: the influence of salinity on host-parasite interactions between Daphnia magna and Pasteuria ramosa. Oecologia 2013; 171: 789– 796. 41. De la Paz Go´mez-Dı´az M, Martı´nez-Jero´ nimo F. Modification of the acute toxic response of Daphnia magna Straus 1820 to Cr (VI) by the effect of varying saline concentrations, NaCl. Ecotoxicology. 2009; 18: 81–86. 42. 43. Liu X, Steiner CF. Ecotoxicology of salinity tolerance in Daphnia pulex: interactive effects of clonal varia- tion, salinity stress and predation. J. Plankt. Res. 2017; 39: 687–697. Lampert W and Trubetskova I. Juvenile growth rate as a measure of fitness in Daphnia. Funct Ecol 1996; 10: 631–635. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283546 April 4, 2023 13 / 13 PLOS ONE
10.1371_journal.pone.0283930	RESEARCH ARTICLE Coliphages of the human urinary microbiota Elias CrumID Genevieve JohnsonID 1,2☯, Zubia Merchant2☯, Adriana Ene1,2, Taylor Miller-Ensminger1, 1, Alan J. WolfeID 3, Catherine PutontiID 1,2,3* 1 Bioinformatics Program, Loyola University Chicago, Chicago, Illinois, United States of America, 2 Department of Biology, Loyola University Chicago, Chicago, Illinois, United States of America, 3 Department of Microbiology and Immunology, Stritch School of Medicine, Loyola University Chicago, Maywood, Illinois, United States of America ☯ These authors contributed equally to this work. * cputonti@luc.edu Abstract Due to its frequent association with urinary tract infections (UTIs), Escherichia coli is the best characterized constituent of the urinary microbiota (urobiome). However, uropatho- genic E. coli is just one member of the urobiome. In addition to bacterial constituents, the urobiome of both healthy and symptomatic individuals is home to a diverse population of bacterial viruses (bacteriophages). A prior investigation found that most bacterial species in the urobiome are lysogens, harboring one or more phages integrated into their genome (pro- phages). Many of these prophages are temperate phages, capable of entering the lytic cycle and thus lysing their bacterial host. This transition from the lysogenic to lytic life cycle can impact the bacterial diversity of the urobiome. While many phages that infect E. coli (coliphages) have been studied for decades in the laboratory setting, the coliphages within the urobiome have yet to be cataloged. Here, we investigated the diversity of urinary coli- phages by first identifying prophages in all publicly available urinary E. coli genomes. We detected 3,038 intact prophage sequences, representative of 1,542 unique phages. These phages include both novel species as well as species also found within the gut microbiota. Ten temperate phages were isolated from urinary E. coli strains included in our analysis, and we assessed their ability to infect and lyse urinary E. coli strains. We also included in these host range assays other urinary coliphages and laboratory coliphages. The temperate phages and other urinary coliphages were successful in lysing urinary E. coli strains. We also observed that coliphages from non-urinary sources were most efficient in killing urinary E. coli strains. The two phages, T2 and N4, were capable of lysing 83.5% (n = 86) of strains isolated from females with UTI symptoms. In conclusion, our study finds a diverse commu- nity of coliphages in the urobiome, many of which are predicted to be temperate phages, ten of which were confirmed here. Their ability to infect and lyse urinary E. coli strains suggests that urinary coliphages may play a role in modulating the E. coli strain diversity of the urobiome. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Crum E, Merchant Z, Ene A, Miller- Ensminger T, Johnson G, Wolfe AJ, et al. (2023) Coliphages of the human urinary microbiota. PLoS ONE 18(4): e0283930. https://doi.org/10.1371/ journal.pone.0283930 Editor: Mark Eppinger, University of Texas at San Antonio, UNITED STATES Received: January 5, 2023 Accepted: March 20, 2023 Published: April 13, 2023 Copyright: © 2023 Crum et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Funding: CP is supported by the US National Science Foundation (NSF) (award # 1661357). EC, ZM, and AE are supported by the Mulcahy Research Fellowship from Loyola University Chicago. AJW is supported by the National Institutes of Health (NIH) (award # R01- DK104718). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 1 / 16 PLOS ONECompeting interests: I have read the journal’s policy and the authors of this manuscript have the following competing interests: Author AJW discloses the following competing interests: Kimberly Clark Corporation - Investigator Initiated Program; Pathnostics - Scientific Advisory Board; Urobiome Therapeutics - Scientific Advisory Board; VBTech - Investigator Initiated Program. This does not alter our adherence to PLOS ONE policies on sharing data and materials. Coliphages of the Human Urinary Microbiota Introduction While several different bacterial species are known to cause urinary tract infections (UTIs), uropathogenic Escherichia coli (UPEC) is estimated to account for up to 74.4% of all commu- nity-acquired infections [1]. The presence of E. coli is often taken as evidence of UTI. While UTIs can be the result of colonization of the urinary tract by E. coli strains from the gut [2–5], E. coli can also be a member of the urinary microbiota (urobiome) of individuals without lower urinary tract symptoms (LUTS) [6–8]. In fact, recent studies found that E. coli can be the predominant taxon in female without LUTS, particularly in some older females [8, 9]. Fur- thermore, human host genetic make-up can contribute to the presence of E. coli in the urine of females without UTI symptoms [9]. The urobiome of individuals with or without LUTS is home to a wide variety of other bacte- rial taxa (see reviews [10–13]), fungi [14], and viruses [15]. In fact, viruses are the most abun- dant members of the urobiome and recently have been associated with LUTS [16]. Viruses that infect bacteria (bacteriophages or phages) far outnumber human viruses in the urobiome [15, 17, 18]. Similar observations have been made in other organs’ microbiota [19]. Phages can drive bacterial diversity within a community through predation [20–22]. Phages that integrate into their bacterial host’s genome (prophages) can increase the virulence of their host [23, 24], e.g., by encoding for toxins [25, 26]. While studies of phage-bacteria dynamics have yet to be conducted for the urobiome, investigations in the gut microbiota are ongoing [27]. Prior studies of urinary bacterial genomes found that most strains harbor one or more pro- phage sequences [28–30]. This includes bacterial species associated with urinary health, e.g., Lactobacillus species [28, 29], as well as bacterial species associated with UTIs, e.g., E. coli [7, 28] and Proteus mirabilis [28]. These phages replicate with their bacterial host and are either integrated in the bacterial host genome or persist as an extrachromosomal plasmid; this is referred to as the lysogenic life cycle of the phage. Our previous work with urinary strains har- boring prophages found that many of these prophages could switch from the lysogenic life cycle to the lytic (predatory) life cycle [28, 30–32]. This switch, a process called induction, is mediated by intrinsic and/or extrinsic factors that cause the prophage to excise itself from the bacterial genome, replicate to produce mature (lytic) phages, and burst or kill the host cell (see review [33]). Bladder-relevant stressors, e.g., changes in pH, have been shown to be effective in inducing prophages [31]. Prophages and phage genes have been routinely identified in urinary E. coli genome sequences [28, 34–37]. While phages and prophages that infect E. coli (coliphages) in the lab have been studied for decades [38] and coliphages from urine samples have been isolated and characterized [39–42], coliphage diversity within the urobiome has yet to be thoroughly inves- tigated. Here, we present the first catalog of urinary E. coli prophages. All publicly available uri- nary E. coli genomes were examined for prophage sequences and the diversity and genic content of these prophages was explored. Ten urinary prophages were then induced from uri- nary E. coli strains, and we assessed their ability to lyse laboratory and urinary E. coli strains isolated from females with UTI or overactive bladder (OAB) symptoms. Additionally, we tested several urinary coliphages and laboratory coliphages against these same set of urinary E. coli strains. We find that coliphages are ubiquitous in the urobiome, including temperate coli- phages capable of lysing other urinary E. coli strains. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 2 / 16 PLOS ONEColiphages of the Human Urinary Microbiota Methods Prophage identification All publicly available E. coli complete and draft genome assemblies in NCBI that were docu- mented as being collected from urine or the urinary tract in the genome metadata were down- loaded (February 2021). 906 genomes were obtained from NCBI meeting this criterion; 2 sequences were removed due to quality concerns. The final set of 904 E. coli sequences (S1 Table) were examined using PHASTER [43]. A Python script was written to pull results from the PHASTER API and to separate PHASTER predicted “intact,” “questionable,” and “incom- plete” prophage sequences. This script is available at https://github.com/putonti/phaster_ commands. All intact prophage sequences were compared via local blastn queries to a database of all complete and partial phage genome sequences in GenBank as of February 2021 (Advance Query Fields—Organism: “Virus” and Division: “PHG”). This database includes 26,381 sequences. Results with a query coverage greater than 50% and a percent identity greater than 70% were considered high confidence hits and the associated taxonomies of the GenBank phage records were used to predict the taxonomies of the predicted urinary phages. Furthermore, all intact prophage sequence were screened for antibiotic resistance genes using the RGI tool v.5.2.1, which uses CARD [44], exploring perfect, strict, and loose hits, and virulence factors via the Virulence Factor Database (VFDB) [45]. RGI was installed locally using conda, and the CARD database (v.3.1.4) was downloaded. The RGI program was run using default parameters. For virulence factor screening, the full dataset of known bacterial vir- ulence factor gene sequences was downloaded from VFDB on March 2021 (http://www.mgc. ac.cn/VFs/download.htm). These sequences were made into a local database via the make- blastdb command (BLAST+ v.2.9.0). Prophage sequences were queried against this database using BLASTn with the parameter -evalue 0.001. Hits of <90% sequence identity and query length <90% of the virulence factor gene sequence were removed from further consideration. The virulence factor gene descriptions were determined using VFDB. The PATRIC online tool was used to annotate the intact prophage sequences with the “Annotation Recipe” parameter set to “Bacteriophage” [46]. All annotated genes containing “integrase” in the description were added to a multi-FASTA file, and Kalign v.2.04 was used to produce a multiple sequence alignment with default parameters [47]. The multiple sequence alignment was manually inspected using Geneious Prime 2021.2.2 (Dotmatics, Auckland, NZ). Partial integrase gene sequences were identified through this manual inspection and were removed from the data set. The remaining integrase gene sequences were aligned again with Kalign. This alignment aided in confirming the identity of integrase genes within the pre- dicted prophage sequences. Prophage network construction Using Anvi’o v.6.2 [48], the intact prophage sequences were annotated, and a coliphage pan- genome was constructed. Prophage sequences were made into an Anvi’o database annotated using the ‘anvi-run-hmms’ command. The annotated prophage sequences were then used to produce a coliphage pan-genome using the Anvi’o ‘anvi-pan-genome’ command with an mcl- inflation of 2 and a minbit of 0.35 to identify homologous genes among the prophage sequences. An R script (www.R-project.org) was written to derive a network of prophages. The meth- ods described here for phage network construction were adapted from our prior work [49]. Using the output result (mcl-cluster.txt) of the Anvi’o mediated clustering using the Markov PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 3 / 16 PLOS ONEColiphages of the Human Urinary Microbiota Clustering Algorithm (MCL), the MCL results were translated into a genome-gene presence/ absence matrix, P, in which each entry {i,j} was 1 if virus genome i contained a homolog found in gene cluster j. This matrix is equivalent to the adjacency matrix for a bipartite network of phage genomes and genes. Adjacency matrices for the genome and gene level networks were then created as Agenome = sign(P × PT) and Agene = sign(PT × P), where T indicates the matrix transpose. The sign() function replaced all nonzero entries resulting from the original matrix products with a 1, converting the matrices from weighted to unweighted adjacency matrices. These matrices were then transformed into undirected graphs and corresponding edge lists using igraph (https://igraph.org/). Thus, for the genome-level network, two genomes are con- sidered connected if they share any genes. The connections were filtered using a normalization calculation: w ¼ ð# of shared genes between genomes 1 & 2Þ= of the genome i. By designating a minimum value of w (minw) that allows for an edge to be drawn between two genomes only if w > minw, the edges were filtered to construct networks of differing connectivity. , where li is the size ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Þ ð l1 � l2 p The edge-lists constructed from the edge-drawing Rscript were then visualized using Cytos- cape [50]. Different values of minw were considered for the PHASTER predicted prophages. Prophage sequence clustering All intact prophage sequences were clustered using cd-hit-est v.4.6 [51, 52]. The following parameters were used: sequence identity threshold = 80% (0.8), length of difference cut- off = 80% (0.8), word length = 4. The “accurate but slow mode” algorithm was used. For each cluster, cd-hit-est selected one sequence as a representative sequence (indicated by “*” in the output .clstr file). For more details regarding the process implemented by CD-HIT algorithms for selecting representative sequences, see Huang et al. [53]. Sequences of representatives of each cluster are available upon request. Comparison of urinary coliphages to gut phageome All phage sequences from metagenomic data sets from the gut microbiome were retrieved from the mMGE database [54] (accessed July 2021). Intact urinary prophage sequences were compared against this database using a local blastn query. For these results, our thresholds were query coverage � 50% and sequence identity � 70%, although all results meeting the query coverage threshold had a sequence identity � 80%. Prophage induction Nine strains of E. coli isolated from urine samples were used for induction experiments. The urinary strains were selected based upon their genome analysis by PHASTER [43]; all 9 were predicted to contain at least one intact prophage sequence. For pH-based induction, we used the previously described protocol [31]. Briefly, urinary E. coli strains were grown overnight in LB at 37˚C with shaking. These strains include E. coli UMB0527, UMB6653, UMB6721, UMB9006, UMB9105, UMB9208, UMB9344, UMB9346, and UMB9930. While UMB9006 was obtained from a “clean-catch” voided urine sample, the remaining 8 samples were isolated from catheterized urine samples. These strains were obtained through prior IRB-approved studies (Loyola University Chicago: LU204195 and LU209545, and University of California San Diego: 170077AW) as part of prior separate studies [55, 56]. The overnight culture was then subcultured into 3 mL of LB adjusted to different pH values, pH = 4, 7, and 9 and grown overnight at 37˚C with shaking. These pH values were selected informed by our prior work [31]. These pH-adjusted cultures were filtered using a 0.22um CA syringe filter. Filtrate was PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 4 / 16 PLOS ONEColiphages of the Human Urinary Microbiota next spotted onto lawns of naïve laboratory strains of E. coli B, E. coli C, and E. coli K-12. For each lawn, 500 uL of turbid (overnight) E. coli culture + 3 mL of soft (0.7%) LB agar were mixed and spread atop a 1.7% LB agar plate. The spot plates were then incubated overnight at 37˚C. Plaques were harvested and used to reinfect the laboratory strain, i.e., the bacteria of the lawn from which the plaque was harvested, and incubated again overnight at 37˚C. These cul- tures were filtered as previously described and plated using the pour plate technique (100 uL of phage lysate + 500 uL of turbid E. coli culture + 3 mL of soft LB agar); a single plaque was picked, suspended in LB, vortexed, filtered, and added to turbid E. coli overnight cultures. This process was repeated at least 3 times to plaque purify the phage and subsequently to create higher titer stocks of the induced urinary prophage. Prophage identification was performed by PCR. Primers were designed using Primer- BLAST [57] to amplify the PHASTER predicted intact prophage sequences. A primer pair was designed for each individual predicted intact prophage sequence; thus, a strain harboring mul- tiple predicted intact prophage sequences would have multiple pairs (S2 Table). Preference was given to primers that amplified coding regions with a predicted protein function, i.e., not hypothetical proteins. All predicted prophage sequences were annotated using the RAST server [58] and visualized using Geneious Prime. If more than one intact prophage was pre- dicted for a given strain, primers were designed for each predicted prophage. Primers were synthesized by Eurofins Genomics LLC (Louisville, KY USA). The original bacterial strain was used as a positive control for PCR reactions of its respective induced prophages. Amplification was confirmed via agarose gel. The PHASTER predicted sequences were queried online against the nr/nt viruses (taxid:10239) to identify the closest related sequenced phage. Phage host range Urinary and laboratory E. coli strains were grown in LB overnight at 37˚C with shaking. Each was lawned, following the protocol listed previously, and 10 uL of phage lysate (at titer 109 phage per mL) was spotted on each lawn. Each phage lysate was spotted onto each E. coli strain a minimum of 4 times (technical replicates). In addition to assessing the host range for the induced urinary phages, two other urinary coliphages previously isolated by our group [41] as well as non-urinary coliphages were spotted on urinary and laboratory E. coli strains. These coliphages include Escherichia phage Greed and Escherichia phage Lust, both isolated from urobiome samples, and non-urinary coliphages K30, P22, T2, T3, T6, T7 and N4 obtained from the Fe´lix d’He´relle Reference Center for Bacterial Viruses (Quebec City, Quebec Can- ada). Phages T2 and N4 were further tested on urinary E. coli strains collected from UTI-posi- tive females (S3 Table). Plaques of T2 and N4 spots were confirmed via PCR of the plaque. T2 primers: 5’-aaacaggtgcctttggtgtc-3’ and 5’-ccacaatacccgcttcagtt-3’; N4 primers: 5’-tgctcttgatac- cagaggcaatg-3’ and 5’-tacgttggttcaacttcttggtt-3’. Primers were synthesized by Eurofins Geno- mics LLC. For all phages producing plaques for the host range spot assays, infection was again tested by harvesting the plaque and replating using the pour plates technique previously described. Results Prophages in urinary E. coli genomes 904 urinary E. coli draft and complete genome sequences were obtained from NCBI and screened for the presence of prophage via PHASTER [43]. Sixty-nine of these genome assem- blies are of strains from our own collection. Of the 8,452 predicted prophage sequences found, 3,038 prophages were identified as intact, 1,508 as questionable, and 3,906 as incomplete. We focused our analysis on those identified as intact. 45 of the 904 genomes examined did not PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 5 / 16 PLOS ONEColiphages of the Human Urinary Microbiota contain any intact prophages, while the majority (95%) of the urinary E. coli genomes harbored at least one intact prophage (S1 Table). 1,807 (59.48%) of the intact prophage sequences con- tained an identifiable integrase gene sequence (S1 Table). Each intact prophage sequence was compared to all annotated phage genome sequences to predict their taxonomy (Table 1). Virulence factor genes were identified in 25% (n = 765) of the intact coliphage sequences. These virulence-carrying intact prophage sequences were harbored by 710 of the 904 E. coli genomes examined. The most frequently identified virulence factor gene was aaiQ (n = 137), a pseudogene that has been linked to enteroaggregative E. coli pathogenesis [59]. The iron uptake sit operon was the next most prevalent virulence factor identified in 61 of the intact prophage sequences. When screened for antibiotic resistance associated genes, 299 genes were identified. These genes were found in 178 different intact prophage sequences (5.86%) from 165 different E. coli genomes. The most frequently identified antibiotic resistance gene was the ethidium multidrug resistance protein E (emrE), found in 47 of the prophage sequences. The transcription factor marA was the second most frequent antibiotic resistance associated gene found; 37 intact prophage sequences encoded for marA. While most strains had just one or two antibiotic resistance genes, E. coli Combat11I9 (GCF_002952095.1) includes one prophage sequence with 14 antibiotic resistance genes. S1 Table lists information about the virulence factors and antibiotic resistance associated genes found within the prophage sequences. Intact prophages were annotated and the number of homologous genes between each pro- phage was calculated. Prophage similarity was assessed using a network approach in which nodes in the network represent a single prophage. Two nodes are connected by an edge in the network if they share a homologous gene sequence. We introduced a threshold for these edges such that only edges in which the two prophages (nodes) shared at least 30% of their genic con- tent were visualized (see Methods). 3,025 of the 3,038 predicted prophages met this threshold, connected by 414,291 edges. Fig 1 displays the network representation of the urinary prophages. Nodes, representative of prophages, have been colored in this network according to their predicted taxonomic family. Over half of the identified prophages most closely resembled tailed phages (order Caudovirales), including the families Myoviridae, Podoviridae, and Sipho- viridae. However, 43.9% of the predicted intact prophage sequences in the urinary E. coli genomes did not have significant similarity to previously isolated and sequenced phages. Our analysis of the shared genetic content of the predicted urinary prophages reveals nine con- nected components (Fig 1, labeled A-I). The individual connected components do not gener- ally share homologous genes (see Methods). While the largest connected component, labeled A in Fig 1, includes the majority of Caudovirales, our network contains two distinct connected components of siphoviruses (Fig 1D and 1G, orange); these are separate from the siphoviruses within the primary (largest) connected component. There also are six distinct connected Table 1. Taxonomic classification of predicted prophages in urinary E. coli based on blastn sequence similarity to sequenced phages. Taxonomy Myoviruses Podoviruses Siphoviruses Unclassified Caudovirales Unclassified bacterial viruses Unknown https://doi.org/10.1371/journal.pone.0283930.t001 # Predicted Prophages 788 113 738 44 20 1,335 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 6 / 16 PLOS ONEColiphages of the Human Urinary Microbiota Fig 1. Diversity of urinary E. coli prophages. Each node corresponds with a predicted intact coliphage. Edges connecting nodes represent shared gene content. Individual connected components, clusters of prophages that share gene content, are indicated by letters A through I. https://doi.org/10.1371/journal.pone.0283930.g001 components of prophages for which we were unable to predict their taxonomic lineage (shown in Fig 1B, 1C, 1E, 1F, 1H and 1I in blue for “Unknown”). Next, we investigated the network of prophages to identify clusters of similar prophage sequences, i.e., closely related prophages likely representative of the same species/strain. In total, 1,542 unique clusters of prophage sequences were identified (S4 and S5 Tables). The size of these clusters varied from singletons, i.e., single representative prophage sequences (n = 1,115, 72.31%) to a single cluster with 127 representatives (length ~43–50 Kbp; 87.73% nucleotide sequence identity between members). The consensus sequence of this large cluster was queried against the nr/nt Viruses database, identifying the top hit as a MAG sequence from the human metagenome (GenBank Accession No.: BK034715.1; 76% query coverage and 99.98% identity). Most of the clusters were small; 95% of the clusters had 5 or less prophage sequences. Four of the clusters had two prophage sequences identified from the same genome sequence. While two of these were the only instances of these prophage sequences in the data set, the other two were found in other genomes, including the cluster of 127 representatives. There is no association between cluster size and prophage sequence length (S5 Table). Sequences for the cluster representative are provided in S1 File. Comparison to gut E. coli prophages Given prior evidence of UTIs by colonization of E. coli from the gut, we investigated whether the same or similar phage populations were harbored by urinary E. coli strains and phages from the gut. The 3,038 intact urinary prophages were screened against the phage database of the gut microbiome. These gut microbiome phage sequences include phages identified from PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 7 / 16 PLOS ONEColiphages of the Human Urinary Microbiota metagenomic studies and, as such, include phages infectious of bacteria other than E. coli; fur- thermore, the host genus or species for most of these phage sequences is not known. As a result of this comparison, we found 2,006 urinary coliphages that exhibited � 50% query coverage and � 80% nucleotide sequence identity to phage sequences from the gut. This is representa- tive of 863 of the 1,542 unique clusters of urinary prophages, and 560 of these clusters only had one representative sequence from the urinary coliphages. Furthermore, 117 intact urinary pro- phages, representative of 49 unique clusters, had a query coverage of 100% to a gut phage sequences (� 96% sequence identity) and 21 were identical (100% query coverage and sequence identity) to gut phage sequences (S6 Table). Thirteen of these 21 prophage sequences shared no significant sequence similarity to an annotated phage genome sequence. Thus, no taxonomic classification could be assigned. These 21 prophages sequences were also screened for antibiotic resistance genes and virulence factors, although none were found. The 21 identi- cal prophages represent 12 unique clusters, including small clusters with just 2 members as well as the largest observed cluster with 127 representatives. Inducing urinary E. coli prophages Nine urinary E. coli strains in our collection were selected for induction assays. These strains were selected as they were predicted to contain prophage sequences from singleton clusters, i.e., the urinary strain was the only genome predicted to contain this prophage, as well as pro- phage sequences from larger clusters, including clusters with prophages from E. coli strains from collections other than our own. For those prophages in singleton clusters, induction would lay the groundwork for future characterization of this phage strain. For those prophages in larger clusters, induction would provide insight into the putative temperance of the phage across many different urinary E. coli strains. Using changes in pH, 10 induced prophages were identified as they were efficient in completely lysing one or more of the naïve laboratory strains tested–E. coli B, E. coli C, and E. coli K-12. We were able to identify the prophage that was induced via PCR (see Methods). Eight of the induced prophage sequences included recognizable integrase genes; i527 and i9930-2 did not. i6721 and i6653 were induced at all three pH conditions tested, pH = 4, 7 and Table 2. pH-induced prophages from urinary E. coli strains and their closest characterized and sequenced phage. Phages are named as i plus the E. coli strain number from which they were induced. As two phages were isolated from E. coli UMB9930, they are signified as “-1” and “-2”. Phage ID pH Condition (E. coli host) Predicted Product Amplified by PCR Description Closest Blast Hit Accession No. Query i527 i6653 4 (E. coli C) 4, 7 and 9 (E. coli C) DUF2560 family protein Enterobacteria phage CUS-3 CP000711.1 Phage terminase, endonuclease subunit GpM Bacteriophage L-413C AY251033.1 Coverage 60% 77% i6721 4, 7 and 9 (E. coli C) Phage terminase, endonuclease subunit GpM Enterobacteria phage fiAA91-ss NC_022750.1 75% i9006 4 (E. coli C) i9105 i9208 i9344 i9346 i9930-1 i9930-2 7 (E. coli C) 4 (E. coli C) 4 (E. coli C) 7 (E. coli C) 7 (E. coli C) 7 (E. coli K-12) Phage replication protein GpA, endonuclease Escherichia virus P2_4E6b NC_049389.1 79% Phage tail fiber protein GpH Bacteriophage L-413C Phage tail fiber protein GpH Phage major tail tube protein GpFII Phage lysis regulatory protein, LysA Phage baseplate assembly protein GpJ Phage head, portal protein B Escherichia phage pro147 Escherichia virus P2_4C9 Bacteriophage R18C Escherichia virus P2_4C9 Stx-1a-converting phage Stx1_499 AY251033.1 KR073660.1 NC_049388.1 NC_049461.1 NC_049388.1 LC567825.1 88% 77% 73% 68% 73% 96% https://doi.org/10.1371/journal.pone.0283930.t002 Sequence Identity 96.04% 97.80% 93.48% 96.52% 99.97% 96.87% 96.24 97.31% 96.24 94.88% PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 8 / 16 PLOS ONEColiphages of the Human Urinary Microbiota 9; the phages isolated from pH = 7 were used for subsequent testing. The closest characterized and sequenced phage record in GenBank was identified for each of the induced prophage sequences (Table 2). Upon further inspection, it was found that the induced prophage from UMB9344 (i9344) is identical–query coverage and sequence identity = 100%–to one of the two induced prophages from UMB9930, i9930-1. The induced prophages from UMB5563 (i6653) and UMB6721 (i6721) also were similar, belonging to the same cluster (S7 Table). A visualiza- tion of the induced prophage genomes is provided in S1 Fig. Phage host range Each induced urinary coliphage was tested for its ability to completely lyse laboratory strains of E. coli, as well as urinary E. coli strains representative of the diversity of E. coli phylotypes found within the female bladder [7]. In parallel, we tested two lytic urinary siphoviruses previ- ously isolated by our group [41], Escherichia phage Greed and Escherichia phage Lust, along- side several well-studied lytic coliphages routinely used in the laboratory—Escherichia phage K30, Enterobacteria T3 and Enterobacteria T7 (family Autographiviridae), Salmonella virus P22 (family Podoviridae), Enterobacteria phage T2 and T6 (family Myoviridae), and Escheri- chia virus N4 (family Schitoviridae). The results of these host range assays are shown in Table 3. The well-studied phages T2 and N4 were most efficient in completely lysing E. coli strains isolated from UTI patients. Each was capable of completely lysing eight of the ten urinary Table 3. Phage efficacy of lysing laboratory and urinary bacteria. If a phage completely lysed the E. coli strain, the table lists “+”. The phylotype for each urinary E. coli strain is listed [7]. E. coli UMB0103 was isolated from a female with OAB symptoms; the other urinary E. coli strains were isolated from females with UTI symptoms. LABORATORY E. COLI STRAINS A B1 B2 D F URINARY E. COLI STRAINS B C K-12 1358 1180 1195 5924 1220 1162 1225 1337 7431 103 Induced Phages i527 i6653 i6721 i9006 i9105 i9208 i9344 i9346 i9930-1 i9930-2 Urinary Phages Lust Greed Laboratory Phages K30 P22 T2 T3 T6 T7 N4 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + https://doi.org/10.1371/journal.pone.0283930.t003 + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 9 / 16 PLOS ONEColiphages of the Human Urinary Microbiota strains tested. Given this success, we further tested T2 and N4 against an additional 103 E. coli strains isolated from urine samples from females with UTI symptoms. A list of the strains tested, including serotype and phylotype, are listed in S3 Table. Sixty-three strains were completely lysed by T2, 71 strains were completely lysed by N4, 48 strains were completely lysed by both T2 and N4, and 17 strains were not completely lysed by either phage (Table 4). Thus, 83.5% of the E. coli strains tested from females with UTI symptoms would be susceptible to a phage cocktail containing T2 and N4. Discussion While our prior catalog of prophages in urinary bacteria considered just four E. coli strains [28], here we have evaluated the incidence of lysogeny among a significantly larger (n = 906) sampling of urinary E. coli strains. While all four of the genomes examined in our prior study were found to harbor prophage sequences, only one was predicted to include an intact pro- phage [28]. Here, the majority (95%) of the urinary E. coli genomes harbored at least one intact prophage sequence indicating that lysogeny is prevalent among E. coli strains of the urobiome. While some (~25%) of these prophage sequences carried virulence factor genes, very few encoded antibiotic resistance genes (S1 Table). One of the lysogen-associated proteins, the integrase, was identified in many (59.48%) of these intact prophage sequences. Those lacking the integrase necessitate further investigation to ascertain if they are temperate phages or pro- phage relics. We found that only 56.1% of the predicted intact prophage sequences resembled character- ized phage sequences. The remaining prophage sequences either represent novel prophages or highly mosaic prophages infectious of E. coli. This concurs with our prior examination of genomes of other taxa from the urobiome, which found a high percentage of unknown, novel prophages [28]. It is important to note that the intact prophages examined here do not include any representatives of inoviruses. Current prophage prediction tools–including PHASTER– frequently fail to identify inovirus sequences [60–62]. Previously, we identified an inovirus harbored by Enterobacteriaceae including urinary E. coli strains [63]. Inovirus sequences could be included in the questionable and/or incomplete prophage sequence predictions excluded from our analysis. We thus hypothesize that the intact prophage sequences examined here underestimates the genetic diversity of prophages within urinary E. coli strains. It does, however, provides a good approximation of the diversity of tailed phages infectious of urinary E. coli. Our network-based analysis of the shared genetic content between the predicted urinary prophages identified nine connected components (Fig 1). The largest connected component contains the majority of the tailed phages identified. From the network, we can posit taxo- nomic classification of “Unknown” (Fig 1, blue) prophages within this connected component as members of Caudovirales. It is important to note that our network analysis includes a threshold for the minimum percentage of gene content shared for an edge to be drawn. Thus, edges represent “modules” shared between prophages. When this threshold is removed, such Table 4. Susceptibility of E. coli strains from females with UTI symptoms to bacteriophages T2 and N4. Both T2 & N4 N4 only T2 only Neither # of E. coli lysed 48 23 15 17 % of E. coli lysed 46.6% 22.3% 14.6% 16.5% https://doi.org/10.1371/journal.pone.0283930.t004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 10 / 16 PLOS ONEColiphages of the Human Urinary Microbiota that every homologous gene shared between two prophages (nodes) is represented by an edge, all prophages are connected producing a single connected component. Previous analyses of phage genomes have taken a similar modular approach [64, 65]. In this prior work, modules in temperate phages were found to correspond to functional modules [64]. These modules often serve as markers corresponding to the evolutionary history of related phages [66]. Of the 1,542 unique clusters of urinary prophage sequences identified, many (55.97%, n = 863) shared sequence similarity to phages identified in the gut microbiota suggesting that similar coliphages infect E. coli strains found in both niches. The interconnectedness of the urobiome and gut microbiota is an open question. Prior studies have shown that E. coli causing UTIs can come from the gastrointestinal tract [2–5]. However, a previous examination of the bacterial constituents of the gut and urinary tract did not find the two niches to be connected [67]. To date, the interconnectedness of the phage populations between these two microbiota have not been examined. The similarities observed here serve as the impetus for future studies. Our ability to induce prophages from the urinary E. coli strains supports the working hypothesis that the PHASTER predictions of intact prophages are temperate phages, rather than prophage relics. Changes in pH were able to stress the bacterial cell and trigger the induc- tion process. Changes in pH are particularly relevant to the urinary tract and its microbiota. Lactobacillus species are dominant members of the healthy female bladder [8], and lactobacilli reduce the pH of the urogenital environment (see review [68]). Thus, fluctuations in lactoba- cilli abundances within the urinary tract could lead to changes in pH, which in turn could trig- ger induction. While nine out of the ten induced urinary prophages completely lyse the naïve laboratory strain E. coli C, they have varied abilities to completely lyse urinary E. coli strains. Two main observations can be made from our host range assays (Table 3). First, no single induced urinary phage was capable of completely lysing all of the urinary E. coli strains tested. Second, induced urinary phages capable of completely lysing one representative of a phylotype were not necessarily able to completely lyse all of the strains tested for that same phylotype. While only a single strain from phylotype A, B1, and F were tested, multiple representatives of phylotypes B2 and D, the most common phylotypes in the bladder [7], were tested. It is well documented that phages often are infectious of some but not all strains of a given species (see review [69]). The observed lack of plaquing by these induced urinary phages could be the result of (1) the phage’s inability to infect the E. coli strain, (2) the phage’s inability to evade the host’s defenses (e.g., CRISPR/Cas system), (3) the phage’s inability to completely lyse the bacte- rial cell, (4) entrance of the temperate phage back into the lysogenic life cycle, and/or (5) bacte- rial resistance due to superinfection. In contrast, the lytic laboratory coliphages were far more successful in completely lysing the urinary E. coli strains. They also were far more successful than the lytic urinary phages Lust and Greed. T2 and N4 were able to complete lyse at least one strain from all five phylotypes tested. The susceptibility of the urinary E. coli strains to these lytic laboratory coliphages may be due to the very fact that they are less likely to encounter these phages in the urinary tract than they are to encounter the induced urinary prophages. This suggests that phages that are not native to the urinary tract would be better candidates for use as phage therapies. Phage therapy is increasingly being explored for treatment of bacterial infections including E. coli and UTIs (see reviews [70–72]). Our further tests of T2 and N4 phages found that they were able to completely lyse most UTI-associated E. coli strains tested (Table 4), and thus should be explored for UTI treatments. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 11 / 16 PLOS ONEColiphages of the Human Urinary Microbiota Supporting information S1 Table. Urinary E. coli genomes retrieved from NCBI and screened for prophage sequences. (XLSX) S2 Table. PCR primers used to identify induced prophages. (XLSX) S3 Table. Urinary E. coli strains tested for susceptibility to T2 and N4. S3 Strains were either isolated from an individual with an acute urinary tract infection (UTI) or recurrent UTI (rUTI). (XLSX) S4 Table. Prophage clusters identified in each urinary E. coli genomes. (XLSX) S5 Table. Cluster sizes and length of the representative sequence for the cluster. (XLSX) S6 Table. Urinary prophages identical to gut phage sequences. (XLSX) S7 Table. Cluster information for induced prophages. (XLSX) S1 Fig. Annotations of induced prophages. (PDF) S1 File. Nucleotide sequences of the cluster representatives. (FASTA) Acknowledgments The authors would like to thank Dr. Jason Shapiro for his comments regarding analysis of the prophage sequences. We acknowledge the Loyola Urinary Education and Research Collabora- tive (LUEREC) clinicians for prior participant recruitment and the participants who provided the samples used in this study. Author Contributions Conceptualization: Catherine Putonti. Data curation: Elias Crum. Formal analysis: Elias Crum, Zubia Merchant, Adriana Ene, Taylor Miller-Ensminger. Investigation: Elias Crum, Zubia Merchant, Genevieve Johnson. Resources: Alan J. Wolfe, Catherine Putonti. Supervision: Alan J. Wolfe, Catherine Putonti. Writing – original draft: Elias Crum. Writing – review & editing: Zubia Merchant, Adriana Ene, Taylor Miller-Ensminger, Genevieve Johnson, Alan J. Wolfe, Catherine Putonti. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 12 / 16 PLOS ONEColiphages of the Human Urinary Microbiota References 1. Foxman B. Urinary tract infection syndromes: occurrence, recurrence, bacteriology, risk factors, and disease burden. Infect Dis Clin North Am. 2014; 28: 1–13. https://doi.org/10.1016/j.idc.2013.09.003 PMID: 24484571 2. Yamamoto S, Tsukamoto T, Terai A, Kurazono H, Takeda Y, Yoshida O. Genetic evidence supporting the fecal-perineal-urethral hypothesis in cystitis caused by Escherichia coli. J Urol. 1997; 157: 1127– 1129. PMID: 9072556 3. Hooton TM. Recurrent urinary tract infection in women. Int J Antimicrob Agents. 2001; 17: 259–268. https://doi.org/10.1016/s0924-8579(00)00350-2 PMID: 11295405 4. Chen SL, Wu M, Henderson JP, Hooton TM, Hibbing ME, Hultgren SJ, et al. Genomic diversity and fit- ness of E. coli strains recovered from the intestinal and urinary tracts of women with recurrent urinary tract infection. Sci Transl Med. 2013; 5: 184ra60. https://doi.org/10.1126/scitranslmed.3005497 PMID: 23658245 5. 6. Tha¨ nert R, Reske KA, Hink T, Wallace MA, Wang B, Schwartz DJ, et al. Comparative Genomics of Anti- biotic-Resistant Uropathogens Implicates Three Routes for Recurrence of Urinary Tract Infections. mBio. 2019;10. https://doi.org/10.1128/mBio.01977-19 PMID: 31455657 Thomas-White KJ, Gao X, Lin H, Fok CS, Ghanayem K, Mueller ER, et al. Urinary microbes and post- operative urinary tract infection risk in urogynecologic surgical patients. Int Urogynecol J. 2018; 29: 1797–1805. https://doi.org/10.1007/s00192-018-3767-3 PMID: 30267143 7. Garretto A, Miller-Ensminger T, Ene A, Merchant Z, Shah A, Gerodias A, et al. Genomic Survey of E. coli From the Bladders of Women With and Without Lower Urinary Tract Symptoms. Frontiers in Micro- biology. 2020;11. https://doi.org/10.3389/fmicb.2020.02094 PMID: 33013764 8. Price TK, Hilt EE, Thomas-White K, Mueller ER, Wolfe AJ, Brubaker L. The urobiome of continent adult women: a cross-sectional study. BJOG. 2020; 127: 193–201. https://doi.org/10.1111/1471-0528.15920 PMID: 31469215 9. Adebayo AS, Ackermann G, Bowyer RCE, Wells PM, Humphreys G, Knight R, et al. The Urinary Tract Microbiome in Older Women Exhibits Host Genetic and Environmental Influences. Cell Host Microbe. 2020; 28: 298–305.e3. https://doi.org/10.1016/j.chom.2020.06.022 PMID: 32697939 10. Whiteside SA, Razvi H, Dave S, Reid G, Burton JP. The microbiome of the urinary tract—a role beyond infection. Nature Reviews Urology. 2015; 12: 81–90. https://doi.org/10.1038/nrurol.2014.361 PMID: 25600098 11. Brubaker L, Wolfe A. The urinary microbiota: a paradigm shift for bladder disorders? Curr Opin Obstet Gynecol. 2016; 28: 407–412. https://doi.org/10.1097/GCO.0000000000000298 PMID: 27379439 12. Mueller ER, Wolfe AJ, Brubaker L. Female urinary microbiota. Curr Opin Urol. 2017; 27: 282–286. https://doi.org/10.1097/MOU.0000000000000396 PMID: 28234750 13. Neugent ML, Hulyalkar NV, Nguyen VH, Zimmern PE, De Nisco NJ. Advances in Understanding the Human Urinary Microbiome and Its Potential Role in Urinary Tract Infection. mBio. 2020;11. https://doi. org/10.1128/mBio.00218-20 PMID: 32345639 14. Ackerman AL, Underhill DM. The mycobiome of the human urinary tract: potential roles for fungi in urol- ogy. Annals of Translational Medicine. 2017; 5: 31–31. https://doi.org/10.21037/atm.2016.12.69 PMID: 28217696 15. Garretto A, Miller-Ensminger T, Wolfe AJ, Putonti C. Bacteriophages of the lower urinary tract. Nat Rev Urol. 2019; 16: 422–432. https://doi.org/10.1038/s41585-019-0192-4 PMID: 31073244 16. Sun Q, Li L, Zhou H, Wu Y, Gao Y, Wu B, et al. The detection of urinary viruses is associated with aggravated symptoms and altered bacteriome in female with overactive bladder. Front Microbiol. 2022; 13: 984234. https://doi.org/10.3389/fmicb.2022.984234 PMID: 36212847 17. Santiago-Rodriguez TM, Ly M, Bonilla N, Pride DT. The human urine virome in association with urinary tract infections. Front Microbiol. 2015; 6: 14. https://doi.org/10.3389/fmicb.2015.00014 PMID: 25667584 18. Garretto A, Thomas-White K, Wolfe AJ, Putonti C. Detecting viral genomes in the female urinary micro- biome. J Gen Virol. 2018; 99: 1141–1146. https://doi.org/10.1099/jgv.0.001097 PMID: 29889019 19. Barr JJ. A bacteriophages journey through the human body. Immunological Reviews. 2017; 279: 106– 122. https://doi.org/10.1111/imr.12565 PMID: 28856733 20. Buckling A, Rainey PB. Antagonistic coevolution between a bacterium and a bacteriophage. Proc Biol Sci. 2002; 269: 931–936. https://doi.org/10.1098/rspb.2001.1945 PMID: 12028776 21. Rodriguez-Valera F, Martin-Cuadrado A-B, Rodriguez-Brito B, Pasić L, Thingstad TF, Rohwer F, et al. Explaining microbial population genomics through phage predation. Nat Rev Microbiol. 2009; 7: 828– 836. https://doi.org/10.1038/nrmicro2235 PMID: 19834481 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 13 / 16 PLOS ONEColiphages of the Human Urinary Microbiota 22. Koskella B, Brockhurst MA. Bacteria-phage coevolution as a driver of ecological and evolutionary pro- cesses in microbial communities. FEMS Microbiol Rev. 2014; 38: 916–931. https://doi.org/10.1111/ 1574-6976.12072 PMID: 24617569 23. Hosseinidoust Z, van de Ven TGM, Tufenkji N. Evolution of Pseudomonas aeruginosa virulence as a result of phage predation. Appl Environ Microbiol. 2013; 79: 6110–6116. https://doi.org/10.1128/AEM. 01421-13 PMID: 23892756 24. Wendling CC, Piecyk A, Refardt D, Chibani C, Hertel R, Liesegang H, et al. Tripartite species interac- tion: eukaryotic hosts suffer more from phage susceptible than from phage resistant bacteria. BMC Evol Biol. 2017; 17: 98. https://doi.org/10.1186/s12862-017-0930-2 PMID: 28399796 25. Wagner PL, Waldor MK. Bacteriophage control of bacterial virulence. Infect Immun. 2002; 70: 3985– 3993. https://doi.org/10.1128/IAI.70.8.3985-3993.2002 PMID: 12117903 26. Casas V, Maloy S. Role of bacteriophage-encoded exotoxins in the evolution of bacterial pathogens. Future Microbiol. 2011; 6: 1461–1473. https://doi.org/10.2217/fmb.11.124 PMID: 22122442 27. De Sordi L, Lourenc¸o M, Debarbieux L. The Battle Within: Interactions of Bacteriophages and Bacteria in the Gastrointestinal Tract. Cell Host Microbe. 2019; 25: 210–218. https://doi.org/10.1016/j.chom. 2019.01.018 PMID: 30763535 28. Miller-Ensminger T, Garretto A, Brenner J, Thomas-White K, Zambom A, Wolfe AJ, et al. Bacterio- phages of the Urinary Microbiome. Zhulin IB, editor. Journal of Bacteriology. 2018; 200: e00738–17. https://doi.org/10.1128/JB.00738-17 PMID: 29378882 29. Miller-Ensminger T, Mormando R, Maskeri L, Shapiro JW, Wolfe AJ, Putonti C. Introducing Lu-1, a Novel Lactobacillus jensenii Phage Abundant in the Urogenital Tract. PLoS One. 2020; 15: e0234159. https://doi.org/10.1371/journal.pone.0234159 PMID: 32525961 30. Brassil B, Mores CR, Wolfe AJ, Putonti C. Characterization and spontaneous induction of urinary tract Streptococcus anginosus prophages. J Gen Virol. 2020; 101: 685–691. https://doi.org/10.1099/jgv.0. 001407 PMID: 32310742 31. Miller-Ensminger T, Garretto A, Stark N, Putonti C. Mimicking prophage induction in the body: induction in the lab with pH gradients. PeerJ. 2020; 8: e9718. https://doi.org/10.7717/peerj.9718 PMID: 32944418 32. Johnson G, Putonti C. Genome Sequence of Pseudomonas Phage UMP151, Isolated from the Female Bladder Microbiota. Microbiol Resour Announc. 2019;8. https://doi.org/10.1128/MRA.00853-19 PMID: 31416880 33. Nanda AM, Thormann K, Frunzke J. Impact of spontaneous prophage induction on the fitness of bacte- rial populations and host-microbe interactions. J Bacteriol. 2015; 197: 410–419. https://doi.org/10.1128/ JB.02230-14 PMID: 25404701 34. Beatson SA, Ben Zakour NL, Totsika M, Forde BM, Watts RE, Mabbett AN, et al. Molecular analysis of asymptomatic bacteriuria Escherichia coli strain VR50 reveals adaptation to the urinary tract by gene acquisition. Infect Immun. 2015; 83: 1749–1764. https://doi.org/10.1128/IAI.02810-14 PMID: 25667270 35. Price TK, Mehrtash A, Kalesinskas L, Malki K, Hilt EE, Putonti C, et al. Genome sequences and annota- tion of two urinary isolates of E. coli. Stand Genomic Sci. 2016; 11: 79. https://doi.org/10.1186/s40793- 016-0202-6 PMID: 27777649 36. Subhadra B, Kim DH, Kim J, Woo K, Sohn KM, Kim H-J, et al. Complete genome sequence of uropatho- genic Escherichia coli isolate UPEC 26–1. Genes Genomics. 2018; 40: 643–655. https://doi.org/10. 1007/s13258-018-0665-5 PMID: 29892948 37. Garretto A. Exploring associations between lysogeny and host abundance. MS, Loyola University Chi- cago. 2019. 38. Brockhurst MA, Koskella B. Experimental coevolution of species interactions. Trends Ecol Evol. 2013; 28: 367–375. https://doi.org/10.1016/j.tree.2013.02.009 PMID: 23523051 39. Dallas SD, Kingsbery L. Bacteriophage Plaques on Primary Isolation Media of a Urine Culture Growing Escherichia coli. Clinical Microbiology Newsletter. 1997; 19: 53–56. https://doi.org/10.1016/S0196- 4399(97)87667-2 40. Brown-Jaque M, Muniesa M, Navarro F. Bacteriophages in clinical samples can interfere with microbio- logical diagnostic tools. Scientific Reports. 2016;6. https://doi.org/10.1038/srep33000 PMID: 27609086 41. Malki K, Sible E, Cooper A, Garretto A, Bruder K, Watkins SC, et al. Seven Bacteriophages Isolated from the Female Urinary Microbiota. Genome Announc. 2016;4. https://doi.org/10.1128/genomeA. 01003-16 PMID: 27881533 42. Pacı´fico C, Hilbert M, Sofka D, Dinhopl N, Pap I-J, Aspo¨ ck C, et al. Natural Occurrence of Escherichia coli-Infecting Bacteriophages in Clinical Samples. Front Microbiol. 2019; 10: 2484. https://doi.org/10. 3389/fmicb.2019.02484 PMID: 31736918 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 14 / 16 PLOS ONEColiphages of the Human Urinary Microbiota 43. Arndt D, Grant JR, Marcu A, Sajed T, Pon A, Liang Y, et al. PHASTER: a better, faster version of the PHAST phage search tool. Nucleic Acids Res. 2016; 44: W16–21. https://doi.org/10.1093/nar/gkw387 PMID: 27141966 44. Alcock BP, Huynh W, Chalil R, Smith KW, Raphenya AR, Wlodarski MA, et al. CARD 2023: expanded curation, support for machine learning, and resistome prediction at the Comprehensive Antibiotic Resis- tance Database. Nucleic Acids Res. 2023; 51: D690–D699. https://doi.org/10.1093/nar/gkac920 PMID: 36263822 45. Liu B, Zheng D, Jin Q, Chen L, Yang J. VFDB 2019: a comparative pathogenomic platform with an inter- active web interface. Nucleic Acids Res. 2019; 47: D687–D692. https://doi.org/10.1093/nar/gky1080 PMID: 30395255 46. Davis JJ, Wattam AR, Aziz RK, Brettin T, Butler R, Butler RM, et al. The PATRIC Bioinformatics Resource Center: expanding data and analysis capabilities. Nucleic Acids Research. 2019; gkz943. https://doi.org/10.1093/nar/gkz943 PMID: 31667520 47. Lassmann T, Sonnhammer EL. Kalign–an accurate and fast multiple sequence alignment algorithm. BMC Bioinformatics. 2005; 6: 298. https://doi.org/10.1186/1471-2105-6-298 PMID: 16343337 48. Eren AM, Esen O¨ C, Quince C, Vineis JH, Morrison HG, Sogin ML, et al. Anvi’o: an advanced analysis and visualization platform for ‘omics data. PeerJ. 2015; 3: e1319. https://doi.org/10.7717/peerj.1319 PMID: 26500826 49. Shapiro JW, Putonti C. Gene Co-occurrence Networks Reflect Bacteriophage Ecology and Evolution. MBio. 2018;9. https://doi.org/10.1128/mBio.01870-17 PMID: 29559574 50. Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, et al. Cytoscape: a software environ- ment for integrated models of biomolecular interaction networks. Genome Res. 2003; 13: 2498–2504. https://doi.org/10.1101/gr.1239303 PMID: 14597658 51. 52. Li W, Godzik A. Cd-hit: a fast program for clustering and comparing large sets of protein or nucleotide sequences. Bioinformatics. 2006; 22: 1658–1659. https://doi.org/10.1093/bioinformatics/btl158 PMID: 16731699 Fu L, Niu B, Zhu Z, Wu S, Li W. CD-HIT: accelerated for clustering the next-generation sequencing data. Bioinformatics. 2012; 28: 3150–3152. https://doi.org/10.1093/bioinformatics/bts565 PMID: 23060610 53. Huang Y, Niu B, Gao Y, Fu L, Li W. CD-HIT Suite: a web server for clustering and comparing biological sequences. Bioinformatics. 2010; 26: 680–682. https://doi.org/10.1093/bioinformatics/btq003 PMID: 20053844 54. Lai S, Jia L, Subramanian B, Pan S, Zhang J, Dong Y, et al. mMGE: a database for human metage- nomic extrachromosomal mobile genetic elements. Nucleic Acids Research. 2021; 49: D783–D791. https://doi.org/10.1093/nar/gkaa869 PMID: 33074335 55. Barnes HC, Wolff B, Abdul-Rahim O, Harrington A, Hilt EE, Price TK, et al. A Randomized Clinical Trial of Standard versus Expanded Cultures to Diagnose Urinary Tract Infections in Women. Journal of Urol- ogy. 2021 [cited 17 Aug 2021]. https://doi.org/10.1097/JU.0000000000001949 PMID: 34184930 56. Hochstedler BR, Burnett L, Price TK, Jung C, Wolfe AJ, Brubaker L. Urinary microbiota of women with recurrent urinary tract infection: collection and culture methods. Int Urogynecol J. 2021 [cited 17 Aug 2021]. https://doi.org/10.1007/s00192-021-04780-4 PMID: 33852041 57. Ye J, Coulouris G, Zaretskaya I, Cutcutache I, Rozen S, Madden TL. Primer-BLAST: a tool to design target-specific primers for polymerase chain reaction. BMC Bioinformatics. 2012; 13: 134. https://doi. org/10.1186/1471-2105-13-134 PMID: 22708584 58. Aziz RK, Bartels D, Best AA, DeJongh M, Disz T, Edwards RA, et al. The RAST Server: rapid annota- tions using subsystems technology. BMC Genomics. 2008; 9: 75. https://doi.org/10.1186/1471-2164-9- 75 PMID: 18261238 59. Dudley EG, Thomson NR, Parkhill J, Morin NP, Nataro JP. Proteomic and microarray characterization of the AggR regulon identifies a pheU pathogenicity island in enteroaggregative Escherichia coli. Mol Microbiol. 2006; 61: 1267–1282. https://doi.org/10.1111/j.1365-2958.2006.05281.x PMID: 16925558 60. Roux S, Hallam SJ, Woyke T, Sullivan MB. Viral dark matter and virus–host interactions resolved from publicly available microbial genomes. eLife. 2015;4. https://doi.org/10.7554/eLife.08490 PMID: 26200428 61. Paez-Espino D, Eloe-Fadrosh EA, Pavlopoulos GA, Thomas AD, Huntemann M, Mikhailova N, et al. Uncovering Earth’s virome. Nature. 2016; 536: 425–430. https://doi.org/10.1038/nature19094 PMID: 27533034 62. Paez-Espino D, Chen I-MA, Palaniappan K, Ratner A, Chu K, Szeto E, et al. IMG/VR: a database of cul- tured and uncultured DNA Viruses and retroviruses. Nucleic Acids Res. 2017; 45: D457–D465. https:// doi.org/10.1093/nar/gkw1030 PMID: 27799466 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 15 / 16 PLOS ONEColiphages of the Human Urinary Microbiota 63. Shapiro JW, Putonti C. UPΦ phages, a new group of filamentous phages found in several members of Enterobacteriales. Virus Evolution. 2020. 64. 65. Lima-Mendez G, Toussaint A, Leplae R. A modular view of the bacteriophage genomic space: identifi- cation of host and lifestyle marker modules. Research in Microbiology. 2011; 162: 737–746. https://doi. org/10.1016/j.resmic.2011.06.006 PMID: 21767638 Iranzo J, Krupovic M, Koonin EV. The Double-Stranded DNA Virosphere as a Modular Hierarchical Net- work of Gene Sharing. mBio. 2016; 7: e00978–16. https://doi.org/10.1128/mBio.00978-16 PMID: 27486193 66. Koonin EV, Dolja VV, Krupovic M, Varsani A, Wolf YI, Yutin N, et al. Global Organization and Proposed Megataxonomy of the Virus World. Microbiol Mol Biol Rev. 2020; 84: e00061–19. https://doi.org/10. 1128/MMBR.00061-19 PMID: 32132243 67. Thomas-White K, Forster SC, Kumar N, Van Kuiken M, Putonti C, Stares MD, et al. Culturing of female bladder bacteria reveals an interconnected urogenital microbiota. Nat Commun. 2018; 9: 1557. https:// doi.org/10.1038/s41467-018-03968-5 PMID: 29674608 68. Stapleton AE. The Vaginal Microbiota and Urinary Tract Infection. Microbiol Spectr. 2016;4. https://doi. org/10.1128/microbiolspec.UTI-0025-2016 PMID: 28087949 69. de Jonge PA, Nobrega FL, Brouns SJJ, Dutilh BE. Molecular and Evolutionary Determinants of Bacteri- ophage Host Range. Trends Microbiol. 2019; 27: 51–63. https://doi.org/10.1016/j.tim.2018.08.006 PMID: 30181062 70. Bolocan AS, Callanan J, Forde A, Ross P, Hill C. Phage therapy targeting Escherichia coli-a story with no end? FEMS Microbiol Lett. 2016;363: fnw256. https://doi.org/10.1093/femsle/fnw256 PMID: 27974392 71. Zalewska-Piątek BM, Piątek RJ. Alternative treatment approaches of urinary tract infections caused by uropathogenic Escherichia coli strains. Acta Biochim Pol. 2019; 66: 129–138. https://doi.org/10.18388/ abp.2018_2787 PMID: 31136644 72. Chegini Z, Khoshbayan A, Vesal S, Moradabadi A, Hashemi A, Shariati A. Bacteriophage therapy for inhibition of multi drug-resistant uropathogenic bacteria: a narrative review. Ann Clin Microbiol Antimi- crob. 2021; 20: 30. https://doi.org/10.1186/s12941-021-00433-y PMID: 33902597 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283930 April 13, 2023 16 / 16 PLOS ONE
10.1371_journal.pone.0283603	RESEARCH ARTICLE When crisis hits: Bike-Sharing platforms amid the Covid-19 pandemic Ecem BasakID 1, Ramah Al Balawi1, Sorouralsadat Fatemi2, Ali Tafti2 1 Zicklin School of Business, Baruch College, City University of New York, New York, New York, United States of America, 2 College of Business Administration, University of Illinois at Chicago, Chicago, Illinois, United States of America ecem.basak@baruch.cuny.edu Abstract In this work, we examine the changes in demand for bike-sharing platforms with the onset of the Covid-19 pandemic. Using the fixed-effects regression formulation of difference-in-differ- ences, we evaluate how the demand for bike-sharing platforms changed after the first cases of Covid were discovered and after the first executive orders were implemented. Accounting for weather conditions, socio-economic characteristics, time trends, and fixed effects across cities, our findings indicate that there is an increase in daily bike-sharing trips by 22% on average after the first Covid-19 case diagnosis, and a decrease of 30% after the first execu- tive order implementation in each municipality, using the data up to August 2020. Moreover, we observe a 22% increase in weekday-specific trip frequency after the first Covid-19 case diagnosis and a 28% decrease in weekend-specific trip frequency after the first executive order implementation. Finally, we find that there is an increase in the frequency of trips on bike-sharing platforms in more bike-friendly, transit-friendly, and pedestrian-friendly cities upon both the first Covid-19 case diagnosis and the first executive order implementation. Introduction The Covid-19 pandemic has significantly affected our societal and economic structures. Man- dated lockdowns and voluntary precautions, which are taken to reduce the spread of the virus, have affected the demand for all modes of transportation, including public transport in cities. For example, Aloi et al. [1] indicate a fall of 76% in overall human mobility and a 93% decrease in public transport usage in Santander, Spain. Using aggregated mobility data from mobile phones in numerous urban areas in the U.S., Kishore et al. [2] show a surge in travel out of the cities immediately preceding the stay-at-home advisory. Another study points out a significant reduction in traffic volumes of 30% to 50% for select highways in California compared to prior shelter-in-place orders [3]. Individuals have changed their transportation patterns as personal travel decisions affect the spread of Covid-19 [4]. In response to the social distancing order, people have been less inclined to board packed buses and trains where social distancing is impossible. Accordingly, individuals reevaluate their transportation options in the face of the Covid-19 pandemic and shift to more isolated modes such as biking or walking. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Basak E, Al Balawi R, Fatemi S, Tafti A (2023) When crisis hits: Bike-Sharing platforms amid the Covid-19 pandemic. PLoS ONE 18(4): e0283603. https://doi.org/10.1371/journal. pone.0283603 Editor: Charitha Dias, Qatar University, QATAR Received: February 6, 2022 Accepted: March 13, 2023 Published: April 7, 2023 Copyright: © 2023 Basak et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting information files. Funding: The author(s) received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 1 / 20 PLOS ONEBike-Sharing platforms and Covid-19 With the Covid-19 pandemic, we witness an increasing awareness of bicycles as an alterna- tive means of transport, as many people either avoid using mass transit or encounter reduced mass transit services. In the U.S., sales of bicycles and related equipment also almost doubled in March 2020 compared with the same period in 2019 [5]. In times of crisis, bicycles can pro- vide resilience in transport systems, satisfying our mobility needs when mass transit systems are inaccessible. For instance, during the national public transit strike in France in December 2019, Parisians adapted and learned that bikes are dependable and credible modes of transport. The bike-sharing system in Paris, Ve´lib, gained popularity during the strike [6]. Other exam- ples are the 2005 New York City transit worker strike and Hurricane Sandy in 2012, which severely disrupted New York’s subway system. These events led to an increase in bicycle rider- ship in the city of New York by about 20% [7]. Bicycle sales surged in Japan after earthquakes struck that country in 2011 [8]. With the surge in demand for bikes, the popularity of bike-sharing platforms has also increased in March compared to the same period in the year before. They have become a viable transport alternative that reduces the risk of contracting or spreading the virus and relieves the fear of overcrowding [9, 10]. Compared to other means of transportation systems such as buses or trains, bicycling is an open-air activity and helps to avoid close contact with other travelers. Therefore, people have a more positive attitude toward bike-sharing for traveling amidst the pandemic [11]. For instance, Citi Bike in New York City announced a 67% increase in demand between March 1, 2020, and March 11, 2020, compared with the same period in 2019. Divvy in Chicago has also reported that the number of trips doubled in the same period [12]. A report from Foursquare and Apptopia shows that bike-share mobile application instal- lations in May and June of 2020 were up 15.6% and 23.3%, respectively, compared to the prior year [13]. A recent study by Li et al. [14] analyzed the demand for the bike-sharing platform in London over the period from January 2019 to June 2020. They found that the number of bike- sharing trips in London decreased after the lockdown; however, it was followed by an increase in demand over time. Heydari et al. [15] investigated the impact of the Covid-19 pandemic on the London bike-sharing platform over the period from March 2020 to December 2020. They initially observed a reduction in bike trips between March and April 2020; however, demand increased in May and June 2020. Moreover, Bouhouras et al. [16] found that the demand for bike-sharing platforms in Greek cities such as Igoumenitsa, Chania, and Rhodes increased in a short period of time before the lockdown period and peaked during the lockdown. Wang and Noland [17] exam- ined the impact of Covid-19 on both bike-sharing and subway usage. They found that both subway ridership and bike-sharing usage plummeted at the beginning; however, bike-sharing usage has almost returned to normal, whereas subway ridership has remained substantially below pre-pandemic levels. Other recent studies [18, 19] also revealed that bike-sharing plat- form usage in many cities has reached or surpassed pre-pandemic levels [20]. It is argued that bike-sharing demand has plummeted because of lockdown waves, even though it shows higher resiliency and lower drop than subway systems [10]. Following the stay-at-home advisories, many companies began to allow their employees to work from home, resulting in a significant reduction in travel within cities. Studies suggest that the demand for bike-sharing platforms has also decreased due to increased levels of remote working and stay- at-home advisories, but not as much as other means of transportation. A study conducted in Budapest, Hungary, shows that there has been an 80% decrease in public transport demand and only a 2% reduction in the use of bike-sharing platforms during the pandemic [21]. Another study that used ridership data from New York in 2020 showed that bike-sharing trips have decreased by less than 71%, whereas subway trips have decreased by 90% compared to February and March of 2019 [10]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 2 / 20 PLOS ONEBike-Sharing platforms and Covid-19 Another study by Li et al. [22] examined the changes in demand for micro-mobility services such as bike-sharing platforms in Zurich, Switzerland, before and during the lockdown period. Their spatial and temporal analysis results showed a decrease in the number of trips during the lockdown period. Their study also revealed that leisure- and shopping-related micro-mobility trips decreased while grocery-related trips increased. Apple mobility data shows that, by the end of May 2020, there was a decrease in all modes of transportation including driving, walk- ing, public transit, and bike-sharing; however, the reduction in public transit ridership was down much more than bike-share usage [23]. According to a press release by the Bureau of Transportation Statistics, ridership on eight of the largest docked bike-share systems in the U. S. declined by 44% from March through May 2020, compared to the same period in 2019 [24]. This could be explained by the fact that people have been traveling less due to stay-at-home advisories and limited business operations, and this might be affecting the demand for bike- sharing platforms like other transportation modes. In this paper, we examine how demand for bike-sharing platform usage changed immedi- ately following the first Covid-19 case and the first executive order in the U.S., using the data from January 2019 to July 2020. We use a fixed-effects econometric formulation of the differ- ence-in-differences (DID) estimation framework, which exploits a natural experiment to examine how bike-sharing trips have changed with the introduction of the first Covid-19 case and the first executive order. The DID estimation method is a suitable technique in our con- text, where randomization on the city level is not possible. DID requires panel data, which is part of the fixed-effects strategy, to capture the differences in post-treatment periods across the treatment and control groups [25]. One benefit of the DID model is that it allows us to avoid “the endogeneity problems that typically arise when making comparisons between heteroge- neous individuals” [26]. We also consider how the frequency of bike-sharing platform use can be different on week- days compared to weekends due to the changing travel patterns during the pandemic. In gen- eral, weekday travel is primarily made up of commuting to and from work, whereas weekend travel behavior is motivated by recreational activities. Differences in activity types can lead to different travel patterns that can be hypothesized on weekends and holidays, compared to weekdays [22, 27, 28]. For instance, Agarwal [27] suggests that there is a decrease in vehicle trips on weekends compared to weekdays at the household level. Kim et al. [28] find different weekend and weekday bike-sharing patterns. Their results point out that there is an increase in bike-sharing traffic volume on the weekends at the stations near parks and schools, which can be due to the rise in leisure and school activities on the weekends. In contrast, residential areas and subway stations are found to have less bike-sharing traffic volume on the weekends than on weekdays. Li et al. [22] find that there was a decrease in the number of micro-mobility service trips on weekdays during the lockdown period in Zurich. In contrast, there are only slight changes on weekends compared to before the lockdown period. In addition, we investigate what factors strengthen or weaken the impact of the pandemic on the frequency of bike-sharing use. Transportation infrastructure, land use, and neighborhood attributes contribute to individuals’ preference for bike-sharing [29]. Several studies examine the effects of the built environment, cycling facilities, transit proximity, and transportation facil- ity features on bike-sharing frequency [30–32]. The findings are consistent: More bicycle facili- ties and more excellent transit proximity lead to greater use of bike-sharing. Recent studies [33, 34] also find that better biking infrastructure is linked to higher bike-sharing demand during the Covid-19 pandemic. For instance, according to Bergantino et al. [33], safer cycling condi- tions and the creation of dedicated infrastructures encourage individuals to use bike-sharing platforms during the pandemic. Therefore, we also test the heterogeneous effects depending on such factors as the city’s bike-friendliness, transit-friendliness, and pedestrian-friendliness. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 3 / 20 PLOS ONEBike-Sharing platforms and Covid-19 The remainder of the paper is organized as follows. The next section introduces the data collection, variable definitions, and research method. Section 3 presents the main results. Sec- tion 4 shows the heterogeneous effects of bike-friendliness, transit-friendliness, and pedes- trian-friendliness. Finally, Section 5 concludes with a discussion of the findings. Materials and methods Data and variables We collected data from multiple sources. First, we collected the historical daily trip data avail- able to the public from bike-sharing programs in Austin, Boston, Chicago, Columbus, Minne- apolis, New York, Philadelphia, Pittsburgh, Portland, San Francisco, and Washington, D.C. The daily trip data includes trip duration, start time, end time, starting station, ending station, and subscription type (i.e., member, single rider). Based on this data, we compute our depen- dent variable, the daily trip frequency of bike-sharing platform trips (TripFrequencyij). It is cal- culated as the total daily trips of bike-sharing platform i at time j. During the construction of this variable, we excluded the bike-sharing platform trips with a duration of two minutes or less as there might be an issue while renting the bike (i.e., the bike is in a bad condition). We also excluded the bike-sharing platform trips of forty-five minutes or more, as these trips are more likely to represent leisure and recreational trips. A single ride for subscribers of these ser- vices includes forty-five minutes of ride time. Second, we collected data from various online sources to construct our treatment measures, FirstCaseij and FirstExecutiveOrderij. Our first treatment variable is the first Covid-19 case diagnosis (FirstCaseij), which is coded as 1, indicating that the first Covid-19 case is identified in city i as of day j’ such that j > = j’. The data on Covid-19 cases comes from The New York Times [35], which is based on reports from state and local health agencies. Our second treat- ment variable is FirstExecutiveOrderij, which is coded as 1, indicating that an executive order is issued in city i as of day j’ such that j > = j’. Specifically, we refer to the first executive action taken by the state governments against the Covid-19 pandemic, which is the stay-at-home advisories announced by the state governments. While stay-at-home advisories were lifted before August 2020, restrictions continued in most cities in various forms. For instance, Min- nesota’s stay-at-home advisory expired on May 18, 2020; however, it was replaced with a "stay safe Minnesota" order. Moreover, the state extended the state of emergency by another 30 days. Therefore, as the restrictions were still in effect, cities did not fully reopen before August 2020. Consequently, we used the entire period after the first executive order implementation. Table 1 lists the start date for each of our treatment variables in each city in our study. Table 1. Treatment start dates. City Austin, TX Boston, MA Chicago, IL Columbus, OH Minneapolis, MN New York, NY Philadelphia, PA Pittsburgh, PA Portland, OR San Francisco, CA Washington D.C. First Covid-19 case diagnosis First executive order 03/13/2020 02/01/2020 01/24/2020 03/14/2020 03/12/2020 03/02/2020 03/09/2020 03/14/2020 03/10/2020 03/05/2020 03/08/2020 03/24/2020 03/24/2020 03/21/2020 03/23/2020 03/27/2020 03/22/2020 03/23/2020 03/23/2020 03/23/2020 03/17/2020 04/01/2020 https://doi.org/10.1371/journal.pone.0283603.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 4 / 20 PLOS ONEBike-Sharing platforms and Covid-19 To construct the control variables, we collected weather data from the National Oceanic and Atmospheric Administration (NOAA). Control variables are included in the model as pre-treatment variables, as weather conditions can have different impacts on the demand for bike-sharing trips. Evidence from empirical studies [36–39] indicates that favorable weather conditions, such as higher temperatures, increase bike-sharing platform usage. In contrast, unfavorable conditions, such as precipitation and strong winds, will decrease such use. For example, Gebhart and Noland [36] suggest that cold temperatures, rain, and high humidity levels are likely to reduce the demand for bike-sharing platform trips in Washington, DC. In contrast, high temperatures are linked to an increased number of such trips. Similarly, the findings of Morton [37] point out that higher temperatures are associated with higher demand rates, whereas heavy precipitation, high wind speed, and relative humidity are negatively asso- ciated with the demand for the London bike-sharing system. Consistent with previous studies, An et al. [39] find out that there is a higher demand for the CitiBike bike-sharing platform in NYC in good weather, which is characterized by favorable temperature conditions, lack of winds, humidity, and rain. On the other hand, El-Assi et al. [31] show that weather conditions such as precipitation and high humidity decrease the demand for the Toronto bike-sharing system. Based on the background evidence, first, we control for the following weather-related variables: 1) Temperatureij, a measure of the average temperature for day j in city i in Fahren- heit (˚F); 2) Windij, a measure of the average wind speed for day j in city i in knots; 3) Snowij, a measure of snow depth for day j in city i in inches; 4) Rainij, a measure of total precipitation for day j in city i in inches; and 5) Humidityij, a measure of the average dew point for day j in city i in Fahrenheit (˚F). Furthermore, we collected data on the socio-economic characteristics of the cities from the U.S. Census Bureau. We include the population (Populationij), median income (Incomeij), the number of the elderly population (Elderlyij), the number of houses with two cars (Vehicleij), and the number of people commuting to work with bikes (Commuteij). When combined, we end up with a panel data set that comprises eleven cities spanning from January 2019 through July 2020. To make the interpretation of the socio-economic characteris- tics easier, we include the log-transformed values in our analyses. Following prior literature, we keep the weather-related variables in their original form [36, 37, 39]. Tables 2 and 3 present the summary statistics and the correlation of the critical variables, respectively. We use the log- transformed values of the socio-economic characteristic to interpret linear regression results. Table 2. Descriptive statistics. Variable FirstCase FirstExecutiveOrder ln(TripFrequency) ln(Population) ln(Elderly) ln(Income) ln(Vehicle) ln(Commute) Temperature Rain Snow Wind Humidity https://doi.org/10.1371/journal.pone.0283603.t002 Obs. 6,052 6,052 6,052 6,052 6,052 6,052 6,052 6,052 6,052 6,052 6,052 6,052 6,052 Mean 0.26 0.23 7.27 15.15 13.22 11.28 13.14 9.75 56.89 0.12 0.08 7.23 44.19 Std. Dev. 0.44 0.42 2.06 0.75 0.81 0.17 0.60 0.96 16.49 0.30 0.51 3.45 16.89 Min 0.00 0.00 0.00 14.25 12.05 11.01 12.39 7.81 -13.50 0.00 0.00 0.50 -24.60 Max 1.00 1.00 11.45 16.75 14.92 11.62 14.32 11.07 91.00 4.42 9.10 24.00 77.90 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 5 / 20 PLOS ONEBike-Sharing platforms and Covid-19 Table 3. Correlation. Variable FirstCase FirstExecutiveOrder ln(TripFrequency) ln(Population) ln(Elderly) ln(Income) ln(Vehicle) ln(Commute) Temperature Rain Snow Wind Humidity 1 1.00 0.90 -0.07 0.03 0.05 0.12 0.03 0.02 0.16 0.00 -0.05 0.05 0.12 1 2 3 4 5 6 7 8 9 10 11 12 13 2 3 4 5 6 7 8 9 10 11 12 13 1.00 -0.08 0.00 0.02 0.10 0.00 -0.01 0.24 0.00 -0.07 0.03 0.20 1.00 0.87 0.85 0.51 0.85 0.86 0.13 -0.02 -0.05 0.08 0.08 1.00 0.95 0.27 0.96 0.78 1.00 0.21 0.95 0.75 -0.09 -0.12 0.02 0.07 0.11 0.02 0.07 0.13 -0.12 -0.14 1.00 0.21 0.66 0.08 -0.05 -0.07 0.18 0.05 1.00 0.74 -0.09 0.03 0.09 0.12 -0.13 1.00 -0.04 -0.03 0.00 0.14 -0.06 1.00 0.02 -0.27 -0.23 0.93 1.00 0.01 0.11 0.12 1.00 0.06 -0.22 1.00 -0.24 1.00 https://doi.org/10.1371/journal.pone.0283603.t003 Model-free evidence Before we introduce our model specification, we present visual model-free evidence of the role of the first Covid-19 case diagnosis and the first executive order implementation on the use of the bike-sharing platforms in Figs 1 and 2. It is worth noting that Figs 1 and 2 do not account for time-fixed effects. In Fig 1, we plot the daily trip frequency 60 days before and after the first reported Covid-19 case in each of the eleven cities (excluding Minneapolis, MN, as its bike- sharing systems do not report daily trip data between December and March). The solid vertical line represents the first Covid-19 diagnosis. The dashed horizontal lines represent the average daily bike-sharing trip frequency before and after the first Covid-19 diagnosis. In Fig 1, we observe that the daily bike trip frequency decreases in most cities following their first reported Covid-19 case, except for Boston and Chicago, which had cold winters and were beginning to warm up in the ensuing weeks. In Fig 2, we plot the daily bike trip frequency 60 days before and after the first executive order implementation across the cities. We observe a similar trend in Fig 2. Across all cities, the daily trip frequency declined in the days immediately after the first executive orders were implemented. Similar to the plots in Figs 1 and 2, we plot the difference in the bike-sharing trip frequency before and after the first Covid-19 case diagnosis and the first executive order implementation by weekday and weekend (see Fig A1 and A2 in S1 Appendix for more details). In Fig A1 in S1 Appendix, we show the weekday bike trip frequencies 60 days before and after the first Covid- 19 case reported in each city. We notice a decrease in the trip frequency on weekdays following the first reported Covid-19 case in most cities, with a few exceptions in which we see a close average daily trip frequency after the first Covid-19 case was reported, such as in Boston. In Fig A2 in S1 Appendix, we show the weekend bike trip frequencies 60 days before and after the first Covid-19 case that was reported. When we look at the changes in the weekend trip fre- quency, we see opposing results suggesting an increase in the daily trip frequency in some cit- ies, such as Philadelphia and Pittsburgh. Moreover, we also notice a similar trend in the daily weekday and weekend trip frequency after the first executive order implementation across the cities in this study (see Figs A3 and A4 in S1 Appendix). Finally, Fig A5 in S1 Appendix shows the bike-sharing seasonal trend of Feb- ruary-June 2019 (pre-covid) compared to February-June 2020 (post-covid) for each city in this study. Relative to the patterns observed in 2019, we see a short-term decrease in bike-sharing PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 6 / 20 PLOS ONEBike-Sharing platforms and Covid-19 Fig 1. Frequency of bike-sharing platform trips over time, before and after the first Covid-19 case. https://doi.org/10.1371/journal.pone.0283603.g001 trip frequency following the pandemic’s start (towards the end of the first quarter of 2020). These plots provide further model-free evidence of the changes in the use of the bike-sharing system due to Covid-19. Figs 3 and 4 show dumbbell charts that compare the average daily bike-sharing trip fre- quency before and after each of our two treatments (the first Covid-19 case diagnosis and the first executive order implementation) that occurred over the entire period of study. We use the log transformation of trip frequencies for better visualization. Generally, we see short-term decline in bike-sharing frequency after the first reported infections and the first executive order implementation within the same U.S. cities. The blue point represents the log average daily trip frequency for the period before the treatment occurred, and the red point represents the log average daily trip frequency in the period after the treatment occurred. Fig 3 shows that the frequency of average daily bike-sharing platforms decreases after the first Covid-19 case diagnosis, except for a few cities such as New York, Philadelphia, and Chicago, in which PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 7 / 20 PLOS ONEBike-Sharing platforms and Covid-19 Fig 2. Frequency of bike-sharing platform trips over time, before and after the first executive order implementation. https://doi.org/10.1371/journal.pone.0283603.g002 the average daily bike-sharing trips did not change significantly; and also except for Columbus, in which we see an apparent increase. Fig 4 also shows that the frequency of average daily bike- sharing trips decreases upon the first executive order implementation, again with the afore- mentioned exceptions. However, it is essential to note that these plots do not consider the cit- ies’ weather conditions and socio-economic characteristics. Therefore, in the following subsection, we propose a statistical model to evaluate how bike-sharing frequency changed fol- lowing the Covid-19 pandemic, accounting for weather conditions and socio-economic characteristics. Model specification The model-free evidence shows that the frequency of trips on bike-sharing platforms generally decreased in U.S. cities following the first Covid-19 cases and the first executive order PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 8 / 20 PLOS ONEBike-Sharing platforms and Covid-19 Fig 3. Dumbbell chart comparing the average daily bike-sharing trip frequency before and after the first Covid-19 case. https://doi.org/10.1371/journal.pone.0283603.g003 implementation. However, as mentioned earlier, model-free evidence does not account for many factors that could influence the trip frequency for bike-sharing and the spread of the virus. Therefore, before incorporating such factors into our statistical analysis, we need to know the pandemic’s actual effect on bike-sharing trip frequency. However, we may observe a decrease in the trip frequency in some cities and an increase in others. Thus, we use a fixed-effects econometric formulation of the DID estimation framework to examine how the Covid-19 pandemic affected the frequency of bike-sharing trips. The primary benefit of this estimation model is that we can mimic an experimental design using observa- tional data. This method compares the differences in bike-sharing trip frequency in treated cit- ies before and after the treatment event—the onset of the pandemic—to the differences in the untreated cities (i.e., those cities yet to report a coronavirus case or to implement a first execu- tive order). The longitudinal nature of the data allows us to use the yet untreated observations in the data as controls for the treated observations; that is, those cities that have yet to have a PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 9 / 20 PLOS ONEBike-Sharing platforms and Covid-19 Fig 4. Dumbbell chart comparing the average daily bike-sharing trip frequency before and after the first executive order implementation. https://doi.org/10.1371/journal.pone.0283603.g004 first Covid-19 case or a Covid-related executive order. To facilitate estimation, we use the fixed-effects regression formulation of the DID model, a formulation described in [25], as fol- lows: ln TripFrequency ð Þij ¼ b0 þ b1Treatmentij þ gWij þ yj þ mi þ εij; ð1Þ where ln(TripFrequency)ij is the log-transformed value of our dependent variable in city i dur- ing day j. Treatmentij refers to the treatment variables in city i during day j: FirstCaseij or First- ExecutiveOrderij. They are applied in different cities at different times. To control for existing time-invariant differences among the heterogeneous geographical locations, i.e., cities, we included city-fixed effects, μi, in our model. In addition, we included time-fixed effects, θj, to control for common temporal shocks. This allows for non-linear time-varying effects in the DID model. Wij is the set of control variables, which includes ln(Population), ln(Income), ln PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 10 / 20 PLOS ONEBike-Sharing platforms and Covid-19 (Elderly), ln(Vehicle), ln(Commute), Temperature, Rain, Snow, Wind, and Humidity. Finally, εij is the error term. Results Table 4 reports the coefficient estimates of Eq (1) for the dependent variable ln(TripFre- quency). As shown in Column (1), we estimate an increase in the log of bike-sharing platform trip frequency of 0.196 on average across eleven cities after the first Covid-19 case, adjusted for covariates. An economic interpretation of this result suggests an average adjusted increase in the number of daily bike-sharing trips by 22% (rounded from the following: [exp(0.196)-1] 100 = 21.65%). On the other hand, from Column (2), we observe a decrease in the log of bike- sharing platform trip frequency by 0.353 after the stay-at-home order implementation. Table 4. Bike-sharing trip frequency: Fixed-effects regression results. Dependent Variable Treatment variable ln(Population) ln(Elderly) ln(Income) ln(Vehicle) ln(Commute) Temperature Rain Snow Wind Humidity Observations R2 R2(Adjusted) F-statistic Daily fixed effects City fixed effects p<0.01 p<0.05 * p<0.10 Robust standard errors are given in parentheses. https://doi.org/10.1371/journal.pone.0283603.t004 (1) ln(TripFrequency) First Case 0.196** (0.098) -1.748 (14.555) 7.876 (9.519) 6.751* (3.567) 7.591* (2.505) 0.332 (0.429) 0.043* (0.003) 0.012 (0.022) -0.172* (0.038) -0.026* (0.005) -0.017* (0.003) 6,052 0.284 0.206 197.145* Yes Yes (2) ln(TripFrequency) First Executive Order -0.353* (0.205) -4.171 (14.779) 8.446 (9.667) 7.408 (3.652) 7.355* (2.502) 0.357 (0.437) 0.044* (0.003) 0.012 (0.021) -0.170* (0.038) -0.026* (0.005) -0.017* (0.002) 6,052 0.283 0.205 196.389*** Yes Yes PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 11 / 20 PLOS ONEBike-Sharing platforms and Covid-19 Economically, this result suggests a reduction of 30% (rounded from the following: [exp (-0.353)-1]100 = -29.74%) in the number of daily bike-sharing trips. We also further examine the robustness of our model to temporal trends using a relative time model (see Fig A6 and Fig A7 in S1 Appendix for more details). Furthermore, we divide our dataset into three panels to compare weekday, weekend, and bank holiday travel behavior. We include the bank holidays (i.e., New Year’s Day, Martin Luther King Jr. Day, or Independence Day) observed by the Federal Reserve System. The results are given in Tables 5 and 6. We estimate a 22% (rounded from the following: [exp(0.197)-1] 100 = 21.77%) increase in the trip frequency on average across cities during the weekdays upon the first Covid-19 case (see Column 1 in Table 5), whereas our results do not suggest a Table 5. Bike-sharing trip frequency: Weekday-, weekend-, and bank holiday-specific fixed-effects regression results where the treatment is the first Covid-19 case. (1) (2) ln(TripFrequency) ln(TripFrequency) Weekday First Case 0.197 (0.098) -3.378 (13.776) 10.125 (9.625) 6.780 (2.961) 7.441* (2.145) 0.440 (0.392) 0.037* (0.002) 0.006 (0.016) -0.158* (0.032) -0.024* (0.005) -0.014* (0.002) 4,161 0.271 0.189 126.577* Yes Yes Weekend First Case 0.174 (0.113) 0.892 (18.152) 4.440 (10.328) 7.152 (5.179) 8.527 (3.777) 0.007 (0.674) 0.059* (0.008) 0.019 (0.039) -0.175* (0.036) -0.028* (0.006) -0.025* (0.003) 1,736 0.337 0.257 71.746* Yes Yes Dependent Variable Travel Behavior Treatment variable ln(Population) ln(Elderly) ln(Income) ln(Vehicle) ln(Commute) Temperature Rain Snow Wind Humidity Observations R2 R2(Adjusted) F-statistic Daily fixed effects City fixed effects * p<0.01 p<0.05 * p<0.10 Robust standard errors are given in parentheses. https://doi.org/10.1371/journal.pone.0283603.t005 (3) ln(TripFrequency) Bank Holiday First Case 0.159 (0.420) 0.824 (29.735) -13.653 (17.933) 2.036 (7.694) -0.633 (5.811) 1.003 (1.025) 0.061* (0.010) 0.141 (0.234) -0.199* (0.068) -0.039 (0.016) -0.028* (0.009) 155 0.438 0.273 8.443* Yes Yes PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 12 / 20 PLOS ONETable 6. Bike-sharing trip frequency: Weekday- and weekend-specific fixed-effects regression results where the treatment is the first executive order. Bike-Sharing platforms and Covid-19 (1) ln(TripFrequency) Weekday First Executive Order -0.358 (0.257) -5.982 (2) ln(TripFrequency) Weekend First Executive Order -0.323 (0.156) -1.063 (13.890) 10.773 (9.797) 7.410 (3.070) 7.273* (2.147) 0.466 (0.396) 0.037* (0.002) 0.007 (0.016) -0.155* (0.032) -0.025* (0.005) -0.014* (0.002) 4,161 0.270 0.188 125.921* Yes Yes (18.444) 4.819 (10.404) 7.819 (5.169) 8.181 (3.700) 0.025 (0.684) 0.059* (0.008) 0.021 (0.038) -0.174* (0.037) -0.029* (0.006) -0.025* (0.003) 1,736 0.337 0.257 71.657* Yes Yes Dependent Variable Travel Behavior Treatment variable ln(Population) ln(Elderly) ln(Income) ln(Vehicle) ln(Commute) Temperature Rain Snow Wind Humidity Observations R2 R2(Adjusted) F-statistic Daily fixed effects City fixed effects * p<0.01 ** p<0.05 * p<0.10 Robust standard errors are given in parentheses. https://doi.org/10.1371/journal.pone.0283603.t006 statistical significance of the same effect on weekends (see Column 2 in Table 5) and the bank holidays (see Column 3 in Table 5). Table 6 shows the results when the treatment variable is FirstExecutiveOrder in which we observe opposing results. We estimate a statistically signifi- cant decrease in the trip frequency of 28% (rounded from the following: [exp(-0.323)-1]100 = -27.60%) on the weekends (see Column 2 in Table 6), whereas we find no evidence of the same effect during the weekdays (see Column 1 in Table 6). However, we do not observe any effect for the bank holidays data panel as the variable in question is being omitted by the regression (see Column 3 in Table 6). The reason behind this is that any variables that are constant within every unit are redundant in a fixed-effects model and will be omitted from the model. Due to the launch dates of the first executive order, Memorial Day 2020 and Independence Day 2022 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 13 / 20 PLOS ONEBike-Sharing platforms and Covid-19 Table 7. Description: BikeScore1, WalkScore1, and TransitScore1. 90–100 70–89 50–69 0–49 BikeScore1 Biker’s Paradise Very Bikeable Bikeable Somewhat Bikeable WalkScore1 Walker’s Paradise Very Walkable Walkable Car-Dependent TransitScore1 Rider’s Paradise Excellent Transit Good Transit Some/Minimal Transit https://doi.org/10.1371/journal.pone.0283603.t007 are treated the same for each unit. Therefore, our treatment variable becomes constant for each city and does not create any variation. These results generally show the differences in residents’ travel behavior between weekdays, weekends, and bank holidays. After the first Covid-19 diagnosis, individuals might have started using bike-sharing platforms as an alternative to other modes of transportation on weekdays, especially, for journeys to and from work. However, with the first executive order implementation, on average, individuals might tend to stay inside more rather than go out. We also ran the analysis on daily trip frequencies with fewer than thirty minutes, as a single ride for non-subscribers includes thirty minutes of ride time. The results are consistent. Heterogeneity analyses While our empirical estimations thus far suggest a significant impact of the Covid-19 pan- demic on the frequency of bike-sharing platform trips, it is worth examining the factors that might amplify the strength of the effect. Prior literature [29–31] suggests that transportation infrastructure, land use, built environment, and neighborhood attributes contribute to individ- uals’ preference for bike-sharing systems. One crucial factor that can moderate the impact of Covid-19 on bike-sharing platforms’ trip frequency is the pre-existing biking infrastructure. First, in cities with more bike lanes, longer bike route lengths, fewer hills, higher road connectivity, and bicycle-aware traffic, bike-sharing platforms should more likely be adopted by individuals and used as an alternative transportation mode. Second, in walkable cities with better access to amenities, residents might be embracing these platforms more due to easy and comfortable access to bike stations. Lastly, in cities with access to public transit, bike-sharing platforms might be used more by the residents due to better connectivity of the transit network. Therefore, we test the heterogeneous effects depending on such factors as the city’s bike-friendliness, transit-friendliness, and pedestrian-friendliness. We collected data from Walk Score [40] to measure 1) bike-friendliness (BikeScore1) [41], which measures the built environment’s ability to support biking for a given location, 2) pedes- trian-friendliness (WalkScore1) [41], which measures the walkability of any address by analyz- ing the walking routes to nearby amenities within a 5-minute walk, and 3) transit-friendliness (TransitScore1), which measures how well a location is served by public transit [41]. These measures range from 0 to 100 and divide cities into different groups [1]. Based on the classifi- cation of BikeScore1, WalkScore1, and TransitScore1, the cities in our dataset scored less than 90 in all measures. Detailed information on the groups and descriptive statistics of the scores are given in Tables 7 and 8, respectively. Table 8. Descriptive statistics: BikeScore1, WalkScore1, and TransitScore1. Variable BikeScore1 WalkScore1 TransitScore1 Obs. 6,052 6,052 6,052 Mean 67.71 70.13 61.43 Std. Dev. 9.84 15.54 16.33 Min 49.90 40.50 32.80 Max 83.50 88.30 84.30 https://doi.org/10.1371/journal.pone.0283603.t008 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 14 / 20 PLOS ONEBike-Sharing platforms and Covid-19 Then, we re-estimate Eq (1) incorporating interaction terms for these classifications with the treatment. The new equation including the interaction terms is given below. Note that as the moderators are static, fixed-effects panel regressions do not yield estimates for β2. The results are given in Table 9. ln TripFrequency ð Þij ¼ b0 þ b1Treatmentij þ b2Moderatorj þ b3Treatmentij ∗ Moderatorj þ gWij þ yj þ mi þ εij; ð2Þ Surprisingly, these findings suggest interesting differences. First, we see that the impact of the first Covid-19 case and the first executive order implementation on bike-sharing platforms’ trip frequency is more substantial in bikeable cities. We estimate that the effect of the first Covid-19 case on Trip Frequency is approximately 90% (rounded from the following: [exp (0.640)-1]100 = 89.64%) higher in “very bikeable” cities than in “bikeable” cities (see Table 9, Column 1). We observe that the impact of the first executive order implementation on Trip Frequency is approximately 103% (rounded from the following: [exp(0.708)-1] 100 = 102.99%) higher in “very bikeable” cities than in “bikeable” cities (see Table 9, Column 4). Followed by the first Covid-19 case and the first executive order implementation, we esti- mate a decrease that is respectively greater by approximately 33% (rounded from the following: [exp(-0.405)-1]100 = -33.30%) (see Table 9, Column 1) and 30% (rounded from the following: [exp(-0.356)-1]100 = -29.95%) (see Table 9, Column 4) in Trip Frequency in “somewhat bike- able” as compared to “bikeable” cities. These results might suggest that the residents in bike- friendly cities embrace these platforms more due to better pre-existing biking infrastructure. With safe and comfortable biking afforded by good biking infrastructure, residents are more likely to use bike-sharing platforms for commuting and recreational purposes. Moreover, we observe a more substantial impact on the Trip Frequency in very walkable cities upon both the first Covid-19 case (see Table 9, Column 2) and stay-at-home advisory implementation (see Table 9, Column 5). In pedestrian-friendly cities, residents might be embracing these platforms more as a result of easy and comfortable access to bike stations by walking. Moreover, similar to bike-friendliness and pedestrian-friendliness, we observe a more substantial impact on the Trip Frequency in cities with excellent transit upon both the first Covid-19 case (see Table 9, Column 3) and the first executive order implementation (see Table 9, Column 6). Lastly, unlike the “somewhat bikeable” classification of cities, we do not observe that car dependence or having some transit (compared to good transit) has any moder- ating influence on the effect of the first Covid-19 case (see Table 9, Column 3) and the first executive order implementation (see Table 9, Column 6) on the use of bike-sharing platforms. Discussion and conclusion We used a DID framework formulated as a fixed-effects regression model to examine how bike-sharing trip frequency in the United States changed with the onset of the Covid-19 pan- demic. We modeled the first reported Covid-19 cases and the implementation of the first exec- utive order in each municipality as two treatment events. We also accounted for socio- economic factors, weather, and fixed effects for each day and city. First, our results indicate that, on average, the first reported Covid-19 cases had a positive and statistically significant effect on the frequency of bike trips in U.S. cities. This could be explained by the fact that the existence of the first reported Covid-19 case in U.S. cities has heightened individuals’ sensitiv- ity to cleanliness and social distance. Therefore, individuals were compelled to change their travel behavior and look for alternative systems of mobility that may offer more resilient urban PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 15 / 20 PLOS ONETable 9. Bike-sharing trip frequency: Fixed-effects regression results by heterogeneous effects. The reference indicator for the bike-friendliness scale is Bikeable; the reference indicator for the pedestrian-friendliness scale is Walkable; and the reference indicator for the transit-friendliness is Good Transit. Dependent Variable ln(TripFrequency) ln(TripFrequency) ln(TripFrequency) ln(TripFrequency) ln(TripFrequency) ln(TripFrequency) (1) (2) (3) (4) (5) (6) Bike-Sharing platforms and Covid-19 First Case 0.049 (0.194) First Case 0.078 (0.113) 0.824 (0.312) -0.020 (0.186) -23.398 (20.104) 15.485 (10.788) 9.591 (4.682) 2.517 (2.677) 1.647* (0.624) 0.041* (0.003) 0.010 (0.021) -0.174* (0.038) -0.025* (0.006) -0.016* (0.002) 6,052 0.323 0.248 200.193* Yes Yes 0.797* (0.263) -0.106 (0.133) -13.360 (16.429) -5.067 (11.572) 1.747 (4.748) 2.087 (3.753) 1.225 (0.587) 0.042* (0.003) 0.006 (0.024) -0.173* (0.039) -0.025* (0.005) -0.017* (0.002) 6,052 0.325 0.250 201.941* Yes Yes Treatment variable Interaction: VeryBikeable Interaction: SomewhatBikeable Interaction: VeryWalkable Interaction: CarDependent Interaction: ExcellentTransit Interaction: SomeTransit ln(Population) ln(Elderly) ln(Income) ln(Vehicle) ln(Commute) Temperature Rain Snow Wind Humidity Observations R2 R2(Adjusted) F-statistic Daily fixed effects City fixed effects First Case 0.314 (0.152) 0.640*** (0.228) -0.405* (0.215) -28.236 (21.704) 17.118 (11.362) 10.441 (5.129) 2.032 (2.710) 1.701* (0.639) 0.041* (0.003) 0.004 (0.024) -0.173* (0.038) -0.025* (0.005) -0.017* (0.002) 6,052 0.326 0.252 203.65* Yes Yes * p<0.01 ** p<0.05 * p<0.10 Robust standard errors are given in parentheses. https://doi.org/10.1371/journal.pone.0283603.t009 First Executive Order First Executive Order First Executive Order -0.429* (0.025) -0.387* (0.213) -0.189* (0.112) 0.708*** (0.225) -0.356* (0.194) -14.129 (16.143) -3.335 (11.246) 2.514 (4.512) 2.545 (3.484) 1.182 (0.565) 0.041* (0.003) 0.008 (0.022) -0.173* (0.038) -0.025* (0.005) -0.016* (0.002) 6,052 0.330 0.257 207.319* Yes Yes 0.833 (0.324) -0.023 (0.198) -14.108 (14.231) -4.122 (10.632) 1.668 (3.898) 2.322 (3.314) 1.228 (0.539) 0.042* (0.003) 0.008 (0.023) -0.173* (0.038) -0.025* (0.005) -0.017* (0.002) 6,052 0.326 0.252 203.57* Yes Yes 0.806* (0.272) -0.104 (0.139) -14.436 (14.436) -2.551 (10.408) 2.534 (3.727) 2.659 (3.139) 1.219 (0.524) 0.042* (0.003) 0.011 (0.020) -0.173* (0.038) -0.025* (0.004) -0.017* (0.002) 6,052 0.328 0.254 205.077* Yes Yes PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 16 / 20 PLOS ONEBike-Sharing platforms and Covid-19 transportation. Bike-sharing platforms offer alternative transportation to avoid crowds in the cities. Second, we observe that the first executive order advisories had a negative and statisti- cally significant effect on the frequency of bike trips in U.S. cities. This could be explained by the fact that lockdown restrictions and working from home have led to a decline in commuting bike trips and other modes of transportation such as public transit. We also examined sources of heterogeneity in the effect of the pandemic on bike-sharing use. We compared how bike-sharing use changed between weekends and weekdays with the onset of the pandemic. We observed an increase in weekday-specific trip frequency as a result of the first Covid-19 case diagnosis and a decrease in weekend-specific trip frequency due to the first executive order implementation. We also tested for heterogeneous impacts across a set of city-level characteristics. We found that there is a greater increase in the frequency of bike- sharing trips in more bike-friendly, transit-friendly, and pedestrian-friendly cities upon both the first Covid-19 case diagnosis and the first executive order implementation. We might con- clude that bike-sharing platforms have an essential role in individuals’ travel behavior, espe- cially in cities with better bike and transit infrastructure. These platforms are perceived as a sustainable and resilient transportation option by individuals due to the unprecedented conse- quences of the Covid-19 pandemic. Bike-sharing platforms offer a sustainable and active mode of transportation, and hence it is important to better understand the factors that affect their adoption by the populace. The Covid-19 pandemic represents an opportunity for cities to embrace new paradigms for urban mobility. Bike-sharing platforms represent one way in which cities provide a resilient and adaptive transport network to face the potential challenges of disruptive events like the Covid- 19 pandemic. The pandemic has already highlighted the importance of rethinking the design of urban transit for greater resilience to such disruptive events. Cities may consider how to encourage greater use of bike-sharing platforms. Decisions by city authorities such as offering free or reduced membership could break down barriers to the adoption of bike-sharing. With the support of local authorities in creating more bike lanes and accessibility to public spaces, bike-sharing platforms can attract more individuals. Proper incentives followed by infrastruc- ture adjustments could ensure that individuals will become accustomed to bike-sharing plat- forms and continue to use them even after the pandemic. For instance, in New York, city officials are already planning to expand Citi Bike and add more docking stations in its busiest areas [12]. Investing in bike-sharing platforms and cycling infrastructure could lead to an increase in memberships because individuals’ willingness to bike is closely linked to how safe they feel [42]. It is important to note that our research is subject to several limitations. First, the adjusted R2 values of the models are low, ranging from 0.18 to 0.26. Low values might be an indicator of the models would not be suitable for use in the predictive modeling of the outcome variables. Hence, the aims of our model interpretations are limited to assessing the direction and signifi- cance of coefficient estimates for causal inference. Second, the main challenge with DID estimation is to ensure that no pre-treatment trends in the absence of treatment [25]. The violation of this assumption can lead to biased causal esti- mates. Although there is no statistical test for this assumption, we examine the robustness of our model to temporal trends using a relative time model. Our findings suggest that there are no pre-existing trends in bike-sharing demand across the cities that experience the first Covid- 19 diagnosis (see Fig 6A in S1 Appendix). As seen in Fig 7A in S1 Appendix, we observe that pre-treatment trends exist in bike-sharing demand across the cities experiencing the first exec- utive order implementation. Future research could explore different estimators that could overcome this challenge. There are a few recent papers that focus on different ways to relax this assumption [43–45]. They propose alternative estimators when the parallel trends PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 17 / 20 PLOS ONEBike-Sharing platforms and Covid-19 assumption is violated and examine the robustness of the results to the potential violations of parallel trends. Third, our model only examines the short-term effect of the Covid-19 pandemic on bike- sharing demand due to the data collection period. In the future, depending on data availability by the bike-sharing providers, the effects can be examined for longer periods. Fourth, we use the first executive order implementation and the first Covid-19 case as prox- ies of the Covid-19 pandemic; however, we should keep in mind that the first executive order implementation might be correlated with the first Covid-19 case. This might be also one of the reasons that might explain the pre-treatment trend in Fig 7A in S1 Appendix. Lastly, our reported results are based upon the actual usage of bike-sharing platforms, along with public transit data for eleven cities in the U.S. Despite the lack of data available for other U.S. cities, we believe that the exogenous nature of the Covid-19 pandemic provides robust insights into the relationship between the Covid-19 pandemic and travel behaviors. Given that we are still in the midst of the pandemic, we expect that forthcoming data could reveal more about the pandemic’s long-term effects on travel behavior. It is very plausible that the effects observed thus far may serve as a signal for more lasting changes to come in urban travel behaviors. Supporting information S1 Dataset. (CSV) S1 Appendix. (DOCX) Author Contributions Conceptualization: Ecem Basak, Ali Tafti. Data curation: Ecem Basak, Ramah Al Balawi, Sorouralsadat Fatemi. Formal analysis: Ecem Basak, Ramah Al Balawi. Methodology: Ecem Basak, Ramah Al Balawi, Sorouralsadat Fatemi, Ali Tafti. Project administration: Ecem Basak. Supervision: Ali Tafti. Visualization: Ramah Al Balawi. Writing – original draft: Ecem Basak, Ramah Al Balawi, Sorouralsadat Fatemi, Ali Tafti. Writing – review & editing: Ecem Basak, Ramah Al Balawi, Sorouralsadat Fatemi, Ali Tafti. References 1. Aloi A, Alonso B, Benavente J, Cordera R, Echa´niz E, Gonza´ lez F, et al. Effects of the COVID-19 lock- down on urban mobility: Empirical evidence from the city of Santander (Spain). Sustain. 2020; 12: 1–19. https://doi.org/10.3390/su12093870 2. Kishore N, Kahn R, Martinez PP, De Salazar PM, Mahmud AS, Buckee CO. Lockdowns result in changes in human mobility which may impact the epidemiologic dynamics of SARS-CoV-2. Sci Rep. 2021; 11: 1–12. https://doi.org/10.1038/s41598-021-86297-w PMID: 33772076 3. Shilling F, Waetjen D. 2020 Special Report: Impact of COVID19 on California Traffic Accidents. 2020; 2020: 9. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 18 / 20 PLOS ONEBike-Sharing platforms and Covid-19 4. Oum TH, Wang K. Socially optimal lockdown and travel restrictions for fighting communicable virus including COVID-19. Transp Policy. 2020; 96: 94–100. https://doi.org/10.1016/j.tranpol.2020.07.003 PMID: 32834681 5. Bert J, Schellong, Daniel Hagenmaier, Markus Hornstein, David Wegscheider AK, Palme T. How COVID-19 Will Shape Urban Mobility. BCG. 2020. 6. Bicycle R&D. Paris Public Transport Strike Shows the Potential of the Bicycle. Copenhagenize. 2019. https://copenhagenize.eu/news-archive/2019/12/19/paris-public-transport-strike-shows-the-potential- of-the-bicycle 7. Morris DZ. After coronavirus, bicycles will have a new place in city life. Fortune. 2020. https://fortune. com/2020/06/15/bicycles-coronavirus-cities-lime-citi-bike/ 8. Sagiike H, Kitanaka A. Japan Bike Shop Sales Double on Commuters’ Earthquake Fears. Bloomberg. 2011. https://www.bloomberg.com/news/articles/2011-04-18/japan-bicycle-retailer-s-sales-triple-as- commuters-haunted-by-earthquakeT 9. 10. Jobe J, Griffin GP. Bike share responses to COVID-19. Transp Res Interdiscip Perspect. 2021; 10. https://doi.org/10.1016/j.trip.2021.100353 PMID: 36844003 Teixeira JF, Lopes M. The link between bike sharing and subway use during the COVID-19 pandemic: The case-study of New York’s Citi Bike. Transp Res Interdiscip Perspect. 2020; 6: 100166. https://doi. org/10.1016/j.trip.2020.100166 PMID: 34173457 11. Nikiforiadis A, Ayfantopoulou G, Stamelou A. Assessing the impact of COVID-19 on bike-sharing usage: The case of thessaloniki, Greeceoriadis, Andreas Ayfantopoulou, Georgia Stamelou, Afroditi. Sustainability. 2020; 12. 12. Hu W. A Surge in Biking to Avoid Crowded Trains in NYC. The New York Times. 2020. https://www. nytimes.com/2020/03/14/nyregion/coronavirus-nyc-bike-commute.html 13. Apptopia. Mobility reImagined: Tracking transportation trends in the U.S. using mobile app & location data. 2020. https://www.terrapinn.com/exhibition/move/index.stm 14. Li H, Zhang Y, Zhu M, Ren G. Impacts of COVID-19 on the usage of public bicycle share in London. Transp Res Part A Policy Pract. 2021; 150: 140–155. https://doi.org/10.1016/j.tra.2021.06.010 PMID: 34511745 15. Heydari S, Konstantinoudis G, Behsoodi AW. Effect of the COVID-19 pandemic on bikesharing demand and hire time: Evidence from Santander Cycles in London. PLoS One. 2021; 16: e0260969. https://doi. org/10.1371/journal.pone.0260969 PMID: 34855914 16. Bouhouras E, Basbas S, Ftergioti S, Paschalidis E, Siakantaris H. COVID-190s pandemic effects on bike sharing systems: A new reality for urban mobility? Appl Sci. 2022; 12: 1230. https://doi.org/10. 3390/app12031230 17. Wang H, Noland RB. Bikeshare and subway ridership changes during the COVID-19 pandemic in New York City. Transp Policy. 2021; 106: 262–270. https://doi.org/10.1016/j.tranpol.2021.04.004 PMID: 34975237 18. Kubalˇa´ k S, Kalasˇova´ A, Ha´jnik A. The bike-sharing system in slovakia and the impact of covid-19 on this shared mobility service in a selected city. Sustainability. 2021; 13: 6544. https://doi.org/10.3390/ su13126544 19. Song J, Zhang L, Qin Z, Ramli MA. Spatiotemporal evolving patterns of bike-share mobility networks and their associations with land-use conditions before and after the COVID-19 outbreak. Phys A Stat Mech its Appl. 2022; 592: 126819. https://doi.org/10.1016/j.physa.2021.126819 PMID: 35002051 20. Bustamante X, Federo R, Ferna´ ndez-i-Marin X. Riding the wave: Predicting the use of the bike-sharing system in Barcelona before and during COVID-19. Sustain Cities Soc. 2022; 83: 103929. https://doi. org/10.1016/j.scs.2022.103929 PMID: 35535208 21. Bucsky P. Modal share changes due to COVID-19: The case of Budapest. Transp Res Interdiscip Per- spect. 2020; 8. https://doi.org/10.1016/j.trip.2020.100141 PMID: 34171021 22. Li A, Zhao P, Haitao H, Mansourian A, Axhausen KW. How did micro-mobility change in response to COVID-19 pandemic? A case study based on spatial-temporal-semantic analytics. Comput Environ Urban Syst. 2021; 90: 101703. https://doi.org/10.1016/j.compenvurbsys.2021.101703 PMID: 34629583 23. Apple. Mobility trends reports. [cited 18 Aug 2020]. https://covid19.apple.com/mobility 24. Bureau of Transportation Statistics. Bikeshare ridership down 44% during COVID-19. 2020 [cited 15 Oct 2020]. https://www.bts.dot.gov/newsroom/bikeshare-ridership-down-44-during-covid-19 25. Angrist JD, Pischke J-S. Mostly harmless econometrics: An empiricist’s companion. Princeton Univer- sity Press. 2008. 26. Bertrand M, Duflo E, Mullainathan S. How much should we trust differences-in-differences estimates? Q J Econ. 2004; 19: 249–275. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 19 / 20 PLOS ONEBike-Sharing platforms and Covid-19 27. Agarwal A. A Comparison of Weekend and Weekday Trav el Behavior Characteristics in Urban Areas. 2004. https://scholarcommons.usf.edu/etd/936 28. Kim D, Im H, Park J. Factors Influencing Travel Behaviors in Bikesharing. Transportation Research Board 91st Annual Meeting. 2012. pp. 1–14. 29. Orvin MM, Fatmi MR. Why individuals choose dockless bike sharing services? Travel Behav Soc. 2021; 22: 199–206. https://doi.org/10.1016/j.tbs.2020.10.001 30. Faghih-Imani A, Eluru N, El-Geneidy AM, Rabbat M, Haq U. How land-use and urban form impact bicy- cle flows: Evidence from the bicycle-sharing system (BIXI) in Montreal. J Transp Geogr. 2014; 41: 306– 314. https://doi.org/10.1016/j.jtrangeo.2014.01.013 31. El-Assi W, Salah Mahmoud M, Nurul Habib K. Effects of built environment and weather on bike sharing demand: a station level analysis of commercial bike sharing in Toronto. Transportation (Amst). 2017; 44: 589–613. https://doi.org/10.1007/s11116-015-9669-z 32. Chen P, Zhou J, Sun F. Built environment determinants of bicycle volume: A longitudinal analysis. J Transp Land Use. 2017; 10: 655–674. https://doi.org/10.5198/jtlu.2017.892 33. Bergantino AS, Intini M, Tangari L. Influencing factors for potential bike-sharing users: an empirical analysis during the COVID-19 pandemic. Res Transp Econ. 2021; 86: 101028. https://doi.org/10.1016/ j.retrec.2020.101028 34. Chibwe J, Heydari S, Faghih Imani A, Scurtu A. An exploratory analysis of the trend in the demand for the London bike-sharing system: From London Olympics to Covid-19 pandemic. Sustain Cities Soc. 2021; 69: 102871. https://doi.org/10.1016/j.scs.2021.102871 35. The New York Times. Coronavirus (Covid-19) Data in the United States. 2021 [cited 3 Nov 2020]. https://github.com/nytimes/covid-19-data 36. Gebhart K, Noland RB. The impact of weather conditions on bikeshare trips in Washington, DC. Trans- portation (Amst). 2014; 41: 1205–1225. https://doi.org/10.1007/s11116-014-9540-7 37. Morton C. The demand for cycle sharing: Examining the links between weather conditions, air quality levels, and cycling demand for regular and casual users. J Transp Geogr. 2020; 88: 102854. https://doi. org/10.1016/j.jtrangeo.2020.102854 38. Ashqar HI, Elhenawy M, Rakha HA. Modeling bike counts in a bike-sharing system considering the effect of weather conditions. Case Stud Transp Policy. 2019; 7: 261–268. https://doi.org/10.1016/j.cstp. 2019.02.011 39. An R, Zahnow R, Pojani D, Corcoran J. Weather and cycling in New York: The case of Citibike. J Transp Geogr. 2019; 77: 97–112. https://doi.org/10.1016/j.jtrangeo.2019.04.016 40. Walk Score. 2020 City & Neighborhood Ranking. 2020 [cited 1 Aug 2020]. https://www.walkscore.com/ cities-and-neighborhoods/ 41. Walk Score. Walk Score Methodology. 2020 [cited 1 Aug 2020]. https://www.walkscore.com/ methodology.shtml 42. Hull A, O’Holleran C. Bicycle infrastructure: can good design encourage cycling? Urban, Plan Transp Res. 2014; 2: 369–406. https://doi.org/10.1080/21650020.2014.955210 43. Rambahan A, Roth C. A More Credible Approach to Parallel Trends. 2022; Working paper. 44. Bilinski A, Hatfield, LA. Nothing to see here? Non-inferiority approaches to parallel trends and other model assumptions. 2018; arXiv preprint arXiv:1805.03273 45. Freyaldenhoven S, Hansen C, Shapiro JM. Pre-event trends in the panel event-study design. American Economic Review. 2019; 109: 3307–8. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283603 April 7, 2023 20 / 20 PLOS ONE
10.1371_journal.pone.0286055	RESEARCH ARTICLE Infant and young child feeding practices and associated socioeconomic and demographic factors among children aged 6–23 months in Ghana: Findings from Ghana Multiple Indicator Cluster Survey, 2017–2018 Samson AkanbongaID Bonny2, Ignitius Ezekiel Lim4, Ilias Mahmud2,5 1, Tanvir Hasan2, Uzzal Chowdhury3, Adrita Kaiser2, Fatema Akter 1 Department of Nutrition and Dietherapy, Holy Family Hospital, Techiman, Ghana, 2 BRAC James P Grant School of Public Health, BRAC University, Dhaka, Bangladesh, 3 Save the Children International, Dhaka, Bangladesh, 4 Department of Comparative Biomedical Sciences, School of Veterinary Medicine, Louisiana State University, Baton Rouge, Louisiana, United States of America, 5 Department of Public Health, College of Public Health and Health Informatics, Qassim University, Al Bukairiyah, Saudi Arabia akanbonsam@yahoo.com Abstract Background Association between poor infant and young child feeding (IYCF) practices and malnutrition in infants and young children (IYC) is well established. Furthermore, appropriate IYCF prac- tices are important during the first 1,000 days of life to ensure optimal health and develop- ment. Understanding IYCF practices and associated socioeconomic and demographic factors will inform interventions to achieve the UN 2030 Sustainable Development Goal (SDG) target to end malnutrition in all forms. Objective This study estimates the prevalence of Minimum Dietary Diversity (MDD), Minimum Meal Frequency (MMF), and Minimum Acceptable Diet (MAD), and examines their association with socioeconomic and demographic characteristics among children aged 6–23 months in Ghana. Method We used data from the Ghana Multiple Indicator Cluster Survey 6 (GMICS6) conducted in 2017–18. Participants were recruited through multi-stage stratified cluster sampling. Infor- mation on caregiver’s self-reported breastfeeding status and 24-hour dietary recall of foods IYC were fed with were collected through face-to-face interviews. We estimated the preva- lence of MDD, MMF and MAD with a 95% confidence interval (CI). We investigated the socioeconomic and demographic determinants of MDD, MMF and MAD using univariate and multivariable logistic regression analyses. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Akanbonga S, Hasan T, Chowdhury U, Kaiser A, Akter Bonny F, Lim IE, et al. (2023) Infant and young child feeding practices and associated socioeconomic and demographic factors among children aged 6–23 months in Ghana: Findings from Ghana Multiple Indicator Cluster Survey, 2017–2018. PLoS ONE 18(6): e0286055. https:// doi.org/10.1371/journal.pone.0286055 Editor: Chandan Kumar, TERI School of Advanced Studies, INDIA Received: April 28, 2022 Accepted: May 8, 2023 Published: June 9, 2023 Copyright: © 2023 Akanbonga et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The data for this study is available and can be accessed by registering and logging into mics.unicef.org/ surveys and downloading the dataset Ghana MICS6 SPSS Datasets.zip. The data for this study is contained in the child dataset labelled ch.sav and the following variables were used; HH1, HH2, BD2, BD3, BD4, BD5, BD6, BD7A, BD7A1, BD7B, BD7C, BD7C, BD7D1, BD7E, BD7E1, BD7X, BD8A, BD8A1, BD8B, BD8B1, BD8C, BD8D, BD8E, BD8F, PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 1 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana BD8G, BD8H, BD8I, BD8J, BD8J1, BD8K, BD8L, BD8M, BD8N, BD8O, BD8P, BD8X, BD9, AN1, AN2, AN4, HH6, HH7, HL4, CAGE, CAGE_6, CAGE_11, CAGED, Stratum, melevel, caretakerdis, ethnicity, chweight, windex5 and PSU. The authors of this study had no special privileges in accessing these datasets which other interested researchers would not have. Findings Among 2,585 IYC aged 6–23 months, MDD, MMF and MAD were estimated as 25.46%, 32.82% and 11.72% respectively. Age of the IYC, educational status of the mothers/primary caregivers, and resident regions were found to have positive associations with MDD, MMF and MAD. In addition, the richest household wealth index and urban area of residence were found to have significant positive associations with MDD. Funding: The authors received no specific funding for this work. Conclusion Competing interests: The authors have declared that no competing interest exist. We report a low prevalence of MDD, MMF and MAD. Efforts to improve IYCF practices among children aged 6–23 months in Ghana should focus on multi-sectorial approaches including increasing access to formal education, income-generating activities and address- ing regional and rural-urban inequity. Introduction Appropriate feeding practices for infants and young children (IYC) reduce the risk of diar- rhoea [1] and acute respiratory infections [2] and prevent anthropometric growth failure [3]. Appropriate feeding of IYC is therefore important particularly during the first 1,000 days of life to ensure optimal health and development [4]. The United Nations Convention on the Rights of the Child states that every child has the right to be raised well and get basic needs, which include a balanced diet, adequate clothing, sufficient shelter, and proper healthcare [5]. United Nations Children’s Fund (UNICEF) and the World Health Organisation (WHO) rec- ommend initiation of breastfeeding within one hour of birth, exclusive breastfeeding for the first six months of life and introduction of nutritionally adequate, age-specific and safe com- plementary (solid, semi-solid or soft) foods at six months along with continued breastfeeding up to two years of age and beyond [6]. To assess infant and young child feeding (IYCF) practices, WHO and UNICEF (2021) developed 16 core indicators. These indicators include ever breastfed, early initiation of breast- feeding, exclusively breastfed for the first two days after birth, exclusive breastfeeding for the first six months, feeding formula and/or animal milk in addition to breast milk under six months, continued breastfeeding till 12–23 months, the introduction of solid, semi-solid or soft foods (ISSSF) at six months, minimum dietary diversity (MDD) during 6–23 months, minimum meal frequency (MMF) during 6–23 months, minimum milk feeding frequency (MMFF) for non-breastfed children aged 6–23 months, minimum acceptable diet (MAD) dur- ing 6–23 months, egg and/or flesh food consumption during 6–23 months, sweet beverage consumption during 6–23 months, unhealthy food consumption during 6–23 months, zero vegetable or fruit consumption during 6–23 months and bottle feeding during 0–23 months. MDD, MMF and MAD are among the important indicators recommended by WHO for assessing complementary feeding practices among IYC aged 6–23 months [7]. Despite the WHO/UNICEF recommendation and the critical benefits related to IYCF, most studies have recorded suboptimal rates of appropriate IYCF practices. Analysis of the UNICEF global database reports global rates of 52.2% and 29.4% for MMF and MDD respec- tively, among children aged 6–23 months [8]. Syntheses of national survey data from 80 low and middle-income countries showed that only 21.3%, 56.2% and 10.1% of these countries had a prevalence of more than 50% in MDD, MMF and MAD, respectively [9]. In sub-Saharan Africa, data from demographic and health surveys between 2010 and 2016 revealed that 41.9% of children aged 6–23 months met MMF, 21% met MDD and 9.8% met PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 2 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana MAD [10]. A community-based survey in Iseyin, a rural community in Oyo state, Nigeria reported the prevalence of MDD at 17.7%, MMF at 46.9% and MAD at 14.9% [11]. A similar study in the rural Damot sore district, Southern Ethiopia, found the prevalence of MDD at 16.5%, MMF at 94.5% and MAD at 16.3% [12]. In Ghana, the prevalence of appropriate IYCF practices for IYC aged 6–23 months is far below the WHO recommendation. As of 2008 in Ghana, the prevalence of MAD for children aged 6–23 months was 29.9% [13], but this declined to 13% in 2014 [14]. In a study based on six public health facilities in Ghana’s Accra metropolitan Assembly, MAD was estimated at 32% [15]. A cross-sectional study in the Kpandai district of Ghana reported the prevalence of MAD among IYC aged 6–23 months at 8.5% [7]. A community-based study in Northern Ghana reported the prevalence of MDD at 35.5%, MMF at 57.3% and MAD at 25.5% among children aged 6–23 months [16]. In 2021, an analysis based on the Ghana Micronutrient Sur- vey of IYC aged 6–23 months revealed that 42% met MDD, 38% met MMF and 14% met MAD [17]. Suboptimal IYCF practices have been found to be directly associated with child- hood malnutrition [18, 19]. Underweight (weight-for-age Z-score <-2SD), wasting (weight- for-height Z-score <-2SD), stunting (height-for-age Z-score <-2SD), micronutrient defi- ciency and overweight (weight-for-age Z-score >2SD) account for more than half of all infant and child mortality in developing countries [20]. Governments and stakeholders need nationally representative accurate and timely evidence to understand the current situation of IYCF practices for IYC aged 6–23 months and its associ- ated factors. Knowing the factors influencing IYCF practices will help shape the development and implementation of evidence-based and effective interventions. In this context, our study aimed to provide information about the prevalence of MDD, MMF and MAD among IYC aged 6–23 months and their association with socioeconomic and demographic characteristics of the IYC in Ghana using nationally representative data from the most recent Ghana Multiple Indicator Cluster Survey 6 (GMICS6), conducted in 2017–18. The findings of this study are expected to assist key policymakers and programme implementers to identify areas requiring further improvements and design need-based and evidence-driven interventions. This will help improve appropriate IYCF practices for IYC aged 6–23 months in Ghana and in the long run contribute towards the target of ending malnutrition in all forms under the UN 2030 sec- ond Sustainable Development Goal (SDG) of zero hunger. Materials and methods Study design A secondary data analysis of the Ghana Multiple Indicator Cluster Survey 6 (GMICS6) was done to estimate the prevalence of MDD, MMF and MAD and investigate its associated socio- economic and demographic factors. The GMICS6 was carried out from October 2017 to Janu- ary 2018 by the Ghana Statistical Service (GSS) in partnership with the Ministry of Health, Ministry of Education, Ministry of Sanitation and Water Resources, Ministry of Gender, Chil- dren and Social Protection, Ghana Health Service and the Ghana Education Service [21] with technical support from UNICEF [21]. The global MICS programme is an international multi-purpose household survey initiated by UNICEF in the mid-1990s with the goal of assisting countries to collect internationally comparable data on varied characteristics of women and children such as breastfeeding status and dietary practices. This data is used for evidence-based national development planning and initiatives and for monitoring progress made towards national goals and global commitments such as the SDGs [22]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 3 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana Study setting and participants Ghana is a middle-income West African country which currently has 16 administrative regions. However, at the time GMICS6 was conducted, there were 10 administrative regions namely: Western, Central, Greater Accra, Volta, Eastern, Ashanti, Brong Ahafo, Northern, Upper East and Upper West [21]. This study utilised data from the mother/caregiver-child pair about breastfeeding and die- tary practices of 2,585 IYC aged 6–23 months. Sampling The GMICS6 employed a multi-stage stratified cluster sampling approach using a sampling frame based on the 2010 Population and Housing Census (PHC) [21]. GMICS6 selected cen- sus enumeration areas (primary sampling units) from both urban and rural areas in each of the then 10 regions of Ghana by using systematic probability sampling proportional to size. Further details of the sampling are reported in the 2017–18 MICS report [21]. The overall sample size of the GMICS6 was 13,202 households with a response rate of 99.4% [21]. All children under 5 years of age living in the sampled households were included in the survey. A total of 8,903 children under five years of age were listed in these households with a response rate of 99.7%. Out of these children, 2,585 were in the age range of 6–23 months and were included in this study. The sample flow details are outlined in Fig 1. Survey instruments Mother’s or caregiver’s self-reported breastfeeding status and all foods that children under five years of age were fed with and frequency of feeding within the last 24 hours before the survey alongside socioeconomic and demographic characteristics were collected using the under-5 questionnaire of the GMICS6. Data collection Data for the GMICS6 was collected via face-to-face interviews using Computer-Assisted Per- sonal Interviewing (CAPI) technique. A training of 30 days for field investigators was conducted from 10th September to 10th October 2017. The content of this training included interviewing techniques, contents of the Fig 1. Sample flow details. https://doi.org/10.1371/journal.pone.0286055.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 4 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana questionnaires, CAPI application, skills in asking questions and mock interviews, 5 days of field practice and a one-day full pilot survey in localities around Winneba in the Central Region. Mandatory re-interviewing was implemented in one household per cluster using both random and purposive sampling techniques. Daily observations of the interviewer’s skills and performance were done [21]. Ethical consideration We utilised secondary data from GMICS6 whose ethical considerations have been published elsewhere [21]. Measures/variables Outcome variables. Minimum Dietary Diversity (MDD): IYC aged 6–23 months who were fed foods and beverages from at least five of eight food groups during the previous day. The eight food groups are: breastmilk; dairy products (milk, infant formula, yogurt and cheese); grains, roots, tubers and plantain; legumes or pulses (beans, peas and lentils), nuts and seeds; flesh foods (meat, fish, poultry and liver/organ meats); eggs; vitamin-A rich fruits and vegetables and other fruits and vegetables [23]. Minimum Meal Frequency (MMF): IYC aged 6–23 months who were fed foods and bever- ages 2 times and 3 times for 6–8 months and 9–23 months, respectively, and for non-breastfed IYC aged 6–23 months, 4 times including milk feeds and at least one solid, semi-solid or soft food during the previous day [23]. Minimum Milk Feeding Frequency (MMFF) for non-breastfed children: Non-breastfed IYC aged 6–23 months who were fed with at least two milk feeds (infant formula, animal milk or yogurt) during the previous day [23]. Minimum Acceptable Diet (MAD): Breastfed IYC aged 6–23 months who received MDD and MMF per age during the previous day and non-breastfed children aged 6–23 months who received MDD, MMF per age and MMFF during the previous day [23]. Independent variables. The following socioeconomic and demographic characteristics of IYC aged 6–23 months were included in the analysis: sex of the child; place of residence; administrative region (Western, Central, Greater Accra, Volta, Eastern, Ashanti, Brong Ahafo, Northern, Upper East and Upper West); age in months (6–8, 9–11, 12–17 and 18–23); moth- er’s or caretaker’s educational attainment (None, primary, JSS/JHS/middle, SSS/SHS/Second- ary and higher); household ethnicity (Akan, Ga/Dangme, Ewe, Guan, Gruma, Mole Dagbani, Grusi, Mande and Others) and household wealth index which captures the underlying long- term wealth of the household ranked according to the wealth score of the household based on urban and rural factor scores regressed on assets owned by that household and finally divided into 5 equal parts or quintiles (poorest, poorer, middle, richer and richest). Statistical analyses The GMICS6 dataset was accessed at mics.unicef.org. Each IYCF indicator or outcome vari- able (MDD, MMF and MAD) was dichotomised into a binary variable as “yes = 1” for a child who meets the indicator and “no = 0” for a child who does not meet the indicator. Frequencies, percentages/prevalences and means (± standard deviation) were used to describe the socioeco- nomic and demographic characteristics and outcome variables. Firstly, the Chi-square test was done to investigate the association between socioeconomic and demographic variables as well as the outcome variables. Socioeconomic and demographic variables that showed statistically significant (p � 0.05) association with the outcome variables were also investigated using uni- variate (unadjusted) logistic regression. Statistically significant (p � 0.05) socioeconomic and PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 5 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana demographic variables in the univariate models were then included in the multivariable logis- tic regression model to get the adjusted association of selected socioeconomic and demo- graphic factors after controlling for other potential confounders. The regression outputs were reported in the form of Odds Ratios (ORs) with 95% confidence intervals (CIs). The level of significance was determined at p-value � 0.05. The data were analysed using Stata/SE version 17.0 (Stata Corporation, LLC, College Station, Texas, USA). Stata syntax developed for calcu- lating IYCF indicators [23] was customised and used for the analysis. The Stata command ‘svy- set’ was used to declare the complex sampling design of the data, while all estimations were performed by using the complex survey-specific command ‘svy.’ Results Socioeconomic and demographic characteristics of infants and young children aged 6–23 months Table 1 shows the socioeconomic and demographic characteristics of the IYC. The mean age of the sampled IYC was 14.3 months (±5.3) and over half (50.6%) were males. Among the IYC, 33.9% and 32.1% of them were 18–23 months and 12–17 months respectively while 56.6% were residing in rural areas. The Ashanti region recorded the highest percentage of IYC (23.9%) while the Upper West region recorded the lowest percentage of IYC (2.3%). In terms of education, about one-fifth (22.9%) of mothers or caregivers of the IYC had no formal educa- tion and only 5.9% had above secondary-level education. Majority of the IYC were from the Akan ethnic group (46.4%) while the Mande ethnic group recorded the least IYC (0.1%). Over a fifth (21.8%) of the IYC were from the poorest households and 19.8% were from the richest households. Prevalence of minimum dietary diversity among infants and young children aged 6–23 months in Ghana Over a quarter (25.4%, 95% CI: 22.98, 27.94) of IYC in this study met MDD (Table 2). Almost equal proportions of male and female IYC achieved MDD, 24.7% and 26.2% respectively (Table 3). The majority (33.6%) of the IYC who achieved MDD were in the 12–17 months age group. Almost one-third of the IYC (31.7%) from urban areas achieved MDD whilst a fifth (20.6%) IYC from rural areas also achieved MDD. The Central region of Ghana recorded the highest prevalence of IYC who achieved MDD (34.7%) whilst the Upper West region recorded the least prevalence of MDD (13.9%). Nearly a fifth (18.4%) of IYC whose mothers/caregivers had no formal education achieved MDD whilst over half (52.4%) of IYC whose mothers/care- givers had above secondary-level education achieved MDD. The majority (94.6%) of IYC from the Mande ethnic group did not achieve MDD. In terms of household wealth quintile, nearly a fifth (18.4%) of IYC from the poorest households achieved MDD compared to almost half (46.7%) of IYC in the richest households who achieved MDD. The staples (grains, roots, tubers and plantain) were the food group that most (79%) IYC were fed with during the previous day (Fig 2). The majority of IYC (77%) were also fed breastmilk during the previous 24 hours. Consumption of other food groups by IYC was all below 50% with legumes or pulses, nuts and seeds recording a consumption of 16%. Prevalence of minimum meal frequency among infants and young children aged 6–23 months in Ghana The percentage of IYC who achieved MMF in the 24 hours preceding the survey was found to be 32.8% (95% CI: 29.99, 35.66) (Table 2). Approximately, 3 in every 10 male IYC achieved PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 6 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana Table 1. Socioeconomic and demographic characteristics of infants and young children aged 6–23 months (n = 2,585). Socioeconomic and demographic characteristics Frequency Percentage (%) Sex Male Female Age (months) 6–8 9–11 12–17 18–23 Area of residence Urban Rural Administrative region Western Central Greater Accra Volta Eastern Ashanti Brong Ahafo Northern Upper East Upper West Mother’s/caretaker’s educational level None Primary JSS/JHS/Middle SSS/SHS/Secondary Higher Household ethnicity Akan Ga/Dangme Ewe Guan Gruma Mole Dagbani Grusi Mande Others Household wealth index quintile Poorest Poorer Middle Richer Richest https://doi.org/10.1371/journal.pone.0286055.t001 1,309 1,276 501 377 830 877 1,121 1,464 293 248 242 212 297 620 238 293 83 59 577 554 999 315 140 1,201 180 283 110 108 441 46 3 213 564 503 481 523 514 50.62 49.38 19.37 14.58 32.11 33.94 43.37 56.63 11.34 9.58 9.37 8.21 11.47 23.99 9.20 11.34 3.20 2.30 22.92 21.62 37.13 12.43 5.90 46.44 6.94 10.93 4.29 4.18 17.06 1.81 0.10 8.25 21.84 19.46 18.59 20.22 19.89 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 7 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana Table 2. Prevalence of minimum dietary diversity, minimum meal frequency, minimum milk feeding frequency and minimum acceptable diet among infants and young children aged 6–23 months (n = 2,585). IYCF indicator Minimum Dietary Diversity Minimum Meal Frequency Minimum Milk Feeding Frequency Minimum Acceptable Diet https://doi.org/10.1371/journal.pone.0286055.t002 Frequency Prevalence (%) 658 848 88 303 25.46 32.82 15.91 11.72 95% CI 22.98, 27.94 29.99, 35.66 11.89, 19.93 9.72, 13.72 MMF (31.3%) and 34.3% of female IYC had MMF (Table 4). IYC aged 6–8 months achieved the highest prevalence of MMF (53.7%) while 4 in every 5 IYC (80.2%) in the age group of 18– 23 months did not achieve MMF. There was not much difference in the prevalence of MMF among IYC from urban and rural areas (32.1% and 33.7% respectively). The highest prevalence of MMF was recorded in the Central region (44.5%) while the Brong Ahafo region recorded the lowest prevalence (21.8%). Among the ethnic groups, the highest prevalence of MMF (35.5%) was observed among Mole Dagbani IYC. Contrastingly, none of the IYC from the Mande ethnic group achieved MMF. The prevalence of MMF increased with mother’s/caregiv- er’s educational status. Less than 30% (28.4%) and over half (55.6%) of IYC whose mothers or caregivers had no formal education and above secondary-level education achieved MMF respectively. The prevalence of MMF was higher among IYC from the richest households com- pared to those from the poorest households (39.5% versus 28.2% respectively). Prevalence of minimum acceptable diet among infants and young children aged 6–23 months in Ghana The prevalence of MAD 24 hours prior to the survey was 11.7% (95% CI: 9.7–13.7) (Table 2). Among males IYC, 10.6% achieved MAD while 12.8% of female IYC achieved MAD (Table 5). The highest prevalence of MAD was recorded among the age group of 12–17 months (15.1%) and only 7.6% of IYC in the age group of 6–8 months achieved MAD. About 13 in every 100 IYC (13.5%) from urban areas and about 10 in every 100 (10.3%) IYC from rural areas achieved MAD. In the Upper West region, 4.7% of IYC achieved MAD while 83.8% of IYC from the Ashanti region did not achieve MAD. Among IYC whose mothers or caregivers had no formal education, 8.2% achieved MAD and 33.6% of IYC whose mothers or caregivers had above secondary-level education achieved MAD. About 16 in every 100 IYC (16.8%) from the Mole Dagbani ethnic group achieved MAD while all IYC from the Mande ethnic group did not achieve MAD. The majority of IYC from the richest households achieved MAD compared to their poorest counterparts (18.5% and 9.9% respectively). Socioeconomic and demographic determinants of feeding practices for infant and young children aged 6–23 months Minimum dietary diversity. Table 3 presents socioeconomic and demographic predictors of MDD among IYC aged 6–23 months. IYC from the age groups of 9–11 months (AOR: 2.39; 95% CI: 1.32, 4.33), 12–17 months (AOR: 4.48; 95% CI: 2.69, 7.46) and 18–23 months (AOR: 3.54; 95% CI: 2.18, 5.75) were 2.39, 4.48 and 3.54 times more likely to achieve MDD respec- tively compared to IYC aged 6–8 months. IYC having a mother or caregiver with secondary education (AOR: 2.21; 95% CI: 1.34, 3.65) and with above secondary education (AOR: 2.19; 95% CI: 1.11, 4.33) had 2 times the odds of achieving MDD compared to IYC with mothers or caregivers with no formal education. IYC from the richest households (AOR: 2.28; 95% CI: PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 8 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana Table 3. Socioeconomic and demographic predictors of minimum dietary diversity among infants and young children aged 6–23 months. Socioeconomic and demographic characteristics Achieved MDD (%) Chi-squared test (p-value) Crude Odds Ratio (95% CI) Adjusted Odds Ratio (95% CI) Sex Male Female Age (months) 6–8 9–11 12–17 18–23 Area of residence Urban Rural Administrative region Western Central Greater Accra Volta Eastern Ashanti Brong Ahafo Northern Upper East Upper West Mother’s/caregiver’s educational level None Primary JSS/JHS/Middle SSS/SHS/Secondary Higher Household ethnicity Akan Ga/Dangme Ewe Guan Gruma Mole Dagbani Grusi Mande Others Household wealth index quintile Poorest Poorer Middle Richer Richest p-value�0.05, p-value�0.01 and p-value�0.001 https://doi.org/10.1371/journal.pone.0286055.t003 24.74 26.20 10.24 22.55 33.60 27.69 31.71 20.67 29.34 34.74 31.83 14.32 20.80 29.92 18.52 21.47 19.50 13.90 18.41 18.24 24.64 41.69 52.48 26.49 24.66 22.80 19.40 20.45 27.35 18.75 5.33 27.48 18.49 18.47 23.89 20.16 46.79 0.6107 �0.001 �0.001 �0.001 �0.001 0.7894 �0.001 1.00 2.55 (1.46, 4.45) 4.44 (2.67, 7.33)* 3.36 (2.14, 5.27)* 1.00 2.39 (1.32, 4.33) 4.48 (2.69, 7.46)* 3.54 (2.18, 5.75)* 1.00 0.56 (0.43, 0.73)* 1.00 0.85 (0.63, 1.14) 1.00 1.28 (0.81, 2.02) 1.12 (0.67, 1.88) 0.40 (0.20, 0.80)** 0.63 (0.40, 0.99)* 1.03 (0.66, 1.60) 0.55 (0.33, 0.91)* 0.66 (0.42, 1.02) 0.58 (0.36, 0.93)* 0.39 (0.23, 0.65)*** 1.00 0.99 (0.61, 1.59) 1.45 (1.03, 2.05)* 3.17 (2.07, 4.86) 4.90 (2.44, 9.83)* 1.00 1.41 (0.90, 2.23) 0.64 (0.38, 1.06) 0.51 (0.26, 0.97)* 0.71 (0.45, 1.11) 0.91 (0.59, 1.40) 0.56 (0.32, 0.97)* 0.92 (0.56, 1.51)* 0.74 (0.45, 1.22) 0.49 (0.28, 0.84)* 1.00 0.85 (0.54, 1.33) 1.19 (0.78, 1.80) 2.21 (1.34, 3.65)** 2.19 (1.11, 4.33)* 1.00 1.00 (0.67, 1.48) 1.38 (0.93, 2.05) 1.11 (0.71, 1.76) 3.88 (2.58, 5.82)* 1.00 0.86 (0.56, 1.33) 1.13 (0.73, 1.74) 0.76 (0.45, 1.30) 2.28 (1.33, 3.90) PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 9 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana Fig 2. Percent of IYC fed with various food groups during the previous day. https://doi.org/10.1371/journal.pone.0286055.g002 1.33, 3.90) had over twice the odds of achieving MDD compared to IYC from the poorest households. Compared to the Western region, IYC from the Volta (AOR: 0.51; 95% CI: 0.26, 0.97), Brong Ahafo (AOR: 0.56; 95% CI: 0.32, 0.97), Northern (AOR: 0.92; 95% CI: 0.56, 1.51) and Upper West (AOR: 0.49; 95% CI: 0.28, 0.84) regions respectively had 49%, 44%, 8% and 51% lower odds of achieving MDD. IYC from rural areas had 44% reduced odds of achieving MDD compared to IYC from urban areas. Minimum meal frequency. Our multivariable logistic regression analysis suggests that IYC with mothers or caregivers having secondary education (AOR: 1.92; 95% CI: 1.25, 2.95) and above secondary education (AOR: 3.23; 95% CI: 1.70, 6.14) were 1.92 and 3.23 times more likely to achieve MMF respectively compared to the IYC with mothers or caregivers with no formal education. The odds of achieving MMF among IYC from the Central region (AOR: 1.74; 95% CI: 1.12, 2.71) was 1.74 times those from the Western region (Table 4). On the other hand, IYC in the age groups of 9–11 months (AOR: 0.42; 95% CI: 0.29, 0.60), 12–17 months (AOR: 0.41; 95% CI: 0.29, 0.58) and 18–23 months (AOR: 0.20; 95% CI: 0.13, 0.29) had 58%, 59% and 80% reduced odds of achieving MMF respectively compared to IYC aged 6–8 months. IYC from the Greater Accra region (AOR: 0.58; 95% CI: 0.36, 0.92) had a 42% reduced odds of achieving MMF compared to IYC from the Western region. Minimum acceptable diet. Our analyses suggest that IYC aged 9–11 months (AOR: 1.86; 95% CI: 1.01, 3.41) and 12–17 months (AOR: 2.13; 95% CI: 1.18, 3.83) were 1.86 and 2.13 times more likely to achieve MAD respectively compared to IYC in the age group of 6–8 months (Table 5). Having a mother or caregiver with secondary education (AOR: 3.19; 95% CI: 1.66, 6.10) and with above secondary education (AOR: 5.24; 95% CI: 2.04, 13.49) also put IYC at about 3 and 5 times the odds of achieving MAD compared to those with no education. IYC from the Central region had 1.93 times the odds of achieving MAD (AOR: 1.93; 95% CI: 1.09, 3.42) compared to those from the Western region. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 10 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana Table 4. Socioeconomic and demographic predictors of minimum meal frequency among infants and young children aged 6–23 months. Socioeconomic and demographic characteristics Achieved MMF (%) Chi-squared test (p-value) Crude Odds Ratio (95% CI) Adjusted Odds Ratio (95% CI) 1.00 0.44 (0.31, 0.62)* 0.44 (0.31, 0.63)* 0.21 (0.15, 0.31)* 1.00 0.42 (0.29, 0.60)* 0.41 (0.29, 0.58)* 0.20 (0.13, 0.29)* 1.00 1.77 (1.14, 2.78)* 0.72 (0.45, 1.15) 1.51 (0.95, 2.42) 0.84 (0.48, 1.45) 1.35 (0.87, 2.07) 0.62 (0.36, 1.06) 1.05 (0.68, 1.63) 0.75 (0.45, 1.26) 0.88 (0.58, 1.35) 1.00 0.96 (0.65, 1.41) 1.23 (0.92, 1.64) 1.66 (1.13, 2.45)* 3.15 (1.72, 5.79)*** 1.00 1.74 (1.12, 2.71)* 0.58 (0.36, 0.92)* 1.58 (1.00, 2.50) 0.77 (0.42, 1.41) 1.40 (0.92, 2.12) 0.61 (0.35, 1.04) 1.18 (0.73, 1.93) 0.81 (0.47, 1.38) 0.86 (0.55, 1.36) 1.00 0.94 (0.62, 1.40) 1.20 (0.85, 1.69) 1.92 (1.25, 2.95)* 3.23 (1.70, 6.14)*** 0.3042 �0.001 0.6014 0.0013 �0.001 0.1125 0.1076 Sex Male Female Age (months) 6–8 9–11 12–17 18–23 Area of residence Urban Rural Administrative region Western Central Greater Accra Volta Eastern Ashanti Brong Ahafo Northern Upper East Upper West Mother’s/caregive’s educational level None Primary JSS/JHS/Middle SSS/SHS/Secondary Higher Household ethnicity Akan Ga/Dangme Ewe Guan Gruma Mole Dagbani Grusi Mande Others Household wealth index quintile Poorest Poorer Middle Richer Richest p-value�0.05, p-value�0.01 and *p-value�0.001 https://doi.org/10.1371/journal.pone.0286055.t004 31.31 34.37 53.73 33.66 33.66 19.74 33.70 32.15 31.17 44.52 24.54 40.67 27.49 37.89 21.82 32.27 25.48 28.55 28.48 27.66 32.79 39.83 55.66 34.60 27.10 34.66 28.49 33.53 35.51 26.78 0.00 22.89 28.04 33.27 29.82 33.68 39.58 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 11 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana Table 5. Socioeconomic and demographic predictors of minimum acceptable diet among infants and young children aged 6–23 months. Socioeconomic and demographic characteristics Achieved MAD (%) Chi-squared test (p-value) Crude Odds Ratio (95% CI) Adjusted Odds Ratio (95% CI) Sex Male Female Age (months) 6–8 9–11 12–17 18–23 Area of residence Urban Rural Administrative region Western Central Greater Accra Volta Eastern Ashanti Brong Ahafo Northern Upper East Upper West Mother’s/caregiver’s educational level None Primary JSS/JHS/Middle SSS/SHS/Secondary Higher Household ethnicity Akan Ga/Dangme Ewe Guan Gruma Mole Dagbani Grusi Mande Others Household wealth index quintile Poorest Poorer Middle Richer Richest p-value �0.05, p-value�0.01 and p-value�0.001 https://doi.org/10.1371/journal.pone.0286055.t005 10.61 12.87 7.60 13.49 15.17 10.06 13.55 10.32 11.59 20.14 7.17 10.07 6.92 16.14 5.48 12.85 7.61 4.71 8.23 7.48 10.67 19.25 33.56 11.94 6.88 10.47 10.88 6.81 16.86 5.80 0.00 10.06 9.90 7.96 12.15 10.25 18.50 0.2871 0.0286 0.1147 �0.001 �0.001 0.1662 0.0160 1.00 1.90 (1.05, 3.44) 2.17 (1.18, 3.40)* 1.36 (0.82, 2.26) 1.00 1.86 (1.01, 3.41)* 2.13 (1.18, 3.83)* 1.33 (0.77, 2.31) 1.00 1.92 (1.10, 3.36)* 0.59 (0.26, 1.35) 0.85 (0.40, 1.81) 0.57 (0.27, 1.20) 1.47 (0.80, 2.70) 0.44 (0.21, 0.94)* 1.12 (0.58, 2.18) 0.63 (0.31, 1.27) 0.38 (0.17, 0.84)* 1.00 0.90 (0.49, 1.64) 1.33 (0.85, 2.08) 2.66 (1.51, 4.68) 5.64 (2.20, 14.41)* 1.00 1.93 (1.09, 3.42)* 0.37 (0.16, 0.88)* 1.00 (0.48, 2.05) 0.61 (0.27, 1.36) 1.37 (0.74, 2.53) 0.44 (0.20, 0.96)* 1.50 (0.69, 3.30) 0.68 (0.32, 1.48) 0.41 (0.17, 0.97)* 1.00 0.95 (0.51, 1.78) 1.42 (0.83, 2.41) 3.19 (1.66, 6.10) 5.24 (2.04, 13.49) 1.00 0.79 (0.47, 1.31) 1.26 (0.74, 2.15) 1.04 (0.57, 1.90) 2.06 (1.17, 3.63)* 1.00 0.65 (0.38, 1.13) 1.05 (0.57, 1.94) 0.71 (0.34, 1.51) 1.01 (0.53, 1.93) PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 12 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana On the other hand, IYC from the Greater Accra (AOR: 0.37; 95% CI: 0.16, 0.88), Brong Ahafo (AOR: 0.44; 95% CI: 0.20, 0.96) and Upper West (AOR: 0.41; 95% CI: 0.17, 0.97) regions had 63%, 56% and 59% reduced odds of achieving MAD respectively compared to IYC from the Western region. Discussion We investigated the prevalence and socioeconomic and demographic determinants of MDD, MMF and MAD based on the GMICS 6 (2017/18). The prevalence of MDD, MMF and MAD was found to be 25.4%, 32.8% and 11.7% respectively. Socioeconomic and demographic charac- teristics having a positively significant association with MDD, MMF and MAD were the age of the IYC, educational status of the mothers/primary caregivers, and resident regions. In addition, the household wealth index was also found to have a positive significant association with MDD. The percentage of IYC who achieved MDD in this study was slightly lower than 29.4% which was reported through an analysis of UNICEF global database [8] but comparable to the estimate of 25.1% based on the DHS data of 32 sub-Saharan African countries including Ghana from 2010 to 2020 [24]. The percentage of IYC who achieved MDD in the 24 hours pre- ceding the survey in this study is higher than 21% found in a similar study using data from Ethiopia 2016 DHS [25] and 11.7% using data from Ethiopia 2019 DHS [26]. The percentage of IYC who achieved MDD in the 24 hours preceding the survey in this study is also higher than that reported in cross-sectional studies including 17.7% in Nigeria [11] and 16.5% in Ethiopia [12] but lower than those reported in Ghana including, 35.5% [16], 32% [15] and 34.8% [27]. The difference might be due to the difference in sample size and the season in which a particular survey was conducted. Also, this study used the WHO and UNICEF revised definition of MDD (� five out of eight food groups) considering breastmilk as a distinct food group. Other previous studies however used the old definition (� four out of seven food groups). The use of the revised definition has been shown to result in a decrease in the prevalence of MDD [28]. This should be kept in mind in comparing the prevalence of MDD reported in this study with that of previous studies. However, considering the results of this study, about a quarter of Ghanaian IYC aged 6–23 months were fed foods from at least five food groups during the previous day. This means about 75% of the IYC in this study are at risk of not meeting nutrient requirements which are essential to their physical and mental development. Even though quantities of the foods IYC were fed with were not included in this study and dietary intake was assessed on a one-day basis which may not reflect usual intake, nonetheless MDD has been shown to be a good pre- dictor of nutrient adequacy and nutritional status [29]. This study also found a low intake of egg and flesh food among the IYC during the previous day (20% and 45% respectively). Many countries have also recorded similar low consumption of egg or flesh food [8]. This is worrisome given the fact that WHO recommends that IYC aged 6–23 months be fed with meat, poultry, fish or egg on a daily basis or as frequently as pos- sible [23] to ensure optimal growth [30]. The percentage of IYC who achieved MMF in the 24 hours preceding the survey in this study is lower than 41.9% found in sub-Saharan Africa using DHS data [10] but similar to 33.6% found in a study in Nigeria [31]. Other community or facility-based studies in Ghana reported MMF at 39.5% [7], 46% [13], 38% [17], 58.2% [32] and 57.3% [16]. The difference might be attributed to the nationally representative nature of the data used in this study which allows for a more representative and heterogeneous sample and more reliable prevalence. The lower MMF in this study may be due to a lack of awareness among mothers or caregivers about the number of times IYC should be fed per day. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 13 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana Ghanaian IYC aged 6–23 months in this study who achieved MAD compares favorably with the average of 9.8% found in sub-Saharan Africa using DHS data [10] and 9.2% in South- western Nigeria [31]. However, the percentage of IYC who achieved MAD in this study is also lower than that found in some community or facility-based studies including 27.8% [32], 25.5% [16] and 38.9% [33] in Northern Ghana, 33% in South Kivu, Democratic Republic of Congo [34], 14% in Southern Ghana [17] and 16.3% in Southern Ethiopia [12]. The reason may be that community or facility-based studies use a relatively more homogenous sample of participants compared to the data used in this study which is nationally representative. The low MAD (11.72%) in this study means that more than 80% of IYC were at increased risk of not meeting their micronutrient and meal frequency adequacy because MAD is a composite indicator of MDD and MMF which depicts both quality and quantity of food given to IYC. This low MAD may play a role in childhood malnutrition in Ghana. For instance, it has been observed that between 30% and 50% of all children aged 6–59 months in Ghana were mal- nourished [35]. This could negatively affect the physical growth and neurodevelopment of IYC in Ghana. This is because it has been noted that IYC who do not meet MDD, MMF or MAD are at higher risk of malnutrition [36, 37] which is significantly associated with neurodevelop- ment impairment [38]. The association of socioeconomic and demographic characteristics with MDD, MMF and MAD found in this study may help inform specific interventions to improve IYCF practices among children aged 6–23 months in Ghana. IYC in the age groups of 9–11, 12–17 and 18–23 months were 2.39, 4.48 and 3.54 times more likely to achieve MDD than those aged 6–8 months. Similarly, IYC aged 9–11 and 12–17 months were 1.86 and 2.13 times more likely to achieve MAD than those aged 6–8 months respectively. This is similar to that found in Dabat District, Ethiopia [39], districts of Bopa and Houeyogbe, Benin [40], India [41], Ghana [13], Southern Benin [42] and Northern Ghana [16]. The reason may be that younger infants and children are given only cereal-based porridge which does not contain diverse nutrients but older IYC are given family foods containing diverse nutrients making them more likely to meet MDD and MAD. Behaviour change communication targeting IYCF among children aged 6–23 months in Ghana should focus on increasing the production of locally available foods and how to modify various family foods in terms of consistency such as blending fruits and vegetables and how to enrich traditional porridges made from cereals with other foods such as groundnut, soybean, egg and milk to ensure diverse foods for younger IYC. Older IYC were also less likely to receive MMF than younger IYC (6–8 months). This means that atten- tion is not given to IYC aged 9–23 months in terms of feeding frequency as compared to their younger counterparts. Sensitization is required for mothers/caretakers to make them under- stand that older IYC needs frequent meals to meet their nutrient requirements. IYC of mothers/caregivers with secondary and/or higher education were more likely to achieve MDD, MMF and MAD than IYC of mothers/caregivers with no formal education. These results corroborate earlier studies such as that found in Ethiopia [26, 43–46], Ghana [47], India [41], South Kivu, Democratic Republic of Congo [34], in a systematic review [48] and Rwanda [49]. This may be due to the fact that educated mothers/caretakers can access information about IYCF through various mediums including the print medium. Educated mothers/caretakers also have a higher chance to be gainfully employed and can afford diverse foods for their IYC. Interventions to improve IYCF practices among children aged 6–23 months in Ghana should focus on improving access to formal education, especially for adoles- cent girls. IYC from the richest households were 2.28 times more likely to achieve MDD than their counterparts from the poorest households. This finding is in agreement with that found in Pakistan [50], Ethiopia [44, 51, 52] and in Indonesia [53]. This may be due to the fact that the PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 14 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana wealthiest households are able to afford diverse foods for their IYC as compared to the poorest households. Efforts to improve IYCF practices among children aged 6–23 months in Ghana should focus on providing access to income-generating activities to improve the economic conditions of households and hence improve MDD. IYC from the Central region were more likely to achieve MMF and MAD than those from the Western region. Also, IYC from Volta, Brong Ahafo, Northern and Upper West regions were less likely to achieve MDD compared to IYC from the Western region, IYC from Greater Accra region were less likely to achieve MMF compared to those from Western region and those from Greater Accra, Brong Ahafo and Upper West regions were less likely to achieve MAD compared to those from the Western region. Region of residence as a significant factor to meeting IYCF practices has also been found in previous studies in Ghana [13, 47] and India [41]. This may be due to geographical differences and the kinds of foods produced in a particu- lar region or other dynamics of a particular region. Surprisingly, in this study, only MDD was significantly associated with the area of resi- dence. The urban area of residence has however been shown to be associated with IYCF prac- tices in previous studies in Ethiopia [25, 26, 39], and in South Kivu, Democratic Republic of Congo [34]. In this study, the lack of significant association between MMF and MAD with the area of residence may be due to the fact that the gap between rural and urban areas in Ghana in terms of feeding frequency is being narrowed while dietary diversity remains widely hetero- geneous. Another possible explanation is that rural residents may be increasingly moving away from feeding their IYC with locally available diverse foods to commercially packaged foods. More efforts should be made to further limit the gap between rural and urban areas by improving road networks and other infrastructure. Strengths and limitations of the study This study utilized the most recent nationally representative GMICS6 data with a large sample size (13,202 households with 2,585 IYC aged 6–23 months) and high response rate (99.4%). This study used the WHO and UNICEF definition of MMD which has been recommended for measuring intermediate outcome indicators for tracking progress towards the 2025 global nutrition targets [54]. The data used in this study has a limitation of having only one day of diet recall per child, which may not be representative of the day-to-day dietary intake and also prone to recall bias. In addition, the measure does not take into consideration the amount of food consumed and thus it does not completely measure nutrient adequacy. Nonetheless this method has been shown to be a good predictor of nutrient adequacy and nutritional status [29]. Lastly, the analysis was limited to variables in the dataset and only associations and no causality can be inferred due to the cross-sectional nature of the survey. Conclusion and recommendation The prevalence of MDD, MMF and MAD among Ghanaian IYC aged 6–23 months in this study was low (25.4%, 32.8% and 11.7% respectively). Socioeconomic and demographic char- acteristics of IYC having positively significant associations with MDD were the age of IYC, educational status of mothers/caretakers, household wealth index, administrative region and area of residence. Additionally, the age of IYC, educational status of mothers/caretakers and administrative region were socioeconomic and demographic characteristics of IYC that showed significant positive associations with MMF and MAD. Surprisingly, unlike other pre- vious studies, only MDD had a significant association with the area of residence in this study. Based on the evidence generated in this study, we recommend that interventions towards MDD, MMF and MAD in Ghana should focus on improving access to formal education, PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 15 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana income-generating activities and strengthening behaviour change communication using avail- able channels in the communities to improve awareness and knowledge of IYCF. Acknowledgments We wish to thank the global Multiple Indicator Cluster Survey (MICS) authority for granting us access to its publicly available Ghana MICS 6 (GMICS6) dataset for this study. This study was conducted for the partial fulfillment of the first authors’ MPH degree from the BRAC James P. Grant School of Public Health. Dr. Tanvir Hasan and Dr. Ilias Mahmud supervised the project with assistance from the other co-authors. Author Contributions Conceptualization: Samson Akanbonga, Tanvir Hasan. Data curation: Samson Akanbonga, Uzzal Chowdhury. Formal analysis: Samson Akanbonga, Uzzal Chowdhury, Ignitius Ezekiel Lim. Methodology: Samson Akanbonga. Project administration: Tanvir Hasan, Ilias Mahmud. Resources: Samson Akanbonga, Adrita Kaiser, Fatema Akter Bonny. Supervision: Tanvir Hasan, Ilias Mahmud. Writing – original draft: Samson Akanbonga, Uzzal Chowdhury, Adrita Kaiser, Fatema Akter Bonny. Writing – review & editing: Samson Akanbonga, Tanvir Hasan, Ignitius Ezekiel Lim, Ilias Mahmud. References 1. Ogbo F. A., Ogeleka K. A. P., Woolfenden S., Page A., Eastwood J. and the Global Child Health Research interest group (2017). Infant feeding practices and diarrhoea in sub-Saharan African coun- tries with high diarrhoea mortality. PLos ONE, 12(2):e0171792. https://doi.org/10.1371/journal.pone. 0171792 PMID: 28192518 2. Ahmed K. Y., Page A., Arora A., Ogbo F. A. and the Global Maternal and Child Health Research collab- oration (2020). Associations between infant and young child feeding practices and acute respiratory infection and diarrhoea in Ethiopia: A propensity score matching approach. PLoS ONE, 15(4): e0230978. https://doi.org/10.1371/journal.pone.0230978 PMID: 32236145 3. Perkins J. M., Jayatissa R. and Subramanian S. V. (2018). Dietary diversity and anthropometric status and failure among infants and young children in Sri Lanka. Nutrition, 55–56:76–83. https://doi.org/10. 1016/j.nut.2018.03.049 PMID: 29980091 4. Njeru, C. M., Ngugi, A., Kathomi, C. and Limbe, M. S. (2021). Complementary feeding: Is it the forgotten factor of the first 1000 days of life? Researchsquare.com. http://doi.org/10.21203/rs.3.rs-140037/v1. Accessed on December 26, 2021 @ 17:00 GMT. 5. UNICEF (2019). The state of the world’s children, food and nutrition. Growing well in a changing world. UNICEF. 6. WHO (2021). Infant and young child feeding. WHO. https://www.who.int/news-room/fact-sheets/detail/ infant-and-young-child-feeding on 11/11/2021 at 12:24 GMT. 7. Appiah K. B., Cheyuo E. K., Alhassan A., Ayanore M. A., Kubuga C. K. and Mogre V. (2020). Mothers’ knowledge and attitude regarding child feeding recommendations, complementary feeding practices and determinants of adequate diet. BMC nutrition, 6(67) no page number. 8. White J. M., Begin F., Kumapley R., Murray C. and Krasevec J. (2017). Complementary feeding prac- tices: Current global and regional estimates. Maternal and child nutrition, 13(S2):e12505. https://doi. org/10.1111/mcn.12505 PMID: 29032623 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 16 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana 9. Gatica-Dominguez G, Neves P. A. R., Barros A. J. D. and Victora C. G. (2021). Complementary Feed- ing Practices in 80 Low- and Middle-Income Countries: Prevalence of and Socioeconomic Inequalities in Dietary Diversity, Meal Frequency, and Dietary Adequacy Journal of nutrition, 151:1956–64. https:// doi.org/10.1093/jn/nxab088 PMID: 33847352 10. Gebremedhin S. (2019). Core and optional infant and young child feeding indicators in sub-Saharan Africa: a cross-sectional study. BMJ Open, 9:e023238. https://doi.org/10.1136/bmjopen-2018-023238 PMID: 30782876 11. Ariyo O., Aderibigbe O. R., Ojo T. J. and Sturm B. (2021). Determinants of appropriate complementary feeding practices among women with children aged 6–23 months in Iseyin, Nigeria. Scientific African, 13:e00848. 12. Areja A., Yohannes D. and Yohannis M. (2017). Determinants of appropriate complementary feeding practice among children 6–23 months of age in rural Damot sore district, Southern Ethiopia; a commu- nity based cross sectional study. BMC Nutrition, 3(82). 13. Issaka A. I., Agho K. E., Burns P., Page A., and Dibley M. J. (2014). Determinants of inadequate com- plementary feeding practices among children 6–23 months in Ghana. Public Health Nutrition, 18 (4):669–78. 14. GSS/GHS/ICF International (2015). Ghana Demographic and Health Survey, 2014. GSS/GHS/ICF International. 15. Gyampoh S., Otoo G. E. and Aryeetey R. N. O. (2014). Child feeding knowledge and practices among women participating in growth monitoring and promotion in Accra, Ghana. BMC pregnancy and child- birth, 14:180. https://doi.org/10.1186/1471-2393-14-180 PMID: 24886576 16. Saaka M., Larbi A., Mutaru S. and Hoeschle-Zeledon I. (2016). Magnitude and factors associated with appropriate complementary feeding among children 6–23 months in Northern Ghana. BMC Nutrition, 2 (2).no page number. 17. Donkor W. E. S., Adu-Afarwuah S., Wegmuller R., Bentil H., Petry N., Rohner F. et al. (2021). Comple- mentary feeding indicators in relation to micronutrient status of Ghanaian children aged 6–23 months: Results from a national survey. Life, 11(969) no page number. https://doi.org/10.3390/life11090969 PMID: 34575118 18. Anees M., -ur-Rehman A., Ahmad A. M. R., Bukhari S., Siddique M. H., Farooq U., et al. (2020). Persis- tent inadequacies in infant and young Child feeding practices and their determinants. Journal of Food and Nutrition Research, 8(7):347–354. 19. Marriott B. P., White A., Hadden L., Davies J. C. and Wallingford J. C. (2012). WHO infant and young child feeding indicators: associations with growth measures in 14 low-income countries. Maternal and child nutrition, 8(3):354–70. 20. Walson J. L. and Berkley J. A. (2018). The impact of malnutrition on childhood infections. Current opin- ion in infectious diseases, 31(3):231–6. https://doi.org/10.1097/QCO.0000000000000448 PMID: 29570495 21. GSS (2018). Multiple Indicator Cluster Survey (MICS 2017/18), Survey Findings Report. Accra, Ghana. GSS. 22. Dey N. E. Y., Dziwornu E., Frimpong-Manso K., Duah H. O. and Agbadi P. (2020). Correlates of child functional difficulties status in Ghana: A further analysis of the 2017/18 multiple indicator cluster survey. Heliyon, 6:e05727. https://doi.org/10.1016/j.heliyon.2020.e05727 PMID: 33364496 23. WHO/UNICEF (2021). Indicators for assessing infant and young child feeding practices: definitions and measurement methods. WHO/UNICEF. http://creativecommons.org/licenses/by-nc-sa/3.0/igo. Accessed on November 13, 2021 at 19:27 GMT. 24. Aboagye R. G., Seidu A.-A., Ahinkorah B. O., Arthur-Holmes F., Cadri A., Dadzie L. K., et al. (2021). Dietary diversity and undernutrition in children aged 6–23 months in sub-Saharan Africa. Nutrients, 13:3431. No page number. https://doi.org/10.3390/nu13103431 PMID: 34684435 25. Tassew A. A., Tekle D. Y., Belachew A. B. and Adhena B. M. (2019). Factors affecting feeding 6–23 months age children according to minimum acceptable diet in Ethiopia: A multilevel analysis of the Ethi- opian Demographic Health Survey. PLoS ONE, 14(2):e0203098. https://doi.org/10.1371/journal.pone. 0203098 PMID: 30789922 26. Shagaro S. S., Mulugeta B. T. and Kale T. D. (2018). Complementary feeding practices and associated factors among mothers of children aged 6–23 months in Ethiopia: Secondary data analysis of Ethiopian mini demographic and health survey 2019. Archives of Public Health, 79:205. https://doi.org/10.1186/ s13690-021-00725-x PMID: 34809724 27. Saaka M., Wemakor A., Abizari A. and Aryee P. (2015). How well do WHO complementary feeding indi- cators relate to nutritional status of children aged 6–23 months in rural Northern Ghana? BMC Public Health, 15:1157. https://doi.org/10.1186/s12889-015-2494-7 PMID: 26596246 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 17 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana 28. Roy A., Hossain M., Khan S. A., Hanif A. A. M., Hasan M., Hossaine M., et al. (2021). Differences in estimates of minimum dietary diversity by using the new and old Definition: evidence from Bangla- desh. Current Developments in Nutrition, 5(Supplement 2):808. https://doi.org/10.1093/cdn/ nzab046_105 29. Sealey-Potts C. and Potts Ac (2014). An association of dietary diversity and nutritional status of pre- school children. Austin journal of nutrition and food science, 2(7):1040. 30. Tang M. and Krebs N. F. (2014). High protein intake from meat as complementary food increases growth but not adiposity in breastfed infants: a randomized trial. American Journal of Clinical Nutrition, 100 (5):1322–28. https://doi.org/10.3945/ajcn.114.088807 PMID: 25332329 31. S. Folake O. and I. Ebunoluwa G. (2020). Complementary feeding practices and associated factors among nursing mothers in Southwestern Nigeria. International journal of maternal and child health and AIDS, 9(2):223–31. https://doi.org/10.21106/ijma.363 PMID: 32704409 32. Saaka M. and Abaah I. (2015). Maternal and infant factors associated with child growth in the first year of life. Science, 3(5):775–781. 33. Anin S. K., Saaka M., Fischer F. and Kraemer A. (2020). Association between Infant and Young Child Feeding (IYCF) Indicators and the Nutritional Status of Children (6–23 Months) in Northern Ghana. Nutrients, 12:2565. https://doi.org/10.3390/nu12092565 PMID: 32847027 34. Kambale R. M., Ngaboyeka G. A., Kasengi J. B., Niyitegeka S., Cinkenye B. R., Baruti A., et al. (2021). Minimum acceptable diet among children aged 6–23 months in South Kivu, Democratic Republic of Congo: a community-based cross-sectional study. BMC Pediatrics, 21:239. 35. Kuwornu J., Amoyaw J., Manyanga T., Cooper E., Donkoh E. and Nkrumah A. (2020). Measuring the Overall Burden of Early Childhood Malnutrition in Ghana: A Comparison of Estimates from Multiple Data Sources. International Journal of Health Policy and Management, https://doi.org/10.34172/ijhpm. 2020.253 PMID: 33589568 36. Andina E., Madinar M., and Achadi E. L. (2021). Fulfilment of Minimum Acceptable Diet as Dominant Factor in Wasting in Children Aged 6–23 Months in Central Jakarta, Indonesia, 2019. Indonesia journal of public health nutrition, 1(2). n.p. 37. Benedict L., Hong S. A., Winichagoon P., Tejativaddhana P. and Kasemsup V. (2021). Double burden of malnutrition and its association with infant and young child feeding practices among children under five in Thailand. Public health nutrition, 24(10):3058–65. https://doi.org/10.1017/S1368980020003304 PMID: 33054885 38. Galler J. R., Bringas-Vega M. L., Tang Q., Rabinowitz A. G., Musa K. I., Chai W. J., et al. (2021). Neuro- developmental effects of childhood malnutrition: A neuroimaging perspective. NeuroImage, 231 (117828). https://doi.org/10.1016/j.neuroimage.2021.117828 PMID: 33549754 39. Belew A. K., Ali B. M., Abebe Z. and Dachew B. A. (2017). Dietary diversity and meal frequency among infant and young children: a community based study. Italian Journal of Pediatrics, 43:73 https://doi.org/ 10.1186/s13052-017-0384-6 PMID: 28810887 40. Bodjrènon F. S. U., Hounkpatin W. A., Termote C. Dato G. and Savy M. (2020). Determining factors associated with breastfeeding and complementary feeding practices in rural Southern Benin. Food Sci Nutr., 9:135–144. https://doi.org/10.1002/fsn3.1971 PMID: 33473277 41. Dhami M. V., Ogbo F. A., Osuagwu U. L. and Agho K. E. (2019). Prevalence and factors associated with complementary feeding practices among children aged 6–23 months in India: a regional analysis. BMC Public Health, 19:1034. https://doi.org/10.1186/s12889-019-7360-6 PMID: 31370827 42. Mitchodigni I. M., Hounkpatin W. A., Ntandou-Bouzitou G., Avohou H., Termote C, Kennedy Gina, et al. (2017). Complementary feeding practices: determinants of dietary diversity and meal frequency among children aged 6–23 months in Southern Benin. Food Secience, 9:1117–30. https://doi.org/10.1007/ s12571-017-0722-y 43. Abebe H., Gashu M., Kebede A., Abata H., Yeshaneh A., Workye H. et al. (2021). Minimum acceptable diet and associated factors among children aged 6–23 months in Ethiopia. Italian Journal of Pediatrics, 47:215. https://doi.org/10.1186/s13052-021-01169-3 PMID: 34717712 44. Dagmawit S., Zewdie A. and Tegegne T. K. (2017). Minimum dietary diversity and associated factors among children aged 6–23 months in Addis Ababa, Ethiopia. International Journal for Equity in Health, 16:181. https://doi.org/10.1186/s12939-017-0680-1 PMID: 29025434 45. Molla A., Egata G., Getacher L., Kebede B., Sayih A., Arega M. et al. (2021). Minimum acceptable diet and associated factors among infants and young children aged 6–23 months in Amhara region, Central Ethiopia: community-based cross-sectional study. BMJ Open, 11:e044282. https://doi.org/10.1136/ bmjopen-2020-044284 PMID: 33972337 46. Mulaw G. F., Feleke F. W. and Masresha S. A. (2020). Maternal characteristics are associated with child dietary diversity score in Golina district, Northeast Ethiopia: a community-based cross-sectional study. Journal of nutrition and metabolism, Article ID 6702036. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 18 / 19 PLOS ONEInfant and young child feeding practices and associated socioeconomic and demographic factors in Ghana 47. Anane I., Nie F. and Huang J. (2021). Socioeconomic and geographic pattern of food consumption and dietary diversity among children aged 6–23 months old in Ghana. Nutrients, 13(2):603. https://doi.org/ 10.3390/nu13020603 PMID: 33673212 48. Scaglioni S., De Cosmi V., Ciappolino V., Parazzini F., Brambilla P. and Agostoni C. (2018). Factors Influencing Children’s Eating Behaviours. Nutrients, 10(706). https://doi.org/10.3390/nu10060706 PMID: 29857549 49. Van Der, S. (2015). The Association Between Complementary Feeding Indicators and Linear Child Growth and the Determinants of Inappropriate Feeding Practices Among Children 6–23 Months in Rwanda. Master’s Thesis, Wageningen University, Division of Epidemiology and Public Health, Wagen- ingen, The Netherlands. 50. Ali M., Arif M. and Shah A. A. (2021). Complementary feeding practices and associated factors among children aged 6–23 months in Pakistan. PLOS ONE, 16(2):e0247602. Nutrients, 13(603). https://doi. org/10.1371/journal.pone.0247602 PMID: 33630931 51. Tuji T. S. and Wake A. D. (2021). Sociodemographic determinants of dietary diversity and meal fre- quency among mothers/caregivers paired with infants and child age 6 to 23 months. International jour- nal of nutrition and food sciences, 10(1):1. 52. Woldegebriel A. G., Desta A. A., Gebreegziabiher G., Berhe A. A., Ajemu K. F. and Woldearegay T. W. (2020). Dietary diversity and associated factors among children aged 6–59 months in Ethiopia: analysis of Ethiopian demographic and health survey 2016 (EDHS 2016). International journal of pediatrics, Arti- cle ID 3040845. https://doi.org/10.1155/2020/3040845 PMID: 32908551 53. Zebadia E., Mahmudiono T., Atmaka D. R., Dewi M., Helmyati S. and Yuniar C. (2021). Factors associ- ated with minimum acceptable diet in 6–11 months old Indonesian children using the 2017 IDHS. Open access Macedonian journal of medical sciences, 9(E):1403–12. 54. WHO (2017). Global Nutrition Monitoring Framework: Operational Guidance for Tracking Progress in Meeting Targets for 2025. World Health Organization, Geneva, Switzerland. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286055 June 9, 2023 19 / 19 PLOS ONE
10.1371_journal.pone.0284383	RESEARCH ARTICLE Medical decision making beyond evidence: Correlates of belief in complementary and alternative medicine (CAM) and homeopathy Leonie AßmannID, Tilmann Betsch Department of Psychology, Faculty of Education, University of Erfurt, Erfurt, Germany leonie.assmann@uni-erfurt.de Abstract Many people believe in and use complementary and alternative medicine (CAM) to address health issues or prevent diseases. Empirical evidence for those treatments is either lacking or controversial due to methodological weaknesses. Thus, practitioners and patients primar- ily rely on subjective references rather than credible empirical evidence from systematic research. This study investigated whether cognitive and personality factors explain differ- ences in belief in CAM and homeopathy. We investigated the robustness of 21 predictors when examined together to obtain insights into key determinants of such beliefs in a sample of 599 participants (60% female, 18-81 years). A combination of predictors explained 20% of the variance in CAM belief (predictors: ontological confusions, spiritual epistemology, agreeableness, death anxiety, gender) and approximately 21% of the variance in belief in homeopathy (predictors: ontological confusions, illusory pattern perception, need for cogni- tive closure, need for cognition, honesty-humility, death anxiety, gender, age). Individuals believing in CAM and homeopathy have cognitive biases and certain individual differences which make them perceive the world differently. Findings are discussed in the context of previous literature and in relation to other unfounded beliefs. Introduction Imagine you have a sore throat, runny nose, and itchy eyes—you might have caught a cold. You think it is not yet necessary to see a doctor, but you want to take something for relief and that helps your body to get better. What is your choice of remedy? There are treatments provided by conventional medicine as well as by complementary and alternative medicine (CAM). CAM differs substantially from conventional medicine [1], although a clear defini- tion of CAM is lacking [2] and quite difficult to achieve since CAM comprises approxi- mately 400 procedures that differ widely in methodical approaches [3]. In general, CAM treatments include not only remedies but also a wide range of practices and other modalities —the offers are highly diverse. Some forms can be received as delivered by a practitioner (e.g., acupuncture or chiropractic treatments), whereas others involve self-care practices, i.e., homeopathic remedies, herbal medicines, and vitamins [4]. All treatments are a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Aßmann L, Betsch T (2023) Medical decision making beyond evidence: Correlates of belief in complementary and alternative medicine (CAM) and homeopathy. PLoS ONE 18(4): e0284383. https://doi.org/10.1371/journal. pone.0284383 Editor: Yuh-Yuh Li, National Sun Yat-sen University, TAIWAN Received: October 21, 2022 Accepted: March 29, 2023 Published: April 21, 2023 Copyright: © 2023 Aßmann, Betsch. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting information files. Funding: The authors received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284383 April 21, 2023 1 / 14 PLOS ONECorrelates of belief in complementary and alternative medicine (CAM) and homeopathy commonly not provided within conventional medicine settings, since they do not adhere to the dominant biomedical model of health and evidence-based health care [5–7]. Neverthe- less, many people use CAM to address health issues and prevent diseases. This also applies for people who do not refuse conventional medicine [8, 9]. Hence, the remedies are widely used as complementary treatments to conventional medicine but sometimes also as alterna- tive treatments. The belief in CAM and use of CAM treatments are prominent all over the globe. In India, CAM treatments are part of public health care [10]. In Europe, e.g., Germany and Switzerland, they are covered by public health insurance [11, 12] and universities provide courses in Tradi- tional Chinese Medicine, Anthroposophical Medicine, and homeopathy [13–15]. In other western countries, e.g., in the US, UK, and Australia, a small but considerable number of peo- ple are attracted to CAM treatments [12]. In Europe, approximately 26% of the general popula- tion have experience using CAM. The use depends highly on the country [16]. Especially in Germany, various forms of CAM are used frequently. The most favored form is homeopathy: 55% have experience with the use of homeopathic remedies. Only approximately one in four (26%) report a refusal in using homeopathic treatment [17]. The use of homeopathy often includes not only visits to homeopaths but also purchasing over-the-counter homeopathic medicines [12]. For example, German pharmacies turned over approximately 542 million Euros with homeopathic medicines in 2018 [18]. In many cases, the use of homeopathy goes along with use of other CAM methods [19, 20] and sometimes also with rejection of conven- tional medicine (e.g., vaccinations [6]). Evidence-based medicine, science, CAM, and homeopathy One reason for the popularity of CAM and homeopathy is that such treatments are perceived to be low risk with regard to negative side effects. On the other hand, it is questionable whether they have any effects at all: quite often they have not been empirically tested [6] nor is there empirical evidence for their usefulness beyond placebo effects (e.g., homeopathy [7]). Hence, the main ‘evidence’ for CAM treatments is merely anecdotical [2, 21]. Practitioners and patients primarily rely on subjective references—successful treatment experiences reported by family, friends, and colleagues—rather than credible empirical evidence from systematic research. However, empirical evidence should be used as a guideline in health-related decision making along with other considerations such as an evaluation of risks [22]. Evidence-based medicine is an approach that combines “individual clinical expertise with the best available external evidence from systematic research” [23]. This has become the most prevalent norm not only for medical decision making but also for clinical practice guidelines [3, 24]. In con- ventional medicine, the demonstration that a treatment is relatively safe and effective is a necessary requirement before it is included in public health care [21]. This demonstration is generally accomplished through empirical studies with randomized controlled trials (RCTs). Without such research, both patients and practitioners alike tend to attribute improvements in health status as valid treatment effects. Consequently, CAM “embraces subjective, emotive truth criteria, whereas its detractors demand objective evidence” [21]. In addition, the hypothesized mechanisms are in conflict with those accepted by science. This is fundamental in studies concerning homeopathic remedies. The theoretical explanations for the effective- ness of homeopathy (i.e., the similarity rule and exponentiation) violate fundamental princi- ples of natural sciences. Therefore, it might be considered questionable per se to test “whether magic works” [25]. Notwithstanding its unsound theoretical background, studies on the effects of homeopathic treatments have been accumulating during the recent decades. The PLOS ONE \| https://doi.org/10.1371/journal.pone.0284383 April 21, 2023 2 / 14 PLOS ONECorrelates of belief in complementary and alternative medicine (CAM) and homeopathy majority are methodologically severely flawed. Studies that are high quality and adhere to sci- entific standards find no effects of homeopathy beyond placebo effects (e.g. [26–28]). Further, numerous reviews and meta-analyses converge in showing that there is no empirical evidence that homeopathic remedies have any effect beyond control group level [28–33]. Rather, pla- cebo effects are seen as the most obvious cause for the effects experienced after homeopathic treatments [3, 32–34]. Determinants and correlates of beliefs in CAM and homeopathy When decisions for CAM and homeopathy treatments are beyond evidence, the key question is: why do so many people still believe in CAM as well as homeopathy despite the presence of contrary data? This question is important not only for academia but also for applied contexts and especially public health, because incorrect health-related decisions do not only imply eco- nomic loss but can also have serious consequences for the individual health status. Those con- sequences arise through delayed conventional treatments or unexpected side effects of CAM treatments. Thus, it is necessary to understand what fosters such beliefs. In many studies, the focus was on socio-demographic variables. It was found that primar- ily middle-aged [8], well-educated women use CAM [16, 35] and homeopathic treatments, in particular [4, 8, 19, 35, 36]. In addition, reasoning skills have also been examined in rela- tion to CAM belief and use. In principle, human information processing is often biased by social and emotional factors and motivated to confirm existing beliefs [6, 37, 38]. This is, because rational and experiential knowledge systems work in parallel. Both depend on each other [22] and existing information (scientific and non-scientific) is needed to process new (scientific) information, and to decide if it is accepted, rejected or assimilated [39]. Also, decision making is affected by reasoning skills and cognitive style [40–42]. Beyond situa- tional factors evoking different thinking styles, there are also stable individual differences in the tendency to overcome intuitive responses through additional reflection [42]. Those dif- ferences in the willingness to reason analytically oppose the susceptibility to biases in deci- sion making [6]. Decisions for health-related treatments involve uncertainty and appropriate weighting of probabilities. Dealing with numerical information is often easier for people with higher numeracy and a tendency to reason analytically. In the context of health, people with an analytic cognitive style evaluate treatments more favorably when they are in line with available evidence. Thus, analytical thinking style was negatively related to all forms of CAM, having a significant relationship with homeopathy [6]. In contrast, an intuitive cogni- tive style is associated with impulsive decision making—thus, it is connected to more incor- rect evaluations of CAM treatments [4, 7]. In addition, evidence-based health decisions require a certain understanding of science and scientific methods. Thus, an accurate under- standing of the evidence around CAM and homeopathy is related to individual differences in scientific reasoning. This was a predictor not only for CAM belief but also for the use of CAM [2, 9]. It also contributes to an understanding of causality. One of the most common and invalid assumptions in (medical) decision making is confusing correlation or coinci- dence with causation [21]—for instance, attributing a random effect to a treatment without testing for a causal relationship. Additionally, people tend to perceive illusory patterns in random events [43]. The automatic tendency to see meaningful connections between stimuli arises from the fact that people strive to understand the world. However, these otherwise functional processes can be disrupted, so that connections are mistakenly seen between stimuli that are actually unrelated. Thus, in Illusory Pattern Perception, coherent and mean- ingful connections are seen in a set of random stimuli. These include the perception of false correlations [43, 44]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284383 April 21, 2023 3 / 14 PLOS ONECorrelates of belief in complementary and alternative medicine (CAM) and homeopathy Not only cognitive factors influence our perceptions and decision making. Personality can shape our beliefs as well. In the context of CAM, openness to new experience is an associated personality dimension [4, 7, 45]. Open individuals often favor “a personal, emotional, and spir- itual approach to health decisions” and thus reject objective scientific knowledge [6]. Research findings for the relationship between CAM and neuroticism were mixed—some studies found a positive, others a negative relationship [7]. Certain beliefs—more specifically, unfounded beliefs such as paranormal beliefs or conspiracies—are directly related to CAM belief. Previous studies showed that CAM believers tend to also believe in paranormal phenomena and conspiracies [8, 46, 47]. Para- normal beliefs are usually built on a magical worldview without reasoned review [48], which is shared by CAM. Both belief forms advocate emotional criteria for truth instead of empirical data and logical considerations. Another belief, namely spirituality, is closely related to paranormal beliefs and religiosity, and also associated with being a CAM user [4, 45, 49, 50]. Lindeman [46] found that CAM belief could be best explained by intuitive reasoning, paranormal beliefs, and ontological confusions. Ontological confusions are defined as category mistakes in which properties of living and lifeless entities are mixed [51]. This study Taken together, research has examined different correlates and predictors for belief in and use of CAM. The findings are varying. In most studies, the predictors were investigated sepa- rately (e.g. [2, 4, 8]). Researchers typically focused on a small selection of factors. Thus, there is a particularity—studies so far have investigated relatively small models with only a selected set of predictors in relation to CAM beliefs and beliefs in homeopathy. So far, there is no model for all predictors. This is a gap in the literature. Hence, it is unknown which predictors are strongest—the relative predictive power of each individual predictor remains unknown if it is not assessed in comparison to other potential predictors. Thus, this study is, to the best of our knowledge, a first attempt to investigate most of the variables together along with fur- ther variables. The goal of the current study is to test the predictors’ robustness when exam- ined together. We focus on belief since use is not necessarily linked to belief in effectiveness [7]. We aim to obtain insights into key determinants of belief in CAM as well as belief in home- opathy as an explicit and popular form of CAM. Specifically, the present study aims to assess the degree to which belief in CAM and homeopathy is associated with cognitive and personal- ity factors. Our focus is on rather stable individual factors. Certainly, there are also different environments which influence the process of belief and judgement formation resulting in dif- ferent decisions regarding health products and treatments. However, in this attempt, we aim to find out the individual factors that are more general enduring determinants. These individ- ual differences go beyond specific situations. The effects are tested in an exploratory manner due to two considerations. First, it is a new approach to analyze most of the predictors in a global analysis. Even if we have some informa- tion on the predictors from the reported literature, in these studies, smaller models were used that examined the predictors separately as already mentioned. Therefore, no specific expecta- tions on how the predictive power will change and how strong the predictors will remain in competition with other variables can be derived on this basis. Secondly, we did not preregister hypotheses and the analysis plan. Thus, we follow current recommendations to consider the analyses exploratory (cf. [52–54]). PLOS ONE \| https://doi.org/10.1371/journal.pone.0284383 April 21, 2023 4 / 14 PLOS ONECorrelates of belief in complementary and alternative medicine (CAM) and homeopathy Materials and methods Sample Participants were recruited from a large participant pool (N = 1,632 participants) established by the University of Hagen, Germany. The personality measures (HEXACO), demographic variables, and level of education were already assessed in a basic survey when registering for the participant pool. 599 individuals participated in the study. Based on a post-hoc power analysis (G-Power [55]) we were able to detect small effects (Cohen’s f2 = .02) with 93% sta- tistical power (linear multiple regression, single coefficient, alpha = .05, two-sided test) with this sample size. The sample included 362 women (60%) and 237 (40%) men. Age ranged between 18 und 81 years (M = 33.63 years, SD = 11.38). Forty-two percent of participants reported holding an academic degree. It was the same sample as in Betsch et al. [48], since the goal of the study was to examine predictors for multiple unfounded beliefs, such as para- normal beliefs, CAM beliefs, and beliefs in conspiracy theories. Predictors for paranormal beliefs were presented in Betsch et al. [48]. Predictors for the other criteria have not yet been analyzed. Materials Causality understanding was assessed with a fictitious scenario called ‘Tom in South Amer- ica’ (Betsch et al., [Unpublished]). The scenario described an incidence of a seemingly causal relationship. Participants were asked whether the coincidence sufficed as proof of a causal relationship. Three items were based on scientific criteria for verifying a causal relationship, the other three items ignored these criteria and contained information that did not allow causal conclusions. The sum of the correct answers and the correctly rejected answers formed the indicator for the individual causality understanding. The 3-item Cognitive Reflection Test (CRT-3) was used to examine the cognitive style. The test indicated the ten- dency to override an initial intuitive response by applying analytic thinking skills [40]. It is the most popular instrument based on modern dual process theories of cognition. Epistemo- logical prudence is a concept that describes the testing logic in science and is characterized by critical rationality in thinking [56]. Critical rationalism is one of the epistemological theo- ries by which new knowledge can be gained. It involves forming hypotheses that must be tested for validity. They can be disproved and are thus falsified. An understanding of this testing logic and the realization that the process of gaining knowledge is never complete and scientific results are never final [56] characterizes Epistemological Prudence. It was mea- sured with the Epistemological Prudence Scale (Betsch [Unpublished]). The tendency to see patterns in random events (illusory pattern perception) was examined with random coin tosses as in van Prooijen et al. [43]. Ontological confusions were assessed with the Core Knowledge Confusions Scale [51, 57], which has already been used in previous studies. As in Browne et al. [6], an item combining knowledge and spiritual/religious belief was included in addition to religious belief. The item “The most important knowledge results from reli- gious/spiritual experiences” (7-point Likert scale) should capture an aspect of religiosity/ spirituality that is contradictory to an evidence-based world-view, the so-called spiritual epistemology. The personality dimensions openness, emotionality, extraversion, agreeable- ness, conscientiousness, and honesty-humility were measured with the 100-Item German version of the HEXACO Personality Inventory-Revised (HEXACO-PI-R) [58]. Each dimen- sion consisted of 16 items. Data on numeracy were collected with the DR-Numeracy Test [59]. The ‘Short scale for the assessment of need for cognitive closure’ ([60]; adaption of [61]) was used to assess need for cognitive closure. The ‘Scale for the assessment of need for PLOS ONE \| https://doi.org/10.1371/journal.pone.0284383 April 21, 2023 5 / 14 PLOS ONECorrelates of belief in complementary and alternative medicine (CAM) and homeopathy cognition’ [62] measured need for cognition. To investigate ambiguity tolerance, the ‘Multi- ple Stimulus Types Ambiguity Tolerance Scale II (MSTAT-II)’ [63] was used. Life satisfac- tion was assessed with the ‘Short scale life satisfaction-1’ [64]. The Death Anxiety Scale measured death anxiety with 16 items [65]. The socio-demographic variables, precisely age, gender and education, were assessed with self-report items. The key dependent variables CAM belief and belief in homeopathy– a specific form of CAM belief—measured with single items asking for the strength of belief. The item “Please indicate how much you believe in complementary and alternative medicine (e.g., energetic healing, Bach flowers, healing stones . . .)” as well as the item “Please indicate how much you believe in homeopathy.” were rated on a 6-point Likert scale. Procedure This study was approved by the University of Erfurt Ethic Board and written informed consent was obtained at the beginning of the study. Participants voluntarily agreed to participate in a study on attitudes and beliefs which they could withdraw at any time. The online questionnaire included 121 items in a fixed order. Answering took about 28 minutes. Participants received a flat fee of 5€ in compensation for their participation. Results Reliability of the scales used ranged between.59 and.92 (Cronbach’s alpha). According to con- ventions, all scales were sufficiently reliable and could be utilized. The exact data is provided in S1 Table. Overall, we replicated most of the correlations reported in the literature. These findings indicate that the majority of the variables assessed in our study represented promising candidates for predictors of beliefs in CAM and homeopathy. Results are displayed in Table 1. To assess the degree to which beliefs in CAM and homeopathy are associated with cognitive and personality factors the data were analyzed with a linear multiple forced entry regression analysis containing all of the predictor variables simultaneously. The regression was conducted for CAM belief and for belief in homeopathy each. As a first step, we checked the assumptions of a regression—they were all met. The correla- tions between the predictor variables can be found in S2 Table. No variables correlated too highly and the collinearity diagnostics provide no reason for concern (cf. [66]). The first model tested the predictors in relation to the criterium ‘CAM belief’ resulting in the regression model: yCAM ¼ b0 þ bcausality understanding∗x1 þ bcognitive style∗x2 þ bepistemological prudence∗x3 þbillusory pattern perception∗x4 þ bontological confusions∗x5 þ bspiritual epistemology∗x6 þbopenness to new experiences∗x7 þ bemotionality∗x8 þ bextraversion∗x9 þbagreeableness∗x10 þ bconscientiousness∗x11 þ bhonesty=humility∗x12 þbnumeracy∗x13 þ bneed for cognitive closure∗x14 þ bneed for cognition∗x15 þbambiguity tolerance∗x16 þ blife satisfaction∗x17 þ bdeath anxiety∗x18 þbage∗x19 þ bgender∗x20 þ beducation∗x21 þ �i PLOS ONE \| https://doi.org/10.1371/journal.pone.0284383 April 21, 2023 6 / 14 PLOS ONETable 1. Descriptive data and correlations of criteria and potential predictor variables. Correlates of belief in complementary and alternative medicine (CAM) and homeopathy Variable CAM Homeopathy Causality understanding Cognitive style Epistemological prudence Illusory pattern perception Ontological confusions Spiritual epistemology Openness to experience Emotionality Extraversion Agreeableness Conscientiousness Honesty − humility Numeracy Need for cognitive closure Need for cognition Ambiguity tolerance Life satisfaction Death anxiety Age Gender Education Note. N = 599. Bold values indicate p <.05. https://doi.org/10.1371/journal.pone.0284383.t001 M SD CAM Homeopathy 3.10 3.32 5.35 1.75 4.91 2.42 2.47 2.04 3.48 3.23 3.33 3.01 3.57 3.58 12.26 3.24 5.19 4.49 7.73 8.04 33.63 - - 1.68 1.70 1.10 1.13 .67 1.32 .59 1.37 .60 .59 .66 .59 .57 .72 1.95 .70 .92 .75 2.35 2.86 11.38 - - .72 -.14 -.13 -.10 .12 .29 .20 -.05 .13 .08 .06 -.05 .00 -.12 .04 -.11 -.04 .03 .17 .03 .25 -.08 -.13 -.16 -.16 .15 .29 .13 -.09 .10 .09 .02 -.03 -.04 -.15 .02 -.15 -.003 .07 .17 .02 .24 -.08 The second model tested the predictors in relation to the criterium ‘belief in homeopathy’ resulting in the regression model: yhomeopathy ¼ b0 þ bcausality understanding∗x1 þ bcognitive style∗x2 þbepistemological prudence∗x3 þ billusory pattern perception∗x4 þbontological confusions∗x5 þ bspiritual epistemology∗x6 þbopenness to new experiences∗x7 þ bemotionality∗x8 þ bextraversion∗x9 þbagreeableness∗x10 þ bconscientiousness∗x11 þ bhonesty=humility∗x12 þbnumeracy∗x13 þ bneed for cognitive closure∗x14 þ bneed for cognition∗x15 þbambiguity tolerance∗x16 þ blife satisfaction∗x17 þ bdeath anxiety∗x18 þbage∗x19 þ bgender∗x20 þ beducation∗x21 þ �i Table 2 displays the results for both regression analyses. The predictor model explained 20.2% of the variance in CAM belief, indicating a high goodness-of-fit [67]. Ontological confusions and spiritual epistemology correlated positively with the criterion. Also, agreeableness and death anxiety were both significant predictors of CAM belief and as well positively correlated. Gender, as a socio-demographic variable, was also positively correlated with CAM belief. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284383 April 21, 2023 7 / 14 PLOS ONECorrelates of belief in complementary and alternative medicine (CAM) and homeopathy Table 2. Results of regression analysis with cognitive, personality, and socio-demographic variables as predictors as well as CAM belief and belief in homeopathy as criteria. Model CAM Model Homeopathy Predictors (constant) Causality understanding Cognitive style Epistemological prudence Illusory pattern perception Ontological confusions Spiritual epistemology Openness to experience Emotionality Extraversion Agreeableness Conscientiousness Honesty − humility Numeracy Need for cognitive closure Need for cognition Ambiguity tolerance Life satisfaction Death anxiety Age Gender Education b (SE) .71 (1.5) -.09 (.06) .02 (.07) -.03 (.11) .06 (.05) .59 (.12) .14 (.05) -.14 (.12) .12 (.13) .18 (.11) .31 (.12) -.05 (.12) -.16 (.10) -.01 (.04) -.16 (.14) -.07 (.09) -.002 (.12) .01 (.03) .09 (.03) .01 (.01) .74 (.15) -.05 (.06) β t p -.06 .01 -.01 .05 .21 .11 -.05 .04 .07 .11 -.02 -.07 -.01 -.07 -.04 -.001 .01 .15 .08 .22 -.03 .48 -1.47 .27 -.27 1.20 5.00 2.80 -1.22 .91 1.66 2.55 -.44 -1.65 -.32 -1.11 -.76 -.02 .22 3.41 1.76 5.06 -.82 .63 .14 .79 .79 .23 <.001 .005 .22 .36 .10 .01 .66 .10 .75 .27 .45 .99 .83 .001 .08 <.001 .42 b(SE) 2.73 (1.50) -.07 (.06) .01 (.07) -.17 (.11) .10 (.05) .54 (.12) .05 (.05) -.19 (.12) .06 (.13) .16 (.11) .17 (.12) .05 (.13) -.22 (.10) -.03 (.04) -.29 (.14) -.22 (.09) .12 (.12) .04 (.03) .10 (.03) .02 (.01) .76 (.15) -.05 (.06) β t p -.04 .01 -.07 .08 .18 .04 -.07 .02 .06 .06 .02 -.09 -.03 -.12 -.12 .05 .05 .17 .10 .22 -.03 1.82 -1.09 .14 -1.46 2.03 4.47 1.00 -1.57 .47 1.39 1.39 .38 -2.17 -.73 -2.04 -2.29 .99 1.19 3.94 2.23 5.12 -.81 .07 .28 .89 .14 .04 <.001 .32 .12 .64 .16 .17 .71 .03 .47 .04 .02 .33 .23 <.001 .03 <.001 .42 Note. N = 599. Bold p-values indicate p <.05; Model CAM: F (21, 577) = 6.96; p <.01; R = .45, R2 = .202, SE = 1.53; Model Homeopathy: F (21, 577) = 7.22; p <.01; R = .46,R2 = .208, SE = 1.55. https://doi.org/10.1371/journal.pone.0284383.t002 Approximately 21% of the variance in belief in homeopathy was explained with the predic- tor model, again indicating a high goodness-of-fit [67]. The predictors ontological confusions and illusory pattern perception correlated positively with the criterion. The socio-demographic variables, gender and age, were both positively correlated with belief in homeopathy. Need for cognitive closure, need for cognition, honesty-humility, and death anxiety were significant predictors of belief in homeopathy. Need for cognitive closure, need for cognition, and hon- esty-humility were negatively correlated with the criterion, whereas death anxiety was posi- tively correlated. Discussion We conducted this study because a considerable number of adults hold beliefs in CAM and homeopathy. This can seem harmless and without any severe consequences, especially in rela- tion to milder diseases—however, this is not the case. Those beliefs rather reflect “a misunder- standing of how evidence for effective treatments is generated and what actually constitutes evidence of efficacy” [2]. Thus, we assessed whether cognitive and personality factors are able to explain differences in CAM belief and belief in homeopathy. Precisely, we investigated the predictors’ robustness when examined together to obtain insights into key determinants PLOS ONE \| https://doi.org/10.1371/journal.pone.0284383 April 21, 2023 8 / 14 PLOS ONECorrelates of belief in complementary and alternative medicine (CAM) and homeopathy associated with beliefs in CAM and homeopathy. There were 21 predictors included in the analyses. According to our findings, individuals with CAM belief were mainly women who tended to confuse ontological categories, acquired knowledge from religious and/or spiritual experiences, had a higher level of death anxiety, and were more agreeable which means, among other things, being lenient in judging others. Further, our results indicate that believers in homeopathy were primarily women with increasing age who also showed ontological confu- sions, perceived illusory pattern in unrelated stimuli, had a lower need for cognition but at the same time a lower need for cognitive closure and, among other characteristics, had a strong sense of self-importance. They also felt anxious due to death. As we can see, some variables predicted both belief forms, whereas others were predictors of either one belief. Focusing on the cognitive variables, there were biases predicting the beliefs. Those reflect that people do not think scientifically and in adequate categories. People with such cognitive biases apply the wrong distinctive properties to the superordinate categories, base their knowledge on inappro- priate foundations, and see relations that might only be due to coincidence. Moreover, gender and age can explain the beliefs as in previous studies [4, 8, 16, 19, 35]. Interestingly, these demographic variables describe a group that is generally more interested in and concerned with health issues and basically engages more with health-related topics [2, 46]. One could argue that the predictors’ informative value is therefore rather limited. On the other hand, they can be valuable in the sense that it can be further evaluated why some—and only some—of those generally interested in health topics turn to CAM and homeopathy. Death anxiety was a strong predictor for both belief forms. This is reasonable, since CAM methods and homeo- pathic remedies are used to maintain health or advert diseases and physical suffering that might lead to death. World view In general, the results can be seen in terms of a specific perception of the world. A key char- acteristic of many CAM treatments is the spiritual orientation to knowledge and decision- making [6]. For medical decisions, individuals who agree that important knowledge results from religious or spiritual experiences might not rely on evidence as a proof of efficiency but rather explain it in terms of personal experiences. In a previous study, it was found that some of the primary reasons for the use of homeopathy were having good experiences in the past [19]. Own experiences have a high value and persuasive power but are meaningless from a scientific point of view as individual cases cannot be generalized to others. Another severe problem in relying on personal experiences for proving effectiveness is that there is no con- trol for confounding variables. Thus, experience easily leads to erroneous conclusions about causality due to uncontrolled confounding. This also relates to the predictor illusory pattern perception. Category mistakes in which properties of living and lifeless entities are mixed (ontological confusions) were a stable predictor not only for CAM beliefs [46] but also for other unfounded beliefs such as paranormal beliefs [48, 51, 57]. The confusions occur between the core attri- butes of mental, physical, and biological entities and processes. Mixing up those attributes can easily result in incorrect conclusions about treatments, especially considering their physical and biological processes. Paranormal beliefs and CAM belief do not only share ontological confusions as a predictor, paranormal beliefs are typically the best predictor of CAM belief (e.g. [8, 46]). However, comparing those belief forms, it becomes clear that they share many concepts and approaches to explain reality in an unscientific way. Therefore, both belief forms, paranormal beliefs and CAM beliefs, could also be seen as a result of a world view in which sci- entific evidence is valued less and, instead, emotional and spiritual explanations are consulted. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284383 April 21, 2023 9 / 14 PLOS ONECorrelates of belief in complementary and alternative medicine (CAM) and homeopathy Inattention and ignorance Our results do not replicate previous findings that showed predictive value of certain cognitive variables such as cognitive style (e.g. [4, 6, 7]). An explanation could be that rather inattention to accuracy than inability to consider empirical evidence fosters the beliefs. People might sim- ply not be aware of the absence of evidence. Another possibility is that people are aware of the absence of evidence but are reluctant to engage with it. Practitioners and patients often claim “whatever works is good” or “the main thing is that it works”. Thus, it is ignorance rather than lack of capacity to appropriately process the evidence. Limitations As with most cross-sectional studies using questionnaires, our results are based on self-reports. Additionally, single items were used for measuring belief strength. Even if multi-item mea- sures often have advantages, single items can be advantageous in terms of practical benefits, e.g., adapting to subjects’ limited attention and time resources. There are several single item measures successfully used to measure diverse concepts [68–71] including attitudes [72]. Also, the variance on those items in our sample shows that participants were able to reflect their beliefs and rank them on the scale provided. Another limitation is that the findings are based on regression analyses, which do not provide insight into causality. Thus, the relationship remains correlational. Even if our sample was broader than in many other psychological stud- ies—it was slightly unbalanced, especially in comparison to the German population. It over- represented educated individuals which may led to an inadequate variation of the cognitive variables if we consider the relationship between cognition and education. However, education and the cognitive variables are only weakly correlated. Thus, it can be assumed that the unbal- anced sample did not affect the distribution of cognitive variables to a great extent. Conclusion Our findings show that some of the predictors from previous research replicated whereas oth- ers did not. Demographics and certain cognitive variables seem to be key determinants associ- ated with beliefs in CAM and homeopathy. Those individual differences and cognitive biases might result in a different perception of the world. However, variables related to abilities did not predict the beliefs. Thus, they might not be a result of inability but rather of ignorance. Supporting information S1 Table. Overview of reliability values for the scales used. (PDF) S2 Table. Overview of correlations between the predictors. (PDF) S1 Dataset. Dataset of the sample. (SAV) Acknowledgments We would like to thank Andreas Glo¨ckner for providing access to the data pool and Heather Fiala for copy editing on an earlier draft of this paper. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284383 April 21, 2023 10 / 14 PLOS ONECorrelates of belief in complementary and alternative medicine (CAM) and homeopathy Author Contributions Conceptualization: Leonie Aßmann, Tilmann Betsch. Data curation: Leonie Aßmann. Formal analysis: Leonie Aßmann. Investigation: Leonie Aßmann. Methodology: Leonie Aßmann, Tilmann Betsch. Project administration: Leonie Aßmann. Supervision: Tilmann Betsch. Validation: Tilmann Betsch. Visualization: Leonie Aßmann. Writing – original draft: Leonie Aßmann. Writing – review & editing: Leonie Aßmann, Tilmann Betsch. References 1. Ernst E, Smith K. More harm than good? Cham: Springer International Publishing; 2018. 2. Čavojova´ V, Ersoy S. The role of scientific reasoning and religious beliefs in use of complementary and alternative medicine. Journal of Public Health. 2020; 42(3):e239–e248. https://doi.org/10.1093/ pubmed/fdz120 PMID: 31588497 3. Anlauf M, Hein L, Hense HW, Ko¨bberling J, Lasek R, Leidl R, et al. Complementary and alternative drug therapy versus science-oriented medicine. German medical science: GMS e-journal. 2015; 13(Doc5). https://doi.org/10.3205/000209 PMID: 26161049 4. Thomson P, Jones J, Browne M, Leslie SJ. Psychosocial factors that predict why people use comple- mentary and alternative medicine and continue with its use: a population based study. Complementary Therapies in Clinical Practice. 2014; 20(4):302–310. https://doi.org/10.1016/j.ctcp.2014.09.004 PMID: 25456023 5. Bishop FL, Yardley L, Lewith GT. A systematic review of beliefs involved in the use of complementary and alternative medicine. Journal of health psychology. 2007; 12(6):851–867. https://doi.org/10.1177/ 1359105307082447 PMID: 17956965 6. Browne M, Thomson P, Rockloff MJ, Pennycook G. Going against the herd: Psychological and cultural factors underlying the ‘vaccination confidence gap’. PLOS ONE. 2015; 10(9):e0132562. https://doi.org/ 10.1371/journal.pone.0132562 PMID: 26325522 7. Galbraith N, Moss T, Galbraith V, Purewal S. A systematic review of the traits and cognitions associated with use of and belief in complementary and alternative medicine (CAM). Psychology, Health & Medi- cine. 2018; 23(7):854–869. https://doi.org/10.1080/13548506.2018.1442010 PMID: 29468890 8. van den Bulck J, Custers K. Belief in complementary and alternative medicine is related to age and paranormal beliefs in adults. European journal of public health. 2010; 20(2):227–230. https://doi.org/10. 1093/eurpub/ckp104 PMID: 19587233 9. Stoneman P, Sturgis P, Allum N, Sibley E. Incommensurable worldviews? Is public use of complemen- tary and alternative medicines incompatible with support for science and conventional medicine? PLOS ONE. 2013; 8(1):e53174. https://doi.org/10.1371/journal.pone.0053174 PMID: 23382836 10. Kaur H, Chalia DS, Manchanda RK. Homeopathy in Public Health in India. Homeopathy: The Journal of the Faculty of Homeopathy. 2019; 108(2):76–87. https://doi.org/10.1055/s-0038-1673710 PMID: 30736082 11. Jansen E. The role of complementary and alternative medicine in the healthcare system: A German par- adox. Complementary Medicine Research. 2017; 24(5):290–294. https://doi.org/10.1159/000475549 PMID: 28934737 12. Relton C, Cooper K, Viksveen P, Fibert P, Thomas K. Prevalence of homeopathy use by the general population worldwide: a systematic review. Homeopathy. 2017; 106(2):69–78. https://doi.org/10.1016/j. homp.2017.03.002 PMID: 28552176 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284383 April 21, 2023 11 / 14 PLOS ONECorrelates of belief in complementary and alternative medicine (CAM) and homeopathy 13. Colquhoun D. Science degrees without the science. Nature. 2007; 446(7134):373–374. https://doi.org/ 10.1038/446373a PMID: 17377563 14. Bartens W. Homo¨ opathie: Fragwu¨ rdige Heiler akademisch geadelt. Su¨ ddeutsche Zeitung. 2018;. 15. Grill M, Hackenbroch V. Medizin: Ru¨ ckfall ins Mittelalter. DER SPIEGEL. 2010;. 16. Kemppainen LM, Kemppainen TT, Reippainen JA, Salmenniemi ST, Vuolanto PH. Use of complemen- tary and alternative medicine in Europe: Health-related and sociodemographic determinants. Scandina- vian journal of public health. 2018; 46(4):448–455. https://doi.org/10.1177/1403494817733869 PMID: 28975853 17. Hillienhof A. Mehr als jeder Zweite hat Erfahrungen mit Homo¨ opathie; 2020. 18. Statista. Homoopathische Arzneimittel—Umsatz in Deutschland nach Vertriebsweg bis 2018—Statista; 2021. Available from: https://de.statista.com/statistik/daten/studie/526063/umfrage/apothekenumsatz- mit-homoeopathischen-arzneimitteln-nach-vertriebsweg/. 19. Achstetter K, Teut M. Use of Self-Medication with Homeopathy in Germany: Results of an Online Ques- tionnaire Survey. Complementary Medicine Research. 2018; 25(6):383–390. https://doi.org/10.1159/ 000485077 PMID: 30286479 20. Bu¨ ssing A, Ostermann T, Heusser P, Matthiessen PF. Usage of alternative medical systems, acupunc- ture, homeopathy and anthroposophic medicine, by older German adults. Journal of Chinese Integra- tive Medicine. 2011; 9(8):847–856. 21. Beyerstein BL. Alternative medicine and common errors of reasoning. Academic medicine: journal of the Association of American Medical Colleges. 2001; 76(3):230–237. https://doi.org/10.1097/ 00001888-200103000-00009 PMID: 11242572 22. Slovic P, Finucane ML, Peters E, MacGregor DG. Risk as analysis and risk as feelings: some thoughts about affect, reason, risk, and rationality. Risk analysis: an official publication of the Society for Risk Analysis. 2004; 24(2):311–322. https://doi.org/10.1111/j.0272-4332.2004.00433.x PMID: 15078302 23. Sackett DL, Rosenberg WM, Gray JA, Haynes RB, Richardson WS. Evidence based medicine: what it is and what it isn’t. BMJ (Clinical research ed). 1996; 312(7023):71–72. https://doi.org/10.1136/bmj. 312.7023.71 PMID: 8555924 24. Kent DM, Steyerberg E, van Klaveren D. Personalized evidence based medicine: predictive approaches to heterogeneous treatment effects. BMJ (Clinical research ed). 2018; 363:k4245. https:// doi.org/10.1136/bmj.k4245 PMID: 30530757 25. Gorski DH, Novella SP. Clinical trials of integrative medicine: testing whether magic works? Trends in molecular medicine. 2014; 20(9):473–476. https://doi.org/10.1016/j.molmed.2014.06.007 PMID: 25150944 26. Bonne O, Shemer Y, Gorali Y, Katz M, Shalev AY. A randomized, double-blind, placebo-controlled study of classical homeopathy in generalized anxiety disorder. The Journal of clinical psychiatry. 2003; 64(3):282–287. https://doi.org/10.4088/jcp.v64n0309 PMID: 12716269 27. Aabel S. No beneficial effect of isopathic prophylactic treatment for birch pollen allergy during a low-pol- len season: a double-blind, placebo-controlled clinical trial of homeopathic Betula 30c. The British homoeopathic journal. 2000; 89(4):169–173. https://doi.org/10.1054/homp.1999.0440 PMID: 11055773 28. Mathie RT, Ramparsad N, Legg LA, Clausen J, Moss S, Davidson JRT, et al. Randomised, double- blind, placebo-controlled trials of non-individualised homeopathic treatment: systematic review and meta-analysis. Systematic reviews. 2017; 6(1):63. https://doi.org/10.1186/s13643-017-0445-3 PMID: 28340607 29. 30. Linde K, Clausius N, Ramirez G, Melchart D, Eitel F, Hedges LV, et al. Are the clinical effects of homo- eopathy placebo effects? A meta-analysis of placebo-controlled trials. Lancet. 1997; 350(9081):834– 843. https://doi.org/10.1016/S0140-6736(97)02293-9 PMID: 9310601 Linde K, Melchart D. Randomized controlled trials of individualized homeopathy: a state-of-the-art review. Journal of Alternative and Complementary Medicine. 1998; 4(4):371–388. https://doi.org/10. 1089/acm.1998.4.371 PMID: 9884175 31. Kleijnen J, Knipschild P, ter Riet G. Clinical trials of homoeopathy. BMJ (Clinical research ed). 1991; 302(6772):316–323. https://doi.org/10.1136/bmj.302.6772.316 PMID: 1825800 32. Shang A, Huwiler-Mu¨ ntener K, Nartey L, Ju¨ ni P, Do¨rig S, Sterne JAC, et al. Are the clinical effects of homoeopathy placebo effects? Comparative study of placebo-controlled trials of homoeopathy and allo- pathy. The Lancet. 2005; 366(9487):726–732. https://doi.org/10.1016/S0140-6736(05)67177-2 PMID: 16125589 33. National Health and Medical Research Council. NHMRC information paper: Evidence on the effective- ness of homeopathy for treating health conditions. National Health and Medical Research Council; 2015. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284383 April 21, 2023 12 / 14 PLOS ONECorrelates of belief in complementary and alternative medicine (CAM) and homeopathy 34. Markun S, Maeder M, Rosemann T, Djalali S. Beliefs, endorsement and application of homeopathy dis- closed: a survey among ambulatory care physicians. Swiss medical weekly. 2017; 147:w14505. https:// doi.org/10.4414/smw.2017.14505 PMID: 29039627 35. Bu¨ cker B, Groenewold M, Schofer Y, Scha¨ fer T. The use of complementary alternative medicine (CAM) in 1001 German adults: results of a population-based telephone survey. Gesundheitswesen. 2008; 70 (8-9):e29–36. https://doi.org/10.1055/s-2008-1081505 36. Steel A, Cramer H, Leung B, Lauche R, Adams J, Langhorst J, et al. Characteristics of Homeopathy Users among Internal Medicine Patients in Germany. Forschende Komplementarmedizin (2006). 2016; 23(5):284–289. https://doi.org/10.1159/000450818 37. Tversky A, Kahneman D. Judgment under Uncertainty: Heuristics and Biases. Science. 1974; 185 (4157):1124–1131. https://doi.org/10.1126/science.185.4157.1124 PMID: 17835457 38. Stanovich KE, West RF. Individual differences in reasoning: implications for the rationality debate? The Behavioral and Brain Sciences. 2000; 23(5):645–665; discussion 665-726. https://doi.org/10.1017/ S0140525X00003435 PMID: 11301544 39. Moscovici S. The phenomenon of social representations. In: R M Farr & S Moscovici, editor. Social rep- resentations. Cambridge University Press; 1984. p. 3–69. 40. Frederick S. Cognitive Reflection and Decision Making. The Journal of Economic Perspectives. 2005; 19(4):25–42. https://doi.org/10.1257/089533005775196732 41. Gervais WM, Norenzayan A. Analytic thinking promotes religious disbelief. Science. 2012; 336 (6080):493–496. https://doi.org/10.1126/science.1215647 PMID: 22539725 42. 43. Toplak ME, West RF, Stanovich KE. Assessing miserly information processing: An expansion of the Cognitive Reflection Test. Thinking & Reasoning. 2014; 20(2):147–168. https://doi.org/10.1080/ 13546783.2013.844729 van Prooijen JW, Douglas KM, de Inocencio C. Connecting the dots: Illusory pattern perception predicts belief in conspiracies and the supernatural. European journal of social psychology. 2018; 48(3):320– 335. https://doi.org/10.1002/ejsp.2331 PMID: 29695889 44. Whitson JA, Galinsky AD. Lacking control increases illusory pattern perception. Science (New York, NY). 2008; 322(5898):115–117. https://doi.org/10.1126/science.1159845 PMID: 18832647 45. Smith BW, Dalen J, Wiggins KT, Christopher PJ, Bernard JF, Shelley BM. Who is willing to use comple- mentary and alternative medicine? EXPLORE. 2008; 4(6):359–367. https://doi.org/10.1016/j.explore. 2008.08.001 PMID: 18984547 46. 47. Lindeman M. Biases in intuitive reasoning and belief in complementary and alternative medicine. Psychology & health. 2011; 26(3):371–382. https://doi.org/10.1080/08870440903440707 PMID: 20419560 Lamberty P, Imhoff R. Powerful Pharma and Its Marginalized Alternatives? Social Psychology. 2018; 49(5):255–270. https://doi.org/10.1027/1864-9335/a000347 48. Betsch T, Aßmann L, Glo¨ckner A. Paranormal beliefs and individual differences: story seeking without reasoned review. Heliyon. 2020; 6(6):e04259. https://doi.org/10.1016/j.heliyon.2020.e04259 PMID: 32637687 49. Ellison CG, Bradshaw M, Roberts CA. Spiritual and religious identities predict the use of complementary and alternative medicine among US adults. Preventive Medicine. 2012; 54(1):9–12. https://doi.org/10. 1016/j.ypmed.2011.08.029 PMID: 21893085 50. Hildreth KD, Elman C. Alternative Worldviews and the Utilization of Conventional and Complementary Medicine. Sociological Inquiry. 2007; 77(1):76–103. https://doi.org/10.1111/j.1475-682X.2007.00178.x 51. Lindeman M, Svedholm-Ha¨kkinen AM, Lipsanen J. Ontological confusions but not mentalizing abilities predict religious belief, paranormal belief, and belief in supernatural purpose. Cognition. 2015; 134:63– 76. https://doi.org/10.1016/j.cognition.2014.09.008 PMID: 25460380 52. Klein O, Hardwicke TE, Aust F, Breuer J, Danielsson H, Mohr AH, et al. A Practical Guide for Transpar- ency in Psychological Science. Collabra: Psychology. 2018; 4(1). https://doi.org/10.1525/collabra.158 53. Nosek BA, Alter G, Banks GC, Borsboom D, Bowman SD, Breckler SJ, et al. Scientific standards: Pro- moting an open research culture. Science. 2015; 348(6242):1422–1425. https://doi.org/10.1126/ science.aab2374 54. Wagenmakers EJ, Wetzels R, Borsboom D, van der Maas HLJ, Kievit RA. An Agenda for Purely Confir- matory Research. Perspectives on Psychological Science. 2012; 7(6):632–638. https://doi.org/10. 1177/1745691612463078 PMID: 26168122 55. Faul F, Erdfelder E, Lang AG, Buchner A. G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods. 2007; 39(2):175–191. https://doi.org/10.3758/BF03193146 PMID: 17695343 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284383 April 21, 2023 13 / 14 PLOS ONECorrelates of belief in complementary and alternative medicine (CAM) and homeopathy 56. Albert H. Traktatu¨ ber rationale Praxis. vol. 22 of Die Einheit der Gesellschaftswissenschaften. Tubingen: Mohr; 1978. 57. 58. Lindeman M, Aarnio K. Superstitious, magical, and paranormal beliefs: An integrative model. Journal of Research in Personality. 2007; 41(4):731–744. https://doi.org/10.1016/j.jrp.2006.06.009 Lee K, Ashton MC. Psychometric Properties of the HEXACO-100. Assessment. 2018; 25(5):543–556. https://doi.org/10.1177/1073191116659134 PMID: 27411678 59. Peters E, Dieckmann N, Dixon A, Hibbard JH, Mertz CK. Less is more in presenting quality information to consumers. Medical care research and review: MCRR. 2007; 64(2):169–190. https://doi.org/10. 1177/10775587070640020301 PMID: 17406019 60. Schlink S, Walther E. Kurz und gut: Eine deutsche Kurzskala zur Erfassung des Bedu¨ rfnisses nach kog- nitiver Geschlossenheit. Zeitschrift fu¨ r Sozialpsychologie. 2007; 38(3):153–161. https://doi.org/10. 1024/0044-3514.38.3.153 61. Webster DM, Kruglanski AW. Individual differences in need for cognitive closure. Journal of Personality and Social Psychology. 1994; 67(6):1049–1062. https://doi.org/10.1037/0022-3514.67.6.1049 PMID: 7815301 62. Bless H, Wa¨nke M, Bohner G, Fellhauer RF, Schwarz N. Need for cognition: eine Skala zur Erfassung von Engagement und Freude bei Denkaufgaben: Need for cognition: a scale measuring engagement and happiness in cognitive tasks. Zeitschrift fur Sozialpsychologie. 1994; 25. 63. McLain DL. Evidence of the properties of an ambiguity tolerance measure: the Multiple Stimulus Types Ambiguity Tolerance Scale-II (MSTAT-II). Psychological Reports. 2009; 105(3 Pt 1):975–988. https:// doi.org/10.2466/PR0.105.3.975-988 PMID: 20099561 64. Beierlein C, Kovaleva A, La´ szlo Z, Kemper CJ, Rammstedt B. Eine Single-Item-Skala zur Erfassung der Allgemeinen Lebenszufriedenheit: Die Kurzskala Lebenszufriedenheit-1 (L-1). GESIS-Working Papers, 2014/33. Mannheim: GESIS—Leibniz-Institut fu¨ r Sozialwissenschaften; 2014. 65. 66. Templer DI. The construction and validation of a Death Anxiety Scale. The Journal of general psychol- ogy. 1970; 82(2d Half):165–177. https://doi.org/10.1080/00221309.1970.9920634 PMID: 4394812 Field AP. Discovering statistics using IBM SPSS statistics. 5th ed. SAGE edge. London: SAGE Publi- cations; 2018. 67. Cohen J. Statistical power analysis for the behavioral sciences. 2nd ed. Lawrence Erlbaum; 1988. 68. Davey HM, Barratt AL, Butow PN, Deeks JJ. A one-item question with a Likert or Visual Analog Scale adequately measured current anxiety. Journal of Clinical Epidemiology. 2007; 60(4):356–360. https:// doi.org/10.1016/j.jclinepi.2006.07.015 PMID: 17346609 69. Konrath S, Meier BP, Bushman BJ. Development and validation of the Single Item Narcissism Scale (SINS). PLOS ONE. 2014; 9(8):e103469. https://doi.org/10.1371/journal.pone.0103469 PMID: 25093508 70. Flanagin AJ, Winter S, Metzger MJ. Making sense of credibility in complex information environments: the role of message sidedness, information source, and thinking styles in credibility evaluation online. Information, Communication & Society. 2020; 23(7):1038–1056. https://doi.org/10.1080/1369118X. 2018.1547411 71. Hoeppner BB, Kelly JF, Urbanoski KA, Slaymaker V. Comparative utility of a single-item versus multi- ple-item measure of self-efficacy in predicting relapse among young adults. Journal of Substance Abuse Treatment. 2011; 41(3):305–312. https://doi.org/10.1016/j.jsat.2011.04.005 PMID: 21700411 72. Bergkvist L, Rossiter JR. The Predictive Validity of Multiple-Item versus Single-Item Measures of the Same Constructs. Journal of Marketing Research. 2007; 44(2):175–184. https://doi.org/10.1509/jmkr. 44.2.175 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284383 April 21, 2023 14 / 14 PLOS ONE
10.1371_journal.pone.0286685	RESEARCH ARTICLE Early warning model and prevention of regional financial risk integrated into legal system Yanyu ZhuangID 1, Hua Wei2 1 KoGuan School of Law, Shanghai Jiao Tong University, Shanghai, China, 2 Antai College of Economics and Management, Shanghai Jiao Tong University, Shanghai Development Strategy Research Institute, Shanghai, China zhuangyanyu@sjtu.edu.cn Abstract a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Zhuang Y, Wei H (2023) Early warning model and prevention of regional financial risk integrated into legal system. PLoS ONE 18(6): e0286685. https://doi.org/10.1371/journal. pone.0286685 Editor: Simon Grima, University of Malta, MALTA Received: April 2, 2023 Accepted: May 20, 2023 Published: June 2, 2023 Copyright: © 2023 Zhuang, Wei. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Funding: The author(s) received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. In order to improve the laws and regulations of the financial system, in the construction of laws and regulations, the traditional financial risk Early Warning (EW) model is optimized. The financial prevention and control measures with legal protection are implemented to warn the financial risks, which plays an important role in the construction of the rule of law in the Financial Market (FM) and the establishment of financial risk prevention and control laws and regulations. This paper combines the deep learning model and the Markov regime Switching Vector Auto Regression (MS-VAR) model and constructs a regional financial risk EW model from the following aspects: macroeconomic operation EW indicators, regional economic risk EW indicators, regional financial institution risk EW indicators. The model is empirically researched and analyzed. The results show that the fluctuation trend of the mac- roeconomic pressure index in the time series is relatively large, and the overall fluctuation of the regional economic pressure index is small, and fluctuates around 0 in most periods. After the financial crisis, local governments stepped up their supervision of non-performing corporate and household loans. From 2011 to 2018, the non-performing loan ratio began to decline, and the overall fluctuation of the regional financial comprehensive stress index was small, fluctuating around 0. Due to the lack of legal regulation, from the perspective of the regional economy, the risk level is more likely to change from low risk to moderate risk, while the risk status is less likely to change from high risk to moderate risk. From the perspective of regional financial institutions, the probabilities of maintaining low risk and moderate risk are 0.98 and 0.97, respectively, which is stronger than maintaining the stability of high risk. From the perspective of the state transition of the regional financial risk composite index, the probability of maintaining low risk and high risk is 0.97 and 0.93, which is higher than main- taining the stability of medium risk. The Deep Learning (DL) regional financial risk EW MS- VAR model has strong risk prediction ability. The model can better analyze the conversion probability of regional financial risk EW index and has better risk EW ability. This paper enhances the role of legal systems in financial risk prevention and control. The regional financial risk EW model incorporating financial legal indicators can better describe the regional financial risk level, and the EW results are basically consistent with the actual situa- tion. In order to effectively prevent financial risks and ensure the safety of the financial PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 1 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system system, it is recommended that the government improve local debt management, improve financial regulations and systems, and improve the legislative level of financial legal supervision. Introduction In the decades of social and economic development in China, the financial industry plays a vital role in the market economy and is the key to the market economy. In order to improve the laws and regulations of the financial system, the traditional financial risk early warning model is optimized in the construction of laws and regulations. The implementation of legally protected financial prevention and control measures and early warning of financial risks play an important role in the legal construction of Financial Market (FM) and the establishment of laws and regulations on financial risk prevention and control. FM malfunctions occur from time to time, and the risk probability increases significantly. Financial risk has the characteris- tics of high risk and high crisis, which is likely to cause serious harm to the market economy [1]. In particular, Regional Financial Risks (RFRs) have strong linkage and spread, and it is easy to induce national financial risks and even global financial crisis through ripple effect [2]. Under the downward pressure of huge Economic Growth (EG), various uncertain factors appear. Therefore, Early Warning (EW), prevention and control of RFRs have become an important task of macro-control in China [3]. Therefore, it is essential to emphasize the imple- mentation of effective financial risk prevention measures. Meanwhile, relevant laws and regu- lations must be improved to make FM develop harmoniously and orderly. Although national legal supervision can effectively monitor risks, effective financial risk EW measures can make the FM develop more harmoniously and orderly, make the whole market economy run more smoothly, and bring higher economic benefits to the society [4]. Strengthening RFRs-oriented EW has important practical significance for preventing financial risks and ensuring stable regional EG [5]. Financial behavior is highly related to the economy and society. All walks of life have paid more attention to the risks in the financial field. As a popular technology in Artificial Intelligence (AI), Deep Learning (DL) can model abstract high-level features of various data with multiple processing layers and nonlinear transforma- tion [6]. Meanwhile, DL can find appropriate and effective features from complex data by pro- cessing big data, learning features through training, and using multi-layer perceptron models to supervise unsupervised data learning. The deeper the model is, the more accurate the feature expression will be [7]. Chen et al. (2020) used Deep Neural Network (DNN) to model the evac- uation of subway station buildings, and carried out simulation experiments. Comparing the Convolutional Neural Network (CNN) model with the pre-training model by classifying data sets, the accuracy and training speed of the proposed model are verified [8]. Chen et al. (2021) used DL technology to model the network security system of smart cities, reducing the net- work security risks [9]. DL can also be used in financial analysis to predict commodity prices, financial events, financial risks and other hot issues. For example, Zhou et al. (2021) applied DL to the financial risk early warning of real estate enterprises. They took real estate as an example to make a concrete demonstration and analysis [10]. For RFRs-oriented electronic warfare, scholars have also conducted relevant research on risk causes, propagation paths, indicators, and model selection [11]. Du et al. (2021) established a scientific and effective ori- ented electronic warfare system based on Big Data Technology (BDT). They integrated a lot of data and introduced related risk index [3]. In selecting financial risk EW indexes, the Evalua- tion Index System (EIS) mainly focuses on currency crises, bond crises, or banking crises and PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 2 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system fails to reflect the systemic financial risk fully. Moreover, the traditional financial risk EW sys- tem is mainly based on the linear model [12]. Based on the above theory, this paper improves the selection of financial risk EW indicators and risk EW models. Section 1 describes the purpose of writing the article and literature review. In Section 2, the characteristics, incentive mechanism and risk factors of regional finance are summarized, and the regional financial risk EW system is constructed by using DL model and Markov regime switching vector autoregression (MS-VAR) model. Then, the regional financial risk EW index and its comprehensive index are constructed, and the regional financial risk EW MS-VAR model based on DL is constructed. Section 3 analyzes the comprehensive index of economic stress from two aspects: macro-economy and regional economy, and regional financial stress from two aspects: regional financial institutions and regional finance. This model is used to test the EW of risks in different dimensions and predict regional financial risks. Section 4 summarizes the main results of this study and analyzes the future research direction. The innovation of this paper is to integrate DL model and financial legal system into the construction of regional financial risk EW model, and analyze the com- prehensive indicators of regional financial risk EW model from many aspects. The design aims to make the EW of RFRs conform to the development characteristics of the local financial industry and improve the predictability of the EW model of financial risk. This discovery pro- vides a reference for subsequent scholars to study financial risk EW. Theoretical research and method design of regional financial risk Overview of RFRs theory 1. Basic concepts and characteristics of RFRs. RFRs is a variety of risks in the financial industry within a certain economic range. According to the impact region and performance characteristics, financial risks are divided into macro, regional, and micro levels. In particular, RFRs belongs to the meso level [13]. The formation of financial risks within a specific range generally has three-level factors. The micro-level risk spreads within a specific range and has the characteristics of top-to-bot- tom development. Risks spreading in some highly different economic ranges belong to hori- zontal risk. Finally, the macro-level risks accumulate and spread in the financial industry system [14]. RFRs have the general characteristics of financial risks, such as objective existence, controllability, and negative impact, and [15] at the same time, have unique characteristics. For example, the RFRs formation mechanisms are special. The risk harms limited areas, and there is a remarkable effect in RFRs-oriented EW [16]. Regional financial risks mainly consider the impact of four aspects: the level of macroeco- nomic development, the level of regional industrial development, the development of regional financial institutions, and the financial legal supervision system [17]. The details are shown in Fig 1. As shown in Fig 1, from the perspective of the macroeconomic level, economic risk is an essential factor in financial risks. RFRs is closely related to the market macroeconomic factors of stocks, bonds, currencies, and real estate. The economic slowdown, the industrial structural transformation, the real estate market downturn, and investors’ lack of confidence will all affect regional finance’s development. Local economic development indirectly affects regional financial stability. Regional financial institutions are mainly banks. If the non-performing loan ratio of the banking industry increases, it will induce a run on the banking industry, which will easily induce RFRs. Loopholes in the financial and legal supervision system, the applicable fuzzy boundary of legal norms, and the lack of governance and authority of local regulations are also important factors causing RFRs. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 3 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system Fig 1. Influencing factors of RFRs. https://doi.org/10.1371/journal.pone.0286685.g001 2. Financial risk incentive theory. The theory of financial risk incentives includes five aspects: financial business cycle theory, financial vulnerability theory, financial asset price fluc- tuation theory, bank run theory, and lack of legal supervision [18]. Fig 2 shows the details. As shown in Fig 2, the financial business cycle refers to the periodic changes in financial and economic operation under the joint action of internal and external factors. From the Fig 2. The financial risk incentive theory. https://doi.org/10.1371/journal.pone.0286685.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 4 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system perspective of financial development, this work studies the dynamic change mechanism in dif- ferent stages of the financial cycle. The balance sheet and bank loan mechanism are the pri- mary factors that increase financial risks. The theory of financial fragility refers to the accumulation of internal risks formed by its high debt business model. The risks mainly come from the highly leveraged business strategy, the lack of legal and regulatory constraints of emerging financial institutions, the growth of asset foam caused by speculative investment, and the accelerated accumulation of risks caused by information asymmetry. The fluctuation law of financial asset prices also differs significantly from the trend of macroeconomic opera- tion. The banking industry’s assets mainly come from the deposits of depositors, which makes the liquidity of bank assets worse than that of liabilities. When banks realize asset appreciation in the form of short deposits and long loans, the uncertainty of depositors’ demand is the main factor causing bank runs. The lack of legal supervision is the lack of clear legal regulation of financial supervision, imperfect financial supervision regulations, and lack of legal texts for risk prevention and control. 3. The transmission channel of RFRs. The financial risk transmission mechanism mainly includes the transmission of trade, financial channels, and two-way transmission between the real and financial industries. The main transmission path is the two-way transmission between the real and financial industries [19]. The specific transmission path is demonstrated in Fig 3. According to Fig 3, trade channel transmission mainly refers to the risk transmission between two regions with trade exchanges. Financial channel transmission refers to the out- bound regional capital transfer induced by financial risks through capital flow, bank lending, and financial product investments. Meanwhile, unclear definitions of FM laws and regulations and the division of supervision power unregulated by legal norms and guidelines make the entire FM the most important channel for risk transmission. Essentially, the two-way trans- mission mechanism between the real economy and the financial industry is the two-way risk transmission of financial risks through the real economy and the FM. RFRs-oriented EW system 1. DL model theory. Deep Neural Network (DNN) is an effective machine learning algo- rithm for DL. It learns the inherent laws and representation levels of sample data. It can automati- cally learn data features and complete tasks such as classification and regression [20]. Its ultimate goal is to enable the machine to have the same analysis and autonomous learning ability as people and recognize characters, images, and other data [21]. DL extracts features layer by layer by min- ing the underlying feature distribution of the data, with multiple hidden layers. The hidden layer connects the input and output layers. The structure of the DNN model is unfolded in Fig 4. As explained in Fig 4, unsupervised learning from the input layer to the output layer is used in the DNN model. DNN starts from the input layer and trains layer by layer to the top layer. The parameters of each layer are trained layer by layer without calibration data. This training method can be regarded as an unsupervised training process [22]. The idea of DL feature extraction is applied to the construction of financial risk EW model. The DL data is used to mine features, ana- lyze regional financial risk indicators, and mine financial risk and influencing factor indicators. 2. RFRs-oriented EW model. The traditional financial risk EW model includes the Fre- quency Ratio (FR) model. FR is based on the influencing factors of currency risk. The FR model expressed in Eq (1): PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 PðY ¼ 1Þ ¼ Fðx; qÞ PðY ¼ 0Þ ¼ 1 (cid:0) Fðx; qÞ ð1Þ 5 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system Fig 3. RFRs transmission channels. https://doi.org/10.1371/journal.pone.0286685.g003 In Eq (1), Y is the financial risk variable. Y = 1 indicates the outbreak of financial risk. x rep- resents the influencing factor of financial risk. q denotes the parameter vector of x. The joint probability of induced variables is used to measure the outbreak probability of financial risk. Suppose N countries are measured, and the sample period is 1,2,� � �,T. In that case, p{i, t} is estimated by Eq (2) [23]: ( pfi; tg ¼ 1; When risk occurs in country i at moment t 0; When no risks occur in country i at moment t ð2Þ The comprehensive result of country i in period t is x{i, t}. MS-VAR model can well analyze the structural changes between variables and predict the future data change rules based on summarizing the historical data change rules. The specific model is as follows: PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 6 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system Fig 4. DNN model. https://doi.org/10.1371/journal.pone.0286685.g004 The vector autoregressive model composed of the k-dimensional time series yt = (y1t,� � �, ykt)0 is expressed by Eq (3): yt ¼ v þ A1yt(cid:0) 1 þ � � � þ Apyp(cid:0) 1 þ mt ð3Þ In Eq (3), t = 1,2,� � �,T, μt~IID(0,S), and y0,� � �,yt−p are the determined variable. Suppose the error variable conforms to the normal distribution, or μt~IID(0,S). In that case, Eq (3) can be PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 7 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system expressed as the intercept term of VAR(p) model. Its expression is given in Eq (4) [24]: yt(cid:0) 1 (cid:0) m ¼ A1ðyt(cid:0) 1 (cid:0) mtÞ þ � � � þ Apðyt(cid:0) p (cid:0) mÞ þ mt In Eq (4), μ is the k×1 mean form of yt. The calculation of μ reads: m ¼ ðIk (cid:0) Pp j¼1 AJÞ(cid:0) 1(cid:0) ð4Þ ð5Þ When the time series is affected by the change in the regime system, it is assumed that the regime switch variable St2{1,� � �,M} is a Markov chain in a discrete state. The conversion prob- ability is counted by Eq (6): Pij ¼ PrðSt(cid:0) 1 ¼ j j St ¼ iÞ PM j¼1 Pij ¼ 1 8i; j 2 f1; � � � ; Mg ð6Þ When the order of the MS-VAR model is P, the specific performance of the Regime Switch Model (SRM) reads: yt (cid:0) mðStÞ ¼ A1ðStÞðyt(cid:0) 1 (cid:0) mðSt(cid:0) 1ÞÞ þ � � � þ ApðStÞðyt(cid:0) p (cid:0) mðSt(cid:0) pÞÞmt ð7Þ In Eq (7), μ(St), A1(St), and Ap(St) are parameters of μ. A1� � �,Ap is applicable to the parame- ter function of regime St. μ(St) is calculated by Eq (8): mðStÞ ¼ 8 >>< >>: m1St ¼ 1 . .. mMSt ¼ M ð8Þ The RSM is added to the intercept term to achieve a reasonable state of regime switch and mean smooth switch [25]. The results are reflected in Eq (9): yt ¼ vðStÞ þ A1ðStÞyt(cid:0) 1 þ � � � þ ApðStÞyt(cid:0) p þ mt ð9Þ RFRs-oriented EW index and its Composite Index (CI) construction 1. RFRs-oriented EW index. This section takes Shanghai as an example to construct the RFRs-oriented EW indexes from four aspects: macroeconomic operation risk, regional eco- nomic risk, regional financial institution risk, and financial and legal supervision system risk [26]. Macroeconomic operation index data are selected from government departments, stock market, bond market, and foreign exchange market [27]. Regional economic risk index data take the dimensions of government departments and enterprises as secondary indexes. The regional financial institution index considers the banking and insurance industries as second- ary indexes. By comparison, the financial and legal supervision index takes the local legal norm system and the legal norm of the supervision subject as the secondary indexes. The index selection is specified in Fig 5. Macroeconomic fluctuation is a cyclical change from depression to recovery and then to cli- max. In terms of the leading index, economic indexes represented by macroeconomic prosper- ity and entrepreneur confidence index are selected to improve the RFRs-oriented EIS further. Regional economic risk is mainly reflected in the transfer of local debt risk to the risk of the financial system and the risk transfer from the real industry to the financial system. The strength of risk management and control of financial institutions directly affects the stability of the FM, with banking and insurance as secondary indexes. The local legal norm system is PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 8 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system Fig 5. RFRs-oriented EW indexes. (a) Macroeconomic operation risk EW index; (b) Regional economic risk EW index; (c) Regional financial institution risk EW index; (d) Financial legal supervision system risk EW index. https://doi.org/10.1371/journal.pone.0286685.g005 reflected in two aspects: the division of financial supervision power and the division of punish- ment for illegal acts. The legal norms of the supervision subject are embodied in the classifica- tion and standardization of the supervision subject. 2. RFRs CI. In view of the characteristics of China’s financial system and the regional characteristics of financial risks, taking Shanghai as an example, this paper analyzes the macro- economic and regional economy, regional financial institutions and regional financial risk index from 2002 to 2020. The data are derived from Wind, eastmoney.com and cnfin.com and other relevant economic and financial websites. Against the characteristics of China’s financial system and the regional characteristics of RFRs, the CI method is used to construct the RFRs- oriented EIS. It is used as the basic variable of the financial risk EW model [28]. The CI method can be converted according to variable changes and combined with various risk iden- tification and EW models. Fig 6 lists the details. Fig 6(A) is the macroeconomic and regional economic risk index from 2002 to 2020, and Fig 6(B) is the regional financial institution and regional financial risk index from 2002 to 2020. The economic risk index and the regional financial risk index are variable. With the strengthening of government supervision, the risk index has decreased. DL-based MS-VAR model for RFRs EW Combined with the DL model, the regional financial risk EW indicators are analyzed, and on this basis, the regional financial risk EW MS-VAR model is constructed. In order to strengthen the EW ability of the MS-VAR model, EW inspections are carried out on risks of different PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 9 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system Fig 6. RFRs CI. (a) Economic risk index; (b) RFRs index. https://doi.org/10.1371/journal.pone.0286685.g006 dimensions, that is, risk state transition identification is carried out, and transition probability identification is carried out for three different risk levels. It includes low risk level, medium risk level, and high risk level [29]. The process is as follows: rt ¼ mt þ εt; εt � Nð0; s2 t Þ P3 i¼1 miSits2 t ¼ mt ¼ P3 i¼1 s2 i Sit pij ¼ PrðSt ¼ j j St(cid:0) 1 ¼ i; St(cid:0) 2 ¼ k; St(cid:0) 3 ¼ l; � � � � � �Þ ¼ PrðSt(cid:0) 1 ¼ iÞ ( Sit ¼ 1; St ¼ i 0; others ð10Þ ð11Þ ð12Þ ð13Þ In Eqs (10)—(13), St is the state variable. St = 1,2,3 respectively represent a low-level finan- cial risk, medium level financial risk, and high-level financial risk. {St} denotes a first-order Markov chain. The state variable St−1 at the previous moment determines St at moment t. pij stands for state conversion probability. The conversion probability matrix P of St, is obtained by Eq (14): 2 6 4 P ¼ 3 7 5 p11 p12 p13 p21 p22 p23 p31 p32 p33 ð14Þ In Eq (14), pij2[0,1], and It−1 indicates the information set of rt corresponding to moment t−1. Then, the joint density i¼1 pij ¼ 1; i = 1,2,3. P3 function is expressed by Eq (15): f ðrt j It(cid:0) 1Þ ¼ P3 St(cid:0) 1 P3 St(cid:0) 1¼1 f ðrt j St; St(cid:0) 1; It(cid:0) 1Þ � PrðSt; St(cid:0) 1 j It(cid:0) 1Þ ð15Þ PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 10 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system Suppose rt is known. In that case, the change of joint distribution probability is calculated by Eq (16): PrðSt ¼ j; St(cid:0) 1 j ItÞ ¼ ¼ PrðSt ¼ j; St(cid:0) 1 ¼ i j It(cid:0) 1; rtÞ PrðSt ¼ j; St(cid:0) 1 ¼ i; rt j It(cid:0) 1Þ f ðrt j It(cid:0) 1Þ f ðrt j St ¼ j; St(cid:0) 1 ¼ i; It(cid:0) 1Þ � PrðSt ¼ j; St(cid:0) 1 ¼ i j It(cid:0) 1Þ P3 P3 ¼ St(cid:0) 1 St(cid:0) 1¼1 f ðrt j St ¼ j; St(cid:0) 1 ¼ i; It(cid:0) 1Þ � PrðSt ¼ j; St(cid:0) 1 ¼ i j It(cid:0) 1Þ The filter probability Pr(St = j\|It) is converted, and the result is given in Eq (17): P3 PrðSt ¼ j j ItÞ ¼ i¼1 PrðSt ¼ j; St(cid:0) 1 j ItÞ ð16Þ ð17Þ Based on the filter alternation effect, the smoothing probability Pr(St = j\|It) is calculated. The smaller the smoothing probability estimation is, the smaller the probability that the moment t is at the ith volatility level [30]. Empirical research and analysis on risk EW Economic pressure CI The results are shown in Fig 7 from the macro-economy and regional economy perspectives. As explained in Fig 7, the macroeconomic pressure index has a large fluctuation trend in the time series. The macro-economy is subject to external shocks, and the pressure posed by potential risks continues to accumulate. When the impact utility expands rapidly, a pressure Fig 7. Economic pressure CI. https://doi.org/10.1371/journal.pone.0286685.g007 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 11 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system risk erupts. The overall fluctuation of the regional economic pressure index is small. It fluctu- ates up and down at 0 in most periods, and the larger fluctuation range appears around 2004. In particular, the fluctuation trend of the regional economic pressure index is different from the macroeconomic dimension. This is mainly because the impact of the global financial crisis on regional economic development is significantly less than the macroeconomic fluctuation. Regional financial pressure CI Fig 8 analyzes the regional financial pressure CI from the two aspects of regional financial institutions and regional banking. As shown in Fig 8, the fluctuation range of the dimensional pressure index of regional financial institutions is significantly higher than that of the regional banking dimension. From 2004 to 2010, the non-performing loan ratio of the banking industry was always at a high level. After the financial crisis, the banking risks were released to a certain extent, and the legal supervision of local governments was strengthened. At the same time, the supervision of non- performing enterprises and residents’ loans was strengthened, which greatly reduced the non- performing loan ratio of the regional banking industry. From 2011 to 2018, the non-perform- ing loan ratio began to decline, and the pressure index of financial institutions entered a gentle state. The overall fluctuation range of the regional financial pressure CI is small, floating up and down the zero value. The financial pressure CI has changed more dramatically during 2004–2006, 2007, 2009, and 2016. MS-VAR EW inspection EW tests are performed on risks of different dimensions. Risk state conversion identification is mainly based on three levels: high risk, medium risk, and low risk, as revealed in Fig 9. Fig 8. Financial pressure CI. https://doi.org/10.1371/journal.pone.0286685.g008 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 12 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system Fig 9. EW test of financial risks in different dimensions. (a) Conversion probability of macroeconomic risk level; (b) Conversion probability of regional economic risk level; (c) Conversion probability of regional financial structure dimension. https://doi.org/10.1371/journal.pone.0286685.g009 As revealed in Fig 9, in terms of the macroeconomic dimension, the probability of main- taining the same state of transition among the three regional systems of low, medium and high risk is all 0.91. This indicates that the stability of the transition between regional systems is rela- tively strong, and the probability of maintaining the original risk level is high. It is easier to convert regional economic risk from low risk to moderate risk, but it is more difficult to con- vert from high risk to moderate risk. From the perspective of regional financial institutions, the probability of maintaining low risk, medium risk, and high risk is 0.98, 0.97, and 0.76, respectively, and the stability of low risk and medium risk is stronger than that of high risk. State conversion identification of RFRs CI The state conversion identification of RFRs CI is described in Fig 10. As described in Fig 10, the probability of maintaining low risk is 0.97, and the probability of maintaining high risk is 0.93. The stability of the two-zone system is strong, but the probability of maintaining moderate risk is less than 0.3, indicating that the moderate risk fluctuates greatly. The dimension mainly analyzes the state transition probability of low risk and high risk. The probability of high risk and low risk level conversion is high. RFRs prediction 1. Prediction test of regional financial pressure index. Based on Shanghai, the fitting trend of the regional financial pressure index from 2002 to 2020 is analyzed in Fig 11. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 13 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system Fig 10. Conversion probability between different risk Cis. https://doi.org/10.1371/journal.pone.0286685.g010 As illustrated in Fig 11, the observed value of the fitted regional financial risk EW compre- hensive pressure index has a high degree of coincidence with the fitted value, and the observed value and the fitted value curve have a good agreement as a whole, indicating that the DL regional financial risk EW MS-VAR model has strong predictive ability and high credibility of the predicted data. 2. Prediction test for RFRs. The proposed RFRs-oriented EW model is used to estimate the three-regime conversion probability of the RFRs CI. The results are plotted in Fig 12. According to Fig 12, the probability of low risk maintaining low is 0.90, and the probability of high risk maintaining high is 0.88. The stability of the two-regime system is strong, but the probability of medium risk conversion is relatively low. Thus, the fluctuation of medium risk is large. Therefore, the conversion probability of risk CI mainly analyzes low risk and high risk. The probability of conversion from low risk to medium risk is high. The regional financial risk EW MS-VAR model of DL can better analyze the conversion probability of regional finan- cial risk EW index. Discussion Based on the DL model and MS-VAR model, this paper constructs the regional financial risk EW model from three aspects: macroeconomic operation EW index, regional economic risk EW index, and regional financial institution risk EW index. The model is empirically studied and analyzed. The results show that the macroeconomic pressure index fluctuates greatly in time series, while the regional economic pressure index fluctuates slightly in general, and fluc- tuates around 0 in most periods. From 2011 to 2018, the non-performing loan ratio began to PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 14 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system Fig 11. Fitting results of regional financial pressure index. https://doi.org/10.1371/journal.pone.0286685.g011 Fig 12. Conversion probability of RFRs CI. https://doi.org/10.1371/journal.pone.0286685.g012 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 15 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system decline, and the overall regional financial comprehensive stress index fluctuated slightly, fluc- tuating around 0. EW MS-VAR model of DL regional financial risk has strong risk prediction ability. The model can well analyze the conversion probability of regional financial risk EW indicators and has good risk EW ability. In this research direction, literature [23] gives EW of regional financial risks based on macroeconomic indicators, analyzes the changing trend of macroeconomic indicators by constructing multiple regression models, predicts possible eco- nomic crises and financial risks, and provides EW information for the government and finan- cial institutions. Literature [24] constructs a technical analysis model through stock index, bond price and exchange rate, analyzes the trend of market indicators, predicts possible fluctu- ations and risks in the market, and provides reference for investors and financial institutions. The method in literature [25] has high accuracy. By constructing multi-dimensional evaluation model and supervision index system, the risk status of financial institutions is analyzed and forewarned. Compared with these studies, the advantages of this paper lie in the following points. First, risk characteristics can be captured more accurately. DL model and MS-VAR model can cap- ture the nonlinear relationship and state switching characteristics of data more accurately to describe the characteristics of regional financial risks more accurately and improve the predic- tion accuracy. Second, data processing is more flexible. The EW model of regional financial risk adopts deep learning model and MS-VAR model, which can deal with nonlinear, hetero- geneous and multivariable data flexibly to better meet the forecasting needs of different data types. Then, it is more interpretable. Compared with the traditional economic model, the deep learning model and MS-VAR model are more explanatory, can present the forecast results intuitively, and better provide decision support for decision makers. Finally, the EW model of regional financial risk can be flexibly adjusted and optimized according to the changes of data and forecast demand, and has stronger adaptability. Conclusion Experimental result In order to improve the laws and regulations of the financial system and optimize the regional financial risk EW model, this paper constructs the regional financial risk EW indicators from four aspects: macroeconomic operation EW indicators, regional economic risk EW indicators, regional financial institution risk EW indicators and financial legal supervision system. According to the DL algorithm idea, the financial risk EW indicators are analyzed, the indica- tor system is improved, and the MS-VAR model is constructed. Finally, the regional financial risk EW MS-VAR model based on DL is constructed. Taking Shanghai as the research object, the model is empirically researched and predicted from several aspects, such as the compre- hensive index of economic pressure, the comprehensive index of regional financial pressure, the MS-VAR EW test, and the comprehensive index of regional financial risk. The results show: (1) The overall fluctuation of the regional economic pressure index is small, fluctuating around 0 in most periods, and the periods with large changes in the pressure index are mainly from 2004 to 2006, 2007, 2009 and 2016. The macroeconomic pressure index fluctuates greatly in the time series, and the impact of the global financial crisis on regional economic develop- ment is less than the impact on macroeconomic development. (2) After the financial crisis, the local government increased the supervision of non-performing enterprise and household loans, which greatly reduced the non-performing loan ratio of the regional banking industry. From 2011 to 2018, the non-performing loan ratio began to decline, and the overall fluctuation of the regional financial comprehensive stress index was small, fluctuating around 0. (3) In terms of the macroeconomic dimension, the probability of maintaining risk among low, PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 16 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system medium and high risks is all 0.91, the stability of the transition between risk zones is relatively strong, and the probability of maintaining the original risk level is relatively high. (4) From the perspective of the regional economic dimension, it is easier to convert from low risk to moder- ate risk, but it is more difficult to convert from high risk to moderate risk. From the perspec- tive of regional financial institutions, the probabilities of maintaining low risk and moderate risk are 0.98 and 0.97, respectively, which is stronger than maintaining the stability of high risk. (5) From the perspective of the state transition of the regional financial risk composite index, the maintenance probabilities of high risk and low risk levels are both 0.93 and 0.97, which are higher than the maintenance probability of medium risk. From the regional finan- cial pressure index prediction test, the overall observation value and the fitting value curve are in good agreement, indicating that the DL regional financial risk EW MS-VAR model has strong risk prediction ability. The model can better analyze the regional financial risk EW index conversion probability. Future research direction Due to the limited energy, this paper still has some limitations in the study of regional financial risk EW and prevention and control based on deep learning model. The future study of regional financial risk EW can be started from the following aspects: First, it is necessary to fur- ther strengthen cross-domain cooperation and data integration capabilities. Regional financial risk involves the financial systems of many fields and countries, and it needs interdisciplinary and cross-disciplinary cooperation, integrating various relevant data and establishing a com- prehensive and systematic risk EW model. Second, it is necessary to introduce advanced artifi- cial intelligence technology and big data analysis technology. At present, artificial intelligence technologies such as machine learning and deep learning, as well as big data analysis technolo- gies, have been gradually applied to the financial field. In the future, regional financial risk EW can also be more accurate and efficient through these technical means. In addition, it is neces- sary to deepen the research on the theory and method of risk EW, develop more flexible and operable EW indicators and models, and establish a sound risk EW system. Finally, it is neces- sary to strengthen international cooperation and information sharing and form a global risk monitoring system. Regional financial risks cross national boundaries, so international coop- eration and information sharing are needed to meet the risk challenges, and a global risk mon- itoring system should be established to provide more effective means for preventing and resolving regional financial risks. In a word, the research direction and means of regional financial risk EW in the future should be diversified and comprehensive, which requires cross- disciplinary and interdisciplinary cooperation, giving full play to the power of scientific and technological innovation, establishing a global risk monitoring system, and providing more reliable guarantee for financial stability and economic development. Supporting information S1 Data. The data sets used in Figs 6–12 are all from https://datasetsearch.research.google. com/. (ZIP) Author Contributions Conceptualization: Yanyu Zhuang. Data curation: Yanyu Zhuang. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 17 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system Formal analysis: Yanyu Zhuang. Investigation: Yanyu Zhuang. Methodology: Yanyu Zhuang. Project administration: Yanyu Zhuang. Resources: Yanyu Zhuang. Software: Hua Wei. Supervision: Hua Wei. Validation: Hua Wei. Visualization: Hua Wei. Writing – original draft: Yanyu Zhuang, Hua Wei. Writing – review & editing: Yanyu Zhuang, Hua Wei. References 1. Sun Y, Yang Y, Huang N, Zhou X. The impacts of climate change risks on the financial performance of mining industry: Evidence from listed companies in China. Resources Policy, 2020; 69: 101828. 2. Chen L, Han Q, Qiao Z, Stanley, H. E. Correlation analysis and systemic risk measurement of regional, financial and global stock indices. Physica A: Statistical Mechanics and its Applications, 2020; 542: 122653. 3. Du G, Liu Z, Lu H. Application of innovative risk early warning mode under big data technology in Inter- net credit financial risk assessment. Journal of Computational and Applied Mathematics, 2021; 386: 113260. 4. Suit B Y, Hassan L, Krauss S E, Ramanoon S Z, Ooi P T, Yasmin A R, et al. Exploring the mental model of cattle farmers in disease prevention and control practices. Veterinary sciences, 2020; 7(1): 27. https://doi.org/10.3390/vetsci7010027 PMID: 32121169 5. Zhang H, Luo Y. Enterprise Financial Risk Early Warning Using BP Neural Network Under Internet of Things and Rough Set Theory. Journal of Interconnection Networks, 2022; 2145019. 6. Bartlett P L, Montanari A, Rakhlin A. Deep learning: a statistical viewpoint. Acta numerica, 2021; 30: 87–201. 7. Ranganathan G. A study to find facts behind preprocessing on deep learning algorithms. Journal of Innovative Image Processing (JIIP), 2021; 3(01): 66–74. 8. Chen Y, Hu S, Mao H, Deng W, Gao X. Application of the best evacuation model of deep learning in the design of public structures. Image and Vision Computing, 2020; 102: 103975. 9. Chen D, Wawrzynski P, Lv Z. Cyber security in smart cities: a review of deep learning-based applica- tions and case studies. Sustainable Cities and Society, 2021; 66: 102655. 10. Zhou W, Chen M, Yang Z, Song X. (. Real estate risk measurement and early warning based on PSO- SVM. Socio-Economic Planning Sciences, 2021; 77: 101001. 11. Qi M, Gu Y, Wang Q. Internet financial risk management and control based on improved rough set algorithm. Journal of Computational and Applied Mathematics, 2021; 384: 113179. 12. Ouyang Z, Lai Y. Systemic financial risk early warning of financial market in China using Attention- LSTM model. The North American Journal of Economics and Finance, 2021; 56: 101383. 13. Peng H, Zhao S. The Influence of Fintech on Regional Financial Risk and Its Spatial Spillover Effect. International Journal of Innovation and Applied Studies, 2022; 36(2): 313–319. 14. Zhu K N, Liu D S, Wu J C, Sun L. The research of the regional financial risk early-warning model inte- grating the regression of lagging factors. AASRI Procedia, 2012; 1: 428–434. 15. Yu H. Empirical study on the correlation between real estate environmental engineering price fluctuation and regional financial stability. Nanotechnology for Environmental Engineering, 2022; 7(1): 193–200. 16. Chen Y. The Prevention and Diffusion of Systematic or Regional Financial Risks—The Methodology for China to Resolve Financial Crise. National Finance. Palgrave Macmillan, Singapore, 2021; 217–264. 17. Fan H, Liu R. Research on Financial Systemic Risk in ASEAN Region. Entropy, 2021; 23(9): 1131. https://doi.org/10.3390/e23091131 PMID: 34573758 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 18 / 19 PLOS ONEEarly warning model and prevention of regional financial risk integrated into legal system 18. Barach M A, Golden J M, Horton J J. Steering in online markets: the role of platform incentives and cred- ibility. Management Science, 2020; 66(9): 4047–4070. 19. Gkillas K, Tsagkanos A, Vortelinos D I. Integration and risk contagion in financial crises: Evidence from international stock markets. Journal of Business Research, 2019; 104: 350–365. 20. Zappone A, Di Renzo M, Debbah M. Wireless networks design in the era of deep learning: Model- based, AI-based, or both. IEEE Transactions on Communications, 2019; 67(10): 7331–7376. 21. Khan F A, Gumaei A, Derhab A, Hussain A. A novel two-stage deep learning model for efficient network intrusion detection. IEEE Access, 2019; 7: 30373–30385. 22. Krishnaraj N, Elhoseny M, Thenmozhi M, Selim M, Shankar K. Deep learning model for real-time image compression in Internet of Underwater Things (IoUT). Journal of Real-Time Image Processing, 2020; 17(6): 2097–2111. 23. Kou G, Chao X, Peng Y, Alsaadi F E, Herrera-Viedma E. Machine learning methods for systemic risk analysis in financial sectors. Technological and Economic Development of Economy, 2019; 25(5): 716–742. 24. Tang Y. Financial risk and early warning based on Qingdao marine economic forecast. Journal of Coastal Research, 2020; 112(SI): 195–198. 25. Duprey T, Klaus B. Early warning or too late? A (pseudo-) real-time identification of leading indicators of financial stress. Journal of Banking & Finance, 2022; 138: 106196. 26. Zeng H. Influences of mobile edge computing-based service preloading on the early-warning of financial risks. The Journal of Supercomputing, 2022; 78(9): 11621–11639. 27. Kuek T H, Puah C H, Arip M A, Habibullah M S. Macroeconomic perspective on constructing financial vulnerability indicator in China. Journal of Business Economics and Management, 2021; 22(1): 181– 196. 28. Popa D C S, Popa D N, Bogdan V, Simut, R. Composite financial performance index prediction–a neu- ral networks approach. Journal of Business Economics and Management, 2021; 22(2): 277–296. 29. Ping W, Wang F, Wang A, Huang Y. Risk Early Warning Research on China’s Futures Company. Emerging Markets Finance and Trade, 2021; 57(8): 2259–2270. 30. Yunita P. Dual Banking System Stability Index in the Shadow of COVID-19 Pandemic. International Journal of Islamic Economics and Finance (IJIEF), 2022; 5(1): 151–176. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286685 June 2, 2023 19 / 19 PLOS ONE
10.1371_journal.pone.0286527	RESEARCH ARTICLE Comparison of mortality and hospitalizations of older adults living in residential care facilities versus nursing homes or the community. A systematic review Denis Boucaud-MaitreID Teguo2,4, Jean-Franc¸ ois Dartigues3, He´ lène Amieva3, Maturin Tabue´ -Teguo2,4 1,2, Luc Letenneur3, Moustapha Drame´ ID 2,4, Nadine Taube´ - 1 Centre Hospitalier Le Vinatier, Bron, France, 2 Equipe EPICLIV, Universite´ des Antilles, Fort-de-France, Martinique, 3 Inserm, U1219 Bordeaux Population Health Center, University of Bordeaux, Bordeaux, France, 4 Centre Hospitalo-Universitaire de Martinique, Fort-de-France, Martinique a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 denis.boucaud@gmail.com Abstract OPEN ACCESS Citation: Boucaud-Maitre D, Letenneur L, Drame´ M, Taube´-Teguo N, Dartigues J-F, Amieva H, et al. (2023) Comparison of mortality and hospitalizations of older adults living in residential care facilities versus nursing homes or the community. A systematic review. PLoS ONE 18(5): e0286527. https://doi.org/10.1371/journal. pone.0286527 Editor: Charlotte Beaudart, University of Liege: Universite de Liege, BELGIUM Received: February 24, 2023 Accepted: May 17, 2023 Published: May 31, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0286527 Copyright: © 2023 Boucaud-Maitre et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Residential care facility may provide a transition between living at home and a nursing home for dependent older people or an alternative to nursing homes. The objective of this review was to compare mortality and hospitalizations of older adults living in residential care facili- ties with those living in nursing homes or in the community. We searched Medline, Scopus and Web of Science from inception to December 2022. Fifteen cohort studies with 6 months to 10 years of follow-up were included. The unadjusted relative risk (RR) of mortality was superior in nursing homes than in residential care facilities in 6 of 7 studies (from 1.3 to 1.68). Conversely, the unadjusted relative risk of hospitalizations was higher in residential care facilities in 6 studies (from 1.3 to 3.37). Studies conducted on persons with dementia found mixed results, the only study adjusted for co-morbidities observing no difference on these two endpoints. Compared with home, unadjusted relative risks were higher in residen- tial care facilities for mortality in 4 studies (from 1.34 à 10.1) and hospitalizations in 3 studies (from 1.12 to 1.62). Conversely, the only study that followed older adults initially living at home over a 10-year period found a reduced risk of heavy hospital use (RR = 0.68) for those who temporarily resided in a residential care facilities. There is insufficient evidence to deter- mine whether residential care facilities might be an alternative to nursing homes for older people with similar clinical characteristics (co-morbidities and dementia). Nevertheless, given the high rate of hospitalizations observed in residential care facilities, the medical needs of residents should be better explored. Introduction According to the World Health Organization (WHO), the number of people aged 60 and over will have overtaken the number of children under 5 by 2020 [1]. Population projections esti- mate that the proportion of people aged 60+ will almost double between 2015 and 2050, from 12% to 22% (or 2.1 billion people) [1]. The development of strategies for the care and housing PLOS ONE \| https://doi.org/10.1371/journal.pone.0286527 May 31, 2023 1 / 14 PLOS ONEData Availability Statement: All relevant data are within the paper and its Supporting information files. Funding: The author(s) received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. Mortality and hospitalizations in residential care facilities versus nursing homes or the community of the older people, depending on their individual needs, is a priority. These individual needs depend on the physical and cognitive functions, their psychological state, their comorbidities and their social environment [2]. Older adults want to age at home and avoid institutionaliza- tion. The proportion of community-dwelling older people with functional limitations has increased in recent years [3]. Keeping these individuals at home often requires the implemen- tation of medical (home care services, mobile geriatric teams, hospitalization at home) and other social aids (caregivers, meal delivery) which have an individual and collective cost. For older people who have severe medical and disability problems, the most widespread social model in developed countries remain nursing home care. However, other housing models exist such as residential care facilities or foster families [4]. Residential care facilities (also called “senior housing", "independent living communities”, "assisted living facilities", or "continual care retirement communities") have developed over the past decades. In general, each resident has a private apartment and access to common areas and services. Residential care facilities differ in size, type (residence or village), services offered and costs. These structures aim to promote the autonomy and social life of older people [5]. The socio-demographic and medical characteristics of residents in nursing homes and resi- dential care facilities are not similar. Nevertheless, in the US, the age of residents is comparable and residential care facilities are increasingly admitting residents with functional limitations and/or Alzheimer’s and other dementia. Indeed, the prevalence of Alzheimer’s disease or other dementia in residential care facilities has increased from 5% in 2002 to 42% in 2010 [3]. Thus, determining the most appropriate care for dependent older adults is a public health priority. The place of residential care facilities, as an intermediate stage or final place of resi- dence is therefore a key issue for the older adults, their families and health policies. Depending on the characteristics of the patient, existing models need to be properly assessed and com- pared to determine the effectiveness and cost-utility of each model. This is particularly true for older adults with dementia in the current debate over whether residential care facilities can substitute for, or delay the transfer to nursing homes. The effectiveness of residential care facil- ities, particularly in terms of mortality and hospitalizations, has been poorly studied in the lit- erature. In 2012, a systematic review on patients with dementia found only one study suggesting that mortality and hospitalizations did not differ in residential care facilities and nursing homes [6]. The objective of this review was therefore to compare mortality and hospi- talization rates reported among residents living in residential care facilities, nursing homes and/or in the community and to look for studies conducted on similar patients profiles with regard to dementia status and dependency. Methods Our review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) guidelines [7]. The protocol for this review was regis- tered in the International Prospective Register of Systematic Reviews (PROSPERO; CRD42022327207). Data source We systematically searched Medline, Scopus, and Web of Science from inception up to 31 December 2022. We developed and conducted the literature search, using a combination of the MeSH terms “senior housing” or “independent living communities” or “residential care” or “continual care retirement” in the title or abstract and “mortality” or “death” or “hospitali- zations” in the full text of the articles. Two reviewers (DBM, LL) screened the titles and abstracts. S1 Appendix provides the comprehensive search strategy used to identify original PLOS ONE \| https://doi.org/10.1371/journal.pone.0286527 May 31, 2023 2 / 14 PLOS ONEMortality and hospitalizations in residential care facilities versus nursing homes or the community research articles for inclusion in our systematic review. The same two reviewers independently assessed the full text of the articles for eligibility. Discrepancies were resolved by discussion. We also checked the reference lists of all reviews on this topic to identify articles that might have been missed. We limited the search to articles written in English and French. Eligibility criteria Population. We only selected cohort studies conducted among older persons (�65 years old). Intervention. We selected studies that had a cohort design (prospective or retrospective cohort design) and at least six months of follow-up and that measured mortality or hospitaliza- tions as outcomes. For duplicate publications from the same cohort, we selected those with the largest number of participants. We excluded cross-sectional studies and, reviews. Comparison. We selected cohort studies comparing residential care facilities with nursing homes and/or communities. Outcomes. To be included, we considered studies that reported the number of partici- pants or person years and the number of deaths or hospitalizations in both groups (residential care facilities versus nursing home and/or communities). Data extraction and management After the study selection process, one reviewer (DBM) extracted data from the original cohort studies. The characteristics extracted from each cohort were: name of the first author, year of publication, study design, length of follow-up, number of participants, mean age, percentage of dementia and percentage of participants with disability (functional status assessed by the Instrumental Activities of Daily Living (IADL) scale [8] or the Activities of Daily Living (ADL) scale [9]), number and percentage of deaths, number and percentage of hospitalizations and risk estimates. Quality assessment The risk of bias was assessed using the Quality Assessment Tool for Observational Cohort and Cross-sectional studies [10], a recommended tool for analytic studies [11] by one reviewer (DBM). This process ensures that the quality of included studies is good enough to provide reliable results. Based on a series of questions, the goal was to identify potential flaws in the publication that could affect the measurement of the outcome. The quality of the studies was rated as “poor”, “fair”, or “good”. Question 6 “For the analyses in this paper, were the exposure (s) of interest measured prior to the outcome(s) being measured?”, question 7 “Was the time- frame sufficient so that one could reasonably expect to see an association between exposure and outcome if it existed?” and question 14 “Were key potential confounding variables mea- sured and adjusted statistically for their impact on the relationship between exposure(s) and outcome(s)?” were considered critical because this review focuses on the relationship between mortality/hospitalization and types of accommodations. Potentially confounding variables affecting the results were age, gender, comorbidities, levels of dependency and cognitive func- tion. In cases where studies answered “no” to questions 7 and 14, quality was rated as “poor”. Statistical analysis Unavailable relative risks (RR) and confidence intervals were calculated from the number of events and the number of residents for each study. After reviewing each of the studies included in this review, we considered that a meta-analysis was not relevant, due to the low PLOS ONE \| https://doi.org/10.1371/journal.pone.0286527 May 31, 2023 3 / 14 PLOS ONEMortality and hospitalizations in residential care facilities versus nursing homes or the community methodological quality of the studies, the diversity of study designs and follow-up and the het- erogeneity of populations, for which essential baseline characteristics were not available. Results Our search yielded 8,716 records, of which 6,975 remained after eliminating duplicates (Fig 1). When screening tittles and abstracts, 6,918 records were excluded, leaving 24 full texts to be assessed. Nine studies were excluded, among them two focused on specific mortality endpoints (respiratory mortality and suicide [12, 13]), one related to emergency department visits rather than hospital admission [14], one pooled data from residential care facilities and nursing home data [15], three were systematic review [6, 16, 17] and two were sub-studies of other studies. A total of 15 studies met the eligibility criteria and were included in this review [18–32]. Fig 1. Flow diagram of included studies. https://doi.org/10.1371/journal.pone.0286527.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286527 May 31, 2023 4 / 14 PLOS ONEMortality and hospitalizations in residential care facilities versus nursing homes or the community Characteristics of the included studies The studies were published between 2000 and 2020, 5 studies were prospective cohort studies [18, 19, 21, 23, 24] and 10 were retrospective cohort studies [20, 22, 25–32]. The number of participants ranged from 158 to 691,388 for the studies comparing residential care facilities with nursing homes, and from 808 to 3,366,303 for the studies comparing residential care facil- ities to home-based care. The studies follow-up ranged from 6 months to 10 years. In three studies [22, 25–26], age was not available, and in the remaining 12 studies, the mean age of par- ticipants was greater than 80 years or a majority of participants were 75 and older. For eight studies, MMSE score or the percentage of dementia among participants were not available for one or the both groups of participants. Three studies comparing residential care facilities to nursing homes focused only on dementia patients [19–21]. For the other 5 studies, the oldest adults with dementia were reported in nursing homes rather than in residential care facilities in two studies (72.7% versus 30.3% [20], MMSE score = 17.88 versus 16.03 [24]), there was no difference for one study (MMSE score = 23.3 versus 23.1 [18] and one study reported oldest adults with dementia in residential care facilities rather than at home (33.9% versus 8.4%) [31]. Functional ability was available for three studies, all comparing residential care facilities to nursing homes, with a lower disability score in nursing homes than in residential care facilities [18, 19, 24]. Six studies assessed cohorts from US, 3 from Canada, 3 from United Kingdom, 1 from Ireland, Taiwan and Australia. Risk of bias Of the 15 studies, two were rated “good” [18, 31], six “fair” [18, 20, 22, 28, 29, 32] and seven “poor” [21, 24–27, 30] (S1 Appendix). Quality was considered good when similar clinical base- line characteristics were observed or when adjustments on age, sex, dementia and dependency (question 14) were implemented. We considered that failure to adjust for any of this variables downgraded the quality of the study to “fair”, no adjustment or insufficient time (inferior to one year) were rated as “poor” studies (S1 Annex). 1. Comparison of mortality between residential care facilities and nursing homes. Nine studies compared mortality between residential care facilities and nursing homes [18– 26] (Table 1). Six of the seven studies with at least one year of follow-up described a higher unadjusted relative risk in nursing homes than in residential care facilities, ranging from 1.3 to 1.68. The only other study with more than 1 year of follow-up suggesting a lower risk [26] had in-hospital mortality, not total mortality as its endpoint. Three studies [19–21] focused specifically on patients with dementia and a fourth adjusted for dementia [18]. The study by Thomas et al. [20], comparing 88,867 residents in residential care facilities with 602,521 residents of nursing homes (1-year follow-up) reported a risk of 1.58 [1.56–1.6]. The study by Sloane et al. [19] found no significant difference in mortality between residential care facilities and nursing homes for mild dementia or moderate/severe dementia after adjustment for age, gender, race, education, marital status, length of stay, cogni- tion and comorbidities. Finally, the study by Shah et al. [22] found a gender and age-adjusted ratio of 419 (396–442) for nursing homes alone versus 284 (266–302) for residential care facili- ties. Further standardization for dementia diagnosis reduced the ratio to 309 (292–326) and to 218 (205–232), respectively. 2. Comparison of hospitalizations between residential care facilities and nursing homes. Seven studies [19–21, 26–29] compared hospitalization rates between residential care facilities and nursing homes (Table 2). All studies were North American (3 in Canada, 3 in the US) with an exception of one study from the UK. All described an unadjusted increased rela- tive risk of hospitalizations in residential care facilities compared with nursing homes, ranging PLOS ONE \| https://doi.org/10.1371/journal.pone.0286527 May 31, 2023 5 / 14 PLOS ONEMortality and hospitalizations in residential care facilities versus nursing homes or the community Table 1. Studies comparing residential care facilities versus nursing homes on mortality. Study design and follow-up Participants/ cases Age (mean or %), years Percentage of dementia Percentage of mortality Unadjusted relative risk Adjusted risk ratio or hazard ratio Author, year, Country Pruchno, 2000, US [18] Prospective, 15 months • Nursing Homes: 76 • Residential care facilities:82 Sloane, 2005, US [19] Prospective, 1 year • Nursing Homes: 479 • Residential care facilities: 773 • Nursing Homes: 87.4 • Residential care facilities: 86.2 • Nursing Homes:: 84.9 • Residential care facilities: 84.4 • Nursing Homes: MMSE score = 23,3 • Residential care facilities: MMSE score = 23.1 • Nursing Homes: 18.4% (14/76) • Residential care facilities: 12.2% (10/82) • Nursing Homes: Moderate to severe: 49.3% • Residential care facilities: Moderate to severe: 29.4% • Nursing Homes: 22.5% (108/479) • Residential care facilities: 14.7% (114/773) 1.51 [0.71–3.19] No 1.53 [1.21–1.94] Incidence rate per 100 participants (mild dementia): nursing homes: 4.2 versus residential care facilities: 3.2 (not- significant) Moderate to severe dementia: 4.2 versus 3.7 (non-significant) Incidence rate adjusted for baseline age, gender, race, education, marital status, length of stay, cognition and number of comorbidities 1.58 [1.56–1.6] No - No 1.38 [1.28–1.48] Age and sex-adjusted hazard ratios: 1.48. The ratio for nursing homes alone was 419 (396–442) and that for residential homes was 284 (266–302). Further standardization for dementia diagnosis reduced the ratio to 309 (292–326) for nursing homes and to 218 (205–232) for residential homes. Thomas, 2020, US [20] Retrospective, 1 year Resnick, 2015, US [21] Prospective, 6 months Shah, 2013, England and Wales [22] Retrospective, 1 year • Nursing Homes: >75 years: 80.6% • Residential care facilities: >75 years: 88.2% • Nursing Homes: 83.7 • Residential care facilities: 85.7 Not specified • Nursing Homes: 602,521 • Residential care facilities: 88,867 • Nursing Homes: 103 • Residential care facilities: 93 • Nursing Homes: 4109 • Residential care facilities: 4320 • Nursing Homes: 100% • Residential care facilities: 100% • Nursing Homes: 100% (MMSE score: 8.7) • Residential care facilities: 100% (MMSE score: 5.8) Nursing Homes and residential care facilities combined: 38,9% • Nursing Homes: 31.1% (187350/602521) • Residential care facilities: 19.7% (17491/ 88867) • Nursing Homes: 0% (0/ 103) • Residential care facilities: 0% (0/93) Nursing Homes: 30.8% (1265/ 4109) Residential care facilities: 22.3% (963/4320) McCann, 2009, Ireland [23] Prospective, 5 years Liu, 2010, Taiwan [24] Prospective, 9 months • Nursing Homes: 895 • Residential care facilities: 577 • Nursing Homes: 140 • Residential care facilities: 185 >75 years: • Nursing Homes: 88% • Residential care facilities: 89% >75 years: • Nursing Homes: 65% • Residential care facilities: 73% Not specified • Nursing Homes: MMSE score = 17.88±8.91 • Residential care facilities: MMSE score = 16.03±6.90 • Nursing Homes: 70% (626/895) • Residential care facilities: 54% (311/577) • Nursing Homes: 0% (0/ 140) • Residential care facilities: 3.2% (6/185) 1.3 [1.19–1.42] No - No (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0286527 May 31, 2023 6 / 14 PLOS ONEMortality and hospitalizations in residential care facilities versus nursing homes or the community Table 1. (Continued) Study design and follow-up Participants/ cases Age (mean or %), years Percentage of dementia Percentage of mortality Unadjusted relative risk Adjusted risk ratio or hazard ratio Author, year, Country Rothera, 2002, UK [25] Retrospective, 20 months Godden, 2001, UK [26] Retrospective, 1 year Not specified Not specified Not specified Not specified • Nursing Homes: 499 • Residential care facilities: 866 • Nursing Homes: 1700 • Residential care facilities:1504 • Nursing Homes: 39.1% (195/499) • Residential care facilities: 23.3% (202/866) • Nursing Homes: 5.7% (97/1700) • Residential care facilities: 8.8% (133/1504) 1.68 [1.42–1.97] No No. 0.65 [0.5–0.83] (Hospital death and not total mortality) https://doi.org/10.1371/journal.pone.0286527.t001 from 1.3 to 3.37, except for one small study. Five studies had 1 year follow-up, with hospitaliza- tion rates of 30% to 40% in residential care facilities versus 10% to 30% in nursing homes. Four studies included only residents with dementia. The study with the highest relative risk was conducted in Canada [28]. It found hospitalization rates of 36.1% in residential care facili- ties versus 10.7% in nursing homes. Less severe cognitive impairment (Hazard Ratio: 0.35 [0.18–0.67]) was associated with a lower hospitalization rate in this study. The study by Thomas et al. [20] reported a relative risk of 1.30 [1.29–1.31]. Conversely, in the study by Sloane et al. [19] adjusted for age, gender, ethnicity, education, marital status, length of stay, cognition and number of comorbidities, the risk of hospitalization was higher for patients with mild dementia (14.2% versus 8.4%, p = 0.009) in residential care facilities but not for those with moderate or severe dementia (14.2% versus 10.0%, p = 0.115). 3. Comparison of mortality between residential care facilities and the community. Four studies have been conducted to compare mortality of older patients living in residential care facilities and the community (Australia [30], Ireland [23], two in the US [20, 31]) (Table 3). All suggest a higher mortality rate in residential care facilities, with unadjusted RRs ranging from 1.34 to 10.1. In these four studies, the characteristics of the elderly differed by age and/or dementia between the two groups. The Australian study [30] comparing 3,330,987 older people at home versus 35,316 residents in residential care facilities observed 1-year mortality rates of 3.6% versus 34.6% respectively, giving an age and gender adjusted odd-ratio (OR) of 10.1 (95% CI: 9.8–10.5). The Irish study [23] found an OR of 1.63, adjusted for age, sex, general health and marital status with a 5-year follow-up. Bartley’s study [30] retrieved an OR of 2.4, adjusted for Charlton Comorbidity Index and marital status. Finally, Thomas et al.’s study [20] of patients with dementia reported an unadjusted relative risk of 1.34 [1.33–1.36]. 4. Comparison of hospitalizations of older adults living in residential care facilities and the community. Three 1-year longitudinal studies from the US or the UK suggest a greater risk of hospitalization of older adults living in residential care facilities than in the community ranging from 1.12 to 1.65 [20, 26, 31] (Table 4). One-year hospitalization rates ranged from 31 to 48%. Only one study, by Park et al. [32], followed elderly people initially living at home and compared the 10-year hospitalization rate between those who went to residential care facilities and those who did not. In this study, the risk of hospitalization among those entering residen- tial care facilities was decreased for heavy hospital use (RR 0.68 (p<0.001)) but not for moder- ate hospital use. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286527 May 31, 2023 7 / 14 PLOS ONEMortality and hospitalizations in residential care facilities versus nursing homes or the community Table 2. Studies comparing hospitalizations between residential care facilities and nursing homes. Participants/ cases Age (mean and %), years Percentage of dementia Percentage of hospitalizations Unadjusted relative risk Adjusted risk ratio or hazard ratio Author, year, Country McGregor, 2014, Canada [27] Study design and follow-up duration Retrospective, 3 years Maxwell, 2015, Canada [28] Retrospective, 1 year Hogan, 2014, Canada [29] Retrospective, 1 year • Nursing homes:12209 • Residential care facilities:842 • Nursing homes: 83.1 • Residential care facilities: 81.5 Not specified • Nursing homes: 691 • Residential care facilities: 609 • Nursing homes: 86.4 • Residential care facilities: 85.7 • Nursing homes: 100% • Residential care facilities:: 100% • Nursing homes: 976 • Residential care facilities: 1066 • Nursing homes: Not specified • Residential care facilities: 84.9 • Nursing homes: not specified • Residential care facilities:: 57.1% Sloane, 2005, US [19] Prospective, 1 year • Nursing homes: 479 • Residential care facilities: 773 • Nursing homes: 84.9 • Residential care facilities: 84.4 • Nursing homes: Moderate to severe: 49.3% • Residential care facilities:: Moderate to severe: 29.4% Thomas, 2020, US [20] Retrospective, 1 year Godden, 2001, UK [26] Retrospective, 1 year • Nursing homes: 602521 • Residential care facilities: 88867 • Nursing homes: 1132 • Residential care facilities:1504 • Nursing homes: >75 years: 80.6% • Residential care facilities: >75 years: 88.2% • Nursing homes: 100% • Residential care facilities: 100% Not specified Not specified Resnick, 2015, US [21] Prospective, 6 months • Nursing homes: 103 • Residential care facilities: 93 • Nursing homes: 83.7 • Residential care facilities: 85.7 • Nursing homes: 100% (MMSE score: 8.7) • Residential care facilities: 100% (MMSE score: 5.8) https://doi.org/10.1371/journal.pone.0286527.t002 1.65 [1.57– 1.73] 3.37 [2.65– 4.29] 2.76 [2.32– 3.28] 1.55 [1.11– 2.16] No No No Mild dementia: P = 0.009 adjusted for baseline age, gender, race, education, marital status, length of stay, cognition and number of comorbidities. Moderate to severe dementia: P = 0.115 adjusted for baseline age, gender, race, education, marital status, length of stay, cognition and number of comorbid conditions. 1.30 [1.29– 1.31] No 1.4 [1.22–1.59] No - No • Nursing homes: 42.0% (5125/12209) • Residential care facilities: 69.1% (582/ 842) • Nursing homes: 10.7% (74/691) • Residential care facilities: 36.1% (220/ 609) • Nursing homes: 14.0% (137 /976) • Residential care facilities: 38.7% (413/ 1066) Mild dementia: • Nursing homes: 8.4% (20/243) • Residential care facilities: 14.2% (78/ 546) Moderate to Severe dementia: • Nursing homes: 10.0% (24/236) • Residential care facilities: 14.2% (32/ 227) • Nursing homes: 29% (174323/602521) • Residential care facilities: 37.6% (33457/88867) • Nursing homes: 22,3% (253/1132) • Residential care facilities: 31,2% (469/ 1504) • Nursing homes: 0% (0/103) • Residential care facilities: 0% (0/93) Discussion In this literature review, we analyzed 15 studies comparing older adults living in residential care facilities with older adults living in nursing homes or the community. In general, a twofold increase in mortality was observed in nursing homes compared to res- idential care facilities. This result was expected since nursing homes generally accommodate patients at the end of life, with significant co-morbidities and a severe degree of dependency. We observed that the age of patient was generally similar in studies comparing nursing homes PLOS ONE \| https://doi.org/10.1371/journal.pone.0286527 May 31, 2023 8 / 14 PLOS ONEMortality and hospitalizations in residential care facilities versus nursing homes or the community Table 3. Studies comparing mortality of older adults living in residential care facilities and the community. Study design Participants/ cases Age (mean and %), years Percentage of dementia Percentage of mortality Unadjusted relative risk Adjusted risk ratio or hazard ratio Author, year, Country Inacio, 2020, Australia [30] Retrospective, 1 year Bartley, 2018, USA [31] Retrospective, with age- and sex- matched, 1 year Prospective, 5 years Retrospective, 1 year McCann, 2009, Ireland [23] Thomas, 2020, US [20] • Community: 3330987 • Residential care facilities: 35316 • Community: 404 • Residential care facilities: 404 >75 years: • Community: 63.0% • Residential care facilities: 88.2% (mean: 85 years) • Community: 86.8 • Residential care facilities: 86.8 • Community: 205566 • Residential care facilities: 577 >75 years: • Community: 42% • Residential care facilities: 89% Not specified Community: 8.4% Residential care facilities: 33.9% Not specified • Community: 2074420 • Residential care facilities: 88867 >75 years: Community: 80.6% Residential care facilities: >75 years: 88.2% • Community: 6.1% • Residential care facilities: 30.3% • Community: 3.6% (119815/3330987) • Residential care facilities: 34.6% (12225/35316) • Community: 9.4% (38/404) • Residential care facilities: 20.3% (82/ 404) • Community: 22% (45224/205566) • Residential care facilities: 54% (311/ 577) • Community: 14.7% (303891/ 2074420) • Residential care facilities: 19.7% (17491/88867) 9.62 [9.48– 9.77] OR: 10.1 (95% CI: 9.8– 10.5) adjusted by sex and age 2.16 [1.51– 3.09] OR: 2.4 Adjusted for Charlson Comorbidity Index and marital status 2.45 [2.27– 2.64] HR: 1.63 (1.44–1.85) Adjusted for age, sex, general health and marital status 1.34 [1.33– 1.36] No https://doi.org/10.1371/journal.pone.0286527.t003 and residential care facilities, suggesting that co-morbidities may have a greater impact on mortality than biological age [33]. Unfortunately, in several studies, comorbidities, in particu- lar the level of dependency or severity of dementia, were poorly documented and/or not con- sidered, except in the study by Sloane et al. The studies included in the review also reported an Table 4. Studies comparing hospitalizations of older adults living in residential care facilities and the community. Author, year, Country Study design Participants/ cases Age (mean and %), years Percentage of dementia Percentage of hospitalizations Unadjusted risk ratio Adjusted risk ratio or hazard ratio Park, 2018, US [32] Retrospective, 10 years • Community: 975 • Residential care facilities: 214 • Community: 82.4 • Residential care facilities: 83.3 Not specified Not specified Bartley, 2018, USA [31] Retrospective, with age- and gender- matched, 1 year Retrospective, 1 year • Community: 404 • Residential care facilities: 404 • Community: 86.8 • Residential care facilities: 86.8 • Communities: 8.4% • Residential care facilities: 33.9% • Community: 2074420 • Residential care facilities: 88867 >75 years: 80.6% Residential care facilities: 88.2% • Communities: 6.1% • Residential care facilities: 30.3% Retrospective, 1 year • Community: 83606 • Residential care facilities:1504 Not specified Not specified Thomas, 2020, US [20] Godden, 2001, UK [26] • Community: 31.4% (127/404) • Residential care facilities: 48.3% (195/ 404) • Community: 33.6% (697968/2074420) • Residential care facilities: 37.6% (17491/88867) • Community: 18.9% (15239/80402) • Residential care facilities: 31.2% (469/ 1504) RR: 0.68 (p<0.001) for heavy hospital use RR: 0.89 (NS) for moderate hospital use. Adjusted with death, sociodemographics, health, social support, regions OR: 2.03 [CI: 1.5–2.7] Adjusted for Charlson Comorbidity Index and marital status 1.54 [1.29– 1.83] 1.12 [1.11– 1.13] No 1.65 [1.52– 1.78] No https://doi.org/10.1371/journal.pone.0286527.t004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286527 May 31, 2023 9 / 14 PLOS ONEMortality and hospitalizations in residential care facilities versus nursing homes or the community increased risk of hospitalizations in residential care facilities compared to nursing home, with hospitalization rates of about 30% per year in residential care facilities. From a public health perspective, the cost of these hospitalizations is important to consider when evaluating the effi- ciency of this model. The causes of hospitalizations could be different between these two set- tings, especially concerning falls [34, 35] or polymedication. Targeted geriatric interventions such as telemonitoring [36] could reduce avoidable hospitalizations in residential care facili- ties. Moreover, this high risk of hospitalization raises questions about the ability of residential care facilities to meet the medical needs of older adults. In the US, only 48.2% of community- based residential settings offer a range of services including nursing care, medication assis- tance, meals, laundry, cleaning, transportation, and recreation and 29.1% have access to all these services except for nursing care and medication assistance [37]. It seems important to better define the clinical profile of the older adults who may be candidates for residential care facilities. Indeed, residential care facilities are often considered an appropriate setting for cog- nitively impaired patients in the US and Canada. In these countries, a high rate of residents suffers from dementia (58% in the Canadian study by Maxwell et al. [28] or 68% of individuals in the American study by Watson et al. [38]). Yet dementia is reported to be the most common predisposing factor (>90%) that precipitates the move of older adults to an assisted living or nursing home [39]. Previous research indicates that the percentage of facilities that provide staff training related to psychiatric disorders in older adults is low and generally inadequate in the US [40]. Increased medical and nursing support may be an option to reduce hospitaliza- tion rates. The vast majority of studies have been conducted in the US and Canada. In other countries, the clinical characteristics of residents may be different in residential care facilities. From this point of view, residential care facilities could be a step when patients at home require more care or become frail, before the development of severe cognitive impairment [40]. In the UK, the proportion of resident with severe dementia in residential care facilities appears to be low (2.1% versus 24.1% in nursing homes [41]). In Sweden, only 20% of residents suffered from dementia [42]. In France, residents in residential care facilities tend to be frail (53.7%), without being systematically disabled (mean ADL score; SD = 5.4; 0.9) and the role of residential care facilities is rather to address social isolation, social vulnerability, and loneliness [43]. Further studies in Europe are needed to determine whether the hospitalization rate is comparable to that of the North American studies. Compared to living at home, mortality in residential care facilities was much higher, with RR ranging from 1.34 to 10.1. Again, the methodological quality of these studies was critical, as none of them at least adjusted for dementia status or level of dependency. Studies are needed to compare home care systems with residential care facilities for older adults suffering from social isolation or comorbidity. Indeed, several devices have been developed over time to pro- mote home care, such as remote monitoring, telemedicine, or home care services [44, 45]. In particular, telemedicine has shown encouraging results in the management of care, prevention or management of chronic pathologies (particularly cardiovascular or diabetes) or adherence to medication [46]. With regards to hospitalizations, a higher risk was found in residents living in residential care facilities compared to those living at home in 3 of the 4 studies with a fol- low-up of one year. However, the study by Park et al. [32] contrasts with these results, both because of the methodological quality of the study and because of the results obtained. This study suggests that residential care facilities may reduce the risk of major hospitalization, based on a cohort of patients initially at home, some of whom may or may not enter a residen- tial care facilities during the 10 years of follow-up of the study. The main limitation of this review is the methodological quality of the studies. No study compared these models of care up to institutional entry with minimal adjustment for age, PLOS ONE \| https://doi.org/10.1371/journal.pone.0286527 May 31, 2023 10 / 14 PLOS ONEMortality and hospitalizations in residential care facilities versus nursing homes or the community gender, dementia, and dependency, and only the study by Park et al. [32] analyzed the health trajectory over time of older adults initially living at home. The medical characteristics of resi- dents in terms of dementia and standardized cognitive assessment, activities of daily living, frailty and length of stay were poorly described. Moreover, the type of residential care facilities and services offered may vary from country to country or from institution to institution [47, 48]. Finally, socioeconomic characteristics (marital status, income, etc.), which may influence the choice of institution (public or private), have rarely taken into account in the studies. For all these reasons, we considered that meta-analyses were not relevant because the studies did not compare patients with similar characteristics profile. The level of evidence for the effective- ness of residential care facilities on mortality and hospitalizations compared to nursing home or communities is therefore low. Nevertheless, the particularly high rate of hospitalization in residential care facilities raises the question of the lack of medical and paramedical staff and the cost of these hospitalizations. Nevertheless, the potential benefits of residential care facili- ties versus nursing homes or home care are not limited to mortality and hospitalizations. The effects on physical function, quality of life, happiness, cognition and other aspects of health would need to be compared. A final limitation is that the search strategy was limited to Med- line, Scopus and Web of Science and to articles written in English or French. Nevertheless, It has been established that the exclusion of non-English language articles has only a minimal effect on the overall conclusions of the reviews [49]. A single reviewer conducted the data extraction and quality assessment of the studies, which may reduce the diversity of the studies included. Conclusion This systematic review raises important clinical and policy questions. The place of residential care facilities in the health care pathway of older adults, as an intermediate or alternative step to home or nursing home, has not been sufficiently studied. Although patient’s profiles are likely to differ and care systems are not identical across the world, the particularly high rate of hospitalizations in these settings requires further investigations to assess the effectiveness and efficiency of this model. If residential care facilities are considered as an alternative for older people with mild to moderate dementia, studies of good methodological quality have to be implemented. Preventive and palliative care, depending of levels and types of medical, func- tional, and psychosocial needs, may be useful to reduce avoidable hospitalizations. Supporting information S1 Checklist. PRISMA 2020 checklist. (DOCX) S1 Appendix. Search strategy. (DOCX) S1 Annex. Quality studies. (DOCX) Author Contributions Conceptualization: Denis Boucaud-Maitre, Jean-Franc¸ois Dartigues, He´lène Amieva, Matu- rin Tabue´-Teguo. Formal analysis: Denis Boucaud-Maitre. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286527 May 31, 2023 11 / 14 PLOS ONEMortality and hospitalizations in residential care facilities versus nursing homes or the community Investigation: Denis Boucaud-Maitre, Luc Letenneur. Methodology: Denis Boucaud-Maitre, Jean-Franc¸ois Dartigues, He´lène Amieva. Supervision: Jean-Franc¸ois Dartigues, Maturin Tabue´-Teguo. Validation: Luc Letenneur, Moustapha Drame´, Jean-Franc¸ois Dartigues, He´lène Amieva, Maturin Tabue´-Teguo. Writing – original draft: Denis Boucaud-Maitre. Writing – review & editing: Luc Letenneur, Moustapha Drame´, Nadine Taube´-Teguo, Jean- Franc¸ois Dartigues, He´lène Amieva, Maturin Tabue´-Teguo. References 1. World Health Organization (2020). Decade of Healthy Aging 2020–2030. https://cdn.who.int/media/ docs/default-source/decade-of-healthy-ageing/final-decade-proposal/decade-proposal-final-apr2020- en.pdf?sfvrsn=b4b75ebc_25&download=true 2. Gordon EH, Hubbard RE. Frailty: understanding the difference between age and ageing. Age Ageing. 2022 Aug 2; 51(8):afac185. https://doi.org/10.1093/ageing/afac185 PMID: 35973066 3. Toth M, Palmer L, Bercaw L, Johnson R, Jones J, Love R, et al. US Department of Health and Human Services. Understanding the characteristics of older adults in different residential settings: data sources and trends (October 2020). https://aspe.hhs.gov/system/files/pdf/264196/ResSetChar.pdf 4. Boucaud-Maitre D, Cesari M, Tabue-Teguo M. Foster families to support older people with dependency: a neglected strategy. Lancet Healthy Longev. 2023 Jan; 4(1):e10 https://doi.org/10.1016/S2666-7568 (22)00288-4 PMID: 36610444 5. 6. Jolanki OH. Senior Housing as a Living Environment That Supports Well-Being in Old Age. Front Public Health. 2021; 8:589371. https://doi.org/10.3389/fpubh.2020.589371 PMID: 33614564 Zimmerman S, Anderson W, Brode S, Jonas D, Lux L, Beeber A, et al. Comparison of Characteristics of Nursing Homes and Other Residential Long-Term Care Settings for People With Dementia [Internet]. Rockville (MD): Agency for Healthcare Research and Quality (US); 2012 Oct. Report No.: 12(13)- EHC127-EF. 7. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021 Mar 29; 372:n71. https:// doi.org/10.1136/bmj.n71 PMID: 33782057 8. Lawton MP, Brody EM. Assessment of older people: self-maintaining and instrumental activities of daily living. Gerontologist 1969; 9(3):179–186. PMID: 5349366 9. Katz S, Ford AB, Moskowitz RW, Jackson BA, Jaffe MW. Studies of illness in the aged. The index of ADL: a standardized measure of biological and psychosocial function. JAMA 1963; 185:914–919. https://doi.org/10.1001/jama.1963.03060120024016 PMID: 14044222 10. National Institution of Health: U.S Department of Health and Human Services. Quality assessment tool for observational cohort and cross-sectional studies. https://www.nhlbi.nih.gov/health-topics/study- quality-assessment-tools (2015). 11. Ma LL, et al. Methodological quality (risk of bias) assessment tools for primary and secondary medical studies: What are they and which is better? Military Med. Res. 2020; 7:7. https://doi.org/10.1186/ s40779-020-00238-8 PMID: 32111253 12. Menec VH, MacWilliam L, Aoki FY. Hospitalizations and deaths due to respiratory illnesses during influ- enza seasons: a comparison of community residents, senior housing residents, and nursing home resi- dents. J Gerontol A Biol Sci Med Sci. 2002 Oct; 57(10):M629–35 https://doi.org/10.1093/gerona/57.10. m629 PMID: 12242314 13. Mezuk B, Lohman M, Leslie M, Powell V. Suicide Risk in Nursing Homes and Assisted Living Facilities: 2003–2011. Am J Public Health. 2015 Jul; 105(7):1495–502. https://doi.org/10.2105/AJPH.2015. 302573 PMID: 25973805 14. Ingarfield SL, Finn JC, Jacobs IG, Gibson NP, Holman CD, Jelinek GA, et al. Use of emergency depart- ments by older people from residential care: a population based study. Age Ageing. 2009 May; 38 (3):314–8. https://doi.org/10.1093/ageing/afp022 PMID: 19286676 15. Cloutier DS, Penning MJ, Nuernberger K, Taylor D, MacDonald S. Long-Term Care Service Trajecto- ries and Their Predictors for Persons Living With Dementia: Results From a Canadian Study. J Aging Health. 2019 Jan; 31(1):139–164. https://doi.org/10.1177/0898264317725618 PMID: 28814151 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286527 May 31, 2023 12 / 14 PLOS ONEMortality and hospitalizations in residential care facilities versus nursing homes or the community 16. Boland L, Le´ gare´ F, Perez MM, et al. Impact of home care versus alternative locations of care on elder health outcomes: an overview of systematic reviews. BMC Geriatr. 2017; 17(1):20. https://doi.org/10. 1186/s12877-016-0395-y PMID: 28088166 17. Wysocki A, Butler M, Kane RL, Kane RA, Shippee T, Sainfort F. Long-Term Services and Supports for Older Adults: A Review of Home and Community-Based Services Versus Institutional Care. J Aging Soc Policy. 2015; 27(3):255–79. https://doi.org/10.1080/08959420.2015.1024545 PMID: 25942005. 18. Pruchno RA, Rose MS. The effect of long-term care environments on health outcomes. Gerontologist. 2000 Aug; 40(4):422–8. https://doi.org/10.1093/geront/40.4.422 PMID: 10961031 19. Sloane PD, Zimmerman S, Gruber-Baldini AL, Hebel JR, Magaziner J, Konrad TR. Health and func- tional outcomes and health care utilization of persons with dementia in residential care and assisted liv- ing facilities: comparison with nursing homes. Gerontologist. 2005 Oct; 45 Spec No 1(1):124–32. https://doi.org/10.1093/geront/45.suppl_1.124 PMID: 16230759 20. Thomas KS, Zhang W, Cornell PY, Smith L, Kaskie B, Carder PC. State Variability in the Prevalence and Healthcare Utilization of Assisted Living Residents with Dementia. J Am Geriatr Soc. 2020 Jul; 68 (7):1504–1511. https://doi.org/10.1111/jgs.16410 PMID: 32175594 21. Resnick B, Galik E. Impact of care settings on residents’ functional and psychosocial status, physical activity and adverse events. Int J Older People Nurs. 2015 Dec; 10(4):273–83. https://doi.org/10.1111/ opn.12086 PMID: 26011088 22. Shah SM, Carey IM, Harris T, DeWilde S, Cook DG. Mortality in older care home residents in England and Wales. Age Ageing. 2013 Mar; 42(2):209–15. https://doi.org/10.1093/ageing/afs174 PMID: 23305759 23. McCann M, O’Reilly D, Cardwell C. A Census-based longitudinal study of variations in survival amongst residents of nursing and residential homes in Northern Ireland. Age Ageing. 2009 Nov; 38(6):711–7. https://doi.org/10.1093/ageing/afp173 PMID: 19752201 24. Liu LF, Wen MJ. A longitudinal evaluation of residents’ health outcomes in nursing homes and residen- tial care homes in Taiwan. Qual Life Res. 2010 Sep; 19(7):1007–18. https://doi.org/10.1007/s11136- 010-9667-8 PMID: 20449664 25. Rothera IC, Jones R, Harwood R, Avery AJ, Waite J. Survival in a cohort of social services placements in nursing and residential homes: factors associated with life expectancy and mortality. Public Health. 2002 May; 116(3):160–5. https://doi.org/10.1038/sj.ph.1900832 PMID: 12082598 26. Godden S, Pollock AM. The use of acute hospital services by elderly residents of nursing and residential care homes. Health Soc Care Community. 2001 Nov; 9(6):367–74. https://doi.org/10.1046/j.1365- 2524.2001.00314.x PMID: 11846815 27. McGregor MJ, McGrail KM, Abu-Laban RB, Ronald LA, Baumbusch J, Andrusiek D, et al. Emergency department visit rates and patterns in Canada’s Vancouver coastal health region. Can J Aging. 2014 Jun; 33(2):154–62. https://doi.org/10.1017/S0714980814000038 PMID: 24690211 28. Maxwell CJ, Amuah JE, Hogan DB, Cepoiu-Martin M, Gruneir A, Patten SB, et al. Elevated Hospitaliza- tion Risk of Assisted Living Residents With Dementia in Alberta, Canada. J Am Med Dir Assoc. 2015 Jul 1; 16(7):568–77. https://doi.org/10.1016/j.jamda.2015.01.079 PMID: 25717011 29. Hogan DB, Amuah JE, Strain LA, Wodchis WP, Soo A, Eliasziw M, et al. High rates of hospital admis- sion among older residents in assisted living facilities: opportunities for intervention and impact on acute care. Open Med. 2014 Mar 4; 8(1):e33–45. PMID: 25009683 30. Inacio MC, Lang CE, Khadka J, Watt AM, Crotty M, Wesselingh S, et al. Mortality in the first year of aged care services in Australia. Australas J Ageing. 2020 Dec; 39(4):e537–e544. https://doi.org/10. 1111/ajag.12833 PMID: 32815606 31. Bartley MM, Quigg SM, Chandra A, Takahashi PY. Health Outcomes From Assisted Living Facilities: A Cohort Study of a Primary Care Practice. J Am Med Dir Assoc. 2018 Mar; 19(3):B26. https://doi.org/10. 1016/j.jamda.2017.12.079 PMID: 32774179 32. Park S, Kim B, Kwon E. The Role of Senior Housing in Hospitalizations Among Vulnerable Older Adults With Multiple Chronic Conditions: A Longitudinal Perspective. Gerontologist. 2018 Sep 14; 58(5):932– 941. https://doi.org/10.1093/geront/gnx046 PMID: 28575200 33. Steves CJ, Spector TD, Jackson SH. Ageing, genes, environment and epigenetics: what twin studies tell us now, and in the future. Age Ageing. 2012 Sep; 41(5):581–6 https://doi.org/10.1093/ageing/ afs097 PMID: 22826292 34. Cameron ID, Dyer SM, Panagoda CE, Murray GR, Hill KD, Cumming RG, et al. Interventions for pre- venting falls in older people in care facilities and hospitals. Cochrane Database Syst Rev. 2018 Sep 7; 9 (9):CD005465. https://doi.org/10.1002/14651858.CD005465.pub4 PMID: 30191554 35. Whitney J, Close JC, Lord SR, Jackson SH. Identification of high risk fallers among older people liv- ing in residential care facilities: a simple screen based on easily collectable measures. Arch Gerontol PLOS ONE \| https://doi.org/10.1371/journal.pone.0286527 May 31, 2023 13 / 14 PLOS ONEMortality and hospitalizations in residential care facilities versus nursing homes or the community Geriatr. 2012 Nov-Dec; 55(3):690–5. https://doi.org/10.1016/j.archger.2012.05.010 PMID: 22770712 36. Patel PA, Gunnarsson C. A Passive Monitoring System in Assisted Living Facilities: 12-Month Compar- ative Study. Phys Occup Ther Geriatr. 2012 Mar; 30(1):45–52. https://doi.org/10.3109/02703181.2011. 650298 PMID: 22448082 37. Aldridge MD, Ornstein KA, McKendrick K, Reckrey J. Service Availability in Assisted Living and Other Community-Based Residential Settings at the End of Life. J Palliat Med. 2021 Nov; 24(11):1682–1688. https://doi.org/10.1089/jpm.2020.0625 PMID: 33826855 38. Watson LC, Lehmann S, Mayer L, Samus Q, Baker A, Brandt J, et al. Depression in assisted living is common and related to physical burden. Am J Geriatr Psychiatry. 2006 Oct; 14(10):876–83. https://doi. org/10.1097/01.JGP.0000218698.80152.79 PMID: 17001027 39. Rockwood JKh, Richard M, Garden K, Hominick K, Mitnitski A, Rockwood K. Precipitating and predis- posing events and symptoms for admission to assisted living or nursing home care. Can Geriatr J. 2013 Mar 5; 17(1):16–21. https://doi.org/10.5770/cgj.17.93 PMID: 24596590 40. Dobbs D, Hayes J, Chapin R, Oslund P. The relationship between psychiatric disorders and the ability to age in place in assisted living. Am J Geriatr Psychiatry. 2006 Jul; 14(7):613–20. https://doi.org/10. 1097/01.JGP.0000209268.37426.69 PMID: 16816015 41. Darton R, Ba¨ umker T, Callaghan L, Holder J, Netten A, Towers AM. The characteristics of residents in extra care housing and care homes in England. Health Soc Care Community. 2012 Jan; 20(1):87–96. https://doi.org/10.1111/j.1365-2524.2011.01022.x PMID: 21819475 42. Roos C, Alam M, Swall A, Bostro¨ m AM, Hammar LM. Factors associated with older persons’ percep- tions of dignity and well-being over a three-year period. A retrospective national study in residential care facilities. BMC Geriatr. 2022 Jun 23; 22(1):515. https://doi.org/10.1186/s12877-022-03205-1 PMID: 35739497. 43. De Kerimel J, Tavassoli N, Mathieu C, De Souto Barreto P, Berbon C, Blain H, et al. Seniors living in residential homes: a target population to implement ICOPE (Integrated care for older people) program in primary care. Jour Nursing Home Res 2020; 6:82–88 44. Tappenden P, Campbell F, Rawdin A, Wong R, Kalita N. The clinical effectiveness and cost-effective- ness of home-based, nurse-led health promotion for older people: a systematic review. Health Technol Assess. 2012; 16(20):1–72. https://doi.org/10.3310/hta16200 PMID: 22490205. 45. Coster S, Watkins M, Norman IJ. What is the impact of professional nursing on patients’ outcomes glob- ally? An overview of research evidence. Int J Nurs Stud. 2018 Feb; 78:76–83. https://doi.org/10.1016/j. ijnurstu.2017.10.009 Epub 2017 Oct 19. PMID: 29110907. 46. Changizi M, Kaveh MH. Effectiveness of the mHealth technology in improvement of healthy behaviors in an elderly population-a systematic review. Mhealth. 2017 Nov 27; 3:51. https://doi.org/10.21037/ mhealth.2017.08.06 PMID: 29430455. 47. Leroi I, Samus QM, Rosenblatt A, Onyike CU, Brandt J, Baker AS, et al. A comparison of small and large assisted living facilities for the diagnosis and care of dementia: the Maryland Assisted Living Study. Int J Geriatr Psychiatry. 2007 Mar; 22(3):224–32. https://doi.org/10.1002/gps.1665 PMID: 17044133 48. Manis DR, Rahim A, Poss JW, Bielska IA, Bronskill SE, Tarride JE´ , et al. Association Between Demen- tia Care Programs in Assisted Living Facilities and Transitions to Nursing Homes in Ontario, Canada: A Population-Based Cohort Study. J Am Med Dir Assoc. 2021 Oct; 22(10):2115–2120.e6. 49. Nussbaumer-Streit B, Klerings I, Dobrescu A, et al. Excluding non-English publications from evidence- syntheses did not change conclusions: a meta-epidemiological study. J Clin Epidemiol 2020; 118:42– 54. https://doi.org/10.1016/j.jclinepi.2019.10.011 PMID: 31698064 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286527 May 31, 2023 14 / 14 PLOS ONE
10.1371_journal.pone.0286008	RESEARCH ARTICLE Mixed-gender small-sided recreational team handball games in middle-aged and elderly are physiologically more demanding for women than men Ivone CarneiroID Susana Po´ voas1,2* 1, Peter Krustrup2,3,4,5, Carlo Castagna2,6, Rita PereiraID 7,8, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 1 Research Center in Sports Sciences, Health Sciences and Human Development, CIDESD, University of Maia, Maia, Portugal, 2 Department of Sports Science and Clinical Biomechanics, SDU Sport and Health Sciences Cluster (SHSC), University of Southern Denmark, Odense, Denmark, 3 Danish Institute for Advanced Study (DIAS), University of Southern Denmark, Odense, Denmark, 4 Sport and Health Sciences, University of Exeter, Exeter, United Kingdom, 5 Shanghai University of Sport (SUS), Shanghai, China, 6 Department of Biomolecular Sciences, School of Exercise and Health Sciences, Carlo Bo Urbino University, Urbino, Italy, 7 Laboratory of Metabolism and Exercise (LaMetEx), Research Centre in Physical Activity, Health and Leisure (CIAFEL), Faculty of Sport, University of Porto, Porto, Portugal, 8 University of Maia, Maia, Portugal OPEN ACCESS Citation: Carneiro I, Krustrup P, Castagna C, Pereira R, Po´voas S (2023) Mixed-gender small- sided recreational team handball games in middle- aged and elderly are physiologically more demanding for women than men. PLoS ONE 18(6): e0286008. https://doi.org/10.1371/journal. pone.0286008 Editor: Emiliano Cè, Università degli Studi di Milano: Universita degli Studi di Milano, ITALY Received: January 3, 2023 Accepted: May 5, 2023 Published: June 23, 2023 Copyright: © 2023 Carneiro et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: In order to protect subjects’ confidentiality and privacy, data are only available on request. Interested researchers may contact the Ethics Committee of the Faculty of Sport, Porto University (cefade@fade.up.pt). Funding: IC (SFRH/BD/144132/2019) and RP (SFRH/BD/136789/2018) are supported by grants from the Portuguese Foundation for Science and Technology. This work is supported by national funding through the Portuguese Foundation for * scpovoas@gmail.com Abstract This study examined the physical and physiological demands and perceived experience of a multicomponent exercise mode, recreational team handball (TH), for middle-aged/elderly men and women, played as same- vs. mixed-gender 6v6 game formats. Matches’ heart rate (HR), blood lactate (BL), perceived experience, activity profile, player load and accelerome- ter variables were assessed. Forty-one participants, with at least 12 weeks of experience with recreational TH (22 men; 69±4 years, 19 women; 66±6 years), performed 2 same- and 2 mixed-gender matches on an indoor 40x20 m TH court. A game format-by-gender interac- tion was observed for mean HR (%HRmax), time spent >80 and >90%HRmax, respiratory rat- ing of perceived exertion and for several of the external load variables (p�0.05). During mixed-gender matches, time spent >80 and >90%HRmax, was higher for women vs. men (p�0.017). During same- and mixed-gender matches, BL was lower for women than men (p�0.015). Time spent >90%HRmax was lower for women (p = 0.036), whereas time spent >80%HRmax was higher for men during same- vs. mixed-gender matches (p = 0.034). The frequency, %total match time and distance covered with high-demanding movements were higher for men during same-gender than during mixed-gender matches (p�0.036), and higher for men vs. women in same- and mixed-gender matches (p�0.046). The frequency of high-intensity actions, accelerations, time spent in the higher player load zones and total accumulated player load, were higher for men vs. women during same- and mixed-gender matches (p�0.044). Fun levels were very high (9.1–9.3 AU, 0–10). Mixed-gender small- sided recreational TH games are physiologically more demanding for middle-aged/elderly women compared to men. Men showed higher cardiovascular and activity profile demands when playing same-gender matches, which was opposite to women. Nevertheless, TH is a PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 1 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women high-intensity and motivating exercise mode for both genders, regardless the gender game format, meaning that exercise interventions may use same- and mixed-gender matches to promote participants’ health. Science and Technology, I.P., under project UIDB04045/2020. The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Introduction Multicomponent exercise interventions (i.e. a combination of resistance, endurance, and bal- ance training) have shown to be the best strategy to improve the overall health status of frail elderly individuals [1]. Recreational team handball (TH) is a multicomponent exercise, as it requires high levels of maximal strength and muscle power to sustain the forceful muscle con- tractions requested during TH specific movements, a high aerobic and anaerobic turnover to cope with the intermittent high-intensity nature of the game and specific agility and balance to respond to frequent changes of direction, speed and actions [2,3]. Played as formal (i.e., 7v7) or small-sided games (3v3 to 6v6), TH has shown to improve cardiometabolic health and physical fitness of different populations with different levels of fit- ness, and with or without experience with the sport [4–7]. Additionally, it has proved to induce positive musculoskeletal adaptations in young adult men [5] and women [8] and in postmeno- pausal women [9]. Recreational TH activity profile has only been described in detail for adult/middle-aged formerly trained men [3], male college students, [10] and unexperienced older men [11]. For these participant groups, the average distance covered was 3–5 km during 40–60 min matches and 40–54 specific TH high-intensity game actions were performed, which may provide a combined positive impact on cardiovascular, metabolic, and musculoskeletal health. These physical demands imposed by recreational TH practice alongside with the high physiological stress [mean heart rates (HR): 77–85% of maximal HR (HRmax), time spent >90%HRmax: 4–22% of total match time and average blood lactate (BL): 3.6–4.4 mmol�l-1] have been suggested as the main reasons for the broad-spectrum health benefits observed in the studied populations [3,10,11]. The studies reporting the health and physical fitness effects of recreational TH interventions have only been organized as gender-specific exercise programs. The possibility of having mixed-gender groups in recreational TH interventions, which mainly use match-playing as training tool, is of practical advantage in a community setting where classes frequently include participants of both genders. Although time-course aging-related changes differ in men and women, it is known that the cellular and molecular mechanisms of aging are initially better maintained in women. How- ever, after menopause, women seem to catch up and, in several parameters, reach the same lev- els of aging as men [12]. Nevertheless, elderly men are still stronger and faster than women, which is probably related to higher testosterone levels resulting in higher muscle mass in men [12]. Moreover, in young adult men and women, sex differences in fatigue and ability to recover during high-intensity training are shown even when matching for exercise parameters such as maximal oxygen uptake (VO2max) [13,14]. Despite male superior absolute performance due to genetic differences [15,16], questioning the valence of proposing mixed-gender game formats in recreational team sports-based exercise interventions, similar cardiovascular, meta- bolic and bone health improvements were shown both for men and women, after 16 weeks of mixed-gender recreational football training [17]. Nonetheless, to the best of our knowledge, the physiological demands and activity profile of same- vs. mixed-gender large and small- PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 2 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women sided games have only been analysed for children and adolescents during football practice and in a school setting [18,19]. HR response and perceived experience (namely the perceived effort and fun levels) showed a significantly lower mean value for girls when playing mixed with boys than when playing same-gender matches. No differences were reported in the game demands for boys when playing mixed- or same-gender matches [18]. However, boys per- ceived less fun when playing with girls than when playing within the same gender [18]. Inter- estingly, no significant differences were found in HR response between the genders for 8- to 9-year-old schoolchildren during small-sided team sports [19]. Therefore, understanding whether the physical and physiological demands of middle-aged and elderly men and women playing same- vs. mixed-gender recreational TH game formats differs is of utmost importance. Moreover, since motivation and enjoyment are key factors for long-term adherence to this type of exercise programmes [20], it is also important to ascertain the perceived experience during these different game formats. Thus, the aim of this study was to analyse the physiological response, the activity profile and the perceived experience of middle-aged and elderly men and women playing same- vs. mixed-gender recreational TH game formats. We hypothesized that the demands for men would be higher when playing same- vs. mixed-game formats, while the opposite would occur for women. Materials and methods Participants Forty-one participants (22 men and 19 women) were invited to participate in this study. Descriptive characteristics of the participants are presented in Table 1. Men’s stature, body mass, and distance covered in the Yo-Yo intermittent endurance level 1 test (YYIE1) were higher than women’s (p�0.038), while body mass index (BMI), fat mass values and time expe- rience with recreational TH, were higher for women than men (p<0.001). No differences were shown between the genders for chronological age. Inclusion criteria were: male and female participants aged over 50 years, that were at the moment involved in a recreational TH-based training programme for at least the last 12 weeks, with medical clearance to perform this type of exercise program. All the participants were informed about the study purposes, risks and benefits and signed a written informed consent according to the Declaration of Helsinki. Ethical approval was pro- vided by the local Institutional Review Board (CEFADE 19 2019). Table 1. Chronological age, stature, body composition, aerobic performance and recreational team handball expe- rience (data are presented as mean ± SD (range)) of the men (n = 22) and women (n = 19) that played the recrea- tional team handball matches. Variable Age (years) Stature (cm) Body mass (kg) BMI Fat mass (%) YYIE1 (m) Recreational TH experience (months) Men (n = 22) 69±4 (63–76) 168±5 (159–177) 74.8±8.6 (51–85) 26.3±2.6 (20–30) 25.3±4.8 (15–36) 754±439 (320–1560) 17±7 (4–24) Women (n = 19) 66±6 (56–77) 153±4 (148–161)* 64.7±9.0 (52–79)* 27.6±3.8 (21–34)* 37.3±4.7 (31–44)* 395±158 (200–600)* 28±12 (3–36)* BMI–Body mass index; TH–Team handball; YYIE1 –Yo-Yo intermittent endurance level 1 test. Significantly different from men (p�0.038). https://doi.org/10.1371/journal.pone.0286008.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 3 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women Experimental design The participants were evaluated for anthropometric variables, body composition and YYIE1 performance, in the week before the data collection. All the participants were familiarized with the procedures involved in the evaluations. Additionally, match internal and external load var- iables were monitored for each participant during 4 testing sessions. Each participant per- formed 4 recreational 6v6 (66.6 m2 per player) TH matches, 2 with only same-gender players (i.e., same-gender game format) and 2 with participants of both genders (i.e., mixed-gender game format: 3 men and 3 women), resulting in a total of 20 matches to be analysed. The same- and mixed-gender matches were performed in an indoor TH court (40x20 m). There were 48 hours between each testing session and the participants were asked to refrain from intense physical activity in the 48h before the testing sessions. All testing sessions were per- formed in the morning. To ensure the maintenance of proper hydration throughout the testing sessions, all participants were instructed to be hydrated and to drink water ad libitum. Each testing session started with a standardized 15-min warm-up (consisting of running, coordination, strength, flexibility, and balance exercises), followed by 3x15-min periods of rec- reational TH playing, interspersed by 2-min breaks. Some adaptations to the TH rules were made, such as, no body contact was allowed, and softer and lighter TH balls (47 cm circumfer- ence, GOALCHA, Fredericia, Denmark) than the official ones were used during the matches. There were some exceptions to the official TH rules, namely, no exclusions, no substitutions, no dribbling. Furthermore, the participants rotated positions every 2 min in a random order, including the goalkeeper, and the ball was immediately put back in play by the goalkeeper after a goal. All training sessions were instructed by a professional TH coach and physical education teacher and monitored by the research team. All the data collection and analysis were per- formed by the research team that comprised an experienced group of Sport Science, Physical Exercise and Health and Physical Education Teaching Master and PhD graduates. Participants’ internal load was evaluated as exercise HR, BL concentrations and differential rating of perceived exertion (RPE). Fun levels were also registered at the end of all testing ses- sions. External load was evaluated as frequency (n), percentage of total match duration (%) and, absolute (m) and relative (%) match distance covered in selected locomotor arbitrary cat- egories [3], considering this study participants’ individual speed thresholds, in order to account for inter-individual variability in external load, as well, as total distance covered. Experimental procedures Body (kg) and fat mass (%) were measured in a bioimpedance digital scale (Tanita Inner Scan BC 532, Tokyo, Japan) and stature (cm) was determined using a portable stadiometer (Seca 213, Hamburg, Germany), according to standardized protocols [11]. BMI was calculated as kg/m2. Aerobic performance was evaluated by the YYIE1. The YYIE1 test was performed in the same indoor TH wooden floor court as the matches, according to a protocol already described [11]. The cardiovascular load was monitored in the four testing sessions for each participant. For this purpose, the participants wore a HR monitor (Firstbeat Technologies Ltd., version 4.5.0.2, Jyva¨skyla¨, Finland) and their differential RPE (i.e., respiratory, muscular, and global) [21] and fun levels [22] were recorded immediately after the end of each testing session [23]. The partic- ipants were familiarised with the considered psychometric scales in training sessions per- formed before this study. Individual HRmax was determined as the highest value reached either during a VO2max test, the YYIE1 test or the matches, according to a multiple testing approach [24]. Moreover, capillary blood samples (30 μl) were drawn from the right earlobe at rest and during the last 5 min of the first and third match periods by a portable electroenzymatic lactate device analyser (Lactate Pro 2 LT-1730, Arkray, Amsterdam, The Netherlands), for PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 4 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women measurements of mean and peak BL concentrations. Each of the 41 participants were evalu- ated 4 times during the study (2 during same- and 2 during mixed-gender matches). External load was evaluated as match activities variables using time-motion analysis per- formed by video recordings (SONY-DCR-SX65E, digital video camera recorder) and acceler- ometer variables using Catapult MinimaxX S4 units (MinimaxX S4; Catapult Sports, Canberra, Australia). Frequency of the selected high-intensity match actions, i.e., jumps, throws, stops, changes of direction and one-on-one situations, and total number of actions were registered via video-analysis of the matches during 20 sessions (4 sessions per partici- pant). Players’ displacements were divided into eight locomotor categories: 1) standing still, 2) walking, 3) jogging, 4) fast running, 5) sprinting, 6) sideways medium-intensity, 7) sideways high-intensity, and 8) backwards movement [3]. High-intensity movements were the result of the sum of fast running, sprinting and sideways high-intensity categories. Individual speed thresholds were considered to determine the individual nature of the exercise intensity in each locomotor category [25] and were registered according to the protocol already described [11]. A test-retest analysis was performed for the selected time-motion variables in 8 matches (4 same-gender and 4 mixed-gender; randomly selected) and the analysis was initiated when intraclass correlation coefficient was >0.80. Table 2 shows the average speed thresholds calcu- lated for each gender in the studied population. Accelerometer data was collected using Cata- pult MinimaxX S4 units in indoor mode with global positioning system technology in inactive mode. Data was downloaded and processed using Catapult Sprint Version 5.1.1 (Catapult Innovations, Canberra, Australia). The units were located in a specific vest on players’ upper back. The validity and reliability of the accelerometers have been described elsewhere [26]. Player load (an estimate of the physical demand combining the instantaneous rate of change in acceleration in 3 planes [27]) variables were evaluated at a 100 Hz sampling rate. In this study, time spent in each player load zone, i.e., 0–0.1, >0.1–0.3, >0.3–0.6, >0.6–1.0, >1.0–1.5, >1.5–2.0, >2.0 [27] was presented as percentage of total match time, whereas total accumu- lated player load was presented as arbitrary units [26]. Frequency of accelerations and deceler- ations, categorized as low (1.50 to 2.14 m.s-1), medium (2.14 to 2.78 m.s-1) and high-intensity (>2.78 m.s-1), according to manufacture settings (Catapult Sprint Version 5.1.1 software man- ual, Catapult Innovations, Canberra, Australia), was determined. Statistical analyses Results are presented as means ± standard deviations (SD) and 95% of confidence interval (CI). Students’ unpaired t-test was used to assess differences between genders in chronological age, stature, body composition, BMI, aerobic performance and time experience with recrea- tional TH. To examine differences between the genders during the same- and mixed-gender Table 2. Player’s locomotor speed categories according to gender (average values for this population). Locomotor category Standing Walking Jogging Fast running Sprinting Sideways medium-intensity movements Sideways high-intensity movements Backwards movements https://doi.org/10.1371/journal.pone.0286008.t002 Men 0 km h-1 6 km h-1 9 km h-1 12 km h-1 17 km h-1 8 km h-1 10 km h-1 8 km h-1 Women 0 km h-1 6 km h-1 8 km h-1 11 km h-1 13 km h-1 6 km h-1 7 km h-1 5 km h-1 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 5 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women matches a two-way analysis of variance (ANOVA) for repeated measures with Bonferroni post hoc multiple comparison tests was used. Power calculations were performed to detect an effect size of 0.18 in repeated measures ANOVA (within factors only). With 2 measurements, alpha of 5%, and power of 80%, 39 participants were needed. Effect size was calculated using Cohen d and interpreted as trivial (<0.2), small (0.2–0.5), medium (0.5–0.8) and large (>0.8) [28]. Relative reliability of key HR variables was reported as Intraclass Correlation Coefficients (ICC3,1). Magnitude of the ICC values was rated as excellent, good, and poor for 0.75–1.00, 0.41–0.74, and 0.00–0.40 scores, respectively [29,30]. Statistical Package for the Social Sciences (SPSS Inc., version 23.0) was used for the analyses. The data were tested for normality using the Shapiro-Wilk test. Statistical significance was set at p�0.05. Results Internal load and perceived experience A game format x gender interaction was observed for relative mean HR (p = 0.043; Table 3 and Fig 1), for the percentage of time spent >80%HRmax (p = 0.051) and >90%HRmax (p = 0.011), and for respiratory RPE (p = 0.032). Percentages of time spent >80%HRmax (p = 0.008; 95% CI: -39.96, -6.25; d = 0.360) and >90%HRmax (p = 0.017; 95% CI: -19.41, -2.06; d = 0.451) were higher for women than for men, while playing mixed-gender matches. Men’s percentage of time spent >80%HRmax was higher during same- vs. mixed-gender matches (p = 0.034; 95% CI: -0.53–23.29; d = 0.141; Fig 2), while women’s percentage of time spent >90%HRmax was lower during same- vs. mixed-gender matches (p = 0.036; 95% CI: -12.86–0.44; d = 0.459). Women’s BL values were lower than men’s during same- (mean BL: p = 0.002; 95% CI: 0.49– 1.92; d = 0.980; peak BL: p = 0.011; 95% CI: 0.28–2.02; d = 0.999; first period BL: p�0.001; 95% CI: 0.61–2.08; d = 1.179; third period BL: p = 0.008; 95% CI: 0.29–1.85; d = 0.980) and mixed- gender matches (mean BL: p = 0.005; 95% CI: 0.37–1.97; d = 0.972; peak BL: p = 0.015; 95% CI: 0.29–2.48; d = 0.887; first period BL: p = 0.009; 95% CI: 0.34–2.19; d = 1.094; third period BL: p = 0.010; 95% CI: 0.27–1.88; d = 0.743) (Figs 3 and 4). No significant differences were observed for muscular and global RPE and fun levels between the gender game formats. In men, the ICC across the same- and mixed-gender game formats for mean HR, time >80% HRmax, and time >90% HRmax were 0.61 (0.18–0.84, good), 0.51 (0.12–0.76, good), and 0.35 (0.18–0.84, poor), respectively. For mean HR, time >80% HRmax, and time >90% HRmax, the ICC values for the different gender game formats in the female participants were 0.77 (0.50–0.91, excel- lent), 0.67 (0.31–0.86, good), and 0.72 (0.40–0.88, good), respectively. External load Locomotor activity profile. Participants’ locomotor activity profile is presented in Table 4 and Fig 5. A game format x gender interaction was found for frequency of standing, walking, sprinting, and backwards movements (p�0.048), for the percentage of total match duration spent standing, walking, jogging and in backwards movements (p�0.020), for the absolute and percentage of total match distance covered jogging and in backwards movements (p�0.037) and for the percentage of total match distance covered walking (p = 0.010). A game format effect was observed for frequency of standing, fast running, sideways medium-inten- sity, backwards and high-intensity movements (p�0.050; Table 4), for percentage of total match duration in sideways medium-intensity and backwards movements (p�0.031) and for absolute and percentage of total match distance covered fast running, and in sideways medium-intensity, backwards and high-intensity movements (p�0.022). Men´s frequency of high-demanding movements was higher during same- vs. mixed-gen- der matches (sprinting: p = 0.015; 95% CI: -1.02–0.02; d = 0.707; high-intensity: p = 0.036; 95% PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 6 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women Table 3. Men and women’s internal load during 6v6 same- and mixed-gender recreational team handball game formats (data are presented as mean ± SD). Men Women Game format Gender Interaction Same- gender Mixed-gender Same-gender Mixed-gender (n = 22) (n = 19) Cardiovascular demands Mean HR (%HRmax) Peak HR (%HRmax) Time >80% HRmax (%) Time >90% HRmax (%) Time �60% HRmax (%) Time 61–70% HRmax (%) Time 71–80% HRmax (%) Time 81–90% HRmax (%) BL concentrations First period mean BL (mmol�l-1) Third period mean BL (mmol�l-1) Match mean BL (mmol�l-1) Match peak BL (mmol�l-1) RPE Respiratory RPE (AU, 0–10) Muscular RPE (AU, 0–10) Global RPE (AU, 0–10) Fun (AU, 0–10) 78±5 86±6 36±28 7±16 1±2 17±15 47±21 29±17 4.0±1.4 3.0±1.5 3.5±1.7 4.2±1.5 6.8±1.8 6.8±1.7 6.6±1.8 9.3±0.9 76± 5 85±6 32±28¤ 3±5 2±3 19±12 48±18 29±25 4.1±1.8 3.1±1.6 3.6±1.6 4.5±2.0 5.9±2.2 6.0±2.0 5.6±2.5 9.1±1.1 76±8 88±6 44± 32 5±5 9±13 21±17 27±7 39±28 2.5±0.5¤ 1.7±0.5¤ 2.1±0.5¤ 2.8±0.7¤ 5.9±1.8 5.6±2.1 5.9±1.7 9.3±1.0 79±6 88±6 47±33 9±18# 1±2 14±15 38±18 38±24 2.5±0.7* 1.9±0.8* 2.2±0.7* 2.9±0.9* 6.1±2.9 5.3±2.5 5.9±2.6 9.1±1.5 AU—Arbitrary units; BL–Blood lactate; HR—Heart rate; HRmax−Maximal heart rate; RPE—Rating of perceived exertion. Significantly different from Men Mixed-Gender #Significantly different from Women Same-Gender ¤Significantly different from Men Same-Gender. https://doi.org/10.1371/journal.pone.0286008.t003 p 0.617 0.498 0.332 0.614 0.923 0.728 0.261 0.201 0.344 0.426 0.284 0.103 0.918 0.709 0.659 0.811 p p 0.427 0.340 0.043 0.136 0.565 0.101 0.083 0.059 0.001 0.004 0.001 0.008 0.948 0.543 0.991 0.849 0.043 0.150 0.051 0.011 0.104 0.299 0.231 0.503 0.820 0.989 0.895 0.529 0.032 0.206 0.069 0.335 CI: -5.88–0.43; d = 0.558). During same-gender game formats, men’s frequency of fast running (p = 0.005; 95% CI: 1.89–9.66; d = 0.960) and high-intensity movements was also higher than during mixed-gender matches (p = 0.006; 95% CI: 2.23–10.72; d = 1.011). During mixed-gen- der matches the frequency of fast running (p = 0.005; 95% CI: 1.50–7.80; d = 1.038), sprinting (p = 0.019; 95% CI: -0.18–0.37; d = 0.599) and in high-intensity movements (p = 0.004; 95% CI: 1.45–8.05; d = 1.013) was higher for men than in women. Men’s percentage of total match duration spent standing (p = 0.007; 95% CI: 0.78–7.06; d = 0.663) and walking (p = 0.008; 95% CI: 0.80–7.84; d = 0.740) was higher when playing mixed- vs. same-gender matches, however, in jogging (p = 0.018; 95% CI: -6.31–0.09; d = 0.576), sprint- ing (p = 0.026; 95% CI: -0.12–0.01; d = 0.958) and backwards movements (p<0.001; 95% CI: -7.49, -2.83; d = 1.455) was lower during mixed- vs. same-gender matches. During mixed-gen- der game formats, men’s percentage of total match duration spent fast running (p = 0.002; 95% CI: -0.65–1.27; d = 0.632), sprinting (p = 0.042; 95% CI: -0.02–0.04; d = 0.00), in backwards (p<0.001; 95% CI: -1.67–1.35; d = 0.005) and high-intensity movements (p = 0.002; 95% CI: -0.64–1.29; d = 0.004) was higher than for women. For women, the percentage of total match duration spent walking (p = 0.044; 95% CI: -6.50, -0.41; d = 0.769) was higher, and jogging was lower (p = 0.038; 95% CI: 1.05–4.77; d = 0.928) during same- vs. mixed-gender matches. During mixed-gender matches, men’s absolute match distance covered fast running (p = 0.004; 95% CI: 11.71–121.83; d = 1.022), sprinting (p = 0.046; 95% CI: -2.46–4.63; d = 0.627), and in backwards (p<0.001; 95% CI: -39.95–111.51; d = 0.438) and high-intensity movements PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 7 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women Fig 1. Mean heart rate (mean HR) for men and women in each gender game format (same- and mixed-gender). Data are presented as means±SD. https://doi.org/10.1371/journal.pone.0286008.g001 (p = 0.003; 95% CI: 10.98–124.73; d = 1.023) as well as total distance (p = 0.003; 95% CI: -425.52–826.37; d = 0.098) were higher than women’s. Men’s absolute and percentage of total match distance covered with high-intensity movements, namely, fast running (absolute: p = 0.003; 95% CI: -111.60, -18.06; d = 0.888; percentage: p = 0.012; 95% CI: -2.11, -0.15; d = 0.970) and sprinting (absolute: p = 0.018; 95% CI: -16.36–0.74; d = 0.778; percentage: p = 0.019; 95% CI: -0.33–0.02; d = 1.125) was higher when playing same- vs. mixed-gender matches. For women, percentage of total match distance covered walking was higher (p = 0.037; 95% CI: -7.10, -0.99; d = 1.106), and jogging was lower (p = 0.049; 95% CI: 0.13–6.26; d = 1.016) during same- vs. mixed-gender matches. During mixed-gender matches, the percentage of total match distance covered walking (p = 0.003; 95% CI: -5.57–6.32; d = 0.019), fast running (p = 0.008; 95% CI: 0.23–2.46; d = 1.096), and in backwards (p<0.001; 95% CI: -0.89–2.13; d = 0.006), and high-intensity movements (p = 0.006; 95% CI: 0.23–2.52; d = 1.102) was higher for men than women. During same-gender matches, the absolute and percentage of total match distance covered fast running (absolute: p = 0.019; 95% CI: 41.95–198.25; d = 1.063; percent- age: (p = 0.019; 95% CI: 0.59–3.62; d = 1.067) and in high-intensity movements (absolute: p = 0.021; 95% CI: 48.11–212.51; d = 1.141; percentage: (p = 0.021; 95% CI: 0.71–3.91; d = 1.086) was higher for men than for women. High-intensity game actions. A game format x gender interaction was observed for the frequency of throws (p = 0.045; Table 5). During mixed-gender matches, frequency of jumps (p = 0.009; 95% CI: 1.07–5.05; d = 1.501; Table 5), stops (p = 0.008; 95% CI: 0.06–4.45; d = 0.598), changes of direction (p = 0.004; 95% CI: -0.07–3.16; d = 1.020) and total high-intensity game actions (p = 0.014; 95% CI: 4.44–18.16; d = 0.712) was higher for men than for women. During same-gender matches, men showed higher frequency of jumps (p = 0.004; 95% CI: 0.77–4.96; d = 0.622), throws (p = 0.005; 95% CI: -2.21–2.84; d = 0.537), stops (p = 0.044; 95% PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 8 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women Fig 2. Percentage of total match time spent with heart rates above 80% of individual maximal HR (%HRmax) for men and women in each gender game format (same- and mixed-gender). Data are presented as means±SD. Significantly different from Men Mixed-Gender; ¤Significantly different from Men Same-Gender. https://doi.org/10.1371/journal.pone.0286008.g002 CI: 0.89–5.51; d = 2.104), one-on-one situations (p = 0.028; 95% CI: -0.53–2.59; d = 0.546) and total high-intensity game actions (p = 0.002; 95% CI: 2.30–19.00; d = 0.727) than women. Player load and accelerometer data. A game format x gender interaction was observed for time spent in >0.3–0.6% player load zone (p = 0.042; Table 6). During same- and mixed- gender matches, time spent in 0.0–0.1% player load zone was lower for men than for women (same-gender: p = 0.004; 95% CI: -13.66, -5.05; d = 2.048; mixed-gender: p<0.001; 95% CI: -9.28, -1.91; d = 1.232, respectively; Table 6). However, time spent in >0.1–0.3 (same-gender: p = 0.022; 95% CI: 0.43–8.22; d = 0.898; mixed-gender: p = 0.031; 95% CI: 0.74–8.99; d = 1.323), >1.5–2.0 (same-gender: p<0.001; 95% CI: 1.97–5.61; d = 8.276; mixed-gender: p<0.001; 95% CI: 1.39–5.34; d = 5.183) and >2.0% (same-gender: p = 0.002; 95% CI: 0.62– 3.05; d = 5.484; mixed: p = 0.004; 95% CI: 0.95–3.47; d = 6.458) player load zones and total accumulated player load (same-gender: p = 0.034; 95% CI: 9.77–78.75; d = 1.494; mixed-gen- der: p = 0.013; 95% CI: 3.13–73.34; d = 2.589) was higher for men than for women during same- and mixed-gender matches. Men’s low-intensity (p = 0.012; 95% CI: -4.70–22.49; d = 0.981) and total accelerations (p = 0.040; 95% CI: -0.97–37.80; d = 1.735) were higher than women’s during mixed-gender matches. During same-gender matches, medium (p = 0.014; 95% CI: -0.87–5.87; d = 2.196) and high-intensity (p = 0.005; 95% CI: -1.29–6.71; d = 3.541) accelerations were higher for men than for women. No significant differences were found for decelerations. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 9 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women Fig 3. Mean blood lactate (mmol�l-1) values for men and women in each gender game format (same- and mixed- gender). Data are presented as means±SD. Significantly different from Men Mixed-Gender; ¤Significantly different from Men Same-Gender. https://doi.org/10.1371/journal.pone.0286008.g003 Discussion This is the first study analysing the internal and external load of same- vs. mixed-gender recre- ational TH game formats for middle-aged and elderly men and women. Game format-by-gen- der interactions were found for relative mean HR, time spent >80%HRmax and >90%HRmax and respiratory RPE, and also for several of the external load variables considered. The main findings were that during mixed-gender matches, time spent >80%HRmax and >90%HRmax was higher for women than men, while mean and peak BL values were lower. Furthermore, in same-gender matches, men’s time spent with HR >80%HRmax, as well as several activity pro- file variables, such as high-intensity locomotor movements were higher than in mixed-gender matches. Studies using recreational TH as an exercise intervention have shown that this exercise mode is effective in inducing several health improvements in different age groups and in both genders [4–9]. It has also been shown that the time spent in high HRs zones (i.e., HR >90% HRmax) is positively associated (r = 0.61) with VO2max improvement [7]. Moreover, studies using recreational football have also suggested that time spent with HR >90% HRmax (˜20% of total match time) was the possible cause of the reported improvements in VO2max [31]. In the present study, the participants spent less time with HR >90%HRmax (3–9% of total match time) then the studies reported above. Nevertheless, improvements in VO2max were observed in postmenopausal women by spending around 11% of total match time in HR >90%HRmax [4], meaning that for this age population, spending less than 20% of total match time in HR >90%HRmax is still able to induce improvements in cardiorespiratory fitness. In the present study, during mixed-gender matches, time spent >80% and >90%HRmax was higher for women than for men. This may indicate that during mixed-gender matches, PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 10 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women Fig 4. Peak blood lactate (mmol�l-1) values for men and women in each gender game format (same- and mixed- gender). Data are presented as means±SD. Significantly different from Men Mixed-Gender; ¤Significantly different from Men Same-Gender. https://doi.org/10.1371/journal.pone.0286008.g004 women increase their physical demands in order to keep up with the pace imposed by men. In fact, the ability to perform high-intensity intermittent activity was higher in the male vs. female participants (YYIE1: 754±439 vs. 395±158 m, men and women, respectively). Yo-Yo tests’ per- formance has been positively associated with the amount of high-intensity running during football matches [32]. Men’s performance superiority was observed in the matches’ external demands. In fact, during same- and mixed-gender matches, the frequency of high-intensity movements and specific high-intensity game actions, the time spent in the highest player load zones, the total accumulated player load and frequency of total accelerations were higher for men than for women, which may be the reason why women showed lower mean and peak BL values. Nevertheless, during mixed-gender matches, women showed higher cardiovascular load than men, while performing less external load (e.g., less time spent and lower distance covered in high-intensity movements and lower total distance covered), which can be attrib- uted to their lower aerobic performance level compared to men [32]. During same-gender matches, men’s time spent >80%HRmax was higher than in mixed-gender matches, which is line with the locomotor activity profile. Men performed higher intensity locomotor move- ments during same- than during mixed-gender matches, which may explain the higher HRs shown during same-gender matches. Both during same- and mixed-gender matches, men’s and women’s relative mean and peak HR values were slightly lower than those reported for younger populations [3,5–8]. However, they were in line with those reported for postmenopausal women [4,9], who showed cardiovas- cular and musculoskeletal health improvements after 16 weeks of recreational TH training with those intensities. Relative reliability is a viable strategy to assess the interindividual consis- tency from evaluation-to-evaluation, providing information over the underpinning variability PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 11 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women Table 4. Men and women’s locomotor activity profile during 6v6 same- and mixed-gender recreational team handball game formats (data are presented as mean ± SD). Locomotor categories Men Women Same-Gender Mixed-Gender Same-Gender Mixed-Gender Game format Gender Interaction Freq (n) Standing Walking Jogging Fast running Sprinting Side Med Side High Back High-intensity Total Total match duration (%) Standing Walking Jogging Fast running Sprinting Side Med Side High Back High-intensity Total match distance (m) Walking Jogging Fast running Sprinting Side Med Side High Back High-intensity Total Total match distance (%) Walking Jogging Fast running Sprinting Side Med Side High Back High-intensity 12±8 94±17 74±18 13±8 0.8±1.4 15±9 0.0±0.0 30±17 14±9 240±42 9±7 50±10 28±8 2.2±1.4 0.1±0.2 3.1±2.4 0.0±0.0 7.7±.5 2.3±1.5 2203±443 1804±697 231±165 12.8±24.1 183±143 0.0±0.0 442±339 244±175 4875±713 46±13 36±10 5±3 0.3±0.5 4±3 0.0±0.0 9±6 5±3 (n = 22) (n = 19) 21±5¤ 101±10 74±14 10±6 0.1±0.3¤ 18±8¤ 0.0±0.0 10±6¤ 10±6¤ 234±30 14±5¤ 55±6¤ 24±6¤ 1.6±1.1 0.0±0.0¤ 3.4±2.0 0.0±0.0 2.0±1.5¤ 1.6±1.1 2534±501¤ 1614±494 147±102¤ 2.0±4.8¤ 203±116 0.0±0.0 101±82¤ 149±105¤ 4601±629 55±9¤ 35±9 3±2¤ 0.0±0.1¤ 4±3¤ 0.0±0.0 2±2¤ 3±2¤ 17±6¤ 93±8 73±13 6±4¤ 0.0±0.0 12±6 0.0±0.0 10±8 6±4¤ 211±23 17±5 54±6¤ 25±6 1.0±0.7 0.0±0.0 2.2±1.3 0.0±0.0 2.0±1.8 1.0±0.7 2433±579 1319±375 93±74¤ 0.0±0.0 89±48 0.0±0.0 64±60 93±74¤ 3998±671 61±8 33±7 2±2¤ 0.0±0.0 2±1 0.0±0.0 2±2 2±2¤ 16±7* 92±35 75±30 5±2* 0.0±0.0* 15±7 0.0±0.0 9±6* 5±2* 212±75 15±2* 50±4# 28±3# 1.5±2.6* 0.0±0.0* 3.3±1.5 0.0±0.0 1.9±1.6* 1.5±2.6* 2456±1047 1724±762# 56±30* 0.0±0.0* 152±66 0.0±0.0 70±53* 56±30* 4457±1729* 55±8# 39±7# 1±1 0.0±0.0 4±2 0.0±0.0 2±1* 1±1* p 0.038 0.645 0.965 0.050 0.180 0.009 <0.001 0.050 0.744 0.158 0.706 0.915 0.739 0.209 0.031 <0.001 0.675 0.107 0.499 0.016 0.169 0.020 <0.001 0.014 0.674 0.438 0.421 0.022 0.162 0.009 <0.001 0.018 p 0.913 0.459 0.747 0.002 0.025 0.548 0.008 0.001 0.084 <0.001 0.366 0.904 0.037 0.048 0.703 0.006 0.032 0.875 0.374 0.003 0.059 0.251 <0.001 0.003 0.063 0.478 0.835 0.005 0.062 0.491 0.001 0.004 p <0.001 0.048 0.526 0.500 0.043 0.276 <0.001 0.355 0.371 0.020 0.001 0.002 0.092 0.070 0.586 <0.001 0.076 0.080 0.010 0.088 0.057 0.669 <0.001 0.061 0.084 0.010 0.037 0.232 0.065 0.973 <0.001 0.163 Back–backwards movements; Freq–Frequency; High-intensity–sum of fast running, sprinting and sideways high-intensity movements; Side High–sideways high- intensity movements; Side Med–sideways medium-intensity movements. Significantly different from Men Mixed-Gender #Significantly different from Women Same-Gender ¤Significantly different from Men Same-Gender. https://doi.org/10.1371/journal.pone.0286008.t004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 12 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women Fig 5. Distance covered (m) for the locomotor activity categories in each gender game format (same- and mixed-gender). Data are presented as means ±SD. Back–backwards movements; High-intensity–sum of fast running, sprinting and sideways high-intensity movements; Side High–sideways high-intensity movements; Side Med–sideways medium-intensity movements. Significantly different from Men Mixed-Gender; #Significantly different from Women Same- Gender; ¤Significantly different from Men Same-Gender. https://doi.org/10.1371/journal.pone.0286008.g005 within and between the evaluations. In this study, the ICC value of the key HR variables was studied across the gender game formats (i.e., same- vs mixed-games). The results showed poor-to-good match intensity (HR) consistency for the men and good-to-excellent scores for the women. This data suggests individual monitoring of exercise intensity across the gender game formats when men play with their female counterparts. Interestingly, women’s mean and peak BL values were lower than the men’s (3.5–3.6 and 4.2–4.5 mmol�l-1, respectively), which could be related with the higher frequency, percentage of total match time and distance covered in high-intensity movements by men compared to women, during both gender game formats. Nevertheless, men’s mean and peak BL values were similar to those reported for Table 5. Men and women’s high-intensity game actions during 6v6 same- and mixed-gender recreational team handball game formats (data are presented as means ± SD). Game actions/gender game format Men Women Jumps (n) Throws (n) Stops (n) Changes of direction (n) One-on-one situations (n) Total high-intensity actions (n) Same-Gender Mixed-Gender Same-Gender Mixed-Gender Game format Gender Interaction (n = 22) (n = 19) 9±0 11±1 16±2 12±4 11±0 43±15 7±2 9±1 15±8 12±5 12±5 44±11 6±7¤ 8±7¤ 8±5¤ 8±4 8±8¤ 32±11¤ 2±3* 6±3 11±4* 9±2* 7±4 33±11* p 0.402 0.947 0.161 0.332 0.182 0.363 p 0.002 0.121 0.007 0.007 0.029 0.003 p 0.829 0.045 0.414 0.063 0.399 0.841 Significantly different from Men Mixed-Gender; ¤Significantly different from Men Same-Gender. https://doi.org/10.1371/journal.pone.0286008.t005 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 13 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women Table 6. Men and women’s player load and accelerometer data during 6v6 same- and mixed-gender recreational team handball game formats (data are presented as mean ± SD). Accelerometer variables/gender game formats Men Women Same-Gender Mixed-Gender Same-Gender Mixed- Gender Game format Gender Interaction (n = 22) (n = 19) p p p 12±6 37±4 21±1 9±2 10±2 8±0 2±1 353±22 24±6 14±5 18±6 55±17 13±0 6±0 6±5 24±6 17±3 40±0 19±2 8±2 11±1 5±0 1±0 317±4 14±7 12±3 17±8 43±18 14±5 7±5 8±5 29±14 25±4¤ 34±1¤ 20±4 8±4 11±5¤ 2±0¤ 0±0¤ 278±51¤ 10±6 6±1¤ 6±0¤ 21±6 11±9 6±6 21±28 38±43 22±2 36±5* 23±6# 7±3 10±6 2±1* 0±0* 263±29* 10±0* 7±0 5±1 22±1* 7±8 5±4 2±1 13±13 0.511 0.577 0.129 0.206 0.611 0.275 0.404 0.533 0.829 0.115 0.673 0.753 0.982 0.431 0.094 0.275 <0.001 0.018 0.773 0.414 0.131 <0.001 0.001 0.008 0.042 0.016 0.006 0.011 0.734 0.479 0.627 0.598 0.082 0.694 0.042 0.448 0.169 0.554 0.478 0.740 0.721 0.708 0.182 0.531 0.962 0.737 0.841 0.951 Player load zones Time 0.0–0.1 (%) Time >0.1–0.3 (%) Time >0.3–0.6 (%) Time >0.6–1.0 (%) Time >1.0–1.5 (%) Time >1.5–2.0 (%) Time >2.0 (%) Total accumulated (AU) Accelerations Low-intensity (n) Medium-intensity (n) High-intensity (n) Total (n) Decelerations Low-intensity (n) Medium-intensity (n) High-intensity (n) Total (n) AU–arbitrary units. *Significantly different from Men Mixed-Gender #Significantly different from Women Same-Gender ¤Significantly different from Men Same-Gender. https://doi.org/10.1371/journal.pone.0286008.t006 middle-aged former TH players (3.6 and 4.2 mmol�l-1, respectively) [3]. Based on this study results, playing same- or mixed-gender matches results in internal load values in the range to induce cardiovascular adaptations seen in intervention studies with other age-groups using recreational TH as exercise mode. During same-gender matches, the lower BL values observed for women when compared to men, could be related with the higher frequency of TH high-demanding game-specific actions such as jumps, throws, stops, one-on-one situations and total high-intensity actions, more time spent in higher player load zones (>1.5–2.0 and >2.0), higher total accumulated player load, and higher frequency of medium and high-intensity accelerations found in men. The BL values were also lower for women in mixed-gender matches, which may have been the result of men also showing higher frequency of TH high-intensity game-specific actions such as jumps, stops, changes of direction and total high-intensity actions, spending more time in higher player load zones (>1.5–2.0 and >2.0), showing higher total accumulated player load and performing more low-intensity and total accelerations than women, indicating that men were more involved in the matches than women and, consequently, had a higher level of par- ticipation in the matches. This is important to maintain the participants’ motivation to keep playing [33]. Accordingly, women could have felt less involved and motivated in the game. Nevertheless, this was not the case, since differential RPE were similar for men and women as PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 14 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women well as fun levels, which were very high (9.1–9.3 AU on a 0–10 scale) in both gender game for- mats. Therefore, for this population, both gender game formats could be recommended for both genders, contrary to children, with boys reporting more fun when playing football in same- vs. mixed-gender matches [18]. Our study showed that in recreational TH, men and women spent around 50–55% of the total match time walking and 24–28% jogging. The intensities alternate during a TH game due to its intermittent nature, with the locomotor activity pattern changing from standing and walking, jogging and moderate running, fast running and sprinting, sideways and backwards movements [34,35], requiring a high level of endurance to keep up with the game demands. Nevertheless, TH performance is also highly influenced by cognitive, tactical, and social factors [2]. This may explain the higher amount of time spent jogging (24–28% vs. 16%, respectively) and the less time spent in high-intensity movements (1–2% vs. 15%, respectively), by this study participants, that have a clear lack of knowledge and experience with the sport, when compared to former TH players [3], whose technical-tactical sport background may have allowed them to better manage and control the game than the unexperienced older participants. From a physiological point of view, in order to achieve the highest intensities, recreational TH interventions should be preferentially organized as same-gender game formats for men and as mixed-gender for women. Nevertheless, men’s and women’s HRs in both same- and mixed-gender TH matches, were in line with studies showing cardiovascular health improve- ments, which from a practical perspective, means that both gender game formats may be used to induce cardiovascular adaptations for this population. Perhaps, it would be of interest to organize different recreational TH training sessions by changing the gender game formats, allowing men and women to alternate and benefit from different match demands given by the different gender game formats. Strengths and limitations The main strength of this study is that it is the first to analyse the differences in internal and external load variables in both men and women playing same- vs. mixed-gender recreational TH games. This is important since this multicomponent exercise programme is usually imple- mented in community settings, aiming at men and women, and therefore understanding the physical and physiological demands and the perceived experience for each gender when play- ing same vs. mixed-gender games is crucial to plan and organize the best training settings for this population to achieve broad health improvements. Nevertheless, one study limitation is the fact that during this study the participants were only evaluated during 4 TH training ses- sions (2 same- and 2 mixed-gender matches). Given the practical value of organising mixed-gender games-based exercise interventions, future randomized controlled studies aiming at assessing participants’ health impact are needed. Conclusion Cardiovascular demands were higher for middle-aged and elderly women than for age- matched men during mixed-gender matches. In men, same-gender game formats were more demanding than mixed-gender game formats, while in women the inverse occurred. Interest- ingly, fun levels were reported as being very high for both genders, independently of the gen- der game format, which can possibly lead to greater long-term adherence to this exercise program. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 15 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women It could be concluded that recreational TH is an intermittent high-intensity and motivating exercise mode with potential to induce several health improvements for middle-aged and elderly men as well as for women, regardless the gender game format. From a practical point of view, same- and mixed-gender matches may be organised with the aim to promote health and physical fitness for this population. Since women are more physically and physiologically challenged when playing with men, a lead-in period with same-gender formats may be recom- mended for women, when implementing mixed-gender recreational TH-based exercise interventions. Acknowledgments This study is part of the Handball for Health project, which has the support of the Portuguese Handball Federation, the European Handball Federation, Porto Sports Medicine Center (IPDJ, IP) and Gaia City Hall. We would like to thank all the participants for their committed participation. We would also like to express our gratitude to the members of the Handball4- Health project. Author Contributions Conceptualization: Ivone Carneiro, Peter Krustrup, Carlo Castagna, Susana Po´voas. Data curation: Ivone Carneiro, Rita Pereira, Susana Po´voas. Formal analysis: Ivone Carneiro. Funding acquisition: Peter Krustrup, Susana Po´voas. Investigation: Ivone Carneiro, Rita Pereira, Susana Po´voas. Methodology: Ivone Carneiro, Peter Krustrup, Carlo Castagna, Susana Po´voas. Project administration: Susana Po´voas. Resources: Peter Krustrup, Susana Po´voas. Supervision: Peter Krustrup, Carlo Castagna, Susana Po´voas. Validation: Ivone Carneiro, Carlo Castagna, Susana Po´voas. Visualization: Ivone Carneiro, Rita Pereira, Susana Po´voas. Writing – original draft: Ivone Carneiro, Carlo Castagna. Writing – review & editing: Ivone Carneiro, Peter Krustrup, Rita Pereira, Susana Po´voas. References 1. Cadore EL, Rodrı´guez-Mañas L, Sinclair A, Izquierdo M (2013) Effects of different exercise interven- tions on risk of falls, gait ability, and balance in physically frail older adults: a systematic review. Rejuve- nation Res 16: 105–114. https://doi.org/10.1089/rej.2012.1397 PMID: 23327448 2. Wagner H, Finkenzeller T, Wu¨rth S, von Duvillard SP (2014) Individual and team performance in team- handball: a review. J Sports Sci Med 13: 808–816. PMID: 25435773 3. Po´ voas SCA, Castagna C, Resende C, Coelho EF, Silva P, et al. (2017) Physical and Physiological Demands of Recreational Team Handball for Adult Untrained Men. Biomed Res Int 2017: 6204603. https://doi.org/10.1155/2017/6204603 PMID: 28466014 4. Pereira R, Krustrup P, Castagna C, Coelho E, Santos R, et al. (2020) Effects of a 16-week recreational team handball intervention on aerobic performance and cardiometabolic fitness markers in postmeno- pausal women: A randomized controlled trial. Prog Cardiovasc Dis 63: 800–806. https://doi.org/10. 1016/j.pcad.2020.10.005 PMID: 33222989 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 16 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women 5. Hornstrup T, Lowenstein FT, Larsen MA, Helge EW, Po´ voas S, et al. (2019) Cardiovascular, muscular, and skeletal adaptations to recreational team handball training: a randomized controlled trial with young adult untrained men. Eur J Appl Physiol 119: 561–573. https://doi.org/10.1007/s00421-018-4034-5 PMID: 30474740 6. Hornstrup T, Po´ voas S, Helge JW, Melcher PS, Fristrup B, et al. (2020) Cardiovascular and metabolic health effects of team handball training in overweight women: Impact of prior experience. Scand J Med Sci Sports 30: 281–294. https://doi.org/10.1111/sms.13563 PMID: 31596971 7. Po´ voas SCA, Castagna C, Resende C, Coelho EF, Silva P, et al. (2018) Effects of a Short-Term Recre- ational Team Handball-Based Programme on Physical Fitness and Cardiovascular and Metabolic Health of 33-55-Year-Old Men: A Pilot Study. Biomed Res Int 2018: 4109796. 8. Hornstrup T, Wikman JM, Fristrup B, Po´voas S, Helge EW, et al. (2018) Fitness and health benefits of team handball training for young untrained women-A cross-disciplinary RCT on physiological adapta- tions and motivational aspects. J Sport Health Sci 7: 139–148. https://doi.org/10.1016/j.jshs.2017.09. 007 PMID: 30356492 9. Pereira R, Krustrup P, Castagna C, Coelho E, Santos R, et al. (2021) Effects of recreational team hand- ball on bone health, postural balance and body composition in inactive postmenopausal women—A ran- domised controlled trial. Bone 145: 115847. https://doi.org/10.1016/j.bone.2021.115847 PMID: 33450430 10. Stojiljković N, Scanlan A, Dalbo V, Stankovic R, Milanović Z, et al. (2020) Physiological responses and activity demands remain consistent irrespective of team size in recreational handball. Biol Sport 37: 69–78. https://doi.org/10.5114/biolsport.2020.92516 PMID: 32205912 11. Carneiro I, Krustrup P, Castagna C, Pereira R, Coelho E, et al. (2022) Acute physiological response to different recreational team handball game formats in over 60-year-old inactive men. PLOS ONE 17: e0275483. https://doi.org/10.1371/journal.pone.0275483 PMID: 36227920 12. Ha¨gg S, Jylha¨ va¨ J (2021) Sex differences in biological aging with a focus on human studies. eLife 10: e63425. https://doi.org/10.7554/eLife.63425 PMID: 33982659 13. Astorino TA, Allen RP, Roberson DW, Jurancich M, Lewis R, et al. (2011) Adaptations to high-intensity training are independent of gender. European Journal of Applied Physiology 111: 1279–1286. https:// doi.org/10.1007/s00421-010-1741-y PMID: 21132441 14. Mageean AL, Alexander RP, Mier CM (2011) Repeated Sprint Performance in Male and Female Col- lege Athletes Matched for VO(2)max Relative to Fat Free Mass. Int J Exerc Sci 4: 229–237. PMID: 27182366 15. Perez-Gomez J, Rodriguez GV, Ara I, Olmedillas H, Chavarren J, et al. (2008) Role of muscle mass on sprint performance: gender differences? European Journal of Applied Physiology 102: 685–694. https://doi.org/10.1007/s00421-007-0648-8 PMID: 18084774 16. Thibault V, Guillaume M, Berthelot G, Helou NE, Schaal K, et al. (2010) Women and Men in Sport Per- formance: The Gender Gap has not Evolved since 1983. J Sports Sci Med 9: 214–223. PMID: 24149688 17. Mohr M, Skoradal M-B, Andersen TR, Krustrup P (2019) Gender-dependent evaluation of football as medicine for prediabetes. European Journal of Applied Physiology 119: 2011–2024. https://doi.org/10. 1007/s00421-019-04188-5 PMID: 31346789 18. Po´ voas S, Randers MB, Krustrup P, Larsen MN, Pereira R, et al. (2018) Heart Rate and Perceived Experience Differ Markedly for Children in Same- versus Mixed-Gender Soccer Played as Small- and Large-Sided Games. Biomed Res Int 2018: 7804642. https://doi.org/10.1155/2018/7804642 PMID: 30175142 19. Bendiksen M, Williams CA, Hornstrup T, Clausen H, Kloppenborg J, et al. (2014) Heart rate response and fitness effects of various types of physical education for 8- to 9-year-old schoolchildren. Eur J Sport Sci 14: 861–869. https://doi.org/10.1080/17461391.2014.884168 PMID: 24533471 20. Elbe AM, Strahler K, Krustrup P, Wikman J, Stelter R (2010) Experiencing flow in different types of physical activity intervention programs: three randomized studies. Scand J Med Sci Sports 20 Suppl 1: 111–117. https://doi.org/10.1111/j.1600-0838.2010.01112.x PMID: 20210905 21. McLaren SJ, Smith A, Spears IR, Weston M (2017) A detailed quantification of differential ratings of per- ceived exertion during team-sport training. J Sci Med Sport 20: 290–295. https://doi.org/10.1016/j. jsams.2016.06.011 PMID: 27451269 22. Grant S, Aitchison T, Henderson E, Christie J, Zare S, et al. (1999) A comparison of the reproducibility and the sensitivity to change of visual analogue scales, Borg scales, and Likert scales in normal sub- jects during submaximal exercise. Chest 116: 1208–1217. https://doi.org/10.1378/chest.116.5.1208 PMID: 10559077 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 17 / 18 PLOS ONERecreational team handball for middle-aged and elderly men and women 23. Castagna C, Bizzini M, Po´ voas SCA, D’Ottavio S (2017) Timing Effect on Training-Session Rating of Perceived Exertion in Top-Class Soccer Referees. Int J Sports Physiol Perform 12: 1157–1162. https:// doi.org/10.1123/ijspp.2016-0626 PMID: 28121188 24. Po´ voas SCA, Krustrup P, Pereira R, Vieira S, Carneiro I, et al. (2019) Maximal heart rate assessment in recreational football players: A study involving a multiple testing approach. Scand J Med Sci Sports 29: 1537–1545. https://doi.org/10.1111/sms.13472 PMID: 31112342 25. Casamichana D, Morencos E, Romero-Moraleda B, Gabbett TJ (2018) The Use of Generic and Individ- ual Speed Thresholds for Assessing the Competitive Demands of Field Hockey. J Sports Sci Med 17: 366–371. https://doi.org/10.1234/12345678 PMID: 30116109 26. Boyd LJ, Ball K, Aughey RJ (2011) The reliability of MinimaxX accelerometers for measuring physical activity in Australian football. Int J Sports Physiol Perform 6: 311–321. https://doi.org/10.1123/ijspp.6.3. 311 PMID: 21911857 27. Randers MB, Hagman M, Brix J, Christensen JF, Pedersen MT, et al. (2018) Effects of 3 months of full- court and half-court street basketball training on health profile in untrained men. J Sport Health Sci 7: 132–138. https://doi.org/10.1016/j.jshs.2017.09.004 PMID: 30356444 28. Cohen J (1988) Statistical Power Analysis for the Behavioral Sciences. Lawrence Erlbaum Associates, Hillsdale, NJ, USA 2nd edition. 29. Hopkins WG (2000) Measures of reliability in sports medicine and science. Sports Med 30: 1–15. https://doi.org/10.2165/00007256-200030010-00001 PMID: 10907753 30. Weir JP (2005) Quantifying test-retest reliability using the intraclass correlation coefficient and the SEM. J Strength Cond Res 19: 231–240. https://doi.org/10.1519/15184.1 PMID: 15705040 31. Milanović Z, Pantelić S, Čović N, Sporisˇ G, Krustrup P (2015) Is Recreational Soccer Effective for Improving VO2max A Systematic Review and Meta-Analysis. Sports Med 45: 1339–1353. https://doi. org/10.1007/s40279-015-0361-4 PMID: 26210229 32. Krustrup P, Mohr M, Amstrup T, Rysgaard T, Johansen J, et al. (2003) The yo-yo intermittent recovery test: physiological response, reliability, and validity. Med Sci Sports Exerc 35: 697–705. https://doi.org/ 10.1249/01.MSS.0000058441.94520.32 PMID: 12673156 33. Nielsen G, Wikman JM, Jensen CJ, Schmidt JF, Gliemann L, et al. (2014) Health promotion: The impact of beliefs of health benefits, social relations and enjoyment on exercise continuation. Scand J Med Sci Sports 24: 66–75. https://doi.org/10.1111/sms.12275 PMID: 24944133 34. Michalsik LB, Aagaard P, Madsen K (2013) Locomotion characteristics and match-induced impairments in physical performance in male elite team handball players. Int J Sports Med 34: 590–599. https://doi. org/10.1055/s-0032-1329989 PMID: 23258606 35. Po´ voas SC, Seabra AF, Ascensão AA, Magalhães J, Soares JM, et al. (2012) Physical and physiologi- cal demands of elite team handball. J Strength Cond Res 26: 3365–3375. https://doi.org/10.1519/JSC. 0b013e318248aeee PMID: 22222325 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286008 June 23, 2023 18 / 18 PLOS ONE
10.1371_journal.pone.0282086	RESEARCH ARTICLE No evidence for a mixing benefit—A registered report of voluntary dialect switching Mathieu Declerck1, Neil W. KirkID 2* 1 Linguistics and Literary Studies, Vrije Universiteit Brussel, Brussels, Belgium, 2 Division of Psychology & Forensice Sciences, Abertay University, Dundee, Scotland, United Kingdom a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 * n.kirk@abertay.ac.uk Abstract This is a Registered Report and may have an associated publication; please check the article page on the journal site for any related articles. OPEN ACCESS Citation: Declerck M, Kirk NW (2023) No evidence for a mixing benefit—A registered report of voluntary dialect switching. PLoS ONE 18(5): e0282086. https://doi.org/10.1371/journal. pone.0282086 Editor: Jie Wang, Education University of Hong Kong, HONG KONG Received: May 21, 2022 Accepted: February 7, 2023 Published: May 4, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0282086 Copyright: © 2023 Declerck, Kirk. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All data files are available from the Open Science Framework at Previous language production research with bidialectals has provided evidence for similar language control processes as during bilingual language production. In the current study, we aimed to further investigate this claim by examining bidialectals with a voluntary lan- guage switching paradigm. Research with bilinguals performing the voluntary language switching paradigm has consistently shown two effects. First, the cost of switching lan- guages, relative to staying in the same language, is similar across the two languages. The second effect is more uniquely connected to voluntary language switching, namely a benefit when performing in mixed language blocks relative to single language blocks, which has been connected to proactive language control. While the bidialectals in this study also showed symmetrical switch costs, no mixing effect was observed. These results could be taken as evidence that bidialectal and bilingual language control are not entirely similar. Introduction Previous research has indicated that when bidialectals (i.e., speakers of a regional dialect that are also fluent in a standard language variety) produce language, both the standard language and dialect are activated (e.g., [1, 2]), which is assumed to lead to competition among both lan- guage varieties. Similar to bilinguals (for reviews, see [3, 4]), a language control process is assumed to be implemented to deal with the competition between language varieties in bidia- lectals [1–2, 5, 6]. Some studies have suggested that the language control process implemented by bidialectal speakers of closely related language varieties is similar to the bilingual language control process [1, 2, 5]. In the current study, we set out to further investigate the language control process implemented by bidialectals and its relation to the bilingual language control process by letting bidialectals perform in a voluntary language switching paradigm. Voluntary language switching [7–16] usually requires participants to name pictures in one of two languages. Unlike other variants of the language switching paradigm (for a review, see [4]), which indicate the language that should be used for each stimulus through cues (e.g., dif- ferently colored frames around the stimuli; [17]), alternating languages (e.g., AABBAABB, with A and B referring to trials in different languages; [18]), or written words (e.g., [19]), vol- untary language switching allows the participants to choose the language on each trial. PLOS ONE \| https://doi.org/10.1371/journal.pone.0282086 May 4, 2023 1 / 14 PLOS ONEhttps://osf.io/wmr3d/ (DOI: 10.17605/OSF.IO/ WMR3D). Funding: This research was funded by a Carnegie Trust for the Universities of Scotland Research Incentive Grant (https://www.carnegie-trust.org/) awarded to NWK (RIG009864). The funders did not have a role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Voluntary language switching in bidialectals Similar to other variants of language switching (e.g., [17, 20–23]), voluntary language switching with bilinguals generally results in a cost when switching languages, relative to stay- ing in the same language across trials ([8–16]; however, see [7]). Indicative of the voluntary language switching paradigm with bilinguals is that these switch costs are similar across lan- guages (i.e., symmetrical switch costs [8–14]). This is in contrast with other variants of the lan- guage switching paradigm, where asymmetrical switch costs, which entails larger switch costs in the first language (L1) than in the second language (L2) and is typically used as a measure of inhibitory control (e.g., [17, 24–26] for a review, see [27]), are relatively often found. Accord- ing to Gollan and Ferreira [11], the absence of asymmetrical switch costs with the voluntary language switching paradigm is because this paradigm allows bilinguals to name “easier” words in their L2. This should result in a more similar L1 and L2 activation level for the pro- duced words regardless of language proficiency, and thus might lead to symmetrical switch costs. Another measure of language control are mixing costs (for a review, see [28]). Mixing costs entail worse performance in repetition trials in mixed language blocks relative to performance in single language blocks. This measure has been explained with control processes that are implemented in anticipation of any upcoming cross-language competition (i.e., proactive lan- guage control; e.g., [26]) and the mental cost to maintain and monitor two languages (e.g., [8]). Mixing costs are a highly stable effect in all variants of language switching (e.g., [20, 25, 26, 29–31]), with the exception of voluntary language switching. Voluntary language switching studies actually tend to show a mixing benefit (i.e., worse performance in single language blocks relative to performance in repetition trials in mixed language blocks) in one [11, 12] or both languages [8, 9, 13, 14]. The mixing benefit in L2 observed by Gollan and Ferreira [11] was explained by assuming that only “easier” words are produced in L2 in the voluntary lan- guage switching paradigm. Based on this explanation, one would expect that the mixing bene- fit was only observed for words that were consistently named in L2, but that was not the case [11]. An alternative explanation that can account for a mixing benefit across both languages comes from de Bruin et al. [8]: Production in a single language block requires substantial pro- active inhibition of the non-target language [32, 33]. When producing in a mixed language block with voluntary language switching, bilinguals do not require substantial proactive con- trol processes to guide language production, since participants can choose which language to use on any given trial. So, because the implementation of more proactive control processes during single than mixed language blocks is more taxing, better performance is expected in the latter block type when using a voluntary language switching paradigm. From this overview, it appears that voluntary vs. involuntary language switching has a large impact on measures of the bilingual language control process. In the current study, we set out to investigate if this is also the case for bidialectals by asking bidialectals to perform in a voluntary language switching paradigm. The few studies that inves- tigated control processes with bidialectals and bilinguals provide evidence for a shared lan- guage control process. Similar to bilinguals, bidialectals show a cost to switching between language varieties, relative to staying in the same language variety across languages [1, 2, 5, 6]. Kirk and colleagues [5], for instance, let bidialectals (English-Orcadian) name pictures during a language switching task, where the language variety on each trial was indicated by differently colored frames corresponding to each language variety (cf. involuntary language switching). The results showed that switching between language varieties results in worse performance than staying in the same language variety across trials. Similar to bilinguals (e.g., [17]), these bidialectals showed asymmetrical switch costs, with larger switch costs in their more dominant language variety (dialect) than in their less dominant language variety (standard language; see also [2]). Asymmetrical switch costs were even found with new bidialectals (English- PLOS ONE \| https://doi.org/10.1371/journal.pone.0282086 May 4, 2023 2 / 14 PLOS ONEVoluntary language switching in bidialectals Dundonian), whereas more fluent bidialectals showed symmetrical switch costs [1]. The latter pattern is similar to that observed with second language learners and highly proficient bilin- guals, respectively ([21, 34]). Finally, Kirk and colleagues also showed worse performance in repetition trials in mixed language blocks than in single language blocks [5]. So, along the lines of prior bilingual studies (e.g., [26, 29–31]), mixing costs can be observed with bidialectals dur- ing involuntary language switching. The similarities of bidialectals and bilinguals in previous studies that relied on involuntary language switching seem to indicate that similar language control processes are implemented by these two groups during language production. While it might seem obvious that bilinguals and bidialectals rely on the same language control processes, previous related research indi- cates that is not necessarily the case. For instance, control processes used within the same lan- guage have been shown to be different to control used between languages [35]. Even more damning for the assumption that similar control processes are used throughout language pro- cessing is that some studies found evidence that different language pairs do not necessarily converge when it comes to language control [36]. A similar discrepancy in language control has been observed across modalities within the same bilinguals (e.g., [37]). These studies pro- vide evidence against the claim that language control is domain general [17, 24], as the control processes within a domain (i.e., language processing) are sometimes even different. There have also been attempts to objectively distinguish languages from dialects (and thus bilinguals from bidialectals) on a cognitive level using the picture word interference paradigm, which initially suggested dialect items were processed as within-language competitors, akin to synonyms [38]. However, more recent evidence has challenged some of these findings [39, 40]. Thus, the extent to which bidialectals are similar to bilinguals is still unclear and can have theo- retical and methodological implications for research comparing bilinguals and monolinguals. For example, research suggesting that there is a general executive control advantage for bilin- guals over monolinguals as a result of the regular engagement of language control mechanisms [41], could be invalidated by the presence of bidialectal speakers who also use these mecha- nisms, but who are erroneously categorized as monolingual [42]. To further investigate the issue of whether there are similar language control processes used in bilingual and bidialectal language production, we set out to examine whether a similar pat- tern can be observed with bidialectals as with bilinguals in a voluntary language switching par- adigm. In the current study, we relied on Scottish speakers of a specific type of Scots as the dialect of interest. Although Scots is recognized by the European Charter for Regional or Minority Languages as a separate minority language (from English), it is generally not given this status, with many facing ridicule for suggesting that Scots and English are separate lan- guages [43]. Even a majority of speakers themselves do not hold this view, with one Scottish Government [44] report demonstrating that 65% of respondents consider their use of Scots as “just a way of speaking”. Consequently, these speakers are likely to identify as monolingual rather than bilingual—or even bidialectal–especially if language background measures are not sensitive to the existence of non-standard varieties (see, [1, 5]). More specifically, we will test speakers of Dundonian Scots and (Scottish) Standard English. Dundonian is an urban dialect used in and around the Scottish city of Dundee. Like other urban Scots dialects, it exists as a lower status variety in a diglossic situation with (Scottish) Standard English as the prestige variety [45]. Whereas Dundonian Scots overlaps substantially with its corresponding standard language, there are several notable differences. First, it is char- acterized by phonetic differences relative to the standard language, such as vowel differences (e.g., Standard English ‘pie’ / paɪ / vs. Dundonian ‘peh’ /pε/) and monophthongisation (e.g., Standard English ‘mouse’ /maʊs/ vs. Dundonian ‘moose’ /mu:s/). Second, and most important for the current study, there are also words entirely different in Dundonian than its PLOS ONE \| https://doi.org/10.1371/journal.pone.0282086 May 4, 2023 3 / 14 PLOS ONEVoluntary language switching in bidialectals corresponding translation equivalent in Standard English (e.g., “crying” in Standard English would be “greetin” in Dundonian). If bidialectals and bilinguals rely on similar language control processes, we expect to observe symmetrical switch costs in the voluntary language switching paradigm with the English-Dundonian bidialectals, similar to the pattern observed with bilinguals. This finding might simply indicate that bidialectals tend to produce “easier” words in the less proficient lan- guage [11]. Observing a mixing benefit with English-Dundonian bidialectals in a voluntary language switching paradigm would be a more unique finding, as this effect has only reliably been observed with bilinguals in a voluntary language switching paradigm. A mixing benefit would indicate that bidialectals implement proactive language control in single language blocks, whereas this is less the case in mixed language blocks [8]. Method Participants A power analysis on the mixing benefit was run to determine the required number of partici- pants and trials. We chose the mixing benefit, since this effect seems more uniquely connected to voluntary language switching. Along the approach suggested by Brysbaert and Stevens [46], we ran 200 (Monte Carlo) simulations with the simr package [47] on the voluntary language switching data of Jevtović et al. [14], who tested 40 bilinguals with 20 distinct stimuli in 80 tri- als in single language blocks and 180 trials in voluntary language switching blocks for each par- ticipant. The results showed that the setup of Jevtović et al. [14] had a 99.5% chance of showing a mixing benefit. In the current study, we relied on the same number of stimuli as Jev- tović and colleagues. Furthermore, we relied on a similar number of participants and trials. The number of voluntary language switching trials per participant was slightly decreased rela- tive to Jevtović et al. (from 180 to 160 trials per participant). That way, we had a similar num- ber of repetition trials in the voluntary language switching blocks and overall trials in the single language blocks. To make sure that we had at least the same number of voluntary lan- guage switching block trials across participants as Jevtović and colleagues, we increased the number of participants from 40 to 46. In line with this target, we collected useable data from 46 bidialectal speakers of (Scottish) Standard English and Dundonian Scots (16 identified as men, 29 as women, and 1 as non- binary), who were recruited through a local media campaign and word of mouth. An addi- tional two participants completed the study but did not produce useable data–one had an uploading error resulting in no audio recordings being stored and another was excluded at the accuracy/variety coding stage due to eating throughout the task, which interfered with reaction time extraction. Using a similar questionnaire for the bidialectals as in Kirk et al. [1] and Declerck et al. [48], these speakers reported using Dundonian Scots around 26% of the time. Following the experi- ment, the participants also completed an English vocabulary test on the basis of a lexical deci- sion task (i.e., LexTale [49] and a 10-item Dundonian Scots test where they had to select the correct definition of a word from four options). This information and a summary of demo- graphic information is shown in Table 1. The study received approval from Abertay Univer- sity’s research ethics committee (EMS4259). Materials and task Along the lines of Jevtović et al. [14], 20 pictures were presented to the bidialectal participants. These pictures corresponded to non-cognate names between Standard English (average num- ber of syllables across the English names: 1.55; average Zipf frequency of the English names: PLOS ONE \| https://doi.org/10.1371/journal.pone.0282086 May 4, 2023 4 / 14 PLOS ONETable 1. Overview of participants’ demographic information (SD in brackets). Voluntary language switching in bidialectals N Age (years) Dundonian use (Current %) Dundonian use (Childhood %) Self-rated Dundonian comprehension* Self-rated English comprehension* Self-rated Dundonian production* Self-rated English production* Dundonian Word Test (%) English LexTale (%) Dundonian Switch Rate (%) Dundonian Repetition Rate (%) English Switch Rate (%) English Repetition Rate (%) 46 38.3 (10.7) 25.9 (13.6) 27.4 (20.4) 6.4 (0.8) 7 (0) 5.7 (1.3) 6.9 (0.2) 87.8 (12.3) 91.4 (10.7) 22.9 (2.9) 33.4 (11.9) 22.4 (3.3) 21.2 (12.3) Note.* Self-rated scores are on a scale of 1 (low proficiency) to 7 (high proficiency). ** Switch and Repetition rates are based on percentage of total number of valid coded trials in voluntary language switching blocks, before reaction time clean up. The variety name refers to whether the trial was named in Dundonian or English. Switch refers to having used the other variety in the previous trial and Repetition refers to having remained within that variety from the previous trial. https://doi.org/10.1371/journal.pone.0282086.t001 4.24; [50]) and Dundonian (average number of syllables across the Dundonian names: 1.60; Zipf frequency was not calculated for Dundonian names as no database with word frequency exists for the Dundonian dialect. See S1 Table for the stimulus list). Each picture was presented twice, in non-consecutive trials, throughout each block. Procedure The picture naming study was presented online on the Gorilla platform [51] and is publicly available as Open Materials (https://app.gorilla.sc/openmaterials/341909). After providing informed consent, participants performed a microphone check, in which they named a sen- tence and then listened to their own recording. A familiarization block followed the micro- phone check, in which all 20 pictures were presented together with the corresponding names in Standard English and Dundonian. Because there is no standardized written form of the Dundonian dialect, participants had the option to listen to a recording of the Dundonian word spoken by a local speaker. The familiarization phase was followed by the actual experiment, which consisted of two single language blocks of 40 trials each and four voluntary language switching blocks of 40 trials each. There are several ways to present the order of single language blocks and mixed language blocks to obtain mixing costs or a mixing benefit [28]. We opted for a setup in which participants first saw one single language block, followed by the four vol- untary language switching blocks, and then again one single language block in the other lan- guage variety than the first single language block. The language variety of the single language blocks was counterbalanced across participants. Prior to each of the three block types (i.e., English language block, Dundonian language block, and the voluntary language switching blocks), instructions were displayed pertinent for that block type, emphasizing speed and accuracy to name each picture. Moreover, in the single language blocks, the bidialectals were instructed to name each picture in the corresponding language throughout the block. Prior to the voluntary language switching blocks the following PLOS ONE \| https://doi.org/10.1371/journal.pone.0282086 May 4, 2023 5 / 14 PLOS ONEVoluntary language switching in bidialectals sentences were presented (for similar instructions, see [8, 14]): “In the following section, you can name the pictures in either Standard English or Dundonian. You are free to choose which language variety to use for each picture. However, please do not use the same language variety throughout the whole task.”. The instructions were followed by a short demonstration of the task before completing a short practice block consisting of 8 trials. Finally, the participants per- formed the experimental block(s). Each trial started with a fixation cross in the middle of the screen. After 250 ms, the fixation cross was replaced by the stimulus for a maximum of 3000 ms. Each trial ended with a brief 50 ms blank screen before the onset of the next trial. Following the main task, participants completed a Language Background Questionnaire, the English LexTale and a 10-item test of Dundonian Scots words (also available from: https:// app.gorilla.sc/openmaterials/341909). Data analyses The raw data and analyses scripts are available on the Open Science Framework (https://osf.io/ wmr3d/). The first trial in each voluntary language switching block was excluded from the reaction time (RT) analyses, since this is neither a switch nor repetition trial. Furthermore, error trials and trials immediately following an error were also excluded from the RT analyses. Trials with RTs faster than 150 ms, slower than 3000 ms, or three standard deviations above participant mean were also removed. Similar to previous bilingual voluntary language switching studies (e.g., [9, 10]), we pro- vided an overview of the mean switch rate of the participants in all conditions (see Table 1). Furthermore, two analyses were conducted on the RT data. The Switching analysis was con- ducted on the data of the voluntary language switching blocks and consisted of the factors Trial type (switch vs. repetition trials) and Language variety (Standard English vs. Dundonian). The Mixing analysis was conducted on the “pure” trials from the single language blocks and the repetition trials in the voluntary language switching blocks. The latter analysis consisted of Block type (repetition trials from voluntary language switching blocks vs. trials from single lan- guage block) and Language variety (Standard English vs. Dundonian). The RTs were analyzed using linear mixed-effects regression modeling [52]. Both partici- pants and items were considered random factors with all fixed effects and their interactions varying by all random factors [53]. No, convergence issues were registered, hence we did not need to rely on the buildmer package [54] to simplify the model. For all two-level factors we used effect coding (i.e., -0.5 and 0.5). Finally, t- and z-values larger or equal to 1.96 were deemed significant [55]. The error data were not analyzed because we did not observe enough errors for meaningful analysis (< 5%). However, we did report descriptive statistics below. Results Error rates Trials were coded as errors if the wrong word was produced, if no utterance was produced, or if the utterances contained extraneous noises (e.g., “umm”). Additionally, in the Single lan- guage blocks, trials produced in the wrong variety were coded as errors (this was not relevant in the Voluntary language switching blocks as the participants were free to produce in either variety from trial to trial). Overall, error rates reached 3.5% of valid trials, thus we did not conduct any error analyses in line with our threshold of 5% errors (or more). For a breakdown of the error rates, see Table 2. PLOS ONE \| https://doi.org/10.1371/journal.pone.0282086 May 4, 2023 6 / 14 PLOS ONEVoluntary language switching in bidialectals Table 2. Breakdown of percentage errors in each block type. Single Language Block 4.97% Voluntary Language Block 2.70% Standard 4.33% Dialect 5.62% https://doi.org/10.1371/journal.pone.0282086.t002 Reaction times Switching analysis. The model yielded a main effect of Trial Type, with slower responses in switch (990.3 ms, SD = 289.2 ms) than in repetition trials (962.6 ms, SD = 268.9 ms; see Table 3 and Fig 1), indicating that voluntary language switching with bidialectals still produces switch costs. We also found a main effect of Variety with Standard English items (982.3 ms, 286.7 ms) being named slower on average than Dundonian Scots dialect items (969.5 ms, SD = 272.0 ms). There was no interaction between Trial Type and Variety, indicating a sym- metrical switch cost pattern across language varieties. Mixing analysis. The model yielded no main effect of Block Type, indicating similar reac- tion times for the repetition trials from the voluntary language switching blocks and the trials from the single language blocks, thus providing no indication of either a mixing benefit or cost. There was a main effect of Variety with Standard English items (989.7 ms, SD = 283.5 ms) being named slower than Dundonian Scots items (959.8 ms, SD = 267.4 ms; see Table 3 and Fig 2). There was no significant interaction between Block Type and Variety. Discussion In the current study, we set out to examine whether a similar language control process is implemented by bilinguals and bidialectals by relying on voluntary language switching. Prior bilingual studies have shown that voluntary language switching is easier (at least in one lan- guage) than producing in a specific language (i.e., mixing benefit; e.g., [8, 9, 11–14]), an effect linked to proactive language control [8]. Additionally, bilingual voluntary language switching tends to result in symmetrical switch costs across the two languages [8–14]. We reasoned that similar patterns should be observed with bidialectals if bilinguals and bidialectals rely on a sim- ilar language control process. On the one hand, the results showed no mixing benefit with bidialectals. On the other hand, symmetrical switch costs were found during bidialectal volun- tary language switching, similar to most voluntary language switching studies with bilinguals. Table 3. Parameter estimates and results of significance tests in mixed-effects models. Fixed effects Switching Model: Reaction Times ~ Variety * TrialType + (Variety * TrialType\| Participant) + (Variety * TrialType\| Picture) SE β p t Intercept Variety (Standard vs Dialect) 991.47 -42.02 29.30 15.80 33.84 -2.66 < .001 .013 Trial Type (Switch vs Repeat) Trial Type * Variety Mixing Model: Reaction Times ~ Variety * BlockType + (Variety * BlockType \| Participant) + (Variety* BlockType \| Picture) -24.64 12.93 -3.13 -0.09 -1.16 7.88 .929 .003 Intercept Variety (Standard vs Dialect) Block Type (Single vs Voluntary) Block Type * Variety 979.90 -54.69 -3.14 20.18 27.43 18.07 13.05 19.88 35.72 -3.03 -0.24 1.02 < .001 .005 .811 .316 https://doi.org/10.1371/journal.pone.0282086.t003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0282086 May 4, 2023 7 / 14 PLOS ONEVoluntary language switching in bidialectals Fig 1. Violin plot showing the distribution of reaction times across switch and repetition trials in the voluntary language switching block, indicative of switching costs for Standard English (Standard) and Dundonian Scots (Dialect). The boxplot shows the interquartile range, the horizontal line represents the median, and the dot indicates the mean for each condition. https://doi.org/10.1371/journal.pone.0282086.g001 Bidialectal vs. bilingual voluntary mixing effect Regarding the unique mixing benefit patten (i.e., worse performance in single language blocks than in mixed language blocks, at least for one language) observed in bilingual studies that relied on voluntary language switching [8, 9, 11–14], no such pattern was observed in the cur- rent bidialectal study. More specifically, we observed no significant mixing effect. Since the mixing benefit with a bilingual voluntary language switching task is typically explained in terms of proactive language control [8], the lack of such a mixing benefit with bidialectals could be interpreted as there being no proactive language control, or at least no difference in proactive language control between single- and mixed-language contexts. It is surprising that we observed no voluntary mixing benefit with bidialectals because prior bidialectal studies that investigated language control [1, 2, 5] typically showed quite some over- lap with the findings of similar bilingual studies. Yet, it should be noted that several studies have shown that qualitative differences in language control can be observed across bilingual groups (e.g., [56, 57]). In a recent study by Declerck et al. [56], for instance, a different ERP pattern was observed when bimodal bilinguals (i.e., bilinguals proficient in a signed and PLOS ONE \| https://doi.org/10.1371/journal.pone.0282086 May 4, 2023 8 / 14 PLOS ONEVoluntary language switching in bidialectals Fig 2. Violin plot showing the distribution of reaction times across trials in the single language blocks and repetition trials in the voluntary language switching blocks, indicative of a mixing effect for Standard English (Standard) and Dundonian Scots (Dialect). The boxplot shows the interquartile range, the horizontal line represents the median, and the dot indicates the mean for each condition. https://doi.org/10.1371/journal.pone.0282086.g002 spoken language) switched languages involuntarily relative to what is typically observed with unimodal bilinguals (i.e., bilinguals proficient in two spoken languages), thus showing a differ- ence in language control between these two groups of bilinguals. Similarly, the results of the current study could be interpreted as bidialectals and bilinguals not relying on exactly the same language control processes. However, any language mixing difference observed here might also be due to differences between the specific languages used in prior bilingual PLOS ONE \| https://doi.org/10.1371/journal.pone.0282086 May 4, 2023 9 / 14 PLOS ONEVoluntary language switching in bidialectals voluntary language switching studies (Catalan-Spanish, Spanish-Basque, and Spanish-English) and the language varieties used in the current study. So, more research based on different lan- guages and language varieties should result in more conclusive evidence, especially since the few bilingual studies that investigated the voluntary mixing benefit relied on a small number of language combinations. An alternative explanation is that the absence of a significant mixing benefit with bidialec- tals during voluntary language switching is due to the same underlying mechanism that results in a mixing benefit with bilinguals that perform a voluntary language switching task [8, 9, 11– 14]. As we have indicated in the introduction, both bidialectals [5] and bilinguals [20, 25, 26, 29–31] show a mixing cost when involuntarily switching between languages. With bilinguals this turns into a mixing benefit, for at least one language, when switching languages voluntar- ily. This reversal of the mixing effect could be due to little to no language control being neces- sary in voluntary mixed language blocks, whereas some control processes are necessary in involuntary mixed language blocks. Based on the current study, no such mixing benefit pattern is found with bidialectals. However, relying on voluntary language switching does change the typical mixing cost effect observed with involuntary language switching into a non-significant effect. Therefore, not only does voluntary language switching impact the mixing effect of bilin- guals, but also bidialectals, when compared to involuntary language switching. It might be that the underlying mechanism that turns a mixing cost into a mixing benefit due to bilingual vol- untary language switching could also change the mixing cost pattern observed during bidia- lectal involuntary language switching to the point that this effect is not significant anymore (instead of becoming a mixing benefit with bilinguals) when relying on voluntary language switching. This would entail that there is no qualitative difference in (proactive) language con- trol between bidialectals and bilinguals, but merely a quantitative difference. (A)symmetrical switch costs Along the lines of previous involuntary language switching studies with bidialectals [1, 2, 5, 6], the present study showed that a switch cost pattern can be observed when bidialectals voluntarily switch between their dialect and the standard language. Similar to bilingual voluntary language switching studies [8–14], the size of the switch costs was similar for the dialect and the standard language. This could be taken as evidence that at least some language control processes imple- mented by bidialectals and bilinguals in a voluntary language switching context are similar. Another interpretation would be that bilinguals and bidialectals have a tendency to produce easier words in their less proficient language during voluntary language switching [11]. This should lead to more similar activation levels across the two languages, at least for the used items, and thus could result in similar switch costs. Hence, it could be that that this pattern simply indicates that bilinguals and bidialectals tend to rely on the same strategy during volun- tary language switching. Furthermore, it should be taken into account that prior studies with a similar group of bidia- lectals performing involuntary language switching tasks also resulted in symmetrical switch costs [1, 48]. This is not to say that bidialectals never show asymmetrical switch costs when involuntarily switching languages (see [5]), but members of this particular group of bidialectals (i.e., Dundonian-English bidialectals) might generally be similar to highly proficient bilinguals (e.g., [21, 34]), which should have a smaller difference between the base activation of their two language varieties, and thus show similar switch costs across language varieties. Along the same lines, it is still unclear what the boundary conditions are to observe asym- metrical switch costs (for reviews, see [4, 27, 58]). So, finding symmetrical switch costs might also be due to characteristics of this study other than voluntary language switching. PLOS ONE \| https://doi.org/10.1371/journal.pone.0282086 May 4, 2023 10 / 14 PLOS ONEVoluntary language switching in bidialectals Another reason why not too much weight should be put upon finding a similar symmetrical switch cost pattern during voluntary language switching with bilinguals and bidialectals is that two very recent bilingual voluntary language switching studies observed asymmetrical switch costs [59, 60]. Hence, it would seem as if the symmetrical switch cost pattern in bilingual vol- untary language switching is less stable than we originally thought at the start of this project. This makes it more difficult to make any comparisons between bilingual and bidialectal volun- tary language switching results based on this effect. Conclusion In sum, a different mixing pattern was observed in this Registered Report (see Declerck & Kirk [61] for the stage 1 Registered Report of the current study) with bidialectals when relying on voluntary language switching (i.e., no significant mixing effect) than what is generally observed with bilinguals that perform a voluntary language switching task (i.e., a mixing benefit in at least one language). While this could be explained with the notion that these two groups rely on, at least partly, different language control processes, it might just be due to a quantitative difference. Regarding the symmetrical switch costs observed with bidialectals in a voluntary language switching task, which is similar to what has typically been found with bilinguals, it is prudent to be careful to interpret this pattern as evidence that bilinguals and bidialectals rely to some degree on the same language control processes, as other interpretations (cf. similar strategies, high proficiency in both language varieties, and/or specific experiment characteristics) are also valid. Supporting information S1 Table. Proposed standard English and Dundonian non-cognate stimuli. (PDF) Acknowledgments We wish to thank Bethany Lane for assistance with coding responses for accuracy and variety. Author Contributions Conceptualization: Mathieu Declerck, Neil W. Kirk. Formal analysis: Mathieu Declerck, Neil W. Kirk. Funding acquisition: Neil W. Kirk. Investigation: Mathieu Declerck, Neil W. Kirk. Methodology: Mathieu Declerck. Project administration: Neil W. Kirk. Visualization: Neil W. Kirk. Writing – original draft: Mathieu Declerck, Neil W. Kirk. Writing – review & editing: Mathieu Declerck, Neil W. Kirk. PLOS ONE \| https://doi.org/10.1371/journal.pone.0282086 May 4, 2023 11 / 14 PLOS ONEVoluntary language switching in bidialectals References 1. Kirk N. W., Kempe V., Scott-Brown K. C., Philipp A., & Declerck M. (2018). Can monolinguals be like bilinguals? Evidence from dialect switching. Cognition, 170, 164–178. https://doi.org/10.1016/j. cognition.2017.10.001 PMID: 29024916 2. Vorwerg C. C., Suntharam S., & Morand M. A. (2019). Language control and lexical access in diglossic speech production: Evidence from variety switching in speakers of Swiss German. Journal of Memory and Language, 107, 40–53. 3. Calabria M., Costa A., Green D. W., & Abutalebi J. (2018). Neural basis of bilingual language control. Annals of the New York Academy of Sciences, 1426, 221–235. https://doi.org/10.1111/nyas.13879 PMID: 29917244 4. Declerck M., & Philipp A. M. (2015). A review of control processes and their locus in language switching. Psychonomic Bulletin & Review, 22, 1630–1645. https://doi.org/10.3758/s13423-015-0836-1 PMID: 25917142 5. Kirk N. W., Declerck M., Kemp R., & Kempe V. (2022). Language control in dialect speakers–“monolin- guals” by name, bilinguals by nature? Bilingualism: Language and Cognition. 6. Scaltritti M., Peressotti F., & Miozzo M. (2017). Bilingual advantage and language switch: What’s the linkage?. Bilingualism: Language and Cognition, 20, 80–97. 7. Blanco-Elorrieta E., & Pylkka¨ nen L. (2017). Bilingual language switching in the laboratory versus in the wild: The spatiotemporal dynamics of adaptive language control. Journal of Neuroscience, 37, 9022– 9036. https://doi.org/10.1523/JNEUROSCI.0553-17.2017 PMID: 28821648 8. 9. de Bruin A., Samuel A. G., & Duñabeitia J. A. (2018). Voluntary language switching: When and why do bilinguals switch between their languages?. Journal of Memory and Language, 103, 28–43. de Bruin A., Samuel A. G., & Duñabeitia J. A. (2020). Examining bilingual language switching across the lifespan in cued and voluntary switching contexts. Journal of Experimental Psychology: Human Per- ception and Performance, 46, 759–788. https://doi.org/10.1037/xhp0000746 PMID: 32324035 10. Gollan T. H., Kleinman D., & Wierenga C. E. (2014). What’s easier: Doing what you want, or being told what to do? Cued versus voluntary language and task switching. Journal of Experimental Psychology: General, 143, 2167–2195. https://doi.org/10.1037/a0038006 PMID: 25313951 11. Gollan T. H., & Ferreira V. S. (2009). Should I stay or should I switch? A cost–benefit analysis of volun- tary language switching in young and aging bilinguals. Journal of Experimental Psychology: Learning, Memory, and Cognition, 35, 640–665. https://doi.org/10.1037/a0014981 PMID: 19379041 12. Gross M., & Kaushanskaya M. (2015). Voluntary language switching in English–Spanish bilingual chil- dren. Journal of Cognitive Psychology, 27, 992–1013. https://doi.org/10.1080/20445911.2015. 1074242 PMID: 26889376 13. Grunden N., Piazza G., Garcı´a-Sa´ nchez C., & Calabria M. (2020). Voluntary language switching in the context of bilingual aphasia. Behavioral Sciences, 10, 141. https://doi.org/10.3390/bs10090141 PMID: 32962153 14. Jevtović M., Duñabeitia J. A., & de Bruin A. (2020). How do bilinguals switch between languages in dif- ferent interactional contexts? A comparison between voluntary and mandatory language switching. Bilingualism: Language and Cognition, 23, 401–413. 15. Kleinman D., & Gollan T. H. (2016). Speaking two languages for the price of one: Bypassing language control mechanisms via accessibility-driven switches. Psychological Science, 27, 700–714. https://doi. org/10.1177/0956797616634633 PMID: 27016240 16. Liu H., Tong J., de Bruin A., Li W., He Y., & Li B. (2020). Is inhibition involved in voluntary language switching? Evidence from transcranial direct current stimulation over the right dorsolateral prefrontal cortex. International Journal of Psychophysiology, 147, 184–192. https://doi.org/10.1016/j.ijpsycho. 2019.12.002 PMID: 31830498 17. Meuter R. F., & Allport A. (1999). Bilingual language switching in naming: Asymmetrical costs of lan- guage selection. Journal of Memory and Language, 40, 25–40. 18. Jackson G. M., Swainson R., Cunnington R., & Jackson S. R. (2001). ERP Correlates of executive con- trol during repeated language-switching. Bilingualism: Language and Cognition, 4, 169–178. 19. Macizo P., Bajo T., & Paolieri D. (2012). Language switching and language competition. Second Lan- guage Research, 28, 131–149. 20. Christoffels I. K., Firk C., & Schiller N. O. (2007). Bilingual language control: An event-related brain potential study. Brain Research, 1147, 192–208. https://doi.org/10.1016/j.brainres.2007.01.137 PMID: 17391649 PLOS ONE \| https://doi.org/10.1371/journal.pone.0282086 May 4, 2023 12 / 14 PLOS ONEVoluntary language switching in bidialectals 21. Costa A., & Santesteban M. (2004). Lexical access in bilingual speech production: Evidence from lan- guage switching in highly proficient bilinguals and L2 learners. Journal of Memory and Language, 50, 491–511. 22. Declerck M., Koch I., & Philipp A. M. (2012). Digits vs. pictures: The influence of stimulus type on lan- guage switching. Bilingualism: Language and Cognition, 15, 896–904. 23. Declerck M., Koch I., & Philipp A. M. (2015). The minimum requirements of language control: Evidence from sequential predictability effects in language switching. Journal of Experimental Psychology: Learn- ing, Memory, and Cognition, 41, 377–394. https://doi.org/10.1037/xlm0000021 PMID: 24999708 24. Green D. W. (1998). Mental control of the bilingual lexico-semantic system. Bilingualism: Language and Cognition, 1, 213–229. 25. Declerck M., Philipp A. M., & Koch I. (2013). Bilingual control: Sequential memory in language switch- ing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 39, 1793. https://doi.org/ 10.1037/a0033094 PMID: 23773181 26. Ma F., Li S., & Guo T. (2016). Reactive and proactive control in bilingual word production: An investiga- tion of influential factors. Journal of Memory and Language, 86, 35–59. 27. Bobb S. C., & Wodniecka Z. (2013). Language switching in picture naming: What asymmetric switch costs (do not) tell us about inhibition in bilingual speech planning. Journal of Cognitive Psycholology, 25, 568–585. 28. Declerck M. (2020). What about proactive language control?. Psychonomic Bulletin & Review, 27, 24– 35. https://doi.org/10.3758/s13423-019-01654-1 PMID: 31410740 29. Jylkka¨ J., Lehtonen M., Lindholm F., Kuusakoski A., & Laine M. (2018). The relationship between gen- eral executive functions and bilingual switching and monitoring in language production. Bilingualism: Language and Cognition, 21, 505–522. 30. Peeters D., & Dijkstra T. (2018). Sustained inhibition of the native language in bilingual language pro- duction: A virtual reality approach. Bilingualism: Language and Cognition, 21, 1035–1061. 31. Stasenko A., Matt G. E., & Gollan T. H. (2017). A relative bilingual advantage in switching with prepara- tion: Nuanced explorations of the proposed association between bilingualism and task switching. Jour- nal of Experimental Psychology: General, 146, 1527–1550. https://doi.org/10.1037/xge0000340 PMID: 28714710 32. Degani T., Kreiner H., Ataria H., & Khateeb F. (2020). The impact of brief exposure to the second lan- guage on native language production: Global or item specific? Applied Psycholinguistics, 41, 153–183. 33. Wodniecka Z., Szewczyk J., Kałamała P., Mandera P., & Durlik J. (2020). When a second language hits a native language. What ERPs (do and do not) tell us about language retrieval difficulty in bilingual lan- guage production. Neuropsychologia, 141, 107390. https://doi.org/10.1016/j.neuropsychologia.2020. 107390 PMID: 32057934 34. Costa A., Santesteban M., & Ivanova I. (2006). How do highly proficient bilinguals control their lexicali- zation process? Inhibitory and language-specific selection mechanisms are both functional. Journal of Experimental Psychology: Learning, Memory, and Cognition, 32, 1057–1074. https://doi.org/10.1037/ 0278-7393.32.5.1057 PMID: 16938046 35. Declerck M., Ivanova I., Grainger J., & Duñabeitia J. A. (2020). Are similar control processes imple- mented during single and dual language production? Evidence from switching between speech regis- ters and languages. Bilingualism: Language and Cognition, 23, 694–701. 36. Kaufmann E., Mittelberg I., Koch I., & Philipp A. M. (2018). Modality effects in language switching: Evi- dence for a bimodal advantage. Bilingualism: Language and Cognition, 21, 243–250. 37. Blanco-Elorrieta E., & Pylkka¨ nen L. (2016). Bilingual language control in perception versus action: MEG reveals comprehension control mechanisms in anterior cingulate cortex and domain-general control of production in dorsolateral prefrontal cortex. Journal of Neuroscience, 36, 290–301. https://doi.org/10. 1523/JNEUROSCI.2597-15.2016 PMID: 26758823 38. Melinger A. (2018). Distinguishing languages from dialects: a litmus test using the picture-word interfer- ence task. Cognition, 172, 73–88. https://doi.org/10.1016/j.cognition.2017.12.006 PMID: 29232596 39. Melinger A. (2021). Do elevators compete with lifts?: Selecting dialect alternatives. Cognition, 206, 104471. https://doi.org/10.1016/j.cognition.2020.104471 PMID: 33075569 40. Dylman A. S., & Barry C. (2018). When having two names facilitates lexical selection: Similar results in the picture-word task from translation distractors in bilinguals and synonym distractors in monolinguals. Cognition, 171, 151–171. https://doi.org/10.1016/j.cognition.2017.09.014 PMID: 29172105 41. Bialystok E., Craik F. I., Klein R., & Viswanathan M. (2004). Bilingualism, aging, and cognitive control: evidence from the Simon task. Psychology and Aging, 19, 290. https://doi.org/10.1037/0882-7974.19. 2.290 PMID: 15222822 PLOS ONE \| https://doi.org/10.1371/journal.pone.0282086 May 4, 2023 13 / 14 PLOS ONEVoluntary language switching in bidialectals 42. Kempe V., Kirk N. W., & Brooks P. J. (2015). Revisiting theoretical and causal explanations for the bilin- gual advantage in executive functioning. Cortex, 73, 342–344. https://doi.org/10.1016/j.cortex.2015.07. 021 PMID: 26298266 43. McDermott P. (2019). From ridicule to legitimacy? ‘Contested languages’ and devolved language plan- ning. Current Issues in Language Planning, 20, 121–139. 44. Scottish Government (2010). Public attitudes towards the Scots language. Available from: http://www. gov.scot/Publications/2010/01/06105123/0 45. Jauriberry T. (2021). Variation and change of middle-class/r/in Standard Scottish English. Lingua, 103059. 46. Brysbaert M., & Stevens M. (2018). Power analysis and effect size in mixed effects models: a tutorial. Journal of Cognition, 1, 9. https://doi.org/10.5334/joc.10 PMID: 31517183 47. Green P., & MacLeod C. J. (2016). SIMR: an R package for power analysis of generalized linear mixed models by simulation. Methods in Ecology and Evolution, 7, 493–498. 48. Declerck M., O¨ zbakar E., & Kirk N. W. (2021). Is there proactive inhibitory control during bilingual and bidialectal language production?. Plos one, 16, e0257355. https://doi.org/10.1371/journal.pone. 0257355 PMID: 34520485 49. 50. Lemho¨fer K., & Broersma M. (2012). Introducing LexTALE: A quick and valid lexical test for advanced learners of English. Behavior Research Methods, 44, 325–343. https://doi.org/10.3758/s13428-011- 0146-0 PMID: 21898159 van Heuven W. J., Mandera P., Keuleers E., & Brysbaert M. (2014). SUBTLEX-UK: A new and improved word frequency database for British English. Quarterly Journal of Experimental Psychology, 67, 1176–1190. https://doi.org/10.1080/17470218.2013.850521 PMID: 24417251 51. Anwyl-Irvine A. L., Massonnie´ J., Flitton A., Kirkham N., & Evershed J. K. (2020). Gorilla in our midst: An online behavioral experiment builder. Behavior Research Methods, 52, 388–407. https://doi.org/10. 3758/s13428-019-01237-x PMID: 31016684 52. Baayen R. H., Davidson D. J., & Bates D. M. (2008). Mixed-effects modeling with crossed random effects for subjects and items. Journal of Memory and Language, 59, 390–412. 53. Barr D. J., Levy R., Scheepers C., & Tily H. J. (2013). Random effects structure for confirmatory hypoth- esis testing: Keep it maximal. Journal of Memory and Language, 68, 255–278. https://doi.org/10.1016/ j.jml.2012.11.001 PMID: 24403724 54. Voeten C. C. (2021). Buildmer: Stepwise elimination and term reordering for mixed- effects regression. https://CRAN.R-project.org/package=buildmer. 55. Baayen R. (2008). Analyzing Linguistic Data: A practical introduction to statistics. Cambridge University Press. 56. Declerck M., Meade G., Midgley K. J., Holcomb P. J., Roelofs A., & Emmorey K. (2021). Language con- trol in bimodal bilinguals: Evidence from ERPs. Neuropsychologia, 161, 108019. https://doi.org/10. 1016/j.neuropsychologia.2021.108019 PMID: 34487737 57. Kaufmann E., Mittelberg I., Koch I., & Philipp A. M. (2018). Modality effects in language switching: Evi- dence for a bimodal advantage. Bilingualism: Language and Cognition, 21, 243–250. 58. Declerck M., & Koch I. (2022). The concept of inhibition in bilingual control. Psychological Review. https://doi.org/10.1037/rev0000367 PMID: 35420846 59. Liu H., Li W., de Bruin A., & He Y. (2021). Should I focus on self-language actions or should I follow oth- ers? Cross-language interference effects in voluntary and cued language switching. Acta Psychologica, 216, 103308. https://doi.org/10.1016/j.actpsy.2021.103308 PMID: 33892263 60. Sa´ nchez L. M., Struys E., & Declerck M. (2022). Ecological validity and bilingual language control: vol- untary language switching between sentences. Language, Cognition and Neuroscience. 61. Declerck M., & Kirk N. W. (2021). Is it easier to use one language variety at a time, or mix them? An investigation of voluntary language switching with bidialectals. PloS one, 16, e0256554. https://doi.org/ 10.1371/journal.pone.0256554 PMID: 34495987 PLOS ONE \| https://doi.org/10.1371/journal.pone.0282086 May 4, 2023 14 / 14 PLOS ONE
10.1371_journal.pone.0286962	RESEARCH ARTICLE Impact of end-stage knee osteoarthritis on perceived physical function and quality of life: A descriptive study from Jordan Sumayeh AbujaberID 1, Ibrahim AltubasiID 1, Mohammad Hamdan2, Raed Al-Zaben3 1 Department of Physiotherapy, School of Rehabilitation Sciences, The University of Jordan, Amman, Jordan, 2 Department of Special Surgery, Division of Orthopaedics, School of Medicine, The University of Jordan, Amman, Jordan, 3 Department of Orthopaedic Surgery, Royal Medical Services, Amman, Jordan a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 s.abujaber@ju.edu.jo Abstract Objective OPEN ACCESS Citation: Abujaber S, Altubasi I, Hamdan M, Al- Zaben R (2023) Impact of end-stage knee osteoarthritis on perceived physical function and quality of life: A descriptive study from Jordan. PLoS ONE 18(6): e0286962. https://doi.org/ 10.1371/journal.pone.0286962 Editor: Aqeel M. Alenazi, Prince Sattam bin Abdulaziz University, SAUDI ARABIA Received: December 31, 2022 Accepted: May 21, 2023 Published: June 9, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0286962 Copyright: © 2023 Abujaber et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Of the present study was to evaluate the impact of end-stage knee OA on patient’s percep- tion of their functional abilities and quality of life (QoL) using the self-reported questionnaire; the Knee Injury and Osteoarthritis Outcome Score (KOOS), and to determine the contribu- tion of knee pain on patient’s perceived outcomes. Methods Patients with end-stage knee OA who are on the waiting list for total knee arthroplasty were recruited in this cross-sectional study. Patients were asked to fill out the KOOS question- naire. Knee pain for both sides was quantified on a continuous scale from 0–10. Age, and anthropometric data were recorded. Descriptive statistics were calculated for patients’ char- acteristics, and for the scores of each KOOS subscale. Hierarchical linear regression mod- els were created to determine the contributions of knee pain on two KOOS subscales; the function in daily living (KOOS-ADL), and the knee-related quality of life (KOOS-QoL). Results Patients in this study scored low across KOOS subscales (27.7% - 54.2%) with the QoL sub- scale being the lowest. After accounting for age and BMI, hierarchical linear regressions revealed that knee pain in both sides were determinants of self-perceived KOOS-ADLs, while only knee pain in the most-affected side significantly contributed to lower KOOS-QOL scores. Conclusion End-stage knee OA negatively impact the patients’ perceived function and quality of life. Patients’ KOOS scores were similar to those reported in other countries, with QoL being the domain most affected. Our findings demonstrate that the level of knee pain has a determi- nant effect on our patients’ perceptions of functional abilities and QoL. As waiting-list PLOS ONE \| https://doi.org/10.1371/journal.pone.0286962 June 9, 2023 1 / 12 PLOS ONEKnee pain and perceived function in patients with end-stage knee osteoarthritis patients, addressing knee pain with a targeted regimen prior to TKA, as well as increasing patient’s awareness about knee pain management, may improve/ or minimize deterioration in perceived functional ability and QoL while awaiting TKA. Funding: This study was financially supported by the Deanship of Scientific Research at the University of Jordan. This fund was received by SA. https://research.ju.edu.jo/Home.aspx The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Introduction Osteoarthritis (OA) is a multifactorial, progressive disease that involves degeneration of the hyaline cartilage and surrounding tissue within the joint. OA is the most common joint disor- der in adults [1], with the knee being the most affected joint [2]. Knee OA is a highly prevalent musculoskeletal condition, affecting approximately 22.9% of individuals aged 40 and over [3]. Along with hip OA, knee OA is considered the 11th highest contributor to global disability in the elderly [4]. Functional limitations, persistent pain, and poor quality of life are conse- quences of Knee OA [5, 6]. When knee OA progresses to severe end-stage, and conservative treatment fails to reduce pain and improve the patient’s functions, then total knee arthroplasty (TKA) becomes the treatment of choice. In the USA, this surgery was performed for 4.55% of the population �50 years of age [7], and the estimated annual incidence of primary TKA will be 3.5 million by 2030 [8]. However, patients with end-stage knee OA may wait for several months before they undergo the TKA procedure due to various reasons. Waiting time for TKA may further deteri- orate the patient’s condition in terms of pain, physical function, and health-related quality of life [9]. A self-reported questionnaire is a tool that measures the patient’s perception regarding their functional abilities and limitations. It allows the patients to express their satisfaction level, physical limitation or symptoms, and quality of life. Self-reported questionnaires are recom- mended to use as they are easy to administer, inexpensive, and have high internal consistency [10]. Although patients may over- or underestimate their abilities compared to their actual performed physical function, patient perception is certainly an essential part of the manage- ment process when describing the patient’s functional status and monitoring the outcomes progression. Many disease-specific self-reported questionnaires are available and commonly utilized in OA and TKA populations including Knee Injury and Osteoarthritis Outcome Score (KOOS) [11], Knee Outcome Survey (KOS) [12], and Western Ontario and McMaster Univer- sities Osteoarthritis Index (WOMAC) [13]. Although several studies have examined the functional abilities of patients with knee OA in different countries, the impact of knee OA may vary among different populations due to dif- ferent lifestyles, everyday activities, and sociocultural differences. Knee OA is a common joint disorder in Jordan, a developing country in the Middle East region. Although studies on its prevalence are lacking, it has been reported that Jordanians have a higher percentage of end- stage knee OA compared to the developed world [14]. Particularly, studies describing the per- ceived function in Jordanian patients with knee OA are scarce. Jordanians with knee OA showed limited knowledge of the disease pathology and management options that is partly attributed to inappropriate service delivery, lack of education, and cultural factors [15]. These factors could influence the experience of people with knee OA, potentially worsening their pain and joint deterioration. This could ultimately have a negative impact on their functional ability. Therefore, the purpose of this study is to evaluate the impact of knee OA on patients’ per- ception of functional abilities and quality of life in Jordanians with end-stage knee OA, who are on the waiting list for TKA. The second purpose was to determine how joint pain PLOS ONE \| https://doi.org/10.1371/journal.pone.0286962 June 9, 2023 2 / 12 PLOS ONEKnee pain and perceived function in patients with end-stage knee osteoarthritis influences the patient’s perceived functional ability and quality of life. The results from this study will help to characterize how patients perceived their quality of life and functional capac- ity considering their knee condition. It will also provide evidence regarding the contribution of joint pain to a patient’s perception, which might serve to design a targeted rehabilitation program that minimizes functional limitations during the waiting time for TKA. Methods Study design and participants This cross-sectional study is part of a longitudinal study that evaluates changes in physical function before and after TKA. Waiting lists for patients with end-stage knee OA scheduled for TKA were obtained from local hospitals. Patients with end-stage knee OA between 50 and 85 years, who are scheduled for unilateral TKA were included. Patients were screened for eligi- bility using a telephone interview conducted by a research assistant. Exclusion criteria were if patients have other musculoskeletal, neurological, or cardiovascular pathologies that affect their ability to perform daily activities, and if they have a history of cancer in the lower extrem- ity. Moreover, patients who plan to have surgery on their “less affected” knee (staged bilateral TKA) or on other lower extremity joints within a year or had previous arthroplasty surgery less than 1 year were excluded from the study to avoid the potential confounding influence of other joint impairments. Eligible patients were asked to fill out a self-reported questionnaire; the Arabic version of KOOS to evaluate their perceived pain and function. Evaluation session was performed 1 day to several weeks before TKA and took place either at the Physiotherapy department/ School of Rehabilitation Sciences at the University of Jordan or the University of Jordan Hospital. For those who were unable to reach the sitting, a web-based questionnaire was sent to them to fill out online. The study was approved by the Institutional Review Board at the University of Jordan hospital, and patients signed an informed consent form before par- ticipation, while those who completed the questionnaire online gave their informed consent by agreeing to answer the questionnaire’s questions. Participants’ characteristics Age, weight, height, and sex were recorded, and body mass index (BMI) was calculated for each subject. Knee pain, for the affected and less affected sides, was assessed on a continuous scale from 0 to 10, patients were specifically asked to verbally “rate your average pain over the past week from 0 to 10, where 0 is no pain and 10 is the worst imaginable pain”. Evaluation of symptoms, physical function, and quality of life The Arabic version of the KOOS was used to evaluate pain, symptoms, physical function, and quality of life [16]. The KOOS is a self-reported tool used to evaluate symptoms and function in patients with knee-related pathologies [11, 17]. The KOOS is a joint- and limb-specific ques- tionnaire, patients in this study provide answers regarding their “most affected” knee only. KOOS includes 42 items across 5 subscales: 1) pain frequency and severity during functional activities (KOOS-Pain; 9 items), 2): other symptoms such as the severity of knee stiffness and the presence of swelling (KOOS-Symptoms; 7 items), 3) difficulties when performing activi- ties of daily living (KOOS-ADL, 17 items), 4) difficulty experienced during sports and recrea- tional activities (KOOS-Sport; 5 items), and 5) knee-related quality of life (KOOS-QOL; 4 items). Each subscale is scored separately; the answer for each question is rated using a 5-point Likert scale thus scoring from 0 (no problem) to 4 (extreme problem). PLOS ONE \| https://doi.org/10.1371/journal.pone.0286962 June 9, 2023 3 / 12 PLOS ONEKnee pain and perceived function in patients with end-stage knee osteoarthritis A total score for each subscale is calculated and normalized to 100; with 100 indicating no symptoms or difficulties, and 0 indicating severe symptoms or difficulties. This normalization method is used in accordance with the KOOS scoring guideline 2012 (available at www.koos. nu), which calculates the total score as shown in Eq 1. 100 (cid:0) ðaverage of items=4Þ∗100 ð1Þ Patients were asked to rate the above subscales based on the previous week. The KOOS has been shown to have excellent reliability and good content and construct validity when used for follow-up of a knee injury, including OA [10]. Alfadhel et al. (2018) translated and cross-cul- turally adapted the original KOOS questionnaire into Arabic in Saudi Arabic patients with knee OA. The psychometric properties of the KOOS’s Arabic version were evaluated in that study, and the results showed that the Arabic version of the KOOS is well accepted and has excellent reliability, internal consistency, and construct validity [16]. The KOOS has similar items as the WOMAC physical function subscale with the addition of questions about sport and recreation and knee-related Quality of life. Data analysis Descriptive statistics (mean and SD) were calculated for patients’ characteristics (age, height, weight, BMI, and knee pain), and each KOOS subscale. The median for each item’s score within each subscale was calculated to indicate the perceived difficulty/severity for that item (based on Likert score), and the percentage of patients who rated that item with a given Likert point was calculated too. To determine the influence of knee pain (independent variable), measured by KOOS pain- subscale, on the functional abilities measured by KOOS ADLs-subscale and on the quality of life measured by KOOS-QOL (dependent variables), hierarchical linear regression models were created. Separate regression models were created for each dependent variable. Subject characteristics (age and BMI) were potential confounders to the relationships between pain and functional outcomes. In these regression models, age and BMI were first entered into the models. Because pain in the less affected side may affect the dependent outcome, it was added in the second step in the hierarchical regression (this was assessed using the 0 to 10 continuous scale), followed by the addition of knee pain of the affected side reported in KOOS-Pain sub- scale in the third step. The change in R2 informs us whether the addition of the variable in each step provides significant additional predictive information when accounting for the vari- ance explained by the variables in the preceding steps. The significance of each model, as well as the significant change in R2 between each step, was recorded. Statistical analyses were con- ducted using the Statistical Package for the Social Sciences (IBM SPSS 23.0, Chicago, IL). Sig- nificance level was set at 0.05. Results This study sample included 52 participants. We were only able to recruit four male participants out of the 52 participants, who were drawn from a population with a high female-to-male ratio. The data from the four male participants was included in this analysis, as we did not originally plan to exclusively focus on female patients. Additionally, we conducted the analysis with and without the male participants data and found that results remained unchanged. The mean age of the sample was 67 years old (range 50–78). The anthropometric character- istics of the patients were summarized in Table 1. Patients expressed a higher level of pain on the “most affected” side (8/10) compared to that on the “less affected” side (4/10) as measured using the 0–10 continuous scale (Table 1). PLOS ONE \| https://doi.org/10.1371/journal.pone.0286962 June 9, 2023 4 / 12 PLOS ONEKnee pain and perceived function in patients with end-stage knee osteoarthritis Table 1. Anthropometric characteristics, pain and KOOS scores (N = 52). Height (cm) Weight (Kg) BMI (Kg/m2) Age (yrs.) Knee Pain “most affected” (0–10) Knee Pain “less affected” (0–10) KOOS-symptom (%) KOOS-pain (%) KOOS-ADL (%) KOOS-QOL (%) https://doi.org/10.1371/journal.pone.0286962.t001 Mean 156.05 82.15 33.67 66.62 7.96 4.15 54.20 45.75 47.05 27.76 SD 7.81 13.85 5.00 6.68 1.73 2.91 20.70 20.16 20.14 16.71 Range 125–172 41–129 23.4–53 50–78 3–10 0–10 10.71–92.86 2.78–84.38 4.69–91.67 0–87.5 Patients in our sample reported impairments and limitations on the KOOS. The KOOS subscales’ scores ranged between 27.7% - 54.2%, with the quality-of-life subscale having the lowest score. The KOOS-Sport subscale was not quantified as most patients did not answer its questions or indicated that it does not apply to them. Table 1 shows the mean and standard deviation of the KOOS subscales. The median for each item within a specific subscale is presented in Table 2. Regarding the KOOS-Symptom subscale, the items’ median ranged from 0–4. Item S5 “Can you bend your knee fully?” showed the highest median (4) with 61.6% of patients responding with “rarely” or “never able to do so”. The median on the other items ranged between 0–2. In the KOOS-Pain subscale, items’ median ranged between 1–4, P1 item “How often do you experience knee pain?” showed the highest median (4), with 94.2% of patients responding with “daily” or “always”. This was followed by the P6 item “Going up or down stairs” which showed a median of 3, with 59.6% of patients experiencing “severe” to “extreme pain” when perform- ing that task. Regarding the KOOS-ADL, the elements with the highest median were for: A8 “Going shop- ping” and A16 “Heavy domestic duties (moving heavy boxes, scrubbing floors, etc)” in which 50% and 42.3% of patients reported severe to extreme difficulty in doing these activities, and were not rated by 40.4% and 51.9% of patients, respectively. Those elements were followed by A1 “Descending stairs”, A2 “Ascending stairs”, and A4 “Standing” elements with a median of 3, in which 48.1%, 61.5%, and 59.6% of patients reported severe to extreme difficulty in these activities, in order. The median for the rest of the activities was between 1–2, in which 15.4– 48.1% of patients reported severe to extreme difficulty. In the KOOS-QoL subscale, the items’ median ranged between 2–4. Patients reported the highest median when describing how often they are aware of their knee problem (Q1. How often are you aware of your knee problem?), with 98.1% reporting “daily” or “constantly”. Moreover, 69.2% of patients reported severe to extreme lack of confidence in their affected knee (Q3. How much are you troubled with lack of confidence in your knee?), and 75% reported severe to extreme general difficulty with their knee (Q4. In general, how much difficulty do you have with your knee?). However, 42.3% of participants have “severely” to “totally” modified their lifestyles to avoid potentially damaging activities (Q2. Have you modified your lifestyle to avoid potentially damaging activities?). Hierarchical linear regressions revealed that knee pain on the less affected side significantly improved the model and explained an additional 19% of the variance in KOOS-ADL scores after accounting for age and BMI (Table 3). The addition of knee pain on the affected side (KOOS-Pain) also significantly improved the model and explained a further 50% of the PLOS ONE \| https://doi.org/10.1371/journal.pone.0286962 June 9, 2023 5 / 12 PLOS ONETable 2. KOOS-subscales with the calculated median for each item, the percentage of patients rated the item with a specific Likert point, and percentage of patients who did not rate an item. Knee pain and perceived function in patients with end-stage knee osteoarthritis Subscale Symptoms Pain Function, daily living Quality of Life Item Median (0) None % (1) Mild % (2) Moderate % (3) Severe % (4) Extreme % Did not rate the item %, (N) 1 2 3 4 5 6 7 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 1 2 3 4 2 2 2 0 4 2 2 4 3 1 3 2 3 2 1 2 3 3 2 3 2 2 2 4 1 2 1 2 2 1 2 4 2 4 2 3 3 36.5 19.2 36.5 75.0 15.4 26.9 32.7 0 5.8 32.7 13.5 19.2 0 25.0 34.6 15.4 1.9 0 15.4 7.7 21.2 21.2 5.8 0 36.5 13.5 46.2 9.6 15.4 36.5 28.8 1.9 5.8 0 9.6 5.8 3.8 5.8 9.6 9.6 1.9 5.8 3.8 5.8 0 5.8 21.2 13.5 21.2 9.6 7.7 21.2 9.6 7.7 7.7 13.5 1.9 7.7 13.5 13.5 1.9 15.4 17.3 7.7 15.4 9.6 15.4 19.2 0 9.6 0 23.1 3.8 1.9 26.9 21.2 25 1.9 7.7 32.7 23.1 5.8 15.4 13.5 9.6 28.8 19.2 21.2 30.8 34.6 36.5 26.9 34.6 30.8 19.2 36.5 34.6 7.7 25.0 28.8 26.9 25.0 17.3 28.8 28.8 3.8 30.8 1.9 25.0 21.2 19.2 13.5 9.6 17.3 9.6 5.8 26.9 26.9 40.4 17.3 15.4 11.5 23.1 26.9 34.6 11.5 25.0 23.1 32.7 26.9 48.1 17.3 17.3 30.8 17.3 15.4 23.1 13.5 30.8 15.4 13.5 13.5 5.8 23.1 25.0 32.7 34.6 53.8 17.3 38.5 11.5 9.6 55.8 9.6 11.5 53.8 19.3 9.6 26.9 7.7 32.7 11.5 1.9 15.4 25.0 28.8 7.7 11.5 21.2 9.6 15.4 32.7 1.9 17.3 1.9 17.3 13.5 5.8 9.6 36.5 19.2 73.1 9.6 34.6 21.2 0 1.9% (1) 0 1.9% (1) 9.6% (5) 0 0 0 36% (19) 7.7% (4) 25% (13) 0 11.5% (6) 0 0 0 5.8% (3) 3.8% (2) 1.9% (1) 0 13.5% (7) 1.9% (1) 0 40.4% (21) 5.8% (3) 0 3.8% (2) 1.9% (1) 28.8% (15) 0 0 51.9% (27) 11.5% (6) 0 0 0 0 https://doi.org/10.1371/journal.pone.0286962.t002 variance in this measure (Table 3). Regarding the KOOS-QoL score, pain in the less affected knee did not contribute to this measure. While pain in the affected knee significantly predicted the KOOS-QoL score and explained 41% of the variance after accounting for variables in pre- ceding steps (BMI and age, knee pain on the less affected side) (Table 3). Discussion The goal of this study is to investigate the impact of end-stage knee OA on the perceived physi- cal function and quality of life using the KOOS tool, and to examine the influence of knee pain PLOS ONE \| https://doi.org/10.1371/journal.pone.0286962 June 9, 2023 6 / 12 PLOS ONEKnee pain and perceived function in patients with end-stage knee osteoarthritis Table 3. Hierarchical linear regression analysis with KOOS-ADLS and KOOS-QOL as dependent variables, and age, BMI, knee pain as independent variables. R R2 R2 change Significance of change Significance of model KOOS-ADLS Model 1 2 3 KOOS-QoL Model 1 2 3 0.252 0.503 0.869 0.201 0.292 0.705 0.063 0.253 0.755 0.041 0.085 0.496 0.063 0.189 0.502 0.041 0.045 0.411 0.214 0.001 0.000 0.378 0.141 0.000 0.214 0.004 0.000 0.378 0.247 0.000 * Model 1 = body mass index (BMI) and age; Model 2 = BMI and age, pain in the “less affected” knee; Model 3 = BMI and age, pain in the” less affected” knee, and pain in the “most affected” knee https://doi.org/10.1371/journal.pone.0286962.t003 on functional outcomes and QoL. Although several studies have examined the functional abili- ties of patients with knee OA in different countries, to our knowledge this is the first study conducted on the Jordanian population with end-stage knee OA. Only four males took part in this study, which may reflect the fact that more women than men in Jordan have knee OA and are awaiting TKA. In a hospital-based cohort study in Jor- dan, women constitute 61.4% of patients diagnosed with severe radiographic knee OA [14]. Besides, in the USA, women are 40% more likely than men to develop knee osteoarthritis [18] and over 60% of TKA patients are females [8]. Findings of this study showed that the knee condition negatively impacts the functional abilities and quality of life as perceived by patients in the current study. In general, patients in this study scored low across KOOS subscales (28%-54%), indicating limited abilities in per- forming daily activities and poor quality of life. Although no population-based reference data for KOOS are available from the Jordanian population, patients in this study demonstrated poorer outcomes when compared to those reported in Southern Sweden [19], scoring 42%, 30%, 39%, and 60% lower in: pain, symptoms, ADL, and quality of life subscales, respectively (given that lower scores indicate worse outcomes). Regarding the patient population, various studies have evaluated the reported function by KOOS in those with knee OA [20, 21], and pre-operative to TKA [22–24]. The KOOS scores in the current study are slightly lower than those reported by Sivachidambaram et al. [20]; a study that examined the relationship between KOOS with the performance-based function in patients with primary Knee OA, with a mean difference ranging from 15 to 21 points across all KOOS subscales. Similarly, compared to the pre-operative KOOS scores reported in a study that evaluated the effect of preoperative physio- therapy on the functional outcomes after TKA [22], our scores were lower across all subscales, with a mean difference of 3–27 points. On the other hand, our patients’ scores were slightly better compared to those reported by patients with symptomatic bilateral knee OA in Sabirli and colleagues study [21], with a mean difference of 1–10 points across KOOS subscales. Moreover, when compared to the scores reported by Rastogi [24], a study conducted on patients before undergoing primary TKA, our scores were very similar to theirs. In the current study, each subscale’s element and its difficulty level have been thoroughly examined. Patients in our sample reported that the most impairment is in their ability to bend their knee fully. Moreover, most patients (94.2%) experienced knee pain that persists daily or always. When looking at the elements of the daily activities’ subscale, there were dissimilar responses across the elements indicating that some activities are more difficult to perform than others. The most challenging activities were when going shopping and doing heavy domestic PLOS ONE \| https://doi.org/10.1371/journal.pone.0286962 June 9, 2023 7 / 12 PLOS ONEKnee pain and perceived function in patients with end-stage knee osteoarthritis duties, followed by ascending and descending stairs and standing. It is noticeable that 50% and 42.3% of responders reported: “difficulty to extreme difficulty” when going shopping and per- forming heavy domestic duties, in order. At the same time, a large proportion of patients (40.4% and 51.9%, respectively) did not rate the difficulty they encountered when performing these two activities, which could be due to their inability or unwillingness to perform these tasks. Patients in current study are older adults; in our culture, it is uncommon for the elderly to engage in such activities. Moreover, ascending stairs, standing, and descending stairs activi- ties seem challenging in which 61.5%, 59.6% and 48.1% of patients recognized their perfor- mance as difficult to extremely difficult. On the other hand, some tasks are the easiest for our patients including “Putting on socks”, “Taking off socks”, and “sitting”. Overall, our patients expressed difficulties in tasks that are required daily. Regarding the impact on health related QoL, osteoarthritis has a strong clinically important negative effect [25], and patients with knee OA have significantly lower QoL compared with healthy controls [26, 27]. Our findings were similar; patients scored very low on the QoL sub- scale. Nearly all patients in this study reported they are aware of their knee condition on a daily or ongoing basis, and most of them are severely to extremely troubled because they lack confidence in their knee and have severe to extreme general difficulty with their knee. This indicates how the knee OA condition mostly influences not only the physical but also the emo- tional functioning and the level of confidence as measured by the KOOS-QoL subscale. Inter- estingly, after excluding the sport and recreation subscale, the findings of this study are consistent with a general trend observed in earlier studies with the QoL subscale having the lowest score [28–30], and the symptom subscale having the maximum score [28, 29] among the KOOS subscales. According to the findings of this study, knee pain on the "less affected" and on the "most affected" sides were determinants of self-perceived function on the KOOS-ADLs. While only pain in the “most affected” side was a determinant for the knee related QoL. These results are consistent with those reported in previous studies; joint pain was a predictor to self-reported function in patients with knee OA [31–33], and in those with hip OA [34]. Mizner and col- leagues also reported that pain was strongly related to the patient-reported KOS-ADLs score [35]. Besides, a study by Stonga et al revealed that patients with end-stage knee OA are at greater risk of falling compared to healthy older adults, with pain and limited functional ability negatively influencing the patient’s quality of life and increased the fall risk [36]. It is expected that functional abilities and QoL in our sample will continue to worsen during the waiting period. It has been shown that patients with knee OA demonstrated a reduced quality of life compared to the age-matched norms, and as the disease progresses, the QoL declines [27, 37]. It was reported by Ackerman and colleagues that health-related QoL contin- ues to deteriorate in 75% of patients awaiting for TKA during the mean 10-month waiting period before surgery [37]. Moreover, patients waiting for total hip and knee arthroplasty per- ceived the surgical wait as a contributor to the amount of health deterioration [38]. It is clearly evident that primary objective in OA rehabilitation is to reduce joint pain, increase strength, and improve the functional abilities and quality of life [39]. Given that pain is a determinant of functional ability and QoL, our findings may emphasize the importance and benefit of provid- ing a therapeutic intervention to help relieve or control osteoarthritis-related knee pain which may mitigate the potential deterioration in functional and QoL levels during waiting time. Patients in the current study, however, reported no exercise or physical therapy intervention while awaiting TKA. Lifestyle modification, such as regular low-impact exercise, mass reduction, and avoidance of activities that significantly aggravate the patients’ symptoms, is considered a core part of the conservative treatment for knee OA [40, 41]. These modifications can help in reducing knee PLOS ONE \| https://doi.org/10.1371/journal.pone.0286962 June 9, 2023 8 / 12 PLOS ONEKnee pain and perceived function in patients with end-stage knee osteoarthritis pain and slowing the progression of the condition [41]. We noticed that 33% of patients in cur- rent study either did not or only slightly modified their lifestyle while 42% of patients have severely to totally modified their lifestyle (Table 2, QoL element 2). This variability in responses may be attributed to the patient’s level of pain; the severity of the knee pain may necessitate lifestyle modifications. However, it may also indicate a lack of awareness about the importance of adopting such modifications to minimize further potential damage to their knees. It was reported by Al-Khlaifat et al. that Jordanian patients with knee OA primarily rely on medication as first option to manage their pain, which may be attributed to several factors including a lack of knowledge about pain management, a lack of education on the importance of therapeutic exercises, ineffective service delivery process, and inappropriate exercise pre- scription [42]. Hence, increasing patient’s awareness about pain management and lifestyle modifications to reduce pain is warranted. It is of utmost importance to encourage the patient to participate in rehabilitation program to reduce pain and improve function while waiting for TKA which may enhance/ or slow down the deterioration their functional and QoL outcomes, with physicians, surgeons, and therapists having a crucial role in doing so. Moreover, it is a message to the orthopedic surgeons to consider the patient’s perceived functional outcomes in clinical decision-making when determining the timing and priority of TKA. Limitations and future directions This study in not without any limitations. Our sample may not be representative of all patients with end-stage knee OA across Jordan. Even though knee OA is more prevalent among female patients, the overrepresentation of females in our study participants and the exclusion of comorbid conditions that are common in this age group may limit the generalizability of our findings to all individuals with knee OA. The effect of gender on KOOS outcome couldn’t be evaluated in this study due to the limited number of male participants, and this could be inves- tigated in a future study with a more diverse and representative sample. Additionally, in this analysis, we looked at the relation between pain with the perceived function and quality of life, however; many other variables could also be considered as contributing factors but could not be included in this study, such as muscle strength, socioeconomic status, psychological factors, disease duration of knee OA, and level of physical activity. However, it was not possible in this study to cover all the complex variables influencing knee related perceived function and qual- ity of life. Future studies that investigate the influence of such variables on knee-related per- ceived function and quality of life, could provide a more comprehensive understanding of the subject, which could, in turn, inform the development of more effective interventions for patients with end-stage knee OA. Conclusion Although our findings are parallel to previous studies from different countries, an accurate comparison cannot be made due to different inclusion criteria and severity of the knee condi- tion on those studies. Overall, our study indicates that Jordanian patients with end-stage knee OA showed a substantial impairment in perceived function and knee-related quality of life. The perception of functional abilities and quality of life is negatively influenced by knee pain level in our study group. This study emphasizes the importance of addressing knee pain through a targeted regimen prior to TKA, as well as raising patient awareness about pain man- agement and the importance of exercise, which may positively affect the perceived functional ability and quality of life while waiting for TKA. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286962 June 9, 2023 9 / 12 PLOS ONEKnee pain and perceived function in patients with end-stage knee osteoarthritis Supporting information S1 File Dataset. . (XLSX) Acknowledgments We acknowledge the research assistant; Lubna Alnajjar for her contribution in recruiting sub- jects and collecting data. We would like to thank the University of Jordan Hospital, Depart- ment of Orthopaedics, and particularly the referring surgeons for their continued support for this work. We also thank all the participants in this study. Author Contributions Conceptualization: Sumayeh Abujaber, Ibrahim Altubasi. Data curation: Sumayeh Abujaber, Ibrahim Altubasi. Formal analysis: Sumayeh Abujaber, Ibrahim Altubasi. Funding acquisition: Sumayeh Abujaber. Investigation: Sumayeh Abujaber. Methodology: Sumayeh Abujaber, Ibrahim Altubasi. Project administration: Sumayeh Abujaber. Resources: Mohammad Hamdan, Raed Al-Zaben. Writing – original draft: Sumayeh Abujaber, Ibrahim Altubasi. Writing – review & editing: Sumayeh Abujaber, Ibrahim Altubasi, Mohammad Hamdan, Raed Al-Zaben. References 1. Lawrence RC, Felson DT, Helmick CG, Arnold LM, Choi H, Deyo RA, et al. Estimates of the prevalence of arthritis and other rheumatic conditions in the United States. Part II. Arthritis Rheum. 2008; 58: 26– 35. https://doi.org/10.1002/art.23176 PMID: 18163497 2. Oliveria SA, Felson DT, Reed JI, Cirillo PA, Walker AM. Incidence of symptomatic hand, hip, and knee osteoarthritis among patients in a health maintenance organization. Arthritis Rheum. 1995; 38: 1134– 1141. https://doi.org/10.1002/art.1780380817 PMID: 7639811 3. Cui A, Li H, Wang D, Zhong J, Chen Y, Lu H. Global, regional prevalence, incidence and risk factors of knee osteoarthritis in population-based studies. EClinicalMedicine. 2020;29–30: 100587. https://doi. org/10.1016/j.eclinm.2020.100587 PMID: 34505846 4. Cross M, Smith E, Hoy D, Nolte S, Ackerman I, Fransen M, et al. The global burden of hip and knee osteoarthritis: Estimates from the Global Burden of Disease 2010 study. Ann Rheum Dis. 2014; 73: 1323–1330. https://doi.org/10.1136/annrheumdis-2013-204763 PMID: 24553908 5. Guccione AA, Felson DT, Anderson JJ, Anthony JM, Zhang Y, Wilson PWF, et al. The effects of specific medical conditions on the functional limitations of elders in the Framingham study. Am J Public Health. 1994; 84: 351–358. https://doi.org/10.2105/ajph.84.3.351 PMID: 8129049 6. Peat G, McCarney R, Croft P. Knee pain and osteoarthritis in older adults: A review of community bur- den and current use of primary health care. Ann Rheum Dis. 2001; 60: 91–97. https://doi.org/10.1136/ ard.60.2.91 PMID: 11156538 7. Kremers HM, Larson DR, Crowson CS, Kremers WK, Washington RE, Steiner CA, et al. Prevalence of total hip and knee replacement in the United States. J Bone Jt Surg—Am Vol. 2014. https://doi.org/10. 2106/JBJS.N.01141 PMID: 26333733 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286962 June 9, 2023 10 / 12 PLOS ONEKnee pain and perceived function in patients with end-stage knee osteoarthritis 8. Kurtz S, Ong K, Lau E, Mowat F, Halpern M. Projections of primary and revision hip and knee arthro- plasty in the United States from 2005 to 2030. J Bone Jt Surg. 2007; 89: 780–785. https://doi.org/10. 2106/JBJS.F.00222 PMID: 17403800 9. Desmeules F, Dionne CE, Belzile E, Bourbonnais R, Fre´mont P. The burden of wait for knee replace- ment surgery: effects on pain, function and health-related quality of life at the time of surgery. Rheuma- tology (Oxford). 2010; 49: 945–954. https://doi.org/10.1093/rheumatology/kep469 PMID: 20144931 10. Collins NJ, Misra D, Felson DT, Crossley KM, Roos EM. Measures of knee function: International Knee Documentation Committee (IKDC) Subjective Knee Evaluation Form, Knee Injury and Osteoarthritis Outcome Score (KOOS), Knee Injury and Osteoarthritis Outcome Score Physical Function Short Form (KOOS-PS), Knee Ou. Arthritis Care Res. 2011; 63: 208–228. https://doi.org/10.1002/acr.20632 PMID: 22588746 11. Roos EM, Lohmander LS. The Knee injury and Osteoarthritis Outcome Score (KOOS): From joint injury to osteoarthritis. Health Qual Life Outcomes. 2003; 1: 64. https://doi.org/10.1186/1477-7525-1-64 PMID: 14613558 12. Irrgang JJ, Snyder-Mackler L, Wainner RS, Fu FH, Harner CD. Development of a patient-reported mea- sure of function of the knee. J Bone Jt Surg. 1998; 80: 1132–1145. https://doi.org/10.2106/00004623- 199808000-00006 PMID: 9730122 13. Bellamy N, Buchanan WW, Goldsmith CH, Campbell J, Stitt LW. Validation study of WOMAC: A health status instrument for measuring clinically important patient relevant outcomes to antirheumatic drug therapy in patients with osteoarthritis of the hip or knee. J Rheumatol. 1988; 15: 1833–1840. PMID: 3068365 14. Hawamdeh ZM, Al-Ajlouni JM. The clinical pattern of knee osteoarthritis in Jordan: A hospital based study. Int J Med Sci. 2013; 10: 790–795. https://doi.org/10.7150/ijms.5140 PMID: 23630445 15. Al-Khlaifat L, Okasheh R, Muhaidat J, Hawamdeh ZM, Qutishat D, Al-Yahya E, et al. Knowledge of Knee Osteoarthritis and Its Impact on Health in the Middle East: Are They Different to Countries in the Developed World? A Qualitative Study. Rehabil Res Pract. 2020;2020. https://doi.org/10.1155/2020/ 9829825 PMID: 32455026 16. Alfadhel SA, Vennu V, Alnahdi AH, Omar MT, Alasmari SH, AlJafri Z, et al. Cross-cultural adaptation and validation of the Saudi Arabic version of the Knee Injury and Osteoarthritis Outcome Score (KOOS). Rheumatol Int. 2018; 38: 1547–1555. https://doi.org/10.1007/s00296-018-4072-7 PMID: 29882042 17. Roos EM, Roos HP, Lohmander LS, Ekdahl C, Beynnon BD. Knee Injury and Osteoarthritis Outcome Score (KOOS)—Development of a self-administered outcome measure. J Orthop Sports Phys Ther. 1998; 28: 88–96. https://doi.org/10.2519/jospt.1998.28.2.88 PMID: 9699158 18. Losina E, Weinstein AM, Reichmann WM, Burbine SA, Solomon DH, Daigle ME, et al. Lifetime risk and age at diagnosis of symptomatic knee osteoarthritis in the US. Arthritis Care Res. 2013; 65: 703–711. https://doi.org/10.1002/acr.21898 PMID: 23203864 19. Paradowski PT, Bergman S, Sunde´ n-Lundius A, Lohmander LS, Roos EM. Knee complaints vary with age and gender in the adult population. Population-based reference data for the Knee injury and Osteo- arthritis Outcome Score (KOOS). BMC Musculoskelet Disord. 2006; 7: 1–8. https://doi.org/10.1186/ 1471-2474-7-38 PMID: 16670005 20. Sivachidambaram K, Ateef M, Tahseen S. Correlation of Self-Reported Questionnaire (KOOS) with Some Objective Measures in Primary OA Knee Patients. ISRN Rheumatol. 2014; 2014: 1–5. https:// doi.org/10.1155/2014/301485 PMID: 24527222 21. Sabirli F, Paker N, Bugdayci D. The relationship between Knee Injury and Osteoarthritis Outcome Score (KOOS) and Timed Up and Go test in patients with symptomatic knee osteoarthritis. Rheumatol Int. 2013; 33: 2691–2694. https://doi.org/10.1007/s00296-012-2512-3 PMID: 22955800 22. Mat Eil-Ismail MS, Sharifudin MA, Ahmed Shokri A, Ab Rahman S. Preoperative physiotherapy and short-term functional outcomes of primary total knee arthroplasty. Singapore Med J. 2016; 57: 138– 143. https://doi.org/10.11622/smedj.2016055 PMID: 26996450 23. Tilbury C, Haanstra TM, Verdegaal SHM, Nelissen RGHH, De Vet HCW, Vliet Vlieland TPM, et al. Patients’ pre-operative general and specific outcome expectations predict postoperative pain and func- tion after total knee and total hip arthroplasties. Scand J Pain. 2018; 18: 457–466. https://doi.org/10. 1515/sjpain-2018-0022 PMID: 29794270 24. Rastogi R, Chesworth BM, Davis AM. Change in patient concerns following total knee arthroplasty described with the International Classification of Functioning, Disability and Health: A repeated mea- sures design. Health Qual Life Outcomes. 2008; 6. https://doi.org/10.1186/1477-7525-6-112 PMID: 19077246 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286962 June 9, 2023 11 / 12 PLOS ONEKnee pain and perceived function in patients with end-stage knee osteoarthritis 25. Parker L, Moran GM, Roberts LM, Calvert M, McCahon D. The burden of common chronic disease on healthrelated quality of life in an elderly communitydwelling population in the UK. Fam Pract. 2014; 31: 557–563. https://doi.org/10.1093/fampra/cmu035 PMID: 24987022 26. Alkan BM, Fidan F, Tosun A, Ardic¸ogˇ lu O¨ . Quality of life and self-reported disability in patients with knee osteoarthritis. Mod Rheumatol. 2014; 24: 166–171. https://doi.org/10.3109/14397595.2013.854046 PMID: 24261774 27. Desmeules F, Dionne CE, Belzile E´ , Bourbonnais R, Fre´mont P. Waiting for total knee replacement sur- gery: Factors associated with pain, stiffness, function and quality of life. BMC Musculoskelet Disord. 2009; 10: 1–10. https://doi.org/10.1186/1471-2474-10-52 PMID: 19457252 28. Naili JE, Iversen MD, Esbjo¨ rnsson AC, Hedstro¨ m M, Schwartz MH, Ha¨ger CK, et al. Deficits in func- tional performance and gait one year after total knee arthroplasty despite improved self-reported func- tion. Knee Surgery, Sport Traumatol Arthrosc. 2017; 25: 3378–3386. https://doi.org/10.1007/s00167- 016-4234-7 PMID: 27436196 29. Huang KH, Hsieh RL, Lee WC. Pain, physical function, and health in patients with knee osteoarthritis. Rehabil Nurs. 2017; 42: 235–241. https://doi.org/10.1002/rnj.234 PMID: 26391619 30. Stevens-Lapsley JE, Schenkman ML, Dayton MR. Comparison of Self-Reported Knee Injury and Oste- oarthritis Outcome Score to Performance Measures in Patients After Total Knee Arthroplasty. PM R. 2011; 3: 541–549. https://doi.org/10.1016/j.pmrj.2011.03.002 PMID: 21665167 31. Kauppila AM, Kyllo¨nen E, Mikkonen P, Ohtonen P, Laine V, Siira P, et al. Disability in end-stage knee osteoarthritis. Disabil Rehabil. 2009; 31: 370–380. https://doi.org/10.1080/09638280801976159 PMID: 18608423 32. Maly MR, Costigan PA, Olney SJ. Determinants of self-report outcome measures in people with knee osteoarthritis. Arch Phys Med Rehabil. 2006; 87: 96–104. https://doi.org/10.1016/j.apmr.2005.08.110 PMID: 16401446 33. Stratford PW, Kennedy DM. Performance measures were necessary to obtain a complete picture of osteoarthritic patients. J Clin Epidemiol. 2006; 59: 160–7. https://doi.org/10.1016/j.jclinepi.2005.07.012 PMID: 16426951 34. Joseph Z, Abujaber S, Pozzi F, Raisis L. Relationship between strength, pain, and different measures of functional ability in patients with end-stage hip osteoarthritis. Arthritis Care Res. 2014; 66: 1506– 1512. https://doi.org/10.1002/acr.22329 PMID: 24664893 35. Mizner RL, Petterson SC, Clements KE, Zeni JA, Irrgang JJ, Snyder-Mackler L. Measuring functional improvement after total knee arthroplasty requires both performance-based and patient-report assess- ments: a longitudinal analysis of outcomes. J Arthroplasty. 2011; 26: 728–37. https://doi.org/10.1016/j. arth.2010.06.004 PMID: 20851566 36. Tsonga T, Michalopoulou M, Malliou P, Godolias G, Kapetanakis S, Gkasdaris G, et al. Analyzing the history of falls in patients with severe knee osteoarthritis. CiOS Clin Orthop Surg. 2015; 7: 449–456. https://doi.org/10.4055/cios.2015.7.4.449 PMID: 26640627 37. Ackerman IN, Bennell KL, Osborne RH. Decline in Health-Related Quality of Life reported by more than half of those waiting for joint replacement surgery: A prospective cohort study. BMC Musculoskelet Dis- ord. 2011;12. https://doi.org/10.1186/1471-2474-12-108 PMID: 21605398 38. Snider MG, MacDonald SJ, Pototschnik R. Waiting times and patient perspectives for total hip and knee arthroplasty in rural and urban Ontario. Can J Surg. 2005; 48: 355–360. PMID: 16248132 39. Nguyen C, Lefèvre-Colau MM, Poiraudeau S, Rannou F. Rehabilitation (exercise and strength training) and osteoarthritis: A critical narrative review. Ann Phys Rehabil Med. 2016; 59: 190–195. https://doi. org/10.1016/j.rehab.2016.02.010 PMID: 27155923 40. Wall C, Johnson T, de Steiger R. Symptom management for patients awaiting joint replacement sur- gery. Aust J Gen Pract. 2020; 49: 444–446. https://doi.org/10.31128/AJGP-03-20-5286 PMID: 32600002 41. Erin Connelly A, Tucker AJ, Kott LS, Wright AJ, Duncan AM. Modifiable lifestyle factors are associated with lower pain levels in adults with knee osteoarthritis. Pain Res Manag. 2015; 20: 241–248. https:// doi.org/10.1155/2015/389084 PMID: 26125195 42. Al-Khlaifat L, Okasheh R, Muhaidat J, Hawamdeh ZM, Qutishat D, Al-Yahya E, et al. Perceptions and performance of exercise in people with knee osteoarthritis in the Middle East: are they different to coun- tries in the developed world? A qualitative study in Jordan. Physiother Theory Pract. 2022; 38: 55–66. https://doi.org/10.1080/09593985.2020.1728793 PMID: 32077786 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286962 June 9, 2023 12 / 12 PLOS ONE
10.1371_journal.pone.0286828	RESEARCH ARTICLE RNAi of Complex I and V of the electron transport chain in glutamate neurons extends life span, increases sleep, and decreases locomotor activity in Drosophila melanogaster Jessie E. LandisID, Kevin Sungu, Hannah Sipe, Jeffrey M. CopelandID* Department of Biology, Eastern Mennonite University, Harrisonburg, VA, United States of America a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 * jeffrey.copeland@emu.edu Abstract OPEN ACCESS Citation: Landis JE, Sungu K, Sipe H, Copeland JM (2023) RNAi of Complex I and V of the electron transport chain in glutamate neurons extends life span, increases sleep, and decreases locomotor activity in Drosophila melanogaster. PLoS ONE 18(6): e0286828. https://doi.org/10.1371/journal. pone.0286828 Editor: Efthimios M. C. Skoulakis, Biomedical Sciences Research Center Alexander Fleming, GREECE Received: December 16, 2022 Accepted: May 24, 2023 Published: June 15, 2023 Copyright: © 2023 Landis et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. Funding: J.M.C. is supported by a Small Project Research Grant from the Virginia Academy of Science. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.” RNAi targeting the electron transport chain has been proven to prolong life span in many dif- ferent species, and experiments specifically with Drosophila melanogaster and Caenorhab- ditis elegans have shown a distinct role for neurons. To determine which subset of neurons is implicated in this life span extension, we used the GAL4/UAS system to activate RNAi against genes of Complex I and Complex V. We found life span extension of 18–24% with two glutamate neuron (D42 and VGlut) GAL4 lines. We used the GAL80 system to deter- mine if the overlapping set of glutamate neurons in these two GAL4 lines imparts the life span extension. Limiting GAL4 activity to non-VGlut glutamate neurons in the D42 back- ground failed to extend life span, suggesting that glutamate neurons have an important role in aging. Interestingly, RNAi of the electron transport chain in D42 glutamate neurons also caused an increase in daytime and nighttime sleep and a decrease in nighttime locomotor activity. Changes to sleep patterns and prolonged life span were not accompanied by any changes in female fertility or response to starvation. Our findings demonstrate that a small subset of neurons can control life span, and further studies can look into the contributions made by glutamate neurons. Introduction Aging is marked by progressive tissue dysfunction, with common hallmarks in mammals, flies, worms, and fish. Such tissue dysfunction includes neurological impairments in cognitive function and memory and an increased likelihood of neurodegenerative diseases [1]. Within cells, aging manifests as mitochondrial dysfunction, telomere loss, cell senescence, stem cell exhaustion, and disrupted cell signaling [2]. Mitochondrial dysfunction is a critical determinant in physiological aging, and the electron transport chain (ETC) is essential to mitochondrial activity. The ETC is comprised of five large protein complexes and produces most of the ATP and reactive oxygen species (ROS) in the cell. Changes by RNAi or mutations in individual components of the ETC are known to extend life span in many disparate species [3]. Loss of ETC activity is not the only means to PLOS ONE \| https://doi.org/10.1371/journal.pone.0286828 June 15, 2023 1 / 14 PLOS ONECompeting interests: The authors have declared that no competing interests exist. RNAi of the ETC extends life span and increases sleep in Drosophila prolong life span. Overexpression of Ndi-1, a single component acting as NADH dehydroge- nase/ubiquinone oxidoreductase I, can likewise extend life span in Drosophila [4, 5]. One con- nection between aging, ETC function, and life span is the production of ROS in the mitochondria. It has been observed that low concentrations of ROS in the cell stimulates mito- hormesis, an adaptive cytoprotective response, but high concentrations accelerate aging [6, 7]. The concentration of ROS in particular is determined by the individual ETC component involved; a highly reduced pool of ubiquinone drives reverse electron transport and can pro- long life span [8]. Even though targeting ETC components by RNAi is sufficient to extend life span, the production of mitochondrial ROS has yet to be measured in Drosophila and C. elegans. While the creation and the cellular response to ROS are critical for aging, so is the removal of dysfunctional mitochondria. With advancing age, the number of dysfunctional mitochon- dria accumulate within a cell, with the older mitochondria characterized by their swollen size, fragmented appearance, and decreased respiratory ability [9]. Concurrent with the increase of faulty mitochondria is the increased response of mitophagy. Mitophagy is one type of autop- hagy in which dysfunctional mitochondria are selectively sequestered and degraded by the lysosome. Several studies have uncovered that elevated levels of mitochondrial fission extend life span through the recovery of dysfunctional mitochondria [10, 11]. All of these studies underscore the importance of healthy mitochondria in promoting functional cellular activity and normal health span. Particular subsets of cells can aging of the entire animal. For example, RNAi manipulations of the ETC only in neurons can extend life span in both Drosophila and C. elegans [12, 13]. Once again, the importance of mitochondria in neuronal aging is not determined by activity of the ETC alone as changes in the proteins involved in mitophagy show similar effect [10, 11]. In addition, changes to mitochondrial fission in neurons have effects elsewhere. For example, upregulation of the mitophagy protein BNIP3 reduces age-related protein aggregation in mus- cles and improves intestinal barrier function [11]. Certainly not all neurons affect life span equally. Activated RNAi of the ETC in dopamine and serotonin neurons shortens life span in Drosophila [14]. Given the effects that neurons have on aging, questions remain as to which particular neurons impart these changes. Glutamate neurons are involved in several behaviors impacted by age, like movement, sleep, and circadian rhythms. A decline in motor activity corresponds with advanced aging, and a few studies in C. elegans and Drosophila have shown that prolonged life span can nega- tively affect movement [15, 16]. Likewise, sleep patterns change with increasing age in both humans and flies, as observed specifically in fragmented and weakened sleep:wake cycles [17, 18]. Though sleep is a complex behavior involving many neuron clusters in the adult brain, glutamate neurons, specifically the dorsal clock neurons (DN1s), lateral posterior neurons (LPNs), and superior lateral protocerebrum neurons (SPNs), have been shown to have sleep promoting effects [19–21]. Given these connections between life span, movement, sleep, and glutamate neurons, we activated RNAi against two components of the ETC in glutamate neu- rons and explored the effects on life span and these behaviors. Our results demonstrate that RNAi modification of components of the ETC in glutamate neurons extends life span, decreases nighttime locomotor activity, and increases daytime and nighttime sleep in flies. Materials and methods Fly strains used The D42-GAL4 (BL-8816), VGlut-GAL4 (OK371; BL-26160), ChAT-GAL4 (BL-6798), and VGlut-GAL80 (BL-58448) fly strains were obtained from the Bloomington Stock Center (NIH PLOS ONE \| https://doi.org/10.1371/journal.pone.0286828 June 15, 2023 2 / 14 PLOS ONERNAi of the ETC extends life span and increases sleep in Drosophila P40OD018537, Bloomington, IN). The RNAi lines targeting the electron transport chain genes ND-20/CG9172 (23256) and ATPsynF062L/CG5389 (22112) were purchased from the Vienna Drosophila RNAi Center (Vienna, Austria) and were backcrossed ten times to the white1118 laboratory strain to minimize hybrid vigor [22]. Life span assays Flies for the life span assays were kept in a temperature-controlled (25˚C) container with 12:12 light:dark cycle at an average density of 25 flies per vial. Male and female adult flies were main- tained separately. Flies were maintained on standard cornmeal food and flipped to fresh food twice a week. All life span experiments were conducted in duplicate. Prism9 (GraphPad, San Diego, CA) and Excel (Microsoft) software were used to analyze the life span data. Activity monitor 10 day old males with D42-GAL4 activated RNAi against CG9172 and CG5389 or crossed with white1118 laboratory strain were used for behavioral analysis. Flies were monitored with a 2 μW/cm2 intensity green LED light for 7 days at 12:12 light:dark cycle, followed by 7 days of 24 hrs dark cycle in Trikinetics activity monitors (Waltham, MA). Flies were maintained at 22˚C, 60% relative humidity. 32 male flies were used per condition, and activity monitor tubes contained a plug of fly food at one end. Experiments using the activity monitor were con- ducted in duplicate. Data were analyzed using the ShinyR-DAM program [23] and Microsoft Excel. Fertility To measure female fertility in the D42-GAL4 activated RNAi lines, 3-day old virgin females were mated to 2-day old w1118 males. Each fly cross contained 5 adult females and 7 adult males and each experiment had three vials. Parental flies were transferred to new vials 2–3 times per week, and progeny were counted every day after eclosion. Experiments were con- ducted in duplicate. Stress resistance Response to starvation were recorded in the D42-GAL4 activated electron transport chain RNAi lines. Starvation conditions were created by placing 10-day old females on 1% agar. Experiments were conducted in duplicate. Results In previous studies, RNAi targeting individual components of the electron transport chain only to neurons was sufficient to prolong Drosophila life span by approximately 15%, a differ- ence of approximately 10 days [12]. Given these results, we set out to uncover which neuronal subtypes are important for life span extension. In this study, we used the GAL4/UAS system to activate RNAi in specific subsets of neurons. Specifically, the D42-GAL4 and the VGlut-GAL4 (OK371-GAL4) driver lines were used to broadly target an overlapping set of glutamate neu- rons, and ChAT-GAL4 to target cholinergic neurons. Using both D42- and VGlut-GAL4 driver lines was important as these lines show overlapping expression in adult and larval brains and in motor neurons. RNAi was used to affect the activity of the Complex I gene CG9172 and the Complex V gene CG5389. The life span of flies with activated RNAi was compared to that of heterozygous GAL4 flies without the UAS-RNAi elements. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286828 June 15, 2023 3 / 14 PLOS ONERNAi of the ETC extends life span and increases sleep in Drosophila Fig 1. RNAi of electron transport chain components in glutamate neurons is sufficient for female life span extension in Drosophila. Females of the indicated GAL4 driver were mated to a white1118 control (open circles) or UAS-CG9172-RNAi (light blue), or UAS-CG5389-RNAi (dark blue) lines. (A, B) Survival curves of RNAi against CG9172 and CG5389 RNAi in D42-GAL4 glutamate neurons show comparable 23% and 18% (p < 0.0001, log rank test) life span extension. (C, D) In VGlut-GAL4 glutamate neurons, RNAi of CG9172 and CG5389 leads to an extension of 24% and 22%, respectively (p < 0.0001). (E, F) RNAi of CG9172 and CG5389 in ChAT-GAL4 acetylcholine neurons extends life span by 12% (p < 0.0001). (G, H) Activating RNAi against CG9172 and CG5389 to the set of non-overlapping set of glutamate neurons (D42-GAL4; VGlut-GAL80) minimizes life span extension to 6% and 5%, respectively (p < 0.0001). https://doi.org/10.1371/journal.pone.0286828.g001 We observed a life span extension of 23% and 18% (p < 0.0001), respectively, when RNAi targeted CG9172 and CG5389 in D42-GAL4 glutamate neurons (Fig 1A and 1B; Table 1). Simi- lar life span extension of 24% and 22% (p < 0.0001) was observed using the VGlut-GAL4 line PLOS ONE \| https://doi.org/10.1371/journal.pone.0286828 June 15, 2023 4 / 14 PLOS ONERNAi of the ETC extends life span and increases sleep in Drosophila Table 1. Descriptive statistics of female life span in neuron-specific RNAi of the electron transport chain. GAL4 driver RNAi target Mean lifespan (days) % change p value (log rank) Maximum average lifespan (days) % change p value (t test) N D42-GAL4 VGlut-GAL4 ChAT-GAL4 VGlut-GAL80; D42- GAL4 Control CG9172 CG5389 Control CG9172 CG5389 Control CG9172 CG5389 Control CG9172 CG5389 60.4 74.4 17.3 60.6 75.3 73.8 68 76.4 75.9 63.9 67.6 67.2 https://doi.org/10.1371/journal.pone.0286828.t001 23.2 18 24.4 21.8 12.4 11.6 5.8 5.2 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 79.9 87.7 84.8 80.8 94.2 92.4 90.6 92.8 92 78.2 82.7 80.5 9.8 6.1 16.6 14.4 2.4 1.5 5.8 2.9 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 593 471 548 437 387 387 342 356 361 645 633 636 (Fig 1C and 1D). A smaller life span extension of about 12% was observed with RNAi activa- tion in cholinergic neurons (p < 0.0001). Maximum life span, as measured as the life span of the top 10% longest living cohort, was also prolonged in flies with RNAi activated by both glu- tamate GAL4 lines. A marginal increase in maximum life span was seen with RNAi in cholin- ergic neurons (Table 1). Although the response was not as consistent as for female flies, average male life span was prolonged by 11 to 28% in ETC RNAi activated by the D42- and VGlut-GAL4 lines, showing that the response is conserved among the sexes (S1 Fig and S1 Table). Life spans of unactivated UAS-CG5389-RNAi and UAS-CG9172-RNAi showed only a 6–7% increase when compared to a white1118 control line, indicating that extension is not sim- ply an artifact of the UAS-RNAi element itself (S2 Fig). To determine if the overlapping set of glutamate neurons are responsible for the life exten- sion, the GAL4/GAL80 system was employed to block RNAi activation in these cells. In our experiment, ETC RNAi was activated in D42-GAL4 glutamate neurons, but inhibited in VGlut-GAL80 expressing neurons. In these flies, life span extension was significantly reduced in activated CG9172 and CG5389 RNAi lines (Fig 1G and 1H). These results highlight the cru- cial role that glutamate neurons play in life span extension. We did notice that activated RNAi in cholinergic neurons had no effect on life span in males (S1 Fig and S1 Table). With this mixed effect in cholinergic neurons, we focused our experiments using the D42 glutamate neu- ron GAL4 driver line, though we recognize the potential importance of cholinergic neurons in aging. Decreased Drosophila motor activity during dark phase in a 12 hour light/ dark cycle With such a noticeable life span extension from glutamate neurons with activated ETC RNAi, we wanted to explore a possible relationship between activity and aging. We used Trikinetics activity monitors (Waltham, MA) on flies with RNAi targeting CG9172 and CG5389. Cohorts of 10 day old male flies were placed in the activity monitor in 12:12 light/dark (LD) conditions for 7 days and then switched to all dark (DD) settings for 7 days. Even though female flies showed a more robust life span extension and might show a greater difference in activity, we used male flies instead to eliminate egg laying behavior and inclusion of progeny in the mea- surements. Overall average activity was measured for 24 hour periods, as well as activity specif- ically during the light phase and dark phase. As a control group, flies carrying only the D42- GAL4 genetic element were included. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286828 June 15, 2023 5 / 14 PLOS ONERNAi of the ETC extends life span and increases sleep in Drosophila Fig 2. RNAi of electron transport chain components in glutamate neurons decreases nighttime activity in 10 day old flies. Adult male flies of either the D42-GAL4 control line (white) and activated CG9172 RNAi (light blue) and CG5389 RNAi (dark blue) were tested for their activity starting at 10 days of adulthood. Data are presented as means +/- SEM and compared using one-way ANOVA with a Tukey’s post-hoc test. (A) Average activity over 24 hours in a LD cycle saw a decrease of 32% (p = 0.029) and 39% (p = 0.005) in RNAi targeting CG9172 and CG5389. (B) Targeting CG9172 and CG5389 by RNAi leads to a decrease in average activity during the dark phase of a 24 hour LD cycle by 64% and 69% (p < 0.0001), respectively. (C) Average activity during the light phase was not statistically altered in CG9172 and CG5389 RNAi activated flies (p = 0.86, p = 0.54), respectively. (D) Representative activity profile over 24 hours. (E) RNAi targeting of CG9172 and CG5389 in a 24 DD cycle decreased activity by 21% (p = 0.0011) and 32% (p < 0.0001), respectively. https://doi.org/10.1371/journal.pone.0286828.g002 CG9172 and CG5389 RNAi flies were significantly less active compared to the D42-GAL4 control flies. In the 24 hour LD cycle, CG9172 RNAi flies were 32% less active and CG5389 RNAi flies were 40% less active (p < 0.0001) (Fig 2 and S2 Table). This decrease in activity was specific for nighttime. During the 12 hour dark phase, the CG9172 RNAi and CG5389 RNAi flies were 64% and 69% (p < 0.0001) less active than the control group. In the light phase, the decline in activity did not meet statistical significance. The activity of the UAS-CG5389-RNAi and UAS-CG9172-RNAi control flies failed to show any statistical difference with the D42-GAL4 group, indicating that the UAS-RNAi elements are not responsible for the decreased activity (S3 Fig and S2 Table). In general, all flies tended to be more active in the DD setting than in the 24 hour LD phase, but the activated RNAi flies were consistently less active than the D42-GAL4 control line. In DD, the CG9172 RNAi flies were 21% less active (p < 0.0011) and the CG5389 RNAi flies were 32% less active (p < 0.0011). Increased sleep observed during a 12 hour light/dark cycle in electron transport chain RNAi flies The most pronounced effect on activity in the ETC RNAi lines was during the dark phase in the LD conditions. Given the timing of this effect, we wanted to explore a possible connection with sleep. Sleep was defined as bouts of inactivity longer than 5 minutes [18]. Using the data PLOS ONE \| https://doi.org/10.1371/journal.pone.0286828 June 15, 2023 6 / 14 PLOS ONERNAi of the ETC extends life span and increases sleep in Drosophila Fig 3. RNAi of electron transport chain components in glutamate neurons increases sleep in 10 day old flies. Adult male flies of either the D42-GAL4 control line (white) and activated CG9172 RNAi (light blue) and CG5389 RNAi (dark blue) were tested for overall sleep starting at 10 days of adulthood. Sleep data is presented as average time asleep per 30 minutes, shown as means +/- SEM, and compared using one-way ANOVA with a Tukey’s post-hoc test. (A) The average amount of sleep over 24 hours in a LD cycle increases by 95% (p < 0.0001) and 115% (p < 0.0001) in RNAi targeting CG9172 and CG5389. (B) Average sleep amount during the nighttime phase increases by 95% and 107% (p < 0.0001). (C) A similar increase in sleep amounts is seen during the light phase for CG9172 RNAi (94%, p = 0.035) and CG5389 RNAi (130%, p = 0.0022) adults. (D) Representative sleep profile over 24 hours. (E) RNAi targeting of CG9172 and CG5389 increases length of a sleep bout by 31% (p < 0.0001) and 41% (p < 0.0001), respectively. https://doi.org/10.1371/journal.pone.0286828.g003 collected from the activity monitors, the average amount of sleep was calculated per fly for a 30 minute period. Data was also analyzed depending on the phase of the 24 hour light/dark cycle. The data clearly shows a dramatic increase in the amount of sleep in the ETC RNAi flies. The average amount sleep for the 24 hour LD period increased by 95% and 115% (p < 0.0001) for the CG9172 RNAi and CG5389 RNAi lines, respectively (Fig 3 and S3 Table). The increase in sleep was not limited to a particular phase of the LD cycle, but the changes during the dark phase (p < 0.0001) were more statistically significant than the light phase (p = 0.03–0.002). The activated RNAi flies also exhibited a significantly longer sleep bout length than the control. Bout length increased 31.1% and 41.8% (p < 0.0001) in the CG9172 and CG5389 RNAi lines. The unactivated UAS-RNAi elements are not the cause for the increased sleep since these flies sleep comparably to the D42-GAL4 control group (S3 Fig). No effect on female fertility and starvation with electron transport chain RNAi in glutamate neurons In addition to monitoring activity and sleep, fertility rate and response to starvation conditions were studied to assess possible physiological impacts of ETC RNAi. Oftentimes, extended life span causes a reciprocal decline in reproductive capacity [24]. For example, decrease in ETC functioning in Caenorhabditis elegans shows a decrease in fertility as well as movement [15, 25]. In this study, we measured female fertility by counting the amount of adult progeny pro- duced over the first 30 days of adulthood in flies with RNAi activated in D42-GAL4 glutamate PLOS ONE \| https://doi.org/10.1371/journal.pone.0286828 June 15, 2023 7 / 14 PLOS ONERNAi of the ETC extends life span and increases sleep in Drosophila Fig 4. Activation of components of the electron transport chain RNAi in glutamate neurons confers no significant effect on female fertility and starvation response. Females with the D42-GAL4 driver were mated to a white1118 control line (open circles) or RNAi targeting CG9172 (light blue) or CG5389 (dark blue) lines. (A, B) Activation of RNAi to CG9172 (p = 0.98, two way ANOVA) and CG5389 (p = 0.47) in glutamate neurons does not affect female fertility. (C, D). Starvation in females with activated RNAi to the electron transport chain genes specifically in D42- GAL4 neurons has a minimal effect on overall survival (CG9172, no change, p = 0.32; CG5389, decrease 4%, p = 0.0005). https://doi.org/10.1371/journal.pone.0286828.g004 neurons. RNAi of CG9172 or CG5389 did not affect female fertility in comparison to the D42- GAL4 control line (Fig 4). These results are consistent with a previous study that noted no change in female fertility in pan-neuronal RNAi of the same electron transport chain genes [12]. To extend the study of possible physiological costs for prolonged life span, we examined the ETC RNAi flies response to starvation. Starvation conditions were created by placing 10 day old females on 1% agar. ETC RNAi flies had life spans 4% shorter than the D42-GAL4 con- trol lines, though these results were only statistically significant in the CG5389-RNAi flies. Unactivated UAS-RNAi do not affect fertility or response to starvation as these elements have similar response as the D42-GAL4 control group (S4 Fig). Given the minimal effect and the difference between the two ETC RNAi lines, these results do not show any meaningful sensiti- zation to starvation. All data collected for the life span, activity, sleep, fertility, and starvation experiments are included in the supplementary information (S4–S9 Tables) Discussion In this report, we present evidence for a role of glutamate neurons in modulating Drosophila life span. RNAi targeting individual components of the ETC in glutamate neurons was suffi- cient to extend life span by 18 to 24%. RNAi of the same ETC components caused a remark- able increase in sleep and a decrease in activity specifically during the dark phase of a 24 LD cycle. Previous work has established that RNAi of the ETC in neurons alone is sufficient to extend life span. Specifically, activation of RNAi in neurons of components from Complex I, IV, and V and ectopic expression of the single Complex I subunit Ndi1 prolongs life span [12, 26]. The PLOS ONE \| https://doi.org/10.1371/journal.pone.0286828 June 15, 2023 8 / 14 PLOS ONERNAi of the ETC extends life span and increases sleep in Drosophila results presented here show a more refined understanding of neurons in aging. Though we see a more robust life span extension with glutamate neurons, our results are comparable to previ- ous studies using a pan-neuronal GAL4 driver and even within this study with acetylcholine neurons [12]. This is not to say that all neurons participate equally in aging. ETC RNAi in dopamine and serotonin neurons shortens life span [14]. The significant life span extension observed in males ETC RNAi in glutamate neurons raises the possibility that glutamate neu- rons have a more prominent role. We do recognize that the results in males could be an artifact of differences in GAL4 expression between driver lines. The D42-GAL4 and VGlut-GAL4 are routinely used to drive gene expression in glutamate neurons and are known to be active in an overlapping set of cells [27, 28]. In our experiments, we notice a significantly smaller life span extension with the addition of the VGlut-GAL80 element to flies with RNAi active in D42- GAL4 glutamate neurons. These results strongly suggest that glutamate neurons have an important role in life span extension. The fact that alterations of mitochondrial function in a specific tissue affect Drosophila life span is not surprising, as changes in the intestines have similar effects. Intestine-specific expression of Ndi1 or RNAi inactivation of a Complex V subunit can significantly lengthen life span in flies and nematodes, respectively [5, 25]. Life span extension is not limited directly to the ETC as promoting mitochondrial biogenesis in the intestines had similar effects [29]. It is worth noting the complex association between longevity and intestinal integrity. Dysfunction of the intestinal barrier precedes aging-related decline and can cause alterations of the micro- biota, though changes to the microbiota are not driving life span [30, 31]. Our results might have connection with the demonstrated importance of the intestines in life span determina- tion. Other studies have demonstrated that changes in the nutrient sensor AMPK in neurons can induce intestinal autophagy, decrease intestinal barrier dysfunction, and prolong life span [32]. Glutamate motor neurons are known to innervate the digestive tract hindgut and proven- triculus, potentially connecting these two aging-important tissues [33, 34]. Certainly, it is worth exploring the importance of intestinal integrity in life span extension observed in flies with glutamate neuron-specific ETC RNAi. In this work, we confirm the previously noted role of the electron transport chain in sleep and motor activity, though we identify a particular role of ETC function in glutamate neurons [35]. We expected an overall decrease in fly activity with ETC RNAi targeting glutamate neu- rons, as most motor neurons in flies are glutamatergic. It was surprising, however, that the effect on activity was limited to the dark phase in the LD cycle. Conversely, RNAi of compo- nents of Complex I and V drove the flies to sleep more throughout the 24 hr cycle, with longer sleep bouts to account for the increase in total sleep. It is entirely possible that the increase in sleep is a direct consequence of decreased motor activity, but is intriguing to note that the peri- ods with an increase in sleep did not perfectly overlap with the periods of lower activity. Per- haps the changes in activity and sleep are two separate effects associated with changes in ETC activity. Our results showing a genetic connection between sleep and aging join a significant body of research tying the two. In typical Drosophila aging, sleep becomes more disrupted with age, and younger flies have prolonged sleep bouts and more nighttime sleep [17]. In our experi- ments, 10 day old flies with activated ETC RNAi slept much longer, similar to young flies car- rying a Complex V point mutation [36]. The effect of ETC RNAi on sleep during the aging process is still to be explored as our experiments only used 10 day old flies. Other studies have uncovered neuronal genes involved in both sleep and life span, with the genes affecting sleep and aging in a direct relationship [37]. The fact that glutamate neurons play an important role in sleep is well established, but there are some conflicting studies about whether these neurons promote sleep or wakefulness [20, 38, 39]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286828 June 15, 2023 9 / 14 PLOS ONERNAi of the ETC extends life span and increases sleep in Drosophila Why does RNAi of components of the ETC in glutamate neurons affect sleep and longevity? It is well established, though counterintuitive, that genetic disruption of the ETC by RNAi does not affect overall ATP levels [12, 25]. Certainly, it is worth exploring what long term effect ETC RNAi has on mitochondrial activity in glutamate neurons and how the RNAi affects neu- ronal activity or age-related loss. One means to address possible changes to glutamate neuron activity is through the use of a heat-activated allele of the dTrpA1 cation channel to over-acti- vate glutamate neurons. Alternatively, the use of a temperature-sensitive shibire allele could inhibit glutamate neuron function and alter life span. Sangston and Hirsh have shown that the TH-GAL4 driver can change fly activity contingent on dTrpA1 expression [40]. It is important to note that our results strongly suggest that the entire electron transport chain is involved in the aging process and sleep rather than just one subunit. RNAi targeting separate components of Complex I and Complex V had quite similar and robust effects. It is therefore unlikely that any one component would have an independent role outside of the ETC. Additionally, it is unlikely that the changes presented in this report are an artifact of RNAi off-targeting. The hairpin DNA sequences used to target CG9172 and CG5389 share no sequence similarity (Vienna Drosophila Resource Center, Vienna, Austria). The role of the nervous system in regulating life span is certainly complex and each neuronal subtype does not contribute equally. The results presented here show that RNAi targeting the elec- tron transport chain in glutamate neurons is sufficient to prolong life span. RNAi of the same elec- tron transport chain components leads to a sharp increase in daytime and nighttime sleep and a specific decrease in nighttime activity. It is possible that aging and sleep are independent biological phenomena, but this report provides another genetic link between these processes. Given that only a small subset of neurons can influence such biologically important behaviors, understanding the precise role of the electron transport chain is of significant importance. Supporting information S1 Fig. RNAi of the electron transport chain in glutamate neurons extends life span in male Drosophila. Males of the indicated GAL4 driver were mated to a white1118 control (open circles) or UAS-CG9172-RNAi (light blue), or UAS-CG5389-RNAi (dark blue) lines. (A, B) Survival curves of RNAi against CG9172 and CG5389 RNAi in D42-GAL4 glutamate neurons show comparable 11% and 17% (p < 0.0001, log rank test) life span extension. (C, D) In VGlut-GAL4 glutamate neurons, RNAi of CG9172 and CG5389 leads to an extension of 28% and 26%, respectively (p < 0.0001). (E, F) RNAi of CG9172 and CG5389 in ChAT-GAL4 ace- tylcholine neurons extends life span by 2% and 5%, respectively (p = 0.81, p = 54). (G, H) Acti- vating RNAi against CG9172 and CG5389 to the set of non-overlapping set of glutamate neurons (D42-GAL4; VGlut-GAL80) minimizes life span extension to 2%, respectively (p = 0.91, p = 0.44). (TIF) S2 Fig. Controls for the UAS RNAi lines slightly extend life span in female Drosophila. Females of the indicated UAS RNAi lines were mated to a white1118 control (closed circles) and the white1118 control line was used for comparison (open circles). (A) Survival curve of outcrossed UAS-CG9172-RNAi compared to the white1118 control shows 6% (p < 0.0001, log rank test) life span extension. (B) Survival curve of outcrossed UAS-CG5389-RNAi compared to the white1118 control shows 7% (p < 0.0001, log rank test) life span extension. (TIF) S3 Fig. Controls for the UAS RNAi lines show similar average activity and sleep in 10 day old males. Adult male flies of either the D42-GAL4 control line (white) and unactivated RNAi PLOS ONE \| https://doi.org/10.1371/journal.pone.0286828 June 15, 2023 10 / 14 PLOS ONERNAi of the ETC extends life span and increases sleep in Drosophila to CG9172 (light blue) or CG5389 (dark blue) lines were tested for their activity and sleep. Data are presented as means +/- SEM and compared using one-way ANOVA with a Tukey’s post-hoc test. (A-C) No significant differences were observed in average activity during a 24 hour period (p = 0.89, p = 0.75), the dark phase (p = 0.57, p = 0.29), or the light phase (p = 0.92, p = 0.87) to unactivated CG9172 and CG5389 flies, respectively, in a 24 hour LD cycle. (D-F) No significant differences were observed in total sleep during a 24 hour period (p = 0.96, p = 0.68), the dark phase (p = 0.99, p = 0.17), or the light phase (p = 0.71, p = 0.99) to unactivated CG9172 and CG5389 flies, respectively, in a 24 hour LD cycle. (TIF) S4 Fig. Controls for the UAS RNAi lines have no significant effect on female fertility and starvation response. 10 day old females with the D42-GAL4 driver (open circles) were com- pared to females with unactivated RNAi to CG9172 (light blue) or CG5389 (dark blue) lines. (A, B) The unactivated RNAi line to CG9172 (p = 0.55, two way ANOVA) and CG5389 (p = 0.61) does not affect female fertility. (C, D). Starvation in females with unactivated RNAi to the electron transport chain genes had no effect on overall survival (CG9172, no change, p = 0.64; CG5389, p = 0.05). (TIF) S1 Table. Descriptive statistics of male life span in neuron-specific RNAi of the electron transport chain. (XLSX) S2 Table. Descriptive statistics of male motor activity in glutamate neuron (D42) specific RNAi of the electron transport chain. (XLSX) S3 Table. Descriptive statistics of male sleep in glutamate neuron (D42) specific RNAi of the electron transport chain. (XLSX) S4 Table. Raw data for female life spans. (XLSX) S5 Table. Raw data for male life spans. (XLSX) S6 Table. Raw data for male activity. (XLSX) S7 Table. Raw data for male sleep. (XLSX) S8 Table. Raw data for female fertility. (XLSX) S9 Table. Raw data for female starvation. (XLSX) Acknowledgments The authors would like to thank Eli Wenger and Derek Harnish for their assistance with the life span experiments and Jay Hirsh (University of Virginia) for the use of his Trikinetics PLOS ONE \| https://doi.org/10.1371/journal.pone.0286828 June 15, 2023 11 / 14 PLOS ONERNAi of the ETC extends life span and increases sleep in Drosophila activity monitors. Stocks obtained from the Bloomington Drosophila Stock Center (NIH P40OD018537) were used in this study. Author Contributions Conceptualization: Jeffrey M. Copeland. Funding acquisition: Jeffrey M. Copeland. Investigation: Kevin Sungu, Jeffrey M. Copeland. Supervision: Jeffrey M. Copeland. Writing – original draft: Jessie E. Landis, Hannah Sipe, Jeffrey M. Copeland. Writing – review & editing: Jessie E. Landis, Kevin Sungu, Jeffrey M. Copeland. References 1. Jagust W. Vulnerable Neural Systems and the Borderland of Brain Aging and Neurodegeneration. Neu- ron. 2013 Jan; 77(2):219–34. https://doi.org/10.1016/j.neuron.2013.01.002 PMID: 23352159 2. Piper MDW, Partridge L. Drosophila as a model for ageing. Biochimica et Biophysica Acta (BBA)— Molecular Basis of Disease. 2018 Sep; 1864(9):2707–17. 3. Wang Y, Hekimi S. Mitochondrial dysfunction and longevity in animals: Untangling the knot. Science. 2015 Dec 4; 350(6265):1204–7. https://doi.org/10.1126/science.aac4357 PMID: 26785479 4. Sanz A, Soikkeli M, Portero-Otı´n M, Wilson A, Kemppainen E, McIlroy G, et al. Expression of the yeast NADH dehydrogenase Ndi1 in Drosophila confers increased lifespan independently of dietary restric- tion. Proc Natl Acad Sci USA. 2010 May 18; 107(20):9105–10. 5. Hur JH, Bahadorani S, Graniel J, Koehler CL, Ulgherait M, Rera M, et al. Increased longevity mediated by yeast NDI1 expression in Drosophila intestinal stem and progenitor cells. Aging. 2013 Sep 6; 5 (9):662–81. https://doi.org/10.18632/aging.100595 PMID: 24038661 6. Yun J, Finkel T. Mitohormesis. Cell Metabolism. 2014 May; 19(5):757–66. https://doi.org/10.1016/j. cmet.2014.01.011 PMID: 24561260 7. Lennicke C, Cocheme´ HM. Redox signalling and ageing: insights from Drosophila. Biochemical Society Transactions. 2020 Apr 29; 48(2):367–77. 8. Scialò F, Sriram A, Ferna´ ndez-Ayala D, Gubina N, Lõhmus M, Nelson G, et al. Mitochondrial ROS Pro- duced via Reverse Electron Transport Extend Animal Lifespan. Cell Metabolism. 2016 Apr; 23(4):725– 34. https://doi.org/10.1016/j.cmet.2016.03.009 PMID: 27076081 9. Mattson MP, Arumugam TV. Hallmarks of Brain Aging: Adaptive and Pathological Modification by Meta- bolic States. Cell Metabolism. 2018 Jun; 27(6):1176–99. https://doi.org/10.1016/j.cmet.2018.05.011 PMID: 29874566 10. Rana A, Oliveira MP, Khamoui AV, Aparicio R, Rera M, Rossiter HB, et al. Promoting Drp1-mediated mitochondrial fission in midlife prolongs healthy lifespan of Drosophila melanogaster. Nat Commun. 2017 Dec; 8(1):448. https://doi.org/10.1038/s41467-017-00525-4 PMID: 28878259 11. Schmid E. Neuronal induction of BNIP3-mediated mitophagy slows systemic aging in Drosophila.: 28. https://doi.org/10.1038/s43587-022-00214-y PMID: 36213625 12. Copeland JM, Cho J, Lo T, Hur JH, Bahadorani S, Arabyan T, et al. Extension of Drosophila Life Span by RNAi of the Mitochondrial Respiratory Chain. Current Biology. 2009 Oct; 19(19):1591–8. https://doi. org/10.1016/j.cub.2009.08.016 PMID: 19747824 13. Durieux J, Wolff S, Dillin A. The Cell-Non-Autonomous Nature of Electron Transport Chain-Mediated Longevity. Cell. 2011 Jan; 144(1):79–91. https://doi.org/10.1016/j.cell.2010.12.016 PMID: 21215371 14. Keppley LJW, Nafziger AJ, Liu YT, Hirsh J, Copeland JM. RNAi targeting of the respiratory chain affects Drosophila life span depending on neuronal subtype. BIOS. 2018 Jun 1; 89(2):35. 15. Dillin A, Hsu AL, Arantes-Oliveira N, Lehrer-Graiwer J, Hsin H, Fraser AG, et al. Rates of Behavior and Aging Specified by Mitochondrial Function During Development. Science. 2002 Dec 20; 298 (5602):2398–401. https://doi.org/10.1126/science.1077780 PMID: 12471266 16. Rybina OY, Sarantseva SV, Veselkina ER, Bolschakova OI, Symonenko AV, Krementsova AV, et al. Tissue-specific transcription of the neuronal gene Lim3 affects Drosophila melanogaster lifespan and locomotion. Biogerontology. 2017 Oct; 18(5):739–57. https://doi.org/10.1007/s10522-017-9704-x PMID: 28466185 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286828 June 15, 2023 12 / 14 PLOS ONERNAi of the ETC extends life span and increases sleep in Drosophila 17. Koh K, Evans JM, Hendricks JC, Sehgal A. A Drosophila model for age-associated changes in sleep: wake cycles. Proc Natl Acad Sci USA. 2006 Sep 12; 103(37):13843–7. 18. Sehgal A, Mignot E. Genetics of Sleep and Sleep Disorders. Cell. 2011 Jul; 146(2):194–207. https://doi. org/10.1016/j.cell.2011.07.004 PMID: 21784243 19. Guo F, Yu J, Jung HJ, Abruzzi KC, Luo W, Griffith LC, et al. Circadian neuron feedback controls the Dro- sophila sleep–activity profile. Nature. 2016 Aug; 536(7616):292–7. https://doi.org/10.1038/nature19097 PMID: 27479324 20. Ly S, Pack AI, Naidoo N. The neurobiological basis of sleep: Insights from Drosophila. Neuroscience & Biobehavioral Reviews. 2018 Apr; 87:67–86. https://doi.org/10.1016/j.neubiorev.2018.01.015 PMID: 29391183 21. Ni JD, Gurav AS, Liu W, Ogunmowo TH, Hackbart H, Elsheikh A, et al. Differential regulation of the Dro- sophila sleep homeostat by circadian and arousal inputs. eLife. 2019 Feb 5; 8:e40487. https://doi.org/ 10.7554/eLife.40487 PMID: 30719975 22. Dietzl G, Chen D, Schnorrer F, Su KC, Barinova Y, Fellner M, et al. A genome-wide transgenic RNAi library for conditional gene inactivation in Drosophila. Nature. 2007 Jul; 448(7150):151–6. https://doi. org/10.1038/nature05954 PMID: 17625558 23. Cichewicz K, Hirsh J. ShinyR-DAM: a program analyzing Drosophila activity, sleep and circadian rhythms. Commun Biol. 2018 Dec; 1(1):25. https://doi.org/10.1038/s42003-018-0031-9 PMID: 29911688 24. Partridge L, Gems D, Withers DJ. Sex and Death: What Is the Connection? Cell. 2005 Feb; 120 (4):461–72. https://doi.org/10.1016/j.cell.2005.01.026 PMID: 15734679 25. Xu C, Hwang W, Jeong DE, Ryu Y, Ha CM, Lee SJV, et al. Genetic inhibition of an ATP synthase sub- unit extends lifespan in C. elegans. Sci Rep. 2018 Dec; 8(1):14836. https://doi.org/10.1038/s41598- 018-32025-w PMID: 30287841 26. Bahadorani S, Cho J, Lo T, Contreras H, Lawal HO, Krantz DE, et al. Neuronal expression of a single- subunit yeast NADH-ubiquinone oxidoreductase (Ndi1) extends Drosophila lifespan. Aging Cell. 2010 Apr; 9(2):191–202. 27. Parkes TL, Elia AJ, Dickinson D, Hilliker AJ, Phillips JP, Boulianne GL. Extension of Drosophila lifespan by overexpression of human SOD1 in motorneurons. Nat Genet. 1998 Jun; 19(2):171–4. https://doi.org/ 10.1038/534 PMID: 9620775 28. Mahr A, Aberle H. The expression pattern of the Drosophila vesicular glutamate transporter: A marker protein for motoneurons and glutamatergic centers in the brain. Gene Expression Patterns. 2006 Mar; 6 (3):299–309. https://doi.org/10.1016/j.modgep.2005.07.006 PMID: 16378756 29. Rera M, Bahadorani S, Cho J, Koehler CL, Ulgherait M, Hur JH, et al. Modulation of Longevity and Tis- sue Homeostasis by the Drosophila PGC-1 Homolog. Cell Metabolism. 2011 Nov; 14(5):623–34. https://doi.org/10.1016/j.cmet.2011.09.013 PMID: 22055505 30. Clark RI, Salazar A, Yamada R, Fitz-Gibbon S, Morselli M, Alcaraz J, et al. Distinct Shifts in Microbiota Composition during Drosophila Aging Impair Intestinal Function and Drive Mortality. Cell Reports. 2015 Sep; 12(10):1656–67. https://doi.org/10.1016/j.celrep.2015.08.004 PMID: 26321641 31. Schinaman JM, Rana A, Ja WW, Clark RI, Walker DW. Rapamycin modulates tissue aging and lifespan independently of the gut microbiota in Drosophila. Sci Rep. 2019 Dec; 9(1):7824. https://doi.org/10. 1038/s41598-019-44106-5 PMID: 31127145 32. Ulgherait M, Rana A, Rera M, Graniel J, Walker DW. AMPK Modulates Tissue and Organismal Aging in a Non-Cell-Autonomous Manner. Cell Reports. 2014 Sep; 8(6):1767–80. https://doi.org/10.1016/j. celrep.2014.08.006 PMID: 25199830 33. Cognigni P, Bailey AP, Miguel-Aliaga I. Enteric Neurons and Systemic Signals Couple Nutritional and Reproductive Status with Intestinal Homeostasis. Cell Metabolism. 2011 Jan; 13(1):92–104. https://doi. org/10.1016/j.cmet.2010.12.010 PMID: 21195352 34. Zhang W, Yan Z, Li B, Jan LY, Jan YN. Identification of motor neurons and a mechanosensitive sensory neuron in the defecation circuitry of Drosophila larvae. eLife. 2014 Oct 30; 3:e03293. https://doi.org/10. 7554/eLife.03293 PMID: 25358089 35. Kempf A, Song SM, Talbot CB, Miesenbo¨ ck G. A potassium channel β-subunit couples mitochondrial electron transport to sleep. Nature. 2019 Apr; 568(7751):230–4. 36. Fogle KJ, Mobini CL, Paseos AS, Palladino MJ. Sleep and circadian defects in a Drosophila model of mitochondrial encephalomyopathy. Neurobiology of Sleep and Circadian Rhythms. 2019 Jan; 6:44–52. https://doi.org/10.1016/j.nbscr.2019.01.003 PMID: 30868108 37. Bushey D, Hughes KA, Tononi G, Cirelli C. RSelseeaerchpa,rtiacleging, and lifespan in Drosophila. 2010;18. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286828 June 15, 2023 13 / 14 PLOS ONERNAi of the ETC extends life span and increases sleep in Drosophila 38. Robinson JE, Paluch J, Dickman DK, Joiner WJ. ADAR-mediated RNA editing suppresses sleep by act- ing as a brake on glutamatergic synaptic plasticity. Nat Commun. 2016 Apr; 7(1):10512. https://doi.org/ 10.1038/ncomms10512 PMID: 26813350 39. Zimmerman JE, Chan MT, Lenz OT, Keenan BT, Maislin G, Pack AI. Glutamate Is a Wake-Active Neu- rotransmitter in Drosophila melanogaster. Sleep [Internet]. 2017 Feb 1 [cited 2022 May 24]; 40(2). Avail- able from: https://academic.oup.com/sleep/article/doi/10.1093/sleep/zsw046/2667755 PMID: 28364503 40. Sangston R, Hirsh J. An inexpensive air stream temperature controller and its use to facilitate tempera- ture-controlled behavior in Drosophila. BioTechniques. 2019 Mar; 66(3):159–61. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286828 June 15, 2023 14 / 14 PLOS ONE
10.1371_journal.pone.0280975	RESEARCH ARTICLE Biochemical characterization of a GDP- mannose transporter from Chaetomium thermophilum Gowtham Thambra Rajan PremageethaID Sucharita BoseID Samuel GrandfieldID Subramanian RamaswamyID 4, Vinod NayakID 1,2* 2, Lavanyaa Manjunath2, Deepthi Joseph2, Aviv PazID 4, 2, Luis M. Bredeston5, Jeff Abramson4, 1,2,3, KanagaVijayan Dhanabalan1,2, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 1 Biological Sciences, Purdue University, West Lafayette, Indiana, United States of America, 2 Institute for Stem Cell Science and Regenerative Medicine, Bengaluru, Karnataka, India, 3 Manipal Academy of Higher Education, Manipal, Karnataka, India, 4 Department of Physiology, David Geffen School of Medicine at UCLA, Los Angeles, CA, United States of America, 5 Departamento de Quı´mica Biolo´gica-IQUIFIB, Facultad de Farmacia y Bioquı´mica, Universidad de Buenos Aires-CONICET, Ciudad Auto´ noma de Buenos Aires, Junı´n, Argentina OPEN ACCESS Citation: Premageetha GTR, Dhanabalan K, Bose S, Manjunath L, Joseph D, Paz A, et al. (2023) Biochemical characterization of a GDP-mannose transporter from Chaetomium thermophilum. PLoS ONE 18(4): e0280975. https://doi.org/ 10.1371/journal.pone.0280975 Editor: Michael Massiah, George Washington University, UNITED STATES Received: January 11, 2023 Accepted: April 4, 2023 Published: April 20, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0280975 Copyright: © 2023 Premageetha et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. * subram68@purdue.edu Abstract Nucleotide Sugar Transporters (NSTs) belong to the SLC35 family (human solute carrier) of membrane transport proteins and are crucial components of the glycosylation machinery. NSTs are localized in the ER and Golgi apparatus membranes, where they accumulate nucleotide sugars from the cytosol for subsequent polysaccharide biosynthesis. Loss of NST function impacts the glycosylation of cell surface molecules. Mutations in NSTs cause several developmental disorders, immune disorders, and increased susceptibility to infec- tion. Atomic resolution structures of three NSTs have provided a blueprint for a detailed molecular interpretation of their biochemical properties. In this work, we have identified, cloned, and expressed 18 members of the SLC35 family from various eukaryotic organisms in Saccharomyces cerevisiae. Out of 18 clones, we determined Vrg4 from Chaetomium thermophilum (CtVrg4) is a GDP-mannose transporter with an enhanced melting point tem- perature (Tm) of 56.9˚C, which increases with the addition of substrates, GMP and GDP- mannose. In addition, we report—for the first time—that the CtVrg4 shows an affinity to bind to phosphatidylinositol lipids. Introduction Glycosylation is the process that adds glycans to lipids and proteins. Most of these glycosyla- tion reactions occur in the lumen of the endoplasmic reticulum (ER) and Golgi compartments. The building blocks for glycan biosynthesis are nucleotide sugars (NS), which function as sub- strates for glycosyltransferases to append sugar residues onto glycoproteins or glycolipids. In general, NS are synthesized in the cytosol, except CMP-sialic acid [1], and transported across the ER/Golgi membrane by Nucleotide Sugar Transporters (NST). NSTs function as PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 1 / 15 PLOS ONEFunding: RS thanks support from DBT-B-life grant, Grant/Award Number: BT/PR5081/INF/156/2012, DBT-Indo Swedish Grant, Grant/Award Number: BT/IN/SWEDEN/06/SR/2017-18, ESRF Access Program of RCB, Grant/Award Number: BT/INF/22/ SP22660/2017. AP and JA were supported by grant 5R35GM135175-03 from the National Institute of General Medical Sciences. Scientific- Technological Cooperation Program MINCyT- Argentina and DST-India (Grant Award Number IN/ 14/09 to LMB and RS). RS thanks support from SERB, India for Grant/Award Number EMR/2016/ 001825. KV thanks support from SERB, India, for a National post-doctoral fellowship. The funders had no role in study design, data collection, analysis, publication decision, or manuscript preparation. The authors declare that they do not have any competing interests. Competing interests: The authors have declared that no competing interests exist. GDP-mannose transporter from C. thermophilum antiporters where they transport nucleotide sugars into the lumen of ER/Golgi in exchange for nucleoside mono/di-phosphate (NMP/NDP) back to the cytosol for regeneration [2]. NSTs belong to the solute carrier SLC35 family of membrane transporters. This family is subdivided into seven subfamilies (SLC35A−G) that are further delineated by the specificity of the sugars they transport [3]. Humans have nine sugars—glucose (Glc), galactose (Gal), N-ace- tyl glucose (GlcNAc), N-acetyl galactose (GalNAc), glucuronic acid (GlcA), xylose (Xyl), man- nose (Man), and fucose (Fuc)—conjugated to either GDP or UDP nucleotides. CMP-sialic acid is the lone monosaccharide available as a nucleotide monophosphate [4]. Although GDP- mannose is a naturally occurring NS in humans, no NST that transports it is found. Hence, it provides a unique opportunity to target GDP-mannose transporters for fighting fungal infec- tions in humans where mannose is the most abundant sugar of the fungal cell wall, which directly supports the integrity of the cellSince NSTs serve as the primary transporters of NS, their loss of function has several consequences for human health and disease, resulting in Con- genital Disorders of Glycosylation (CDG). Two well-documented autosomal recessive disor- ders linked to NSTs are leukocyte adhesion deficiency syndrome II [5] and Schneckenbecken dysplasia [6, 7], which results from a loss of function in the GDP-fucose (SLC35A1) and UDP- sugar transporters (SLC35D1) respectively. Additionally, NSTs have been linked to develop- mental disorders in invertebrates [8, 9] and pathogenicity and survival of lower eukaryotes [10]. Thus, a detailed structure and functional analysis are required. After more than four decades of research on NSTs, the atomic resolution structure of the GDP-mannose transporter from Saccharomyces cerevisiae was determined in 2017 [11, 12]. More recently, NST structures of the maize CMP-sialic acid transporter [13] and the mouse CMP-Sialic acid transporter [14] have been determined. Despite their functional and sequence disparity, these NST structures reveal some common salient features. The crystal structures reveal that NSTs comprise ten transmembrane (TM) alpha helices where TM 1−5 is related to TM 6−10 via a pseudo twofold axis. As seen with many transporters, the inverted repeats share structural homology with little to no sequence similarities. To date, all structures of NSTs reside in an outward facing conformation (i.e., opening to lumen) where both substrates (nucleotide sugars and the corresponding NMP) bind to NST in a similar manner. Lipids play a crucial role in altering NST’s function, stability, conformation dynamics, and oligomeric state [15], yet no lipid-binding site(s) have been structurally resolved. A key aspect of NST’s function is its interactions with lipids. Vrg4 from S. cerevisiae prefers short-chain lip- ids for its function [11]. Despite the similar structural architecture of the NSTs, it remains unclear the role lipids play in augmenting NSTs function and how minor changes in amino acid sequences correspond to sugar specificity. In this manuscript, we characterize a GDP- mannose transporter from Chaetomium thermophilum (CtVrg4) and screened several lipids for specific protein-lipids interaction. We show CtVrg4 prefers phosphatidylinositol species such as phosphatidylinositol-(3)-phosphate (PI3P), phosphatidylinositol-(4)-phosphate (PI4P), and phosphatidylinositol-(5)-phosphate (PI5P). Results NSTs have proven to be difficult to express, purify, and structurally resolve. To overcome these limitations, we adopted the ‘funnel approach’ initially proposed by Lewinson et al., 2008 [16]. We screened 18 NSTs, from different organisms, in an effort to find the ones more amenable to crystallization and characterization. Of these 18 constructs, we selected Vrg4 from C. ther- mophilum based on its expression, detergent extraction, and stability (S1 Fig and S1 Table) as a crystallization target. During the course of our work, the structure of the GDP-mannose trans- porter from S. cerevisiae (ScVRG4) was determined, causing us to refocus our priorities toward PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 2 / 15 PLOS ONEGDP-mannose transporter from C. thermophilum Fig 1. Sequence alignment of C. thermophilum CtVrg4, S. cerevisiae ScVrg4 and other GDP-mannose transporter homologs. Clustal Omega [17] was used for multiple sequence alignment of selected GDP-mannose transporters. Identical residues are highlighted in red, and highly conserved residues (>0.7) are highlighted in blue boxes. The nucleotide-binding and sugar recognition motifs are highlighted with star and closed circles, respectively, at the bottom of the alignment. The positions of the transmembrane domains are indicated and colored as blue and red, corresponding to ScVrg4 and CtVrg4 structures, respectively. https://doi.org/10.1371/journal.pone.0280975.g001 a detailed biochemical characterization of CtVrg4 [11]. The amino acid sequences of CtVrg4 and ScVrg4 are 53.6% identical (S2 Table), and both have the characteristic FYNN and GALNK GDP-mannose binding motifs (Fig 1). Due to these similarities, we generated a homology model for identifying structural components of CtVrg4 function. Chymotrypsin-cleaved CtVrg4 (cCtVrg4) Initial crystallization trials were performed using a mosquito crystallization robot where puri- fied CtVrg4 was equally mixed with 576 commercial crystallization screening conditions. Ini- tial crystals were identified in 5 conditions and optimized by varying the contents of the crystallization components. Of these 5 conditions, the optimized condition of 0.07M sodium citrate pH 4.8, 75mM sodium fluoride, and 25% PEG300 yielded crystal that diffracted to 3.8Å. Unfortunately, these crystals proved difficult to reproduce and took almost a month for crystal PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 3 / 15 PLOS ONEGDP-mannose transporter from C. thermophilum Fig 2. Chymotrypsin cleaved CtVrg4 purification, crystallization, and diffraction pattern. (a) Size exclusion profile of chymotrypsin cleaved CtVrg4 compared with full-length CtVrg4 protein. Inset shows SDS-PAGE of chymotrypsin cleaved CtVrg4 (lane 1) and full-length protein (lane 3), markers in lane 4. (b) CtVrg4 crystal indicated by the arrow. (c) Diffraction pattern of CtVrg4 crystal. https://doi.org/10.1371/journal.pone.0280975.g002 formation leading to difficulties in obtaining more detailed characterization of CtVrg4 crystals. Further analysis of the CtVrg4 crystals by SDS-PAGE showed a significantly reduced molecular weight (~25 kDa) when compared to full-length purified protein, which has a molecular weight of 37 kDa (Fig 2A, gel insert) We speculated that CtVrg4 suffered from pro- teolysis and attempted to mimic this modification through incubation with chymotrypsin (cCtVrg4). After chymotrypsin induced proteolysis, cCtVrg4 elutes as a monodisperse peak at 75.44mL from the size exclusion column, where the full-length protein elutes at 74ml (Fig 2A). To test the functionality of cCtVrg4, we reconstituted the protein into liposomes made of Yeast Polar Lipid (YPL) and carried out transport assay. The proteoliposome transport assay showed that the cleaved protein is functional with a Km value of 32.07μM for GDP-mannose (S2 Fig). Additionally, cCtVrg4 crystalized reproducibly but did not diffract to high resolution for structure determination (Fig 2B and 2C). AlphaFold2 model of CtVrg4 We used AlphaFold2 to generate a model of the CtVrg4 [18] to aid in biochemical interpreta- tion and to better assist structural comparisons with known NST structures [11, 13, 14]. The structural features are very similar and the confidence values of the prediction in the conserved regions are very high (S3 Fig). The homology model shows the anticipated ten transmembrane helical structure corre- sponding to the NSTs fold. Additionally, the model predicted a 34 amino acids long stretch of disordered region and a short helix at the N-terminal and similarly a 17 amino acids long dis- ordered region at the C-terminal of CtVrg4. This elongated stretch of disordered region is not seen in other NSTs (Fig 3A). Superimposition of the Alphafold2 model onto the ScVrg4 apo structure (PDB-5OGE Chain A) resulted in a good alignment of all the transmembrane helices with an RMSD value of 1.6Å (299 Cα atom pairs), barring unstructured and short helices at the terminus (indicated by arrowhead in Fig 3B). PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 4 / 15 PLOS ONEGDP-mannose transporter from C. thermophilum Fig 3. AlphaFold2 model of CtVrg4. (a), AlphaFold model of CtVrg4 with transmembrane helix colored from blue (N-terminal) to red (C-terminal) and numbered from 1 to 10. (b), Structural comparison of CtVrg4 model (Cyan) and ScVrg4 apo crystal structure (Maroon). Arrowhead indicates the unaligned structural element of the AlphaFold2 model with respect to the ScVrg4 structure. https://doi.org/10.1371/journal.pone.0280975.g003 Complementation assay of CtVrg4 The functionality of CtVrg4 was assessed by complementation assay using a hygromycin B- sensitive yeast strain, NDY5, which lacks the Vrg4-2 gene [19]. Yeast that lacks or with an inhibited Vrg4 gene show defects in the glycosylation and the outer cell membrane becoming sensitive to hygromycin. This assay is a good proxy for the measurement of integrity of glycan structure. The assay shows that both CtVrg4 and ScVrg4 rescue NDY5 in hygromycin (Fig 4A). Additionally, based on the AlphaFold2 model CtVrg4Δ17(1−368 amino acids)—a truncation of 17 amino acids from the C-terminal end, which is a disordered loop region flanking the transmembrane helix 10 is able to rescue NDY5 in hygromycin. This indicates that the unstructured c-terminus is not necessary for function (Fig 4A). Based on the ScVrg4 crystal structure (PDB-5OGK Chain A) published by Parker and Newstead [11], four residues—N220 and N221 form hydrogen bonds with the guanine moiety and Y28 and Y281 coordinate the ribose sugar (Fig 4B)—were identified for further functional characterization. In CtVrg4, alanine substitution of these amino acids shows either no (Y54A, Y310A) or only partial rescue (N245A, N246A) in the NDY5 assay (Fig 4A). More conservative substitution of Y310F and N246S also had limited ability for rescue. This complementation assay result agrees well with the in vitro transport assays of ScVrg4 mutants [11, 12] and estab- lishes CtVrg4 as a GDP-mannose transporter. Transport kinetics of CtVrg4 Based on the ScVrg4 structure, the hydroxyl moiety of Y310 coordinates the ribose component of NS. To resolve the significance of the hydroxyl moiety in GDP-mannose binding and trans- port kinetics, we generated a protein with Y310F mutation. Proteins with CtVrg4 WT and Y310F mutation were reconstituted into the liposomes. The GDP-mannose IC50 for WT CtVrg4 is 25.45μM and 47.05μM for Y310F (Fig 5A). To further characterize CtVrg4 WT and Y310F mutant, a thermal shift assay was performed to determine the dissociation constant (Kd) in the presence of their substrates, GMP and GDP-mannose [20]. The Kd values in Y310F mutant are similar for both substrates yielding a Kd of 139.2μM for GMP and 129.95μM for GDP-mannose. However, the WT protein has a Kd PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 5 / 15 PLOS ONEGDP-mannose transporter from C. thermophilum Fig 4. Functional Characterization of CtVrg4 and its mutants by a Hygromycin B based in-vivo assay: (a) Complementation assay of NDY5 by CtVrg4 WT and various functional mutants. Transformed yeast cells were serially diluted and spotted on synthetic agar media in the presence and absence of 100μg/mL of Hygromycin B. (b), Close-up view of ScVrg4 crystal structure. Equivalent amino acids that were chosen to be mutated in CtVgr4 are highlighted and shown in magenta stick color representation. GDP-Mannose is shown in yellow. Transmembrane helixes are numbered. https://doi.org/10.1371/journal.pone.0280975.g004 of 143.2μM for GMP but is reduced to 74.26μM for GDP-mannose (Fig 5B). Taken together, these suggests that the hydroxyl group is not critical for binding or transport of GDP-mannose by CtVrg4. We discuss later our idea that the importance of Y310 may come from its role in positioning Y54, which is critical for binding to the ribose sugar. Lipid binding activity and kinetics of CtVrg4 WT and CtVrg4Δ31 construct Earlier studies showed that 1,2-dimyristoyl-sn-glycerol-3-phosphocholine (DMPC), is essen- tial for ScVrg4 function [11]. We therefore screened several lipids for specific interaction with CtVrg4 wildtype protein using an established lipid blot assay [21]. In short, specific lipids are immobilized on strips (100 pmol) and subsequently bathed with purified protein to determine if there are protein/lipid interactions. After the incubation period, the strips are washed to remove nonspecific binding and the presence of protein is detected using antibody that recog- nizes the protein’s poly-histidine affinity tag fused to the N-terminus. This assay revealed that PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 6 / 15 PLOS ONEGDP-mannose transporter from C. thermophilum Fig 5. Proteoliposome and thermal shift assay of CtVrg4. (a), Representative GDP-mannose IC50 curve for CtVrg4 WT and CtVrg4 Y310F. IC50 values are shown at the top of the graph. (b), Thermal shift assay for CtVrg4 WT and CtVrg4 Y310F with varying concentrations of GMP and GDP-mannose. Kd values are shown at the bottom of the graph. Calculated IC50 and Kd values are the means of two independent biological repeats (each done in technical duplicate or triplicate), errors are indicated as S.D. https://doi.org/10.1371/journal.pone.0280975.g005 CtVrg4 specifically binds to three phosphatidylinositol lipids—phosphatidylinositol-(3)-phos- phate (PI3P), phosphatidylinositol-(4)-phosphate (PI4P), and phosphatidylinositol-(5)-phos- phate (PI5P) (Fig 6A). These results suggest that CtVrg4 is possibly present in the Golgi membrane. MD simulation by Parker et al. 2019 [12] predicted two lipid binding sites in the ScVrg4 protein. The first site is at the dimer interface formed by two transmembrane helices, TM5 and TM10, and the second is at the shallow groove between TM1, TM9, and TM10. Identifying TM10 as an integral component of lipid binding, we probed its’ role by deleting the last 31 amino acid segments from the C-terminus, including the predicted Golgi retrieval signal (K355VRQKA), which leaves most of the TM10 buried in the membrane (S3 Fig). This signal harbors several positively charged amino acids that could potentially bind to negatively charged phosphatidylinositol species. The new construct (CtVrg4Δ31) still showed binding for the same set of lipids as the WT protein, indicating the possibility of an additional lipid bind- ing site. This surprising result led to further characterization of the protein with CtVrg4Δ31 muta- tion. Both the proteoliposome and thermal shift assays for the truncated protein construct showed similar kinetics as the WT protein (Fig 6B and 6C). GDP-mannose IC50 was found to be 48μM, and the Kd for GMP and GDP-mannose was 63.2μM and 58.5μM, respectively. These results suggest that the C-terminal 31 residues are not critical for transport. Discussion Our efforts to study the structure-function relationship of NST led to the identification of a thermostable GDP-mannose transporter, CtVrg4. CtVrg4 shares 53.6% sequence identity with PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 7 / 15 PLOS ONEGDP-mannose transporter from C. thermophilum Fig 6. Lipid binding activity and kinetics of CtVrg4Δ31. (a), Lipid blot. Left–lipid binding activity of CtVrg4 WT and CtVrg4Δ31 protein. Right–legend with lipids present in the corresponding PIP strip spots. (b), Representative GDP-mannose IC50 curve for CtVrg4Δ31. The IC50 value is shown at the top of the graph. (c), Thermal shift assay for CtVrg4Δ31 with varying concentrations of GMP and GDP-mannose. Kd values are shown at the bottom of the graph. Calculated IC50 and Kd values are mean of two independent biological repeats (each done in technical duplicate or triplicate), errors are indicated as S.D. Lipid binding assay was done in at least two biological replicates. https://doi.org/10.1371/journal.pone.0280975.g006 ScVrg4 and maintains the conserved substrate binding motifs, FYNN and GALNK, indicative that CtVrg4 is a GDP-mannose transporter as well. We further confirmed that CtVrg4 is a GDP-mannose transporter utilizing a hygromycin-based complementation assay and substan- tiated our finding by in vitro proteoliposome transport assays. The IC50 value of GDP-man- nose was found to be 25.45μM, which is approximately three times higher than that for ScVrg4 (7.7μM). Our melting point assay in DDM showed CtVgr4 is significantly more stable (56.9˚C) (Fig 7A) than ScVrg4 (37.9˚C). This is not surprising as CtVrg4 is isolated from a thermostable fungus. During crystallization trials, we found that the chymotrypsin cleaved CtVrg4 produced bet- ter diffraction quality crystals than the full-length protein. Proteoliposome transport assay of chymotrypsin cleaved CtVrg4 showed that the cleaved protein is functional with a GDP-man- nose Km value of 32.07 μM, suggesting that limited proteolysis does not affect the core of the protein. The observation that a shorter construct is still functional suggests it might be worth reattempting the structure determination, which we did not pursue further after the structure of ScVrg4 was published. Molecular replacement with the 3.8 Å data did not produce good solutions. Given the similarity in sequence between CtVrg4 and ScVrg4 (53%), and the mod- ern structure prediction tools like AlphaFold2, the predicted structure would be a good model (compared to a 3.8 Å structure). In ScVrg4, Y281 forms a hydrogen bond with the ribose sugar of GDP-mannose and a π- stacking interaction with Y28, which along with S269, further coordinates the ribose sugar. The alanine mutants Y28A, and Y281A of ScVrg4, resulted in the complete abolishment of transport activity [11]. In our complementation assay, we found that Y310F (equivalent to Y281 in ScVrg4) conferred a partial rescue, whereas no complementation was observed in the PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 8 / 15 PLOS ONEGDP-mannose transporter from C. thermophilum Fig 7. Thermal stability analysis of CtVrg4. (a), Bar graph of melting temperature (Tm) of CtVrg4 WT, Y310F, and CtVrg4Δ31. (b), Kd values of GMP/GDP-mannose binding to CtVrg4 WT, Y310F, and CtVrg4Δ31 were estimated by a thermal shift assay. (c), Electrostatic surface representation of the AlphaFold2 model for CtVrg4 highlighting positively charged residues in the Golgi retrieval sequence in TM10. The orientation of CtVrg4 in the lipid bilayer was simulated through the OPM server [22] (d), GMP/GDP-mannose binding to CtVrg4 was estimated by a shift in Tm after the addition of 25 mM GMP/GDP-mannose to WT, Y310F, and CtVrg4Δ31. https://doi.org/10.1371/journal.pone.0280975.g007 case of Y310A, Y54A, and Y54F. The Kd for GDP-mannose almost doubled in the Y310F mutant compared with WT, but the Kd for GMP remained the same (Fig 7B). The IC50 for GDP-mannose also doubled compared with the WT, further validating the decreased binding efficiency of GDP-mannose in the Y310F mutation. These results suggest that Y310 also forms a hydrogen bond with the ribose sugar and helps position the Y54 residue to bind the ribose sugar, whose substitution is lethal for transport activity. It was previously reported that ScVrg4 localizes to the Golgi using its C-terminal Golgi retrieval sequence, which binds to COPI vesicles [23]. CtVrg4 has a Golgi retrieval sequence (K355VRQKA) like ScVrg4 in the cytoplasmic end of TM10. The Golgi retrieval sequence is positioned on membrane boundaries to interact with the membrane lipids (Fig 7C). The Golgi retrieval sequence is rich in positively charged residues, which could be a potential binding site for negatively charged PIP lipids to bind. However, our results show that CtVrg4Δ31 (devoid PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 9 / 15 PLOS ONEGDP-mannose transporter from C. thermophilum of the last 31 amino acids from the C-terminal end, including the Golgi retrieval sequence) has the same binding affinity as the full-length protein—suggesting the Golgi retrieval sequence may not play any role in PIP binding. Further studies are needed to pinpoint the exact location of PIP binding to the transporter and its structural or functional effect on CtVrg4. We report that CtVrg4Δ31 is less thermostable than the WT and Y310F mutant with a 3–4˚C lower melting temperature (Fig 7A). Interestingly, both GMP and GDP-mannose bind to CtVrg4Δ31 and show similar thermal shift values compared with WT and Y310F (Fig 7D). This result suggests that although overall stability is affected by truncation, the ability of the substrate to bind and stabilize the core protein is not affected. Our work reported here suggests that further structure function studies of NSTs are needed in order to understand the mecha- nism of transport and the role of lipids in localization, stability and NS transport. Materials and methods Cloning and expression of CtVrg4 wildtype, mutants, and truncated constructs in a yeast expression system The sequence of CtVrg4 from C. thermophilum (NCBI GenBank XM_006692792.1) was iden- tified through a homology search against the ScVrg4 sequence. CtVrg4 is 385 amino acids long. The gene was cloned and expressed in a modified pDDGFP yeast expression vector with an N-terminal 12 histidine tag followed by a SmaI restriction site for homologous recombina- tion-based gene insertion and a stop codon. All CtVrg4 mutants were generated via site-directed mutagenesis using the PCR method and further confirmed by sequencing. Two C-terminal truncated versions of CtVrg4 were con- structed, CtVrg4Δ17 (constituting amino acids from 1 to 368) and CtVrg4Δ31 (comprising amino acids from 1 to 354). The cloned genes were expressed in S. cerevisiae haploid strain FGY217 (MATa, ura3-52, lys2_201, and pep4). Expression and purification of CtVrg4 The primary cultures of CtVrg4 WT, mutations, and truncations were grown in synthetic media without uracil in 2% glucose. The primary culture was diluted into a secondary culture (1L) to the final OD of 0.2 in 0.1% glucose. The culture was induced with 2% galactose at 0.6 to 0.8 OD to express the protein and further grown for 22–24h before harvesting the cells by cen- trifugation. Cells were resuspended in membrane resuspension buffer (75mM Tris pH 8.0, 150 mM NaCl, and 5% glycerol) and then lysed using a cell disruptor (Constant Systems Ltd) at 28, 32, 36, and 39 kpsi. Membranes were isolated by centrifugation at 200,000 xg for 1.5h. The pro- tocol for protein purification was adapted from Drew et al., 2008 with modifications [24]. The membranes were solubilized in dodecyl β-D-maltopyranoside (DDM, Anatrace) at a 1: 0.2 (w/ w) ratio of the membrane to DDM for 2 h in membrane resuspension buffer along with a pro- tease inhibitor cocktail. For crystallization, the protein was solubilized in 2% Decyl β-D-malto- pyranoside (DM) for two hours in the membrane resuspension buffer. The solubilized membrane was centrifuged at 200,000 xg for 30 minutes, and the supernatant was loaded onto a 5 mL His-Trap FF/HP column (Cytiva). The protein was eluted with 300 mM imidazole. The eluted fractions were desalted using a Hiprep 26/10 desalting column (Cytiva) to remove the imidazole from the buffer. The protein was concentrated using a 50 kDa cut-off Amicon device (Millipore). The concentrated protein was centrifuged at 14,000 xg for 15 mins before injecting onto a Superdex 200 (10/300) size-exclusion column (Cytiva) pre-equilibrated with liposome assay buffer (20 mM HEPES pH 7.4, 50 mM KCl, and 2 mM MgSO4) containing 0.1 mM EDTA and 0.013% DDM. PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 10 / 15 PLOS ONEGDP-mannose transporter from C. thermophilum Chymotrypsin cleavage and crystallization Purified CtVrg4 was subjected to limited proteolysis using chymotrypsin (50:1 ratio) for 2h at 18˚C. The reaction was stopped with 0.1mM phenylmethylsulfonyl fluoride followed by ultra- centrifugation at 150,000 xg for 30 minutes. The chymotrypsin cleaved protein eluted at 75.44ml compared to 74ml for the full-length protein on a Superdex 200 (16/600) column (Cytiva). The cleaved protein’s peak fractions were pooled and concentrated using 50kDa cut- off Amicon (Millipore). Purified chymotrypsin cleaved CtVrg4 (12mg/ml) was mixed in a 4:1 (protein/bicelle) ratio with a 25% (2.8:1) DTPC/CHAPSO bicellar solution for 45 minutes on ice, yielding 9.6 mg/ml of chymotrypsin cleaved CtVrg4 in 5% bicelles. The best crystals grew to a size of 0.07 mm X 0.07 mm X 0.02 mm at 18˚C in 0.07 M sodium citrate pH 4.8, 75 mM sodium fluoride, and 25% PEG300. The crystals were cryoprotected with 30% PEG400 before flash-cooling in liquid nitrogen. X-ray diffraction data were collected at the PROXIMA-1 beamline, SOLEIL synchro- tron source (France), at a wavelength of 0.97857 Å using a 100 K nitrogen stream. The best crystal diffracted to 3.8 Å resolution. Protein reconstitution into liposomes Chymotrypsin cleaved CtVrg4 was reconstituted into yeast polar lipid (YPL, Avanti Polar Lip- ids) liposomes in 20 mM HEPES pH 7.5, 100 mM KCl, as described by Parker et al., 2017 [11]. CtVrg4 wildtype/Y310F/truncates were reconstituted into liposomes using a modified proto- col from Majumdar et al., 2019 [25]. Briefly, YPL were suspended in chloroform, dried using a nitrogen stream, and left in a vacuum desiccator overnight. The lipid film was resuspended at 10 mg/mL in liposome assay buffer, and then bath sonicated to form small multilamellar vesi- cles. Vesicles were extruded for 10 cycles through 400 nm polycarbonate membranes (Avanti Polar Lipids). For reconstitution, purified (CtVrg4 wildtype/Y310F/truncate) protein in DDM (at 8 to 16 mg/mL) was added to the extruded YPL at a final lipid: protein ratio (w/w) of 80:1. Sodium cholate (Anatrace) was added at a concentration of 0.65−0.75% to this mixture and incubated for one hour at room temperature, then for a further 30 minutes on ice. As a control, liposomes without protein were resuspended in the assay buffer containing 0.013% of DDM. The protein-lipid mixture was passed through an 8.3 mL PD10 column (Cytiva) pre-equili- brated with 0.5 mg YPL in liposome assay buffer. A fraction volume of 2.8 mL was collected after the column’s void volume (2.6 mL) and centrifuged at 150,000 xg for 30 minutes. Subse- quently, the pellet containing proteoliposomes was resuspended with liposome assay buffer, flash-frozen in liquid N2, and stored at -80˚C. The protein reconstitution into the lipids was verified by Western blot with anti-His antibody. Transport assay CtVrg4 wildtype/Y310F/truncate proteoliposomes were thawed, and the desired concentration of internal cold substrate (1 mM GDP-mannose) was added and then subjected to six rounds of freeze-thaw in liquid nitrogen to load the proteoliposomes with the substrate. For chymo- trypsin-cleaved CtVrg4, the internal concentration of GDP-mannose ranged from 1 to 1000 μM. Unloaded GDP-mannose was removed by ultracentrifugation at 150,000 xg for 30 minutes. The pellets were resuspended in a cold liposome assay buffer. For a typical 50 μl reac- tion, 10 μl of loaded proteoliposomes containing approximately 4 μg of protein was added to 40 μl of liposome assay buffer containing 0.5 μM [3H]GMP (American Radiolabeled Chemi- cals, Inc). For the chymotrypsin cleaved protein in proteoliposomes, the concentration of [3H] GMP was 0.384 μM. The addition of GMP initiated the exchange reaction. For IC50 value determination, the external competing substrate was varied from 2 μM to 2.5 mM. The PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 11 / 15 PLOS ONEGDP-mannose transporter from C. thermophilum mixture was then incubated at 25˚C for 20 minutes. For chymotrypsin cleaved CtVrg4, the reaction was performed at room temperature and terminated after 10 min. The uptake of the radiolabeled substrate was stopped by adding 800 μl of cold water and rapidly filtering onto 0.22 micron mixed cellulose esters filters (Millipore), which were then washed three times with 2 mL of ice-cold water. A liquid scintillation counter (Perkin Elmer) measured the amount of [3H]GMP transported inside the liposomes. All experiments were done in two biological repeats, each in technical duplicate or triplicate. Kinetic parameters were calculated by nonlin- ear fit using the GraphPad Prism software (GraphPad Software, Inc., San Diego, CA, USA). Hygromycin B-based in vivo assay For in vivo functional characterization, wildtype/mutant/truncated proteins were expressed in the yeast NDY5 strain (MAT ura3–52a, leu2–211, vrg4–2), which is a vrg4Δ mutant. The trans- formants were selected based on the uracil resistance marker. Cells grown overnight were seri- ally diluted and spotted on the synthetic agar media in the presence and absence of 100 μg/mL of Hygromycin B. Protein expression was induced with 2% galactose, and phenotype was observed after three days. For negative control, transformed cells were spotted in 2% glucose and compared with 2% galactose. Lipid blot assay Lipid blot assay was carried out using PIP StripsTM (Echelon Biosciences) following the manu- facturer’s instructions. Briefly, the strips were blocked overnight with PBST buffer (1x PBS with 0.05% Tween-20) containing 5% non-fat dry milk at 4˚C with mild rocking. The strips were then incubated with 50 μg of CtVrg4 WT/CtVrg4Δ31 in PBST for two hours at room temperature. After three washes with PBST, the strips were incubated with conjugated anti- polyHistidine−Peroxidase antibody (Sigma A7058, in a 1:2000 ratio) for one hour at room temperature. After three more washes with PBST, the bound proteins were detected with the ClarityTM Western ECL kit (Bio-Rad Laboratories). Thermal shift assay Following size exclusion chromatography, detergent-solubilized CtVrg4 wildtype/mutant/ truncate (final concentration of 0.5mg/mL) were incubated with 0 mM to 50 mM GMP/GDP- mannose at room temperature for 15 minutes before measuring the Tm using a Tycho NT.6 instrument (NanoTemper Technologies, Germany). All experiments were done in two biologi- cal repeats and three technical repeats. Kinetic parameters were calculated by nonlinear fit using the GraphPad Prism software. AI prediction and superimposition The AlphaFold2 structure prediction feature in UCSF ChimeraX was used to predict the struc- ture of CtVrg4. Superimposition and all protein model figures were generated with UCSF Chi- meraX [26]. ESPript was used to render sequence similarities and secondary structure information onto multiple sequence alignments [27]. Supporting information S1 Fig. A schematic of the funnel approach that was used to screen for crystallizable NSTs. (TIF) S2 Fig. Exchange velocities of internal GDP-man with external [3H]GMP in proteolipo- somes. Values are the means of three independent biological repeats (each done in technical PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 12 / 15 PLOS ONEGDP-mannose transporter from C. thermophilum triplicate). Errors are indicated as SD. Km was calculated by non-linear fit using the GraphPad Prism software. (TIF) S3 Fig. AlphaFold2 model for CtVrg4. Colors represent the predicted confidence value of the structure. The confidence value decreases as the color changes from red to blue. The figure shows that the transmembrane regions are predicted with high confidence–in red. The N-ter- minus region is floppy and is predicted poorly. Note that the C-terminal helix is predicted with intermediate confidence (green). Amino acid Alanine 27 (A27) and the C-terminal resi- due (S385) are labeled for reference. A354, is the last residue in the CtVrg4Δ31 construct. (TIF) S1 Table. NSTs identified for crystallization through homology search. (DOCX) S2 Table. Percentage sequence identity of known and putative GDP-mannose transporters from different species of fungi—alignment carried out using Clustal Omega. (DOCX) S1 Raw image. (PDF) Acknowledgments We thank Dr. Leonard Chavas from SOLEIL Synchrotron (Proxima-1 beamline) for providing the beamtime for data collection. The Tycho experiment was done in the Chemical Genomics Facility at Purdue Institute for Drug Discovery and the NIH-funded Indiana Clinical and Translational Sciences Institute. Author Contributions Conceptualization: KanagaVijayan Dhanabalan, Luis M. Bredeston, Jeff Abramson, Subramanian Ramaswamy. Data curation: Gowtham Thambra Rajan Premageetha, KanagaVijayan Dhanabalan, Jeff Abramson. Formal analysis: Gowtham Thambra Rajan Premageetha, KanagaVijayan Dhanabalan, Sucharita Bose, Lavanyaa Manjunath, Vinod Nayak. Funding acquisition: Luis M. Bredeston, Jeff Abramson, Subramanian Ramaswamy. Investigation: Gowtham Thambra Rajan Premageetha, KanagaVijayan Dhanabalan, Sucharita Bose, Lavanyaa Manjunath, Deepthi Joseph, Aviv Paz, Samuel Grandfield, Vinod Nayak, Luis M. Bredeston, Jeff Abramson, Subramanian Ramaswamy. Methodology: Gowtham Thambra Rajan Premageetha, KanagaVijayan Dhanabalan, Sucharita Bose, Lavanyaa Manjunath, Vinod Nayak. Supervision: Luis M. Bredeston, Jeff Abramson, Subramanian Ramaswamy. Validation: KanagaVijayan Dhanabalan. Writing – original draft: Gowtham Thambra Rajan Premageetha, KanagaVijayan Dhanabalan, Jeff Abramson, Subramanian Ramaswamy. PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 13 / 15 PLOS ONEGDP-mannose transporter from C. thermophilum Writing – review & editing: Gowtham Thambra Rajan Premageetha, KanagaVijayan Dhanabalan, Sucharita Bose, Lavanyaa Manjunath, Deepthi Joseph, Aviv Paz, Vinod Nayak, Luis M. Bredeston, Jeff Abramson, Subramanian Ramaswamy. References 1. Coates SW, Gurney T, Sommers LW, Yeh M, Hirschberg CB. Subcellular localization of sugar nucleo- tide synthetases. J Biol Chem. 1980; 255: 9225–9229. PMID: 6251080 2. Hadley B, Maggioni A, Ashikov A, Day CJ, Haselhorst T, Tiralongo J. Structure and function of nucleo- tide sugar transporters: Current progress. Computational and Structural Biotechnology Journal. 2014; 10: 23–32. https://doi.org/10.1016/j.csbj.2014.05.003 PMID: 25210595 3. Hadley B, Litfin T, Day CJ, Haselhorst T, Zhou Y, Tiralongo J. Nucleotide Sugar Transporter SLC35 Family Structure and Function. Computational and structural biotechnology journal. 2019; 17: 1123– 1134. https://doi.org/10.1016/j.csbj.2019.08.002 PMID: 31462968 4. Mikkola S. Nucleotide Sugars in Chemistry and Biology. Molecules. 2020; 25: 5755. https://doi.org/10. 3390/molecules25235755 PMID: 33291296 5. Liu L, Xu YX, Hirschberg CB. The role of nucleotide sugar transporters in development of eukaryotes. Seminars in Cell and Developmental Biology. 2010; 21: 600–608. https://doi.org/10.1016/j.semcdb. 2010.02.002 PMID: 20144721 6. Hiraoka S, Furuichi T, Nishimura G, Shibata S, Yanagishita M, Rimoin DL, et al. Nucleotide-sugar trans- porter SLC35D1 is critical to chondroitin sulfate synthesis in cartilage and skeletal development in mouse and human. Nat Med. 2007; 13: 1363–1367. https://doi.org/10.1038/nm1655 PMID: 17952091 7. Rautengarten C, Quarrell OW, Stals K, Caswell RC, De Franco E, Baple E, et al. A hypomorphic allele of SLC35D1 results in Schneckenbecken-like dysplasia. Human molecular genetics. 2019 Nov 1; 28 (21):3543–51. https://doi.org/10.1093/hmg/ddz200 PMID: 31423530 8. Cipollo JF, Awad AM, Costello CE, Hirschberg CB. srf-3, a mutant of Caenorhabditis elegans, resistant to bacterial infection and to biofilm binding, is deficient in glycoconjugates. Journal of Biological Chemis- try. 2004; 279: 52893–52903. https://doi.org/10.1074/jbc.M409557200 PMID: 15452127 9. Goto S, Taniguchi M, Muraoka M, Toyoda H, Sado Y, Kawakita M, et al. UDP–sugar transporter impli- cated in glycosylation and processing of Notch. Nat Cell Biol. 2001; 3: 816–822. https://doi.org/10.1038/ ncb0901-816 PMID: 11533661 10. Gaur U, Showalter M, Hickerson S, Dalvi R, Turco SJ, Wilson ME, et al. Leishmania donovani lacking the Golgi GDP-Man transporter LPG2 exhibit attenuated virulence in mammalian hosts. Experimental Parasitology. 2009; 122: 182–191. https://doi.org/10.1016/j.exppara.2009.03.014 PMID: 19328787 11. Parker JL, Newstead S. Structural basis of nucleotide sugar transport across the Golgi membrane. Nature. 2017; 551: 521. https://doi.org/10.1038/nature24464 PMID: 29143814 12. Parker JL, Corey RA, Stansfeld PJ, Newstead S. Structural basis for substrate specificity and regulation of nucleotide sugar transporters in the lipid bilayer. Nat Commun. 2019; 10: 4657. https://doi.org/10. 1038/s41467-019-12673-w PMID: 31604945 13. Nji E, Gulati A, Qureshi AA, Coincon M, Drew D. Structural basis for the delivery of activated sialic acid into Golgi for sialyation. Nature Structural and Molecular Biology. 2019; 26: 415–423. https://doi.org/10. 1038/s41594-019-0225-y PMID: 31133698 14. Ahuja S, Whorton MR. Structural basis for mammalian nucleotide sugar transport. eLife. 2019; 8. https://doi.org/10.7554/eLife.45221 PMID: 30985278 15. Corradi V, Sejdiu BI, Mesa-Galloso H, Abdizadeh H, Noskov SY, Marrink SJ, et al. Emerging Diversity in Lipid-Protein Interactions. Chem Rev. 2019; 119: 5775–5848. https://doi.org/10.1021/acs.chemrev. 8b00451 PMID: 30758191 16. Lewinson O, Lee AT, Rees DC. The Funnel Approach to the Precrystallization Production of Membrane Proteins. Journal of Molecular Biology. 2008; 377: 62–73. https://doi.org/10.1016/j.jmb.2007.12.059 PMID: 18241890 17. Sievers F, Wilm A, Dineen D, Gibson TJ, Karplus K, Li W, et al. Fast, scalable generation of high-quality protein multiple sequence alignments using Clustal Omega. Molecular systems biology. 2011; 7(1):539. Tunyasuvunakool K, Adler J, Wu Z, Green T, Zielinski M, Zˇ ı´dek A, et al. Highly accurate protein struc- ture prediction for the human proteome. Nature. 2021; 596: 590–596. https://doi.org/10.1038/s41586- 021-03828-1 PMID: 34293799 18. 19. Dean N. Yeast glycosylation mutants are sensitive to aminoglycosides. Proc Natl Acad Sci USA. 1995; 92: 1287–1291. https://doi.org/10.1073/pnas.92.5.1287 PMID: 7877969 PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 14 / 15 PLOS ONEGDP-mannose transporter from C. thermophilum 20. Vivoli M, Novak HR, Littlechild JA, Harmer NJ. Determination of protein-ligand interactions using differ- ential scanning fluorimetry. JoVE (Journal of Visualized Experiments). 2014 Sep 13(91):e51809. https://doi.org/10.3791/51809 PMID: 25285605 21. 22. Jin W, Ge WP, Xu J, Cao M, Peng L, Yung W, et al. Lipid binding regulates synaptic targeting of PICK1, AMPA receptor trafficking, and synaptic plasticity. Journal of Neuroscience. 2006 Mar 1; 26(9):2380– 90. https://doi.org/10.1523/JNEUROSCI.3503-05.2006 PMID: 16510715 Lomize MA, Pogozheva ID, Joo H, Mosberg HI, Lomize AL. OPM database and PPM web server: resources for positioning of proteins in membranes. Nucleic acids research. 2012 Jan 1; 40(D1):D370– 6. https://doi.org/10.1093/nar/gkr703 PMID: 21890895 23. Abe M, Noda Y, Adachi H, Yoda K. Localization of GDP-mannose transporter in the Golgi requires retrieval to the endoplasmic reticulum depending on its cytoplasmic tail and coatomer. Journal of Cell Science. 2004; 117: 5687–5696. https://doi.org/10.1242/jcs.01491 PMID: 15494368 24. Drew D, Newstead S, Sonoda Y, Kim H, Von Heijne G, Iwata S. GFP-based optimization scheme for the overexpression and purification of eukaryotic membrane proteins in Saccharomyces cerevisiae. Nature protocols. 2008 May; 3(5):784–98. https://doi.org/10.1038/nprot.2008.44 PMID: 18451787 25. Majumder P, Khare S, Athreya A, Hussain N, Gulati A, Penmatsa A. Dissection of Protonation Sites for Antibacterial Recognition and Transport in QacA, a Multi-Drug Efflux Transporter. Journal of Molecular Biology. 2019; 431: 2163–2179. https://doi.org/10.1016/j.jmb.2019.03.015 PMID: 30910733 26. Pettersen EF, Goddard TD, Huang CC, Meng EC, Couch GS, Croll TI, et al. UCSF ChimeraX: Structure visualization for researchers, educators, and developers. Protein Science. 2021 Jan; 30(1):70–82. https://doi.org/10.1002/pro.3943 PMID: 32881101 27. Robert X, Gouet P. Deciphering key features in protein structures with the new ENDscript server. Nucleic acids research. 2014 Jul 1; 42(W1):W320–4. https://doi.org/10.1093/nar/gku316 PMID: 24753421 PLOS ONE \| https://doi.org/10.1371/journal.pone.0280975 April 20, 2023 15 / 15 PLOS ONE
10.1371_journal.pone.0285953	RESEARCH ARTICLE High post-exposure prophylaxis (PEP) uptake among household contacts of pertussis patients enrolled in a PEP effectiveness evaluation – United States, 2015–2017 1, Amy B. Rubis1, Lucia Pawloski1, Elizabeth Briere1, Lucy A. McNamaraID Lara Misegades1¤, Aurora A. BrusseauID Kari Burzlaff4, Victor Cruz5, Lucia Tondella1, Tami H. Skoff1, for the Pertussis Post-Exposure Prophylaxis Study Team¶ 2, Sandra Peña1,2, Karen Edge2,3, Rachel Wester4, 1 Division of Bacterial Diseases, National Center for Immunization and Respiratory Diseases, US Centers for Disease Control and Prevention, Atlanta, GA, United States of America, 2 New Mexico Department of Health, Santa Fe, NM, United States of America, 3 Colorado Department of Public Health and Environment, Denver, CO, United States of America, 4 New York State Department of Health, Albany, NY, United States of America, 5 Minnesota Department of Public Health, St. Paul, MN, United States of America ¤ Current address: Office of Refugee Resettlement, Administration for Children and Families, Washington, DC, United States of America ¶ Membership of the Pertussis Post-Exposure Prophylaxis Study Team is provided in the Acknowledgments. xdf4@cdc.gov Abstract Background Post-exposure prophylaxis (PEP) for pertussis is recommended for household contacts of pertussis cases in the United States within 21 days of exposure, but data on PEP effective- ness for prevention of secondary cases in the setting of widespread pertussis vaccination are limited. We implemented a multi-state evaluation of azithromycin PEP use and effective- ness among household contacts. Methods Culture- or PCR-confirmed pertussis cases were identified through surveillance. Household contacts were interviewed within 7 days of case report and again 14–21 days later. Inter- viewers collected information on exposure, demographics, vaccine history, prior pertussis diagnosis, underlying conditions, PEP receipt, pertussis symptoms, and pertussis testing. A subset of household contacts provided nasopharyngeal and blood specimens during interviews. Results Of 299 household contacts who completed both interviews, 12 (4%) reported not receiving PEP. There was no evidence of higher prevalence of cough or pertussis symptoms among contacts who did not receive PEP. Of 168 household contacts who provided at least one nasopharyngeal specimen, four (2.4%) were culture or PCR positive for B. pertussis; three a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: McNamara LA, Rubis AB, Pawloski L, Briere E, Misegades L, Brusseau AA, et al. (2023) High post-exposure prophylaxis (PEP) uptake among household contacts of pertussis patients enrolled in a PEP effectiveness evaluation – United States, 2015–2017. PLoS ONE 18(5): e0285953. https://doi.org/10.1371/journal.pone.0285953 Editor: Farhana Haque, LSHTM: London School of Hygiene & Tropical Medicine, UNITED KINGDOM Received: December 29, 2022 Accepted: May 4, 2023 Published: May 18, 2023 Copyright: This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication. Data Availability Statement: The informed consent forms for this study indicated that participants’ data would be shared only in aggregate, so we are unable to share an individual- level dataset. The aggregated data are available in the manuscript tables and figures. If researchers are interested in obtaining the data in this manuscript aggregated in a different way, they can reach out to MVPDB inquiries (mvpdb@cdc.gov) to discuss potential data access. A data use agreement may be required. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285953 May 18, 2023 1 / 14 PLOS ONEPost-exposure prophylaxis uptake among household contacts of pertussis patients Funding: This work was conducted as part of the Enhanced Pertussis Surveillance through the Emerging Infections Program Network (EIP). The EIP is supported through a Centers for Disease Control and Prevention cooperative agreement. Funder website: https://www.cdc.gov/ Competing interests: The authors have declared that no competing interests exist. of these received PEP prior to their positive test result. Of 156 contacts with serologic results, 14 (9%) had blood specimens that were positive for IgG anti-pertussis toxin (PT) antibodies; all had received PEP. Conclusions Very high PEP uptake was observed among household contacts of pertussis patients. Although the number of contacts who did not receive PEP was small, there was no differ- ence in prevalence of pertussis symptoms or positive laboratory results among these con- tacts compared with those who did receive PEP. Introduction Bordetella pertussis causes pertussis, or whooping cough, a highly contagious respiratory dis- ease with secondary attack rates of up to 80% among susceptible individuals. Introduction of pertussis vaccines during the 1940s significantly decreased the burden of disease in the United States; since the late 1980s, however, the reported number of pertussis cases has gradually increased and epidemic peaks in disease have been reported [1]. Many factors are likely con- tributing to this reported increase including increased provider recognition, changes in diag- nostic testing and reporting, and waning immunity from pertussis vaccines [2]. Periodic evaluation of pertussis prevention and control strategies is critical, especially in the setting of an ongoing pertussis resurgence. Waning immunity from pertussis vaccines is well documented, particularly since the transition to acellular vaccines in the United States during the 1990s [1, 3–8]. Current acellular vaccines have high short-term effectiveness and continue to protect against severe disease; however, many reported pertussis cases in the United States are among fully vaccinated persons [3, 4, 9]. Until improved pertussis vaccines with longer duration of protection are available, alternative pertussis prevention and control strategies such as post-exposure prophylaxis (PEP) can be used alongside vaccination to help control pertussis transmission, especially in the setting of outbreaks among fully vaccinated individuals. PEP is recommended for many pathogens to prevent secondary cases of disease follow- ing exposure to an infected individual [10]. In the United States, pertussis prophylaxis within 21 days of exposure is recommended for household contacts of a confirmed case and for contacts outside the household who are at high risk for severe disease, such as infants <1 year of age [11]; however, implementation of this guidance may vary across U.S. health departments. Macrolide antibiotics, most commonly azithromycin, erythromycin, or clari- thromycin, are recommended for pertussis PEP; co-trimoxazole can be used when macro- lides are contraindicated [12, 13]. While studies have demonstrated the effectiveness of PEP at preventing secondary trans- mission of pertussis [14–18], available data have focused primarily on erythromycin which, compared to the newer macrolide azithromycin, is less commonly used because it requires a longer treatment course and has a less favorable side effect profile, resulting in lower compli- ance [19–22]. It is also unclear whether PEP is needed in the setting of pertussis vaccination, despite waning vaccine-induced immunity [18]. Evaluating the current effectiveness of PEP in preventing secondary pertussis transmission is important to assess whether the benefits of PEP outweigh the potential harms of increased antibiotic use. We therefore implemented a multi- PLOS ONE \| https://doi.org/10.1371/journal.pone.0285953 May 18, 2023 2 / 14 PLOS ONEPost-exposure prophylaxis uptake among household contacts of pertussis patients state study with the objective of describing the use of azithromycin PEP and its effectiveness for preventing secondary cases of pertussis among household contacts of U.S. pertussis cases. Methods Culture- or PCR-confirmed pertussis cases with cough of any duration were identified through routine Enhanced Pertussis Surveillance (EPS) as part of the Emerging Infections Program Network between March 1, 2015 and December 31, 2017. The study catchment area included four EPS sites: Colorado (5-county Denver metro area), Minnesota (statewide), New Mexico (statewide), and New York (15 Albany and Rochester area counties). The sites were selected for study participation based on site interest. EPS captures all reported pertussis cases within the catchment area that meet the above case definition; however, only cases reported and inter- viewed by health department staff within 21 days of cough onset were considered for enroll- ment. Household contacts were recommended a five-day course of azithromycin PEP based on local and state health department recommendations. Following reporting of a pertussis case, the patient’s household was assessed for study eligi- bility as part of the public health investigation of the case. Case households were defined as the location where the index case resided at least 50% of the time during the reference period, defined as the time from index case cough onset through the first study interview date (Fig 1A). Households were ineligible if the index case lived alone, the household was previously enrolled for an earlier pertussis case, the health department was unable to contact the case for the initial public health case investigation within 21 days of cough onset, or any household contacts had acute cough illness or a pertussis diagnosis in the 7–28 days prior to index case cough onset (Fig 1A). Within eligible households, individual contact eligibility was determined by screening household contacts within 7 days of the date the case was reported to the health department. Household contacts were defined as individuals who resided in the same household at least 50% of the time as the index patient during the reference period. Household contacts were ineligible if they could not be screened within 7 days of case report, had an allergy to macrolide antibiotics, had liver disease, or were taking antibiotics other than azithromycin at the time of first study interview. Household contacts were also ineligible if they had acute cough illness with at least one pertussis symptom (paroxysmal cough, whoop, post-tussive vomiting, or apnea) or were diagnosed with pertussis during the reference period (Fig 1A). Of note, if the onset of illness in the household contact was between 7 days prior to cough onset in the index case and the time of the first study interview, the contact was ineligible but other contacts in the household remained eligible to participate (Fig 1A). Eligible household contacts who were screened and consented to participate were inter- viewed in person by study staff within 7 days of the date the index case was reported to the health department (Fig 1B). The initial study interview collected relationship and length of exposure to the case in the household with earliest cough onset, age, gender, pertussis vaccine history, history of physician-diagnosed pertussis prior to the reference period, underlying con- ditions (including immunodeficiencies, chronic respiratory issues, and neuromuscular disor- ders), and pregnancy status. Participating household contacts could optionally provide a nasopharyngeal (NP) and a blood specimen for laboratory testing for evidence of B. pertussis infection. NP specimens were collected using NP swabs and placed in Regan-Lowe transport media, then stored in a cooler with ice packs for up to 24 hours before being transferred to the public health laboratory or CDC. Blood was collected through venipuncture or finger stick and left at room temperature to clot, then transferred to the public health laboratory and PLOS ONE \| https://doi.org/10.1371/journal.pone.0285953 May 18, 2023 3 / 14 PLOS ONEPost-exposure prophylaxis uptake among household contacts of pertussis patients Fig 1. Timeline of household and contact eligibility and implementation of study procedures. a. Top, household eligibility based on day that the index case is reported relative to the day of index case cough onset. Bottom, household or individual household contact eligibility based on day of onset of acute cough illness in a household contact relative to day of cough onset in the index case. If any household contact had acute cough illness or pertussis diagnosis with illness onset 7 to 28 days prior to cough onset in the index case, the household is ineligible. If a household contact had acute cough illness with one or more other pertussis symptom and/or a diagnosis of pertussis with onset between 7 days prior to cough onset in the index case and the time of the first study interview, the contact is ineligible but other contacts in the household remain eligible to participate. b. Timing of the two study interviews relative to the day the index case was reported to the health department. https://doi.org/10.1371/journal.pone.0285953.g001 centrifuged within 24 hours of collection to separate and aliquot the serum. Sera were stored at -40-80˚C and batch-shipped frozen to CDC for testing. At the first study interview, household contacts were given a paper symptom diary (daily symptom checklist) and instructed to record daily information on pertussis symptoms experi- enced between the first and second study interviews as well as antibiotic adherence for partici- pants who received PEP. Household contacts were re-interviewed by study staff in person 14–21 days after the initial study interview to capture potential development of pertussis symptoms during most or all of the reported incubation period for pertussis (4–21 days) (Fig 1B). The second study interview collected information on PEP receipt, the number of PEP doses taken, onset of any new per- tussis symptoms, and any pertussis testing since the initial study interview. At this time, partic- ipants could again optionally provide a NP and blood specimen. However, household contacts were not eligible to have a second set of laboratory specimens collected if they were taking anti- biotics other than azithromycin at the time of the second study interview. The symptom diary was returned to study staff at the second visit and was used to validate the symptom questions on the follow-up questionnaire. Participants were classified as having a symptom by the time of the second study interview if the symptom was noted either during the second study inter- view or on the symptom diary. Household contacts were classified as asymptomatic if they did not report cough or another pertussis symptom at any point during the study period. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285953 May 18, 2023 4 / 14 PLOS ONEPost-exposure prophylaxis uptake among household contacts of pertussis patients NP specimens were tested by culture and PCR. For culture, specimens were plated on Regan-Lowe medium with and without cephalexin and incubated for up to 10 days at 37˚C with high humidity; at the New York site, specimens were also plated on Bordet-Gengou media under similar conditions. The PCR assay used by the Colorado, Minnesota, and New Mexico sites was a combination of a multiplex reaction (IS481, hIS1001, pIS1001) and a single- plex reaction (ptxS1) that distinguishes among B. pertussis, B. holmesii and B. parapertussis [23]. At the New York site, a multiplex PCR assay targeting IS481, BP283, and IS1001 was used instead. The New York assay also distinguishes B. pertussis from B. parapertussis and B. holme- sii [24] and New York State Department of Health unpublished data. Serum specimens were tested at CDC using an anti-PT IgG ELISA [25]. Samples were tested at a 1:100 dilution and run on two plates with triplicate wells per plate. Final concentration was based on the average of two valid tests. Diagnostic cut-offs were <49 International Units (IU)/ mL, 49–93 IU/mL, and �94 IU/mL for negative, indeterminate, and positive results, respec- tively [26]. Specimens collected within 180 days of pertussis vaccination were excluded to ensure that serologic results reflected B. pertussis infection rather than vaccination [27]. Pertussis vaccine history was verified where possible using participant vaccination cards, immunization registries, or provider records. Age-appropriate pertussis vaccination was defined as described previously [9]. Additional case information, including patient age, sex, pertussis vaccine history, and history of previous pertussis diagnosis, was collected through routine EPS surveillance. Data were analyzed using SAS 9.4. Differences in proportions were tested using Fisher’s exact test; differences in means were assessed using a t-test. Patient consent statement: Written informed consent was obtained from all study partici- pants. The study protocol and informed consent forms were reviewed and approved by the Institutional Review Boards of all participating sites before the start of the study (see 45 C.F.R. part 46; 21 C.F.R. part 56). Results Of 2,295 pertussis cases identified in the study catchment area during the study period, 180 pertussis case households (7.8%), which included 424 household contacts, agreed to participate and met household eligibility criteria (Fig 2). Reasons for non-enrollment of identified cases were not systematically captured; however, health departments reported that they were fre- quently unable to contact index case households within seven days of case notification or and index patients had often been coughing for more than 21 days by the time the household was contacted. In addition, at least 300 eligible households declined to participate. Among the 424 household contacts in the 180 households that agreed to participate, 49 con- tacts (12%) were ineligible, most commonly because of pertussis symptoms (31, 7.3%) or diag- nosis (4, 0.9%) before the first study interview. Ten additional household contacts (2.4%) declined participation and 4 (0.9%) failed to complete the first study interview, leaving 361 household contacts (85%) from 169 households (94%) who completed the first study interview. Of these 361 household contacts, 168 (47%) provided an NP specimen and 165 (46%) provided a blood specimen at the first study interview. Two hundred ninety-nine (83%) household con- tacts also completed the second study interview; of these, 128 (43%) provided an NP specimen and 125 (42%) provided a blood specimen at the second study interview. Most participating households included 1–2 eligible contacts; the majority of participating index cases and household contacts resided in Minnesota and Colorado (Table 1). Over half of index cases were in patients aged 11–18 years, while a majority of household contacts were aged 30–64 years. Approximately one third of household contacts were mothers of the index PLOS ONE \| https://doi.org/10.1371/journal.pone.0285953 May 18, 2023 5 / 14 PLOS ONEPost-exposure prophylaxis uptake among household contacts of pertussis patients Fig 2. Household contact eligibility, enrollment, and study participation among contacts identified within the 180 initially enrolled households. NP, nasopharyngeal. https://doi.org/10.1371/journal.pone.0285953.g002 patients and another third were siblings; fathers and other relatives each accounted for a smaller proportion. Seventy-two percent of index cases and over 80% of contacts had ever received a pertussis vaccine and 62–67% had received age-appropriate pertussis vaccination. Of note, most remaining participants had unknown vaccination status; only 4–6% reported having never been vaccinated (Table 1). One index case and seven household contacts reported a previous pertussis diagnosis, which occurred 4–45 years prior to the study. Only 23 (6%) contacts who participated in the first study interview and 12 (4%) who partici- pated in both study interviews reported never receiving PEP (Table 1). The 12 contacts who had not received PEP by the second study interview were aged 1–57 years. Two of these 12 contacts, both adults, had never been vaccinated for pertussis; none of the 12 reported a previ- ous pertussis diagnosis. The 12 contacts were associated with five households. In three of these households (containing one, four, and five eligible contacts), no eligible household contacts received PEP. The remaining two contacts who did not receive PEP were both fathers of the index patient; each belonged to a household where all other contacts (n = 1 or 3) did receive PEP. Only one household contact, the 57-year-old father of an index patient, reported receiv- ing a PEP prescription but not filling it. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285953 May 18, 2023 6 / 14 PLOS ONEPost-exposure prophylaxis uptake among household contacts of pertussis patients Index cases First interview contacts Second interview contacts N % N % N % Table 1. Index case and household contact characteristics. Total N Eligible contacts per household 1 2 3 4 or more Age group <1 year 1–6 years 7–10 years 11–18 years 19–29 years 30–64 years > = 65 years Gender Male Female Missing State CO MN NM NY Relationship to case Mother Father Sibling Other Missing Ever received pertussis vaccine Yes No Unknown Age-appropriate pertussis vaccination Yes No Unknown / too young* Previous pertussis diagnosis Yes No Unknown Received azithromycin PEP** Yes, <7 days after index case onset Yes, 7–13 days after index case onset Yes, �14 days after index case onset Yes, timing unknown 169 61 58 26 24 7 25 30 89 2 13 3 76 75 18 47 103 7 12 - - - - - 122 9 38 104 20 45 1 147 21 - - - - - - 36.1 34.3 15.4 14.2 4.1 14.8 17.8 52.7 1.2 7.7 1.8 45.0 44.4 10.7 27.8 61.0 4.1 7.1 - - - - - 72.2 5.3 22.5 61.5 11.8 26.6 0.59 87.0 12.4 - - - - - 361 - - - - - 6 31 41 56 25 198 4 147 212 2 119 203 16 23 125 66 121 48 1 217 15 29 237 36 88 7 344 10 59 185 84 1 - - - - - - 1.7 8.6 11.4 15.5 6.9 54.8 1.1 40.7 58.7 0.6 33.0 56.2 4.4 6.4 34.6 18.3 33.5 13.3 0.3 83.1 5.7 11.1 65.7 10.0 24.4 1.9 95.3 2.8 16.3 51.2 23.3 0.3 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285953 May 18, 2023 299 82.8 - - - - - 5 24 33 49 14 171 3 120 178 1 105 158 13 23 106 53 107 33 0 266 12 21 199 30 70 7 283 9 56 150 75 1 - - - - - 1.7 8.0 11.0 16.4 4.7 57.2 1.0 40.1 59.5 0.3 35.1 52.8 4.3 7.7 35.5 17.7 35.8 11.0 0 89.0 4.0 7.0 66.6 10.0 23.4 2.3 94.6 3.0 18.7 50.2 25.1 0.3 (Continued ) 7 / 14 PLOS ONEPost-exposure prophylaxis uptake among household contacts of pertussis patients Table 1. (Continued) No Unknown Index cases First interview contacts Second interview contacts N % N % N % - - - - 23 9 6.4 2.5 12 5 4.0 1.7 Too young <2m old at time of interview (1 case and 1 contact) By time of the last study interview the contact participated in https://doi.org/10.1371/journal.pone.0285953.t001 Of those who received azithromycin PEP, only three household contacts reported not com- pleting the full five-day course; all three took PEP for four days. The mean time between index case cough onset and contact PEP administration shortened slightly over the study period, with somewhat faster PEP receipt among household contacts aged <6 or �65 years compared to household contacts in other age groups (S1 Table). PEP receipt was also faster among house- hold contacts who were pregnant (S1 Table). By the second study interview, 22% of enrolled household contacts had developed cough, and 6.5% had developed cough with at least one additional pertussis symptom (Table 2). How- ever, among the 12 household contacts who had not received PEP by the second study inter- view, only one (8.3%) reported cough and none reported cough plus additional pertussis symptoms (Table 2). The prevalence of cough or cough plus additional pertussis symptoms was similar among all PEP recipients regardless of the timing of PEP receipt after cough onset in the index patient. We next examined vaccination and other factors that were potentially associated with per- tussis symptoms among household contacts. While no factors were significantly associated with development of cough alone by the second study interview, household contacts who had never received a pertussis vaccine were significantly more likely to have cough and at least one additional pertussis symptom by the second study interview (S2 Table). There was no signifi- cant difference in the mean age of household contacts who did or did not develop cough (27.1 vs. 31.0 years, p = 0.12) or cough and another pertussis symptom (25.8 vs. 30.6 years, p = 0.26) or who were or were not vaccinated (28.4 vs. 32.8 years, p = 0.33). Of the 168 household contacts who provided at least one NP specimen, one (0.6%; Fig 3, contact a) was culture-positive for B. pertussis and three (1.8%; Fig 3, contacts b-d) were PCR- positive at either the first or second study interview. A fifth household contact (Fig 3, contact 3) was indeterminate by PCR at both the first and second study interviews. The culture-posi- tive individual (contact a) had not received PEP at any point; the remaining four received PEP one to six days prior to their positive or indeterminate PCR result. One of the PCR-positive individuals (contact c) had a second positive PCR result 16 days after PEP initiation. None of these five household contacts had any of the underlying conditions (immunodeficiencies, chronic respiratory issues, and neuromuscular disorders) assessed in the study. Of the four household contacts with positive NP specimens, two (contacts b and c) devel- oped cough, including one with paroxysmal cough. The remaining two household contacts with positive NP specimens (contacts a and d) were asymptomatic at each study interview. One asymptomatic contact (contact a) was culture-positive at the first study interview; how- ever, this individual did not participate in the second study interview and we were therefore unable to assess whether pertussis symptoms developed at a later time point. The second asymptomatic individual (contact d) was PCR-positive at the first study interview but PCR- negative by the second study interview. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285953 May 18, 2023 8 / 14 PLOS ONEPost-exposure prophylaxis uptake among household contacts of pertussis patients Table 2. Pertussis symptoms and positive or indeterminate nasopharyngeal swab or serologic results among household contacts of pertussis cases, by receipt of post-exposure prophylaxis (PEP). Second interview participants Any nasopharyngeal culture or PCR results available Any serologic results available Total Cough Cough and additional symptom Total Positive Indeterminate Total Positive Indeterminate N N % N % N N % N % N N % N % 279 54 150 74 1 12 4 63 11 34 18 0 1 1 295 65 23 20 23 24 0 8.3 25 22 18 3 10 5 0 0 1 19 6.5 5.7 6.7 6.8* 0 0 25 6.5† 148 20 81 47 0 13 7 168 3 0 2 1 0 1 0 4 2.0 0 2.5 2.1 – 7.7 0 2.4 1 0 0 1 0 0 0 1 0.68 0 0 2.1 – 0 0 142 14 19 76 47 0 9 5 3 6 5 0 0 0 0.6 156 14 9.9 16 7.9 11 – 0 0 9 14 1 6 7 0 1 1 16 9.9 5.3 7.9 15 – 11 20 10 Received PEP (days after index case onset) Any <7 7-13 > = 14 Unknown No PEP Unknown PEP Total Of 277 with results available Of 53 with results available **Of 73 with results available †Of 293 with results available https://doi.org/10.1371/journal.pone.0285953.t002 Serum specimens from participants were tested for serologic evidence of B. pertussis infec- tion. Of 156 household contacts who provided blood specimens, 14 (9%) had blood specimens that were positive for IgG anti-PT antibodies at any study interview (Table 2); all had received PEP. Of these, six (43%) had cough and three (21%) had cough with at least one additional per- tussis symptom. Among seropositive individuals, there was no significant difference in anti- body concentrations between those with or without cough (t-test p = 0.2). Eleven of the 14 individuals with positive serology also provided one or more NP swab specimens; of these, one was positive by PCR at the first study interview but negative at the second, and one was inde- terminate by PCR at both study interviews. Interestingly, the individual who was PCR-positive for B. pertussis at both study interviews was negative for IgG anti-PT antibodies at both time points (Fig 3). Because very few household contacts had NP swabs that were positive for B. pertussis by cul- ture or PCR, or serology specimens positive for anti-pertussis toxin antibodies, the relationship between PEP receipt and positive NP or serum specimens could not be assessed. Discussion In concordance with CDC guidance for pertussis prevention and control, we found high and rapid PEP uptake among household contacts of pertussis patients enrolled in our study. While this high PEP uptake was encouraging, especially in an era of pertussis resurgence, it unfortu- nately limited our ability to address our primary study question, which was to assess the effec- tiveness of PEP at preventing pertussis among household contacts. However, among the few contacts who did not receive PEP in our study, none had cough with other pertussis symptoms at the time of the second study interview. There was also no difference in prevalence of cough and other pertussis symptoms among those who received PEP earlier vs. later after index case cough onset, suggesting that delays in PEP administration may not increase the secondary transmission of pertussis to household contacts. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285953 May 18, 2023 9 / 14 PLOS ONEPost-exposure prophylaxis uptake among household contacts of pertussis patients Fig 3. Timing of key events for the five household contacts with positive PCR or culture for B. pertussis on nasopharyngeal swabs. Contacts a and d were asymptomatic at all study interviews in which they participated. NP, nasopharyngeal; PCR, polymerase chain reaction; PEP, post-exposure prophylaxis. https://doi.org/10.1371/journal.pone.0285953.g003 Our findings contrast with the results of a recent study conducted in Spain, which demon- strated a significant decrease in secondary attack rate among contacts who received azithromy- cin PEP within seven days of index case symptom onset [28]. Since the Spanish study included more than five times as many contacts as our study, this difference is likely driven by sample size. Differences between the enrolled study populations may also have contributed; the Span- ish study included contacts who resided outside the index patient household and the distribu- tions of contact age, gender, and relationship to the index case also differed. Unfortunately, the potential contribution of differential vaccination coverage to the difference in findings cannot be readily assessed as the Spanish study characterized vaccination status only among contacts �18 years of age. Overall, only 6.5% of enrolled household contacts developed cough with an additional per- tussis symptom during this study, and only four (2.4%) of the 168 contacts who provided at least one NP specimen had a positive culture or PCR result for B. pertussis. This low secondary attack rate is consistent with that observed in the Spanish study [28] and contrasts with previ- ously reported secondary attack rates of 80% or more among unvaccinated household contacts [29, 30], suggesting that pertussis secondary attack rates are reduced in a setting of high PEP and vaccination uptake. While the proportion of household contacts with cough with or with- out additional symptoms (22%) was higher than the proportion that had cough with an addi- tional symptom, this non-specific presentation likely includes individuals with other PLOS ONE \| https://doi.org/10.1371/journal.pone.0285953 May 18, 2023 10 / 14 PLOS ONEPost-exposure prophylaxis uptake among household contacts of pertussis patients respiratory illnesses besides pertussis. Of note, household contacts who had never been vacci- nated pertussis had a 25% prevalence of cough and another pertussis symptom at the second study interview, compared with 5.4% among those who had ever received a pertussis vaccine. This finding further underscores the importance of pertussis vaccination despite challenges with waning of vaccine-induced immunity. The limited evidence for PEP effectiveness in the context of high pertussis vaccination uptake and widespread community transmission supports the U.S. Centers for Disease Con- trol and Prevention stance for limiting pertussis PEP to household and high-risk contacts. Even with targeted PEP administration, there remain concerns that the benefits of PEP may be outweighed by the drawbacks of promoting antibiotic resistance through overuse of azithro- mycin. Additionally, while infants are at highest risk for severe pertussis morbidity and mortal- ity during the early months of life, before they are old enough to be vaccinated, and therefore arguably might benefit the most from PEP, an association between azithromycin receipt and infantile hypertrophic pyloric stenosis has been reported [31]. Based on the unclear benefits and potential drawbacks of PEP, some countries as well as some U.S. state health departments have implemented more restrictive variations of this pertussis PEP guidance, such as only rec- ommending PEP for high-risk contacts both within and outside the household setting [32]. A major limitation to this study was the extremely low power to detect a difference in per- tussis secondary attack rate between household contacts who did or did not receive PEP. Whereas the target sample size to achieve adequate statistical power was 1,424 household con- tacts, and over 2,000 case households were originally identified during the study time period, only 169 households and 361 contacts were ultimately included in the study. Low enrollment was due in large part to difficulties in identifying and contacting cases within the time limits prescribed by the study; however, the refusal rate was also high, with over 300 eligible house- holds declining to participate. The high refusal rate not only limited the sample size but also the generalizability of our results, as households that refuse to participate may have been less likely to have high uptake of PEP. This concern about generalizability is exacerbated by the fact that a large majority of study participants were recruited from only two states, Colorado and Minnesota. Finally, among households that did participate, PEP uptake was very high and rapid, so the group of individuals who had not received PEP by the first or second study inter- views was very small. Considering the high PEP administration rates observed among house- hold contacts in our study sites, further attempts to study PEP effectiveness might be more readily conducted in locations with more restrictive pertussis PEP policies to increase the size of the non-PEP recipient comparison group. While the very small number of non-PEP recipients precluded our ability to statistically assess the effectiveness of PEP, our study does suggest that the combined strategy of pertussis vaccination and PEP reduces pertussis infection among household contacts to levels far lower than observed historically [29, 30]. The prevalence of pertussis symptoms among household contacts was similar regardless of timing of PEP receipt after index case onset; in contrast, the frequency of cough plus an additional symptom was increased among unvaccinated or inade- quately vaccinated household contacts. These findings thus reaffirm that despite the resur- gence of pertussis and concerns around waning immunity from vaccines, adhering to pertussis vaccine recommendations remains an effective way to prevent additional cases of pertussis. Meanwhile, further study of PEP effectiveness in a setting of less uniform PEP uptake would help inform the best use of this strategy as the global community continues to wait on the development of next-generation pertussis vaccines. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285953 May 18, 2023 11 / 14 PLOS ONEPost-exposure prophylaxis uptake among household contacts of pertussis patients Supporting information S1 Table. PEP receipt and timing of PEP administration among household contacts of per- tussis cases, by contact characteristics. (DOCX) S2 Table. Development of pertussis symptoms among household contacts of pertussis cases by vaccination status, relationship to case, and timing of case antibiotic treatment. (DOCX) Acknowledgments The lead authors for the Pertussis Post-Exposure Prophylaxis Team are indicated by name in the author list. The additional members of the Pertussis Post-Exposure Prophylaxis Team are: Bernadette Albanese, MD, MPH; Lisa Devries, RN, BSN; Michael Ryan, RN, BSN; Lindsay Bel- lamy, RN, BSN; and Bryce Alderson, RN, BSN, MS, Tri-County Health Department. Ann Shen, RN, BSN; Ran Tao, RN, DNP, MPH; Katie VanHoosen, RN, BSN; Erin Blau, RN, DNP, MPH; and Christine Schmidt, RN, MS, Jefferson County Health Department. Carol McDon- ald, RN, Denver Public Health. Katherine Herpin, MPA, MSEd; Kate Ott, MPH, BSN, RN; Kathleen Root, RN, BSN; Sarah Shayne, RN; Karen Kulas, BS; Kara Mitchell, PhD; Elizabeth Nazarian, PhD; Glenda Smith, MPH; and Nancy Spina, MPH, New York State Department of Health. Hong Ju, PhD and Marsenia Mathis, MS, MPH, US Centers for Disease Control and Prevention. The authors also acknowledge Xarviera Appling, Rachel Sweet, Ezra Menon, Emily Laurent, and Kelsea Keep, Minnesota Department of Public Health. The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention. Author Contributions Conceptualization: Elizabeth Briere, Lara Misegades, Lucia Tondella, Tami H. Skoff. Data curation: Amy B. Rubis. Formal analysis: Lucy A. McNamara, Lucia Pawloski. Funding acquisition: Tami H. Skoff. Investigation: Lucia Pawloski, Aurora A. Brusseau, Sandra Peña, Karen Edge, Rachel Wester, Kari Burzlaff, Victor Cruz. Methodology: Elizabeth Briere, Lara Misegades, Lucia Tondella, Tami H. Skoff. Project administration: Lucy A. McNamara, Amy B. Rubis, Elizabeth Briere, Lara Misegades, Aurora A. Brusseau, Sandra Peña, Karen Edge, Rachel Wester, Kari Burzlaff, Victor Cruz, Lucia Tondella, Tami H. Skoff. Resources: Lucia Tondella, Tami H. Skoff. Supervision: Lucy A. McNamara, Elizabeth Briere, Lucia Tondella, Tami H. Skoff. Visualization: Lucy A. McNamara. Writing – original draft: Lucy A. McNamara, Amy B. Rubis, Tami H. Skoff. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285953 May 18, 2023 12 / 14 PLOS ONEPost-exposure prophylaxis uptake among household contacts of pertussis patients Writing – review & editing: Lucy A. McNamara, Amy B. Rubis, Lucia Pawloski, Elizabeth Briere, Lara Misegades, Aurora A. Brusseau, Sandra Peña, Karen Edge, Rachel Wester, Kari Burzlaff, Victor Cruz, Lucia Tondella, Tami H. Skoff. References 1. Skoff TH, Hadler S, Hariri S. The Epidemiology of Nationally Reported Pertussis in the United States, 2000–2016. Clin Infect Dis. 2019 May 2; 68(10):1634–1640. https://doi.org/10.1093/cid/ciy757 PMID: 30169627 2. Clark T. Changing pertussis epidemiology: everything old is new again. The Journal of infectious dis- eases. 2014; 209(7):978–81. https://doi.org/10.1093/infdis/jiu001 PMID: 24626532 3. Misegades L, Winter K, Harriman K, Talarico J, Messonnier N, Clark T, et al. Association of childhood pertussis with receipt of 5 doses of pertussis vaccine by time since last vaccine dose, California, 2010. JAMA: the Journal of the American Medical Association. 2012; 308(20):2126–32. https://doi.org/10. 1001/jama.2012.14939 PMID: 23188029 4. Acosta AM, DeBolt C, Tasslimi A, Lewis M, Stewart LK, Misegades LK, et al. Tdap Vaccine Effective- ness in Adolescents During the 2012 Washington State Pertussis Epidemic. Pediatrics. 2015; 135 (6):981–9. https://doi.org/10.1542/peds.2014-3358 PMID: 25941309 5. Breakwell L, Kelso P, Finley C, Schoenfeld S, Goode B, Misegades L, et al. Pertussis Vaccine Effective- ness in the Setting of Pertactin-Deficient Pertussis. Pediatrics. 2016; 137(5). https://doi.org/10.1542/ peds.2015-3973 PMID: 27244813 6. Tartof S, Lewis M, Kenyon C, White K, Osborn A, Liko J, et al. Waning immunity to pertussis following 5 doses of DTaP. Pediatrics. 2013; 131(4):e1047–e52. https://doi.org/10.1542/peds.2012-1928 PMID: 23478868 7. Klein NP, Bartlett J, Rowhani-Rahbar A, Fireman B, Baxter R. Waning Protection after Fifth Dose of Acellular Pertussis Vaccine in Children. New England Journal of Medicine. 2012; 367(11):1012–9. https://doi.org/10.1056/NEJMoa1200850 PMID: 22970945 8. Klein N, Bartlett J, Fireman B, Baxter R. Waning Tdap Effectiveness in Adolescents. Pediatrics. 2016; 137(3):e20153326–e. https://doi.org/10.1542/peds.2015-3326 PMID: 26908667 9. McNamara L, Skoff T, Faulkner A, Miller L, Kudish K, Kenyon C, et al. Reduced Severity of Pertussis in Persons With Age-Appropriate Pertussis Vaccination-United States, 2010–2012. Clinical infectious dis- eases. 2017; 65(5):811–8. https://doi.org/10.1093/cid/cix421 PMID: 29017283 10. Bader MS, McKinsey DS. Postexposure prophylaxis for common infectious diseases. American family physician. 2013; 88(1):25–32. PMID: 23939603 11. CDC. Postexposure Antimicrobial Prophylaxis: Information for Health Professionals [updated August 7, 2017. Available from: https://www.cdc.gov/pertussis/outbreaks/pep.html. 12. TIWARI T, MURPHY T, MORAN J. Recommended antimicrobial agents for the treatment and postex- posure prophylaxis of pertussis 2005 CDC guidelines. Morbidity and Mortality Weekly Report. 2005; 54 (RR 14):1–16. PMID: 16340941 13. American Academy of Pediatrics. Pertussis (Whooping Cough). In: Kimberlin DW BM, Jackson MA, Long SS, eds, editor. Red Book: 2018 Report of the Committee on Infectious Diseases. 31st ed. Itasca, IL2018. p. 620–34. 14. Halperin SA, Bortolussi R, Langley JM, Eastwood BJ, De Serres G. A randomized, placebo-controlled trial of erythromycin estolate chemoprophylaxis for household contacts of children with culture-positive bordetella pertussis infection. Pediatrics. 1999; 104(4):e42. https://doi.org/10.1542/peds.104.4.e42 PMID: 10506267 15. Dodhia H, Miller E. Review of the evidence for the use of erythromycin in the management of persons exposed to pertussis. Epidemiology and infection. 1998; 120(02):143–9. https://doi.org/10.1017/ s0950268897008571 PMID: 9593483 16. Altunaiji SM, Kukuruzovic RH, Curtis NC, Massie J. Antibiotics for whooping cough (pertussis). Cochrane Database of Systematic Reviews. 2007(3). https://doi.org/10.1002/14651858.CD004404. pub3 PMID: 17636756 17. Godoy P, Garcı´a Cenoz M, Toledo D, Carmona G, Caylà J, Alsedà M, et al. Factors influencing the spread of pertussis in households: a prospective study, Catalonia and Navarre, Spain, 2012 to 2013. Euro surveillance. 2016; 21(45). https://doi.org/10.2807/1560-7917.ES.2016.21.45.30393 PMID: 27918260 18. Goins W, Edwards K, Vnencak Jones C, Rock M, Swift M, Thayer V, et al. A comparison of 2 strategies to prevent infection following pertussis exposure in vaccinated healthcare personnel. Clinical infectious diseases. 2012; 54(7):938–45. https://doi.org/10.1093/cid/cir973 PMID: 22238169 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285953 May 18, 2023 13 / 14 PLOS ONEPost-exposure prophylaxis uptake among household contacts of pertussis patients 19. Cherry JD. Treatment of Pertussis—2017. Journal of the Pediatric Infectious Diseases Society. 2017; 7 (3):e123–e5. 20. Tanaka H, Kaji M, Higuchi K, Shinohara N, Norimatsu M, Kawazoe H, et al. Problems associated with prophylactic use of erythromycin in 1566 staff to prevent hospital infection during the outbreak of pertus- sis. 2009; 34(6):719–22. 21. Tornese G, Patarino F, Marchetti F. Erythromycin in whooping cough. 2009; 338:b2615. 22. Devasia RA, Jones TF, Collier B, Schaffner W. Compliance With Azithromycin Versus Erythromycin in the Setting of a Pertussis Outbreak. The American Journal of the Medical Sciences. 2009; 337(3):176– 8. https://doi.org/10.1097/maj.0b013e318184e72f PMID: 19301451 23. Tatti K, Sparks K, Boney K, Tondella M. Novel multitarget real-time PCR assay for rapid detection of Bordetella species in clinical specimens. Journal of Clinical Microbiology. 2011; 49(12):4059–66. https://doi.org/10.1128/JCM.00601-11 PMID: 21940464 24. Probert WS, Ely J, Schrader K, Atwell J, Nossoff A, Kwan S. Identification and evaluation of new target sequences for specific detection of Bordetella pertussis by real-time PCR. J Clin Microbiol. 2008; 46 (10):3228–31. https://doi.org/10.1128/JCM.00386-08 PMID: 18753352 25. Menzies SL, Kadwad V, Pawloski LC, Lin TL, Baughman AL, Martin M, et al. Development and analyti- cal validation of an immunoassay for quantifying serum anti-pertussis toxin antibodies resulting from Bordetella pertussis infection. Clinical and vaccine immunology: CVI. 2009; 16(12):1781–8. 26. Baughman AL, Bisgard KM, Edwards KM, Guris D, Decker MD, Holland K, et al. Establishment of diag- nostic cutoff points for levels of serum antibodies to pertussis toxin, filamentous hemagglutinin, and fim- briae in adolescents and adults in the United States. ClinDiagnLab Immunol. 2004; 11(6):1045–53. https://doi.org/10.1128/CDLI.11.6.1045-1053.2004 PMID: 15539504 27. Pawloski L, Kirkland K, Baughman A, Martin M, Talbot E, Messonnier N, et al. Does tetanus-diphtheria- acellular pertussis vaccination interfere with serodiagnosis of pertussis infection? Clinical and Vaccine Immunology. 2012; 19(6):875–80. https://doi.org/10.1128/CVI.05686-11 PMID: 22539469 28. Alvarez J, Godoy P, Plans-Rubio P, Camps N, Carol M, Carmona G, et al. Azithromycin to Prevent Per- tussis in Household Contacts, Catalonia and Navarre, Spain, 2012–2013. Emerg Infect Dis. 2020; 26 (11):2678–84. https://doi.org/10.3201/eid2611.181418 PMID: 33079034 29. Lambert HJ. Epidemiology of a small pertussis outbreak in Kent County, Michigan. Public Health Rep. 1965; 80(4):365–9. PMID: 14279983 30. Broome CV, Preblud SR, Bruner B, McGowan JE, Hayes PS, Harris PP, et al. Epidemiology of pertus- sis, Atlanta, 1977. J Pediatr. 1981; 98(3):362–7. https://doi.org/10.1016/s0022-3476(81)80696-8 PMID: 7009816 31. Eberly M, Eide M, Thompson J, Nylund C. Azithromycin in early infancy and pyloric stenosis. Pediatrics. 2015; 135(3):483–8. https://doi.org/10.1542/peds.2014-2026 PMID: 25687145 32. Skoff TH, Brown C. S., and Amirthalingam G. Public health consequences. In: Pertussis: epidemiology, immunology, & evolution. In: Scarpino PRaSV, editor.: Oxford University Press; 2019. p. 241–56. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285953 May 18, 2023 14 / 14 PLOS ONE
10.1371_journal.pone.0283609	RESEARCH ARTICLE Skin microbiome alterations in upper extremity secondary lymphedema Adana-Christine CampbellID Jinyeon Shin1, Kevin Kuonqui1, Stav Brown1, Ananta Sarker1, Raghu P. Kataru1, Babak J. MehraraID 1‡, Teng Fei2‡, Jung Eun Baik1¤, Hyeung Ju Park1, 1* a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Campbell A-C, Fei T, Baik JE, Park HJ, Shin J, Kuonqui K, et al. (2023) Skin microbiome alterations in upper extremity secondary lymphedema. PLoS ONE 18(5): e0283609. https:// doi.org/10.1371/journal.pone.0283609 Editor: Jose´ Anto´nio Baptista Machado Soares, Universidad San Francisco de Quito, ECUADOR Received: October 12, 2022 Accepted: March 13, 2023 Published: May 17, 2023 Copyright: © 2023 Campbell et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Funding: This research was supported in part by the NIH through R01 HL111130 awarded to B.J.M., T32CA009501 (stipend for A.C.C.), and the Cancer Center Support Grant P30 CA008748. The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. 1 Division of Plastic and Reconstructive Surgery, Memorial Sloan Kettering Cancer Center, Department of Surgery, New York, NY, United States of America, 2 Department of Epidemiology and Biostatistics, Memorial Sloan Kettering Cancer Center, New York, NY, United States of America ¤ Current address: Department of Biotechnology, Levatio Therapeutics, San Diego, California, United States of America ‡ ACC and TF contributed equally and also share first authorship to this work. * mehrarab@mskcc.org Abstract Lymphedema is a chronic condition that commonly occur from lymphatic injury following sur- gical resection of solid malignancies. While many studies have centered on the molecular and immune pathways that perpetuate lymphatic dysfunction, the role of the skin micro- biome in lymphedema development remains unclear. In this study, skin swabs collected from normal and lymphedema forearms of 30 patients with unilateral upper extremity lymph- edema were analyzed by 16S ribosomal RNA sequencing. Statistical models for micro- biome data were utilized to correlate clinical variables with microbial profiles. Overall, 872 bacterial taxa were identified. There were no significant differences in microbial alpha diver- sity of the colonizing bacteria between normal and lymphedema skin samples (p = 0.25). Notably, for patients without a history of infection, a one-fold change in relative limb volume was significantly associated with a 0.58-unit increase in Bray-Curtis microbial distance between paired limbs (95%CI = 0.11,1.05, p = 0.02). Additionally, several genera, including Propionibacterium and Streptococcus, demonstrated high variability between paired sam- ples. In summary, we demonstrate high compositional heterogeneity in the skin microbiome in upper extremity secondary lymphedema, supporting future studies into the role of host- microbe interactions on lymphedema pathophysiology. 1. Introduction Secondary lymphedema (LE) is a chronic condition of the lymphatic system that is character- ized by fibroadipose tissue deposition, chronic inflammation, and, in some cases, recurrent infections [1, 2]. In fact, nearly 40% of patients with LE develop recurrent cellulitis and lym- phangitis requiring antibiotic treatment and hospitalization [3, 4]. In some cases, LE-related infections can be severe, resulting in sepsis and even death [5]. For example,92% of the 165,055 LE-related hospital admissions in the US between 2012–2017 were for treatment of cellulitis and had an associated inpatient mortality of 0.03% [6]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283609 May 17, 2023 1 / 12 PLOS ONESkin microbiome in secondary lymphedema Patients with secondary LE have impaired immune responses to bacterial and viral anti- gens, making recurrent infections more likely; however, the mechanisms that underlie this increased risk remain largely unknown [7, 8]. In the past ten years our lab and others have shown that chronic T-helper 2 (TH2) immune responses are an important pathological response in LE and these responses are known to cause barrier disruption in other chronic inflammatory skin disorders [9, 10]. Impaired barrier function may thus provide a port of entry for bacteria since the skin is an important defense against infections [11]. In addition, the accumulation of protein-rich fluid in LE provides an optimal environment for bacterial colonization [12]. Alterations in the skin microbiome are associated with cutaneous skin disorders, such as atopic dermatitis and psoriasis [13–16]. Atopic dermatitis results in a dysbiosis that favors expansion of Staphylococcus aureus, which correlates with the severity of disease [13, 17–19]. Interestingly, the Th2-based immunologic changes that drive atopic dermatitis share signifi- cant similarities with secondary LE and may thus implicate a role for inflammation-driven dysbiosis in LE pathogenesis. However, to determine if observed microbiome changes contrib- ute to infection risk in secondary LE patients, alternative methods for differential abundance analysis are required, such as metagenomic sequencing to clarify the functional profile of the microbes detected [20]. To date, only one previous study [21] has analyzed bacterial dysbiosis and infection risk in LE resulting from filarial infections. Here, using high-throughput genomic sequencing, we analyzed skin microbiome composi- tion in paired affected/unaffected skin samples from patients with unilateral upper extremity cancer-related LE. We show high compositional heterogeneity in the skin microbiome in LE and that variations in relative abundance relate, in part, to relative limb volume difference between the normal and LE limb. Our results highlight a new area of study for LE pathology. 2. Materials and methods 2.1 Patient demographics and skin sample collection Patients were recruited from the Plastic and Reconstructive Surgery Lymphedema Clinic at Memorial Sloan Kettering Cancer Center (MSK). The inclusion criteria were: (1) age >18 years; (2) unilateral upper extremity lymphedema, with a >10% difference in volume and tex- ture of the affected limb; and (3) moderate to severe severity according to the International Society of Lymphology (ISL). Patients with an acute inflammatory condition, such as flu-like illness, skin infection, or fever-associated illness within six weeks of sample collection were excluded. Additional exclusion criteria included: (1) history of metastatic or untreated breast cancer; (2) systemic or topical antibiotic treatment within six weeks of sample collection; (3) history of chronic skin disease or open wounds of the upper extremities; and (4) recent use of antiseptic topical applications. In total, thirty (28 female and 2 male) patients with LE were selected based on the inclusion criteria. All female patients had a primary breast cancer diag- nosis. For the two male patients, the underlying diagnoses included squamous cell cancer of the left axilla of unknown primary and left midback melanoma. All patients provided written informed consent. The study was approved by the Memorial Sloan-Kettering Cancer Center’s Institutional Review Board/Privacy Board-A and Institutional Review Board/ Privacy Board-B (IRB 18–536) Participants were contacted 48 hours prior to sample collection to confirm eligibility according to the inclusion criteria. Twenty-four hours prior to clinic arrival, patients were instructed as follows: (1) do not shower, bathe, or wash forearms with soap or water; (2) avoid creams, moisturizers, perfumes, and lotion applications to the forearms; and (3) limit the use of compressive garments. Skin swabs were collected from the proximal forearm of the normal PLOS ONE \| https://doi.org/10.1371/journal.pone.0283609 May 17, 2023 2 / 12 PLOS ONESkin microbiome in secondary lymphedema and LE limbs according to the skin sampling protocol outlined by the Children’s Hospital of Philadelphia (CHOP) Microbiome Center, Division of Gastroenterology, Hepatology, and Nutrition. Briefly, Copan flexible flocked swabs (FLOQSwab1553-C, Copan Diagnostics Inc.) were moistened with sterile PBS, and the volar forearm of the LE limb stroked 60 times, alter- nating directions vertically and horizontally, over a sampling diameter of <4 cm2. The maneu- ver was repeated for the normal limb using a clean FLOQSwab. Four swabs moistened with sterile PBS alone were collected as negative controls. The swabs were then placed in a dry col- lection tube, appropriately labeled by paired sample number, and placed in a mobile liquid nitrogen container for overnight shipment to the CHOP Microbiome Center. 2.2 DNA extraction, library construction, and 16S rRNA sequencing DNA sequencing of the bacterial 16S rRNA gene V1-V3 region was carried out at the CHOP Microbiome Center. DNA was extracted using the PowerSoil kit (Qiagen). DNA library prepa- ration was performed using dual-barcoded primers targeting the V1-V3 regions of the bacte- rial 16S rRNA gene. PCR products were sequenced as 300 base-pair reads using the Illumina MiSeq instrument16S rRNA marker gene sequence data was analyzed using the QIIME2 pipe- line (v2019.7) with default parameters [22]. Denoising and selection of the amplicon sequence variants (ASVs) were performed with DADA2 software [23]. Taxonomic assignments were generated using a Naïve Bayes classifier trained on the Greengenes reference database (v13_8) [24]. 2.3 Statistical analysis All statistical analyses were performed in R 4.1.1 (R Core Team, 2021). Descriptive statistics for the study population are reported, including median and interquartile range (IQR) for con- tinuous variables and percentages for categorical variables. Missing data were omitted from descriptive statistics. Shannon index and Inverse Simpson index were calculated for microbial α-diversity. Wilcoxon signed-rank test was used to test paired differences of α-diversity between normal and LE samples. Bray-Curtis distance (BCD) and Aitchison’s distance (AD) were calculated for microbial β-diversity. Principal coordinate analysis (PCoA) and corre- sponding 2D visualization plots were conducted based on BCD matrix. Multivariable linear regression was applied to investigate the association between patient-specific paired microbial distances (BCD or AD) and patient clinical characteristics. Microbial variability analysis was conducted to reveal which taxa had high variation between paired LE and normal samples, where variability was defined as the absolute value of the relative abundance difference between paired samples. Linear decomposition model [25] was applied to test taxa differential abundance between paired LE and normal samples, adjusting for previously described clinical variables. The obtained p-values were adjusted for multiple testing by sequential Monte Carlo multiple testing procedure [26]. P-values <0.05 were considered statistically significant. 3. Results 3.1 Microbial profiles are similar in normal and LE skin We collected skin swabs from the proximal forearm of the normal and LE limb in 30 patients with unilateral upper extremity LE (Table 1). Analysis by high-throughput 16S RNA sequenc- ing demonstrated no significant difference in microbial α-diversity between normal and LE skin as measured by Shannon and Inverse Simpson’s indices (Fig 1A and 1B). Moreover, there was no consistent difference in α-diversity when comparing paired normal and LE limbs (Fig 1A and 1B; grey lines)—some patients had higher diversity in the LE limb, while others PLOS ONE \| https://doi.org/10.1371/journal.pone.0283609 May 17, 2023 3 / 12 PLOS ONETable 1. Patient demographics and clinical characteristics. Skin microbiome in secondary lymphedema Characteristic Sex Female Male Unknown Age (years) Unknown Race African-American Asian Indian White Unknown History of infection No Yes Duration of LE (months) Unknown Absolute volume differential Unknown Relative volume differential (%) Unknown ISL stage Unknown N = 30 27 (93%) 2 (6.9%) 1 59 (12) 1 1 (3.6%) 1 (3.6%) 26 (93%) 2 10 (36%) 18 (64%) 82 (55) 3 671 (482) 4 30 (22) 4 2.0 (0.0) 2 Data are n(%) or Mean (SD) LE: lymphedema; ISL: International Society of Lymphology https://doi.org/10.1371/journal.pone.0283609.t001 had a higher diversity in the normal limb. Consistent with these findings, principal coordinate analysis (PCoA) showed no clear separation of LE and normal samples according to the Bray- Curtis distance (BCD) (Fig 1C). 3.2 Clinical factors are associated with microbial dissimilarity and heterogeneity in LE We next investigated the degree of microbial heterogeneity between normal and LE limbs. BCD quantifies compositional distance between two samples on a scale of 0 to 1, where dis- tances closer to 0 indicate similar microbial compositions, while distances closer to 1 imply highly different profiles. Overall, there was a high degree of heterogeneity in our cohort (Fig 2A)—for example, subject 18 had dramatically different microbial profiles between limbs (BCD = 0.88), while subject 6 had very similar microbial compositions between limbs (BCD = 0.15). In addition, 4/5 patients with the highest paired BCDs had no history of infection. We used multivariable linear regression to further investigate the association between clini- cal variables and BCD. Interestingly, we found that increases in limb volume were associated with increased BCD in patients who did not have a history of infection (Table 2A); a 1-fold increase in volume between the LE and normal limbs resulted in a 0.58-unit increase in paired BCD for patients without a history of infection (95% CI = 0.11, 1.05; p = 0.02). In contrast, in patients with a history of infection, a 1-fold increase in limb volume was associated with a PLOS ONE \| https://doi.org/10.1371/journal.pone.0283609 May 17, 2023 4 / 12 PLOS ONESkin microbiome in secondary lymphedema Fig 1. Microbial diversity of normal and lymphedema limbs in 30 patients with upper extremity secondary lymphedema. (a) Boxplot of Shannon index for lymphedema and normal limbs; paired samples are connected by grey lines. P-values obtained by Wilcoxon signed-rank test for paired data. (b) Boxplot of inverse Simpson’s index for lymphedema and normal limbs; paired samples connected by grey lines. P-values obtained by Wilcoxon signed-rank test for paired data. (c) Principal component analysis (PCoA) plot of the first two principal coordinates (PC1, PC2) based on Bray-Curtis distance matrix. https://doi.org/10.1371/journal.pone.0283609.g001 non-significant, 0.07-unit increase in paired BCD (95% CI = -0.26, 0.39; p = 0.69). The correla- tion between relative volume differential and BCD in patients with and without a history of LE-related infection is shown in Fig 2B. The duration of LE shows positive but insignificant association with BCD (p = 0.11; Table 2A and Fig 2C). The multivariable linear model for the Aitchison’s distance indicated similar associations, where a 1-fold increase in volume between the LE and normal limbs resulted in a significant increase in AD for patients with a history of Fig 2. Association of microbial distance between paired limbs and clinical covariates. (a) Swimmer plot of Bray-Curtis distance between lymphedema and normal limbs, with history of infection indicated. (b) Scatter plot of paired Bray-Curtis distance versus relative volume differential, with fitted line and confidence band from marginal linear regression, stratified by history of infection. (c) Scatter plot of paired Bray-Curtis distance versus duration of lymphedema in months, with fitted line and confidence band from marginal linear regression. https://doi.org/10.1371/journal.pone.0283609.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283609 May 17, 2023 5 / 12 PLOS ONESkin microbiome in secondary lymphedema Table 2. Point estimates, 95% confidence intervals, and corresponding p-values of patient clinical characteristics in the multivariable linear regression model of (a) paired Bray-Curtis distance and (b) paired Aitchison’s distance. (a) Model for Bray-Curtis distance Characteristic History of infection No Yes Duration of LE Relative volume differential No history of infection History of infection (b) Model for Aitchison’s distance Characteristic History of infection No Yes Duration of LE Relative volume differential No history of infection History of infection CI: confidence interval Estimate 95% CI p-value — -0.043 0.001 0.580 0.066 — -0.237, 0.152 0.000, 0.002 0.111, 1.048 -0.257, 0.388 0.651 0.112 0.018 0.689 Estimate 95% CI p-value — 4.403 0.087 11.30 11.23 — -4.672, 13.48 -0.011, 0.185 0.635, 21.97 1.231, 21.22 0.323 0.078 0.039 0.028 https://doi.org/10.1371/journal.pone.0283609.t002 infection (11.30-unit increase; 95% CI = 0.635, 21.97; p = 0.04) and without (11.23-unit increase; 95% CI = 1.231, 21.22; p = 0.03; Table 2b). 3.3 Taxa-specific analysis demonstrates high genus-level variability between normal and LE limbs Across all samples, 872 genus-level bacterial taxa were identified. To identify taxa with the highest variation between paired normal and LE samples, we calculated the absolute difference of taxa relative abundance between each LE and normal pair (Table 3 and Fig 3A). In particu- lar, the genera Propionibacterium and Streptococcus demonstrated high variations between paired limbs, with average variability in relative abundance of 10% and 5%, respectively. The microbial variability also varied across patients, further indicating high heterogeneity among LE patients. On the other hand, the direction of relative abundance changes was not uniform across patients (Fig 3B), where the average abundance differences were close to zero (Table 3). Moreover, no obvious directional shifts were observed for these highly unstable taxa for patients with differing histories of infection (Fig 3C). Finally, taxa differential abundance test- ing [25] also showed that no taxa were significantly differentially abundant between normal and LE after adjusting for potentially confounding clinical variables and false discovery rate [26] (Tables 3 and 4). Mean variability was calculated as average absolute difference of taxon- specific relative abundance between paired normal and LE samples. Mean differential abun- dance was calculated as average signed difference of taxon-specific relative abundance between paired normal and LE samples. Taxa direction indicates the limb with the higher abundance. 4. Discussion A better understanding of the etiology and pathogenesis of LE is critical for developing novel treatment modalities aimed at a cure for the 1 in 1000 Americans affected by the disease [27]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283609 May 17, 2023 6 / 12 PLOS ONESkin microbiome in secondary lymphedema Table 3. Top 10 variable taxa found in microbial variability analysis. Top ten most variable taxa (p: phylum; g: genus; c: class) Mean Variability Mean Differential Abundance Taxa direction Normal (-) Lymphedema (+) Unadjusted p- value Adjusted p- value p__Actinobacteria g__Propionibacterium p__Firmicutes g__Streptococcus p__Firmicutes c__Bacilli p__Firmicutes g__Staphylococcus p__Actinobacteria g__Corynebacterium p__Firmicutes g__Veillonella p__Firmicutes g__Finegoldia p__Fusobacteria g__Fusobacterium p__Actinobacteria g__Kocuria p__Proteobacteria g__Xanthomonas https://doi.org/10.1371/journal.pone.0283609.t003 0.105 0.052 0.036 0.035 0.033 0.028 0.024 0.016 0.015 0.014 -0.041 0.004 0.009 0.013 0.008 0.018 -0.014 0.008 0.007 0.003 - + + + + + - + + - 0.237 0.702 0.540 0.418 0.394 0.038 0.479 0.822 0.780 0.480 0.793 0.893 0.793 0.793 0.793 0.793 0.793 0.921 0.905 0.793 Although LE appears to be mediated by a predominant Th2 inflammatory response, it is evi- dent that a combination of intrinsic and extrinsic factors plays a role in disease development and progression. In this study, we utilized high-throughput sequencing to investigate the role of the skin microbiome in LE pathophysiology. We found that differences in microbial compo- sition of the normal and LE limb is heterogeneous among patients with varying histories of infection and is related to relative limb volume changes between limbs. Bacterial dysbiosis, or a disruption in the balance of resident microbes, has been implicated in a variety of cutaneous diseases [28]. Particularly, in atopic dermatitis, a loss of microbial diversity is associated with disease severity [13]. Similarly, in filarial LE, the most common form of secondary LE worldwide caused by Wuchereria bancrofti infection, an increase in Staphylococcal aureus is observed in filarial skin when compared to skin of healthy controls Fig 3. Patient-specific taxa variability between normal and lymphedema limbs for the ten most variable taxa. (a) Stacked bar chart of patient-specific taxa variability, defined as the absolute difference of taxa relative abundance. (b) Directional bar chart of patient-specific difference of taxa relative abundance. (c) Directional bar chart of patient-specific difference of taxa relative abundance, stratified by history of infection. p: phylum; g: genus; c:class. https://doi.org/10.1371/journal.pone.0283609.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283609 May 17, 2023 7 / 12 PLOS ONESkin microbiome in secondary lymphedema Table 4. Top 10 genera with smallest unadjusted p-values obtained by the linear decomposition model (LDM) test of differential abundance for paired data. Genera with unadjusted p < 0.05 (p: phylum; g: genus) Mean Differential Abundance Taxa direction Normal (-) Lymphedema (+) Unadjusted p- value Adjusted p- value p__Proteobacteria g__Methylobacterium p__Proteobacteria g__Janthinobacterium p__Actinobacteria g__Arthrobacter p__Proteobacteria g__Rhodoplanes p__Proteobacteria g__Enhydrobacter p__Proteobacteria g__Sphingomonas p__Proteobacteria g__Pseudomonas p__Proteobacteria g__Caulobacter p__Firmicutes g__Veillonella p__Proteobacteria g__Brevundimonas https://doi.org/10.1371/journal.pone.0283609.t004 -0.006 -0.001 -0.000 -0.000 -0.002 -0.004 -0.001 -0.000 0.018 -0.008 - - - - - - - - + - 0.002 0.012 0.013 0.017 0.028 0.031 0.034 0.037 0.038 0.039 0.793 0.793 0.793 0.793 0.793 0.793 0.793 0.793 0.793 0.793 [21]. In contrast to the former study, our study aimed to characterize the bacterial skin micro- biome in individuals with non-filarial upper extremity secondary LE. Secondly, whereas the previous group relied on culture and mass spectrometry techniques to draw associations between skin commensal diversity and infection risk, we utilized high-throughput sequencing technology for taxa identification, which reduces the risk of underestimating or misidentifying the species present in a sample [21, 29]. An additional strength of our study, the paired sample study design, accounts for confounders that may otherwise be present when comparing to healthy controls. Our findings indicate that multiple genera of the phyla Firmicutes demonstrate high vari- ability between normal and LE samples (Table 3). Specifically, the most variable genera observed in the LE limb included Streptococcus, Staphylococcus, Veillonella, Fusobacterium, and Anaerococcus. The microbial variability analysis performed in this study is inspired by the concept of microbial volatility, which describes the temporal instability of the microbiome [30, 31]. Traditionally, volatility has been studied in the context of the gut microbiome, particularly as it relates to inflammatory bowel diseases [32]. Interestingly, observed volatility in intestinal physiology has been shown to influence inflammatory activity at distant organ sites, namely the skin barrier, leading to this concept of the gut-skin axis [32, 33]. The gut-skin axis, or the involvement of the gut microbiome in regulating health and disease states of the skin, has been linked with the development of chronic inflammatory skin conditions, such as psoriasis, rosa- cea, and acne [33]. Disturbances in the gut microbiome may contribute to the microbial vari- ability that we observe between patients with LE. However, further studies investigating the gastrointestinal health of patients that develop disease is warranted if a bidirectional relation- ship between gut dysbiosis and LE development is to be established. More recently, volatility has been studied in the context of microbial variations in response to elevated levels of stress, which is relevant to the microbiome-gut-brain axis [30]. Bastiaans- sen and colleagues observed significant positive correlations between chronic psychosocial stress and the degree of gut microbiome volatility in mice and humans. They speculate that hosts with the most volatile microbiomes are most susceptible to stress-associated symptoms. Although no causal link has been established between stress levels and LE development, chronic stress has been recognized as a barrier to effective management of LE [34]. Notably, in a single-center clinical trial evaluating the effect of combined psychosocial relaxation tech- niques and comprehensive decongestive therapy (R-CDT) to comprehensive decongestive therapy (CDT)alone on depression scores and the volume of edema in 31 patients with breast cancer-related lymphedema, a significant reduction in depression scores (p = 0.024) and a PLOS ONE \| https://doi.org/10.1371/journal.pone.0283609 May 17, 2023 8 / 12 PLOS ONESkin microbiome in secondary lymphedema downtrend in mean edema volume at 9-week follow-up (p = 0.470) was observed in the RCDT group when compared to CDT alone [35]. Taken together with the findings from our current study, it is possible that the variability observed between paired samples is influenced by psy- chosocial stress levels and that limb volume changes may, in part, contribute to microbial com- position detected in skin swabs. Given the dynamic nature of the microbiome and the fact that our samples were taken at a single timepoint, our observations of paired sample variability may not completely reflect the temporal variations of the microbial communities analyzed. We also utilized multivariable statistical modeling to study how clinical factors may influ- ence microbial composition between normal and LE limbs. Our results indicate a significant association between microbial distances (BCD and AD) and relative limb volume differential. The accumulation of lymph fluid can alter skin integrity and facilitate the entry of external pathogens; thus, this observation supports a likely relationship between the degree of arm swelling and bacterial populations present on the skin [36, 37]. One could hypothesize that the gradual increase in limb swelling over time observed in LE is in part related to a disruption of the bacterial microbiota. However, because we did not observe any consistent changes in the microbial composition, it is unclear if dysbiosis in favor of a single genera can serve as a marker for the disease. Kwarteng et al determined that seasonal variations in the microbiota, favoring a shift towards an over-population of Staphylococcus aureus is present in filarial lymphedema lesions. They speculate that the observed dysbiosis in combination with a diminished local skin immune system influences the infectious attacks that are frequent to this population. Addi- tionally, their study and others demonstrate that topographical location on the body is a defin- ing factor of bacterial diversity [38, 39]. In our study, skin swabs were obtained from the volar forearm of the LE and the normal limbs. Compared to other areas of the upper limb, the volar forearm is known to harbor a diverse microbial community, making it an ideal region for comparative sequencing studies at symmetric sites [38, 40, 41]. An interesting investigation would be to compare the microbial composition of the LE limb in areas where edema is most apparent and likely correlates with a weakened skin barrier. Using indocyanine green (ICG) lymphography, a minimally invasive diagnostic tool that shows patterns of dermal backflow, may help facilitate this type of study [42, 43]. In recognizing study limitations, the small sample size of 30 patients limited the statistical power of testing taxa differential abundance. Additionally, including multiple timepoints for skin swab analysis and varying locations for sample collection along the lymphedema limb would enhance the robustness of this study. In addition, although 16S amplicon sequencing is a standard approach for characterizing the taxonomic profile of the microbiome in LE, utiliza- tion of shotgun metagenomics technology would better discriminate those bacterial communi- ties that play a functional role in the disease process [44, 45]. Future studies with larger sample sizes will allow deeper investigation of differential abundance, patient heterogeneity, and lon- gitudinal dynamics of the microbiota associated with lymphedema. In conclusion, 16S rRNA microbiome sequencing shows high compositional heterogeneity in the skin microbiome between the normal and diseased limbs among patients with upper extremity secondary LE. We encourage further studies into host-microbiome interactions in secondary LE, with a focus on the implications for LE diagnosis and management. Supporting information S1 Table. Genus-level relative abundance data for paired lymphedema and normal sam- ples. (XLSX) PLOS ONE \| https://doi.org/10.1371/journal.pone.0283609 May 17, 2023 9 / 12 PLOS ONESkin microbiome in secondary lymphedema Acknowledgments The authors would like to acknowledge the technical assistance provided by the University of Pennsylvania and Children’s Hospital Philadelphia, PennCHOP Microbiome program. Author Contributions Conceptualization: Adana-Christine Campbell, Jung Eun Baik, Ananta Sarker. Data curation: Adana-Christine Campbell, Teng Fei, Jinyeon Shin. Formal analysis: Teng Fei, Jinyeon Shin. Investigation: Adana-Christine Campbell, Jung Eun Baik. Methodology: Teng Fei. Resources: Stav Brown. Software: Teng Fei. Supervision: Hyeung Ju Park, Raghu P. Kataru, Babak J. Mehrara. Writing – original draft: Adana-Christine Campbell. Writing – review & editing: Hyeung Ju Park, Kevin Kuonqui, Ananta Sarker, Raghu P. Kataru, Babak J. Mehrara. References 1. Li CY, Kataru RP, Mehrara BJ. Histopathologic Features of Lymphedema: A Molecular Review. Int J Mol Sci. 2020; 21(7). https://doi.org/10.3390/ijms21072546 PMID: 32268536 2. Yuan Y, Arcucci V, Levy SM, Achen MG. Modulation of Immunity by Lymphatic Dysfunction in Lymph- edema. Front Immunol. 2019; 10:76. https://doi.org/10.3389/fimmu.2019.00076 PMID: 30761143 3. Vignes S, Poizeau F, Dupuy A. Cellulitis risk factors for patients with primary or secondary lymph- edema. J Vasc Surg Venous Lymphat Disord. 2022; 10(1):179–85.e1. https://doi.org/10.1016/j.jvsv. 2021.04.009 PMID: 33957278 4. Wiser I, Mehrara BJ, Coriddi M, Kenworthy E, Cavalli M, Encarnacion E, et al. Preoperative Assess- ment of Upper Extremity Secondary Lymphedema. Cancers (Basel). 2020; 12(1). https://doi.org/10. 3390/cancers12010135 PMID: 31935796 5. Connor MP, Gamelli R. Challenges of cellulitis in a lymphedematous extremity: a case report. Cases J. 2009; 2:9377. https://doi.org/10.1186/1757-1626-2-9377 PMID: 20062550 6. Lopez M, Roberson ML, Strassle PD, Ogunleye A. Epidemiology of Lymphedema-related admissions in the United States: 2012–2017. Surg Oncol. 2020; 35:249–53. https://doi.org/10.1016/j.suronc.2020. 09.005 PMID: 32932222 7. Garcı´a Nores GD, Ly CL, Savetsky IL, Kataru RP, Ghanta S, Hespe GE, et al. Regulatory T Cells Medi- ate Local Immunosuppression in Lymphedema. J Invest Dermatol. 2018; 138(2):325–35. https://doi. org/10.1016/j.jid.2017.09.011 PMID: 28942366 8. Huang JJ, Gardenier JC, Hespe GE, Garcı´a Nores GD, Kataru RP, Ly CL, et al. Lymph Node Trans- plantation Decreases Swelling and Restores Immune Responses in a Transgenic Model of Lymph- edema. PLoS One. 2016; 11(12):e0168259. https://doi.org/10.1371/journal.pone.0168259 PMID: 27942023 9. Ba¨ sler K, Brandner JM. Tight junctions in skin inflammation. Pflugers Arch. 2017; 469(1):3–14. https:// doi.org/10.1007/s00424-016-1903-9 PMID: 27853878 10. Biedermann T, Skabytska Y, Kaesler S, Volz T. Regulation of T Cell Immunity in Atopic Dermatitis by Microbes: The Yin and Yang of Cutaneous Inflammation. Front Immunol. 2015; 6:353. https://doi.org/ 10.3389/fimmu.2015.00353 PMID: 26217343 11. Gallo RL. Human Skin Is the Largest Epithelial Surface for Interaction with Microbes. J Invest Dermatol. 2017; 137(6):1213–4. PLOS ONE \| https://doi.org/10.1371/journal.pone.0283609 May 17, 2023 10 / 12 PLOS ONESkin microbiome in secondary lymphedema 12. Fife CE, Farrow W, Hebert AA, Armer NC, Stewart BR, Cormier JN, et al. Skin and Wound Care in Lymphedema Patients: A Taxonomy, Primer, and Literature Review. Adv Skin Wound Care. 2017; 30 (7):305–18. https://doi.org/10.1097/01.ASW.0000520501.23702.82 PMID: 28617750 13. Nakatsuji T, Gallo RL. The role of the skin microbiome in atopic dermatitis. Ann Allergy Asthma Immu- nol. 2019; 122(3):263–9. https://doi.org/10.1016/j.anai.2018.12.003 PMID: 30550810 14. Gao Z, Tseng CH, Strober BE, Pei Z, Blaser MJ. Substantial alterations of the cutaneous bacterial biota in psoriatic lesions. PLoS One. 2008; 3(7):e2719. https://doi.org/10.1371/journal.pone.0002719 PMID: 18648509 15. Gupta M, Weinberg JM, Yamauchi PS, Patil A, Grabbe S, Goldust M. Psoriasis: Embarking a dynamic shift in the skin microbiota. J Cosmet Dermatol. 2022; 21(4):1402–6. https://doi.org/10.1111/jocd.14273 PMID: 34089221 16. van Rensburg JJ, Lin H, Gao X, Toh E, Fortney KR, Ellinger S, et al. The Human Skin Microbiome Asso- ciates with the Outcome of and Is Influenced by Bacterial Infection. mBio. 2015; 6(5):e01315–15. https://doi.org/10.1128/mBio.01315-15 PMID: 26374122 17. Edslev SM, Agner T, Andersen PS. Skin Microbiome in Atopic Dermatitis. Acta Derm Venereol. 2020; 100(12):adv00164. https://doi.org/10.2340/00015555-3514 PMID: 32419029 18. Williams MR, Gallo RL. The role of the skin microbiome in atopic dermatitis. Curr Allergy Asthma Rep. 2015; 15(11):65. https://doi.org/10.1007/s11882-015-0567-4 PMID: 26404536 19. SanMiguel A, Grice EA. Interactions between host factors and the skin microbiome. Cell Mol Life Sci. 2015; 72(8):1499–515. https://doi.org/10.1007/s00018-014-1812-z PMID: 25548803 20. Wensel CR, Pluznick JL, Salzberg SL, Sears CL. Next-generation sequencing: insights to advance clini- cal investigations of the microbiome. J Clin Invest. 2022; 132(7). https://doi.org/10.1172/JCI154944 PMID: 35362479 21. Kwarteng A, Wireko S, Asiedu SO, Kini P, Aglomasa BC, Amewu EKA, et al. Shift in the skin micro- biome among individuals presenting with filarial lymphedema compared to non-filarial healthy individu- als in Ghana. Scientific African. 2022; 16:e01237. 22. Bolyen E, Rideout JR, Dillon MR, Bokulich NA, Abnet CC, Al-Ghalith GA, et al. Reproducible, interac- tive, scalable and extensible microbiome data science using QIIME 2. Nat Biotechnol. 2019; 37(8):852– 7. https://doi.org/10.1038/s41587-019-0209-9 PMID: 31341288 23. Callahan BJ, McMurdie PJ, Rosen MJ, Han AW, Johnson AJ, Holmes SP. DADA2: High-resolution sample inference from Illumina amplicon data. Nat Methods. 2016; 13(7):581–3. https://doi.org/10. 1038/nmeth.3869 PMID: 27214047 24. McDonald D, Price MN, Goodrich J, Nawrocki EP, DeSantis TZ, Probst A, et al. An improved Green- genes taxonomy with explicit ranks for ecological and evolutionary analyses of bacteria and archaea. Isme j. 2012; 6(3):610–8. https://doi.org/10.1038/ismej.2011.139 PMID: 22134646 25. Zhu Z, Satten GA, Mitchell C, Hu YJ. Constraining PERMANOVA and LDM to within-set comparisons by projection improves the efficiency of analyses of matched sets of microbiome data. Microbiome. 2021; 9(1):133. https://doi.org/10.1186/s40168-021-01034-9 PMID: 34108046 26. Sandve GK, Ferkingstad E, Nygård S. Sequential Monte Carlo multiple testing. Bioinformatics. 2011; 27(23):3235–41. https://doi.org/10.1093/bioinformatics/btr568 PMID: 21998154 27. Sleigh BC, Manna B. Lymphedema. StatPearls. Treasure Island (FL): StatPearls Publishing Copyright © 2022, StatPearls Publishing LLC.; 2022. 28. DeGruttola AK, Low D, Mizoguchi A, Mizoguchi E. Current Understanding of Dysbiosis in Disease in Human and Animal Models. Inflamm Bowel Dis. 2016; 22(5):1137–50. https://doi.org/10.1097/MIB. 0000000000000750 PMID: 27070911 29. Hiergeist A, Gla¨sner J, Reischl U, Gessner A. Analyses of Intestinal Microbiota: Culture versus Sequencing. Ilar j. 2015; 56(2):228–40. https://doi.org/10.1093/ilar/ilv017 PMID: 26323632 30. Bastiaanssen TFS, Gururajan A, van de Wouw M, Moloney GM, Ritz NL, Long-Smith CM, et al. Volatil- ity as a Concept to Understand the Impact of Stress on the Microbiome. Psychoneuroendocrinology. 2021; 124:105047. https://doi.org/10.1016/j.psyneuen.2020.105047 PMID: 33307493 31. Weinstock GM. The volatile microbiome. Genome Biol. 2011; 12(5):114. https://doi.org/10.1186/gb- 2011-12-5-114 PMID: 21635714 32. Halfvarson J, Brislawn CJ, Lamendella R, Va´zquez-Baeza Y, Walters WA, Bramer LM, et al. Dynamics of the human gut microbiome in inflammatory bowel disease. Nat Microbiol. 2017; 2:17004. https://doi. org/10.1038/nmicrobiol.2017.4 PMID: 28191884 33. Thye AY, Bah YR, Law JW, Tan LT, He YW, Wong SH, et al. Gut-Skin Axis: Unravelling the Connection between the Gut Microbiome and Psoriasis. Biomedicines. 2022; 10(5). https://doi.org/10.3390/ biomedicines10051037 PMID: 35625774 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283609 May 17, 2023 11 / 12 PLOS ONESkin microbiome in secondary lymphedema 34. Ridner SH. The psycho-social impact of lymphedema. Lymphat Res Biol. 2009; 7(2):109–12. https:// doi.org/10.1089/lrb.2009.0004 PMID: 19534633 35. Abbasi B, Mirzakhany N, Angooti Oshnari L, Irani A, Hosseinzadeh S, Tabatabaei SM, et al. The effect of relaxation techniques on edema, anxiety and depression in post-mastectomy lymphedema patients undergoing comprehensive decongestive therapy: A clinical trial. PLoS One. 2018; 13(1):e0190231. https://doi.org/10.1371/journal.pone.0190231 PMID: 29304095 36. Jones D, Meijer EFJ, Blatter C, Liao S, Pereira ER, Bouta EM, et al. Methicillin-resistant Staphylococcus aureus causes sustained collecting lymphatic vessel dysfunction. Sci Transl Med. 2018; 10(424). https://doi.org/10.1126/scitranslmed.aam7964 PMID: 29343625 37. Schommer NN, Gallo RL. Structure and function of the human skin microbiome. Trends Microbiol. 2013; 21(12):660–8. https://doi.org/10.1016/j.tim.2013.10.001 PMID: 24238601 38. Gao Z, Tseng CH, Pei Z, Blaser MJ. Molecular analysis of human forearm superficial skin bacterial biota. Proc Natl Acad Sci U S A. 2007; 104(8):2927–32. https://doi.org/10.1073/pnas.0607077104 PMID: 17293459 39. Zeeuwen PL, Boekhorst J, van den Bogaard EH, de Koning HD, van de Kerkhof PM, Saulnier DM, et al. Microbiome dynamics of human epidermis following skin barrier disruption. Genome Biol. 2012; 13(11): R101. https://doi.org/10.1186/gb-2012-13-11-r101 PMID: 23153041 40. Costello EK, Lauber CL, Hamady M, Fierer N, Gordon JI, Knight R. Bacterial community variation in human body habitats across space and time. Science. 2009; 326(5960):1694–7. https://doi.org/10. 1126/science.1177486 PMID: 19892944 41. Ruocco V, Ruocco E, Brunetti G, Sangiuliano S, Wolf R. Opportunistic localization of skin lesions on vul- nerable areas. Clin Dermatol. 2011; 29(5):483–8. https://doi.org/10.1016/j.clindermatol.2011.03.003 PMID: 21855722 42. de Godoy JM, de Godoy MF, Valente A, Camacho EL, Paiva EV. Lymphoscintigraphic evaluation in patients after erysipelas. Lymphology. 2000; 33(4):177–80. PMID: 11191659 43. Mihara M, Hara H, Araki J, Kikuchi K, Narushima M, Yamamoto T, et al. Indocyanine green (ICG) lym- phography is superior to lymphoscintigraphy for diagnostic imaging of early lymphedema of the upper limbs. PLoS One. 2012; 7(6):e38182. https://doi.org/10.1371/journal.pone.0038182 PMID: 22675520 44. Bharti R, Grimm DG. Current challenges and best-practice protocols for microbiome analysis. Brief Bioinform. 2021; 22(1):178–93. https://doi.org/10.1093/bib/bbz155 PMID: 31848574 45. Stehlikova Z, Kostovcik M, Kostovcikova K, Kverka M, Juzlova K, Rob F, et al. Dysbiosis of Skin Micro- biota in Psoriatic Patients: Co-occurrence of Fungal and Bacterial Communities. Front Microbiol. 2019; 10:438. https://doi.org/10.3389/fmicb.2019.00438 PMID: 30949136 PLOS ONE \| https://doi.org/10.1371/journal.pone.0283609 May 17, 2023 12 / 12 PLOS ONE
10.1371_journal.pone.0286598	RESEARCH ARTICLE Enteral nutrition management in critically ill adult patients and its relationship with intensive care unit-acquired muscle weakness: A national cohort study Ignacio Zaragoza-Garcı´aID Daniel Martı´5☯, Elisabet Gallart6☯, Alicia San Jose´ -Arribas7☯, Tamara Raquel Velasco- Sanz1,8☯, Eva Blazquez-Martı´nez9☯, Marta Raurell-Torredà10☯ 1,2☯, Susana Arias-Rivera3☯, Marı´a Jesu´ s Frade-Mera1,4☯, Joan 1 Department of Nursing, Faculty of Nursing, Physiotherapy and Podology, University Complutense of Madrid, Madrid, Spain, 2 Invecuid, Instituto de Investigacio´n Sanitaria Hospital 12 de Octubre (imas12), Madrid, Spain, 3 University Hospital of Getafe, CIBER Enfermedades Respiratorias, Instituto de Salud Carlos III, Getafe, Spain, 4 Department of Critical Care, 12 Octubre University Hospital, Madrid, Spain, 5 Clinic University Hospital, Barcelona, Spain, 6 Department of Critical Care, Vall Hebron University Hospital, Barcelona, Spain, 7 Escola Universitaria d’Infermeria Sant Pau, Hospital de la Santa Creu i Sant Pau, Barcelona, Spain, 8 Department of Critical Care, San Carlos University Hospital, Madrid, Spain, 9 Bellvitge University Hospital, Hospitalet de Llobregat, Llobregat, Spain, 10 Department d’Infermeria Fonamental i medicoquiru´ rgica, Facultat d’Infermeria, Universitat de Barcelona, Barcelona, Spain ☯ These authors contributed equally to this work. izaragoz@ucm.es Abstract Objective To assess the incidence and determinants of ICU-acquired muscle weakness (ICUAW) in adult patients with enteral nutrition (EN) during the first 7 days in the ICU and mechanical ventilation for at least 48 hours. Methods A prospective, nationwide, multicentre cohort study in a national ICU network of 80 ICUs. ICU patients receiving invasive mechanical ventilation for at least 48 hours and EN the first 7 days of their ICU stay were included. The primary outcome was incidence of ICUAW. The secondary outcome was analysed, during days 3–7 of ICU stay, the relationship between demographic and clinical data to contribute to the onset of ICUAW, identify whether energy and protein intake can contribute independently to the onset of ICUAW and degree of com- pliance guidelines for EN. Results 319 patients were studied from 69 ICUs in our country. The incidence of ICUAW was 153/ 222 (68.9%; 95% CI [62.5%-74.7%]). Patients without ICUAW showed higher levels of active mobility (p = 0.018). The logistic regression analysis showed no effect on energy or protein intake on the onset of ICUAW. Overfeeding was observed on a significant proportion a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Zaragoza-Garcı´a I, Arias-Rivera S, Frade- Mera MJ, Martı´ JD, Gallart E, San Jose´-Arribas A, et al. (2023) Enteral nutrition management in critically ill adult patients and its relationship with intensive care unit-acquired muscle weakness: A national cohort study. PLoS ONE 18(6): e0286598. https://doi.org/10.1371/journal.pone.0286598 Editor: Sebastien Kenmoe, University of Buea, CAMEROON Received: February 11, 2023 Accepted: May 19, 2023 Published: June 7, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0286598 Copyright: © 2023 Zaragoza-Garcı´a et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286598 June 7, 2023 1 / 14 PLOS ONEFunding: This work was supported by 2018 european federation of critical care nursing associations (EfCCNa) Research Awards. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Enteral nutrition management and ICUAW of patient-days, while more overfeeding (as per US guidelines) was found among patients with obesity than those without (42.9% vs 12.5%; p<0.001). Protein intake was deficient (as per US/European guidelines) during ICU days 3–7. Conclusions The incidence of ICUAW was high in this patient cohort. Early mobility was associated with a lower incidence of ICUAW. Significant overfeeding and deficient protein intake were observed. However, energy and protein intake alone were insufficient to explain ICUAW onset. Relevance to clinical practice Low mobility, high incidence of ICUAW and low protein intake suggest the need to train, update and involve ICU professionals in nutritional care and the need for early mobilization of ICU patients. Introduction Patients admitted to Intensive Care Units (ICUs) are subject to increased metabolic stress. Ele- vated catabolism requires nutritional resources for the body to perform anabolism adequately [1]. If oral intake is not possible, enteral nutrition (EN) is recommended over parenteral nutri- tion, because it has fewer complications [2]. Inappropriate management of enteral nutrition support in these patients can lead to malnutrition, a common finding in ICU patients [3], for which the incidence ranges from 39% to 50% of patients, depending on the country and ICU type [4]. Various authors have described possible causes of malnutrition in critically ill patients. Delayed initiation of nutrition support has been found in 60% of cases. In addition, an incor- rect EN regimen can lead to under- or overfeeding, which, together with the inflammatory response typical for this metabolic state, can contribute to hyperglycaemia, loss of muscle mass and strength, prolonged rehabilitation, as well as an increase in comorbidities resulting in deteriorated quality of life in the long term [5]. Loss of muscle mass together with other factors, such as physical immobility, can lead to the onset of bilateral and symmetric neuromuscular complications, referred to as ICU- acquired muscle weakness (ICUAW), which contributes to significant functional impairment. Specifically, the muscles of the limbs and the diaphragm may become weak and atrophic, impairing patients’ autonomy, prolonging mechanical ventilation, and increasing weaning time and length of hospital stays [6, 7]. The most studied predictors in ICUAW are related to gender, time on mechanical ventila- tion, length of ICU stay, age, more days on renal replacement therapy. On the other hand, the presence of delirium and being actively mobilised during the first 5 days in the ICU are consid- ered protective factors [8, 9]. Some international bodies specialised in EN suggest the need for research on the relationship between EN and ICUAW, but due to lack of evidence, they do not yet make any recommendations in this regard [2, 10]. As a result, various international nutrition-related societies publish specific recommenda- tions for critically ill patients. Recent studies suggest that diet-only interventions are insuffi- cient to improve patients’ nutritional status and reduce comorbidities, and this is now reflected in current recommendations [2]. To mitigate this deterioration, early mobilization in PLOS ONE \| https://doi.org/10.1371/journal.pone.0286598 June 7, 2023 2 / 14 PLOS ONEEnteral nutrition management and ICUAW the ICU is recommended [5]. The combination of nutrition plus exercise may modify the cata- bolic effects of critical illness, muscle wasting, and the development of ICUAW, which has been identified as a research priority [11]. Currently, no national multicentre studies have evaluated the management of EN in criti- cally ill patients or the degree of mobility of these patients related to the incidence of ICUAW. The aim of this study was to assess the incidence and determinants of ICUAW in adult patients with EN during the first 7 days in the ICU and receiving mechanical ventilation for at least 48 hours. Materials and methods Design A prospective multicentre observational cohort study was conducted during four months (2019–2020) in a Spanish national ICU network of 80 ICUs. Data collection Patients were recruited consecutively. The data were collected starting from day 3 of ICU admission. Inclusion criteria were adult patients receiving invasive mechanical ventilation (IMV) for at least 48 hours in an ICU and EN for at least the first 7 days of their ICU stay. Exclusion criteria were pregnant women, patients <18 years, those referred to the ICU from other hospitals, patients with primary neurologic or neuromuscular pathology, those unable to walk, recent limb amputees, users of orthopaedic devices and patients with body mass index (BMI) >35. Sample/Participants The minimum sample size was 316, calculated according to the 46% incidence of ICUAW found in a sample of 1421 patients by Stevens et al. [12], a confidence level of 95%, an esti- mated standard error of 5 and an expected loss of 5%. Ethical considerations The study was approved by the Ethics and Clinical Research Committees of the participating sites under reference protocol PI16/00771. Written informed consent was obtained. The rele- vant STROBE checklist was followed for reporting the study. Research variables and measures Primary outcome. The primary outcome was incidence of ICUAW, assessed by the Med- ical Research Council Scale (MRC-Sum score) following the assessment protocol described by Hermans [13]. ICUAW was diagnosed for values lower than 48 out of 60 (the maximum score) in the first measure of MRC (baseline MRC) [14]. The measure of MRC was conducted after the first awakening of the patient, with the patient fully awake. See S1 File. Measurement tools. Secondary outcomes. The secondary outcome were, on the one hand, analysis of the rela- tionship between demographic and clinical data contributing to the onset of ICUAW during days 3–7 of ICU stay. On the other hand, we proceeded to identify whether energy or protein intake during 3–7 days of the ICU stay, taking into account the US and European recommen- dation, can contribute independently to the onset of ICUAW. Finally, the degree of compli- ance with current US and European guidelines for target dietary intake in EN during the acute phase (days 3–7) of ICU admission was analysed. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286598 June 7, 2023 3 / 14 PLOS ONEEnteral nutrition management and ICUAW Specifically, the following recommendations were used for reference in the study [2, 15]: Target energy and protein intake: According to ASPEN (American Society for Parenteral and Enteral Nutrition) guidelines [15]: target energy and protein intake should be 25–30 kcal/ kg/day and 1.2–2 g/kg/day, respectively. During the first week, trophic EN is permitted. For patients with BMI �30 kg/m2 the energy target is 11–14 kcal/kg/day actual body weight/day and the protein target is 2 g/kg ideal body weight/day. According to ESPEN (European Society for Clinical Nutrition and Metabolism) guidelines [2]: target energy and protein intake is 20–25 kcal/kg/day and 1.3 g/kg/day delivered progres- sively, respectively. During the first week, trophic EN is permitted. Actual body weight is used for patients with BMI �25 kg/m2 and adjusted body weight for BMI >25 kg/m2. Other recommendations discussed were interruptions to EN should be avoided. It is rec- ommended that stopping feeding to evaluate oral tolerance should be limited to once daily at the most. In addition, gastric residual volume < 500 mL indicates EN tolerance. Finally, insu- lin therapy should be used to control blood glucose levels and blood glucose levels should be maintained at <180 mg/dL. Variables. Independent variables related to the patient’s baseline condition as well as hos- pital admission variables were collected. Specifically, age, gender, and BMI, diagnosis on admission, Barthel and Charlson index, and APACHE II scores were collected. All parameters were collected from the medical records by a collaborating research nurse. See supplementary material for definitions and classifications. The principal dependent variable that was collected was presence of ICUAW according to MRC sum-score, conducted by a physiotherapist. Secondary dependent variables were energy and protein intake via EN, level of mobility, continuous renal replacement therapy (CRRT), airway management, ICUAW-related drugs, vasopressors, moderate and severe hyperglycaemia. All these variables were collected during days 3–7 of ICU stay: ICU Mobility Scale (IMS) score. IMS is a 10-point scale ranges from 0 (patient immobile lying in bed) to 10 (independent ambulation). The IMS was categorized using a binary system (where <4 represents, in-bed activities, and �4, active out-of-bed mobilization); Days on which the patient requires CRRT; Type of airway management (invasive mechanical ventilation (IMV) or no IMV); ICUAW- related drugs, understood as cumulative doses of drugs such as neuromuscular blocking agents, steroids [methylprednisolone, dexamethasone, and hydrocortisone in mg equivalent dose] and aminoglycosides; Administered doses of Vasopressors (epinephrine, adrenaline, noradrenaline, dopamine and dobutamine). In both cases above intravenous administration is considered (continuous infusion, stat dose, and bolus injection on demand); Moderate (gly- caemia >181 and �215 mg/dl) or severe hyperglycaemia (�216 mg/dl) of the total blood glu- cose results on day 3–7 of the ICU stay multiplied by 100. A team of trained professionals recorded the variables. Detailed of the measurement tools are provided in the S1 File. Data analysis. Categorical variables were expressed as frequency and percentage, using Fisher or Chi-squared test for comparison between groups. Quantitative variables were expressed as mean and standard deviation (SD) or median and interquartile range (IQR), and groups were compared using Student-t or Mann–Whitney U test. To study the correlation between quantitative variables (actual body weight, energy and protein intake), Pearson or Spearman was used. A multivariate analysis was used to investigate the association between EN during 3–7 days of the ICU stay (energy and protein administration, days with overfeeding and days with protein >0.8 g/kg/day) and ICUAW, also controlling other explanatory vari- ables: baseline variables (age, gender, BMI, Barthel and Charlson scores) and those related to ICUAW onset during days 3–7 of the ICU stay (days with CCRT, doses of ICUAW-related PLOS ONE \| https://doi.org/10.1371/journal.pone.0286598 June 7, 2023 4 / 14 PLOS ONEEnteral nutrition management and ICUAW drugs and vasopressors, days with IMS �4, and days with moderate and severe hyperglycae- mia). Data were analysed using SPSS 25.0. Results We analysed 319 patients, corresponding to 1595 EN days and 69 ICUs in our country (Fig 1). The incidence of ICUAW was 68.9% (153/222 patients; 95% CI [62.5%-74.7%]). In 30.4% (97/319 patients; 95% CI [25.6%-35.7%]), the MRC assessment was unfeasible. Among the patients with ICUAW, females were at higher risk than males and the most prevalent diagnosis was sepsis. Patients with ICUAW had higher rates of comorbidity (Charlson), were more dependent (Barthel) and had greater disease severity (APACHE), but these results were not statistically significant (Table 1). More overweight and, conversely, fewer obese patients devel- oped ICUAW (p<0.05 in both cases) (Table 1). On ICU days 3–7, although the general cohort had low active mobility out of bed (IMS�4), ICUAW patients had significantly lower values during this period (p<0.001) (Table 2). In addition, ICUAW patients received significantly more vasopressors (p = 0.029) and had more days of IMV (p = 0.032). No significant differences were found in median days of CRRT, of severe or moderate hyperglycaemia, or of administration of ICUAW-related drugs (Table 2) in the same period. Of the patients who developed ICUAW, 67.4% (273/405 patient/days) had deep sedation (RASS-3-5) vs 47.7% (83/174 patient/days) of those who did not have ICUAW. Patients in whom ICUAW could not be assessed had significantly more days of deep sedation than those in whom ICUAW could be assessed (87.3% (261/299) vs 61.5% (356/579); p<0.001). Energy intake during days 3–7 was similar among patients who did and did not develop ICUAW, independently of which guidelines were followed (ASPEN or ESPEN). Likewise, no differences were observed in the percentage of patients with overfeeding or number of days of overfeeding when comparing patients with and without ICUAW (Table 3). Patients receiving propofol had a median energy intake of 188.0 kcal/day [62.7–380.6 kcal/day] over a total of 727 patient/days. With regard to protein intake during days 3–7 in all groups, independently of which guide- lines were followed (ASPEN or ESPEN), no differences were observed between number of days with >0.8 g/kg/day and ICUAW onset (Table 3). Median protein intake was below 0.8 g/kg/day, and the median in obese patients with ICUAW (as per ASPEN guidelines) was closer to the recommended value, at 0.77 g/kg/day [0.48–0.99] (Table 3). The logistic regression analysis for ICU days 3–7 showed no effect of energy or protein intake on the onset of ICUAW. Neither could ICUAW be explained by the increase in days with overfeeding (S2 File). The days with overfeeding on ICU days 3–7 showed an OR of 1.085 [0.934–1.261]; p = 0.286. This remained after adjusting for baseline variables (OR: 1.109 [0.948–1.296]; p = 0.197). The results were similar after adjusting for ICU stay variables (OR: 1.106 [0.945–1.294]; 0.209) and after adjusting for all variables (baseline and ICU stay vari- ables) (OR:1.128 [0.956–1.332]; p = 0.154). Median daily energy intake was close to recommended levels during the first week of the ICU stay, except for obese patients, who were found to receive slightly above the recom- mended energy intake levels according to the US guidelines (S1 Fig). The degree of compliance with energy intake depends on which recommendations are con- sidered. Overfeeding was observed according to US and European guidelines. Using the US guidelines and considering patients with BMI <30kg/m2, overfeeding was found on 12.5% patients/day; 95% CI [10.8%-14.5%] whereas for patients with BMI �30kg/m2 the rate was PLOS ONE \| https://doi.org/10.1371/journal.pone.0286598 June 7, 2023 5 / 14 PLOS ONEEnteral nutrition management and ICUAW Fig 1. Flow diagram showing patients’ movement through the study. https://doi.org/10.1371/journal.pone.0286598.g001 42.9% patients/day; 95% CI [38.0%-48.0%] and according to the European guidelines the rate was 43.1% patients/day; 95% CI [40.7%-45.5%] (S1 Table). Median protein administration was low throughout days 3–7 of the ICU stay according to US and European guidelines (S1 Fig and S1 Table). A third of patients received less than 0.5 g/ kg/day of protein during the first week (S1 Table). A high percentage of patient-days showed glycaemia <180mg/dl. Patients without ICUAW had a significantly higher proportion of patient-days with glycaemia <180 mg/dl (p = 0.006) PLOS ONE \| https://doi.org/10.1371/journal.pone.0286598 June 7, 2023 6 / 14 PLOS ONEEnteral nutrition management and ICUAW Table 1. General characteristics of the study population. Patients with EN who developed ICUAW n = 153 (48.0%) Patients with EN who did NOT develop ICUAW n = 69 (21.6%) Female Sepsis Trauma Neurosurgery Cardiovascular surgery Other surgeries Overdose Other medical patients Underweight Normal Overweight Obese Gender Age, years Dx. on Admission BMI (kg/m2) BMI Barthel Charlson index APACHE IIa 52 (34.0%) 68.0 [55.0–76.0] 32 (20.9%) 10 (6.5%) 4 (2.6%) 15 (9.8%) 18 (11.8%) 3 (2.0%) 71 (46.4%) 27.1 [24.3–30.3] 2 (1.3%) 46 (30.1%) 66 (43.1%) 39 (25.5%) 100 [95–100] 5.0 [2.0–7.0] 23 [18–28] 14 (20.3%) 63.0 [47.5–74.5] 9 (13.0%) 7 (10.1%) 0 (0%) 4 (5.8%) 8 (11.6%) 2 (2.9%) 39 (56.5%) 26.7 [24.0–30.8] 0 (0.0%) 23 (33.3%) 24 (34.8%) 22 (31.9%) 100 [95–100] 4.0 [1.0–6.0] 21 [16–27] p value 0.041 0.079 <0.001 0.467 - 0.012 0.050 0.655 0.002 0.752 - 0.006 <0.001 0.030 0.933 0.247 0.537 EN: enteral nutrition; ICUAW: intensive care unit-acquired muscle weakness; n: sample; %: percentage; BMI: body mass index; MRC: Medical Research Council scale. aAPACHE II (assessed in 45 patients without ICUAW, 67 with ICUAW and 59 with missing ICUAW data). Categorical variables are expressed as frequency and percentage (n (%)) and quantitative variables with non-normal distribution as median [25th -75th percentile] https://doi.org/10.1371/journal.pone.0286598.t001 (S2 Table). On most patient/days there was one or zero interruptions or pauses in EN. Gastric residual volume (GRV) was <500 ml on most patient-days. No differences were found between patients with or without ICUAW for glycaemia or GRV (S2 Table). A weak correlation was found between patients’ actual body weight and energy (kcal/kg/ day) (r = -0.121; p<0.031) and proteins (g/kg/day) (r = -0.112; p<0.045) delivery. Table 2. ICU variables, by ICUAW onset on days 3–7 of ICU stay. Days with IMS�4 Days with CRRT Days according to Airway management • IMV • no IMV ICUAW-related drugs (mg) Vasopressors (mg) Moderate hyperglycaemia (rate) Severe hyperglycaemia (rate) Patients with EN who developed ICUAW n = 153 (48.0%) Patients with EN who did NOT develop ICUAW n = 69 (21.6%) 0.0 [0.0–0.0] / 0.02±0.14 0.0 [0–0] 0.0 [0.0–0.0] / 0.12±0.44 0.0 [0–0] 5.0 [5.0–5.0] 0.0 [0.0–0.0] 60.0 [0.0–307.4] 24.2 [0.0–123.5] 5.6 [0.0–20.9] 0.0 [0.0–18.0] 5.0 [4.0–5.0] 0.0 [0.0–1.0] 0.0 [0.0–240.0] 7.7 [0.0–39.2] 5.9 [0.0–17.7] 0.0 [0.0–11.5] p value <0.001 0.327 0.032 0.028 0.111 0.029 0.386 0.223 EN: enteral nutrition; ICUAW: intensive care unit-acquired muscle weakness; n: sample; %: percentage; IMV: invasive mechanical ventilation; IMS: ICU mobility scale; CRRT: continuous renal replacement therapy; SD: standard deviation. Quantitative variables are expressed as median [25th -75th percentile] or mean and SD. https://doi.org/10.1371/journal.pone.0286598.t002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286598 June 7, 2023 7 / 14 PLOS ONETable 3. Energy and protein intake via EN and association with ICUAW on ICU days 3–7. Patients with EN who developed ICUAW n = 153 (48.0%) Patients with EN who did NOT develop ICUAW n = 69 (21.6%) p value Enteral nutrition management and ICUAW Energy. 2016 ASPEN guidelines Energy (BMI <30 kg/m2) Overfeeding (BMI <30 kg/m2)n (%) Days with overfeeding (BMI <30 kg/m2) Energy (BMI >30 kg/m2) Overfeeding (BMI >30 kg/m2)n (%) Days with overfeeding (BMI >30 kg/m2) Energy. 2019 ESPEN guidelines Energy Overfeeding n (%) Days with overfeeding Protein. 2016 ASPEN guidelines Prot (BMI <30 kg/m2) Days with protein >0.8 g/kg/day (BMI <30 kg/m2) Protein (BMI >30 kg/m2) Days with protein >0.8 g/kg/day (BMI >30 kg/m2) Protein. 2019 ESPEN guidelines Protein Days with protein >0.8 g/kg/day 16.8 [10.6–22.4] 10 (8.8%) 0.0 / [0.0–0.0] 12.6 [9.1–18.3] 12 (30.8%) 2.0 [0.0–4.0] 16.7 [10.7–23.0] 62 (40.5%) 2 [0–4] 0.65 [0.46–0.89] 1.0 [0.0–3.3] 0.77 [0.48–0.99] 3.0 [0.0–4.0] 0.69 [0.46–0.92] 2 [0–4] 16.2 [11.1–22.5] 5 (10.6%) 0.0 / [0.0–0.0] 12.0 [9.7–17.2] 8 (36.4%) 1.5 [0.0–3.0] 15.9 [12.0–22.4] 26 (37.7%) 1 [0–3] 0.69 [0.44–0.91] 2.0 [0.0–4.0] 0.69 [0.52–0.87] 1.0 [0.0–3.0] 0.68 [0.46–0.91] 2 [0–4] 0.899 0.555 0.998 0.988 0.778 0.787 0.916 0.767 0.330 0.873 0.598 0.409 0.238 0.793 0.654 EN: enteral nutrition; ICUAW: intensive care unit-acquired muscle weakness; n: sample; %: percentage. Energy is calculated as Kcal/Kg/day; protein as g/Kg/day. Categorical variables are expressed as frequency and percentage (n (%)) and quantitative variables as median [25th -75th percentile]. https://doi.org/10.1371/journal.pone.0286598.t003 Discussion The 68.9% incidence of ICUAW found in this study was higher than the 40% incidence (95% CI [38–42]) reported in a systematic review [16]. However, the percentage of patients without MRC assessment during the ICU stay was similar (26% IC 95% [16, 24–28]). Missing ICUAW data is explained by the patients in whom it was impossible to perform MRC due to insuffi- cient awakening and comprehension (97/97 [100%] patients), which in itself is considered a factor that hinders early mobilization [9, 16, 17]. As in other studies, ICUAW was found predominantly in females and patients with sepsis [18]. We found an association between overweight and ICUAW onset, although the opposite occurred in the case of obesity. A study conducted in obese and non-obese septic mice [19] found that sepsis reduced body weight similarly in both groups, but there was attenuated mus- cle wasting and weakness in the obese mice. This is known as the ‘obesity paradox’. Mechanical ventilation can lead to a daily muscle loss of 1–2% [12]. In addition, the side effects of inappropriate nutrition support include hyperglycaemia, muscle loss, prolonged weaning from MV, and delayed rehabilitation [1, 5]. A review suggests that EN support alone is insufficient to reduce early muscle catabolism, proposing a combination of early mobilization and optimal rehabilitation [20]. A current study, investigated the association between these variables, finding that high protein intake and early mobilization preserves muscle mass [21]. Similarly, other study conducted a clinical trial with three groups (early mobilization, early mobilization and enteral nutrition protocol based on ESPEN guidelines, and control group), and found an improvement in ICUAW in PLOS ONE \| https://doi.org/10.1371/journal.pone.0286598 June 7, 2023 8 / 14 PLOS ONEEnteral nutrition management and ICUAW both intervention groups versus the control group [10]. Despite this, there was little difference between the intervention groups, except for the improvement in muscle strength found in the enteral nutrition and early mobilization group versus only early mobilization. The patients in our study had deficient protein intake and few achieved early active mobility. According to Hermans [22], higher levels of patient mobilization are achieved when physiotherapists lead mobilization decisions. Our study found that despite the low active mobility on days 3–7, low mobility is associated with the onset of ICUAW (p = 0.018). We found no differences in drug administration (neuromuscular blocking agents, steroids and aminoglycosides) with regard to ICUAW, as corroborated by other recent study [9] except for the administration of vasopressors, which was also described by other authors [23]. Our study population was found a high percentage of overfeeding, but insufficient protein intake. Despite there being some controversy, some authors suggest that protein deficiency may lead to muscle deterioration and risk for ICUAW [21, 24], yet we found no differences in protein intake between patients who developed ICUAW or not. This finding may, however, be due to generalized low protein delivery [25]. In our case, a third of patient/days were below 0.5 g pro- tein/kg/day, which is defined in the European guidelines as a low protein diet. Therefore, according to our results, the onset of ICUAW appears to be unrelated to protein intake, because although protein intake was low in most patients, some of those patients did not develop ICUAW. ICUAW-related guideline recommendations: Energy and protein administration A high percentage of patients received trophic EN or were below 80% of the US recommenda- tions [15] for target energy during the first week. However, current evidence and the European recommendations [2], along with the most recent US guideline update [26], show a tendency towards lower energy intake during this period. Arabi et al. [27], noted that anorexia is a com- mon characteristic of critically ill patients. However, during the acute phase, full nutrient pro- vision can be detrimental because it inhibits autophagy, giving cause for concern considering that in our study, overfeeding, defined as “energy administration of 110% above the defined target” [2], was found on almost half of patient/days (43.1%), applying the European recom- mendations, and in 42.9% of obese patients, applying the 2016 US recommendations. Despite this, in our study we have not been able to establish overfeeding as defined in the European guidelines as an explanation for ICUAW. Although some authors question the optimal amount of protein to deliver, most agree that early initiation is more important than energy provision [28]. According to both guidelines, protein intake was insufficient in most patients in our sam- ple. Cahill et al. [29] audited 20 countries to evaluate protein support and concluded that only 2.5% of hospitals achieved >80% of the protein target. Similarly, more recent studies have found below-target protein intake, specifically 52% (±30%) of the prescribed goal [30] and 10– 12% of the total calorie intake, instead of 24–32% [31]. Furthermore, although our study had few patients with CRRT patient/days, ongoing use of these therapies may reduce the protein available for muscle formation [15, 32], and this would worsen protein intake deficiency. Several studies have described various barriers to delivering the nutritional target in criti- cally ill patients. A study identified three factors involved in compliance, which are related to patient, clinical, and site-specific considerations [30]. A Canadian study reported on a nutri- tional improvement programme in the ICU whereby patients attained over 80% of recom- mended target energy and protein intakes [33]. This success was attributed to 1) Presence of registered dietitians in the ICU; 2) Education of the clinical team regarding the need for good PLOS ONE \| https://doi.org/10.1371/journal.pone.0286598 June 7, 2023 9 / 14 PLOS ONEEnteral nutrition management and ICUAW nutritional practice; 3) Encouragement of a culture of interest in bedside nutritional care among all ICU staff. Furthermore, in our results, a weak inverse correlation between weight and kcal/kg/day and proteins/kg/day may suggest that EN was administered through a stan- dard regimen, regardless of weight or patient state. Similar results were reported in a study conducted in 46 countries over 7 years, observing that patients were undernourished because EN was not guided by weight or disease status [34]. Furthermore, Peterson et al. [35] found that no enteral product is able to provide adequate protein intake without excess calorie intake. This observation is important because the current trend is for permissive underfeeding [36]. McCall et al. [33] reported that they increased pro- tein intake by delivering additional protein in powder boluses. Other authors have proposed the use of parenteral amino acids [37], although this route of administration appears to result in lower protein availability (83% vs complete protein) [38]. Other recommendations related to ICUAW prevention Patients with ICUAW had fewer patient-days with glycaemia <180 mg/dl. Although hypergly- caemia was not found to be a risk factor for developing ICUAW in this study, other authors have found an independent relationship between ICUAW and more than 3 days of hypergly- caemia [22, 39]. EN cessation was observed on a third of patient/days, which contrasts with few patient/days with GRV >500 ml and a high energy vs poor protein intake. Unlike other authors’ findings [40], this study appears to show that EN cessations are not the main cause of inappropriate nutrition support. In view of various authors’ findings and ours, it seems reasonable to combine various actions, including the use of up-to-date EN protocols in all ICUs and the presence in ICUs of professionals trained in critical care nutrition, and with ICU care team members trained and motivated to provide early mobilization, who can monitor patients throughout their stay and be involved in discharge plans [6, 41–43]. Limitations The lack of a cohort of patients attaining protein goals limits the results on a potential associa- tion with ICUAW. Measuring ICUAW by means of the MRC scale requires patients’ coopera- tion, which may have caused a delay or absence in diagnosing ICUAW. The patient’s actual weight was only recorded on admission and at no other time during the patient’s stay. Like other authors, we found a lack of reliable instruments available in the ICUs to measure body weight. In addition, weight estimation is hard because of fluid loss and gain, and changes in lean tissue mass [44]. Patients’ actual energy requirements could not be measured due to a gen- eralised absence of indirect calorimetry techniques in the ICUs. Instead, energy requirements were estimated following the general recommendations in international guidelines. Finally, use of parenteral nutrition alone or in combination with EN was not investigated, and some authors have found that parenteral nutrition is detrimental in ICUAW prevention [14, 18]. Implications and recommendations for practice This study highlights the need for better adherence to international enteral feeding guidelines among patients admitted to the ICU. The existence of low mobility, high incidence of ICUAW and low protein intake suggest a need to continue future research to further inform the nutri- tion–early mobilization binomial, which has recently been observed for the first time. Such an approach–considering nutrition as a priority but never alone–will enable us to overcome the PLOS ONE \| https://doi.org/10.1371/journal.pone.0286598 June 7, 2023 10 / 14 PLOS ONEEnteral nutrition management and ICUAW undesirable effects of ICUAW. We believe that it is necessary to train, update and involve ICU professionals in nutritional care and the need for early mobilization of ICU patients. Conclusions The incidence of ICUAW was high in patients receiving EN for at least one week. Early mobilization is associated with lower incidence of ICUAW. Energy and protein intake alone was insufficient to explain the onset of ICUAW. The influence of protein intake on ICUAW was unclear, because significant protein deficiency was found in almost all patients throughout days 3–7 of the ICU stay. Although overfeeding was a common finding in this patient population, we were unable to confirm an association between overfeeding and ICUAW onset. Despite adequate compliance with some recommendations, a high percentage of patients were malnourished according to the guidelines. Future studies are needed in which early mobilization is more widely implemented and nutritional requirements are calculated according to individual patients’ baseline situation and clinical condition, thereby permitting further investigation of the onset of ICUAW in critically ill patients. Supporting information S1 File. Measurement tools. (PDF) S2 File. Logistic regression. (PDF) S1 Fig. Daily energy and protein intake via enteral nutrition, measured by kg of body weight and day. (TIF) S1 Table. Energy and protein intake by target recommendation. (PDF) S2 Table. Other recommendations for enteral nutrition on ICU days 3 to 7. (PDF) Acknowledgments To Sociedad Española de Enfermerı´a Intensiva y Unidades Coronarias (SEEIUC) who pro- moted the study. To the MoviPre Group who collaborated in the data collection. Author Contributions Conceptualization: Ignacio Zaragoza-Garcı´a, Susana Arias-Rivera, Marı´a Jesu´s Frade-Mera, Joan Daniel Martı´, Elisabet Gallart, Alicia San Jose´-Arribas, Tamara Raquel Velasco-Sanz, Eva Blazquez-Martı´nez, Marta Raurell-Torredà. Data curation: Ignacio Zaragoza-Garcı´a, Susana Arias-Rivera, Marı´a Jesu´s Frade-Mera, Joan Daniel Martı´, Elisabet Gallart, Alicia San Jose´-Arribas, Tamara Raquel Velasco-Sanz, Eva Blazquez-Martı´nez, Marta Raurell-Torredà. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286598 June 7, 2023 11 / 14 PLOS ONEEnteral nutrition management and ICUAW Formal analysis: Ignacio Zaragoza-Garcı´a, Susana Arias-Rivera, Marı´a Jesu´s Frade-Mera, Ali- cia San Jose´-Arribas, Tamara Raquel Velasco-Sanz, Eva Blazquez-Martı´nez, Marta Raurell- Torredà. Funding acquisition: Alicia San Jose´-Arribas, Marta Raurell-Torredà. Investigation: Ignacio Zaragoza-Garcı´a, Susana Arias-Rivera, Marı´a Jesu´s Frade-Mera, Alicia San Jose´-Arribas. Methodology: Ignacio Zaragoza-Garcı´a, Marı´a Jesu´s Frade-Mera, Joan Daniel Martı´, Elisabet Gallart, Eva Blazquez-Martı´nez, Marta Raurell-Torredà. Supervision: Susana Arias-Rivera, Marı´a Jesu´s Frade-Mera, Elisabet Gallart, Alicia San Jose´- Arribas, Marta Raurell-Torredà. Writing – original draft: Ignacio Zaragoza-Garcı´a, Marta Raurell-Torredà. Writing – review & editing: Ignacio Zaragoza-Garcı´a, Susana Arias-Rivera, Marı´a Jesu´s Frade-Mera, Joan Daniel Martı´, Elisabet Gallart, Alicia San Jose´-Arribas, Tamara Raquel Velasco-Sanz, Eva Blazquez-Martı´nez, Marta Raurell-Torredà. References 1. Sharma K, Mogensen KM, Robinson MK. Pathophysiology of Critical Illness and Role of Nutrition. Nutr Clin Pract. 2019; 34(1):12–22. https://doi.org/10.1002/ncp.10232 PMID: 30580456 2. Singer P, Blaser AR, Berger MM, Alhazzani W, Calder PC, Casaer MP, et al. ESPEN guideline on clini- cal nutrition in the intensive care unit. Clin Nutr. 2019; 38(1):48–79. https://doi.org/10.1016/j.clnu.2018. 08.037 PMID: 30348463 3. Patkova A, Joskova V, Havel E, Kovarik M, Kucharova M, Zadak Z, et al. Energy, Protein, Carbohy- drate, and Lipid Intakes and Their Effects on Morbidity and Mortality in Critically Ill Adult Patients: A Sys- tematic Review. Adv Nutr. 2017; 8(4):624–634. https://doi.org/10.3945/an.117.015172 PMID: 28710148 4. Shabanpur M, Nachvak SM, Moradi S, Hedayati S, Hosseinikia M, Pasdar Y, et al. Nutritional Care in Iranian Intensive Care Units. Clin Nutr Res. 2018; 7(2):136–145. https://doi.org/10.7762/cnr.2018.7.2. 136 PMID: 29713622 5. Nydahl P, Schuchhardt D, Ju¨ ttner F, Dubb R, Hermes C, Kaltwasser A, et al. Caloric consumption dur- ing early mobilisation of mechanically ventilated patients in Intensive Care Units. Clin Nutr. 2020; 39 (8):2442–2447. https://doi.org/10.1016/j.clnu.2019.10.028 PMID: 31732289 6. 7. Jang MH, Shin MJ, Shin YB. Pulmonary and Physical Rehabilitation in Critically Ill Patients. Acute Crit Care. 2019; 34(1):1–13. https://doi.org/10.4266/acc.2019.00444 PMID: 31723900 Jung B, Moury PH, Mahul M, de Jong A, Galia F, Prades A, et al. Diaphragmatic dysfunction in patients with ICU-acquired weakness and its impact on extubation failure. Intensive Care Med. 2016; 42(5):853– 861. https://doi.org/10.1007/s00134-015-4125-2 PMID: 26572511 8. Yang Z, Wang X, Chang G, Cao Q, Wang F, Peng Z, et al. Development and validation of an intensive care unit acquired weakness prediction model: A cohort study. Front Med (Lausanne). 2023; 10:1122936. https://doi.org/10.3389/fmed.2023.1122936 PMID: 36910489 9. Raurell-Torredà M, Arias-Rivera S, Martı´ JD, Frade-Mera MJ, Zaragoza-Garcı´a I, Gallart E, et al. Care and treatments related to intensive care unit-acquired muscle weakness: A cohort study. Aust Crit Care. 2021; 34(5):435–445. https://doi.org/10.1016/j.aucc.2020.12.005 PMID: 33663950 10. Zhou W, Yu L, Fan Y, Shi B, Wang X, Chen T, et al. Effect of early mobilization combined with early nutrition on acquired weakness in critically ill patients (EMAS): A dual-center, randomized controlled trial. PLoS One. 2022; 17(5):e0268599. https://doi.org/10.1371/journal.pone.0268599 PMID: 35617287 11. Wandrag L, Brett SJ, Frost GS, Bountziouka V, Hickson M. Exploration of muscle loss and metabolic state during prolonged critical illness: Implications for intervention?. PLoS One. 2019; 14(11): e0224565. Published 2019 Nov 14. https://doi.org/10.1371/journal.pone.0224565 PMID: 31725748 12. Stevens RD, Marshall SA, Cornblath DR, Hoke A, Needham DM, de Jonghe B, et al. A framework for diagnosing and classifying intensive care unit-acquired weakness. Crit Care Med. 2009; 37(10 Suppl): S299–S308. https://doi.org/10.1097/CCM.0b013e3181b6ef67 PMID: 20046114 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286598 June 7, 2023 12 / 14 PLOS ONEEnteral nutrition management and ICUAW 13. Hermans G. Assessment protocol of limb muscle strength in critically ill patients admitted to the ICU: the Medical Research Council Scale. Available in: http://links.lww.com/CCM/A780. Last accessed Feb- ruary 2, 2023. 14. Elliott D, Denehy L, Berney S, Alison JA. Assessing physical function and activity for survivors of a criti- cal illness: a review of instruments. Aust Crit Care. 2011; 24(3):155–166. https://doi.org/10.1016/j.aucc. 2011.05.002 PMID: 21723143 15. McClave SA, Taylor BE, Martindale RG, Warren MM, Johnson DR, Braunschweig C, et al. Guidelines for the Provision and Assessment of Nutrition Support Therapy in the Adult Critically Ill Patient: Society of Critical Care Medicine (SCCM) and American Society for Parenteral and Enteral Nutrition (A.S.P.E. N.) [published correction appears in JPEN J Parenter Enteral Nutr. 2016 Nov;40(8):1200]. JPEN J Par- enter Enteral Nutr. 2016; 40(2):159–211. https://doi.org/10.1177/0148607115621863 PMID: 26773077 16. Appleton RT, Kinsella J, Quasim T. The incidence of intensive care unit-acquired weakness syndromes: A systematic review. J Intensive Care Soc. 2015; 16(2):126–136. https://doi.org/10.1177/ 1751143714563016 PMID: 28979394 17. Parry SM, Knight LD, Connolly B, Baldwin C, Puthucheary Z, Morris P, et al. Factors influencing physi- cal activity and rehabilitation in survivors of critical illness: a systematic review of quantitative and quali- tative studies. Intensive Care Med. 2017; 43(4):531–542. https://doi.org/10.1007/s00134-017-4685-4 PMID: 28210771 18. Yang T, Li Z, Jiang L, Wang Y, Xi X. Risk factors for intensive care unit-acquired weakness: A system- atic review and meta-analysis. Acta Neurol Scand. 2018; 138(2):104–114. https://doi.org/10.1111/ane. 12964 PMID: 29845614 19. Goossens C, Marques MB, Derde S, Vander Perre S, Dufour T, Thiessen SE, et al. Premorbid obesity, but not nutrition, prevents critical illness-induced muscle wasting and weakness. J Cachexia Sarcope- nia Muscle. 2017; 8(1):89–101. https://doi.org/10.1002/jcsm.12131 PMID: 27897405 20. van Gassel RJJ, Baggerman MR, van de Poll MCG. Metabolic aspects of muscle wasting during critical illness. Curr Opin Clin Nutr Metab Care. 2020; 23(2):96–101. https://doi.org/10.1097/MCO. 0000000000000628 PMID: 31904602 21. Nakamura K, Nakano H, Naraba H, Mochizuki M, Takahashi Y, Sonoo T, et al. High protein versus medium protein delivery under equal total energy delivery in critical care: A randomized controlled trial. Clin Nutr. 2021; 40(3):796–803. https://doi.org/10.1016/j.clnu.2020.07.036 PMID: 32800385 22. Hermans G, De Jonghe B, Bruyninckx F, Van den Berghe G. Interventions for preventing critical illness polyneuropathy and critical illness myopathy. Cochrane Database Syst Rev. 2014; 2014(1):CD006832. Published 2014 Jan 30. https://doi.org/10.1002/14651858.CD006832.pub3 PMID: 24477672 23. Wolfe KS, Patel BK, MacKenzie EL, Giovanni SP, Pohlman AS, Churpek MM, et al. Impact of Vasoac- tive Medications on ICU-Acquired Weakness in Mechanically Ventilated Patients. Chest. 2018; 154 (4):781–787. https://doi.org/10.1016/j.chest.2018.07.016 PMID: 30217640 24. Montejo Gonza´ lez JC, Sa´ nchez-Bayton Griffith M, Orejo´n Garcı´a L. Muscle in critically ill patients. Nutr Hosp.2019; 36(Spec No2), 12–17. https://doi.org/10.20960/nh.02676 25. Looijaard WGPM Dekker IM, Beishuizen A Girbes ARJ, Oudemans-van Straaten HM Weijs PJM. Early high protein intake and mortality in critically ill ICU patients with low skeletal muscle area and -density. Clin Nutr. 2020; 39(7):2192–2201. https://doi.org/10.1016/j.clnu.2019.09.007 PMID: 31669003 26. Compher C, Bingham AL, McCall M, Patel J, Rice TW, Braunschweig C, et al. Guidelines for the provi- sion of nutrition support therapy in the adult critically ill patient: The American Society for Parenteral and Enteral Nutrition [published correction appears in JPEN J Parenter Enteral Nutr. 2022;46(6):1458– 1459]. JPEN J Parenter Enteral Nutr. 2022; 46(1):12–41. https://doi.org/10.1002/jpen.2267 PMID: 34784064 27. Arabi YM, Reintam Blaser A, Preiser JC. Less is more in nutrition: critically ill patients are starving but not hungry. Intensive Care Med. 2019; 45(11):1629–1631. https://doi.org/10.1007/s00134-019-05765- 0 PMID: 31531714 28. Weijs PJ, Looijaard WG, Beishuizen A, Girbes AR, Oudemans-van Straaten HM. Early high protein intake is associated with low mortality and energy overfeeding with high mortality in non-septic mechan- ically ventilated critically ill patients. Crit Care. 2014; 18(6):701. https://doi.org/10.1186/s13054-014- 0701-z PMID: 25499096 29. Cahill NE, Dhaliwal R, Day AG, Jiang X, Heyland DK. Nutrition therapy in the critical care setting: what is "best achievable" practice? An international multicenter observational study. Crit Care Med. 2010; 38 (2):395–401. https://doi.org/10.1097/CCM.0b013e3181c0263d PMID: 19851094 30. Ridley EJ, Peake SL, Jarvis M, Deane AM, Lange K, Davies AR, et al. Nutrition Therapy in Australia and New Zealand Intensive Care Units: An International Comparison Study. JPEN J Parenter Enteral Nutr. 2018; 42(8):1349–1357. https://doi.org/10.1002/jpen.1163 PMID: 29701877 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286598 June 7, 2023 13 / 14 PLOS ONEEnteral nutrition management and ICUAW 31. Auiwattanakul S, Chittawatanarat K, Chaiwat O, Morakul S, Kongsayreepong S, Ungpinitpong W, et al. Characters of Nutrition Status and Energy-delivery Patterns of the University-based Surgical Intensive Care Units in Thailand (Multi-center THAI-SICU Study). Med Arch. 2018; 72(1):36–40. https://doi.org/ 10.5455/medarh.2018.72.36-40 PMID: 29416216 32. Nystrom EM, Nei AM. Metabolic Support of the Patient on Continuous Renal Replacement Therapy. Nutr Clin Pract. 2018; 33(6):754–766. https://doi.org/10.1002/ncp.10208 PMID: 30320418 33. McCall ME, Adamo A, Latko K, Rieder AK, Durand N, Nathanson T. Maximizing Nutrition Support Prac- tice and Measuring Adherence to Nutrition Support Guidelines in a Canadian Tertiary Care ICU. J Inten- sive Care Med. 2018; 33(3):209–217. https://doi.org/10.1177/0885066617749175 PMID: 29284322 34. Bendavid I, Singer P, Theilla M, Themessl-Huber M, Sulz I, Mouhieddine M, et al. NutritionDay ICU: A 7 year worldwide prevalence study of nutrition practice in intensive care. Clin Nutr. 2017; 36(4):1122– 1129. https://doi.org/10.1016/j.clnu.2016.07.012 PMID: 27637833 35. Peterson SJ, Sheean PM, Braunschweig CL. Orally fed patients are at high risk of calorie and protein deficit in the ICU. Curr Opin Clin Nutr Metab Care. 2011; 14(1): 182–185. https://doi.org/10.1097/MCO. 0b013e3283428e65 PMID: 21178611 36. Owais AE, Kabir SI, Mcnaught C, Gatt M, MacFie J. A single-blinded randomised clinical trial of permis- sive underfeeding in patients requiring parenteral nutrition. Clin Nutr. 2014; 33(6):997–1001. https://doi. org/10.1016/j.clnu.2014.01.005 PMID: 24467878 37. Van Zanten ARH, Petit L, De Waele J, Kieft H, de Wilde J, van Horssen P, et al. Very high intact-protein formula successfully provides protein intake according to nutritional recommendations in overweight critically ill patients: a double-blind randomized trial. Crit Care. 2018; 22(1):156. https://doi.org/10.1186/ s13054-018-2070-5 PMID: 29895309 38. Hoffer LJ. How much protein do parenteral amino acid mixtures provide? Am J Clin Nutr. 2011; 94(6): 1396–1398. https://doi.org/10.3945/ajcn.111.023390 PMID: 22011458 39. Diaz Ballve LP, Dargains N, Urrutia Inchaustegui JG, Bratos A, Milagros Percaz M, Bueno Ardariz C, et al. Weakness acquired in the intensive care unit. Incidence, risk factors and their association with inspiratory weakness. Observational cohort study. Rev Bras Ter Intensiva. 2017; 29(4):466–475. https://doi.org/10.5935/0103-507X.20170063 PMID: 29236843 40. Kozeniecki M, Pitts H, Patel JJ. Barriers and Solutions to Delivery of Intensive Care Unit Nutrition Ther- apy. Nutr Clin Pract. 2018; 33(1):8–15. https://doi.org/10.1002/ncp.10051 PMID: 29323759 41. Crossfield CL, Russo PL, Bucknall TK. Enteral nutrition feeding practices by intensive care nurses: A retrospective evaluation. Nurs Crit Care. 2022; 27(5):676–681. https://doi.org/10.1111/nicc.12609 PMID: 33605511 42. Berger MM. Nutrition and Micronutrient Therapy in Critical Illness Should Be Individualized. JPEN J Par- enter Enteral Nutr. 2020; 44(8):1380–1387. https://doi.org/10.1002/jpen.2002 PMID: 32829498 43. Huang J, Yang L, Zhuang Y, Qi H, Chen X, Lv K. Current status and influencing factors of barriers to enteral feeding of critically ill patients: A multicenter study. J Clin Nurs. 2019; 28(3–4):677–685. https:// doi.org/10.1111/jocn.14667 PMID: 30182514 44. Schneider AG, Thorpe C, Dellbridge K, Matalanis G, Bellomo R. Electronic bed weighing vs daily fluid balance changes after cardiac surgery. J Crit Care. 2013; 28(6):1113.e1-1113.e11135. https://doi.org/ 10.1016/j.jcrc.2013.07.056 PMID: 24144961 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286598 June 7, 2023 14 / 14 PLOS ONE
10.1371_journal.pone.0286321	RESEARCH ARTICLE Effects of physical activity on Chinese overseas students’ mental health during the COVID-19: A multi-country cross-sectional analysis Yingjun NieID 1‡, Yuanyan Ma2‡, Xiaozhi Yao1‡, Yijia Gao3, Xiangping Zheng1 1 College of the Physical Education, Wuhan Sports University, Wuhan, China, 2 College of Sports, Huazhong Normal University, Wuhan, China, 3 College of the Arts, Wuhan Sports University, Wuhan, China ‡ YN, YM and XY are co-first authors on this work. * 29907939@qq.com (YN); 248585348@qq.com (XZ) Abstract a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Background Citation: Nie Y, Ma Y, Yao X, Gao Y, Zheng X (2023) Effects of physical activity on Chinese overseas students’ mental health during the COVID-19: A multi-country cross-sectional analysis. PLoS ONE 18(5): e0286321. https://doi. org/10.1371/journal.pone.0286321 Editor: Christian Napoli, Sapienza University of Rome, ITALY Received: January 10, 2023 Accepted: May 13, 2023 Published: May 30, 2023 Copyright: © 2023 Nie et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The data that support the findings are within the paper. In order to protect subjects’ confidentiality and privacy, additional detailed data are only available on request. Interested researchers may contact the Academic Ethics Committee of Wuhan Sports University (2013028@whsu.edu.cn) directly. Funding: This work was supported by the young and middle-aged scientific research team of Wuhan Sports University (No. 21KT13) and Humanities and Social Sciences Youth Foundation, Ministry of COVID-19 caused severe effects on the psychological well-being of Chinese students over- seas (COS). Physical activity (PA) is critical to strengthen immunity, prevent infection, and reduce the psychological burden caused by COVID-19. However, there is a severe lack of effective PA intervention for mental health in most countries, and COS have limited access to mental healthcare during the pandemic. Objective We aim to examine the effects of PA on COS’ mental health during the pandemic abroad and to better understand that certain types of PA might be associated with a greater reduc- tion in psychological burdens during the pandemic. Methods and results In a multi-country cross-sectional analysis, a questionnaire was distributed to COS living in 37 foreign countries via WeChat Subscription using a snowball sampling strategy. A total of 10,846 participants were included. Descriptive statistics and Binary logistic regression anal- ysis were used for statistical analysis. We found that COS had negative psychology during the pandemic, especially with fear (2.90, 95% CI 2.88−2.92), anxiety (2.84, 95% CI 2.82 −2.85), and stress (2.71, 95% CI 2.69−2.73). PA had meaningful effects on reducing COS self-reported mental health burdens (3.42, 95% CI 3.41–3.44) during the pandemic. The largest associations were seen for recreational and home-based PA (i.e., family games, home aerobic exercise), individual outdoor PA (i.e., walking or running, rope skipping), and PA with a duration of 30 to 70 min per session at frequencies of 4 to 6 times and a total of 150 to 330 min of moderate and vigorous intensity per week tends to be an optimal choice during social distancing times. PLOS ONE \| https://doi.org/10.1371/journal.pone.0286321 May 30, 2023 1 / 14 PLOS ONEEffects of PA on Chinese overseas students’ mental health Education of the People’s Republic of China (No.20YJC890017). The funders provided support in the form of salaries for Yingjun Nie, but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the “author contributions” section. Competing interests: The authors have declared that no competing interests exist. Conclusions COS had several poor mental health conditions during the pandemic. The improvement of PA on COS’ psychology was positively effective during the pandemic. Specific types, inten- sities, durations, and frequencies of PA might have advantages over others for improving COS’ mental health during periods of public health emergencies, and the topic may merit interventional study to reveal multiple factors causing COS’ psychological burdens and enrich the PA forms for all COS’ mental health improvement (i.e., infected, recovered, and asymptomatic COS). Background COVID-19 has undergone several mutations, forcing over 861.7 million students out of school [1], and is less rigorously controlled in the majority of countries worldwide. Colleges in various countries have implemented urgent strategies, such as temporarily closing schools and switch- ing to online classes, to deal with the global public health emergency. And multiple factors, including an increasing number of infected people, asymptomatic patients, social distance, and even delaying graduation, might cause deleterious psychological burdens for students, especially for oversea college students who might experience higher rates of mental health dis- orders than adults with average age during the pandemic [2]. Now there are about 15.3 million Chinese overseas students (COS) studying abroad [3]. Related research showed that COS expe- rienced varying degrees of psychological challenges during the pandemic (e.g., depression, anxiety, and loneliness) [4, 5], and they were even confronted with allegations of being deemed as potential COVID-19 carriers [5]. For example, recent studies have shown that nearly 50% of COS experienced psychological pressure after the onset of COVID-19 in North America. Moreover, students with serious psychological problems may be more likely to have unhealthy lifestyles, such as drinking alcohol and smoking, and even being reported hopelessness, which could cause a vicious circle [6, 7]. However, COS had limited access to mental healthcare dur- ing the pandemic in most countries [3]. There are still many uncertainties in the treatment, infection, and transmission of COVID- 19 presently. And these uncertainties are likely to heighten psychological worries. The best way may be to effectively prevent infection and improve self-immunity. Scientific evidence has demonstrated the positive effect of PA on preventing infection and reducing the negative psy- chological symptoms caused by COVID-19 [8, 9], such as increasing self-mastery to control negative emotions and relaxing mental stress by providing an emotion-communication envi- ronment [9]. However, most people experienced restrictions on PA, and their amount of PA decreased sharply due to various worries and many changes of their lifestyles and PA methods [10]. For example, the rate of PA insufficiency among Chinese residents during the outbreak period of Covid-19 increased about 5 times that during the non-epidemic period [11]. And the high prevalence of physical inactivity is a key risk factor for inducing related diseases [12]. Due to being forced to change PA methods and the lack of effective guidance about what forms of PA are beneficial to health status, most people are confused about whether PA is significant and how to participate in PA to their well-being, and even severely lack of PA guidance and mitigation strategies to advance the knowledge and role of PA for mental health in most coun- tries during the epidemic [13], and there also is a severe lack of effective PA intervention for mental health [11]. Therefore, finding effective PA to achieve significant benefits for COS’ PLOS ONE \| https://doi.org/10.1371/journal.pone.0286321 May 30, 2023 2 / 14 PLOS ONEEffects of PA on Chinese overseas students’ mental health psychological well-being, as well as a better understanding of the relationship between certain PA and mental health improvement, is critical during the pandemic. Given that the degree of psychological improvement may vary as a function of different PA, for instance, recreational group PA may be of higher effectiveness than individual PA in emo- tional relaxation and stress relief [14], and moderate intensity PA may lead to better mental health improvements including relief of depression and anxiety during the pandemic [11], we intended to explore the answers to the following questions: what types of PA are beneficial to psychological improvement? Whether certain forms of PA may be better than others during the epidemic? As a result, we used a multi-country representative sample survey to discover the relationship between PA and COS’ psychological well-being during the pandemic. Investi- gations such as the present study are essential for providing important references and evidence to promote PA and timely crisis-oriented psychological PA interventions for COS during the pandemic. Methods Study population COS are mainly distributed in North America, Western Europe, Africa, Oceania (Chinese Ministry of Education, 2020). Therefore the study used a snowball sampling strategy to recruit COS not infected with COVID-19, who lived in 37 foreign countries across Asia, North Amer- ica, Oceania, Europe, and Africa during the pandemic, and a questionnaire was distributed via the WeChat with high usage rate of all young people in China. To avoid the exclusion of stu- dents who do not use WeChat during the COVID-19, some organizers of Chinese student societies in various college of oversea countries were invited to participate in questionnaire dis- tribution using their social platforms, including enrollment via both online and offline. Through these ways, ensure that COS from different countries participated and to maximize the diversity and representativeness of the COS participating in the survey. Individuals were asked to recall their mental health and PA during the pandemic. The data collection period was from December 1, 2020, to February 28, 2021. Furthermore, full ethical approval was obtained from Wuhan Sports University. All participants gave their informed consent in writing while completing the questionnaire. Survey From April to September 2020, the COVID-19 pandemic reached its peak globally, and over 10 million confirmed cases and 430 thousand deaths were reported in the above 210 countries and regions (WHO, 2020). Therefore, this study focuses on COS’ PA and mental health in this period, and all the participants were asked to recall their PA and mental health status from September 1st to 30th, 2020. PA data were collected by an original self-report scale of IPAQ-C (the Chinese version of the International PA Questionnaire). The Cronbach’s alpha coefficient was 0.89, and the test- retest reliability was 0.82, indicating the good reliability of the questionnaire. All data were managed and screened according to standard methods and the guidelines for data processing and analysis in the IPAQ. COS were asked to recall the type, intensity, frequency, and duration of various PA that they engaged in during the period of September 1st to 30th, 2020. Mental health was assessed using the 50-item Self-evaluation Table for COS’ mental health during the pandemic (a semi-standardized test), which was compiled based on Symptom Checklist 90 (SCL-90). The Cronbach’s alpha coefficient was 0.92, and the test-retest reliability was 0.80, indicating good reliability of the questionnaire. COS were asked to recall the improvement of PA in their psychology during the period of September 1st to 30th, 2020, PLOS ONE \| https://doi.org/10.1371/journal.pone.0286321 May 30, 2023 3 / 14 PLOS ONEEffects of PA on Chinese overseas students’ mental health including anxiety, fear, depression, somatization, and stress. The Cronbach’s alpha coefficients for each sub-scale were 0.87 for anxiety, 0.91 for fear, 0.92 for depression, 0.88 for somatiza- tion, and 0.90 for stress, respectively. The test-retest reliabilities were 0.77 for anxiety, 0.79 for fear, 0.82 for depression, 0.81 for somatization, and 0.83 for stress, respectively. Both scores indicate good reliability of all sub-scales. Each item was rated on a five-point Likert scale, rang- ing from 1 (very slightly or not at all) to 5 (extremely), to indicate the extent to which the par- ticipants felt that each psychology item pertained to them. Statistical analysis SPSS software 26.0 (IBM Inst., Chicago, IL, USA) is employed to analyze all data. Descriptive statistics (i.e., percentages, 95% CI, means, and standard deviations) were calculated for cate- gorical variables and continuous variables to reflect the demographic characteristics of the sur- vey population. Binary logistic regression analysis was conducted to reveal the effect of PA on improving COS’ psychology during the pandemic. Results Survey respondents The final analysis included 10,846 participants from 138 institutions in 37 countries (Table 1), ranging in age from 18 to 49 years old (Mean = 26.3 years old, Standard Deviation = 8.46 years). The participants included 5225 undergraduates, 2174 postgraduates, 1885 Ph.D. stu- dents, and 1562 visiting scholars, including 5200 females and 5646 males (Table 2). The influence of COVID-19 on COS’ mental health The COVID-19 pandemic has significantly negative effects on COS’ psychology. During the pandemic, COS’ psychology was affected to varying degrees (Table 3) with fear (2.90, 95% CI 2.88−2.92), anxiety (2.84, 95% CI 2.82−2.85), stress (2.71, 95% CI 2.69−2.73), depression (2.41, 95% CI 2.39−2.43), and somatization (2.21, 95% CI 2.19−2.22). Accordingly, fear, anxiety, and stress might be more seriously affected. Countries Estonia Netherlands Finland Bosnia and Herzegovina Russia Italy Czech Republic Albania Germany Hungary France Poland Count (percent) Countries Count (percent) 52 (0.47%) 170 (1.56%) 256 (2.36%) 132 (1.22%) 879 (8.1%) 321 (2.96%) 89 (0.82%) 65 (0.6%) 562 (5.18%) 102 (0.94%) 467 (4.31%) 369 (3.4%) United Kingdom 1022 (9.42%) India Uzbekistan Turkey Thailand Malaysia Kazakhstan Korea Georgia Philippines Japan Singapore 32 (0.3%) 19 (0.17%) 21 (0.19%) 20 (0.18%) 16 (0.15%) 49 (0.45%) 542 (5%) 32 (0.29%) 12 (0.11%) 598 (5.51%) 554 (5.11%) Table 1. Distribution of respondents in counties. Countries Canada America New Zealand Australia Cote D’Ivoire Lesotho Tanzania Sudan Mozambique Togo Greece Cyprus Montenegro Count (percent) 616 (5.67%) 1279 (11.79%) 872 (8.03%) 1321 (12.17%) 52 (0.47%) 36 (0.33%) 26 (0.23%) 39 (0.35%) 26 (0.23%) 93 (0.85%) 21 (0.19%) 16 (0.14%) 10 (0.09%) https://doi.org/10.1371/journal.pone.0286321.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286321 May 30, 2023 4 / 14 PLOS ONEEffects of PA on Chinese overseas students’ mental health Table 2. Demographic characteristics of respondents. states Asia North America Oceania Europe Africa Frequency 1895(17.47%) 2193(20.22%) 1953(18.01%) 4533(41.79%) 272(2.51%) https://doi.org/10.1371/journal.pone.0286321.t002 Education level Undergraduate Postgraduate PhD student Visiting scholar Frequency 5225(48.17%) 2174(20.04%) 1885(17.38%) 1562(14.40%) Age (years) 18–24 25–29 30–39 40–49 Frequency 3994(36.82%) 3086(28.45%) 2777(25.60%) 989(9.12%) The influence of PA on COS’ mental health The positive effect of PA on COS’ mental health. This study found that PA had positive effects on reducing COS’ mental health burdens (3.42, 95% CI 3.41−3.44) during the pandemic (Table 4), including somatization (3.31, 95% CI 3.29−3.33), anxiety (3.53, 95% CI 3.51−3.54), depression (3.22, 95% CI 3.20−3.24), stress (3.54, 95% CI 3.52−3.56) and fear (3.52, 95% CI 3.51−3.54). And their correlations with PA were respectively 0.51, 0.53, 0.65, 0.58 and 0.62, all greater than 0.5. Influence of different PA types on improving mental health. Binary logistic regression was used to analyze the effect of PA types on improving mental health during COVID-19. The results showed that some types of PA had a greater improvement in mental health burden than others during the pandemic (Fig 1). Including Yoga (2.10, 95% CI 1.85−2.35), Family games (2.20, 95% CI 2.06−2.41), Gymnastics, Freehand exercises, Dancing (2.25, 95% CI 1.90 −2.51), Home aerobic exercise (2.32, 95% CI 2.11−2.47), Rope skipping (2.43, 95% CI 2.21 −2.77), Resistance training (2.62, 95% CI 2.16−3.18), and Walking or running (3.13, 95% CI 2.61− 3.67) (Table 5). Walking or running (3.13, 95% CI 2.81−3.45) have the greatest effect on mental health, Rope skipping exert the greatest influence on somatization (3.12, 95% CI 2.90 −3.34), Home aerobic exercise on anxiety (2.81, 95% CI 2.52−3.10) and fear (3.05, 95% CI 2.87 −3.23), Gymnastics, Freehand exercises, Dancing on depression (3.23, 95% CI 3.01−3.45), Resistance training on stress (2.93, 95% CI 2.63−3.23) (Table 6). Influence of different PA intensities on improving COS’ mental health. According to the standards of the IHO [15], PA intensity is divided into light intensity (approximately 35 −59% of the maximum heart rate), moderate intensity (approximately 60−69% of the maxi- mum heart rate), and vigorous intensity (greater than or equal to 80% of the maximum heart rate). We found that all types of PA intensity had positive effects on improving COS’ mental health, and the effect of moderate-intensity exercise (3.57, 95% CI 3.24−3.75, p<0.05) and vig- orous-intensity (3.58, 95% CI 3.29−4.06, p<0.05) on improving COS’ psychology might be more effectual than that of light-intensity (3.12, 95% CI 2.76−3.30, p<0.05) during the COVID-19 (Fig 2). Table 3. COS’ psychology status during the pandemic. Fear Anxiety Stress Depression Somatization M±SD 2.90 ± 0.93 2.84 ± 0.93 2.71 ± 1.02 2.41 ± 0.96 2.21 ± 0.86 Variance 95% CI (LL) 95% CI (UL) CV 0.87 0.87 1.04 0.92 0.74 2.88 2.82 2.69 2.39 2.19 2.92 2.85 2.73 2.43 2.22 32.16% 32.93% 37.77% 39.72% 38.99% M = mean; SD = standard deviation; CI = Confidence Interval; CV = coefficient of variation. The negative effects of COVID-19 on COS’ psychology were estimated with lower limits (LL) and upper limits (UL) of 95% confidence interval (CI). https://doi.org/10.1371/journal.pone.0286321.t003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286321 May 30, 2023 5 / 14 PLOS ONEEffects of PA on Chinese overseas students’ mental health Table 4. The positive influence of PA on COS’ psychology status. Improved somatization Decreased anxiety Decreased depression Decreased stress Decreased fear Improved mental health M±SD 3.31±0.99 3.53±0.90 3.22±0.94 3.54±0.92 3.52±0.87 3.42±0.81 Variance 95% CI (LL) 95% CI (UL) 0.99 0.81 0.89 0.85 0.76 0.66 3.29 3.51 3.20 3.52 3.51 3.41 3.33 3.54 3.24 3.56 3.54 3.44 CV 30.18% 25.56% 29.33% 25.98% 24.73% 23.75% M = mean; SD = standard deviation; CI = Confidence Interval; CV = coefficient of variation. The positive effects of PA on COS’ mental health were estimated with lower limits (LL) and upper limits (UL) of 95% confidence interval (CI). The type, intensity, frequency, and duration of various PA was self-reported and was included in the logistic regression as a continuous variable. https://doi.org/10.1371/journal.pone.0286321.t004 Influence of PA duration per session on improving COS’ mental health. Overall, the improvement rate of different psychological conditions might reach a climax when a PA dura- tion per session approaches from 30 to 70 min (Fig 3). However, for COS who engaged in less than 10 min or more than 90 min of PA per session, the improvement of psychological condi- tions might be significantly lower during the pandemic. According to Fig 4, a PA frequency of 4 to 6 times per week (p<0.05) might be better for ameliorating mental conditions. On the contrary, less than 2 times or more than 8 times of PA per week might be associated with a lower effect on psychological improvement. For all types of mental health burdens, better-improving psychology was seen for COS who engaged in a total time of PA per week from 150 to 330 min during the pandemic (Fig 5, p<0.05), regardless of intensity. And lower improving psychology was seen for COS who engaged in less than 30 min or more than 360 min of PA per week during the pandemic. Fig 1. The effect of different PA types on improving COS’ mental health. https://doi.org/10.1371/journal.pone.0286321.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286321 May 30, 2023 6 / 14 PLOS ONEEffects of PA on Chinese overseas students’ mental health Table 5. The effect of PA types on improving mental health. PA types WR = Walking or running RT = Resistance training RS = Rope skipping HA = Home aerobic exercise GF = Gymnastics, Freehand exercises, Dancing FG = Family games YO = Yoga BV = Basketball, volleyball, or football SG = Sensory-motor games and whole body vibration RE = Random exercise BI = Bicycle GB = Golf ball FB = Fight drill, boxing SW = Swim RS = Roller-skating FE = Fencing CC = Chess classes SE 1.41 0.10 0.15 0.14 0.11 0.13 0.12 0.17 0.36 0.19 0.25 1.45 0.34 0.34 1.41 1.55 0.35 Z -3.44 -8.55 -6.77 -10.46 -14.12 -8.75 -12.78 -1.61 -5.53 -9.22 -12.04 -0.89 -7.74 -8.74 -3.99 0.02 -4.91 P 0.001 0.000 0.000 0.000 0.000 0.000 0.000 0.106 0.000 0.000 0.000 0.37 0.000 0.000 0.000 0.983 0.000 OR 3.13 2.62 2.43 2.32 2.25 2.20 2.10 1.79 1.63 1.45 1.31 1.25 1.20 1.05 0.93 0.81 0.79 95% CI Lower Upper 2.61 2.16 2.21 2.11 1.90 2.06 1.85 1.54 1.53 1.28 1.21 1.10 0.105 0.90 0.70 0.69 0.50 3.67 3.18 2.77 2.47 2.51 2.41 2.35 1.96 1.78 1.65 1.51 1.40 0.13 1.32 1.26 1.03 0.97 SE = Standard Error; Z = Z-Score; P = P-value; OR = Odds Ratio; CI = Confidence Interval. The effect of PA types on improving mental health was estimated with lower limits (LL) and upper limits (UL) of 95% confidence interval (CI). OR was introduced to reflect the relative merits of different types of PA. https://doi.org/10.1371/journal.pone.0286321.t005 Discussion Analysis of the impact of COVID-19 on mental health This study showed that COVID-19 had harmed COS’ psychological status. During the pan- demic, COS living abroad experienced different forms of psychological challenges, including somatization, anxiety, depression, stress, and fear. This phenomenon may be shared with all overseas students during the pandemic. Overall, prolonged school closure, cancellation of clas- ses, new modes of instruction, daily lifestyle changes, and much more might accentuate or cre- ates new stressors for COS’ mental health who lived aboard during the pandemic [16]. Firstly, epidemic severity might be the most important factor causing negative psychologi- cal impacts. COVID-19 has undergone several mutations. Various factors could lead to more Table 6. The types with the greatest effect on different mental health indicators. Mental health improvement PA types with the greatest effect SE Z P Improved mental health Improved somatization Decreased anxiety Decreased fear WR = Walking or running RS = Rope skipping HA = Home aerobic exercise HA = Home aerobic exercise Decreased depression GF = Gymnastics, Freehand exercises, Dancing Decreased stress RT = Resistance training 1.41 1.12 0.45 0.63 0.19 0.21 -3.44 -7.56 -6.54 -9.73 -8.36 -7.23 0.001 0.000 0.000 0.001 0.000 0.000 OR 3.13 3.12 2.81 3.05 3.23 2.93 95% CI Lower Upper 2.81 2.90 2.52 2.87 3.01 2.63 3.45 3.34 3.10 3.23 3.45 3.23 SE = Standard Error; Z = Z-Score; P = P-value; OR = Odds Ratio; CI = Confidence Interval. The effect of PA types on improving mental health was estimated with lower limits (LL) and upper limits (UL) of 95% confidence interval (CI). https://doi.org/10.1371/journal.pone.0286321.t006 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286321 May 30, 2023 7 / 14 PLOS ONEEffects of PA on Chinese overseas students’ mental health Fig 2. The effect of different PA intensities on improving COS’ psychology. https://doi.org/10.1371/journal.pone.0286321.g002 serious psychological symptoms of COS living abroad, including the most possible of being asymptomatic carriers for youth [17], the growing numbers of confirmed COVID-19 cases and deaths, the increasing population being infected after vaccination, and some Covid-19 recovered persons being re-infected. The worry of being infected positively correlates with depression, anxiety, and virus aversion among COS. Besides, COS might face discrimination, alienation, and isolation in some countries due to being deemed as potential SARS-CoV-2 car- riers [18]. Research reported that since the WHO declared the COVID-19 outbreak in March 2020, stigma and discrimination against the Chinese has been on the rise [19], which may lead to stress, anxiety, and fear of COS. Secondly, the worry about academic performance might positively correlate with COS’ psy- chological problems during COVID-19, and mental health issues also are the leading impedi- ment to academic success, which could cause a vicious circle [4]. For example, prolonged school closures and class postpones might have negative effects on the contact with lecturers and the support from classmates, then lead to increased fear of passing their examinations [20]. Fig 3. The effect of different PA duration per session on improving COS’ psychology. https://doi.org/10.1371/journal.pone.0286321.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286321 May 30, 2023 8 / 14 PLOS ONEEffects of PA on Chinese overseas students’ mental health Fig 4. The different effects of PA frequency per week on improving COS’ psychology. https://doi.org/10.1371/journal.pone.0286321.g004 Fig 5. The different effects of weekly total PA on improving COS’ psychology. (ME = mental health; SO = somatization; AN = anxiety; DE = depression; ST = stress; FE = fear). https://doi.org/10.1371/journal.pone.0286321.g005 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286321 May 30, 2023 9 / 14 PLOS ONEEffects of PA on Chinese overseas students’ mental health In addition, disruptions of their research projects and internships during school closure might delay their graduation and undermine their competitiveness in the job market, which could cause negative effects on COS’ mental health during COVID-19. Furthermore, many colleges switched to online classes during the COVID-19 pandemic, but some COS were unfamiliar with the new online learning methodology, and the virtual classes were stressful for them to adapt to web-based teaching platforms and technology rapidly [21]. These procedures for pass- ing course exams and presenting their work via online platforms might also be stressful for them [22]. Moreover, the significant increase in screen time mainly being spent on the com- puter and Internet usage may be an additional source of COS’ psychological symptoms [23]. Thirdly, the epidemic prevention environment of the host government, including infection control policies and preventive measures, are generally different among counties, which may be additional challenges to assessing the impact of COVID-19 on COS’ psychological well- being. In fact, COVID-19 has caused a massive burden on governments and individuals around the world. Governments in various countries have implemented urgent national con- tainment policies to prevent the spread of the pandemic. However, the spread of the pandemic is not effectively controlled, and the number of COVID-19 cases and deaths is increasing sig- nificantly in some countries [11]. The uncontrollability of COVID-19 likely arouses COS’ worry about their health [24], and COS tend to experience more anxiety if someone they know is tested positive for COVID-19 [2]. In addition, populations differ on many dimensions between and within countries, such as socioeconomic status, views about COVID-19, and reli- gious beliefs, which all may lead to inconsistent and even conflicting views on COVID-19 and various epidemic-prevention measures between COS and the locals, such as whether to wear a mask outdoor and whether to get vaccinated. That is to say, if COS cannot accept or adapt to the epidemic prevention environment of the host government, their level of anxiety likely increases. Finally, obstruction of return trips might make some COS more prone to psychopathology. Because some counties closed or reduced international air at the beginning of the COVID-19 outbreak or at spikes occur in COVID-19 cases, it was difficult for many COS to buy air tickets to China on time. Some COS even have concerns about infection, COVID-19 tests, and fears of being quarantined on the way back to China. Moreover, some COS might also struggle with the expensive cost of returning to China [25]. Analysis of the PA on improving COS’ mental health This study showed that adequate PA had a positive and significant effect on improving COS’ mental health during the pandemic, and the result of mental health improvement may vary with different PA types, intensities, durations, and frequencies. Related research also reported that PA might reduce the mental health burden caused by COVID-19 [9]. In the context of the worry of being infected and the social distancing strategy, most COS might likely choose the recreational and home-based PA, and individual outdoor PA, which might have a greater ben- efit to emotional relaxation and psychological burden relief. Our other research showed that most people participated in Chinese traditional sports in home quarantine and social isolation during the COVID-19 outbreak in China, such as Chinese martial arts, Taijiquan and Qigong [8], and scientific evidence shows that Chinese martial arts are of great benefit for emotion control, and mental improvement, as well as for the prevention, treatment, and rehabilitation of COVID-19, due to a sequence of movements and postures with the regulation of the breath- ing rhythm and pattern, musculoskeletal stretching and relaxation [26–28]. Nevertheless, few COS seem to participate in Chinese martial arts during the pandemic. It may be necessary to strengthen the education and dissemination of Chinese martial arts for COS, and the PLOS ONE \| https://doi.org/10.1371/journal.pone.0286321 May 30, 2023 10 / 14 PLOS ONEEffects of PA on Chinese overseas students’ mental health establishment of a library of traditional Chinese sports may be an effective strategy to improve COS’ mental health. The form of PA is the key to creating pleasant psychology. This study showed the form of COS’ PA was relatively simple and mainly focused on individual PA during the pandemic. Countries differ politically and economically, the strategies adopted and the difficulties faced vary within each country, as well as the pandemic severity. For example, the UK adopted a pol- icy of herd immunity; students were required to move out of on-campus housing in some col- leges in the USA. Consequently, most COS experienced different restrictions on PA and were forced to change their PA methods during the pandemic [29]. Thus, positive guidance and strategies likely need to be tailored to promote COS’ PA and timely crisis-oriented psychologi- cal PA interventions for COS during the pandemic. For example, in the absence of contact with lecturers and support from classmates, the use of network visualization to participate in collective PA with friends or family who are in China may provide COS with emotional com- munication. In addition, it may be potentially useful for COS to take other professional PA like respiratory exercise, which has been confirmed to have great effects on the prevention, treatment, and rehabilitation of COVID-19 [30]. Related research reported the mental health burden might vary as a function of PA type and there may be an inverted U-shaped relationship between mental health improvement and PA duration as well as PA frequency [31], and unreasonable and inappropriate PA may even damage the immune system and lead to poor mental health [11]. Our study showed the stron- gest associations between COS’ mental health and PA with a duration of 30 to 70 min per ses- sion, frequencies of 4 to 6 times, and a total of 150 to 360 min of moderate and vigorous intensity per week during the pandemic. In our previous study, we found that most people between 12 to 69 years old primarily participated in moderate-intensity PA with 30–60 min duration, 3 to 5 times and a total of 120 to 270 min of PA per week, which was associated with better mental health in China during the COVID-19 outbreak when home quarantine increased severe restrictions and inconvenience on PA [11]. Many COS between 18 to 49 years old tended to participate in moderate and vigorous intensity outdoor PA abroad during the pandemic. The inconsistent findings might be due to differences in the PA environment and participants’ ages. Research also showed that people of different ages may have different PA intensity levels and content during the COVID-19 outbreak period [14]. The IHO also recommended that the population engage in a total of 150 min of moderate- intensity aerobic PA or 75 min of high-intensity PA per week [32]. However, prolonged stren- uous PA has the potential risk of suppressing immune system function, leading to increased susceptibility of infections and the appearance of latent viral reactivation [33, 34]. Our investi- gation showed that more than 8 times or 360 min of PA per week might be associated with a lower effect on psychological improvement for COS during the pandemic. Research also reported PA with more than 6 hours per week or more than 2 hours per session may cause worse mental health burdens [15]. Thus the degree of effectiveness of PA on COS’ mental health may be related to specific PA intensities, durations, and frequencies during the pan- demic. Therefore, during the pandemic, it may be meaningful and necessary for COS to signif- icantly strengthen comprehensive PA monitoring, including duration and frequency, especially the high-intensity PA. Limitations and prospect of research This study had several limitations. Firstly, although the questionnaires were distributed through the populous COS social media platform to maximize the diversity and representation of the subjects, the demographic distribution of participants was certain unevenness and PLOS ONE \| https://doi.org/10.1371/journal.pone.0286321 May 30, 2023 11 / 14 PLOS ONEEffects of PA on Chinese overseas students’ mental health incompleteness. Secondly, the study sample was targeted as individuals who are not infected with COVID-19. For infected people, recovered people, and asymptomatic patients with COVID-19, further research is necessary and meaningful concerning effective PA and mental health. Thirdly, the study tended to rely largely on online self-reported questionnaires to obtain the intensity and content of PA and psychological status, reporting bias cannot be excluded. Finally, this study, as a cross-sectional data study across countries, may not capture differences in PA and mental health outcomes across countries with different epidemic poli- cies over the same time period at the time the results were obtained. Conclusions COS living aboard showed different degrees of psychological challenges during the pandemic, especially fear, anxiety, and stress. Appropriate PA was associated with meaningful improve- ments in self-reported mental health issues, such as stress, anxiety, depression, and somatiza- tion. It is suggested that multiple factors may cause COS’ psychological burdens, such as the epidemic severity, infection rate and deaths, the effectiveness of epidemic prevention policies, and medical readiness in host countries. Clearly, specific types, intensities, durations, and fre- quencies of PA were found to be more effective for improving mental health during the pan- demic. The largest associations were seen for recreational and home-based PA, and individual outdoor PA, with a duration of 30 to 70 min per session, as well as PA at frequencies of 4 to 6 times and nearly 150 to 330 min of moderate and vigorous intensity per week. Above all, it is reassuring for COS to experiment with more PA forms and strength monitoring of PA volume in such intense circumstances. Acknowledgments We also would like to express our sincere thanks to the editor and reviewers for their helpful comments. Author Contributions Conceptualization: Yingjun Nie, Xiangping Zheng. Formal analysis: Yuanyan Ma, Xiaozhi Yao, Yijia Gao. Funding acquisition: Yingjun Nie. Investigation: Yingjun Nie, Yuanyan Ma, Yijia Gao. Methodology: Yingjun Nie, Yuanyan Ma, Xiaozhi Yao. Project administration: Xiaozhi Yao. Writing – original draft: Yingjun Nie, Yuanyan Ma, Xiaozhi Yao. Writing – review & editing: Yingjun Nie, Yuanyan Ma, Xiaozhi Yao, Xiangping Zheng. References 1. Hammami A, Harrabi B, Mohr M, Krustrup P. Physical activity and coronavirus disease 2019 (COVID- 19): specific recommendations for home-based physical training. Managing Sport and Leisure. 2020; 27: 1–6. https://doi.org/10.1080/23750472.2020.1757494 2. Rogowska AM, Kuśnierz C, Bokszczanin A. Examining Anxiety, Life Satisfaction, General Health, Stress and Coping Styles During COVID-19 Pandemic in Polish Sample of University Students. Psy- chology Research and Behavior Management. 2020; 13: 797–811. https://doi.org/10.2147/PRBM. S266511 PMID: 33061695 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286321 May 30, 2023 12 / 14 PLOS ONEEffects of PA on Chinese overseas students’ mental health 3. Wang Y, Teng Y, Peng K, Hu X. The Influencing Factors and Intervention Strategies of Mental Health of Chinese International Students during COVID-19. Chinese Journal of Applied Psychology. 2022; 28: 134–146. https://kns.cnki.net/kcms/detail/detail.aspx?FileName=YXNX202202005&DbName= CJFQ2022 4. Son C, Hegde S, Smith A, Wang X, Sasangohar F. Effects of COVID-19 on college students’ mental health in the united states: Interview survey study. Journal of Medical Internet Research. 2020; 22: 1– 14. https://doi.org/10.2196/21279 PMID: 32805704 5. Zhao Y. Psychological Impacts of the COVID-19 Outbreak on Chinese International Students: Examin- ing Prevalence and Associated Factors. World Journal of Educational Research. 2020; 7: p45. https:// doi.org/10.22158/wjer.v7n3p45 6. Nguyen HT, Do BN, Pham KM, Kim GB, Dam HTB, Nguyen TT, et al. Fear of COVID-19 Scale—Asso- ciations of Its Scores with Health Literacy and Health-Related Behaviors among Medical Students. International Journal of Environmental Research and Public Health. 2020; 17: 4164. https://doi.org/10. 3390/ijerph17114164 PMID: 32545240 7. Arslan G, Yıldırım M, Zangeneh M. Coronavirus Anxiety and Psychological Adjustment in College Stu- dents: Exploring the Role of College Belongingness and Social Media Addiction. International Journal of Mental Health and Addiction. 2021; 20: 1–14. https://doi.org/10.1007/s11469-020-00460-4 PMID: 33500688 8. 9. Zaka A, Shamloo SE, Fiorente P, Tafuri A. COVID-19 pandemic as a watershed moment: A call for sys- tematic psychological health care for frontline medical staff. Journal of Health Psychology. 2020; 25: 883–887. https://doi.org/10.1177/1359105320925148 PMID: 32370621 Jime´nez-Pavo´ n D, Carbonell-Baeza A, Lavie CJ. Physical exercise as therapy to fight against the men- tal and physical consequences of COVID-19 quarantine: Special focus in older people. Progress in Car- diovascular Diseases. 2020;63. https://doi.org/10.1016/j.pcad.2020.03.009 PMID: 32220590 10. Qin F, Song Y, Nassis GP, Zhao L, Dong Y, Zhao C, et al. Physical Activity, Screen Time, and Emo- tional Well-Being during the 2019 Novel Coronavirus Outbreak in China. International Journal of Envi- ronmental Research and Public Health. 2020; 17: 5170. https://doi.org/10.3390/ijerph17145170 11. Nie Y, Ma Y, Wu Y, Li J, Liu T, Zhang C, et al. Association Between Physical Exercise and Mental Health During the COVID-19 Outbreak in China: A Nationwide Cross-Sectional Study. Frontiers in Psy- chiatry. 2021;12. https://doi.org/10.3389/fpsyt.2021.722448 12. Stanton R, To QG, Khalesi S, Williams SL, Alley SJ, Thwaite TL, et al. Depression, Anxiety and Stress during COVID-19: Associations with Changes in Physical Activity, Sleep, Tobacco and Alcohol Use in Australian Adults. International Journal of Environmental Research and Public Health. 2020; 17: 4065. https://doi.org/10.3390/ijerph17114065 PMID: 32517294 13. Xiang Y-T, Yang Y, Li W, Zhang L, Zhang Q, Cheung T, et al. Timely mental health care for the 2019 novel coronavirus outbreak is urgently needed. The Lancet Psychiatry. 2020; 7: 228–229. https://doi. org/10.1016/S2215-0366(20)30046-8 PMID: 32032543 14. Nie Y, Ma Y, Li X, Wu Y, Liu W, Tan Z, et al. PA during the COVID-19 outbreak in China: a cross-sec- tional study. Neural Computing and Applications. 2021;1. https://doi.org/10.1007/s00521-021-06538-x PMID: 34611383 15. World Health Organization. Physical activity. In: World Health Organization [Internet]. World Health Organization; October 5 2022 [cited January 9 2023]. https://www.who.int/news-room/fact-sheets/ detail/physical-activity 16. O¨ lc¸ek G, C¸ elik İ, Başoglu Y, Serbetc¸ioglu MB. Investigation of the Effects of the COVID-19 Pandemic on Audiology Students in Turkey: A Cross-Sectional Survey Study in the COVID-19 Era. Frontiers in Public Health. 2021;9. https://doi.org/10.3389/fpubh.2021.650981 17. Pan X, Chen D, Xia Y, Wu X, Li T, Ou X, et al. Asymptomatic cases in a family cluster with SARS-CoV-2 infection. The Lancet Infectious Diseases. 2020; 20: 410–411. https://doi.org/10.1016/S1473-3099(20) 30114-6 PMID: 32087116 18. Zhai Y, Du X. Mental health care for international Chinese students affected by the COVID-19 outbreak. The Lancet Psychiatry. 2020; 7: e22. https://doi.org/10.1016/S2215-0366(20)30089-4 PMID: 32199511 19. Misra S, Le PD, Goldmann E, Yang LH. Psychological impact of anti-Asian stigma due to the COVID-19 pandemic: A call for research, practice, and policy responses. Psychological Trauma: Theory, Research, Practice, and Policy. 2020;12. https://doi.org/10.1037/tra0000821 20. Heijnen S, Hommel B, Kibele A, Colzato LS. Neuromodulation of Aerobic Exercise—A Review. Fron- tiers in Psychology. 2016;6. https://doi.org/10.3389/fpsyg.2015.01890 21. Zhang Y, Zhang H, Ma X, Di Q. Mental Health Problems during the COVID-19 Pandemics and the Miti- gation Effects of Exercise: A Longitudinal Study of College Students in China. International Journal of Environmental Research and Public Health. 2020; 17: 3722. https://doi.org/10.3390/ijerph17103722 PMID: 32466163 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286321 May 30, 2023 13 / 14 PLOS ONEEffects of PA on Chinese overseas students’ mental health 22. Rizun M, Strzelecki A. Students’ Acceptance of the COVID-19 Impact on Shifting Higher Education to Distance Learning in Poland. International Journal of Environmental Research and Public Health. 2020; 17: 6468. https://doi.org/10.3390/ijerph17186468 PMID: 32899478 23. Gao J, Zheng P, Jia Y, Chen H, Mao Y, Chen S, et al. Mental health problems and social media expo- sure during COVID-19 outbreak. Hashimoto K, editor. PLOS ONE. 2020;15. https://doi.org/10.1371/ journal.pone.0231924 24. Cao W, Fang Z, Hou G, Han M, Xu X, Dong J, et al. The psychological impact of the COVID-19 epi- demic on college students in China. Psychiatry Research. 2020;287. https://doi.org/10.1016/j.psychres. 2020.112934 25. 26. Feng S, Zhang Q, Ho SMY. Fear and anxiety about COVID-19 among local and overseas Chinese uni- versity students. Health & Social Care in the Community. 2021; 249–258. https://doi.org/10.1111/hsc. 13347 PMID: 33825220 Feng F, Tuchman S, Denninger JW, Fricchione GL, Yeung A. Qigong for the Prevention, Treatment, and Rehabilitation of COVID-19 Infection in Older Adults. The American Journal of Geriatric Psychiatry. 2020; 28: 812–819. https://doi.org/10.1016/j.jagp.2020.05.012 PMID: 32425471 27. Chiu H-Y, Chiang P-C, Miao N-F, Lin E-Y, Tsai P-S. The Effects of Mind-Body Interventions on Sleep in Cancer Patients. The Journal of Clinical Psychiatry. 2014; 75: 1215–1223. https://doi.org/10.4088/jcp. 13r08918 28. Buffart LM, van Uffelen JG, Riphagen II, Brug J, van Mechelen W, Brown WJ, et al. Physical and psy- chosocial benefits of Yoga in cancer patients and survivors, a systematic review and meta-analysis of randomized controlled trials. BMC Cancer. 2012;12. https://doi.org/10.1186/1471-2407-12-559 29. Constandt B, Thibaut E, De Bosscher V, Scheerder J, Ricour M, Willem A. Exercising in Times of Lock- down: An Analysis of the Impact of COVID-19 on Levels and Patterns of Exercise among Adults in Bel- gium. International Journal of Environmental Research and Public Health. 2020; 17: 4144. https://doi. org/10.3390/ijerph17114144 PMID: 32532013 30. Liu K, Zhang W, Yang Y, Zhang J, Li Y, Chen Y. Respiratory rehabilitation in elderly patients with COVID-19: A randomized controlled study. Complementary Therapies in Clinical Practice. 2020; 39: 101166. https://doi.org/10.1016/j.ctcp.2020.101166 PMID: 32379637 31. Chekroud SR, Gueorguieva R, Zheutlin AB, Paulus M, Krumholz HM, Krystal JH, et al. Association between physical exercise and mental health in 1�2 million individuals in the USA between 2011 and 2015: a cross-sectional study. The Lancet Psychiatry. 2018; 5: 739–746. https://doi.org/10.1016/s2215- 0366(18)30227-x 32. Ammar A, Brach M, Trabelsi K, Chtourou H, Boukhris O, Masmoudi L, et al. Effects of COVID-19 Home Confinement on Eating Behaviour and Physical Activity: Results of the ECLB-COVID19 International Online Survey. Nutrients. 2020;12. https://doi.org/10.3390/nu12061583 33. Crisafulli A, Pagliaro P. Physical activity/inactivity and COVID-19. European Journal of Preventive Car- diology. 2020; 204748732092759. https://doi.org/10.1177/2047487320927597 34. Goodman-Casanova JM, Dura-Perez E, Guzman-Parra J, Cuesta-Vargas A, Mayoral-Cleries F. Tele- health Home Support During COVID-19 Confinement for Community-Dwelling Older Adults With Mild Cognitive Impairment or Mild Dementia: Survey Study. Journal of Medical Internet Research. 2020; 22: e19434. https://doi.org/10.2196/19434 PMID: 32401215 PLOS ONE \| https://doi.org/10.1371/journal.pone.0286321 May 30, 2023 14 / 14 PLOS ONE
10.1371_journal.pone.0285717	RESEARCH ARTICLE Vancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice Mariam Hantash Abdel JalilID Khawla Abu Hammour1, Maria Hasan Matalqah2, Jwan Saleh Khaleel Albadaineh3, Shrouq Khaled AlOmoush2, Montaha Al-IedeID 1, Rima Ηijazeen1, Farah Khaled Abu-Mahfouz1, 4,5 1 Department of Biopharmaceutics and Clinical Pharmacy, The University of Jordan, Amman, Jordan, 2 School of Pharmacy Jordan University of Science and Technology, Irbid, Jordan, 3 School of Pharmacy Mu’tah University, Kerak, Jordan, 4 Department of Pediatrics, Jordan University Hospital, Amman, Jordan, 5 School of Medicine, The University of Jordan, Amman, Jordan a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 m.abdeljalil01@ju.edu.jo Abstract Background OPEN ACCESS Citation: Abdel Jalil MH, Ηijazeen R, Khaled Abu- Mahfouz F, Abu Hammour K, Hasan Matalqah M, Saleh Khaleel Albadaineh J, et al. (2023) Vancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice. PLoS ONE 18(5): e0285717. https://doi.org/ 10.1371/journal.pone.0285717 Editor: Mona Nabulsi, American University of Beirut Medical Center, LEBANON Received: November 26, 2022 Accepted: May 2, 2023 Published: May 17, 2023 Copyright: © 2023 Abdel Jalil et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The datasets generated during and/or analysed during the current study are not publicly available due to constraints imposed by the consent form and the IRB committee. However, a request to access the data can be requested from the research ethics committee at JUH hospital (sssaleh@ju.edu.jo). Vancomycin prescription and monitoring guidelines have been reported to be poorly fol- lowed by various centers. Aims Identifying barriers to compliance with vancomycin dosing and therapeutic drug monitoring guidelines (TDM) and possible ways to enhance compliance based on the healthcare pro- viders’ (HCPs) perspective. Methods A qualitative study based on semi-structured interviews with HCP (physicians, pharmacists, and nurses) was conducted at two Jordanian Teaching Hospitals. Interviews were audio- recorded and analyzed through thematic analysis. The COREQ criteria for qualitative research were utilized to report the study findings. Results A total of 34 HCPs were interviewed. HCP perceived several factors as barriers to guide- line recommendation compliance. Such factors included negative perception towards pre- scription guidelines, lack of knowledge regarding TDM guidelines, the hierarchy of medication management, work pressure, and ineffective communication among health- care providers. Potential strategies to optimize guidelines adaptation included providing HCPs with more training and decision support tools in addition to activating the role of clini- cal pharmacists. Conclusions Funding: This project was funded by the deanship of scientific research at the University of Jordan. The main barriers to guideline recommendations uptake were identified. Interventions should address those barriers related to the clinical environment, including enhancing PLOS ONE \| https://doi.org/10.1371/journal.pone.0285717 May 17, 2023 1 / 14 PLOS ONEVancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice Competing interests: NO authors have competing interests. interprofessional communication related to vancomycin prescription and TDM, reducing workload and providing support systems, promoting educational and training programs, in addition to adopting guidelines suitable for the local environment. Introduction Vancomycin, a glycopeptide antibiotic, is of the most frequently used antibiotics, especially in critically ill patients and pediatric wards [1–3]. It is considered the first-line antibiotic for managing serious methicillin-resistant Staphylococcus aureus (MRSA) infections such as sepsis, meningitis, and endocarditis. Unfortunately, vancomycin has been commonly implicated in causing medication related-patient harm due to its narrow therapeutic index [4]. Prescribing antibiotics, in general, can be challenging for prescribers. For instance, underdosing has been frequently described [5, 6]. Furthermore, overuse of antibiotics is prevalent. This might be associated with the fact that physicians lacking enough experience can have the perception that the benefits of antibiotic treatment out ways long-term risks, which include resistance and reduced efficacy [7]. For vancomycin as an example, many researchers reported excessive and prolonged empiric prescription [8, 9]. To guide its use and minimize adverse drug effects due to overdosing and inefficacy due to underdosing, Therapeutic Drug Monitoring (TDM) is recommended for prolonged courses. At the study institutions, vancomycin TDM is based mainly on trough monitoring. Despite its simplic- ity, compared to other TDM methods, many researchers reported suboptimal trough moni- toring practice. In previous research of ours [10], we measured the appropriateness of vancomycin dosing and monitoring practices at Jordan University (JUH) teaching hospital. This study revealed several defects in the vancomycin clinical practice, which most prominently included pro- longed antibiotic treatment, inaccurate documentation of TDM parameters, and poor clinical action in response to markedly low drug levels (< 7 mg/L). Similar results were also reported by other researchers, and several studies evaluated barriers to appropriate dosing and/or moni- toring of antibiotics among hospital physicians [7, 11, 12]. However, vancomycin dosing, administration, and TDM require collaborative efforts between physicians, pharmacists, and nurses. For this reason, the present study aimed to explore perceived barriers to appropriate vancomycin prescribing, administration, and TDM practices by all healthcare providers (HCP). Furthermore, it aimed to explore possible strategies to optimize these practices as per- ceived by the HCP. Methodology Study design and ethics The current study was a qualitative study based on face-to-face semi-structured interviews with HCPs involved in vancomycin prescription/ TDM. The study was conducted over a period of two months, starting on August 2021 and extending until October 2021. The study was conducted in two university teaching hospitals in Jordan. The first was the Jordan Univer- sity Hospital (JUH). JUH is the first university teaching hospital in Amman, the Capital of Jor- dan, with a capacity of up to 600 beds. The second is the King Abdallah University Teaching Hospital (KAUH). It is considered the largest medical structure in the north of Jordan, with a bed capacity of 678. Institutional Review Board approvals were obtained from participating centers (JUH and KAUH). PLOS ONE \| https://doi.org/10.1371/journal.pone.0285717 May 17, 2023 2 / 14 PLOS ONEVancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice Data collection and interpretation Residents, pharmacists, and nurses with specific experiences and roles in vancomycin dosing/ TDM and were willing to provide a detailed description of their personal experience were eligi- ble to participate, provided that they signed written informed consent. Consultant physicians were not included in the present study. Based on our previous research on vancomycin prescription and monitoring [10], residents are more involved in daily dosing and monitoring. This approach was also adopted by other researchers [13]. Fur- thermore, due to their important influence on the prescribing behavior of residents, it was deemed better by the research team to evaluate their opinions in a separate study that includes more study sites. Initially, eligible HCP were sent an invitation letter to participate in this semi-structured interview. Later, a convenience snowball sampling method was used as HCPs recommended colleagues that could be willing to take part. Participation was voluntary, and no compensation for the time spent in the interview was offered to participants. HCPs were provided with written information about the study and asked to arrange a loca- tion of the HCP’s choice and a mutually convenient time for the interview. The number of interviews was determined by the point where the data was deemed saturated (i.e., no new themes emerged). An interview guide was developed based on a review of the literature [12, 14] and input from an expert in qualitative research (R.H, Ph.D.) and clinical pharmacokinetics and phar- macy practice (M.A, K. A) (S1 File). The interview guide was written in Arabic language and was formulated as open, non-leading questions and focused on vancomycin prescription prac- tices and associated barriers and challenges faced during usual vancomycin prescribing and TDM practices, in addition to potential means for improving vancomycin prescribing and TDM practice based on HCP views. Prompts and follow-up questions were used as appropri- ate. The interview guide was pilot-tested prior to finalization (results not included in the analy- sis). One female clinical pharmacist with previous experience in research (F. K) conducted the interviews at JUH, while senior PharmD students, both females (M.M, S.A), conducted the interviews at KAUH. All of them received training on interviewing and conducted pilot inter- views with (R.H.) before starting the study. None of them had an established relationship with the participant before the start of the study. The Consolidated Criteria for Reporting Qualitative Research (COREQ) criteria for qualita- tive research were utilized to report the study findings [15]. All interviews were audio-recorded and de-identified. The interviews took place at the workplace during working hours, usually in a meeting room or a physician’s office. The interview schedule began after the interviewer introduced herself as a clinical pharmacist/research assistant. Later a summary of the research rationale and objectives and an overview of the interview guide were provided. Anonymized audio records were transcribed verbatim. The data collected were translated from Arabic to English and then back-translated again to Arabic to assure accuracy. No interviews were repeated, nor did participants hear their records after the interview due to limitations of time. Transcripts were double-checked for accuracy following the transcribing. Data were analyzed using thematic framework analysis [16]. The researchers (F.K, S.K, M.M) kept a record of reflections and thoughts about interpretations of the collected data, which included illustra- tions of possible relationships between emerging themes to guide or reflect the analysis. Those themes were revised and refined through discussion with all authors to ensure that the analysis was supported by the data. Similar to the approach of Chan et al. [12], the data were not repre- sented quantitatively to prevent misrepresentation due to the versatile type of discussions in each semi-structured interview. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285717 May 17, 2023 3 / 14 PLOS ONEVancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice Results A total of 34 HCPs working at different hospital wards, such as pediatric wards, internal wards, surgical and intensive care units, we interviewed in the present study. Due to the design of the study response rate could not be determined. The interviews ranged in duration from 11 to 38 minutes, and data saturation was achieved. Table 1 shows the general characteristics of interview participants. Barriers to optimal vancomycin prescribing and administering practice The barriers the HCPs perceived concerning the optimal prescribing and TDM practice were mainly related to the following major sub-themes: (1) barriers pertaining to the healthcare pro- fessionals; (2) barriers pertaining to the healthcare system and working environment. Barriers pertaining to the healthcare professionals Healthcare professionals’ concerns. HCPs concerns that can lead to noncompliance to guideline recommendations regarding vancomycin prescription were categorized into three subthemes: a) Family’s response to stopping antibiotics and the fear of being blamed as a barrier to optimal vancomycin use in real daily clinical practice. ". . . the clinical picture and other factors control our use of antibiotics, not just the guidelines; for example, some families are nagging and troublemakers and you (the doctor) are afraid Table 1. General characteristics of participants. Characteristics Gender Male Female Age group 22–29 30–39 40–49 Healthcare professional Physician Pharmacist Nurse Years of experience in the current position 1–5 years 6–10 years 11–15 years 16–20 years >20 years Healthcare organization Jordan University Hospital King Abdullah University Hospital https://doi.org/10.1371/journal.pone.0285717.t001 N (%) 8 (23.53%) 26 (76.47%) 17 (50.0%) 11 (32.4%) 6 (17.6%) 12 (35.29) 12 (35.29) 10 (29.42) 15 (44.1%) 8 (23.5%) 5 (14.7%) 5 (14.7%) 1 (3.0%) 20 (58.8%) 14 (41.2%) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285717 May 17, 2023 4 / 14 PLOS ONEVancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice that if you stop the antibiotic and the patient relapses that you would be blamed, so if the kid- ney function is good, you would keep the patient on vancomycin." (Participant 11—neurosurgery resident). b) Fear of vancomycin adverse effects "In the NICU, the problem is that neonatologists don’t usually increase the vancomycin dose to reach a target of 15–20 mg/L even when it is indicated because they are afraid of vancomy- cin’s toxicity" (Participant 12—clinical pharmacist) c) Fear of potential undiagnosed infections "We feel that we overprescribe antibiotics against guideline recommendations. For instance, we know that the case is a case of drug-induced fever or autoimmune disease, but still, we give antibiotics because deep inside you will not be comfortable if the patient needs an antibiotic for some reason and you don’t give it, so frankly we overprescribe antibiotics (vancomycin). We give fluids and take levels to keep the patients safe" (Participant 18—chief internal resident). As for TDM, the same resident reported that TDM might be overused in the ICU due to concerns related to the critical situation of patients "We may perform TDM on a daily basis for ICU patients because they are critical and their kidney function changes continuously, so they need adjustment every day . . .we do it even if the level is not valid because serum creatinine changes. . . we also may tell nurses to delay the next dose until we tell them to give it depending on the level results" (Participant 18—chief internal resident). Perceived lack of benefit of TDM. Some HCPs perceived vancomycin TDM as unneces- sary, which may lead to noncompliance with TDM guidelines "Vancomycin TDM is not really necessary. There are more important things to do. . . you would do TDM for a patient on digoxin, or patients taking neuro-medicines or antipsychotics, for those medications we need to take levels" (Participant 28 -internal resident). Negative perception towards prescribing guidelines. Interview findings revealed that residents and clinical pharmacists seemed to be aware of available international guidelines and support resources for antibiotic prescribing in general and vancomycin in specific and were careful to familiarise themselves with all the evidence and knowledge about vancomycin dos- ing and monitoring. However, they expressed concerns about the simplicity of guidelines and their lack of updated information. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285717 May 17, 2023 5 / 14 PLOS ONEVancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice "We have guidelines, but they are not used, they are outdated, and they don’t include everything" (Participant 20 -pediatric resident) "If something is straightforward, we use guidelines, but we don’t see them useful for compli- cated cases, we rely on experience" (Participant 7- emergency resident) Furthermore, another HCP pointed out how prescribers questioned the applicability of evi- dence from international guidelines to real-world prescribing population, as indicated below: "We have international guidelines, there are reasons why doctors don’t follow guidelines, their recommendations may contradict their experiences and they may consider that the manage- ment of diseases in actual practice in Jordan is different from USA or U.K. because resistance patterns differ" (Participant 27- pediatric resident). Barriers pertaining to the healthcare system and working environment Lack of guidelines to guide TDM. HCPs perceived the lack of clarity of guidelines in van- comycin TDM as a barrier to optimal practice: "Again, the barriers would be the fact that we don’t have clear guidelines here. . . for changing the dose based on the trough level, we would depend more on the experience, so we might change, for example, the dose by 10–15% of the original dose depending on the case. Actually, there is no fixed ratio." (Participant 8: internal resident) Prescribing culture. The decision to initiate vancomycin is usually made by consultant physicians, especially in complicated cases. However, vancomycin can also be started by resi- dents in simple cases or during night admissions, and then the decision is confirmed by the consultant. "Consultants along with residents prescribe vancomycin . . .. It depends on the situation. If I am on a night shift and the case is clear for example, bacterial pneumonia, it is not reasonable to call the consultant in the middle of the night, we start it, and the following day during the round, we finalize the decision. . ..I don’t always agree with consultants regarding their pre- scription of vancomycin, I think there is abuse" (Participant 20 -pediatric resident) Residents expressed their hesitancy to share their concerns regarding vancomycin prescrip- tion with consultants and stated that they would just follow the consultant’s recommendation without questioning their judgment, even if they thought that the decision was not correct. "Residents are responsible for prescription unless in complicated cases. . . We have a protocol in our NICU unit. Some consultants may give vancomycin even when not indicated. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285717 May 17, 2023 6 / 14 PLOS ONEVancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice Vancomycin is frequently prescribed (in about 20% of patients). I am not always convinced with vancomycin prescription decisions, we have misused, but I don’t share my opinion with consultants." (Participant 11: neurosurgery resident) "If the consultant insists on prescribing vancomycin, we just follow his recommendation even if it doesn’t follow the guideline, we don’t question his judgment" (Participant 32, internal resident) Lack of effective communication among healthcare providers. Ineffective communica- tion between healthcare providers may result in ineffective dose adjustment, delayed doses, or misinterpreted samples. "As pharmacists, if we have a recommendation about a dose change, we have to communicate with the medical team in charge of the patient, you have to communicate with the senior resi- dent or the consultant, and sometimes this is difficult, and the required recommendation is not implemented" (Participant 26 -clinical pharmacist) "As nurses, we suffer until the junior physicians come to do a scheduled sample; you call them and tell them to come, sometimes they come, and sometimes they don’t, and sometimes they come late. So we may do the samples to avoid further delay". (Participant 9—nurse) "Sometimes nurses give the dose before we have the chance to take a trough sample, you will not find this in records, and we wouldn’t know unless we have an odd level, for example, your patient’s level is 6-7mg/L daily, suddenly you would find that the level is 50 mg/L. Usu- ally, this gives you an indication that the trough sample was withdrawn at an inaccurate time" (Participant 10 -internal resident). Lack of support strategies: On-rotation versus out-of-working hours. Prescribers declared that they are better supported during working hours either by the availability of con- sultants or by requesting consultation on the spot from other colleagues. For example, they can refer to infectious disease specialists and clinical pharmacists who may offer prescribing advice for challenging decisions that are not covered in the local antimicrobial guidelines. However, out of working hours and night shifts, HCPs were less supported and had fewer resources. "During consultant rounds in the morning, there are infectious disease specialists that we can refer to if there is something not clear or if the patient is not responding; this is help- ful. However, during night shifts, such support is not available, but we may contact the fellow" (Participant 10 -internal resident) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285717 May 17, 2023 7 / 14 PLOS ONEVancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice "During working hours, I am in the ward, and I provide support regarding appropriate dosing when needed, however outside working hours, there is no communication; I don’t have to answer my phone outside my working hours they may just give a stat dose at the time" (Participant 22 -clinical pharmacist) Work pressure and shortage of staff. Furthermore, for many HCPs, lack of time due to work pressure per se was perceived as a barrier to optimal vancomycin administration. For instance, a nurse revealed that work overload and time pressure would make it easy to make mistakes. "I don’t justify, it is not a justification, but the reality is something else. For example, as a nurse, you have 4, or 5,6,7 patients on antibiotics. This number is problematic, and work pres- sure causes stress that generates errors. Errors are usually missed information that is impor- tant or risky to the patient." (Participant 29 -nurse) The workload was also reported consistently to hinder HCPs from optimal TDM practice. "it is difficult to adhere to TDM guidelines in terms of accurate recording of the time of dose administration and sample collection. We have a lot of pressure. If we were in a private hospi- tal and each of us handled 2–3 patients, that would be easy. I have to administer medications for 20 patients if we assume that 3–4 patients are on vancomycin it will be difficult to record every dose, there is no time and no staff" (Participant 14-nurse) "Some mix-ups in vancomycin TDM may occur due to work pressure. A pharmacist counts the doses of vancomycin once prescribed and marks when vancomycin TDM should be done, but they may miss this mark, or the pharmacist may not mark the medication sheet if vanco- mycin was started during the weekend, and the TDM may be forgotten. It depends on the degree of work pressure in the ward" (Participant 17 NICU pharmacist). Lack of institution-wide training opportunities. HCPs seemed to have assumed that, in principle, lack of training was a barrier to optimal TDM performance. Interviews with HCPs revealed that they were keen to participate in training workshops that were assumed to support their needs and fill in the gap in current practice. "I don’t remember receiving any training regarding antibiotics; as you know, this mainly related to physicians, but I remember once we took a lecture about administration, I would like more training about antibiotics and TDM. I think it will be useful" (Participant 30- nurse) Logistics. HCP reported that sometimes logistics might interfere with the efficiency of TDM and the ability to comply with its protocol PLOS ONE \| https://doi.org/10.1371/journal.pone.0285717 May 17, 2023 8 / 14 PLOS ONEVancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice "The problem is that multiple staff is involved in the process of TDM, and this confuses things, sometimes I can’t do my job because the porter didn’t take the sample to the lab. . .sometimes this has delayed my work for 4 hours, it is not easy to go by the book in actual practice" (Participant 1 -clinical pharmacist) Potential strategies to optimise vancomycin prescribing and TDM practice Healthcare professionals were asked to suggest strategies and recommendations for improve- ment, to address their needs. Their needs were summarized as follows: Provide training sessions. Overall, there seemed to be a general sense of the need for training sessions for HCP. Most HCPs suggested that providing education and training in van- comycin dosing and monitoring might have the potential to increase their knowledge and con- fidence during their daily practice. "There must be continuous training, especially for staff working with pediatrics as their doses are small and not easy to prepare also we see a lot of side effects such as red man syndrome, more training would be useful" (Participant 3—nurse). Provide a decision support system. HCPs seemed to have assumed that, in principle, the use of decision support systems would be useful and could potentially improve adherence to TDM guidelines to support their needs and fill in the gap in current practice. "I think any sort of computerized program is good, as far as it has an alarm system that reminds you of the samples that must be drawn for your patient, and a calculator where you put patient’s characteristics, and it gives you the new dose, so it’s more organized and more structured. . . it will be good . . ." (Participant 10 -internal resident) Moreover, a clinical pharmacist suggested that such a system would be useful for HCPs who are not familiar with doing vancomycin calculations. "We took a course about vancomycin calculations, but we didn’t complete it. We do the calcu- lations based on the equations we took at the university, and I don’t feel confident doing them alone. If there is a support system, it would be useful. (Participant 22 -clinical pharmacist) Activating the role of clinical pharmacist. HCPs explicitly referred to the potential role of clinical pharmacists in the vancomycin prescribing process. For instance, a resident spoke from her experience, as indicated below: "I feel that it is very useful that we have a clinical pharmacist in NICU. She always reviews the medication sheet, checks the levels, and checks the doses of babies because their weight changes day by day, so she follows up on increasing or decreasing doses for each patient and checks the duration of therapy. As a resident, I am very busy, so I can’t catch up on everything they help PLOS ONE \| https://doi.org/10.1371/journal.pone.0285717 May 17, 2023 9 / 14 PLOS ONEVancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice us. Honestly, I feel that it is very important to improve practice to have a clinical pharmacist in every ward; this is what I am sure of. In NICU, their presence is very helpful" (Participant 7: pediatric resident) "I think clinical pharmacists must be more involved in medication reconstitution, especially those that need to be prepared under sterile conditions, we are under a lot of pressure, and they are more knowledgeable regarding drugs and their interactions. I think it would be help- ful if they were more involved in drug preparation, including vancomycin. That would be very helpful to us". (Participant 3 –nurse) Need for national guidelines and protocols. HCPs highlighted the need for local guide- lines and protocols that would potentially support their prescribing practice performance. "I think it would help if there are local guidelines and protocols based on local antibiograms. This would always be helpful to improve practice." (Participant 20 -paediatric resident) An HCP also said that the availability of local guidelines would potentially improve their vancomycin prescribing by reducing variation in vancomycin treatment and management and helping to streamline clinical workflows. ". . ..it also standardizes the practice. It is very important, help to streamline workflow" (Participant 18 –chief internal resident) Discussion The present study highlights the complexity of vancomycin prescribing and therapeutic drug monitoring and the need for interprofessional communication, activation of the role of clinical pharmacists, and further education and support to optimize its use. HCPs were generally not aware of the presence of any guidelines related to vancomycin TDM. Nevertheless, prescribers were aware of the presence of vancomycin prescription guidelines (mainly international). However, they were negatively perceived as outdated and noncomprehensive, and they highlighted the need for guidelines based on local sus- ceptibility patterns. This contrasts with the findings of Chan et al. [12], where prescribes found guidelines clear and valuable, which facilitates adaptation of guideline recommenda- tions in their daily practice. In our study, some prescribers reported that they find their patient population different, so it can’t fit the structured guideline recommendations for the average patient. Thus, their prescription may differ. In addition, they doubted the abil- ity of guidelines to guide them in the management of complicated cases. This finding comes in line with the finding of Livorsi et al. [17], where nonadherence recommendations were also found to be related to the need for individualized patient care and skepticism towards guideline recommendations. Thus, it is essential for healthcare institutions not only to have evidence-based guidelines but also to ensure that these guidelines are continu- ously updated and modified to enable them to operate within the local environmental needs. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285717 May 17, 2023 10 / 14 PLOS ONEVancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice The major role in vancomycin, dosing, discontinuation, and TDM was reported to be for the consultants. This "etiquette" of vancomycin prescribing and medication management that relies on prescribing autonomy and the principle of non-interference is another dominant bar- rier in guideline acceptance and renders experience and routine ahead of guidelines when making decisions regarding the prescription. Furthermore, it minimizes the important role of interprofessional communication in patient care, which may compromise patient outcomes [18]. Although the medication management and decision-making process in clinical practice revolves around consultants, a space should be made available for both interprofessional sup- port and autonomous decision-making for juniors to improve practice and reduce errors [19]. An area of possible improvement involves restricting unnecessarily prolonged antibiotic use. Unfortunately, this is a commonly reported practice in various clinical fields due to fear of pos- sible complications if antibiotic treatment is terminated on time [9, 20, 21]. Several centers report that activating the role of a clinical pharmacist improves guideline compliance. For instance, the expansion of pharmacists’ role in an orthopedic unit in Australia ensured correct weight-based dosing and SAP duration [22]. In another recent quasi-experimental study, phar- macists had a pivotal role in optimizing the initial vancomycin doses and dose adjustments [23]. TDM process coordination, especially in wards where nurses were responsible for dose administration and junior physicians were responsible for sample collection, was lacking. Despite the theoretical understanding of HCP of the different TDM steps and the importance of TDM sample timing relative to trough collection for correct dose adjustment decisions, some physicians didn’t take the TDM process seriously due to their belief that it is not detri- mental to optimizing patient outcomes. Interprofessional communication seemed also lacking; these communication issues seemed not only to happen outside working hours but also during working hours, as there was a lack of standardized laboratory ordering times and linked post- TDM action. Organizing the TDM process efficiently might be difficult as there is a deficiency in a structured hospital system that coordinates sampling times relative to dosing time. Such deficiencies in TDM were also reported in the TDM of immunosuppressant tacrolimus, where trough levels were drawn appropriately only in 25.9% of the cases. Furthermore, only 38.1% of the drug administrations occurred within one hour of laboratory study collection [24]. Our previous work revealed that in pediatric wards at JUH [10], discrepancies between the recorded and actual times of dose administration were noted in 83.9% of audited occasions. In comparison, such discrepancies were noted in 82.7% of audited times of sample collection. In the present study, HCPs were aware that recorded times were inaccurate. However, they explained that the main driver for this inaccuracy was related to work pressure, as recording accurate times seemed to have a lower priority compared to other daily tasks for which HCPs are responsible. Several optimization strategies have been suggested, including continuous training and pro- viding local guidelines, which is essential. However, they need to be implemented in addition to other measures, such as reducing workload and activating the role of clinical pharmacists. Additional measures include providing support systems to better perform vancomycin dosing and monitoring. Such a system contains alarms to remind them of scheduled blood sample collection and calculators to help them optimize the dose selection. This can be useful since dose changes in actual practice are empirically based on a slight increase or decrease in dosage regimen. Initial pilot studies evaluating such tools in clinical practice reported enchantments in monitoring processes (reduced number of blood samples and side effects) during the study period. However, long-term effects still need to be evaluated [25]. The main limitation of the present study is the convenience sampling technique, as only HCPs interested in vancomycin prescribing and TDM were included in the study. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285717 May 17, 2023 11 / 14 PLOS ONEVancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice Conclusions For enhancing vancomycin prescription and TDM, the present study highlights the need for creating effective communication networks between HCP, reducing workload, and adapting local guidelines in clinical practice suitable for implementation in the local context. For TDM specifically, a strong organizational structure for the TDM process is needed to consolidate dif- ferent TDM aspects and accommodate effective interprofessional communication. This is quite important as TDM is not only unique for vancomycin, and such structure, if proven use- ful, can be utilized for the TDM of other drugs. Supporting information S1 File. Interview guide (translated version). (DOCX) Author Contributions Conceptualization: Mariam Hantash Abdel Jalil, Rima Ηijazeen, Farah Khaled Abu-Mahfouz, Khawla Abu Hammour, Maria Hasan Matalqah, Jwan Saleh Khaleel Albadaineh, Shrouq Khaled AlOmoush, Montaha Al-Iede. Data curation: Mariam Hantash Abdel Jalil, Rima Ηijazeen, Farah Khaled Abu-Mahfouz, Khawla Abu Hammour, Maria Hasan Matalqah, Jwan Saleh Khaleel Albadaineh, Shrouq Khaled AlOmoush, Montaha Al-Iede. Formal analysis: Mariam Hantash Abdel Jalil, Rima Ηijazeen, Khawla Abu Hammour, Maria Hasan Matalqah, Jwan Saleh Khaleel Albadaineh, Shrouq Khaled AlOmoush. Funding acquisition: Mariam Hantash Abdel Jalil. Methodology: Mariam Hantash Abdel Jalil, Rima Ηijazeen, Farah Khaled Abu-Mahfouz, Khawla Abu Hammour, Maria Hasan Matalqah, Montaha Al-Iede. Supervision: Mariam Hantash Abdel Jalil. Validation: Mariam Hantash Abdel Jalil, Rima Ηijazeen, Farah Khaled Abu-Mahfouz, Khawla Abu Hammour, Maria Hasan Matalqah, Jwan Saleh Khaleel Albadaineh, Shrouq Khaled AlOmoush, Montaha Al-Iede. Writing – original draft: Mariam Hantash Abdel Jalil, Rima Ηijazeen, Farah Khaled Abu- Mahfouz. Writing – review & editing: Mariam Hantash Abdel Jalil, Rima Ηijazeen, Farah Khaled Abu- Mahfouz, Khawla Abu Hammour, Maria Hasan Matalqah, Jwan Saleh Khaleel Albadaineh, Shrouq Khaled AlOmoush, Montaha Al-Iede. References 1. Oskarsdottir K, Haraldsson A, Thorkelsson T, Oskarsdottir T, Gunnarsson P, Thors V. Children may need higher vancomycin doses to achieve therapeutic levels. Acta Paediatr 2021; 110:3077–82. https:// doi.org/10.1111/apa.16025 PMID: 34233034 2. Stone SB, Benner K, Utley A, Maclennan P, Coghill CH. Achieving Vancomycin Troughs Within Goal Range in Low Birth Weight Neonates. J Pediatr Pharmacol Ther 2021; 26:56–61. https://doi.org/10. 5863/1551-6776-26.1.56 PMID: 33424501 3. Abraham J, Sinnollareddy MG, Roberts MS, Williams P, Peake SL, Lipman J, et al. Plasma and intersti- tial fluid population pharmacokinetics of vancomycin in critically ill patients with sepsis. Int J Antimicrob Agents 2019; 53:137–42. https://doi.org/10.1016/j.ijantimicag.2018.09.021 PMID: 30296581 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285717 May 17, 2023 12 / 14 PLOS ONEVancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice 4. Mishra V, Chouinard M, Keiser J, Wagner B, Yen MS, Banas C, et al. Automating Vancomycin Monitor- ing to Improve Patient Safety. Jt Comm J Qual Patient Saf 2019; 45:757–62. https://doi.org/10.1016/j. jcjq.2019.07.001 PMID: 31526711 5. Patel J, Lucas CJ, Ryan J, Jenkins M, Martin JH. Vancomycin therapeutic drug monitoring in paediat- rics. J Paediatr Child Health 2020; 56:563–70. https://doi.org/10.1111/jpc.14683 PMID: 31721353 6. Di Pentima MC, Chan S, Eppes SC, Klein JD. Antimicrobial prescription errors in hospitalized children: role of antimicrobial stewardship program in detection and intervention. Clin Pediatr (Phila) 2009; 48:505–12. https://doi.org/10.1177/0009922808330774 PMID: 19224865 7. Parker HM, Mattick K. The determinants of antimicrobial prescribing among hospital doctors in England: a framework to inform tailored stewardship interventions. Br J Clin Pharmacol 2016; 82:431–40. https:// doi.org/10.1111/bcp.12953 PMID: 27038778 8. Tavakoli-Ardakani M, Ghassemi S, Alizadeh AM, Salamzadeh J, Ghadiani M, Ghassemi S. Effects of Pharmacist Intervention on the Utilization of Vancomycin in a Teaching Hospital. Iran J Pharm Res IJPR 2015; 14:1281. PMID: 26664398 9. Kim NH, Koo HL, Choe PG, Cheon S, Kim MS, Lee MJ, et al. Inappropriate continued empirical vanco- mycin use in a hospital with a high prevalence of methicillin-resistant Staphylococcus aureus. Antimi- crob Agents Chemother 2015; 59:811–7. https://doi.org/10.1128/AAC.04523-14 PMID: 25403664 10. Abdel Jalil M, Khaled F, Qaryouti F, Abu Hammour K, Alsous M, Al-Iede M. Vancomycin audit in the paediatric population: Patterns of use and appropriateness of therapeutic drug monitoring. Basic Clin Pharmacol Toxicol 2023. https://doi.org/10.1111/bcpt.13849 PMID: 36847106 11. Cortoos PJ, De Witte K, Peetermans WE, Simoens S, Laekeman G. Opposing expectations and subop- timal use of a local antibiotic hospital guideline: a qualitative study. J Antimicrob Chemother 2008; 62:189–95. https://doi.org/10.1093/jac/dkn143 PMID: 18397925 12. Chan JOS, Baysari MT, Carland JE, Sandaradura I, Moran M, Day RO. Barriers and facilitators of appropriate vancomycin use: prescribing context is key. Eur J Clin Pharmacol 2018; 74:1523–9. https:// doi.org/10.1007/s00228-018-2525-2 PMID: 30056569 13. Newham R, Thomson AH, Semple Y, Dewar S, Steedman T, Bennie M. Barriers to the safe and effec- tive use of intravenous gentamicin and vancomycin in scottish hospitals, and strategies for quality improvement. Eur J Hosp Pharm 2015; 22:32–7. https://doi.org/10.1136/EJHPHARM-2014-000483/-/ DC1 14. Van Dort BA, Baysari MT, Carland JE, Stocker SL, Braithwaite HE, Fernon AR, et al. Education to improve vancomycin use: the perspectives of educators and education recipients. Intern Med J 2020; 50:565–72. https://doi.org/10.1111/imj.14408 PMID: 31211885 15. Tong A, Sainsbury P, Craig J. Consolidated criteria for reporting qualitative research (COREQ): a 32- item checklist for interviews and focus groups. Int J Qual Heal Care 2007; 19:349–57. https://doi.org/10. 1093/intqhc/mzm042 PMID: 17872937 16. Pope C, Ziebland S, Mays N. Analysing qualitative data. BMJ 2000; 320:114–6. 17. Livorsi D, Comer AR, Matthias MS, Perencevich EN, Bair MJ. Barriers to guideline-concordant antibiotic use among inpatient physicians: A case vignette qualitative study. J Hosp Med 2016; 11:174–80. https://doi.org/10.1002/jhm.2495 PMID: 26443327 18. Chan B, Reeve E, Matthews S, Carroll PR, Long JC, Held F, et al. Medicine information exchange net- works among healthcare professionals and prescribing in geriatric medicine wards. Br J Clin Pharmacol 2017; 83:1185. https://doi.org/10.1111/bcp.13222 PMID: 28009444 19. Creswick N, Westbrook JI. Who Do Hospital Physicians and Nurses Go to for Advice About Medica- tions? A Social Network Analysis and Examination of Prescribing Error Rates. J Patient Saf 2015; 11:152–9. https://doi.org/10.1097/PTS.0000000000000061 PMID: 24583953 20. Abdel Jalil MH, Abu Hammour K, Alsous M, Awad W, Hadadden R, Bakri F, et al. Surgical site infections following caesarean operations at a Jordanian teaching hospital: Frequency and implicated factors. Sci Rep 2017; 7. https://doi.org/10.1038/s41598-017-12431-2 PMID: 28939862 21. Abdel Jalil MH, Abu Hammour K, Alsous M, Hadadden R, Awad W, Bakri F, et al. Noncompliance with surgical antimicrobial prophylaxis guidelines: A Jordanian experience in cesarean deliveries. Am J Infect Control [Epub Ahead Print] 2017. https://doi.org/10.1016/j.ajic.2017.06.033 PMID: 28800838 22. Ierano C, Thursky K, Peel T, Rajkhowa A, Marshall C, Ayton D. Influences on surgical antimicrobial pro- phylaxis decision making by surgical craft groups, anaesthetists, pharmacists and nurses in public and private hospitals. PLoS One 2019; 14:e0225011. https://doi.org/10.1371/journal.pone.0225011 PMID: 31725771 23. Hussain K, Ikram R, Ambreen G, Salat MS. Pharmacist-directed vancomycin therapeutic drug monitor- ing in pediatric patients: a collaborative-practice model. J Pharm Policy Pract 2021; 14:100. https://doi. org/10.1186/s40545-021-00383-y PMID: 34847951 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285717 May 17, 2023 13 / 14 PLOS ONEVancomycin prescribing and therapeutic drug monitoring: Challenges of real clinical practice 24. Strohbehn GW, Pan WW, Petrilli CM, Heidemann L, Larson S, Aaronson KD, et al. Large-scale variabil- ity of inpatient tacrolimus therapeutic drug monitoring at an academic transplant center: A retrospective study. Ther Drug Monit 2018; 40:394–400. https://doi.org/10.1097/FTD.0000000000000526 PMID: 29750738 25. Carland JE, Elhage T, Baysari MT, Stocker SL, Marriott DJE, Taylor N, et al. Would they trust it? An exploration of psychosocial and environmental factors affecting prescriber acceptance of computerised dose-recommendation software. Br J Clin Pharmacol 2021; 87:1215–33. https://doi.org/10.1111/bcp. 14496 PMID: 32691902 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285717 May 17, 2023 14 / 14 PLOS ONE
10.1371_journal.pone.0285756	RESEARCH ARTICLE Chromosome-scale genome sequence assemblies of the ‘Autumn Bliss’ and ‘Malling Jewel’ cultivars of the highly heterozygous red raspberry (Rubus idaeus L.) derived from long- read Oxford Nanopore sequence data R. Jordan Price1, Jahn DavikID Samantha LynnID Muath Alsheikh5, Richard J. HarrisonID 2, Felicidad Fernande´ z Fernande´ z3, Helen J. BatesID 1, 3, Charlotte F. Nellist1, Matteo ButiID 4, Dag Røen5, Nada Sˇ urbanovski3, 1, Daniel James SargentID 3* 1 Cambridge Crop Research, NIAB, Cambridge, United Kingdom, 2 Division of Biotechnology and Plant Health, Norwegian Institute of Bioeconomy Research, Ås, Norway, 3 Department of Genetics, Genomics and Breeding, NIAB, East Malling, Kent, United Kingdom, 4 Department of Agriculture, Food, Environment and Forestry (DAGRI), University of Florence, Florence, Italy, 5 Graminor Breeding Ltd., Ridabu, Norway * dan.sargent@niab.com a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Abstract Citation: Price RJ, Davik J, Fernande´z Fernande´z F, Bates HJ, Lynn S, Nellist CF, et al. (2023) Chromosome-scale genome sequence assemblies of the ‘Autumn Bliss’ and ‘Malling Jewel’ cultivars of the highly heterozygous red raspberry (Rubus idaeus L.) derived from long-read Oxford Nanopore sequence data. PLoS ONE 18(5): e0285756. https://doi.org/10.1371/journal.pone.0285756 Editor: Hidenori Sassa, Chiba Daigaku, JAPAN Received: November 18, 2022 Accepted: May 1, 2023 Published: May 16, 2023 Copyright: © 2023 Price et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All raw sequencing data and genome assemblies presented here are available at the NCBI under the Bioproject IDs PRJNA886864, PRJNA886865 and PRJNA886875. Funding: The author(s) received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. Red raspberry (Rubus idaeus L.) is an economically valuable soft-fruit species with a rela- tively small (~300 Mb) but highly heterozygous diploid (2n = 2x = 14) genome. Chromo- some-scale genome sequences are a vital tool in unravelling the genetic complexity controlling traits of interest in crop plants such as red raspberry, as well as for functional genomics, evolutionary studies, and pan-genomics diversity studies. In this study, we devel- oped genome sequences of a primocane fruiting variety (‘Autumn Bliss’) and a floricane variety (‘Malling Jewel’). The use of long-read Oxford Nanopore Technologies sequencing data yielded long read lengths that permitted well resolved genome sequences for the two cultivars to be assembled. The de novo assemblies of ‘Malling Jewel’ and ‘Autumn Bliss’ contained 79 and 136 contigs respectively, and 263.0 Mb of the ‘Autumn Bliss’ and 265.5 Mb of the ‘Malling Jewel’ assembly could be anchored unambiguously to a previously pub- lished red raspberry genome sequence of the cultivar ‘Anitra’. Single copy ortholog analysis (BUSCO) revealed high levels of completeness in both genomes sequenced, with 97.4% of sequences identified in ‘Autumn Bliss’ and 97.7% in ‘Malling Jewel’. The density of repetitive sequence contained in the ‘Autumn Bliss’ and ‘Malling Jewel’ assemblies was significantly higher than in the previously published assembly and centromeric and telomeric regions were identified in both assemblies. A total of 42,823 protein coding regions were identified in the ‘Autumn Bliss’ assembly, whilst 43,027 were identified in the ‘Malling Jewel’ assembly. These chromosome-scale genome sequences represent an excellent genomics resource for red raspberry, particularly around the highly repetitive centromeric and telomeric regions of the genome that are less complete in the previously published ‘Anitra’ genome sequence. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 1 / 17 PLOS ONERed raspberry ONT sequence assembly Introduction Red raspberry (Rubus idaeus L.) is a popular, highly heterozygous diploid (2n = 2x = 14) peren- nial crop plant, with 822,493 tonnes of raspberries harvested annually throughout the world (http://www.fao.org/faostat/en/#data/QC). As such, red raspberry is one of the most economi- cally valuable soft-fruit species, and interest in the development of new varieties through breeding and selection has led to many breeding programmes for the species being established globally. There are two main flowering habits in commercial red raspberry germplasm: flori- cane fruiting (such as the cultivar ‘Malling Jewel’) in which canes grown the previous season produce lateral shoots in the second year of growth that bear the flowers and fruits; and primo- cane fruiting (such as the cultivar ‘Autumn Bliss’), in which the first-year canes bear a limited number of flowers and fruits in the late summer or early autumn. Conventional breeding of red raspberry germplasm has led to many commercially-successful varieties being released, however genetic improvement in this highly heterozygous outbreeding species is slow [1] with several factors, including inbreeding depression, deleterious recessive alleles [2] and loss of fer- tility, amongst others, making the breeding process challenging. Plant genomics has the poten- tial to significantly increase the precision and accuracy of breeding and selection of crop plants through the development and application of molecular markers, marker-assisted breeding and genomic selection. As such, significant molecular resources have been developed to date to support the breeding effort in red raspberry [3]. Chromosome-scale genome sequences are a vital tool in unravelling the genetic complexity controlling traits of interest in crop plants, as well as for functional genomics, evolutionary studies, and pan-genomics diversity studies. However, there are significant challenges in assembling genomes of highly heterozygous species using short-read sequence data because heterozygosity significantly increases the complexity of the de Bruijn graph structure predomi- nantly used in short-read assemblers. Additionally, sequence length may make correct resolu- tion of haplotigs in highly heterozygous genomes intractable, leading to fragmented assemblies containing many small contigs. Despite a relatively small haploid genome size of ~300 Mb, a diploid genome structure, and good progress in early sequencing efforts for red raspberry [4], the highly heterozygous nature of the genome, and the relatively high cost of early short-read sequencing data, meant that initial assemblies were highly fragmented and consisted of many thousands of scaffolds [5]. Thus, the development of chromosome-scale genomics resources for red raspberry has lagged behind those of closely-related species with genomes of a similar size such as F. vesca [6, 7], Potentilla micrantha [8] and Rubus occidentalis [9]. The advent of long-read sequencing technologies such as the platforms offered by Pacific Biosciences (PacBio) and Oxford Nanopore Technologies (ONT) has permitted high-quality, chromosome-scale genome sequence assemblies of many plant species, including those with relatively complex genomes such as the allo-octoploid strawberry F. × ananassa [7]. Long-read PacBio sequence data was recently used to construct chromosome-scale genome sequences of R. idaeus cultivars ‘Joan J’ [10] and ‘Anitra’ [11]. In the study of Davik et al., [11] the genome sequence covered 291.7 Mb, more than 99% of the estimated genome size of R. idaeus, with 85% of the sequence contigs resolved incorporated into seven chromosome-scale scaffolds, and 98% of Benchmarking Universal Single-Copy Orthologs (BUSCO) genes present in the assembly. The sequence of ‘Joan J’ [10] covered a total of ~300 Mb with a BUSCO complete- ness of 95.3%, but summary statistics for the genome were not available, and to date, the sequence from that publication has not been deposited in public sequence data repositories. These chromosome-scale genome sequences represent an excellent genomics resource for red raspberry, however additional sequence data and assemblies will help improve overall genome coverage and completeness, particularly around the highly repetitive centromeric and PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 2 / 17 PLOS ONERed raspberry ONT sequence assembly telomeric regions of the genome. More complete sequences will increase the value of these genomics resources for breeding and selection, as well as for more fundamental studies within the species. Here we present the genome sequences for two red raspberry cultivars, ‘Autumn Bliss’ and ‘Malling Jewel’, along with associated gene predictions. ‘Malling Jewel’ [‘Preussen’ x (S1‘Lloyd George’ x S1‘Pyne’s Royal’ F2)] is a floricane red raspberry variety that was released in 1949 and represents a ‘pure’ R. idaeus genotype with 25% genetic contribution from R. idaeus strigo- sus (from ‘Superlative’) in a background of R. idaeus vulgatus. In contrast, ‘Autumn Bliss’, released in 1983 is a strict primocane variety with a complex hybrid genetic background [12] which has been used extensively as a parent in breeding worldwide for its early and productive primocane season, its aphid resistance, and its tolerance to soil-borne pathogens. The genome sequences produced in this investigation were compared to the recently-published genome sequence of the cultivar ‘Anitra’ [11]. The use of long-read ONT sequencing data yielded sig- nificantly longer read lengths than the PacBio sequence employed in the assembly presented by Davik et al. [11], which permitted more complete genome sequence assemblies to be resolved and in which the majority of the centromeric regions were defined. Materials and methods DNA extraction and sequencing DNA was extracted from fresh, young leaf material collected from a single plant of the R. idaeus cultivars ‘Autumn Bliss’ and ‘Malling Jewel’ using a high molecular weight genomic DNA extraction protocol [13]. Long-read sequencing libraries were prepared using the SQK-LSK108 Ligation Sequencing Kit (Oxford Nanopore Technologies) from approximately 1 μg of high molecular weight genomic DNA, following the manufacturer’s protocol. Long- read libraries were sequenced on R9.4.1 Spot-On Flow cells (FLO-MIN106) using the Grid- ION X5 platform (Oxford Nanopore Technologies) set to high accuracy base calling. A PCR free short read Illumina sequencing library was prepared for each of the two cultivars using an insert size of 450 bp. The ’Malling Jewel’ libraries were sequenced with 250 bp paired end reads on the Illumina HiSeq 2500 platform at the Earlham Institute (Norwich, UK) and the ‘Autumn Bliss’ libraries were sequenced with 300 bp paired end reads on the Illumina MiSeq platform at NIAB East Malling. RNA extraction and sequencing Developing fruits of the red raspberry cultivar ‘Anitra’ were collected at three maturity stages; unripe, turning, and fully mature. Fruit samples from each maturity stage were collected at Graminor Njøs (Norway), where they were divided into three biological replicates, snap frozen in liquid nitrogen and stored at -80˚C until the samples were processed further. Tissue samples were ground to a fine powder under liquid nitrogen, and total RNA was extracted using the RNeasy Plant Mini Kit (Qiagen, Germany) following the manufacturer’s instructions. The concentration and purity of the resultant RNA was measured using a QIAxpert spectropho- tometer (Qiagen, Germany) and the integrity of the RNA was determined using a Qubit 4.0 fluorimeter (Thermo Fisher Scientific, UK). Samples with an RNA integrity number (RIN) value above 7.0 were submitted for subsequent RNA-Seq. Library preparations were per- formed using the NEB Next1 ultra RNA Library Prep Kit (Biolabs, Inc., Beijing, China) and 150 bp paired-end sequencing was performed by Norwegian Sequencing Centre (Oslo Univer- sity Hospital, Norway) using the HiSeq2500 platform (Illumina Inc., Beijing, China) to yield between 12.6 and 19.4 Gb of data per sample. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 3 / 17 PLOS ONERed raspberry ONT sequence assembly Genome assembly Before assembling the sequence data, the genome size of the ‘Autumn Bliss’ and ‘Malling Jewel’ genomes was estimated by counting k-mers (n = 27) in the Illumina reads in each dataset and calculating a histogram of the k-mer frequencies vs. counts using KAT [14], which were plotted using GenomeScope [15]. Long reads were quality controlled using NanoPlot v1.30.1 [16] and adapters were trimmed using Porechop v0.2.4 (https://github.com/rrwick/Porechop) using default parameters. Reads shorter than 1 Kb or with a quality score less than Q9 were removed using Filtlong v0.2.1 (https://github.com/rrwick/Filtlong). Long reads were assembled using NECAT v0.0.1_update20200803 [17] using a genome size of 300 Mb; all other parameters were left as default. Following assembly, heterozygous contigs and contig overlaps were identified and removed using Purge_Dups v1.0.1 [18] with default settings. Error correction was performed by aligning the long reads to the ‘Autumn Bliss’ and ‘Malling Jewel’ assemblies with Minimap2 v2.17-r941 [19] to inform one iteration of Racon v1.4.20 [20], followed by one iteration of Medaka v1.5.0 (https://github.com/ nanoporetech/medaka) using the r941_min_high_g360 model. Illumina paired-end reads were qual- ity controlled using FastQC v0.11.9 (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/), and adapters and low-quality regions were trimmed using Trimmomatic v0.39 [21]. Short reads were aligned to the purged and corrected long-read assemblies using Bowtie2 v2.4.4 [22] and Samtools v1.12 [23] to allow for three iterations of polishing using Pilon v1.24 [24]. Reference-guided assembly was performed using RagTag v2.1.0 [25] with the R. idaeus cv. ‘Anitra’ chromosome-level assembly [11] as the reference. Assembly statistics were generated using a custom Python script, and single copy ortholog analysis was performed using BUSCO v5.2.2 [26], using the eudicots_odb10 database. Gene prediction and annotation Repetitive and low complexity sequences in the reference-guided assemblies for ‘Autumn Bliss’ and ‘Malling Jewel’ were identified and soft masked using Red version: 05/22/2015 [27]. The R. idaeus cv. ‘Anitra’ RNA-seq reads were quality controlled using FastQC v0.11.9 (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/), and adapters and low-quality regions were trimmed using Trimmomatic v0.39 [21]. Assemblies were indexed and RNA-seq reads were aligned using HISAT2 v2.2.1 [28] using default settings. Gene prediction was per- formed using BRAKER2 v2.1.6 [29] under the–etpmode setting with the RNA-seq and the eudicots protein database as evidence. Annotation completeness of the genome was assessed using BUSCO v5.2.2 [26], using default parameters and the gene families set defined for the eudicots_odb10 database. Gene predictions for the purged alternative contigs in each assembly were performed and gene predictions unique to the these contigs were identified using blastn. The repeat co-ordinates file output from Red and the ‘gene’ co-ordinates from the annotations file output from BRAKER2 were plotted using pyCircos (https://github.com/ponnhide/ pyCircos) to visualise the distribution and density of repeat and coding regions. Tandem repeat content Tandem Repeats Finder [30] was used to analyse the tandem repeat regions in the assemblies of ‘Autumn Bliss’, ‘Malling Jewel’ and ‘Anitra’ using the following settings; match = 2, mis- match = 7, delta = 7, match probability = 80, indel probability = 10, minimum alignment score = 50, max period = 2000. Functional annotation Pairwise sequence comparison of the predicted proteomes of each genome, along with the genes found uniquely in the alternative purged contigs was performed using the BLAST PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 4 / 17 PLOS ONERed raspberry ONT sequence assembly + blastp-fast algorithm [31] through the Galaxy platform [32] against the NCBI nr, SwissProt, RefSeq, TrEMBL and Araport11 protein databases using an expectation value cutoff of 1e-6. InterProScan v5 [33] was used to assign InterPro domains and Gene Ontology (GO) terms, whilst BlastKOALA v2.2 [34] and eggNOG-mapper v2 [35] were used to map Kyoto Encyclo- paedia of Genes and Genomes (KEGG) orthologs and KEGG pathways respectively. GO enrichment analyses were carried out using Cytoscape 3.9.1 [36] and the BinGO 3.0.3 plug-in [37] performing a hypergeometric test with False Discovery Rate (FDR) correction, a 0.05 sig- nificance level and GO Full ontology. Comparative genomics Syntenic alignments to the ‘Anitra’ genome sequence [11] were generated and plotted using D-genies [38] implementing Minimap2 v2.17-r941 [19] for the main assemblies for both ‘Autumn Bliss’ and ‘Malling Jewel’. Additionally, the alternative contigs purged from heterozy- gous regions of the genome during assembly were plotted against the primary assemblies of each cultivar to determine their genomic positions. OrthoFinder [39] was used to identify gene families in the ‘Autumn Bliss’, ‘Malling Jewel’ and ‘Anitra’ genome sequences [11], along with the genome sequences of the related species R. chingii v1.0 [40], R. occidentalis v3.0 [41] and Fragaria vesca v4.0.a2 [42]. Presence or absence of genes in orthogroups was used to deter- mine species- and cultivar-specific gene families. Results Long- and short-read sequencing data Sequencing of high molecular weight genomic DNA from ‘Autumn Bliss’ and ‘Malling Jewel’ using long read ONT libraries yielded 15.8 Gb and 12.0 Gb of data for the two cultivars respec- tively. Following filtering, the sequencing datasets contained 833,623 reads for ‘Malling Jewel’ and 991,952 reads for ‘Autumn Bliss’, representing 55× and 42× coverage, respectively. The read length N50 was 37.5 Kb for ‘Malling Jewel’ and 18.4 Kb for ‘Autumn Bliss’, with maximum read lengths of 456,177 bp and 366,503 bp, respectively for the two cultivars (Table 1). After adapter trimming and the removal of low-quality sequence data, a total of 18.7 Gb of 250 bp paired-end Illumina sequencing data was produced for ‘Malling Jewel’ and 5.0 Gb of 300 bp paired-end Illumina sequencing data was produced for ‘Autumn Bliss’, representing 69× and 19× genome coverage, respectively. Genome sequence assembly The k-mer analysis performed for ‘Autumn Bliss’ returned a predicted heterozygosity of 1.54% and a total predicted genome length of 196.8–198.5 Mbp. The data for ‘Malling Jewel’ returned a predicted heterozygosity of 0.45% and a total predicted genome length of 294.9–298.6 Mbp Table 1. Filtered Oxford Nanopore reads used for the ‘Malling Jewel’ and ‘Autumn Bliss’ genome sequence assemblies. ‘Autumn Bliss’ ‘Malling Jewel’ Mean read length (bp) Mean read quality Number of reads Read length N50 Total bases Longest read (bp) 11,484 13.4 991,952 18,356 11,391,903,736 366,503 https://doi.org/10.1371/journal.pone.0285756.t001 18,004 13.2 833,623 37,581 15,008,949,175 456,177 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 5 / 17 PLOS ONERed raspberry ONT sequence assembly Table 2. Assembly statistics for the de novo assemblies of the Rubus idaeus ‘Malling Jewel’ and ‘Autumn Bliss’ genome sequences. ‘Autumn Bliss’ ‘Malling Jewel’ Total sequence length (bp) Number of contigs Longest contig (bp) Shortest contig (bp) GC content (%) Contig N50 (bp) Contig L50 Gap (%) Complete BUSCOs (%) Single copy BUSCOs (%) Duplicated BUSCOs (%) Fragmented BUSCOs (%) Missing BUSCOs (%) 268,668,490 146 9,660,097 3,465 37.8 3,253,875 25 0 97.7 89 8.7 0.4 1.8 265,724,784 83 40,892,532 3,226 37.85 9,899,270 7 0 97.6 93.6 4 0.4 2 https://doi.org/10.1371/journal.pone.0285756.t002 (S1 Fig). Following long-read assembly, removal of heterozygous contigs and polishing using long- and short- read sequence data, the resulting assemblies for ‘Autumn Bliss’ and ‘Malling Jewel’ were 268.7 Mb and 265.7 Mb in length, and were composed of 146 and 86 contigs, respectively. The ‘Malling Jewel’ assembly had an N50 of 9.9 Mb with an L50 of seven contigs, whilst the ‘Autumn Bliss’ assembly had an N50 of 3.3 Mb and an L50 of 25 contigs (Table 2). A total of 339 contigs in ‘Autumn Bliss’ and 154 contigs in ‘Malling Jewel’ were purged from the heterozygous regions in the sequence assemblies. The total length of the purged contigs was 202.9 Mb and 235.6 Mb ‘Autumn Bliss’ and ‘Malling Jewel’ respectively. BUSCO analysis revealed 97.7% complete single copy orthologs in ‘Autumn Bliss’ and 97.6% in ‘Malling Jewel’, indicating that both assemblies contained high levels of gene space completeness (Table 2). Genome sequence scaffolding The de novo assemblies of ‘Autumn Bliss’ and ‘Malling Jewel’ were scaffolded against the seven pseudochromosomes of the ‘Anitra’ genome sequence. Following scaffolding, 136 contigs of ‘Autumn Bliss’ and 79 contigs of ‘Malling Jewel’ were anchored to the seven ‘Anitra’ pseudo- chromosomes (Table 3). In total, 263.0 Mb (97.9%) of the ‘Autumn Bliss’ assembly was anchored to the ‘Anitra’ chromosomes, with a largest scaffold of 50.7 Mb and an N50 of 35.7 Mb, whilst 265.5 Mb (99.9%) of the ‘Malling Jewel’ assembly was anchored to the ‘Anitra’ chro- mosomes, with a largest scaffold of 44.5 Mb and an N50 of 36.1 Mb (Table 4). Of the ten contigs Table 3. Rubus idaeus ‘Malling Jewel’ and ‘Autumn Bliss’ contigs aligning to ‘Anitra’ genome pseudomolecules. ‘Autumn Bliss’ ‘Malling Jewel’ pseudochromosome 1 pseudochromosome 2 pseudochromosome 3 pseudochromosome 4 pseudochromosome 5 pseudochromosome 6 pseudochromosome 7 Total 16 19 23 12 19 30 17 136 https://doi.org/10.1371/journal.pone.0285756.t003 9 13 5 15 10 15 12 79 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 6 / 17 PLOS ONETable 4. Assembly statistics for the reference scaffolded assemblies of the Rubus idaeus ‘Anitra’, ‘Malling Jewel’ and ‘Autumn Bliss’ genome sequences. ‘Anitra’ ‘Autumn Bliss’ ‘Malling Jewel’ Red raspberry ONT sequence assembly Total sequence length (bp) Number of scaffolds Longest scaffold (bp) Shortest scaffold (bp) GC content (%) Scaffold N50 (bp) Scaffold L50 Gap (%) Masked sequence (%) Complete BUSCOs (%) Single Copy BUSCOs (%) Duplicated BUSCOs (%) Fragmented BUSCOs (%) Missing BUSCOs (%) Contigs aligned to ‘Anitra’ reference Sequence aligned to ‘Anitra’ reference (%) https://doi.org/10.1371/journal.pone.0285756.t004 247,480,545 7 43,293,752 28,685,549 37.62 34,491,998 4 0.1 31.7 97.4 93.3 4.1 0.5 2.1 262,950,453 7 50,701,465 32,095,515 37.79 35,733,366 4 0 32.6 97.4 90.1 7.3 0.3 2.3 136 97.9 265,504,618 7 44,467,746 34,604,543 37.84 36,052,904 4 0 34.8 97.6 93.9 3.8 0.3 2 79 99.9 that were not anchored from the ‘Autumn Bliss’ assembly, nine were included in other contigs, whilst the longest contig (4.6 Mb) contained highly repetitive sequences. The four contigs from the ‘Malling Jewel’ assembly that were not anchored to the ‘Anitra’ chromosomes repre- sented fragmented sequences of the chloroplast and mitochondrial genomes. The scaffolded, anchored assemblies were 15.5 Mb (6.3%) and 18 Mb (7.3%) larger than the previously pub- lished ‘Anitra’ genome sequence for ‘Autumn Bliss’ and ‘Malling Jewel’ respectively. Single copy ortholog analysis (BUSCO) revealed high levels of completeness in both genomes, with 97.4% of the sequences identified in ‘Autumn Bliss’ and 97.7% of the sequences identified in ‘Malling Jewel’ (Table 4). The seven pseudochromosomes assembled for each genome were numbered according to the ‘Anitra’ chromosomes which in turn are concordant with the chro- mosomes of other Rosoideae genomes such as F. vesca [6] and Rosa chinensis [43]. All raw sequencing data and genome assemblies presented here are available at the NCBI under the Bioproject IDs PRJNA886864, PRJNA886865 and PRJNA886875. The assemblies of the ‘Autumn Bliss’ and ‘Malling Jewel’ genomes, along with the transcriptome annotations of the ‘Autumn Bliss’, ‘Malling Jewel’ and ‘Anitra’ have also been deposited on the Genome Data- base for Rosaceae [44]. Assembly completeness The assembled ‘Autumn Bliss’ and ‘Malling Jewel’ contigs were highly syntenic and concor- dant with the previously published R. idaeus cv. ‘Anitra’ genome sequence of Davik et al. [11], despite being significantly larger than the ‘Anitra’ pseudochromosome assembly (Fig 1). S2 Fig contains the names and positions of the ‘Autumn Bliss’ and ‘Malling Jewel’ scaffolds along the ‘Anitra’ assembly, along with their lengths in base pairs. S1 File shows the relative positions of the alternative contigs purged from the primary assembly from heterozygous regions of the genome in relation to the associated primary assembly of each cultivar. The additional length of the primary assemblies presented was partly due to 15% of the total ‘Anitra’ assembly length not being scaffolded into the seven main pseudochromosomes, however there were also differ- ences in the abundance of repetitive sequence within the genomes. Extensive repetitive regions were identified throughout the pseudochromosomes of all three assemblies (Fig 2) and the PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 7 / 17 PLOS ONERed raspberry ONT sequence assembly Fig 1. The red raspberry genome assemblies of ‘Autumn Bliss’ and ‘Malling Jewel’ show synteny to the previously published ‘Anitra’ genome. The dot plots show the alignments of ‘Autumn Bliss’ and ‘Malling Jewel’ genome assemblies to the ‘Anitra’ genome sequence. The plots show the synteny between ‘Autumn Bliss’ and ‘Malling Jewel’ contigs (y-axis) aligning to the ‘Anitra’ pseudochromosomes (x-axis). ‘Anitra’ chromosome names are given along with total assembly lengths of the ‘Anitra’, ‘Autumn Bliss (a) and ‘Malling Jewel’ (b) chromosomes in Mbp. The hashed horizontal lines indicate the positions along the y- axis of the assembled scaffolds in each assembly. ‘Autumn Bliss’ and ‘Malling Jewel’ scaffold names and sizes are given in S2 Fig. https://doi.org/10.1371/journal.pone.0285756.g001 relative abundance of sequence within highly repetitive regions in the assemblies was com- pared to assess genome completeness. Regions with a higher density of repetitive sequence and lower gene density were observed near the centre of each of the seven ‘Anitra’ pseudomole- cules, which we inferred correspond to the centromeres of each the seven chromosomes. These same regions were clearly identifiable in the ‘Autumn Bliss’ and ‘Malling Jewel’ pseudo- molecules, and the density of repetitive sequence they contained was significantly higher than in the ‘Anitra’ assembly (Fig 3). The ‘Autumn Bliss’ and ‘Malling Jewel’ assemblies contained a greater total repetitive sequence within the centromeric regions, and more tandem repeat sequences overall on all seven pseudochromosomes than the ‘Anitra’ assembly, except for chromosome 4 of ‘Autumn Bliss’ (Fig 3). Additionally, telomeric repeat sequences ([CCCTAAA]n) were identified at both ends of five pseudochromosomes and at one end of the remaining two pseudochromosomes in ‘Malling Jewel’, whilst one pseudochromosome of the ‘Autumn Bliss’ assembly contained identifiable telomeric repeats at both ends, and a fur- ther three contained identifiable repeats at a single end. Gene prediction and annotation A total of 41,800 protein coding regions were identified in the ‘Autumn Bliss’ assembly, whilst 40,491 were identified in the ‘Malling Jewel’ assembly. The distribution of these protein coding regions across the seven pseudochromosomes for each cultivar is shown in Fig 2. A further 1,023 genes in ‘Autumn Bliss’ and 1,227 genes in ‘Malling Jewel’ not present in the primary assemblies were found in the purged contigs. A BUSCO analysis of the proteome of the two assemblies identified 96.9% queried proteins in the ‘Autumn Bliss’ proteome, and 97.7% in the ‘Malling Jewel’ proteome (Table 5). The ‘Anitra’ genome sequence was re-annotated using the PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 8 / 17 PLOS ONERed raspberry ONT sequence assembly Fig 2. Extensive repetitive regions were identified throughout the pseudochromosomes of the ‘Anitra’, ‘Autumn Bliss’ and ‘Malling Jewel’ genome sequence assemblies. Circos plots showing the distribution of coding regions (green) and repetitive regions (red) in the ‘Autumn Bliss’, ‘Malling Jewel’ and ‘Anitra’ genome sequence assemblies along with the alignment of syntenic regions of ‘Autumn Bliss’ and ‘Malling Jewel’ to the ‘Anitra’ genome. Major tick marks indicate 5 Mb intervals and minor ticks indicate 1 Mb intervals. https://doi.org/10.1371/journal.pone.0285756.g002 same methodology to ensure consistency in downstream orthology analyses [45]. The re- annotation of the ‘Anitra’ assembly resulted in 41,509 protein coding regions, with a BUSCO completeness score of 97.3% compared to 39,448 (91.3%) in the previous study of [11] indicat- ing a largely complete gene-space characterisation in all three genomes. The statistics for the number of protein-coding gene predictions for each cultivar that returned �1 positive hits after the BlastP analysis with nr, Araport11, RefSeq, SwissProt and TrEMBL databases as sub- jects, are given in Table 6, along with the number of protein-coding gene regions assigned Interpro, GO, KEGG orthology and KEGG pathway terms. Best matches resulting from BlastP homology searches for each dataset are detailed in S2 File Gene prediction BlastP summary, whilst the functional annotation results are given in S3 File Gene prediction functional annotation. Orthology analysis A total of 34,213 orthogroups were identified in the comparison between the proteomes of the three R. idaeus cultivars, R. occidentalis [v3,9], R. chingii [v1, 40] and F. vesca [v4.0.a2, 42]. Of the identified orthogroups, 15,412 were shared across all species, whilst 5,286 orthogroups, containing 17,233 genes, were specific to R. idaeus. Of the R. idaeus specific orthogroups, 2,194 were shared across all three cultivars. Of the remaining groups, 915 were shared between ‘Ani- tra’ and ‘Malling Jewel’, 911 were shared between ‘Anitra’ and ‘Autumn Bliss’, and 890 were shared between ‘Malling Jewel’ and ‘Autumn Bliss’. Finally, 123 orthogroups were specific to ‘Anitra’, 130 to ‘Malling Jewel’ and 123 ‘Autumn Bliss’, representing 322, 303 and 305 genes, respectively (Fig 4). GO enrichment analysis of the seven R. idaeus subgroups using all the PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 9 / 17 PLOS ONERed raspberry ONT sequence assembly Fig 3. The density of repetitive sequences in the ‘Autumn Bliss’ and ‘Malling Jewel’ genome assemblies was substantially higher than in the previously published ‘Anitra’ assembly. Bar charts displaying the number of tandem repeat regions and total repetitive regions identified per chromosome in the ‘Anitra’, ‘Autumn Bliss’ and ‘Malling Jewel’ genomes using Tandem Repeat Finder. https://doi.org/10.1371/journal.pone.0285756.g003 orthogroups as background showed that the orthogroups shared across the three cultivars ‘Malling Jewel’, ‘Anitra’ and ‘Autumn Bliss’ were significantly enriched for metabolic process (GO:0008152), primary metabolic process (GO:0044238), macromolecule metabolic process (GO:0043170), DNA binding (GO:0003677), peptidase activity (GO:0008233) and endopepti- dase activity (GO:0004175) GO categories (Fig 5). Orthogroups shared between ‘Anitra’ and ‘Malling Jewel’, between ‘Anitra’ and ‘Autumn Bliss’, between ‘Malling Jewel’ and ‘Autumn PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 10 / 17 PLOS ONERed raspberry ONT sequence assembly Table 5. A comparison of Benchmarking Universal Single-Copy Orthologs (BUSCO) analysis of the proteome from the previously published ‘Anitra’ genome, the reannotated ‘Anitra’ genome, ‘Malling Jewel’ and ‘Autumn Bliss’. ‘Anitra’ (Published) ‘Anitra’ (This study) ‘Autumn Bliss’ ‘Malling Jewel’ Proteins Complete (%) Single (%) Duplicated (%) Fragmented (%) Missing (%) https://doi.org/10.1371/journal.pone.0285756.t005 39,448 91.3 84 7.3 3.7 5 41,509 97.3 83.2 14.1 0.9 1.9 42,823 96.9 83.4 13.5 0.9 2.1 41,718 97.7 87 10.7 0.6 1.7 Bliss’, as well as orthogroups unique to ‘Malling Jewel’, ‘Autumn Bliss’ and ‘Anitra’, were not significantly enriched for any GO category. Discussion Plant genomes are often highly heterozygous and are composed of a significant proportion of highly repetitive DNA [46]. As such, large and highly contiguous assemblies are often intracta- ble with short-read sequence data alone. In this report, we present two new chromosome-scale genome sequence assemblies for red raspberry (R. idaeus) derived from the cultivars ‘Autumn Table 6. Summary statistics for the number of protein-coding gene predictions for ‘Anitra’, ‘Autumn Bliss’ and ‘Malling Jewel’ that returned �1 positive hit after the BlastP analysis with nr, Araport11, RefSeq, SwissProt and TrEMBL databases as subjects, along with the number of protein-coding gene regions assigned Inter- pro, GO, KEGG orthology and KEGG pathway terms. Cultivar Homology Predicted proteins # BlastP vs NCBI nr: hits # BlastP vs NCBI: hits nr % BlastP vs Araport11: hits # BlastP vs Araport11: hits % BlastP vs RefSeq: hits # BlastP vs RefSeq: hits % BlastP vs SwissProt: hits # BlastP vs Swissprot: hits % BlastP vs Trembl: hits # BlastP vs Trembl: hits % Functional Annotation InterPro: total hits # InterPro: proteins with hits # InterPro: proteins with hits % GO: total hits # GO: proteins with hits # GO: proteins with hits % KEGG ortholog: total hits # KEGG ortholog: proteins with hits # KEGG ortholog: proteins with hits % KEGG pathway: total hits # KEGG pathway: proteins with hits # KEGG pathway: proteins with hits % https://doi.org/10.1371/journal.pone.0285756.t006 ‘Anitra’ ‘Autumn Bliss’ ‘Malling Jewel’ 41,509 36,863 88.60% 32,326 77.90% 36,778 88.60% 28,587 68.90% 36,245 87.30% 95,663 31,379 75.60% 52,755 22,732 54.80% 11,363 11,359 27.40% 36,760 9,190 42,823 37,195 86.86% 32,271 75.36% 37,100 86.64% 28,641 66.88% 36,563 85% 93,586 30,991 74.10% 71,290 22,223 53.20% 11,212 11,208 26.80% 36,726 9,152 41,718 35,983 86.25% 31,129 74.62% 35,878 86% 27,700 66.40% 35,368 84.78% 90,763 29,790 73.60% 49,974 21,554 53.20% 10,780 10,776 26.60% 34,649 8,732 0.221397769 0.218947368 0.215652861 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 11 / 17 PLOS ONERed raspberry ONT sequence assembly Fig 4. A Venn diagram showing the distribution of the 5,286 Rubus idaeus-specific orthogroups identified in this study between the ‘Anitra’, ‘Autumn Bliss’ and ‘Malling Jewel’ genomes. https://doi.org/10.1371/journal.pone.0285756.g004 Fig 5. Overrepresented Gene Ontology (GO) categories in the orthogroups shared across ‘Anitra’, ‘Autumn Bliss’ and ‘Malling Jewel’ cultivars highlight key conserved processes. The circles are shaded based on significance level (yellow = False Discovery Rate (FDR) below 5.00E-2), and the radius of circles is proportional to the number of orthogroups included in each GO category. https://doi.org/10.1371/journal.pone.0285756.g005 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 12 / 17 PLOS ONERed raspberry ONT sequence assembly Bliss’ and ‘Malling Jewel’ which were constructed using long-read ONT sequence data and were compared to the previously published genome sequence assembly of the R. idaeus cultivar ‘Anitra’ [11]. Prior to scaffolding with long-range structural information, both genomes sequenced in this investigation returned a high level of contiguity, with just 143 and 83 contigs and a contig L50 of 25 and 7 returned for the ‘Autumn Bliss’ and ‘Malling Jewel’ assemblies respectively. This is in contrast to the 2,350 contigs in the ‘Anitra’ assembly, with both new assemblies con- taining at least 15 Mb more scaffolded sequence data than the ‘Anitra’ sequence [11]. The lower degree of contiguity within the ‘Autumn Bliss’ assembly is likely due to the differences in the long-read N50 values (37.5 Kb vs 18.4 Kb in ‘Malling Jewel’ and ‘Autumn Bliss’) of the sequence data returned for each cultivar. In contrast to the ONT data used in this investiga- tion, Davik et al. [11] used PacBio long-read sequence data for the chromosome-scale assembly of the ‘Anitra’ genome. The PacBio reads used in that assembly had a mean sub-read length of 9.5 Kb and an N50 length of 13.6 Kb, which is significantly shorter than the N50 read length of the ‘Autumn Bliss’ and ‘Malling Jewel’ datasets presented here. In the final assembly, the highly repetitive centromeric regions of the ‘Anitra’ assembly were less well resolved than the gene- rich regions of the genome and significantly less well resolved than the same regions in the ‘Autumn Bliss’ and ‘Malling Jewel’ genomes presented here. The greater fragmentation in the pre-scaffolded ‘Anitra’ contigs than in the ‘Autumn Bliss’ and ‘Malling Jewel’ assemblies resulted in a significantly shorter final scaffolded sequence for ‘Anitra’, demonstrating the value of very long-read sequence data in resolving highly heterozygous genome sequences. The two new genome assemblies for red raspberry presented here displayed a very high degree of synteny with the genome sequence of ‘Anitra’. Analysis of the tandem repeats in the genome sequences of ‘Autumn Bliss’, ‘Malling Jewel’ and ‘Anitra’ demonstrated that much of the addi- tional sequence data incorporated into the new assemblies was contained in the centromeric regions, which were more accurately assembled in the ‘Autumn Bliss’ and ‘Malling Jewel’ genomes due to the longer read length of the ONT data used compared to the shorter PacBio data used in the ‘Anitra’ assembly. Cultivated red raspberry varieties are derived from the interbreeding of both the European red raspberry (R. idaeus subsp. idaeus L.) and the North American red raspberry (R. idaeus subsp. strigosus L.) as well as other species and subspecies. However, the global genetic struc- ture of red raspberry populations has yet to be extensively studied. Modern red raspberry vari- eties are derived from a relatively narrow genetic base [47], with just 20 clones accounting for the majority of the genetic diversity in 137 varieties of known pedigree that have been released worldwide between 1960 and 1992. However, R. idaeus is naturally an out-crossing species, and as such, high levels of genome differentiation and heterozygosity are a feature of red rasp- berry germplasm. The two varieties for which genome sequence assemblies are presented in this paper are from very different backgrounds within the red raspberry germplasm base. ‘Malling Jewel’ is a floricane with a relatively pure R. idaeus genetic background, containing several important founder clones in its pedigree, whilst ‘Autumn Bliss’ in contrast is a strict primocane variety, with a complex hybrid pedigree including contributions from R. arcticus and R. occidentalis, meaning these two genomes span a significant representative portion of cultivated red raspberry diversity. Many traits that have been selected for during the domestication and breeding of red rasp- berry including large fruit size, flavour, colour and pest and disease resistance, and alleles con- trolling these traits are likely to have been through genetic bottlenecks, with not all favourable alleles passed to all breeding populations of the crop globally. The genes contained within the 5,286 R. idaeus-specific orthogroups identified in this study were enriched for metabolic pro- cess, DNA binding, peptidase activity and endopeptidase activity, however, there is likely to be PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 13 / 17 PLOS ONERed raspberry ONT sequence assembly further differentiation within the cultivated and wild red raspberry germplasm that exists glob- ally. The genetic resources for the red raspberry cultivars ‘Autumn Bliss’ and ‘Malling Jewel’ presented here, along with the previously published sequence of the cultivar ‘Anitra’ are valu- able resources to begin to understand the structure and function of the red raspberry pan- genome and provide a route to unlock the potential of red raspberry germplasm through pre- cise identification and characterisation of genetic loci controlling traits of agronomic importance. Supporting information S1 Fig. Plots of the k-mer distribution determined from analysis of illumine reads generated for the (a) ‘Malling Jewel’ and (b) ‘Autumn Bliss’ genomes. (PNG) S2 Fig. The names and positions of the ‘Autumn Bliss’ and ‘Malling Jewel’ scaffolds along the ‘Anitra’ assembly, along with their lengths in base pairs. (PNG) S1 File. The relative positions of the alternative contigs purged from the primary assembly from heterozygous regions of the (a) ‘Autumn Bliss’ and (b) ‘Malling Jewel’ genomes. (XLSX) S2 File. Gene prediction BlastP summary. (XLSX) S3 File. Gene prediction functional annotation. (XLSX) Author Contributions Conceptualization: R. Jordan Price, Felicidad Fernande´z Fernande´z, Richard J. Harrison, Daniel James Sargent. Data curation: R. Jordan Price, Jahn Davik, Matteo Buti, Muath Alsheikh. Formal analysis: R. Jordan Price, Jahn Davik, Matteo Buti, Nada Sˇurbanovski. Investigation: R. Jordan Price, Jahn Davik, Helen J. Bates, Samantha Lynn, Charlotte F. Nel- list, Matteo Buti, Nada Sˇurbanovski, Daniel James Sargent. Methodology: R. Jordan Price, Helen J. Bates, Charlotte F. Nellist, Matteo Buti, Dag Røen, Nada Sˇurbanovski, Daniel James Sargent. Project administration: R. Jordan Price, Daniel James Sargent. Resources: R. Jordan Price, Matteo Buti, Muath Alsheikh, Richard J. Harrison. Software: R. Jordan Price. Supervision: R. Jordan Price, Helen J. Bates, Richard J. Harrison, Daniel James Sargent. Validation: R. Jordan Price, Felicidad Fernande´z Fernande´z, Matteo Buti, Muath Alsheikh. Visualization: Muath Alsheikh. Writing – original draft: R. Jordan Price, Jahn Davik, Felicidad Fernande´z Fernande´z, Matteo Buti, Nada Sˇurbanovski, Daniel James Sargent. Writing – review & editing: Charlotte F. Nellist, Dag Røen. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 14 / 17 PLOS ONERed raspberry ONT sequence assembly References 1. Graham J, Brennan R. Raspberry: breeding, challenges and advances. 2018. 2. Ward JA, Ponnala L, Weber CA. Strategies for Transcriptome Analysis in Nonmodel Plants. Am J Bot. 2012; 99(2):267–76. https://doi.org/10.3732/ajb.1100334 PMID: 22301897 3. Foster TM, Bassil NV, Dossett M, Leigh Worthington M, Graham J. Genetic and genomic resources for Rubus breeding: a roadmap for the future. Horticulture Research. 2019;6. https://doi.org/10.1038/ s41438-019-0199-2 PMID: 31645970 4. Ward JA, Price JC, Clement M, Schatz M, Weber CA, Swanson JD, et al. Progress on the development of a genome sequence for red raspberry Rubus idaeus. In: 6th Rosaceous Genomics Conference (RGC6), Mezzocorona (TN), 30th September-4th October 2012: 56. 2012. 5. Hackett CA, Milne L, Smith K, Hedley P, Morris J, Simpson CG, et al. Enhancement of Glen Moy x Latham raspberry linkage map using GbS to further understand control of developmental processes leading to fruit ripening. BMC Genet. 2018; 19(1):59. https://doi.org/10.1186/s12863-018-0666-z PMID: 30111279 6. Shulaev V, Sargent DJ, Crowhurst RN, Mockler TC, Folkerts O, Delcher AL, et al. The genome of wood- land strawberry (Fragaria vesca). Nat Genet2011. p. 109–16. 7. Edger PP, Poorten TJ, VanBuren R, Hardigan MA, Colle M, McKain MR, et al. Origin and evolution of the octoploid strawberry genome. Nat Genet. 2019; 51(3):541–7. https://doi.org/10.1038/s41588-019- 0356-4 PMID: 30804557 8. Buti M, Moretto M, Barghini E, Mascagni F, Natali L, Brilli M, et al. The genome sequence and transcrip- tome of Potentilla micrantha and their comparison to Fragaria vesca (the woodland strawberry). Giga- Science. 2018; 7(4):giy010. https://doi.org/10.1093/gigascience/giy010 PMID: 29659812 9. VanBuren R, Bryant D, Bushakra JM, Vining KJ, Edger PP, Rowley ER, et al. The genome of black raspberry (Rubus occidentalis). The Plant Journal. 2016; 87(6):535–47. https://doi.org/10.1111/tpj. 13215 PMID: 27228578 10. Wight H, Zhou J, Li M, Hannenhalli S, Mount SM, Liu Z. Draft genome assembly and annotation of red raspberry Rubus idaeus. BioRxiv. 2019:546135. 11. Davik J, Røen D, Lysøe E, Buti M, Rossman S, Alsheikh M, et al. A chromosome-level genome sequence assembly of the red raspberry (Rubus idaeus L.). Plos one. 2022; 17(3):e0265096. https:// doi.org/10.1371/journal.pone.0265096 PMID: 35294470 12. Keep E, Parker JE, Knight VH. ’Autumn Bliss’, a new early autumn-fruiting raspberry. 1984. 13. Schalamun M, Kainer D, Beavan E, Nagar R, Eccles D, Rathjen JP, et al. A comprehensive toolkit to enable MinION sequencing in any laboratory. bioRxiv. 2018:289579. 14. Mapleson D, Garcia Accinelli G, Kettleborough G, Wright J, Clavijo BJ. KAT: a K-mer analysis toolkit to quality control NGS datasets and genome assemblies. Bioinformatics. 2017; 33(4):574–6. https://doi. org/10.1093/bioinformatics/btw663 PMID: 27797770 15. Vurture GW, Sedlazeck FJ, Nattestad M, Underwood CJ, Fang H, Gurtowski J, et al. GenomeScope: fast reference-free genome profiling from short reads. Bioinformatics. 2017; 33(14):2202–4. https://doi. org/10.1093/bioinformatics/btx153 PMID: 28369201 16. De Coster W D’hert S, Schultz DT, Cruts M, Van Broeckhoven C. NanoPack: visualizing and processing long-read sequencing data. Bioinformatics. 2018; 34(15):2666–9. 17. Chen Y, Nie F, Xie S-Q, Zheng Y-F, Dai Q, Bray T, et al. Efficient assembly of nanopore reads via highly accurate and intact error correction. Nature Communications. 2021; 12(1):1–10. 18. Guan D, McCarthy SA, Wood J, Howe K, Wang Y, Durbin R. Identifying and removing haplotypic dupli- cation in primary genome assemblies. Bioinformatics. 2020; 36(9):2896–8. https://doi.org/10.1093/ bioinformatics/btaa025 PMID: 31971576 19. Li H. Minimap2: pairwise alignment for nucleotide sequences. Bioinformatics. 2018; 34(18):3094–100. https://doi.org/10.1093/bioinformatics/bty191 PMID: 29750242 20. Vaser R, Sović I, Nagarajan N, Sˇ ikić M. Fast and accurate de novo genome assembly from long uncor- rected reads. Genome Res. 2017; 27(5):737–46. https://doi.org/10.1101/gr.214270.116 PMID: 28100585 21. Bolger AM, Lohse M, Usadel B. Trimmomatic: A flexible trimmer for Illumina Sequencing Data. Bioinfor- matics2014. p. 1–7. 22. 23. Langmead B, Salzberg SL. Fast gapped-read alignment with Bowtie 2. Nat Methods. 2012; 9(4):357–9. https://doi.org/10.1038/nmeth.1923 PMID: 22388286 Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, et al. The Sequence Alignment/Map format and SAMtools. Bioinformatics. 2009; 25(16):2078–9. https://doi.org/10.1093/bioinformatics/btp352 PMID: 19505943 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 15 / 17 PLOS ONERed raspberry ONT sequence assembly 24. Walker BJ, Abeel T, Shea T, Priest M, Abouelliel A, Sakthikumar S, et al. Pilon: an integrated tool for comprehensive microbial variant detection and genome assembly improvement. PloS one. 2014; 9(11): e112963. https://doi.org/10.1371/journal.pone.0112963 PMID: 25409509 25. Alonge M, Lebeigle L, Kirsche M, Aganezov S, Wang X, Lippman ZB, et al. Automated assembly scaf- folding elevates a new tomato system for high-throughput genome editing. BioRxiv. 2021. 26. Simão FA, Waterhouse RM, Ioannidis P, Kriventseva EV, Zdobnov EM. BUSCO: assessing genome assembly and annotation completeness with single-copy orthologs. Bioinformatics. 2015; 31(19):3210– 2. https://doi.org/10.1093/bioinformatics/btv351 PMID: 26059717 27. Girgis HZ. Red: an intelligent, rapid, accurate tool for detecting repeats de-novo on the genomic scale. BMC Bioinformatics. 2015; 16(1):1–19. https://doi.org/10.1186/s12859-015-0654-5 PMID: 26206263 28. Kim D, Paggi JM, Park C, Bennett C, Salzberg SL. Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype. Nat Biotechnol. 2019; 37(8):907–15. https://doi.org/10.1038/ s41587-019-0201-4 PMID: 31375807 29. Brůna T, Hoff KJ, Lomsadze A, Stanke M, Borodovsky M. BRAKER2: automatic eukaryotic genome annotation with GeneMark-EP+ and AUGUSTUS supported by a protein database. NAR genomics and bioinformatics. 2021; 3(1):lqaa108. https://doi.org/10.1093/nargab/lqaa108 PMID: 33575650 30. Benson G. Tandem repeats finder: a program to analyze DNA sequences. Nucleic Acids Res. 1999; 27 (2):573–80. https://doi.org/10.1093/nar/27.2.573 PMID: 9862982 31. Camacho C, Coulouris G, Avagyan V, Ma N, Papadopoulos J, Bealer K, et al. BLAST+: architecture and applications. BMC Bioinformatics. 2009; 10(1):1–9. https://doi.org/10.1186/1471-2105-10-421 PMID: 20003500 32. Afgan E, Baker D, Batut B, Van Den Beek M, Bouvier D, Čech M, et al. The Galaxy platform for accessi- ble, reproducible and collaborative biomedical analyses: 2018 update. Nucleic Acids Res. 2018; 46 (W1):W537–W44. https://doi.org/10.1093/nar/gky379 PMID: 29790989 33. Zdobnov EM, Apweiler R. InterProScan–an integration platform for the signature-recognition methods in InterPro. Bioinformatics. 2001; 17(9):847–8. https://doi.org/10.1093/bioinformatics/17.9.847 PMID: 11590104 34. Kanehisa M, Sato Y, Morishima K. BlastKOALA and GhostKOALA: KEGG tools for functional charac- terization of genome and metagenome sequences. J Mol Biol. 2016; 428(4):726–31. https://doi.org/10. 1016/j.jmb.2015.11.006 PMID: 26585406 35. Huerta-Cepas J, Forslund K, Coelho LP, Szklarczyk D, Jensen LJ, Von Mering C, et al. Fast genome- wide functional annotation through orthology assignment by eggNOG-mapper. Mol Biol Evol. 2017; 34 (8):2115–22. https://doi.org/10.1093/molbev/msx148 PMID: 28460117 36. Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, et al. Cytoscape: a software environ- ment for integrated models of biomolecular interaction networks. Genome Res. 2003; 13(11):2498– 504. https://doi.org/10.1101/gr.1239303 PMID: 14597658 37. Maere S, Heymans K, Kuiper M. BiNGO: a Cytoscape plugin to assess overrepresentation of gene ontology categories in biological networks. Bioinformatics. 2005; 21(16):3448–9. https://doi.org/10. 1093/bioinformatics/bti551 PMID: 15972284 38. Cabanettes F, Klopp C. D-GENIES: dot plot large genomes in an interactive, efficient and simple way. PeerJ. 2018; 6:e4958. https://doi.org/10.7717/peerj.4958 PMID: 29888139 39. Emms DM, Kelly S. OrthoFinder: phylogenetic orthology inference for comparative genomics. Genome biology. 2019; 20(1):1–14. 40. Wang L, Lei T, Han G, Yue J, Zhang X, Yang Q, et al. The chromosome-scale reference genome of Rubus chingii Hu provides insight into the biosynthetic pathway of hydrolyzable tannins. The Plant Jour- nal. 2021; 107(5):1466–77. https://doi.org/10.1111/tpj.15394 PMID: 34174125 41. VanBuren R, Wai CM, Colle M, Wang J, Sullivan S, Bushakra JM, et al. A near complete, chromosome- scale assembly of the black raspberry (Rubus occidentalis) genome. Gigascience. 2018; 7(8):giy094. https://doi.org/10.1093/gigascience/giy094 PMID: 30107523 42. Li Y, Pi M, Gao Q, Liu Z, Kang C. Updated annotation of the wild strawberry Fragaria vesca V4 genome. Horticulture research. 2019;6. 43. Raymond O, Gouzy J, Just J, Badouin H, Verdenaud M, Lemainque A, et al. The Rosa genome pro- vides new insights into the domestication of modern roses. Nat Genet. 2018; 50(6):772–7. https://doi. org/10.1038/s41588-018-0110-3 PMID: 29713014 44. Jung S, Lee T, Cheng C-H, Buble K, Zheng P, Yu J, et al. 15 years of GDR: New data and functionality in the Genome Database for Rosaceae. Nucleic Acids Res. 2019; 47(D1):D1137–D45. https://doi.org/ 10.1093/nar/gky1000 PMID: 30357347 45. Weisman CM, Murray AW, Eddy SR. Mixing genome annotation methods in a comparative analysis inflates the apparent number of lineage-specific genes. Curr Biol. 2022. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 16 / 17 PLOS ONERed raspberry ONT sequence assembly 46. Jiao W-B, Schneeberger K. The impact of third generation genomic technologies on plant genome assembly. Curr Opin Plant Biol. 2017; 36:64–70. https://doi.org/10.1016/j.pbi.2017.02.002 PMID: 28231512 47. Dale A, Moore PP, McNicol RJ, Sjulin TM, Burmistrov LA. Genetic diversity of red raspberry varieties throughout the-world. J Am Soc Hort Sci. 1993; 118(1):119–29. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285756 May 16, 2023 17 / 17 PLOS ONE
10.1371_journal.pone.0281837	RESEARCH ARTICLE Parliamentary roll-call voting as a complex dynamical system: The case of Chile Diego Morales-BaderID 4 GarridoID 1,2, Ramo´ n D. CastilloID 1, Ralf F. A. Cox3, Carlos Ascencio- 1 Centro de Investigacio´ n en Ciencias Cognitivas, Facultad de Psicologı´a, Universidad de Talca, Talca, Chile, 2 Facultad de Ingenierı´a y Ciencias, Universidad Adolfo Iba´ ñez, Santiago, Chile, 3 Department of Developmental Psychology, Faculty of Behavioral and Social Sciences, Heymans Institute for Psychological Research, University of Groningen, Groningen, Netherlands, 4 Escuela de Psicologı´a, Universidad Cato´ lica Silva Henrı´quez, Santiago, Chile a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 racastillo@utalca.cl Abstract OPEN ACCESS Citation: Morales-Bader D, Castillo RD, Cox RFA, Ascencio-Garrido C (2023) Parliamentary roll-call voting as a complex dynamical system: The case of Chile. PLoS ONE 18(4): e0281837. https://doi.org/ 10.1371/journal.pone.0281837 Editor: Dan Braha, University of Massachusetts, UNITED STATES Received: August 4, 2022 Accepted: February 1, 2023 Published: April 26, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0281837 Copyright: © 2023 Morales-Bader et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The data underlying the results presented in the study are available from http://opendata.camara.cl https://github.com/ dimoralesb/rollcall_votes_chile A method is proposed to study the temporal variability of legislative roll-call votes in a parlia- ment from the perspective of complex dynamical systems. We studied the Chilean Chamber of Deputies’ by analyzing the agreement ratio and the voting outcome of each vote over the last 19 years with a Recurrence Quantification Analysis and an entropy analysis (Sample Entropy). Two significant changes in the temporal variability were found: one in 2014, where the voting outcome became more recurrent and with less entropy, and another in 2018, where the agreement ratio became less recurrent and with higher entropy. These changes may be directly related to major changes in the Chilean electoral system and the composi- tion of the Chamber of Deputies, given that these changes occurred just after the first parlia- mentary elections with non-compulsory voting (2013 elections) and the first elections with a proportional system in conjunction with an increase in the number of deputies (2017 elec- tions) were held. Introduction It has been proposed that political systems and their behavior are complex systems, and there- fore it is important to study them from this perspective [1–4]. Although there is no formal defi- nition of what a complex system is, we may be referring to systems that have several components interacting with each other, that are difficult to predict and model, and that are non-linear, i.e., they do not always respond in the same way to the exact change or with the same intensity, among other characteristics (e.g., emergence). In these systems, sometimes small changes can generate significant changes at a general level, or large changes can some- times generate no change at all or more minor changes than expected. Legislative roll-call voting records are frequently used to study a political organization’s behavior. These are “yes” or “nay” votes, and depending on the country, there may be absten- tion votes, as is the case of the Chilean parliament. Legislative roll-call analysis has a long his- tory in political science (i.e., [5]). Usually, this kind of data is used to study the polarization of PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 1 / 20 PLOS ONEFunding: RDC and RFAC are funded by the Project for Promotion of International Linkages for Regional Institutions 2022 (FOVI210047), Agencia Nacional de Investigacio´n y Desarrollo de Chile (ANID), and by the Programa de Investigacio´n Asociativa (PIA) en Ciencias Cognitivas (RU-158- 2019), Universidad de Talca; DMB is funded by the Doctorate Scholarship FONDECYT-ANID grant # 21220612. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Roll-call voting as a complex dynamical system political parties [6], their discipline to the party [7], detecting shifts in foreign policy orienta- tions [8, 9], and the influence of controversial votes on elections [10], among other topics. In contrast to these approaches, we propose to study the behavior of legislative roll-call voting variability over time from the perspective of complex systems. Although this approach does not provide us with interpretations as direct as the studies mentioned above, it allows us to visualize changes that may not be detectable with traditional statistical analysis methods. One could argue that the outcome of a legislative roll-call vote is deterministic and, in many cases, predictable because agreements depend on pre-established alliances, which political party has a majority, and what issue is being voted on, among other factors. We propose that this determinism vanishes if we analyze roll-call votes over a sufficiently long time. By adding temporal dynamics to the analysis, we can find various factors of variability that are difficult to predict, such as emerging changes in the political agenda in response to a crisis, the emergence of controversial issues, changes in political alliances, changes in congressional members due to elections, increase or decrease in the number of parliamentarians, citizens’ demands, foreign policy, among several other factors. From this dynamic perspective, we can understand that systems are constantly changing. This constant change makes it challenging to attribute each variation to a particular situation, mainly due to non-linear changes not directly correlated in time. However, our idea is that, despite local variability, we could analyze the global stability of roll-call votes over time. For this, we will take two variables, the result of each vote (approved or rejected) and the propor- tion of agreement among parliamentarians in each vote (this variable will be explained in the next section). Some studies have explored the dynamics of legislative behavior using complex network and social network analysis [11–14]. Given that parliamentarians change in each legislative term, these analyses provide descriptions of local changes. Our proposal seeks to evaluate behavior globally and detect possible significant changes or otherwise to evaluate the stability of the system. If we treat legislative votes as a time series, where each vote is a time point, we can use an analysis method called Recurrence Quantification Analysis (RQA). RQA is a non-linear method used to discover tenuous correlations and repetitive patterns in a time series [15]. Unlike other analyses, it does not require additional treatment or assumptions about data dis- tribution [16–18]. Recurrence is a fundamental property of complex dynamical systems. It is defined as the ability of a system to return to the proximity of the initial point in phase space. RQA enables us to capture the dynamic organization of such systems based on the amount and patterns of such recurrences in phase space. This technique has been used in various domains to characterize temporal patterns of human and non-human behavior [15, 19–21]. However, despite its use in various disciplines, it has been scarcely used in analyzing political variables. For example, one study found that online social cohesion during the Arab Spring in Syria was related to the frequency of protests [22]. For this, they used a variant of RQA, the Cross Recurrence Quantification Analysis, which analyzes the recurrences of two time series instead of analyzing a time series with itself. In an economic study applying cross-RQA, the non-linear relationship between output and unemployment was studied, as well as the tempo- ral dependence and the existence of state changes. Also, was detected the possibility of predict- ing transition phases that coincided with periods in which the economy suffered recessions [23]. Both studies show that the RQA could be useful for detecting patterns and changes in a time series that may be difficult to analyze with traditional statistical methods. An additional useful metric of systems dynamics analysis is entropy. Entropy can be defined as the amount of information in a time series or signal. The more deterministic a signal is, it will contain less information. Sample Entropy (SE) is a typical analysis to study the PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 2 / 20 PLOS ONERoll-call voting as a complex dynamical system randomness or stability of a signal [24]. SE can discriminate stochastic processes and noisy deterministic datasets [25] and is nearly unaffected by low-level noise [26]. SE estimates the randomness of a time series without any previous knowledge about the source generating the data [27]. It measures how much a given data point depends on the values of past data points. Lower values indicate more self-similarity of the time series (approximately similar to a part of itself), while higher values indicate a less predictable series. An example of entropy in political analysis is the study by Marmani et al. [28], who used entropy and the Gini index to analyze municipal and parliamentary voting in Italy. They found high entropy levels but with varia- tions in some election periods. In this study, we considered the roll call votes of the Chilean Chamber of Deputies. We argue that non-linear time series analysis techniques, such as those described above, could pro- vide new insights into the voting dynamics and the system’s behavior [29, 30]. The Chilean political system is a multiparty system where the president is the head of state and head of government. Legislative power is exercised by two chambers of the national con- gress (Chamber of Deputies and Senate) and by the republic’s president. The functions of the Chamber of Deputies are creating and approving laws, which it performs jointly with the Sen- ate. It can supervise the acts of the government. It is currently composed of 155 members, directly elected from 28 electoral districts nationwide for four years. Deputies can express their vote in three ways; voting for, against, or abstaining. There are no obstruction mechanisms, and all absences must be justified. However, a mechanism known as "pairing" allows deputies not to participate in votes. Pairing is an agreement between two deputies from different benches or committees, agreeing not to participate in any vote if one is absent for a certain period. Each legislative term lasts four years, and all members of the Chamber of Deputies are renewed through elections usually held in conjunction with the presidential elections. Depu- ties can be reelected for a maximum of two terms. The board of directors of the Chamber of Deputies is composed of a President, a First Vice President, and a Second Vice President. An absolute majority of the deputies elect the board of directors by secret ballot. The President and Vice-Presidents of the Chamber can be reelected. The board of directors serves until the end of the legislative period; however, there are usually alternation agreements between the political parties to elect and occupy the chamber’s presi- dency for short periods. For example, there may have been four chamber presidents, each occupying the seat for one year in one legislative term. All members of the chamber’s board, including the president, can vote in the sessions. The functions of the president include presiding over the sessions and directing the debates, taking care of the chamber’s rules of procedure, opening, suspending, and adjourning sessions, declaring the inadmissibility of projects and indications following the Organic Constitutional Law of the National Congress, among other functions. In the context of the Chilean presidential regime, where the figure of the president has broad powers and little external control, the effective power possessed by both chambers of congress in terms of determining the agenda and the political life of the country is lower com- pared to the rest of Latin America [31]. Nevertheless, the role of the chamber is fundamental to understanding the development and projections of Chilean politics, where legislative pro- ductivity has been progressively increasing [32]. For example, productivity increases consis- tently in the first half of the presidential term, given the thrust of the incoming government to generate reforms and laws associated with its government plan, which gradually decreases over the last two years of administration [33]. Since 1990, the election system of parliamentarians was a binomial system with compulsory voting based on lists of candidates, which tended to overrepresent the second majority. This PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 3 / 20 PLOS ONERoll-call voting as a complex dynamical system electoral system implied a balanced composition between two major political forces, leaving out the rest of the political movements in a system that could be called forced bipartisanship, making the political parties have as main competition their own allies in the list presented for the elections [34]. A change to the electoral system was approved in 2012 when voting became non-compul- sory (the first election with this change was in 2013). Then in 2015, the binomial system was replaced by a proportional election system following the D’Hont model. In addition, the dis- tricts and their sizes were reorganized, increasing the number of representatives from 120 to 155. This change was implemented for the first time in the elections of 2017. This new electoral system generates greater representativeness in the overall composition of the chamber, but with the cost of greater difficulty for the executive power to achieve stable governance over time [35]. All the characteristics of the Chilean political system described above could set the condi- tions for complex behavior. In this context, the goal of this study is to explore whether some non-linear analysis methods can detect significant changes over time in the legislative work of the Chamber of Deputies of Chile with some complexity measures (recurrence and determin- ism from the RQA, and entropy from Sample Entropy). We will use the roll-call votes recorded from 2002 to the end of 2021. More specifically, generating two variables from the roll-call votes data; the agreement ratio (number of votes of the majority option divided by the total number of votes) and the voting outcome of each vote (approved or rejected). Methods Dataset A total of 19566 legislative votes of the Chilean Chamber of Deputies from 2002-03-19 to 2021-12-22 were obtained (openly available at http://opendata.camara.cl). The dataset initially had 19597 votes, but 31 were omitted due to record errors. By reading the session logs, other votes with recording errors were manually corrected when possible (n = 36). The data was downloaded and processed with a custom Jupyter Notebook, which can be downloaded at the following link: https://github.com/dimoralesb/rollcall_votes_chile. The data contains the number of deputies who voted in favor, against, or abstained, the result of the vote, the type of vote (bill, agreement, resolution, or others), the quorum required (simple quorum, 2/3, 3/5, among others), and the date of the vote. From the data, we calculated the agreement ratio of each vote as a measure of the variability of the agreements. The agreement ratio is obtained by dividing the number of votes of the majority option (in favor or against) by the total number of votes. The minimum agreement ratio was 0.30 (more abstentions than "Yes" or "No" votes), and the maximum was 1 (mean = 0.83, SD = 0.19). The second variable, the voting outcome, is obtained directly from the data. Initially, the data includes some voting results labeled as a tie (n = 22). Since ties imply that the outcome was a rejection, these records were re-labeled as rejected. Consequently, votes labeled as "no quorum” (n = 371) were also classified as rejected according to the rules of the Chamber of Deputies. Then, a numerical recode was made, where value “1” was assigned to approved votes and value “2” to rejected votes. Data analysis First, descriptive analyses were performed on the agreement ratio, voting outcome, and total votes per year. PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 4 / 20 PLOS ONERoll-call voting as a complex dynamical system For the non-linear data analyses, the data were divided into each legislative period, per- forming an RQA and SE for each. Then, for a second analysis, the entire data series was divided into (partly overlapping) windows of 256 data points with steps of 10 points (i.e., win- dow 1: votes 1 to 257, window 2: votes 11 to 267, and so on). In each window, an RQA and a SE were performed. These fixed-size windows include different time ranges since a variable amount of time may elapse in those 256 votes. For example, one window may occur in three months, while another may represent four or five months of data. The purpose of dividing the data into smaller, partially overlapping windows is to observe variations in the variables studied in greater detail. To compare the results without the effect of the time structure, the order of the original series was randomized 30 times [15, 19]. Then, the non-linear analyses were performed for each randomization (RQA and SE). A mean and a confidence interval were calculated for each window and legislative period. We used an algorithm to detect change points in the windowed analyses of our measures to determine when the system significantly changed (change in mean). We aimed to detect the most critical change point in the system since, due to the nature of the data, it could be consid- ered that there are many change points in the system. The R-package "cpm" was used [48]. The Mann-Whitney test statistic with a p value < = 0.01 was selected for change point detection. The recurrence quantification analysis and the Sample Entropy are described below, along with their important parameters. Recurrence Quantification Analysis (RQA). RQA [20, 36–38] was performed for the agreement ratio and voting outcome. Since recurrence is a property of dynamic systems, RQA allows quantifying the number and duration of the recurrences in phase space. RQA is based on procedures for visualizing recurrence patterns in dynamic systems named recurrence plots (RP). In general terms, RP is a two-dimensional representation of a multidimensional phase space trajectory x dynamic properties of the temporal trajectories that a system can generate in its original multi- dimensional space [39]. RPs visualize the recurrence of a state at time i at a different time j in a two-dimensional squared matrix with ones and zeros (or black and white dots in a plot), where both axes are time axes (S1 Appendix). RP can be mathematically expressed as: !. RPs show the geometric and � Ri;j ¼ Y ε(cid:0) k xl ! (cid:0) xj ! k � ; xi 2 Rm; i; j ¼ 1; . . . ; N; ð1Þ Where N is the number of considered states of xi.ε is a threshold distance, k�k is a norm (usu- ally the Euclidean norm), and Θ (�) is the Heaviside function, which is 1 when xi � xj, and 0 otherwise. xi � xj means equality up to an error or distance threshold ε. In short, the matrix tells us when similar states of the underlying system occur. RQA can reconstruct the phase space of a multidimensional system based on Taken’s embedding theorem [40]. This theorem proposes that under some conditions, it is possible to roughly reconstruct the state-space of a dynamical system by delay-embedding only one of its time series as follows: � ! ¼ ui; uiþd; . . . ; uiþdðm(cid:0) 1Þ xi � ð2Þ Where ui is the time series, m is the embedding dimension, and d is the time delay. The delay parameter helps to retrieve the m-dimensions using the delayed embedding method. This parameter considers the properties of the sample of observations, finding points in the time PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 5 / 20 PLOS ONERoll-call voting as a complex dynamical system series that can be used to reconstruct the latent dimensions. The embedded dimensions (m) correspond to an estimate of the latent dimensions that would configure the observed system. Finally, the radius (ε) specifies an interval in which two values are considered equal. In general, the radius is not easy to determine in a continuous series, but it is suggested to set a value that allows counting between 1% to 5% of recurrences [41]. The number of recurrences is related to the established radius, so if it is high (greater freedom), the percentage of recurrence points will be greater. We used the Average Mutual Information (AMI) to estimate the delay to use. AMI helps to quantify the amount of knowledge gained about the value of x (t + d) when observing x (t). AMI creates a histogram of the data using bins. Let pi be the probability that the signal has a value inside the ith bin and let pij (d) be the probability that x (t) is in bin i and x (t + d) is in bin j. Then, AMI for delay d is defined as: AMIðdÞ ¼ X pijlog i;j ! pij pi∗pj ð3Þ Next, to estimate an adequate number of embedded dimensions, we used the False Nearest Neighbor (FNN) algorithm [42, 43]. The goal of this method is to determine the minimal suffi- cient embedding dimensions. If a series is not adequately embedded with its true dimensional- ity, there is a risk of classifying false recurrences. This method assumes that two points in a time series that are near to each other in the sufficient embedding dimension (m) should remain close as the dimension increases. If the reader is interested in how to calculate this iter- ative method, we suggest reviewing the references mentioned here [42, 43]. Conventionally, the first minimum of the Average Mutual Information (AMI) and the False Nearest Neighbor (FNN), or the point at which those functions level-off, are indicative of the optimal delay and embedding dimension [16, 44]. We used the R-package "tseriesChaos" [45], which contains an implementation of the AMI and the FNN methods to estimate the delay and the embedding dimensions for the RQA. To perform the RQA we used the python package "PyRQA" [46]. First, we estimated the agreement ratio’s RQA parameters (delay, embedding dimensions, and threshold). In order to identify the optimal delay for the agreement ratio, the median of the first minimum value of the AMI of each period and window was used. For the embedding dimensions, the median of the global minimum of the FNN was used. These functions indicated that a delay of 1 and an embedded dimension of 6 was optimal for analyzing legislative periods and time-varying windows. For selecting the threshold, a general guideline is that the percentage of recurrence points (REC) should remain low but not so small as to produce a floor effect with values of REC near or at 0.0 [17, 41]. The threshold value of ε = 0.17 was chosen to yield between 1% and 5% (4% average) recurrence points for the agreement ratio. For categorical variables like the voting outcome, a delay and embedding dimension of 1 and a small threshold of 0.0001 is often sufficient [19, 30]. Therefore, estimating these parame- ters with the methods indicated above is not necessary. Several measures can be obtained from the RQA. For this study, we will consider two; the recurrence ratio (RR) and the determinism (DET). RR indicates the probability that a specific state will recur, while DET relates to the predictability of the system [16]. Mathematically, RR is the density of recurrence points in a recurrence plot, while DET is the percentage of recur- rence points that form diagonal lines in the recurrence plot of minimal length ℓmin. PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 6 / 20 PLOS ONE Roll-call voting as a complex dynamical system Considering the Eq 2, REC and DET are estimated as follows: RR ¼ 1 N2 XN i;j¼1 R i; jð Þ DET ¼ PN PN ‘Pð‘Þ ‘¼‘min ‘¼1 ‘Pð‘Þ ð4Þ For DET, ℓmin was set to 6. This number corresponds to the median number of votes parlia- mentarians perform per day. Sample entropy (SE). SE is a complexity measure that quantifies the system fluctuations’ degree of regularity and unpredictability [24]. Mathematically is defined as the negative natu- ral logarithm of the conditional probability that two sequences similar of m points remain sim- ilar at the next point (excluding self-matches). Therefore, a lower value for the SE corresponds to a higher probability indicating more self-similarity. SE is computed as follows: from a vector XN, two sequences of m consecutive points Xm (i) and Xm (j) are selected to compute the maxi- mum distance and compared to tolerance γ for repeated sequences counting. For the sequence Xm (i) its count is defined as Bm i ðgÞ. The maximum distance is computed as follow: � � ð d XmðiÞ; XmðjÞ Þ ¼ max jxi þ k; xj þ kj � gðk 2 ½0; m (cid:0) 1�; g � 0Þ ð5Þ Where the tolerance γ equals to 0.2 * SD (XN) in our case. Bm (γ), is the average amount of ; � and � Bmþ1ðgÞ is the average of m + 1 consecutive points. Then, SE is com- � (cid:0) counts Bm puted as follows: i ðgÞ SEðN; m; gÞ ¼ (cid:0) ln � Bmþ1ðgÞ BmðgÞ � ð6Þ The Python package "nolds" [47], which implements this method, was used with the default parameters to compute the SE. Results Descriptive analysis Descriptive analyses indicated that votes usually have a high agreement ratio among parlia- mentarians (Fig 1, panel B) and a high approval rate (Fig 1, panel C). Besides, the votes per year (Fig 1, panel A) had increased in an approximately linear way (R2 = 0.80) from 2002 (n = 454) to 2021 (n = 1267), dropping almost consistently on election years except for 2013 (election years: 2005, 2009, 2013, 2017 and 2021). Next, the data was divided into legislative periods to observe their variation over time. We found that the percentage of approved votes was higher in the last two legislative periods (Table 1). The 2010–2014 term had the lowest percentage of approved votes (76.0%), whereas the 2014–2018 term had the highest (88.6%). Also, we found that the agreement ratio of approved and rejected votes was lower in the last two legislative periods (Table 1). In other words, more votes were approved in the last two terms than in the previous terms, but with a lower agreement ratio among parliamentarians. Nonlinear analysis We split the time series (agreement ratio and voting outcome) into legislative periods to study potential changes in the voting behavior and political system between these periods. We per- formed the RQA (taking the recurrence rate and determinism) and SE on these periods for both variables. Subsequently, we randomized the order of all points in the time series in each period 30 times (i.e., random shuffling) [15, 19]. We performed the same analyses to compare PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 7 / 20 PLOS ONERoll-call voting as a complex dynamical system Fig 1. Number of votes per year from 2002 to 2021 (Panel A), Histogram of agreement ratio from 2002 to 2021 (Panel B), and frequency of voting outcome from 2002 to 2021 (Panel C). https://doi.org/10.1371/journal.pone.0281837.g001 the resulting measures with those of the original time series. If the data were stochastic, these analyses should indicate a similar result with the original data series and randomly shuffled data. Given the nature of our data, there should be differences from the randomized series. The results of the RQA indicate that the recurrence and determinism of the agreement ratio were relatively stable in the first four legislative periods (Fig 2, Panel A and B). However, since 2018, a lower recurrence ratio and determinism were found, much closer to the PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 8 / 20 PLOS ONERoll-call voting as a complex dynamical system Table 1. Percentage of approved and rejected votes and average agreement ratio in each legislative period from 2002 to 2021. Legislative Period Approved Mean Agreement Ratio of Approved Bills Rejected Mean Agreement Ratio of Rejected Bills 2002–2006 (N = 2505) 2006–2010 (N = 2990) 2010–2014 (N = 4461) 2014–2018 (N = 4464) 2018–2021 (N = 4516) 80.60% 80.90% 76.00% 88.60% 83.10% https://doi.org/10.1371/journal.pone.0281837.t001 0.89 (SD = 0.16) 0.89 (SD = 0.16) 0.87 (SD = 0.17) 0.87 (SD = 0.15) 0.82 (SD = 0.18) 19.40% 19.10% 24.00% 11.40% 16.90% 0.69 (SD = 0.18) 0.66 (SD = 0.17) 0.73 (SD = 0.21) 0.63 (SD = 0.16) 0.57 (SD = 0.11) Fig 2. Analysis of the Agreement Ratio for each legislative period. RQA is used to obtain the Recurrence Rate (Panel A) and Determinism (Panel B). Sample Entropy is used to measure the entropy of the data (Panel C). For each metric, the result obtained with the original data series is shown, along with the average result of randomly shuffling the data for each period 30 times and performing the analyses on each shuffle. The blue dashed vertical line shows when the first parliamentary elections with non-compulsory voting took place. The red dashed vertical line shows the first parliamentary elections with a propositional election system and the increase in the number of deputies. https://doi.org/10.1371/journal.pone.0281837.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 9 / 20 PLOS ONERoll-call voting as a complex dynamical system randomization than in previous years. This variation would indicate a change in the system, being less predictable and less likely for the same state to be repeated than in previous periods. The SE shows that entropy has increased in each legislative period, with a more noticeable increase in the last period (Fig 2, Panel C). In other words, the unpredictability of the agree- ment ratio has been increasing, especially in the last legislative period. This coincides with the findings of the RQA, where the last period marks a different trend with respect to the previous periods. It is important to note that the last legislative period coincides with the change in the parliamentary electoral system and the increase in the number of deputies. On the other hand, the analysis of the voting outcomes (Fig 3, Panel A and B) showed a rel- atively stable recurrence and determinism in the first three periods, with a decrease from 2010–2014, before a notable increase in the 2014–2018 term. While the last term showed a lower recurrence and determinism than the previous one, it is still higher than the first three periods. These results would indicate a change to the system contrary to that observed in the agreement ratio. The system becomes less recurrent and predictable from one term (2010– 2014), then becomes a more deterministic system susceptible to visiting the same states in the last two periods. It is observed that when each period is randomized, the recurrence ratio remains the same. This is because recurrence quantifies the number of events. For categorical variables, the num- ber of events remains the same when randomizing the order of the data within the same period or window. This is not the case for the windowed analysis because we shuffled the order of the whole series. The SE follows the trend found by the RQA for the voting outcome. The unpredictability increased in 2010–2014 and decreased significantly in 2014–2018. Since 2018 the sample entropy has grown again but remained lower than the first three terms. These results indicate a more deterministic system in the last two terms (Fig 3, Panel C). This change in the system occurred during the period when for the first time, deputies were elected in non-compulsory voting elections. Next, to explore the changes and transitions in the system in more detail, we performed the RQA and SE by partially overlapping windows across the entire data. Again, the time series were shuffled 30 times, and the same analyses were performed. A change-point algorithm [48] was used to detect the most significant changes in the time series using the Mann-Whitney test statistic with a p-value < = 0.01. The RQA of the agreement ratio (Fig 4, Panel A and B) shows a decay in recurrence rate and determinism in the last legislative periods. For the recurrence ratio, a change point occurs at window 1453 (p < 0.01), and for determinism, the change occurs at window 1456 (p < 0.01), both shortly after the beginning of the 2018–2021 term. Visual inspection suggests that both indicators are closer to random levels in the last period and sometimes below, more frequently than in previous periods. The SE is consistent with the findings of the RQA. The algorithm indicated a change after the beginning of the 2018–2021 term at window 1462 (p < 0.01). It is also observed that the system’s unpredictability increases in the last period, surpassing the randomized series line for a moment (Fig 4, Panel C). The change points detected in each metric are close to each other. They also occur after the change in the electoral system from a binomial to a proportional one and after the increase in the number of deputies. For the voting outcome, the RQA showed a change point (p < 0.01) near the beginning of the 2014–2018 term (Fig 5, Panel A and B). After this point, the data series increases in recur- rence ratio and determinism. Visually we observe the decay of both indicators before the change point occurs, signaling the increase in the values of the series. While both series PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 10 / 20 PLOS ONERoll-call voting as a complex dynamical system Fig 3. Analysis of the voting outcome for each legislative period. RQA is used to obtain the Recurrence Rate (Panel A) and Determinism (Panel B). Sample Entropy is used to measure the entropy of the data (Panel C). For each metric, the result obtained with the original data series is shown, along with the average result of randomly shuffling the data for each period 30 times and performing the analyses on each shuffle. The blue dashed vertical line shows when the first parliamentary elections with non-compulsory voting took place. The red dashed vertical line shows the first parliamentary elections with a propositional election system and the increase in the number of deputies. https://doi.org/10.1371/journal.pone.0281837.g003 oscillate between the random lines in the first periods, the series remains below this line before the change point. After the change point, the series remained above the random line more fre- quently than in other legislative periods. Again, the SE is consistent with the RQA. The unpredictability increased slightly in 2010– 2014 and decreased significantly in 2014–2018 (Fig 3, Panel C). Since 2018 the sample entropy has grown again but remained lower than the first three terms. These results indicate a more deterministic system from 2014 to 2021. The windowed analysis shows this trend in greater detail (Fig 5, Panel C). First, the series fluctuates above and below the randomized line. For a moment, within the 2010–2014 term, it stays above the randomized line and decreases notably PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 11 / 20 PLOS ONERoll-call voting as a complex dynamical system Fig 4. Analysis of the agreement ratio by partially overlapping windows of 256 data points with steps of 10 points. RQA is used to obtain the Recurrence Rate (Panel A) and Determinism (Panel B) for each window. Sample Entropy is used to measure the entropy of the data in each window (Panel C). For each metric, the result obtained with the original data series is shown in green, along with the average result of randomly shuffling the data 30 times and performing the analyses on each shuffle (red horizontal line). In total, 1932 windows were analyzed. The blue vertical line shows when the first parliamentary elections with non-compulsory voting took place. The red vertical line shows the first parliamentary elections with a propositional election system and the increase in the number of deputies. The purple dashed vertical line indicates the change point detected with the R package algorithm "cpm" with p < = 0.01. https://doi.org/10.1371/journal.pone.0281837.g004 from window 990 (p < 0.01), corresponding to the beginning of the 2014–2018 term. Finally, SE increases slightly towards 2018–2021 but without resembling the levels of the first periods. As was the case with the results of the agreement ratio, in the result of the voting outcome, the change points also occurred around the same point. However, unlike what was found in the agreement ratio, here, the change point occurs after the first elections with non-compul- sory voting were held. However, we can see that before this, there seems to be a distinct vari- ability for a moment in all three metrics. This change point will be discussed in more detail in the next section. PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 12 / 20 PLOS ONERoll-call voting as a complex dynamical system Fig 5. Analysis of the Voting Outcome by partially overlapping windows of 256 data points with steps of 10 points. RQA is used to obtain the Recurrence Rate (Panel A) and Determinism (Panel B) for each window. Sample Entropy is used to measure the entropy of the data in each window (Panel C). For each metric, the result obtained with the original data series is shown in green, along with the average result of randomly shuffling the data 30 times and performing the analyses on each shuffle (red horizontal line). In total, 1932 windows were analyzed. The blue vertical line shows when the first parliamentary elections with non-compulsory voting took place. The red vertical line shows the first parliamentary elections with a propositional election system and the increase in the number of deputies. The purple dashed vertical line indicates the change point detected with the R package algorithm "cpm" with p < = 0.01. https://doi.org/10.1371/journal.pone.0281837.g005 In both Figs 4 and 5, we notice that the metrics constantly vary. This variability may be nat- ural and expected, given that the votes are constantly moving due to different factors. These analyses aim to detect when this variability significatively changes and when the system transi- tions to a new state. For this reason, the analyses are not focused on small local changes with small time scales (a few months in this context) but on changes that move the system’s struc- ture on larger time scales (years or legislative periods). Overall, we found a significant change in the agreement ratio at the beginning of the last legislative period. In this period, the temporal variability of the agreement ratio is less PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 13 / 20 PLOS ONERoll-call voting as a complex dynamical system recurrent and less stable than in previous periods. In addition, the average agreement ratio is the lowest in this period. This change could be related to an important change in the chamber of deputies since it was in that period when the number of parliamentarians increased, and the election system changed from a binomial to a proportional one. We did not find a significant change in the voting outcome in that period. However, we detected a significant change at the beginning of the 2014–2018 term. From that point, the variability of voting outcomes tends to be more predictable and stable compared to previous periods. In this period, the percentage of ballots approved reached its highest point (88.6%). This change seems to match with the first parliamentary elections with non-compulsory voting. However, in the next section, another hypothesis is explored. Discussion In this study, we analyzed 19 years of the temporal variability of the agreement ratio and the voting outcome of the Chilean Chamber of Deputies using non-linear analysis techniques. We argued that these techniques could be useful for detecting significant global changes in the temporal variability of the data. While voting is always in constant change due to various fac- tors, both internal (normal functioning of the system) and external (crises, protests, emerging issues, policy changes, among others), these techniques could help us to detect changes that move away from the traditional states. Indeed, we detected two significant changes, one related to the voting outcome and the other related to the agreement ratio. The first change detected indicates that from 2014 onwards, voting results become more stable and deterministic compared to previous periods. This change matches the first parlia- mentary votes with non-compulsory voting. In this period, the highest percentage of approved votes is reached. One of the reasons that justified the reform of automatic registration and voluntary voting was the aging of the electoral population, which originated mainly from the low registration of young people in the electoral registers. The reform reduced the age bias; however, in the first elections under this modality, the percentage of voter turnout increased by 38% compared to the 2009 elections. We found that in this period (2014–2018), the percentage of deputies that were government supporters (S2 Appendix) was higher than in other periods (59.2%), which would explain why the percentage of approved votes was higher, with the system becoming more recurrent and stable. However, even though recurrence and determinism decreased, and entropy increased in the next period, they are still above the previous periods even though the percentage of dep- uties supporting the government (S2 Appendix) returned to usual percentages (close to 50%). This would indicate that the change detected is not focused only on the 2014–2018 term. An evaluation of the 2022–2026 term could confirm whether this change was global or local. The next step would be to explore what was happening politically in that period. However, we have another hypothesis to explain this change from the complex systems framework. This change could represent a phase transition related to another variable. Phase transitions are a common feature of complex dynamical systems representing a (sudden) change in the global state of the system when a threshold or critical parameter value is reached [41, 49, 50]. We believe this critical parameter could be the number of votes cast by the deputies. In Fig 1, we saw that the number of votes per year increased almost linearly, only decreasing in election years. This observation coincides in part with Badillo et al. [33], who found that the parlia- ment’s productivity decreases gradually during the last two years of its term; however, they do not mention such a marked decrease in the election years we found in this study. PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 14 / 20 PLOS ONERoll-call voting as a complex dynamical system Coincidentally, before the change we detected occurred, the number of votes did not decrease in the election year but decreased the following year. It is clear that the number of votes per year cannot increase unlimitedly, nor does the time of the debate and voting periods increase to meet the demand; therefore, the system must adapt to these changes. This change may be a result of that adaptation, where the voting out- come had to become more stable to respond to the demands of the environment. This can be represented in legislative sessions with the approval or rejection of various items simulta- neously on which there is relative consensus without going into greater detail. It is important to mention that the bills were previously worked on by commissions formed by a group of deputies. They are in charge of preparing and discussing these documents with their political parties, so not all of them require further debate. Furthermore, this hypothesis could explain why, before this significant change occurred, an increase in entropy and a decrease in recurrence and determinism were detected in the win- dow analysis. This could indicate an early warning before the system transitions to a new phase [51]. While these methods, as we apply them here, do not allow us to establish causal relation- ships, they help us to know where to focus our attention, so this significant change detected may be the starting point for future research. The second significant change detected related to the agreement ratio occurred at the beginning of the last period analyzed (2018–2021). In this period, the temporal variability of the agreement ratio became less recurrent and less deterministic, with more entropy, i.e., it became less regular. This change occurred after the electoral system changed from a binomial to a proportional one [52]. This electoral change introduced new political forces into parliament, which might have been more difficult under the binomial system. By introducing new political forces, the dynamics of parliament are likely to change, making it more difficult to reach agreements as there are now more political blocks to negotiate. We cannot attribute a causal relationship to the change found in 2018 because several factors affect the parliament’s behavior. Nevertheless, introducing new political forces is a significant organizational change in the system, bound to affect its dynamics. In addition, the 2017 elections also saw an increase in deputies from 120 to 155. We believe this increase also benefited the entry of new political forces into congress. There is a difficult coexistence between presidential regimes and multi-party systems, which present fewer incentives for party discipline and loyalty, forcing the executive to negoti- ate permanently to achieve governability [53]. This is in line with what is observed in our results since it introduces an additional source of variability. However, this instability would be typical of the combination of presidential regimes such as Chile’s and disaggregated party systems, which should stabilize as a more moderate presidential system is established [35]. Our results could be evidence of how a change in the election system can produce significant changes in the functioning of a parliament. However, more evidence from other parliaments is needed to support this idea. It is important to note that in the 2018–2021 term, several changes and fluctuations in the national political environment occurred. In October 2019, a social outbreak and a political cri- sis, and in 2020 the coronavirus pandemic emerged. These events affected the priorities in the government plans and altered the legislative agenda of the Chamber of Deputies. Despite these changes, the system still exhibits recurrence patterns that distinguish it from a fully entropic system. A complementary analysis using a technique called Detrended Fluctuation Analysis (DFA) found that despite all these changes detected at large and small scales, the system does not lose its self-similarity [54, 55], i.e., the votes continue to maintain a certain structure at dif- ferent time scales (S3 Appendix). PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 15 / 20 PLOS ONERoll-call voting as a complex dynamical system In this study, we rely on the idea that the temporal variability of voting has features of com- plex systems, even though without this time scale, voting may be deterministic. To test this idea, we performed a series of complementary analyses following the method of Olthof et al. [29], which evaluates different characteristics of complex systems (Dependency on past values, long-range and non-stationary temporal correlations, regime shifts, and sensitive dependence on initial conditions). We found that our two variables meet the expected properties of com- plex systems (S4 Appendix). Moreover, in Fig 1, we see a high probability that the agreement ratio is over 90% (panel b) and a high probability that a vote will pass (panel c). Despite this, by including time as a variable, we found distinct temporal patterns even with data that, at first glance, have little variability. According to the findings of this study, we believe that these methods are useful to describe a political organization such as the parliament through the historical record of roll-call votes and, at the same time to detect relevant changes over time. This method could be replicated on voting data from any national or supranational parliament with a historical voting record. A comparative study of the dynamic characteristics of parliaments with different election systems and different numbers of political parties involved (two-party vs. multi-party systems) could lead to valuable insights into the organization and behavior of political systems. Additionally, studying the time gap between each vote in the legislative periods, similar to how interstimulus intervals (ISI) have been studied [56], could provide more details on how parliamentary ses- sions behave and evolve. For example, it would provide information on how chaotic or struc- tured legislative sessions are. Also, specific questions about the relationship between chaotic moments in roll-call voting behavior and particular events in national politics could be addressed. A limitation of this study is that, given its descriptive approach, it is impossible to establish direct relationships between national policy events and the changes found in this study. We can only make conjectures, given the association of the timing of these changes with the timing of the events. Several events and variables may co-occur with internal and external influencing factors. These factors influence in different ways depending on the strength of the external influences relative to the internal influences. Identifying and separating these factors is not trivial; however, efforts have been made by modeling electoral behavior with social influence factors [57]. The next step would be to study the changes detected here in more detail (for example, identify internal and external factors) or to explore local changes in periods of inter- est on a different time scale than the one considered in this study. Conclusion A method was proposed to study the variability of votes in a parliament over time from the perspective of complex systems. The result and the proportion of agreement of each vote of the last 19 years of the Chilean Chamber of Deputies were considered. Two significant changes in temporal variability were found that may be directly related to major changes in the Chilean electoral system and the composition of the Chamber of Deputies. We argue that these meth- ods help to detect changes that could be difficult to observe with traditional statistical methods. Supporting information S1 Appendix. Example of a Recurrence Plot (RP) for the 2006–2010 legislative period. (DOCX) PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 16 / 20 PLOS ONERoll-call voting as a complex dynamical system S2 Appendix. Composition of the chamber of deputies regarding what percentage is con- sidered opposition and what percentage supports the government in each period. (DOCX) S3 Appendix. Detrended Fluctuation Analysis (DFA). (DOCX) S4 Appendix. Analyses to study if our data exhibits memory, regime shifts, and sensitive dependence on initial conditions. (DOCX) Author Contributions Conceptualization: Diego Morales-Bader, Ramo´n D. Castillo, Ralf F. A. Cox. Data curation: Diego Morales-Bader. Formal analysis: Diego Morales-Bader, Ralf F. A. Cox. Funding acquisition: Ramo´n D. Castillo. Investigation: Diego Morales-Bader, Ramo´n D. Castillo, Ralf F. A. Cox, Carlos Ascencio-Garrido. Methodology: Diego Morales-Bader, Ramo´n D. Castillo, Ralf F. A. Cox. Project administration: Diego Morales-Bader. Software: Diego Morales-Bader, Ralf F. A. Cox. Supervision: Diego Morales-Bader, Ramo´n D. Castillo, Ralf F. A. Cox. Validation: Diego Morales-Bader, Ramo´n D. Castillo, Ralf F. A. Cox, Carlos Ascencio-Garrido. Visualization: Diego Morales-Bader, Ralf F. A. Cox. Writing – original draft: Diego Morales-Bader, Ramo´n D. Castillo, Ralf F. A. Cox, Carlos Ascencio-Garrido. Writing – review & editing: Diego Morales-Bader, Ramo´n D. Castillo, Ralf F. A. Cox, Carlos Ascencio-Garrido. References 1. Banerjee S, Erc¸etin ŞŞ, Tekin A, editors. Chaos Theory in Politics. Dordrecht: Springer Netherlands; 2014. https://doi.org/10.1007/978-94-017-8691-1. 2. Complexity Gunitsky S. and theories of change in international politics. Int Theory 2013; 5:35–63. https://doi.org/10.1017/S1752971913000110. 3. Saperstein AM. Alliance Building versus Independent Action: A Nonlinear Modeling Approach to Com- parative International Stability. J Confl Resolut 1992; 36:518–45. https://doi.org/10.1177/ 0022002792036003005. 4. Weisberg HF. Nonlinear models of electoral change: the implications of political time and chaos theory for the study of mass political behavior. Elect Stud 1998; 17:369–82. https://doi.org/10.1016/S0261- 3794(98)00033-X. 5. Lowell AL. The Influence of Party Upon Legislation in England and America. Annu Rep Am Hist Assoc 1901 1902:319–542. 6. McCarty NM, Poole KT, Rosenthal H. Polarized America: the dance of ideology and unequal riches. Cambridge, Mass: MIT Press; 2006. PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 17 / 20 PLOS ONERoll-call voting as a complex dynamical system 7. McCarty N, Poole KT, Rosenthal H. The Hunt for Party Discipline in Congress. Am Polit Sci Rev 2001; 95:673–87. https://doi.org/10.1017/S0003055401003069. 8. Bailey MA, Strezhnev A, Voeten E. Estimating Dynamic State Preferences from United Nations Voting Data. J Confl Resolut 2017; 61:430–56. https://doi.org/10.1177/0022002715595700. 9. Mattes M, Leeds BA, Carroll R. Leadership Turnover and Foreign Policy Change: Societal Interests, Domestic Institutions, and Voting in the United Nations. Int Stud Q 2015; 59:280–90. https://doi.org/10. 1111/isqu.12175. 10. Nyhan B, McGhee E, Sides J, Masket S, Greene S. One Vote Out of Step? The Effects of Salient Roll Call Votes in the 2010 Election. Am Polit Res 2012; 40:844–79. https://doi.org/10.1177/ 1532673X11433768. 11. Brito ACM, Silva FN, Amancio DR. A complex network approach to political analysis: Application to the Brazilian Chamber of Deputies. PLOS ONE 2020; 15:e0229928. https://doi.org/10.1371/journal.pone. 0229928. PMID: 32191720 12. Dal Maso C, Pompa G, Puliga M, Riotta G, Chessa A. Voting Behavior, Coalitions and Government Strength through a Complex Network Analysis. PLOS ONE 2014; 9:e116046. https://doi.org/10.1371/ journal.pone.0116046. PMID: 25549351 13. Puccio E, Pajala A, Piilo J, Tumminello M. Structure and evolution of a European Parliament via a net- work and correlation analysis. Phys Stat Mech Its Appl 2016; 462:167–85. https://doi.org/10.1016/j. physa.2016.06.062. 14. 15. Tam Cho WK, Fowler JH. Legislative Success in a Small World: Social Network Analysis and the Dynamics of Congressional Legislation. J Polit 2010; 72:124–35. https://doi.org/10.1017/ S002238160999051X. Lira-Palma D, Gonza´ lez-Rosales K, Castillo RD, Spencer R, Fresno A. Categorical Cross-Recurrence Quantification Analysis Applied to Communicative Interaction during Ainsworth’s Strange Situation. Complexity 2018; 2018:1–15. https://doi.org/10.1155/2018/4547029. 16. Marwan N, Carmen Romano M, Thiel M, Kurths J. Recurrence plots for the analysis of complex sys- tems. Phys Rep 2007; 438:237–329. https://doi.org/10.1016/j.physrep.2006.11.001. 17. Pellecchia GL, Shockley K. Application of recurrence quantification analysis: Influence of cognitive activity on postural fluctuations. In: Riley MA, Van Orden GC, editors. Tutor. Contemp. Nonlinear Meth- ods Behav. Sci., 2005, p. 95–141. 18. Zbilut JP, Thomasson N, Webber CL. Recurrence quantification analysis as a tool for nonlinear explora- tion of nonstationary cardiac signals. Med Eng Phys 2002; 24:53–60. https://doi.org/10.1016/S1350- 4533(01)00112-6. PMID: 11891140 19. Dale R, Spivey MJ. Unraveling the dyad: Using recurrence analysis to explore patterns of syntactic coordination between children and caregivers in conversation. Lang Learn 2006; 56:391–430. https:// doi.org/10.1111/j.1467-9922.2006.00372.x. 20. Marwan N, Wessel N, Meyerfeldt U, Schirdewan A, Kurths J. Recurrence-plot-based measures of com- plexity and their application to heart-rate-variability data. Phys Rev E 2002; 66:026702. https://doi.org/ 10.1103/PhysRevE.66.026702. PMID: 12241313 21. Castillo RD, Kloos H, Holden JG, Richardson MJ. Long-range correlations and patterns of recurrence in children and adults’ attention to hierarchical displays. Front Physiol 2015; 6. https://doi.org/10.3389/ fphys.2015.00138. PMID: 25999862 22. Chiovaro M, Windsor LC, Windsor A, Paxton A. Online social cohesion reflects real-world group action in Syria during the Arab Spring. PLOS ONE 2021; 16:e0254087. https://doi.org/10.1371/journal.pone. 0254087. PMID: 34270574 23. Caraiani P, Haven E. The Role of Recurrence Plots in Characterizing the Output-Unemployment Rela- tionship: An Analysis. PLOS ONE 2013; 8:e56767. https://doi.org/10.1371/journal.pone.0056767. PMID: 23460814 24. Richman JS, Moorman JR. Physiological time-series analysis using approximate entropy and sample entropy. Am J Physiol-Heart Circ Physiol 2000; 278:H2039–49. https://doi.org/10.1152/ajpheart.2000. 278.6.H2039. PMID: 10843903 25. Pincus SM. Assessing Serial Irregularity and Its Implications for Health. Ann N Y Acad Sci 2001; 954:245–67. https://doi.org/10.1111/j.1749-6632.2001.tb02755.x. PMID: 11797860 26. Zhang X-S, Roy RJ. Derived fuzzy knowledge model for estimating the depth of anesthesia. IEEE Trans Biomed Eng 2001; 48:312–23. https://doi.org/10.1109/10.914794. PMID: 11327499 27. Delgado-Bonal A, Marshak A. Approximate Entropy and Sample Entropy: A Comprehensive Tutorial. Entropy 2019; 21:541. https://doi.org/10.3390/e21060541. PMID: 33267255 28. Marmani S, Ficcadenti V, Kaur P, Dhesi G. Entropic Analysis of Votes Expressed in Italian Elections between 1948 and 2018. Entropy 2020; 22:523. https://doi.org/10.3390/e22050523. PMID: 33286295 PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 18 / 20 PLOS ONERoll-call voting as a complex dynamical system 29. Olthof M, Hasselman F, Lichtwarck-Aschoff A. Complexity in psychological self-ratings: implications for research and practice. BMC Med 2020; 18:1–16. https://doi.org/10.1186/s12916-020-01727-2. 30. Xu TL, de Barbaro K, Abney DH, Cox RFA. Finding Structure in Time: Visualizing and Analyzing Behav- ioral Time Series. Front Psychol 2020; 11:1457. https://doi.org/10.3389/fpsyg.2020.01457. PMID: 32793025 31. Cadena AE, Gayosso RS. Las facultades constitucionales del Ejecutivo en Ame´ rica Latina: entre la concentracio´ n y la dispersio´ n de poder. Estud Polı´ticos 2016; 37:111–41. https://doi.org/10.1016/j. espol.2016.02.005. 32. Visconti G. Cross-time behaviour of the chilean congress: Growth or stagnation of its influence. Rev Cienc Polı´tica Santiago 2011; 31:91–115. https://doi.org/10.4067/S0718-090X2011000100005. 33. Badillo V, Bianchini G, Navia P. La evolucio´n de la influencia del Congreso en Chile, 1990–2018. Una re´plica a Visconti. X Congr. Latinoam. Cienc. Polı´ticas, Asociacio´n Latinoamericana de Ciencias Polı´ti- cas (ALACIP), Asociacio´ n Mexicana de Ciencias Polı´ticas (AMECIP), Instituto Tecnolo´ gico de Estudios Superiores de Monterrey (ITESM); 2011. 34. Duval T, Pertuze´ C. Reforma al sistema binomial: hacia un sistema proporcional compensatorio y com- petitivo. Rev Enfoques Cienc Polı´tica Adm Pu´ blica 2006; 4:141–62. 35. Delgado Sotillos I. La formacio´ n de gobiernos en sistemas multipartidistas: la paradoja del caso español. Teorı´a Real Const 2020: 261. https://doi.org/10.5944/trc.45.2020.27119. 36. Trulla LL, Giuliani A, Zbilut JP, Webber CL. Recurrence quantification analysis of the logistic equation with transients. Phys Lett A 1996; 223:255–60. https://doi.org/10.1016/S0375-9601(96)00741-4. 37. Webber CL, Zbilut JP. Dynamical assessment of physiological systems and states using recurrence plot strategies. J Appl Physiol 1994; 76:965–73. https://doi.org/10.1152/jappl.1994.76.2.965. PMID: 8175612 38. 39. 40. Zbilut JP, Webber CL. Embeddings and delays as derived from quantification of recurrence plots. Phys Lett A 1992; 171:199–203. https://doi.org/10.1016/0375-9601(92)90426-M. Lo´ pez AM, Martı´nez C, Garcı´a A, Romero J. El gra´ fico de recurrencia en el ana´ lisis de series tempor- ales: el programa VRA. REMA 2001; 6:1–8. Takens F. Detecting strange attractors in turbulence. In: Rand D, Young L-S, editors. Dyn. Syst. Turbul. Warwick 1980, vol. 898, Berlin, Heidelberg: Springer Berlin Heidelberg; 1981, p. 366–81. https://doi. org/10.1007/BFb0091924. 41. Webber CL, Zbilut JP. Recurrence quantification analysis of nonlinear dynamical systems. In: Riley MA, Van Orden GC, editors. Tutor. Contemp. Nonlinear Methods Behav. Sci., 2005, p. 26–94. 42. Kennel MB, Brown R, Abarbanel HDI. Determining embedding dimension for phase-space reconstruc- tion using a geometrical construction. Phys Rev A 1992; 45:3403–11. https://doi.org/10.1103/ PhysRevA.45.3403. PMID: 9907388 43. Rhodes C, Morari M. The false nearest neighbors algorithm: An overview. Comput Chem Eng 1997; 21: S1149–54. https://doi.org/10.1016/S0098-1354(97)87657-0. 44. Kantz H, Schreiber T. Nonlinear time series analysis. Cambridge University Press; 1996. 45. Fabio Di Narzo A. tseriesChaos: Analysis of Nonlinear Time Series 2019. 46. Rawald T, Sips M, Marwan N. PyRQA—Conducting recurrence quantification analysis on very long time series efficiently. Comput Geosci 2017; 104:101–8. https://doi.org/10.1016/j.cageo.2016.11.016. 47. Scho¨ lzel Christopher. Nonlinear measures for dynamical systems 2019. https://doi.org/10.5281/ ZENODO.3814723. 48. Ross GJ. Parametric and nonparametric sequential change detection in R: The cpm package. J Stat Softw 2015; 66. https://doi.org/10.18637/jss.v066.i03. 49. Kiyono K, Struzik ZR, Aoyagi N, Togo F, Yamamoto Y. Phase Transition in a Healthy Human Heart Rate. Phys Rev Lett 2005; 95:058101. https://doi.org/10.1103/PhysRevLett.95.058101. PMID: 16090921 50. Opper M, Diederich S. Phase transition and 1/ f noise in a game dynamical model. Phys Rev Lett 1992; 69:1616–9. https://doi.org/10.1103/PhysRevLett.69.1616. 51. Scheffer M, Bascompte J, Brock WA, Brovkin V, Carpenter SR, Dakos V, et al. Early-warning signals for critical transitions. Nature 2009; 461:53–9. https://doi.org/10.1038/nature08227. PMID: 19727193 52. Bormann N-C, Golder M. Democratic Electoral Systems around the world, 1946–2011. Elect Stud 2013; 32:360–9. https://doi.org/10.1016/j.electstud.2013.01.005. 53. Mella-Polanco M. Chile’s Political Party System. In: Farazmand A, editor. Glob. Encycl. Public Adm. Public Policy Gov., Cham: Springer International Publishing; 2019. https://doi.org/10.1007/978-3-319- 31816-5_3170-1. PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 19 / 20 PLOS ONERoll-call voting as a complex dynamical system 54. Bassingthwaighte JB, Liebovitch LS, West BJ. Fractal Physiology. New York, NY: Springer New York; 1994. https://doi.org/10.1007/978-1-4614-7572-9. 55. Brown C, Liebovitch L. Fractal Analysis. 2455 Teller Road, Thousand Oaks California 91320 United States of America: SAGE Publications, Inc.; 2010. https://doi.org/10.4135/9781412993876. 56. Holden JG, Choi I, Amazeen PG, Van Orden G. Fractal 1/ƒ dynamics suggest entanglement of mea- surement and human performance. J Exp Psychol Hum Percept Perform 2011; 37:935–48. https://doi. org/10.1037/a0020991. 57. Braha D, de Aguiar MAM. Voting contagion: Modeling and analysis of a century of U.S. presidential elections. PLOS ONE 2017; 12:e0177970. https://doi.org/10.1371/journal.pone.0177970. PMID: 28542409 PLOS ONE \| https://doi.org/10.1371/journal.pone.0281837 April 26, 2023 20 / 20 PLOS ONE
10.1371_journal.pone.0285835	RESEARCH ARTICLE Validation of the family focused mental health practice questionnaire in measuring health and social care professionals’ family focused practice Anne GrantID Oliver Perra1 1, Susan Lagdon2, John DevaneyID 3, Gavin DavidsonID 4, Joe Duffy4, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 1 School of Nursing and Midwifery, Medical Biology Centre, Queen’s University Belfast, Belfast, Northern Ireland, United Kingdom, 2 School of Psychology, Ulster University, Coleraine, Northern Ireland, United Kingdom, 3 School of Social and Political Science, University of Edinburgh, Edinburgh, United Kingdom, 4 School of Social Sciences, Education and Social Work, Queen’s University Belfast, Belfast, Northern Ireland, United Kingdom OPEN ACCESS Citation: Grant A, Lagdon S, Devaney J, Davidson G, Duffy J, Perra O (2023) Validation of the family focused mental health practice questionnaire in measuring health and social care professionals’ family focused practice. PLoS ONE 18(5): e0285835. https://doi.org/10.1371/journal. pone.0285835 Editor: Muhammad Shahzad Aslam, Xiamen University - Malaysia Campus: Xiamen University - Malaysia, MALAYSIA Received: October 27, 2021 Accepted: April 25, 2023 Published: May 22, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0285835 Copyright: © 2023 Grant et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper. a.grant@qub.ac.uk Abstract Background Parental mental illness is a major public health issue and there is growing evidence that fam- ily focused practice can improve outcomes for parents and their families. However, few reli- able and valid instruments measure mental health and social care professionals’ family focused practice. Objectives To explore the psychometric properties of the Family Focused Mental Health Practice Ques- tionnaire in a population of health and social care professionals. Methods Health and Social Care Professionals (n = 836) in Northern Ireland completed an adapted version of the Family Focused Mental Health Practice Questionnaire. Exploratory factor analysis was used to test the structure of the underlying dimensions in the questionnaire. The results, and theoretical considerations, guided construction of a model that could explain variation in respondents’ items. This model was then validated using confirmatory factor analysis. Results Exploratory factor analysis revealed that solutions including 12 to 16 factors provided a good fit to the data and indicated underlying factors that could be meaningfully interpreted in line with existing literature. From these exploratory analyses, we derived a model that included 14 factors and tested this model with Confirmatory Factor Analysis. The results suggested 12 factors that summarized 46 items that were most optimal in reflecting family PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 1 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire Funding: AG received an award from the NI Health and Social Care Board to complete this study (hscboard.hscbi.net or volorg.monitoring©hscni. net). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. focused behaviours and professional and organizational factors. The 12 dimensions identi- fied were meaningful and consistent with substantive theories: furthermore, their inter-corre- lations were consistent with known professional and organizational processes known to promote or hinder family focused practice. Competing interests: The authors have declared that no competing interests exist. Conclusion This psychometric evaluation reveals that the scale provides a meaningful measure of pro- fessionals’ family focused practice within adult mental health and children’s services, and the factors that hinder and enable practice in this area. The findings, therefore, support the use of this measure to benchmark and further develop family focused practice in both adult mental health and children’s services. Introduction Internationally, it has been estimated that between 12 and 45% of adults receiving treatment from mental health services have children [1–3]. Reports from the United Kingdom (UK) sug- gest that 10% of mothers and 6% of fathers have mental health problems at any given time [4], with more current reports from the UK ‘Understanding Society’ survey [5] suggesting higher rates of 23.6% of mothers and 12.5% of fathers reporting symptoms of emotional distress. Northern Ireland (NI) is currently reported as having the highest levels of maternal mental ill- ness within the UK [6], with one in four children aged 0–16 years exposed to maternal mental illness at any one time, and an estimated 53% of children over 16 having a mother who has been diagnosed with a common (i.e. depression and anxiety) or severe (i.e. psychosis) mental illness at some time [6]. Most recently, the Youth Wellbeing Prevalence Survey [7] found that one in five (22%) parents or caregivers across NI reported a previous diagnosis of any mental health disorder. Parental mental illness (PMI) is an important global public health issue. The needs and issues for parents who have mental health problems, their children and their families are exten- sive and have been documented in numerous studies [8–14]. Bunting et al. found that parental mental health was one of the key factors shown to have a strong association with the develop- ment of mood and anxiety disorders among children and young people in NI. Children whose parents had current mental health problems (as measured by the General Health Question- naire (GHQ-12)) were twice as likely to have an anxiety or depressive disorder themselves [7]. Internationally, it is estimated that 25 to 50% of children who have a parent with a mental ill- ness will experience some psychological disorder during childhood or adolescence and 10– 14% of these children will be diagnosed with a psychotic disorder at some point in their lives [15]. There is also substantial evidence that parental mental health problems may contribute to child maltreatment [15–17]. Moreover, stress from assuming a parenting role may negatively impact parents’ wellbeing [18]. Having to juggle the demands of managing their own mental illness and additional responsibilities of managing their children’s problems, can further heighten the risk of parents relapsing [17]. If parents perceive that they are unable to cope with their parenting role it may have a profound impact on their mood, self-esteem and self-efficacy [19]. A family focused approach to service delivery by professionals can help parents, children and other family members to prevent and/or cope effectively with the difficulties associated PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 2 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire with PMI [20, 21]. Within this approach, professionals engage the service user within the con- text of their immediate connected family relationships and endeavour to meet the needs of both service users and family members (including children) [22]. Central to this is being clear about how the needs of individual family members and the needs of the whole family are con- sidered, especially when a parent’s mental health may be impacting on their child to such a sig- nificant extent that there are concerns about protecting the child. A family focused approach may also involve directly engaging service users’ children around issues related to PMI and promoting their capacity to understand and cope with it. Professionals may also indirectly sup- port children by keeping them in mind while caring for service users, and by referral to other specialist support services as required. Activities can be classified as more or less family focused on a continuum, with direct support of service users’ and their children (i.e. psychoeducation or family therapy) through to indirect support such as referral to other agencies. The types and intensity of activities and processes that professionals use to engage in family focused practice (FFP) are partly determined by the type of service in which professionals practice, their beliefs about the need for and importance of FFP, capacity to engage in it and how they think it should be operationalized [22]. In response to increasing recommendations by researchers, policy makers and professional organisations for health and social care professionals to engage in FFP [23–26], various coun- tries, including the UK, have implemented organisational initiatives to promote FFP. For example, the Health and Social Care Board in NI has endorsed the use of The Family Model (TFM) [17, 27] as a framework to embed FFP within services and provide in-service training to increase awareness and use of the model in practice [28]. Assessment documentation in Northern Irish adult mental health and children’s services has been refined in line with the domains of TFM [28], particularly domain one, where health professionals are prompted to identify children’s needs related to PMI. Similarly, in other parts of the UK, TFM is the approach advocated for use by the Social Care Institute for Excellence [29]. Theoretical foundation of the Family Focused Mental Health Practice Questionnaire The Family Focused Mental Health Practice Questionnaire (FFMHPQ) [30] was developed against the backdrop of increasing recommendations for FFP; the broad variation in practice with families when a parent has a mental illness; and so the need for an agreed minimum skill set for FFP. The questionnaire was originally designed to measure mental health professionals’ FFP working within adult services in Australia. Initially, Maybery et al developed the items for the FFMHPQ in parallel with a systematic review of the literature and with detailed input from the Victorian Families where a Parent has a Mental Illness (FaPMI) and the Co-coordinators and Vichamps project [31]. The review highlighted workplace policy and supports, worker skill and knowledge, and service user factors as central to FFP. These focal domains guided the initial generation of 100 items. These items were then subjected to rigorous review, rewrite and re- review, in Delphi focus groups with FaPMI coordinators. Subsequently, 14 subscales were developed, comprising a total of 45 items, which are scored on a seven-point Likert scale (ranging from strongly disagree = 1 to strongly agree = 7). Five subscales are said to measure professionals’ family focused clinical practices and activities such as providing parenting sup- port, referring family members to services and collaboration with other professionals to meet the needs of the family. The remaining subscales are said to measure organisational influences that can impact these activities such as workplace policies and support, and workload as well as issues related to service user engagement. Principal components analysis of the original scale in an Australian study [30], revealed a sixteen-factor solution, however, there was poor PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 3 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire internal consistency in three of the sixteen subscales. In addition, some of the remaining fac- tors consisted of two items, indicating they may have been unstable [32]. Since its inception, the FFMHPQ has been adapted for use within a number of professional contexts internationally, particularly as the components of FFP are recognised and utilised across many sectors responding to issues associated with PMI. For example [33], adapted the measure for the early educational sector. In their study of preschool teachers and childcare professionals in Australia [33], removed eight of the original sixteen factors based on the research team’s decision that they were not appropriate for teachers. Five of the remaining sub- scales possessed Cronbach’s alpha values below 0.5, so consequently items were removed to improve reliability. Furthermore, in recognition of the need to acknowledge different lan- guage, culture and policy differences across countries, the FFMHPQ has been adapted for use in Ireland with mental health nurses [8] and translated into other languages including in Thai- land [34] and in Norway [35, 36]. While none of these authors undertook psychometric evaluation, exploratory factor analy- sis (EFA) was undertaken as part of a study in NI with health visitors [37] and in Japan with mental health professionals [38]. While Leonard et al. [37] found a two-factor solution related to professional and organisational influences on FFP, Ueno et al. [38] found a 13-factor solu- tion. Family focused practice is characterized by various activities on a continuum [20] and organizational and professional factors that enable and hinder it [8, 39]. The type of service and sector will influence professionals’ understanding and interpretation of FFP and how and to what extent professionals use it in their practice [22]. Hence, the psychometric properties of the scale may vary in accordance with the particular profession and service context in which it is used depending on how FFP has been conceptualized, operationalized and embedded in practice. Maybery et al. [30] recommended that future research using the FFMHPQ should examine and attempt to replicate the psychometric properties of the measure as well as attempt to expand its component structure to include additional important items and factors. Ueno et al. [38] also recommended that further research be conducted to examine FFP within different services, sectors and professionals to determine if there are differences that may impact FFP. Based on previous adaption and validation studies of the FFMHPQ, the aim of this study is to test the psychometric properties and factorial structure of the scale in a population of health and social care (HSC) professionals across NI. Our objective was to explore if family focused practices are influenced by both the individual professional and their core overlapping prac- tices as well as the organizational structures which surround them. To our knowledge this is the first study that has evaluated the psychometric properties of the FFMHPQ when used within two distinct sectors, namely adult mental health and children’s services. Results from this psychometric evaluation of the FFMHPQ can be used to inform future practice bench- marks and organizational developments. Methods Design A cross sectional research design was implemented for the current study, with data collected using a survey approach. The survey consisted of three sections. Section one collected demo- graphic data, section two included items from the FFMHPQ which is designed to measure professionals’ FFP, and section three included items which aimed to capture, through a num- ber of open-ended questions, professionals’ experience of working with parents. Data pertain- ing to section two of the survey will be the focus of analysis within the current study. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 4 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire Measure The FFMHPQ was developed by Maybery et al. [30] within the Australian context and further refined by Grant [39] within the Irish context. Within the current study, professionals responded to 21 subscales (which included 66 items) using a seven-point Likert Scale (ranging from strongly disagree to strongly agree). Subscales were designed to measure different family focused activities and behaviours (e.g. assessing the impact of PMI on the child, providing sup- port to parents, carers and children), and factors that impact these (i.e. workplace support, pol- icy and procedures, inter professional practice and professional development). A low score in a particular subscale suggests a reduced family focus practice in this area, with a high score suggesting increased family focus [30]. Psychometric information of the subscales is detailed in the works of Maybery et al. [30]. The measure was reported as having excellent content and construct validity and good internal subscale reliability. As the FFMHPQ was devised for use in the Australian context, with a variety of professional disciplines (e.g. psychologists, psychiatric nurses, social workers), it required minor adaption and testing in the NI context particularly for health and social care (HSC) professionals prac- ticing within adult mental health and children’s services. Accordingly, the term ‘consumer par- ent’ was changed to ‘service user’ [8]. Further, in conjunction with developers of the original instrument, research team and advisory group, and in response to the emerging literature on PMI, FFP and organisational developments in FFP in NI, three additional subscales (contain- ing 10 additional items) were included within the current survey. These new subscales aimed to measure professionals’ understanding of The Family Model [17] and child protection proto- cols, and their interventions to reduce the impact of the parenting role on parental mental health. These new subscales are further detailed in Table 3 (Factor 4,7 & 9). The validity of the FFMHPQ subscales outside the Australian adult mental health service context was also established. Initially an advisory panel assessed the items in the FFMHPQ subscales for their content validity. Panel members were selected for their expertise in FFP and PMI. All the items to be included were deemed relevant and therefore retained. The final sur- vey including the FFMHPQ was then piloted in one organisation with ten HSC professionals (5 from children’s services and 5 from adult services) not included in the main study to evalu- ate the clarity of the questions and their layout. The main changes made to the survey involved further refinement to the structure and language used particularly in relation to section three of the survey. Participants & procedure Dissemination of the survey among HSC professionals across all five HSC Trusts in NI was achieved in two main ways, (1) online (via Survey Monkey) and (2) hard copy completion. Team managers across adult mental health and children’s services were contacted via email and asked to circulate the information sheet and link to the online survey. The option for hard copy completion was also offered for those who preferred this method. Blank surveys were posted to requested teams using recorded delivery methods and later collected by a member of the research team. A database was created in SPSS for data entry of hard copy surveys; this database was later amalgamated with the online database of completed online surveys. Once both datasets had been merged, correct value range was checked. Additionally, every 25th hard copy survey was audited, and the data compared with the SPSS input data in order to ensure quality and consistency of manually entered data; these checks indicated that the data entry was reliable. We sought to include a wide range of HSC Professionals working across adult and chil- dren’s services with families where a parent has a mental illness. The survey respondents PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 5 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire broadly mirror the relevant workforce which has been the focus of Think Family NI initiatives. The survey was distributed to Approx. 3585 HSC professionals within adult mental health and children’s services across the five HSC Trusts. The minimum number of HSC professionals needed to complete a survey (n = 878) was determined by various factors, including the size of the population to which results are generalizable to, the results of previous research and partic- ularly findings from previous use of the FFMHPQ in different populations and countries and the overall purpose of the current study, which was to compare two groups of HSC profession- als with regard to their FFP. Hence, a two sample comparison of means was used to estimate the overall sample size. We ensured that the characteristics of respondents reflected the population of HSC profes- sionals who fulfilled the inclusion criteria. To promote maximum variation and to secure sam- ple access, a principal investigator (PI) for each Trust was identified along with an independent point of contact for the study. A total of 1088 survey questionnaires were returned by HSC professionals giving a response rate of 30%. However, 119 of these were ineligible based on study inclusion criteria; 48 surveys completed by trainees and support workers, and 71 surveys completed by professionals in inel- igible service areas (e.g. disability services) were excluded. Due to significant missing informa- tion, 101 cases were also removed from the dataset as more than 90% of the survey had not been completed and would not be suitable for inclusion in final analysis. The final sample comprised of 868 HSC professionals, a response rate of 24.2%. The total sample of HSC profes- sionals was derived from all five HSC Trusts and included professionals from both adult men- tal health (n = 493) and children’s social care services (n = 316) (Missing information regarding service area = 59). The largest number of responses were obtained from community mental health teams (28%), followed by children’s services family intervention teams (18.1%), acute mental health and addictions inpatient services (9.3%), initial intake teams for children’s services (Gateway Teams) (9.3%), community addictions teams (6.5%), 16yrs plus teams for adolescents (5.3%), crisis resolution home treatment (4.4%), and single point of access (0.9%). Given the variety of titles and terms attributed to different services across each Trust, the sur- vey offered professionals the option to note their service area under a specialist mental health service or other category (15.2%). Such services included for example unscheduled care, Cog- nitive Behavioural Therapy (CBT), and those working within family centres. A range of professional disciplines across these service areas participated. The most com- mon were Social Workers (n = 473, 54.5%) followed by Nurses (n = 293, 33.8%). Other profes- sionals, included Allied Health professionals (n = 44, 5.1%), Psychiatrists (n = 33, 3.8%), Psychologists (n = 12, 1.4%) and Other, for example, CBT Therapist (n = 13, 1.5). Ethical approval was obtained from the Office for Research Ethics Committees Northern Ireland and Research Governance permission was obtained from the five HSC Trusts which provide statutory HSC services across NI (REC Reference 16/NI/0079). Participants gave informed consent; they were informed of the details of the study in online explanatory state- ments. Implied consent was obtained through participation in the completion of the online or hard copy, anonymous questionnaire. Statistical analysis We used Exploratory Factor Analysis (EFA) to investigate the dimensions underlying the pat- tern of observed responses in the FFMHPQ. Exploratory Factor Analysis was run on the 66 items considering them as ordered categorical items. For this purpose, EFA parameters were estimated using a diagonally Weighted Least Square matrix with Missing Values (WLMSV) estimator, specifically designed for ordered data. We used a Geomin rotation, i.e. an oblique PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 6 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire rotation, which assumed correlations between the underlying dimensions. The use of the WLMSV estimator has the advantage of estimating parameters using all the information pro- vided by respondents, even if information is incomplete (i.e. respondents do not answer all the questions). Approaches like this one, based on Full Information Maximum Likelihood (FIML), have been established to be more reliable and providing less biased results compared to traditional approaches that drop cases with missing data, as long as the reasons that can explain missing data can be assumed to be unrelated to the missing data themselves. Of note, 32 cases did not answer the items on which EFA analyses were based. Thus, results are based on n = 836 respondents that provided complete or incomplete data in the questionnaire. The choice of the final models was based on the eigenvalues for the sample correlation matrix: factors were retained if the eigenvalue of factors was above 1. Furthermore, we esti- mated the model fit indices to check these provided adequate fit. These included the Compara- tive Fit Index (CFI) and the Tucker-Lewis Index (TLI): authors indicate values above 0.90 of these indices signal good model fit [40]. Other indices considered were the Root Mean Square of Error Approximation (RMSEA): a value <0.05 is considered to indicate a “close fit”, while values <0.08 indicate adequate fit. Finally, the Standardized Root Mean Square Residual (SRMR) was also considered: values < 0.08 are considered to indicate adequate model fit. EFA was run using Mplus 7 [41]. The results of the EFA models were inspected and considered alongside theoretical consid- erations [9, 30, 38, 42] to develop a more parsimonious model that could explain the variation in participants’ responses. The fit of this model was tested using Confirmatory Factor Analysis (CFA): CFA is a measurement model whereby the associations between observed items and underlying factors are explicitly modelled to allow items to load only on specific factors (i.e. factors are associated specifically with some items, but not with others). Associations between factors (or lack thereof) can also be modelled, alongside other parameters (e.g. associations between residual variances). For the sake of parsimony, when developing the measurement model through CFA, we retained fewer items in the models, taking into account results from the EFA and theoretical considerations in the selection of the items (i.e. retaining items that had substantive meaning according to our understanding of FFMHPQ). The CFA was run considering the items as categorical ordered variables, similarly to our procedure in EFA. For this reason, we used a WLMSV estimator. Statistics considered in esti- mating model fit were, once again, the CFI, the RMSEA, and the SRMR. Models were esti- mated using Mplus 7 on all n = 835 participants who provided complete and incomplete responses to the items included in the final model. Results Exploratory Factor Analysis Initial review of the scree plot suggested that Factors from 1 to 16 displayed eigenvalues above 1 therefore solutions with 16 factors or less were considered plausible solutions. As can be seen from Fig 1, a sudden shift between the 12 and 13 factor solution was evident, therefore the 12-factor solution was also considered as a plausible option and retained for further inspection. Similarly, the solution with 14 factors was the first to provide adequate fit indices (including TLI) and was also retained for inspection. The fit of different solutions inspected are reported in Table 1. After inspection of these solutions, we considered the solution with 14 factors as the solu- tion that provided the best balance between statistical fit and interpretability of the solution based on evidence based literature [22, 30, 38, 43]. The model with 14 factors displayed PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 7 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire Fig 1. Scree plot of Exploratory Factor Analysis. The horizontal axis reports the factors from factor 1 to factor 25. The vertical axis reports the estimated eigenvalues associated with each factor. Eigenvalue = 1 is indicated by the dotted line. https://doi.org/10.1371/journal.pone.0285835.g001 adequate model fit, as reported in Table 1, and the item loadings provided meaningful and coherent interpretations of the underlying factors, which were in line with our substantive understanding of FFP. Factors seemed to represent some of the previously identified con- structs of FFP such as skills and knowledge of professionals, workplace support and inter-pro- fessional practices [22]. Table 2 reports the items that were most strongly related with each factor. Some correlations between the factors extracted were significant: these correlations are reported in S1 Table. Relationships between subscales measuring professionals’ skills and knowledge (Factors 1, 2, 7, 8, 9), professional practice/ behavioural subscales (Factors 4, 5, 6), and higher order level subscales that describe organization/workplace factors that influence this (Factors 3, 10, 11, 12,) were observed. Confirmatory Factor Analysis (CFA) The 14-factors model described in the previous section was the basis from which we derived a comprehensive more parsimonious model that included 12 factors. In what follows we Table 1. Fit indices of solutions 12 to 16. Model parameters CFI TLI RMSEA SRMR 12-Factor Solution 13-Factor Solution 14-Factor Solution 15-Factor Solution 16-Factor Solution 726 0.957 0.935 0.029 0.028 780 0.962 0.940 0.028 0.026 833 0.966 0.945 0.027 0.025 885 0.970 0.950 0.025 0.023 936 0.974 0.954 0.024 0.022 Chi-Square test of model fit, (degrees of freedom), and p value 2399.12 (1419); p < .0001 2237.13 (1365); p < .0001 2087.59 (1312); p < .0001 1937.59 (1260); p < .0001 1807.36 (1209); p < .0001 https://doi.org/10.1371/journal.pone.0285835.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 8 / 23 PLOS ONETable 2. Factor loadings by items and factors. Loadings with * are significant at p = .05. No Item Factors Validation of the family focused mental health practice questionnaire 1 2 3 4 5 6 7 8 -0.044 -0.106* 0.657* 0.057 0.072 0.003 -0.011 -0.008 9 0 0.203* 0.012 11 12 -0.02 -0.004 13 14 -0.096* 0.093 -0.012 0.061 0.355* -0.013 -0.04 0.205* -0.026 -0.038 0.320* -0.01 -0.028 -0.019 -0.015 0.054 0.003 0.077 0.601* 0.01 -0.073 0.282* -0.091* -0.027 0.02 0.013 -0.01 -0.099 0.006 -0.057 0.06 -0.034 0.144* -0.019 0.266* 0.066 -0.017 0.349* 0.062 -0.047 0.062 -0.187* -0.151* 0.008 0.046 0.107 -0.053 -0.006 0.001 -0.068 0.119* 0.511* 0.267* -0.054 0.063 -0.013 -0.179* -0.09 0.412* 0.082 0.035 -0.038 0.104* -0.116* 0.002 -0.164* -0.01 -0.005 -0.067 -0.186* -0.158* -0.117* -0.056 0.132* 0.087* 0.351* -0.04 -0.003 -0.027 0.259* 0.077 0.089* 0.009 0.045 -0.034 -0.097* -0.061 0.549* -0.074* 0.136* 0.055 0.029 -0.015 0.076 0.263* 0.08 0.068 -0.001 -0.064 -0.059 1 4 6 7 11 12 13 15 My workplace provides mentoring to support health and social care professionals undertaking FFP Government policy regarding FFP is very clear I often receive support from co-workers in regard to FFP I often receive support from co-workers in regard to FFP I am able to determine the developmental progress of children whose parent(s) has mental illness I sometimes wish that I was better able to help parents discuss the impact of their mental illness on their children I am knowledgeable about how parental mental illness impacts on children. I am able to determine the level of importance that parents who have mental illness place on their children maintaining attendance at day to day activities such as school and hobbies (e.g. sport, dance) 17 Working with other health 0.240* 0.269* 0.364* 0.033 0.054 0.049 0.012 0.007 0.016 0.136* 0 -0.006 -0.041 0.065 21 23 24 28 29 and social care professionals enhances my FFP At my workplace, policies and procedures for working with parents who have mental illness on family issues are very clear In my workplace other workers encourage FFP I provide written material (e.g. Think Family educational resources, leaflets) about parenting to parents who have mental illness I am able to assess the level of children’s involvement in their parent’s symptoms I should learn more about how to assist parents who have mental illness with their parenting -0.027 0.017 0.234* 0.038 0.048 0.313* 0.117* -0.014 0.389* 0.125* 0.026 0.015 0.196* 0.008 -0.028 0.163 0.623* 0.017 -0.031 0.212* -0.031 0.004 -0.082 -0.002 0.021 -0.250* 0.099* -0.021 0.029 -0.041 0.236* -0.077 0.169* 0.439* 0.137* 0.197* 0.024 -0.02 0.006 0.05 -0.012 -0.065 -0.05 0.290* 0.038 0.093* 0.054 -0.012 0.041 0.339* 0.305* -0.154* -0.042 0.006 -0.063 0.023 0.615* 0.167* -0.079* -0.023 0.072* -0.041 -0.086* -0.011 0.022 -0.027 0.006 -0.006 -0.081* 0.05 (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 9 / 23 PLOS ONETable 2. (Continued) Validation of the family focused mental health practice questionnaire No 32 33 34 35 36 39 42 43 44 45 46 Item Factors 1 0.076* 2 3 4 5 6 7 8 9 0 11 12 13 14 0.618* -0.059 0.047 0.007 0.139* -0.001 0.102 0.232 0.014 -0.012 -0.017 0.08 0.019 -0.021 0.106 -0.037 0.163* 0.696* -0.049 -0.021 -0.015 0.133* 0.042 0.05 0.007 0.026 -0.002 0.433* 0.280* 0.076 -0.03 0.019 -0.059 0.094* -0.092 0.002 0.190* 0 -0.037 -0.019 0.224* -0.017 0.013 0.034 0.012 0.170* -0.01 -0.034 -0.034 0.052 0.135* 0.563* -0.196* 0.026 -0.024 0.026 -0.014 0.037 -0.049 0.257* 0.447* 0.108* 0.374* 0.011 -0.027 0.041 -0.074 -0.077 0.029 0.853* 0.008 0.01 -0.057 -0.001 0.02 -0.026 0.236* 0.028 0.066* 0 0.011 0.04 -0.006 0.435* 0.230* 0.064 0.046 -0.082 -0.007 0.023 -0.082 -0.079 0.261* 0.023 0.044 -0.053 0.128* 0.157* -0.017 0.018 0.219* 0.484* -0.026 0.057 0.048 -0.008 0.019 -0.045 -0.02 0.04 -0.042 0.978* -0.078* -0.045* 0.041 -0.046 0.018 -0.017 0.281* 0.004 0.03 -0.026 -0.058* 0.146* -0.017 -0.122* -0.007 -0.062 0.349* 0.094* 0.062 -0.005 0.482* 0.008 0.058 0.096* -0.015 0.076 -0.096* 0.697* -0.018 -0.056 0.170* 0.05 0.145* -0.035 0.002 -0.096* -0.069 0.031 -0.02 -0.022 -0.062 I am able to determine the level of importance that parents who have mental illness place on their children maintaining strong relationships with other family members (e.g. other parent, siblings) I refer parents who have mental illness to parent- related programs (e.g. parenting skills) Children and families ultimately benefit if health and social care professionals work together to solve the family’s problems There is time to have regular contact with other agencies regarding parents, families or children (i.e. interface groups such as family support hubs) I regularly provide information (including written materials) about mental health issues to children whose parent(s) has mental illness I would like to undertake future training to increase my skills and knowledge for working with children whose parent(s) has mental illness Team-working skills are essential for all health and social care professionals providing family-focused care I often consider if referral to parent support programme (or similar) is required by parents who have mental illness I would like to undertake training in future to increase my skills and knowledge about helping mentally ill parents with their parenting I am skilled in working with parents who have mental illness in relation to maintaining the wellbeing and resilience of their children I want to have a greater understanding of how to work within the multidisciplinary team to support children and families PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 (Continued ) 10 / 23 PLOS ONETable 2. (Continued) Validation of the family focused mental health practice questionnaire No 47 49 50 51 52 53 54 55 56 57 59 60 54 Item Factors I provide education sessions for adult family members (e.g. about the illness, treatment) I am knowledgeable about the key things that parents who have mental illness could do to maintain the wellbeing (and resilience) of their children I am able to identify how parenthood can precipitate a parent’s mental illness I am able to identify how parenthood can influence a parent’s mental illness I assess the impact of the parenting role on the parent’s mental health I suggest practical strategies to facilitate parents who have mental illness to manage the dual demands of their parenting role and their mental illness or substance misuse I understand how to use Falkov’s Family Model to guide my FFP I perceive that Falkov’s Family Model can guide my FFP I would need to undertake future training to increase my skills and knowledge for using Falkov’s Family Model in practice The regional child protection procedures are clear about when I should be concerned that a parent’s mental illness is impacting negatively on a child I discuss the impact of family functioning, on children’s well-being, with the service user’s adult family members/ carers I would classify my interaction with children whose parent has mental illness as planned, purposeful involvement with therapeutic intervention I understand how to use Falkov’s Family Model to guide my FFP 1 2 3 4 5 -0.027 0.026 -0.03 0.160* -0.056 6 0.505* 7 8 9 0 11 12 13 14 0.055 -0.12 -0.047 0.032 -0.036 0.216 0.122 -0.035 0.01 -0.058 0.023 0.593* 0.008 0.021 -0.061* 0.361* 0.043 -0.002 -0.082* -0.06 0.076 0.101* 0.061* 0.055 0.090* 0.797* -0.150* -0.058 0.054 0.004 0.052 -0.065* -0.019 0.034 0.031 -0.005 0.051 0.045 0.033 0.871* -0.084 -0.078 0.029 -0.074* 0.01 -0.013 0.03 0.023 0.017 0 -0.014 0.028 0 0.649* 0.119 0.073 0.026 0.014 -0.03 0.165* -0.014 0.033 -0.193* 0.007 -0.028 -0.008 -0.051 0.529* 0.119* 0.158* 0.016 0.034 -0.147* 0.229* 0.017 -0.005 -0.205* 0.056 -0.116* -0.02 -0.013 0.064 0.013 0.068 0.914* 0.016 0.03 0.047 0.034 -0.003 -0.037 -0.034 0.046 0.013 -0.02 0.003 0.01 0.049 0.728* -0.038 0.004 0.122* 0.045 0.063 -0.093* 0.045 0.528* -0.024 -0.018 -0.004 0.004 -0.031 -0.381* -0.008 0.021 0.258* 0.023 0.037 -0.113* -0.014 0.03 0.023 0.091 -0.017 -0.121* 0.127 -0.039 -0.012 0.289* 0.534* 0.008 0.002 0.140* 0.007 0.017 0.075 -0.034 0.139* 0.105* 0.043 0.015 0.284* -0.087 0.310* 0.072 -0.212* -0.084 0.071 0.03 0.008 0.015 0.096 -0.043 0.131 0.066 0.446* 0.002 0 0.264* -0.250* -0.074 0.018 0.054 -0.042 0.029 0.005 0.004 -0.081 0.144* 0.022 0.062 0.608* -0.097 -0.109 0.062 -0.079 (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 11 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire Table 2. (Continued) No 66 Item Factors 1 0.095* 2 3 4 5 6 7 8 9 0.027 -0.011 0.114* -0.016 -0.128 0.021 0.077 0.006 0 0.703* 11 12 -0.095 0.04 13 0.03 14 -0.004 I know what to do if I was concerned that a parent’s mental illness was having a significant negative effect on a child 2R In my area we lack services (e.g. other agencies) to refer children to in relation to their parent’s mental illness (i.e. programs for children) 0.024 0.023 -0.027 -0.004 0.006 -0.003 0.099* -0.02 0.042 -0.093 0.152* 0.226* 0.486* -0.027 3R There is no time to work with 0.112* -0.012 0.036 -0.047 0.027 -0.035 0.026 0.181* -0.002 -0.139* 0.512* 0.069 0.169* 0.001 children whose parent has mental illness or substance misuse around issues related to parental mental illness 5R Professional development regarding FFP is not encouraged at my workplace 8R I am not confident working with mentally ill parents on their parenting skills 9R I don’t provide information to the carer and/or family about the service user’s medication and/or treatment 0.093* -0.009 0.621* 0.003 -0.014 -0.107 0.047 0.006 -0.004 -0.034 0.094* 0.340* -0.043 -0.049 -0.059 0.031 0.014 0.045 0.039 0.022 -0.012 0.390* -0.001 0.062 0.036 0.440* -0.011 -0.011 -0.02 0.056 0.085* 0.041 0.03 0.314* 0.035 -0.11 -0.183* 0.01 -0.124* 0.464* 0.061 0.021 10R Many parents who have -0.066 -0.224* -0.027 0.042 -0.110* 0.266* -0.085 0.076 0.023 -0.062 0.023 0.349* -0.038 0.216* mental illness do not consider their illness to be a problem for their children 14R There are no parent-related -0.078 -0.148* 0.012 0.063 0.302* 0.074 -0.016 -0.089 0.012 0.032 0.055 0.234* 0.482* 0.053 programs (e.g. parenting skills) to refer parents with mental illness to 16R I do not refer children whose 0.05 0.035 0.219* -0.047 0.471* -0.045 0.001 0.265* -0.003 -0.117* -0.026 0.120* -0.011 -0.003 parent has mental illness to child focused (e.g. peer support) programs (other than child and adolescent mental health) 18R My workplace does not -0.059 -0.124* 0.706* 0.004 0.062 -0.026 0.037 0.044 0.009 0.052 0.059 0.042 0.055 0.051 provide mentoring to support health and social care professionals undertaking FFP 19R Due to location it is difficult to coordinate families and children with the required services -0.031 0.138* -0.013 -0.102* 0.033 -0.006 0.075 0.037 -0.03 0.048 0.198* 0.003 0.361* 0.001 20R My workload is too high to do -0.024 -0.03 0.039 0.027 -0.05 0.022 0.092* -0.002 -0.027 -0.028 0.736* -0.014 0.027 0.079 family focused work 22R My workplace provides little -0.035 -0.016 0.603* -0.014 0.047 0.004 0.088* 0.056 -0.079* 0.013 0.159* 0.019 0.131* -0.007 support for further training in FFP 25R I am not confident working -0.068 0.143* 0.041 -0.130* 0.025 -0.032 -0.027 0.374* -0.096 0.187* 0.024 0.227* -0.032 -0.038 with families of service user’s (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 12 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire Table 2. (Continued) No Item Factors 26R Rarely do I advocate for the 0.004 1 2 3 0.248* 0.02 4 5 6 7 8 9 0 11 0.025 0.036 0.078 -0.05 0.096 -0.345* 0.237* 0.011 12 0.186* 13 14 0.024 0.019 carers and/or family when communicating with other professionals regarding the service users’ mental illness 27R Discussing issues for the service user with others (including family) would breach their confidentiality 30R I do not have the skills to work with parents who have mental illness about how parental mental illness impacts on children and families 0.006 0.125 0.031 -0.02 -0.056 0.068 0.025 0.035 -0.194* -0.023 0.078 0.016 -0.035 0.172* -0.06 0.121* 0.073* 0.149* -0.016 -0.015 -0.027 0.525* -0.045 0.072 0.026 0.146* 0.089* 0.032 31R There are no family therapy or -0.044 0.015 0.04 0.023 0.251* 0.006 -0.081* 0.025 -0.051 0.013 0.094* -0.035 0.587* 0.016 family counselling services to refer parents who have mental illness and their children to 37R Rarely do I consider if referral 0.044 0.120* 0.093* -0.06 0.389* 0.039 0.113* 0.298* -0.076 -0.104* -0.111* 0.023 0.046 0.081 to peer support program (or similar) is required by children whose parent(s) has mental illness 38R Children often do not want to 0.045 0.075 -0.012 -0.037 0.019 -0.031 0.029 0.133* -0.03 -0.014 0.068 0.065 0.085 0.340* engage with me about their parent’s mental illness 40R I am not experienced in -0.091* 0.038 -0.031 -0.002 0.044 -0.136 0.063* 0.652* -0.036 0.073 -0.047 -0.068 0.032 0.190* working with child issues associated with parental mental illness 41R I am not able to determine the level of importance that parents who have mental illness place on their children maintaining strong relationships with others outside the family (e.g. other children/peers, school) -0.011 0.441* 0.043 0.074 -0.100* -0.031 -0.013 0.460* 0.006 -0.063 -0.046 0.024 0.128* 0.138* 48R I am not confident working 0.016 -0.065 -0.024 -0.014 -0.023 -0.194* -0.01 0.635* 0.007 0.082 0.009 0.077 -0.058 0.105* with children whose parent(s) has mental illness 58R There is no time to work with 0.023 0.024 0.001 0.029 0.016 0.047 -0.031 -0.019 -0.02 -0.048 0.782* 0.003 0.015 0.109* families 61R Parents generally do not want 0.019 -0.025 -0.018 -0.02 0.02 0.033 -0.016 0.049 0.186* -0.03 0.057 0.229* 0.029 0.657* to engage with me about the impact of their mental illness on their children 62R Discussing the impact of parental mental illness on children with parents who have mental illness would compromise rapport with them -0.044 0.045 0.026 0.055 0.023 -0.188* 0.009 0.012 -0.024 0.113 0.013 -0.084 -0.013 0.630* (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 13 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire Table 2. (Continued) No Item 1 2 3 4 5 6 7 8 Factors 63R Insufficient numbers of health -0.161* -0.004 0.056 -0.052 -0.004 -0.074 -0.03 0.068 9 0 0.176* 0.065 11 0.541* 12 13 14 0.051 0.027 0.002 and social care professionals (i.e. nurse, social worker, clinical psychologist) in my service reduces worker’s capacity to address parenting issues 65R I do not understand how to -0.093 0.008 0.070* 0.01 -0.018 -0.08 0.702* 0.054 -0.006 -0.073 -0.063 -0.026 0.104* 0.017 use Falkov’s Family Model to guide my FFP https://doi.org/10.1371/journal.pone.0285835.t002 describe the steps that led to this model. Items were retained based on recommended criteria that item loadings should be above.32, which represent over 10% of item variance explained by the underlying factor. Further, factors with fewer than three items or those items that cross loaded were assigned to the factor determined as a stronger associate with the exception of item 41R (I am not able to determine the level of importance that parents who have mental ill- ness place on their children maintaining strong relationships with others outside the family) which was retained in factors 2 and 4, both of which address child related issues. We consid- ered this as a more parsimonious solution that could adequately represent the variability in participants’ responses particularly as factors 4 and 8 reflect distinctive knowledge regarding either the parent or the child which may also be linked to differences in professional focus and service remit (i.e. Adult mental health and Children services). The items retained in the analy- ses and the underlying factors, together with the factors’ labels and their interpretations are reported in Table 3. In Table 4 we report the fit indices of this 12-factor solution. The model demonstrated adequate fit according to the CFI and the RMSEA. S2 Table displays the factor loadings of the items with the respective underlying factor. Table 3. 12 factor model including items and factor labels and definitions. Factor Items associated with Label for Factors and Cronbach’ alpha Definition of Factors factor 29, 39, 44, 46, 56 15, 32, 41R Training (.83) Connectedness (.67) 1, 5R, 6, 18R, 23, 22R Workplace support (.81) Professional willing to undertake further training Professionals capacity to assess parent awareness of child connectedness The workplace provides support (e.g. supervision and professional development for family focused practice). 49, 50, 51, 52, 53 Skill and Knowledge to support parent and their parenting (.79) Professional’s knowledge and skill regarding impact of parenting on parent’s mental health and activities supporting parent in parenting role 33, 37R, 43 24, 36, 47 54, 55, 65R Referral (.60) Referral to other services Psycho education (.48) Provision of psycho education for parents, children and other adult family members Understanding The Family Model (.78) Awareness of TFM and its relevance in practice 11, 30R, 40R, 41R, 48R Skills and knowledge to support children (.75) Professional’s knowledge and skill to support children directly and via the parent 57, 64, 66 Child protection (.49) Professional’s understanding of child protection protocol 3R, 20R, 35, 58R, 63R Time and workload (.77) Time or workload issues regarding family focused practice 2R, 14R, 31R 38R, 61R, 62R Service availability (.64) Engagement issues (.60) Lack of services to refer parents and children to The opportunity for engagement with parents and children about PMI 1 2 3 4 5 6 7 8 9 10 11 12 https://doi.org/10.1371/journal.pone.0285835.t003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 14 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire Table 4. Model fit parameters of the 12-factor solution used in CFA. Model parameters CFI TLI RMSEA WRMR 12-Factor Solution 389 0.90 0.89 0.049 1.573 Chi-Square test of model fit, (degrees of freedom), and p value 2734.43 (922); p < .0001 https://doi.org/10.1371/journal.pone.0285835.t004 All the items included in the model had generally strong relationships with the underlying factor and were significant at p < .001., as reported in S3 Table. Furthermore, the factors showed patterns of correlations that were consistent with substantive knowledge and theories. These correlations are reported in Table 5. Note that some correlations were negative and sig- nificant (e.g. correlations between Factor 1 Training and Factor 7 Understanding The Family Model = -0.322; Factor 1 Training and Factor 11 Service Availability = -0.321). We suggest that this may be reflecting a relationship between those indicating a need/willingness for further family focused training and those who indicated they had less understanding of The Family Model and that those who felt they needed more training also perceived that there was a lack of services to refer parents and children to. A high number of remaining relationships were positive and strong (r >.32), with 10% of shared variance between these factors. Subscales measuring professionals’ skills and knowledge were positively correlated across one another (e.g. Skills and knowledge to support children WITH Connectedness; Child Protection WITH connectedness). Professional practice/ beha- vioural subscales were also positively correlated with these individual level subscales (e.g. Skills and knowledge to support parent with their parenting WITH Connectedness; Referral WITH Connectedness; Understanding The Family Model WITH Psycho Education and Referral; Skills and knowledge to support children WITH Skill and Knowledge to support parent and their par- enting, Referral and Psycho education; Child protection WITH Skill and Knowledge to support parent and their parenting) as well as across one another (e.g. Referral WITH Skill and Knowl- edge to support parent and their parenting; Psycho education WITH Referral). Higher order level subscales that describe organisation/workplace factors were also posi- tively correlated with individual professionals’ skills and knowledge factors (Understanding The Family Model WITH Workplace support; Engagement issues WITH Skills and knowledge to support children; Time and workload WITH Skills and knowledge to support children) and pro- fessional practice/ behavioural factors (e.g. Referral WITH Workplace support; Psycho educa- tion WITH Workplace support; Time and workload WITH Psycho education) as well as across one another (Time and Workload WITH Workplace support, Service availability WITH Workplace support and Time and Workload; Engagement issues WITH Time and Workload). These findings are reflective of proposed models of family focused practice which suggest that specific family focus practises are influenced by both the individual professional and their core overlapping practices as well as the organisational structures which surround them [20, 22, 44, 45]. Discussion The current study set out to test the psychometric properties and factorial structure of the FFMHPQ in a population of Northern Ireland HSC professionals working within adult mental health and children’s services. Our objective was to explore if family focused practices are PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 15 / 23 PLOS ONETable 5. Correlations between factors of 12 factor model. Correlations Std Error p value Validation of the family focused mental health practice questionnaire Factor #2 Connectedness WITH Training Factor #3 Workplace Support WITH Training F#1 Connectedness F#2 FACTOR #4 Skill and Knowledge to Support Parent and their Parenting WITH Training F#1 Connectedness F#2 Workplace Support F#3 FACTOR #5 Referral WITH Training F#1 Connectedness F#2 Workplace Support F#3 Parenting Support #4 FACTOR #6 Psycho Education WITH Training F#1 Connectedness F#2 Workplace Support F#3 Parenting Support F#4 Referral F#5 FACTOR #7 Understanding The Family Model WITH Training F#1 Connectedness F#2 Workplace Support F#3 Parenting Support F#4 Referral F#5 Psycho Education F#6 FACTOR #8 Skills and Knowledge to Support Children WITH Training F#1 Connectedness F#2 Workplace Support F#3 Parenting Support F#4 Referral F#5 Psycho Education F#6 Understanding The Family Model #7 FACTOR #9 Child Protection WITH Training F#1 Connectedness F#2 Workplace Support F#3 Parenting Support F#4 Referral F#5 Psycho Education F#6 Understanding The Family Model F#7 Skills and Knowledge to Support Children F#8 FACTOR #10 Time and Workload WITH Training F#1 Connectedness F#2 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 0.161* -0.199* 0.193* 0.063 0.565* 0.207* 0.182* 0.459* 0.343* 0.338*** -0.094* 0.275* 0.511* 0.263* 0.564* -0.322* 0.186* 0.344* 0.286* 0.330* 0.444* -0.007 0.491* 0.285* 0.406* 0.520* 0.384* 0.273* 0.249* 0.370* 0.183* 0.484* 0.236* 0.054 0.131 0.246 -0.116 0.137 0.042 0.038 0.04 0.039 0.031 0.036 0.047 0.047 0.043 0.038 0.046 0.048 0.038 0.042 0.044 0.038 0.048 0.038 0.038 0.050 0.040 0.042 0.038 0.035 0.031 0.039 0.042 0.039 0.041 0.044 0.043 0.035 0.042 0.049 0.043 0.045 0.04 0.04 < .001 < .001 < .001 0.1 < .001 < .001 < .001 < .001 < .001 < .001 0.04 < .001 < .001 < .001 < .001 < .001 < .001 < .001 < .001 < .001 < .001 0.875 < .001 < .001 < .001 < .001 < .001 < .001 < .001 < .001 < .001 < .001 < .001 0.265 0.003 < .001 0.004 0.001 (Continued ) 16 / 23 PLOS ONETable 5. (Continued) Validation of the family focused mental health practice questionnaire Workplace Support F#3 Parenting Support F#4 Referral F#5 Psycho Education F#6 Understanding The Family Model F#7 Skills and Knowledge to Support Children F#8 Child Protection F#9 FACTOR #11 Service Availability WITH Training F#1 Connectedness F#2 Workplace Support F#3 Parenting Support F#4 Referral F#5 Psycho Education F#6 Understanding The Family Model F#7 Skills and Knowledge to Support Children F#8 Child Protection F#9 Time and Workload #10 FACTOR #12 Engagement Issues WITH Training F#1 Connectedness F#2 Workplace Support F#3 Parenting Support F#4 Referral F#5 Psycho Education F#6 Understanding The Family Model F#7 Skills and Knowledge to Support Children F#8 Child Protection F#9 Time and Workload #10 Service Availability #11 https://doi.org/10.1371/journal.pone.0285835.t005 Correlations 0.473* 0.074 0.296* 0.381* 0.147* 0.348* 0.005 -0.321* 0.062 0.412* 0.137 0.31* 0.246* 0.149 0.162* 0.044 0.495* -0.058 0.245* 0.268* 0.278* 0.295* 0.178* 0.132 0.563* 0.107* 0.443* 0.271* Std Error 0.032 0.039 0.043 0.041 0.042 0.037 0.045 0.041 0.048 0.039 0.043 0.047 0.049 0.047 0.045 0.049 0.037 0.049 0.049 0.042 0.04 0.052 0.051 0.048 0.036 0.051 0.04 0.053 p value < .001 0.058 < .001 < .001 < .001 < .001 0.913 < .001 0.192 < .001 0.001 < .001 < .001 0.002 < .001 0.362 < .001 0.234 < .001 < .001 < .001 < .001 < .001 0.006 < .001 0.036 < .001 < .001 influenced by both the individual professional and their core overlapping practices as well as the organizational structures which surround them. Following robust statistical analysis including EFA and CFA, results suggest that 12 sub- scales containing a total of 46 items was most optimal using the current sample. These sub- scales are slightly different from the original scale with five related to the individual professional (i.e. knowledge and skill, willingness to undertake training), three describing fam- ily focused behaviours (i.e. referral, psycho education and skill and knowledge to support the parent and parenting), and four organizational factors impacting these behaviours, such as time and workload and workplace support. Test of internal consistency using Cronbach’s alpha suggest that the 46-item version is reli- able (.85) with subscale scores ranging from.60 to.83 with the exception of Psycho Education (.48) and Child protection (.49), each of which had 3 items. As Ueno et al [38] note “In some literatures it is reported that reliability is “acceptable” if Cronbach’s alpha is greater than.60 or.70 but it is recommended that care should be taken when using these subscales” (p.66). Our findings also coincide with that of other studies which found that while the FFMHPQ had PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 17 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire good validity and was reasonably reliable, internal reliability was poorer within some subscales [33, 37–39, 46]. Therefore, like existing research we feel it imperative for future research using the FFMHPQ in the context of multiple service/ sector assessment, to address and improve the reliability of the weaker items. Nonetheless, the 12 subscales identified are meaningful and consistent with substantive the- ories and proposed models of FFP and with our objective, which suggest that specific family focus practises are influenced by both the individual professional and their core overlapping practices as well as the organisational structures which surround them [22, 23, 43, 44]. For example, we found inter-correlations that are consistent with known professional and organi- zational processes identified as either promoting or hindering FFP [8, 12, 38] such as Time and workload with Skills and knowledge to support children, and Referral with Workplace support. We also observed a negative correlation between Training and Understanding The Family Model and Training and Service Availability. We suggest this may reflect a relationship between those indicating a need/willingness for further family focused training, and those who indicated they had less understanding of The Family Model and that those who felt they needed more training also perceived a lack of services to refer parents and children to. Such findings are perhaps similar to Ueno et al. [38] who highlight how professionals who had pre- vious training in FFP had more confidence to engage in FFP and indicated that they undertook more family focused activities including referral of parents and children to other services than those professionals without training. Moreover, current findings also support behavioural intention models such as the Theory of Planned Behaviour which suggests that an individual’s attitudes and knowledge may influ- ence their behaviour [47]. More broadly, our findings coincide with the wider literature on FFP which suggests that if professionals have positive attitudes towards FFP and the necessary knowledge and skills to practice this way, they are more likely to adopt a whole of family approach [8, 12, 48]. This connection between attitudes, knowledge and skill and behaviour highlights the importance of organisations having effective implementation strategies to embed FFP [36, 45]. Future research should further examine these relationships and identify key predictors of FFP within adult mental health and children’s services. The results of the current study are also similar to that of Maybery et al. [30] who identified 14 factors (including professional and organizational) when used in adult mental health ser- vices in Australia and Ueno et al. [38] who identified 13 factors in adult mental health services in Japan. The meaningful clusters and similarity of results, for the most part, between the three countries and cultures in terms of the dimensions identified suggests that the scale is a good measure of family focused behaviours and factors that impact these. Differences between the current results and that of the original measure [30], and an adaption by Ueno et al. [38], relate mostly to labelling of factors and clustering of items within. For instance, while Maybery et al. [30] identified two separate subscales related to workplace support and professional develop- ment, items in the current study loaded into one factor labelled “workplace support” which had good reliability (.81). The differences between results concerning this scale may be explained by contextual factors that pertain to the pathways offered to professionals to engage in FFP. The results of the current study and that of Maybery et al. [30] and Ueno et al. [38], dif- fer substantially to that of Leonard et al. [37] who in a population of health visitors identified two factors related to the individual professional and organisation. These differences suggest that while the measure is flexible it may have more relevance for some services, sectors and professionals than others, depending on the professionals’ remit, understanding and operalisa- tion of FFP. This needs consideration in future research examining FFP in different services other than adult mental health and children’s services. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 18 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire In adapting the measure for the current study, the authors, as previously noted, also devel- oped three additional subscales, one of which aimed to measure professionals’ knowledge and skills in supporting parents to cope with the impact of parenting on their mental health. While Maybery et al. [30] identified a single subscale on knowledge and skills, (primarily related to supporting children), in the current study these items loaded onto two subscales, measuring knowledge and skills to support either parents or children. Two distinct factors related to knowledge and skill within the current study make sense; they reflect the additional items gen- erated and distinction between knowledge and skills to support children and to support parents as discussed in the literature. A key barrier cited in the literature for FFP is the lack of an integrated approach to service provision and lack of interagency cooperation between adult mental health and children’s services [24, 35]. The predominant focus of either service on parents or children as opposed to both together has led to professionals in adult mental health services perceiving that they do not have the skills to support children and for professionals in children’s services to support parents. Our findings further underscore this distinction in knowledge, skills and practice between services and sectors and this should be factored into future benchmarks and organizational support of FFP. The current study also identified two additional individual professional factors, related to knowledge and skills, including understanding of [17, 27] The Family Model and child protec- tion protocols. As previously noted, professionals’ understanding and application of The Fam- ily Model is particularly important in NI considering the HSCB has endorsed its use since 2009 as a framework to embed FFP within services and to structure provision of in service training in this area [8, 28]. The inclusion of this new and reliable subscale on TFM will pro- vide future scope for those organisations who have endorsed this model to monitor and improve its translation in practice. A new subscale on child protection was also developed because while the subscale on connectedness, first identified by Maybery et al. [30], measures professionals’ skill and knowledge capacity beyond assessing abuse and neglect, the original measure did not include items measuring understanding of child protection protocols. It is important to measure professionals’ understanding of child protection protocols in NI and elsewhere because they can be used to inform and encourage an early intervention approach to support parents in their parenting as opposed to being solely applied in response to situations where children’s well-being is at risk [14]. Grant et al. [14] found that HSC professionals per- ceived that they were better able to engage parents around parenting when they highlighted their responsibilities in relation to child protection while at the same time emphasizing the importance of intervening early to support both parents and children and to keeping families together when possible. Limitations Although the professional composition of respondents was similar to those in other studies [30, 38], the sample was drawn from two different sectors making it difficult to directly com- pare findings with previous research. The FFMHPQ is a self-report questionnaire from the professionals’ perspectives and this may not be a fully accurate reflection of actual practice. It may also differ from perspectives of service users, families and managers. Observational research might be conducted to provide additional data regarding the dimensions of FFP and could include the perspectives of family members and managers. While the FFMHPQ had documented validity and reliability in the Australian context [30], in the current study two of the subscales, psychoeducation and child protection, had Cronbach’s alpha coefficients below 0.60, hence care should be taken when using these. In addition, in future research, efforts could be made to improve their reliability. The nature and specificity of the items within these PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 19 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire two subscales could be revised in order to better reflect professional’s experience in relation to the concepts being measured. Conclusion In conclusion, this study identifies a 12-subscale measure of professionals’ FFP in adult mental health and children’s services. The FFMHPQ (NI version) was adapted from the original mea- sure [30], in conjunction with the developers of the instrument in response to the emerging lit- erature on PMI, FFP and organisational developments in FFP in NI. This study has developed further understanding of the dimensions measured within the FFMHPQ and consistency of findings with previous studies. Indeed, the factors identified in our CFA indicate attitudes and behaviours that are known to play a key role in the propensity to engage with FFP. They also represent key contextual factors that can affect the former ones. The associations between fac- tors displayed in our results support the conclusion that the FFMHPQ measure provides a reli- able description of FFP engagement and its supporting factors. The test of internal consistency using Cronbach’s alpha suggest that the 46-item version is reliable (.85) with subscale scores, aside from two, ranging from.60 to.83. Furthermore, the 12-factor model provided good fit to the data, based on common indices used in CFA. Thus, the results of this study indicate the FFMHPQ is a reliable tool that addresses the pau- city of measurement tools in this area, which persists despite increasing recommendations for professionals to engage in FFP and for organizations to benchmark and promote it. Our results indicate that the FFMHPQ can be used to help identify professionals’ key formative needs in relation to FFP as well as what predicts it; thereby presenting a means of benchmarking, moni- toring and evaluating service provision and further developing training programmes. Further research using our adapted version of the FFMHPQ should examine and attempt to replicate the psychometric properties of the measure as well as extend its component structure to any items or factors not included in the current research. Supporting information S1 Table. Correlations matrix of 14 factor solution. (DOCX) S2 Table. Items factor loadings (Est. = Estimate), Standard Errors (S.E.) and z of items by factors in the 12-factor solution. (DOCX) S3 Table. Standardised loadings, standard errors, and p values of item loadings by items and by factors. (DOCX) S1 File. (DOCX) S1 Data. (OUT) S2 Data. (OUT) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 20 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire Acknowledgments We thank the professionals who completed the survey and the staff in the participating services for their assistance with recruitment and logistics. Author Contributions Conceptualization: Anne Grant, Susan Lagdon, Gavin Davidson. Data curation: Anne Grant, Susan Lagdon, John Devaney, Gavin Davidson, Joe Duffy, Oliver Perra. Formal analysis: Anne Grant, Susan Lagdon, Oliver Perra. Funding acquisition: Anne Grant. Investigation: Anne Grant, Susan Lagdon, John Devaney. Methodology: Anne Grant, Susan Lagdon, John Devaney, Gavin Davidson, Oliver Perra. Project administration: Anne Grant, Susan Lagdon. Resources: Anne Grant, Susan Lagdon. Software: Anne Grant, Susan Lagdon, Oliver Perra. Supervision: Anne Grant. Validation: Anne Grant, Susan Lagdon, Oliver Perra. Visualization: Anne Grant, Susan Lagdon, Oliver Perra. Writing – original draft: Anne Grant, Susan Lagdon, Oliver Perra. Writing – review & editing: Anne Grant, Susan Lagdon, John Devaney, Gavin Davidson, Joe Duffy, Oliver Perra. References 1. Maybery D. and Reupert A., "The number of parents who are patients attending adult psychiatric ser- vices," Curr Opin Psychiatry, vol. 31, p. 358–62, 2018. https://doi.org/10.1097/YCO. 0000000000000427 PMID: 29847344 2. Maybery D., Reupert A., Patrick K., Goodyear M. and Crase L., "Prevalence of parental mental illness in Australian families," The Psychiatric Bulletin, vol. 33, no. 1, pp. 22–26, 2009. 3. G. Parker, B. Beresford, S. Clarke, K. Gridley, R. Pitman, G. Spiers, et al, "Technical Report for SCIE research review on the prevalence and incidence of parental mental health problems and the detection, screening and reporting of parental mental health problems (Report No.2247).," York, 2008. 4. Foundation M. H., "What are mental health problems?". 5. England P. H., "Statistical commentary: children living with parents in emotional distress, 2020 update," London, 2020. 6. Abel K., Hope H., Parisi R., Ashcroft D., Kosidou L., Osam C., et al. Pierce and Swift E., "Prevalence of maternal mental illness among children and adolescents in the UK between 2005 and 2017: a national retrospective cohort analysis," Lancet Public Health, vol. 4, p. 291–300, 2019. https://doi.org/10.1016/ S2468-2667(19)30059-3 PMID: 31155222 7. L. Bunting, C. McCartan, G. Davidson, A. Grant, O. McBride, C. Mulholland, J. et al, "The Mental Health of Children and Parents in Northern Ireland: Results of the Youth Wellbeing Prevalence Survey. Final Report.," Belfast, 2020. 8. A. Grant, S. Lagdon, J. Devaney, G. Davidson, J. Duffy, O. Perra, et al, "A Study of Health and Social Care Professionals’ Family Focused Practice with Parents who have Mental Illness, their Children and Families in Northern Ireland. Summary Report," Belfast, 2018. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 21 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire 9. Grant A., Reupert A., Maybery D. and Goodyear M., "Predictors and enablers of mental health nurses’ family-focused practice," International Journal of Mental Health Nursing, vol. 28, p. 140–151, 2018. https://doi.org/10.1111/inm.12503 PMID: 29949231 10. Mennen F., Pohle C., Monro W., Duan L., Finello K., Ambrose S., et al, "The Effect of Maternal Depres- sion on Young Children’s Progress in Treatment," Journal of Child and Family Studies, vol. 24, no. 7, p. 2088–2098, 2015. 11. Nilsson S., Gustafsson L. and Nolbris M., "Young adults’ childhood experiences of support when living with a parent with a mental illness," Journal of Child Health Care, vol. 19, no. 4, p. 444–453, 2015. https://doi.org/10.1177/1367493513519296 PMID: 24486816 12. Maybery D., Goodyear M., Reupert A. and Grant A., "Worker, workplace or families: What influences family focused practices in adult mental health?," Journal of Psychiatric and Mental Health Nursing, vol. 23, p. 163–171, 2016. https://doi.org/10.1111/jpm.12294 PMID: 27170070 13. Ruud T., Maybery D., Reupert A., Weimand B., Foster K., Grant A., et al, "Adult mental health outpa- tients who have minor children: Prevalence of parents, referrals of their children, and patient character- istics," Frontiers in Psychiatry, vol. 10, no. 163, 2019. https://doi.org/10.3389/fpsyt.2019.00163 PMID: 31001149 14. Grant A., Devaney J., Davidson G., Lagdon S. and Duffy J., "Think Family, Think Relationship: factors influencing stronger professional relationships with parents diagnosed with a mental illness. A qualita- tive study," British Journal of Social Work, vol. 51, no. 3, p. 945–963, 2021. 15. Beardslee W., Solantaus T., Morgan B., Gladstone T. and Kowalenko N., "Preventive interventions for children of parents with depression: International perspectives," Medical Journal of Australia, vol. Open 1(Suppl 1), p. 23–25, 2012. 16. N. Stanley and P. Cox, "Parental Mental Health and Child Welfare: Reviews of Policy and Professional Education," Social Care Institute for Excellence, London, 2009. 17. A. Falkov, The Family Model Handbook. An Integrated Approach to Supporting Mentally ill Parents and their Children, Sussex, UK: Pavilion, 2012. 18. Hine R., Maybery D. and Goodyear M., "Challenges of connectedness in personal recovery for rural mothers with mental illness," International Journal of Mental Health Nursing, vol. 27, no. 2, pp. 672– 682, 2017. https://doi.org/10.1111/inm.12353 PMID: 28721701 19. Montgomery P., Mossey S., Bailey P. and Forchuk C., "Mothers with serious mental illness: Their expe- rience of "hitting bottom"," ISRN Nursing, vol. 708318, 2011. https://doi.org/10.5402/2011/708318 PMID: 22007325 20. Foster K. M. D. R. A., Gladstone B., Grant A., Ruud T., Falkov A. and Kowalenko N., "Family-focused practice in mental health care: an integrative review," Child and Youth Services, vol. 37, no. 2, pp. 129– 155, 2016. 21. Siegenthaler E., Munder T. and Egger M., "Effect of preventive interventions in mentally ill parents on the mental health of the offspring: Systematic review and meta-analysis," Journal of the American Acad- emy of Child and Adolescent Psychiatry, vol. 51, no. 1, p. 8–17, 2012. https://doi.org/10.1016/j.jaac. 2011.10.018 PMID: 22176935 22. 23. Lagdon S., Grant A., Davidson G., Devaney J., Donaghy M., Duffy J., et al, "Families with parental men- tal health problems: A systematic narrative review of family focused practice," Child Abuse Review, 2021. Foster K., Goodyear M., Grant A., Weimand B. and Nicholson J., "Family-focused practice with EASE: a practice framework for strengthening recovery when parents have mental illness," International Jour- nal of Mental Health Nursing, vol. 28, pp. 351–60, 2018. 24. A. Grant, A. Falkov and B. Weimand, "Enhancing family focused practice in the voluntary and commu- nity sector in Northern Ireland using The Family Model," 2019. 25. Goodyear M., Hill T., Allchin B., McCormick F., Hine R., Cuff R., et al, "Standards of practice for the adult mental health workforce: meeting the needs of families where a parent has a mental illness," Inter- national Journal of Mental Health Nursing, vol. 24, no. 2, pp. 169–180, 2015. https://doi.org/10.1111/ inm.12120 PMID: 25619407 26. R. C. o. Psychiatrists, "Parents as Patients: Supporting the Needs of Patients who are Parents and their Children," Royal College of Psychiatrists, London, 2011. 27. Falkov A., Crossing Bridges: Training Resources for Working with Mentally ill Parents and their Chil- dren, Brighton: Pavillion Publishing, 1998. 28. M. Donaghy, "Chapter 22 Think Family, Northern Ireland," in Diggins, M. Improving practice and work- ing together across health and social care parental mental health and child welfare work. Learning from success, The Pavillion Annuals, 2016. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 22 / 23 PLOS ONEValidation of the family focused mental health practice questionnaire 29. Social Care Institute for Excellence, "At a glance 55: Think child, think parent, think family: Putting it into practice," London, 2012. 30. Maybery D., Goodyear M. and Reupert A., "The family-focused mental health practice questionnaire," Archives of Psychiatric Nursing, vol. 26, no. 2, pp. 135–144, 2012. https://doi.org/10.1016/j.apnu.2011. 09.001 PMID: 22449561 31. Maybery D. and Reupert A., "Workforce capacity to respond to children whose parents have a mental ill- ness," Australian and New Zealand Journal of Psychiatry, vol. 40, no. 8, pp. 657–664, 2006. https://doi. org/10.1080/j.1440-1614.2006.01865.x PMID: 16866761 32. Costello A. and Osborne J., "Best practices in exploratory factor analysis: four recommendations for getting the most from your analysis," Practical Assessment, Research, and Evaluation, vol. 10, no. 7, 2005. 33. 34. 35. Laletas S., Goodyear M. and Reupert A., "Parental Mental Illness: Cross-Sectional Analysis Of Family Focused Practice within the Early Childhood Sector," J Child Fam Study, vol. 27, p. 1650–1660, 2018. Tungpunkom P., Maybery D., Reupert A., Kowalenko N., Foster K. and et al., "Mental health profession- als’ family-focused practice with families with dependent children: a survey study," BMC Health Serv Res, vol. 17, no. 818, p. 2–8, 2017. https://doi.org/10.1186/s12913-017-2761-7 PMID: 29221455 Lauritzen C. and Reedtz C., "Child responsible personnel in adult mental health services," International Journal of Mental Health Systems, vol. 10, no. 64, 2016. https://doi.org/10.1186/s13033-016-0098-y PMID: 27752281 36. Skogøy B., Sørgaard K., Maybery D., Ruud T., Stavnes K. and Kufås E. e. a., "Hospitals implementing changes in law to protect children of ill parents: a cross-sectional study," BMC Health Serv Res, vol. 18, no. 609, 2018. 37. Leonard R., Linden M. and Grant A., "Psychometric evaluation of the Family Focused Mental Health Practice Questionnaire in measuring home visitors’ family focused practice," PLOSone, vol. 13, no. 9, 2018. https://doi.org/10.1371/journal.pone.0203901 PMID: 30212539 38. Ueno R., Maybery D., Reupert A. and Osada H., "Translation and validation of the family-focused men- tal health practice questionnaire-japanese version," International Journal of Mental Health Promotion, 2020. 39. Grant A., "Registered Psychiatric Nurses’ Practice with Parents who have Mental Illness, their Children and Families, within General Adult Mental Health Services in Ireland," Melbourne, 2014. 40. R. Kline, Principles and Practice of Structural Equation Modeling, Fourth ed. 41. 42. L. K. Muthe´ n and B. O. Muthe´n, "Version 7.1 Mplus Language Addendum," 2013. Leonard R., Linden M. and Grant A., "A multi-perspective exploration of health visitors’ family focused practice with mothers who have mental illness and their families: A mixed methods study protocol," Advances in Mental Health, 2019. 43. Goodyear M., Maybery D., Reupert A., Allchin R., Fraser C., Fernbacher S., et al, "Thinking families: A study of the characteristics of the workforce that delivers family-focussed practice," International Jour- nal of Mental Health Nursing, vol. 26, no. 3, pp. 238–248, 2017. 44. Marston N., Stavnes K., Van Loon L., Drost L., Maybery D., Mosek A., et al, "A content analysis of Inter- vention Key Elements and Assessments (IKEA): What’s in the black box in the interventions directed to families where a parent has a mental illness?," Children & Youth Services, vol. 37, no. 2, p. 112–128, 2016. 45. 46. Falkov A., Goodyear M., Hosman C., Biebel K., Skogøy B. K. N., Wolf T., et al, "A systems approach to enhance global efforts to implement family-focused mental health interventions," Child & Youth Ser- vices, vol. 37, no. 2, pp. 175–193, 2016. van Doesum K., Maia T., Pereira C., Loureiro M., Marau J. and et. al., "The impact of the “SEMENTE” program on the family-focused practice of mental health professionals in Portugal," Frontiers in Psychia- try, vol. 10, no. 305. https://doi.org/10.3389/fpsyt.2019.00305 PMID: 31133896 47. Ajzen I., "Attitude structure and behavior," in Attitude Structure and Function, Lawrence Erlbaum Asso- ciates, Inc., 1989, p. pp. 241–274. 48. Leonard R., Linden M. and Grant A., "Predictors of family focused practice among health visitors: A mixed methods study," vol. 76, no. 5, p. 1255–1265, 2020. https://doi.org/10.1111/jan.14310 PMID: 32012334 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285835 May 22, 2023 23 / 23 PLOS ONE
10.1371_journal.pone.0285841	RESEARCH ARTICLE Factors responsible for Ixodes ricinus presence and abundance across a natural-urban gradient The´ rese Janze´ nID, Monica Hammer, Mona PeterssonID, Patrik Dinne´ tzID So¨ derto¨ rn University, School of Natural Sciences Technology and Environmental Studies, Huddinge, Sweden therese.janzen@sh.se Abstract To better understand the spatial distribution of the common tick Ixodes ricinus, we investi- gated how local site factors and landscape characteristics influence tick presence and abun- dance in different greenspaces along the natural-urban gradient in Stockholm County, Sweden. Ticks and field data were collected in 2017 and 2019 and analyzed in relation to habitat type distributions estimated from land cover maps using geographical information system (GIS). A total of 1378 (992 larvae, 370 nymphs, 13 females, and 3 males) questing ticks were collected from 295 sampling plots in 47 different greenspaces. Ticks were present in 41 of the 47 greenspaces and our results show that both local site features such as vege- tation height, and landscape characteristics like the amount of mixed coniferous forest, sig- nificantly affect tick abundance. Tick abundance was highest in rural areas with large natural and seminatural habitats, but ticks were also present in parks and gardens in highly urbanized areas. Greenspaces along the natural-urban gradient should be included in sur- veillance for ticks and tick-borne diseases, including highly urbanized sites that may be per- ceived by the public as areas with low risk for tick encounters. Introduction Human-environment interactions are crucial components of the epidemiological patterns of vector-borne diseases [1]. The periphery around cities constitutes an environmental gradient ranging from natural and semi-natural areas to highly modified urban landscapes [2]. Natural- urban gradients include a broad variety of greenspaces, spanning from forests to high mainte- nance parks and gardens [3]. Greenspaces are crucial for terrestrial biodiversity and for human well-being, providing urban citizens with access to nature [4]. Greenspaces also pro- vide suitable habitats for insects and arachnids, like mosquitos and ticks [5]. Unfortunately, these organisms are not only bloodsucking parasites, but they can also jeopardize public health and well-being by transmitting vector-borne pathogens to humans and pets [6–10]. Ixodes ricinus (Linnaeus 1758) is a tick species of the family Ixodidae (Acari) commonly found in Europe [11]. I. ricinus has three active life stages, larva, nymph, and adult, that each require a blood-meal from a vertebrate host [12]. Humans and our companion animals are potential hosts for all I. ricinus life stages. Most tick-borne microorganisms that are pathogenic to humans are predominantly horizontally transmitted and can be found only in nymphs, and a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Janze´n T, Hammer M, Petersson M, Dinne´tz P (2023) Factors responsible for Ixodes ricinus presence and abundance across a natural- urban gradient. PLoS ONE 18(5): e0285841. https://doi.org/10.1371/journal.pone.0285841 Editor: Janice L. Bossart, Southeastern Louisiana University, UNITED STATES Received: May 31, 2022 Accepted: May 2, 2023 Published: May 17, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0285841 Copyright: © 2023 Janze´n et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The data underlying the results presented in the study are available from Swedish national dataservice: https://snd.gu. se/en/describe-and-share-data. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 1 / 19 PLOS ONEFunding: This research was supported by the Foundation for Baltic and East European Studies (https://ostersjostiftelsen.se/en/), grant number 52/ 18 to PD, MH, MP. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Factors responsible for Ixodes ricinus presence and abundance adults infected after feeding [13]. Nymphs are more abundant than adult ticks in nature, and due to their smaller size much harder to detect on the human body [14]. Therefore, nymphs are considered the most important stage for the transmission of tick-borne pathogens to humans [15]. The off-host periods for I. ricinus are spent in the leaf litter, humus layer, or in the upper soil layers, where they can hide to avoid desiccation and develop to their next devel- opmental stage [16]. During favorable conditions, I. ricinus climb up on the lower vegetation and use a “sit-and-wait” strategy called questing until a suitable host passes by [17]. To reduce human risk for tick induced diseases in the future, we need a better understand- ing of the underlying factors driving tick population dynamics [18]. Local distribution patterns of questing ticks are largely determined by microclimate and availability of hosts, two factors strongly associated with habitat type [19]. Earlier studies have shown that local habitat condi- tions like tree density and shrub encroachment significantly increased tick abundance, and vegetation height significantly decreased tick abundance [20–22]. At landscape scale, earlier research has shown that the distribution of different forest types affects tick abundance [23– 25]. Different land cover types can also affect tick host species abundance, which has a corre- sponding effect on tick abundance [26]. Many of the earlier studies on tick distributions focused either on small scale local factors, or on large scale landscape characteristics. There have only been a few attempts to bring the different scales together [27, 28]. The aim of this study is to investigate potential factors responsible for I. ricinus presence and abundance in different greenspaces along the natural-urban gradient in Stockholm County, Sweden. Specifically, we test whether variation in I. ricinus presence and abundance can be explained by local habitat factors, land cover characteristics and landscape configura- tion related to different levels of urbanization. Materials and methods Study area Stockholm County is situated in the hemi-boreal forest zone, has a large archipelago in the Bal- tic Sea, and a high coverage of natural and semi-natural areas (Fig 1). The predominant vegeta- tion type is coniferous forest, but mixed stands with conifers and broadleaved trees are regularly present. Pure broadleaved stands are mainly found close to arable land and pastures. Most forests are moist, and wetlands are scattered throughout the County [29]. The Stockholm region is one of the fastest expanding urban areas in Europe, applying a polycentric develop- ment pattern with regional urban cores. Stockholm is planned with ten green wedges stretch- ing from rural areas to more central parts of the city and many residential areas are therefore interconnected with greenspaces that host a wide variety of plant and animal species [30]. Sampling site selection and urbanization gradient In 2017, ticks and field data were collected from 12 different sites around Stockholm County, originally chosen as random controls for a tick-borne encephalitis (TBE) study but never used. In 2019, we made inventories at 35 random sites along the natural-urban gradient around Stockholm. To identify sampling sites along the natural-urban gradient in 2019, we first selected 100 random coordinates within Stockholm County using ArcGIS Pro (Version 2.5.0, ESRI Redland, USA). Using the proportion of Artificial surfaces surrounding each site we cal- culated an urbanization index for each of the 100 coordinates. The urbanization index was cal- culated for a buffer zone with 1000m radius around each coordinate using satellite data maps with 25 m resolution (reference year 2014) for Stockholm County. These satellite images are freely accessible at the Swedish Environmental Protection Agency’s website [31]. The urbani- zation index ranges from 0 to 100 with the value 0 indicating a completely natural or semi- PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 2 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance Fig 1. The hierarchical study design with: (A) sampling sites randomly distributed along the natural-urban gradient in Stockholm County, Sweden, (B) 2m × 2m sampling plots shown as a photo of one plot at one site, and (C) buffer zones in GIS with a 100m–1000m radius around each sampling site used for land cover data information [31], picture shows a 500m radius buffer. In each sampling plot (B) we collected questing Ixodes ricinus and measured the vegetation height, the tree stem density surrounding each plot, and measured the current environmental conditions. https://doi.org/10.1371/journal.pone.0285841.g001 natural site and a value of 100 indicating a landscape with only artificial surfaces. We consid- ered sites with more than 50% Artificial surfaces as urban core areas and therefore excluded them from the site selection. Since Stockholm County is surrounded by water, we subtracted the proportion of Open water to adjust for presence of non-tick habitats (Eq 1). From all coor- dinates with an urbanization index � 50% we randomly selected 35 sampling sites. The selected sampling sites ranged from large greenspaces in rural settings, to parks and small patches of vegetation in highly urbanized areas just outside the city center. Urbanization index ¼ Artificial surfaces %ð Þ = 100 (cid:0) Open water %ð Þ ð Þ � 100 ð1Þ At the first 12 sampling sites in 2017, we randomly selected 10 random 2m × 2m plots for inventories of ticks and field data. At the 35 sampling sites in 2019, we collected ticks and field data in 5 random 2m × 2m plots. All 47 sampling sites were visited once, either in 2017 or in 2019, with a total of 295 sampling plots inventoried for ticks and field data. In addition to the random sampling method, we used equal sampling effort in each sampling plot which allows PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 3 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance for comparisons of factors affecting I. ricinus presence and abundance among sampling sites as long as we control for the number of plots per site. Selection of sampling plots In south and central Sweden, the current climate allows I. ricinus to be active during 6–8 months per year [32]. Nymphs and larvae usually exhibit a bimodal activity pattern with peaks in May—June and August—September, and adult ticks usually exhibit a unimodal activity pat- tern [33]. Field sampling for both years occurred irregularly with most sampling times in June and August. The site coordinates randomly selected in GIS served as starting points to select sampling plots within the sampling sites. If any of the coordinates were located on an artificial surface, the nearest greenspace was identified and a random distance from the entrance into the green- space was chosen to determine a new starting coordinate. For selection of individual 2m × 2m sampling plots at each site we employed a random procedure using compass bearing and dis- tance: starting at the randomly selected GIS generated coordinate, the location for the first sampling plot was determined by holding a compass upside down and randomly rotating the graduation ring to first select a distance between 1m–36m (10˚ on the graduation ring equals 1m), and secondly a random compass direction between 0˚–360˚. The first sampling plot was used as starting point for selecting the second plot, and so on up to five or ten plots per sam- pling site in 2019 and 2017, respectively. The exact coordinates of each sampling plot were determined with a GPS hand receiver (Garmin eTrex 30). Collection of ticks and field data Tick collections were performed with the mopping technique [34]. The mopping technique entails a white flannel blanket measuring 0.7m × 0.7m attached to a mop. In contrast to the common dragging technique, the mop is held in front of the user and is moved anteriorly over the vegetation. The mop handle permits easy adjustments to different vegetation heights [35]. The total area of each plot was swept once. All questing ticks attaching to the blanket were col- lected with tweezers, put into tubes, categorized according to life stage, and later stored at -80ºC. Tick life stage categorization was later confirmed in the laboratory. Seasonal questing activity of I. ricinus ticks can differ between biotopes and life stage [14, 33]. To cover unimodal and bimodal activity patterns, ticks and field data were collected from June to October between 10am and 6pm during days without long-lasting precipitation. We sampled on average three sampling sites per day from different levels of urbanization, ran- domly changing the natural-urban gradient direction on each sampling day. This was done to ensure that sampling at natural, intermediate, and urban areas of the natural-urban gradient was always completed within the same day, and under similar weather conditions. The tem- perature on the sampling days ranged from 3ºC– 27ºC. For each sampling plot (2017 and 2019) we recorded date, time, temperature, weather conditions, number of ticks, vegetation height, and tree stem density surrounding the plot. Tree stem density was estimated using the Bitterlich sampling technique [36]. Land cover data A common procedure for analyzing the effect of different spatial scales is to establish different buffer zone sizes around each sampling location in GIS [37, 38]. To retrieve large scale land- scape characteristics, GPS coordinates from the sampling plots were geocoded and incorpo- rated within a GIS. We established 10 buffer zones with increasing radii from 100m to 1000m around the centroid of all sampling plots at each sampling site. The smallest buffer zones PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 4 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance (100m) represent local habitat land cover at the sampling site. Larger buffer zones represent the land cover distribution in the adjacent landscape around each sampling site. To extract the proportion of the different land cover types surrounding the sampling site we used satellite land cover maps from reference year 2019 [39]. These maps have a spatial resolution of 10m and include the following six main categories: 1) Forest and seminatural areas, 2) Open areas, 3) Arable land, 4) Wetlands, 5) Artificial surfaces and 6) Inland and marine water. These main categories are further divided into subcategories with detailed information regarding the dif- ferent land cover classes (S1 Table). Based on the biological and ecological requirements for I. ricinus and to identify landscape risk factors we decided to use the main categories in the anal- yses, with the exception of Forest and seminatural areas where we included eight individual forest types: Pine forest, Spruce forest, Mixed coniferous forest, Mixed forest, Broadleaved forest, Broadleaved hardwood forest, Broadleaved forest with hardwood forest and Temporarily non- forest (S1 Table). Landscape configuration To calculate landscape configuration metrics at landscape level surrounding each sampling site, we used land cover data from GIS buffers with a radius of 1000m, exported to GeoTIFF format. Buffer zones with a radius of 1000m were used to examine the landscape effects in the adjacent landscape around each sampling site and to include the home ranges of preferred hosts for I. ricinus [40]. Landscape configuration metrics for all 47 sampling sites were esti- mated with FRAGSTATS version 4 [41]. For landscape heterogeneity we used Shannons’ diver- sity index (SHDI), estimating the diversity of all included land cover classes at each sampling site weighted by their proportional coverage. Contagion (CONTAG) of land cover types was used to measure the aggregation of landscape attributes that can influence the suitability of sites for different tick host species. To analyze the influence of forest configuration on tick abundance, we collapsed all forest types into one main forest category. As measures of forest configuration, we used percent of forest cover (PLAND) and total forest edge length (TE). Statistical methods All statistical analyses were performed with R version 4.0.3 [42]. To analyze the effect of possi- ble risk factors for tick abundance in different greenspaces across the natural-urban gradient, we used generalized linear mixed models assuming Poisson distributed residuals. As the data contained a larger proportion of zeros than would be expected according to a Poisson or a neg- ative binomial distribution causing overdispersion [43, 44], we fitted zero-inflated Poisson models using the package glmmTMB (generalized linear mixed models using Template Model Builder) [45]. The glmmTMB package allows for handling of the correlative structure among sampling units within sites and has two main parts: (1) a conditional model that reports the coefficients of the Poisson count model, and (2) a zero-inflated model that reports the probability of a fixed effect resulting in the observation of an extra zero that is not generated by the conditional model [45]. The zero-inflation probability is bounded between zero and one by using a logit link. The zi-formula is a binomial model turned upside down, meaning that a positive coeffi- cient indicates a higher chance of absence. The zero-inflated models deal with the common problem that the mechanisms determining presence/absence data can be different from those that determine abundance. The conditional models incorporate the hierarchical structure that exists in different landscapes by allowing for site specific land cover relationships. The site specificity in the models also accounts for the spatial correlation that may exist in the relation- ship between ticks and their environment as well as for the occurrence of a larger number of PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 5 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance zeros than the standard Poisson distribution can incorporate. Since tick abundance was expected to be affected by factors measured at multiple spatial scales, our models include both sampling plot conditions, local site factors and large landscape characteristics. In the full models, tick abundance was analyzed as a function of the proportion of each land cover type. Larvae and nymphs were analyzed in two separate models due to the clus- tered occurrence of larvae throughout the landscape, and to the large difference in impor- tance as disease vector. Due to very low n-value adult ticks were excluded from the statistical analyses. In both models, the fixed factors Pine forest, Spruce forest, Mixed coniferous forest, Mixed forest, Broadleaved forest, Broadleaved hardwood forest, Broadleaved forest with broadleaved hardwood forest, Temporarily not forest, and the main categories Open areas, Arable land, Wetlands, Artificial surfaces, and Inland and Marine waters, together with the local site conditions, Vegetation height, Tree stem density and Temperature were included. The second order polynomial of Month indicating sampling time (sampling Day caused con- vergence issues) was added to correct for the irregular sampling times. In addition, sampling site was used as a random variable to adjust for the difference in sampling effort between sites, and to handle the correlative structure among plots within the same sampling site in the conditional part of the model. Model selection was performed by stepwise backward model reduction of fixed factors [43] using p-values and Akaike´s Information Criterion (AIC) [46]. The second order polynomial of sampling Month was retained in all models even if the linear and quadratic terms were nonsignificant in the final model. The same pro- cedure was applied separately to the conditional component and to the zero-inflated compo- nent of the model. The same modeling procedure was repeated for each buffer zone size in 100m increments ranging from 100m–1000m radius. Tick abundance was also analyzed as a function of landscape configuration in generalized lin- ear mixed zero-inflated models with the degree of Urbanization, and the landscape configura- tion variables PLAND, TE, CONTAG, and SHDI as fixed landscape scale factors and Vegetation height, Tree stem density and Temperature as fixed local scale factors. We also included the sec- ond order polynomial of Month as a temporal fixed effect, and study Site as random factor. We again made separate models for larvae and for nymphs and excluded adult ticks from the analy- ses. Model selection was performed similarly as for the models analyzing the effects of land cover types. However, in the landscape configuration models, data was only analyzed for buffer zones with 1000m radius. Since TE will be zero when forest cover is 0 or 100% and peak at inter- mediate forest levels, it was difficult to disentangle the forest edge effect in different greenspaces. To account for this, we also included a two-way interaction between PLAND and TE. All models were tested for multicollinearity using the variance inflation factor (VIF) [47, 48]. To provide an indication of the final model’s goodness-of-fit, we estimated the variance explained by the fixed factors (marginal R2) and the variation explained by both the fixed and random factors (conditional R2). Model fit was assessed by the residual distribution of each model using package DHARMa [49]. Results Ticks of the species Ixodes ricinus were present in 41 of the 47 inventoried greenspaces. In total, we collected 1378 ticks, including 992 larvae, 370 nymphs, 13 adult females and 3 adult males. The tick life stage capture ratio for larvae:nymphs:adults was 100:31:1.6. Adult ticks were very few and therefore excluded from the statistical analyses. High VIF values indicating multicollinearity resulted in removal of the landscape factors Pine forest, Temporarily non-for- est, and Open areas in the 800m radius buffer for nymph abundance, and the landscape config- uration factor CONTAG was removed from the analyses of landscape configuration effects for PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 6 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance nymph abundance. Conditional R2 was higher than marginal R2 (Tables 1 and 2), indicating that a substantial part of the explained variance of presence and abundance is due to differ- ences among sampling sites not accounted for by the fixed factors. The conditional and the zero-inflated model parts provided slightly different patterns for the fixed factors that were responsible for I. ricinus presence and abundance, respectively. In general, sampling plot fac- tors were most important for tick abundance, especially the negative effect from Vegetation height (Tables 1–3). Landscape characteristics affected both tick presence and abundance, with negative effects of Artificial surfaces and mixed effects from both open land and forests Table 1. The effect of local scale factors and landscape variables on Ixodes ricinus larvae measured in buffer zones consisting of 100m increments within a 100m– 1000m radius. Presence and abundance of I. ricinus larvae were analyzed in generalized linear mixed zero-inflated models. Estimated coefficients and significance levels for the explanatory variables from both the conditional model part (abundance) and the zero-inflated model part (presence/absence) is reported for each buffer zone size. For easier interpretation we have changed the sign of the estimates for the zero-inflated model part meaning that a positive coefficient should be interpreted as a higher likelihood for tick presence. Local site Surrounding landscape 100m 200m 300m 400m 500m 600m 700m 800m 900m 1000m -0.12* 0.10* 0.12 -0.11 -0.03* 0.12 -0.10 -0.03* 0.16* -0.12 -0.03* -0.03* -0.03* -0.03* -0.05* 0.49 0.48 -0.19 -0.03* -0.20 -0.03* 0.19* -0.14 -0.03*** -0.14* 0.26* -0.13 -0.03* 0.26 0.13 0.22* 0.24* 0.37 0.38* -0.09* -0.13* 0.14*** 0.08* 0.06* 0.37* 0.01 0.31 -0.17* -0.15* 0.08* 0.07* -0.28* 0.29* 0.20 -0.17* -0.15 -0.49 1.07* 0.30* 0.35* -0.28* 0.27 -0.15 47 0.236 0.566 295 47 0.195 0.548 295 47 0.041 0.534 295 47 0.319 0.442 295 47 0.275 0.411 295 47 0.024 0.501 295 47 0.267 0.629 295 0.13 0.05* - 0.98 -0.68 47 0.289 0.455 295 47 0.395 0.635 295 47 0.320 0.545 295 Fixed factors Conditional part (abundance) Spruce forest Mixed coniferous forest Mixed forest Broadleaved hardwood forest Open land Artificial surfaces Vegetation heighta Zero-inflated part (presence/absence) Spruce forest Mixed forest Broadleaved forest Broadleaved hardwood forest Broadleaved forest with broadleaved hardwood forest Temporarily non-forest Wetland Open land Artificial surfaces Water Tree stem densitya Temperature Month Month2 Random factor Site n Marginal R2 Conditional R2 Plot n Significance levels: * P<0.05; P<0.01; * P<0.00 a Local plot factor https://doi.org/10.1371/journal.pone.0285841.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 7 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance Table 2. The effect of local scale and landscape scale variables on Ixodes ricinus nymphs measured in buffer zones consisting of 100m increments within a 100m– 1000m radius. Presence and abundance of I. ricinus nymphs were analyzed in generalized linear mixed zero-inflated models. Estimated coefficients and significance levels for the explanatory variables from both the conditional model part (abundance) and the zero-inflated model part (presence/absence) is reported for each buffer zone size. For easier interpretation we have changed the sign of the estimates for the zero-inflated models meaning that a positive coefficient should be interpreted as a higher likeli- hood for tick presence. Local site Surrounding landscape 100m 200m 300m 400m 500m 600m 700m 800m 900m 1000m Fixed factors Conditional part (abundance) Spruce forest Mixed coniferous forest Broadleaved hardwood forest -0.10 0.08 -0.07* 0.10*** -0.07* 0.08** -0.07* 0.07* 0.10* 0.10* Broadleaved forest with broadleaved hardwood forest -0.17* -0.26* -0.01 -0.05 -0.01 -0.01* -0.08* -0.01* -0.01 -0.01* -0.01 -0.01 -0.01* -0.01 Artificial surfaces Vegetation heighta Zero-inflation part (presence/absence) Pine forest Spruce forest Mixed coniferous Broadleaved forest Broadleaved hardwood forest 0.12** Broadleaved forest with broadleaved hardwood forest -0.20* Temporarily non-forest -0.06* -0.06 -0.10* 47 0.114 0.573 295 47 0.194 0.577 295 47 0.228 0.517 295 47 0.356 0.580 295 Open land Water Artificial surfaces Temperature Month Month2 Random factor Site n Marginal R2 Conditional R2 Plot n Significance levels: * P<0.05; P<0.01; * P<0.00 a Local plot factor https://doi.org/10.1371/journal.pone.0285841.t002 -0.04* -0.07* -0.23* -0.07* -0.18* -0.08* -0.18* -0.12** -0.14* 0.16* 0.51* -0.66* - -0.26* 0.18 -0.11* -0.21* 10.28 -0.73 47 0.261 0.554 295 0.11* -0.20* 0.10 -0.12* -0.13 -0.16 -0.26* 47 0.120 0.538 295 47 0.087 0.512 295 47 0.035 0.537 295 47 0.043 0.526 295 5.69* -0.41* 47 0.131 0.192 295 depending on tick life stage and type of forest (Tables 1–3). In addition, there was a significant curvilinear effect of the factor Month with a peak in late summer for abundance of both larvae and nymphs (Tables 1 and 2). Sampling plot factors The local site factor Vegetation heigh had significant negative effects on both larva and nymph abundance, indicating that we have fewer questing I. ricinus in plots where the field layer vege- tation is high (Fig 2). Ninety percent of all the ticks in this study were collected in vegetation heights of 30cm or lower. Tree stem density surrounding sampling plots had a significant PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 8 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance Table 3. The effect of urbanization and landscape configuration on questing Ixodes ricinus presence and abun- dance measured in 1000m radius buffer zones analyzed in generalized linear mixed zero-inflated models. Esti- mated coefficients and significance levels for the explanatory variables from both the conditional (abundance) and zero-inflated (presence/absence) model components are reported for I. ricinus larvae and nymphs. Parameter estimates and significance levels for landscape factors are estimated based on a 1000m radius buffer zone and for local site factors from individual sampling plots. For easier interpretation we have changed the sign of the estimates for the zero-inflated models meaning that a positive coefficient should be interpreted as a higher likelihood for tick presence and vice versa. Larvae - - 0.06* 0.05* 47 0.167 0.564 295 Nymphs -0.01 -0.10* - - 47 0.040 0.535 295 Fixed factors Conditional part (abundance) Vegetation heighta Zero-inflated part (presence/absence) Urbanization Tree stem densitya PLAND Random factor Sites n Marginal R2 Conditional R2 Plots n Significance levels: P<0.05; * P<0.01; *** P<0.00 a Local plot factor https://doi.org/10.1371/journal.pone.0285841.t003 Fig 2. Effect of vegetation height on number of Ixodes ricinus larvae (A) and nymphs (B). The results from the glmmTMB show significant negative effects of field layer vegetation height with more questing I. ricinus in lower vegetation. Grey area indicates 95% confidence band. https://doi.org/10.1371/journal.pone.0285841.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 9 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance positive effect on the presence of I. ricinus larvae, but no significant effect on the presence of nymphs (Tables 1–3). Temperature had a weak significant negative effect on the presence of both larvae and nymphs (Tables 1 and 2). In our study ticks were actively questing in tempera- tures ranging from 3˚C– 27˚ but most larvae and nymphs (95%) were collected at tempera- tures between 15˚C– 24˚C. Local site factors At local scale, increasing proportions of Open land at the sampling sites had a significant negative effect on both larva and nymph presence (Tables 1 and 2). There was also a signifi- cant negative effect of the habitat factor Spruce forest on larva presence (Table 1). A high pro- portion of Broadleaved hardwood forest had a significant positive effect on nymph presence, and Broadleaved forest stands had a significant negative effect on nymph abundance (Table 2). Surrounding landscape characteristics Our results show significant landscape effects on both presence and abundance of I. ricinus. Large proportions of monocultures of Spruce forests and Pine forests in the largest buffer zones had a significant negative effect on nymph presence, and a significant negative effect on nymph abundance in intermediate buffer zones (Tables 1 and 2). On the other hand, forest stands with Mixed coniferous forest in intermediate and large buffer zones showed significant positive effects on nymph abundance (Tables 1 and 2). The effects from all types of coniferous forests in the landscape on larva distribution did not show any obvious patterns. Presence and abundance of I. ricinus larvae were instead significantly positively affected by Broadleaved forest and Broadleaved hardwood forest (Tables 1 and 2). The effects of broad- leaved forest types on nymphs were varied or lacking. It is important to point out that the over- all proportion of broadleaved forest in Stockholm County is low, still the significant effect on larvae is evident. Temporarily non-forested habitats showed a significant positive effect on larva presence. The proportion of Open land in the landscape had no effect on nymphs (Table 2) but a sig- nificant negative effect on the presence of larvae (Table 1). However, in sampling plots where larvae were present the proportion of Open land in the landscape had a significant positive effect on larva abundance (Table 1). The proportion of Water in intermediate sized buffer zones had a significant positive effect both on larva and nymph presence (Tables 1 and 2). Finally, Artificial surfaces had a significant negative effect on larva abundance (Table 1), and nymph presence (Table 2) in almost all buffer zone sizes. Urbanization and landscape configuration The effects of landscape configuration and urbanization on tick abundance were estimated from the 1000m radius buffer zones surrounding the different sampling sites. Urbanization had a significant negative effect on the presence of nymphs (Table 3). However, even if fewer ticks were found in more urbanized areas, ticks were present also in areas with 30–40% urbani- zation (Fig 3). The percent forest in the landscape (PLAND) had significant positive effects on the presence of larvae (Table 3). Neither the landscape configuration metrics SHDI, CONTAG, TE nor the two-way interaction between PLAND and TE had significant effects on tick abun- dance or tick presence. A simplified summary of the main result from all analyses focusing on the consistent pat- terns displayed by the effects of fixed local and landscape factors is shown in Table 4. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 10 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance Fig 3. The degree of urbanization and Ixodes ricinus abundance at each sampling site in Stockholm County. https://doi.org/10.1371/journal.pone.0285841.g003 Discussion To analyze factors responsible for Ixodes ricinus distribution, we investigated the effects from sampling plot characteristics together with local sampling site factors and effects from the sur- rounding landscape at 47 different sites along the natural-urban gradient in Stockholm County, Sweden. One of the main findings is that sampling plot features mainly are important for variation in tick abundance among sampling sites, while landscape features are important PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 11 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance Table 4. Summary of main results from Tables 1–3. Factors responsible for Ixodes ricinus presence and abundance in greenspaces across the urbanization gradient. Sampling plot conditions include recordings at each site and local site factors include land cover types in the smallest GIS buffer zone (100m radius). Landscape features include land cover types (200-1000m radius) and landscape configuration variables (1000m radius). Fixed factors Sampling plot conditions Vegetation height Tree stem density Local site factors Coniferous forest (monoculture) Open land Landscape features Artificial surfaces Coniferous forest (monoculture) Mixed coniferous forest Broadleaved hardwood forest Urbanization PLAND Larvae Negative Positive Negative Negative Negative Negative - Positive Negative Positive Nymphs Negative - - Negative Negative Negative Positive Positive Negative - https://doi.org/10.1371/journal.pone.0285841.t004 for the occupancy pattern of I. ricinus across sampling sites, but also for tick abundance at sites were I. ricinus is present. Local plot level Local microhabitat characteristics are important for tick survival and reproduction [50–53]. Ticks are sensitive to desiccation and need to hide from the sun during warm weather con- ditions [54]. Most of the ticks collected during our study were found in vegetation lower than 30cm, confirming earlier findings showing that vegetation height has a significant neg- ative effect on tick presence. While this could be a technical field sampling effect, the mop- ping technique we used in our study allowed us to move the blanket both above, beneath, and in between plants, even in high and dense field layers, reducing the risk of missing ticks in hard-to-reach field layers. Interestingly, the sampling plots with the highest tick abun- dance were found in forests with almost no field vegetation, and only a thin layer of leaf litter. I. ricinus mainly move vertically up and down straws and stems in the field layer, risking desiccation if trapped high up in the vegetation [55]. Previous studies have shown that ticks are questing close to the top of the vegetation, preferably at heights between 25cm– 50cm. Veg- etation taller than 75cm is used less for questing, probably due to the desiccation risk [56]. An alternative hypothesis is that high vegetation is a perfect habitat for finding hosts and that plots with high vegetation are already emptied of ticks by large mammal hosts before we conduct our sampling. The latter is a less likely but not impossible scenario and will require experimen- tal studies to be resolved. However, the high abundance of I. ricinus in low vegetation increases the risk for tick infestation of humans because we often avoid tall grass and shrubs and instead prefer areas and paths with low vegetation height that are easier to traverse when we are out hiking. Previous studies consistently found that I. ricinus appear to reduce their questing activity in response to extreme daily temperatures [57–59]. In our study, most ticks (95%) were collected in temperatures between 15º C to 24º C, indicating that this temperature range is beneficial for questing, especially for larvae. However, ticks were actively questing from 3º C to 27º C, PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 12 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance throughout the study period. The statistical analysis did not reveal any consistent temperature effect, though we rarely encountered extreme temperatures during sampling. The main prob- lem during tick sampling is rain or extremely damp conditions, therefore no sampling was conducted during days with long periods of rain. In this study, I. ricinus larvae were often collected in clusters. After being hatched from aggregations of eggs, larvae tend to be found in a clumped distribution. Considering life cycle stages, tick larvae are assumed to be most sensitive to local site conditions, and have the highest mortality rate [60]. This pattern may apply for I. ricinus in many environments, but it was not supported by the tick life stage capture ratio for larvae:nymphs:adults in the present study. However, our study was not optimized to measure the life stage capture ratio, as we used a standardized sampling strategy designed to compare differences among different sites and habitats. Nymphs and adults are known to quest close to the location where individual ticks dropped off from their previous host [61]. Nymphs and adults were mostly found in a scat- tered pattern which likely represents their host-based distribution pattern. Local site factors Land cover types in the smallest buffer zone (100m) represent the local habitat surrounding the inventoried plots. Sampling sites with a large proportion of open land in the smallest buffer zone had significantly lower presence of both larvae and nymphs. We interpret this as a direct negative effect of open land on both tick presence due to drier microclimatic conditions and tick host presence. Previous studies found that trees shadowing the sampling plots may lower the risk of desiccation for I. ricinus [21], corresponding with our findings of significant positive effects of tree stem density around the sampling plots on the presence of larvae. The positive effect of tree stem density in the local habitat surrounding each plot may have been counter- acted by the significant negative effects of dense monocultures of spruce forest or pine forest. More favorable local site conditions for ticks seemed to be present at sampling sites located in mixed coniferous forests or in broad leaved hardwood forests. Coniferous forest is a very com- mon forest type around Stockholm especially at the outer edges of the nature-urban gradient. Broadleaved hardwood forests are rare around Stockholm and more common in highly main- tained park environments in peri-urban and urban environments. Landscape level In this study, greenspaces with surrounding degree of urbanization ranging from 0%– 50% were included in the sampling. The results show significant negative effects of artificial surfaces and urbanization on tick abundance, with higher tick abundance in greenspaces surrounded by large natural and semi-natural habitats. However, we also found ticks in parks and small pockets of vegetation in highly urbanized areas. Previous research and our own observations confirm that urban greenspaces can provide favorable ecological conditions for both small and large vertebrate hosts [5, 6, 62]. This suggests that ticks repeatedly are being transported to these habitats by wildlife, and that ticks can have access to hosts capable of feeding all tick stages also in urban environments [8, 63–65]. In Stockholm County, large green wedges tra- verse the entire natural-urban sector like spokes in a wheel, leading from the rural rim to the city hub. The proportion of forest within these wedges is substantial in many areas and there are established populations of prime tick hosts such as rodents, birds, rabbits, hares, foxes, and roe deer even in the most densely human populated areas [66]. Variation in tick abundance between greenspaces differing in habitat characteristic and location has been well documented [25, 67, 68]. Forest patches affect the survival of ticks by creating a humid microhabitat as well as determining habitat suitability for hosts. Broadleaved PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 13 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance forests and mixed forests harboring a diverse fauna are generally considered ideal habitats for I. ricinus [69]. However, coniferous forests in areas with high rainfall and with a thick moist litter layer can also support high tick densities [70, 71]. The most important land cover type in the surrounding landscape promoting I. ricinus abundance in different green- spaces around Stockholm is mixed coniferous forest. I. ricinus was also present in landscapes with high proportion of broadleaved hardwood forests which is in agreement with previous studies [61, 67]. However, in Stockholm County broadleaved hardwood forest is extremely rare except for highly maintained parks and small remnants of the old farming landscape. The structures of the surrounding landscape influence I. ricinus abundance through varia- tion in abiotic conditions and tick-host dynamics [72]. Despite the importance of these dif- ferent land cover variables on tick abundance, site variation also played an important role, indicating that there are additional factors influencing I. ricinus abundance along the urban- ization gradient. In addition to landscape composition, landscape configuration may also affect tick abun- dance along the natural-urban gradient. Percent forest cover had a positive effect on larval presence, which is probably connected to the environmental conditions affecting density and activity of I. ricinus main hosts [56]. One of the forest configuration factors that we analyzed was the total length of the forest edge. Forest edges have been suggested as important tick habi- tats due to their suitability for many tick hosts [73]. In this study, we could not see any effect of forest edge, neither as an interaction factor with forest area nor as a main factor on its own. Similarly, there was no clear effect of forest aggregation on tick presence or abundance, which in contrast to earlier studies [25, 74]. The lack of forest configuration effect may be explained by the green wedges stretching all the way from rural to urban areas. Due to these wedges, the proportion of forest in buffer zones with a 1km radius is substantial in many areas of Stock- holm County, even close to the city center. However, even if the forest configuration and com- position does not change dramatically along the natural-urban gradient the proportion of artificial surfaces is very different. The natural-urban gradient represents many special ecological characters affecting both ticks and hosts [75]. The presence of animal hosts for all active stages and the ability to disperse within and between habitats are important for the life cycle of I. ricinus [76]. Nymphs, the medically most important tick stage, were found along the entire natural-urban gradient, but adult ticks were absent in highly urbanized greenspaces. Urbanized greenspaces surrounded by built environments typically have reduced diversity of wildlife, especially larger mammals on which adult ticks usually feed [6]. The few adult ticks sampled in this study were found in suburban forests and at these same sites, the other tick life stages were present as well. Subur- ban forests seem to have favorable conditions for all tick stages and options for finding suitable hosts for their blood meals. Future studies should include an assessment of the occurrence and density of wildlife at the different collection sites which might shed more light on the large var- iation in tick presence and abundance between our study sites. Factors responsible for Ixodes ricinus presence and abundance Our study demonstrates the presence of I. ricinus in greenspaces across the entire natural- urban gradient in Stockholm County, but with higher abundance of ticks in intermediate urbanized areas and large natural and seminatural greenspaces. Tree abundance surrounding each plot had a significant positive effect on the presence of larvae, and forest patches consist- ing of mixed coniferous forest or broadleaved hardwood forest in the surrounding landscape significantly increased both larval and nymphal presence and abundance. In many parts of Europe, climate change is altering the growth and development of forest types, with a decline PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 14 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance of coniferous forests and an increase in mixed and broad-leaved forests. A change in forest composition and structure might increase the proportion of habits suitable for ticks in this region, especially larvae, which might consequently influence the epidemiology of tick-borne diseases [21]. Higher tick abundance will increase the risk for human tick encounters. However, the range of vegetation types where we found ticks included greenspaces with very low vegetation height, and highly urbanized parks. This information is relevant to public health, since humans and their pets are more likely to spend more time, and make closer skin contact, with low vege- tation in parks or along trails than with tall rough vegetation in large and remote greenspaces [71]. In a polycentric city like Stockholm with green wedges stretching along the urbanization gradient, ticks are present in a wide variety of greenspaces, even at sites often perceived to be free of tick hazards by the public. Conclusions This study investigated Ixodes ricinus presence and abundance in relation to patterns of urban- ization, and simultaneously considered small scale local environmental conditions, and large- scale landscape characteristics. Our systematic random sampling method can be used for com- parisons among sites within a study as well as with other studies using similar procedures. The main finding from our study is that even if there is a higher abundance of I. ricinus in natural and semi-natural areas compared to urban areas, ticks are still present at highly urbanized sites. We also found that I. ricinus, along the entire natural-urban gradient, is readily abundant in sites with very low field vegetation height. At landscape scale, we found that in the hemi- boreal forest zone around Stockholm in Sweden, mixed coniferous forest seems to be an important driver for I. ricinus. The presence of I. ricinus in urban areas and its preference for low vegetation height makes the probability for human tick encounters high also in urban areas. Based on our results, we recommend greenspaces in urban areas to be included in sur- veillance for ticks and tick-borne diseases. Supporting information S1 Table. Swedish land cover nomenclature including grid code, land cover type and defi- nition. (DOCX) Acknowledgments The authors thank the editor and the anonymous reviewers for their useful input. In addition, thanks to Kari Lethila¨ for help with statistical models and Raphaela Mayerhofer for help with editing the manuscript. Author Contributions Conceptualization: The´rese Janze´n, Monica Hammer, Mona Petersson, Patrik Dinne´tz. Data curation: The´rese Janze´n, Mona Petersson, Patrik Dinne´tz. Formal analysis: The´rese Janze´n, Mona Petersson, Patrik Dinne´tz. Funding acquisition: Monica Hammer, Mona Petersson, Patrik Dinne´tz. Investigation: The´rese Janze´n, Mona Petersson, Patrik Dinne´tz. Methodology: The´rese Janze´n, Patrik Dinne´tz. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 15 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance Project administration: The´rese Janze´n, Patrik Dinne´tz. Supervision: Monica Hammer, Mona Petersson, Patrik Dinne´tz. Visualization: The´rese Janze´n, Patrik Dinne´tz. Writing – original draft: The´rese Janze´n. Writing – review & editing: The´rese Janze´n, Monica Hammer, Mona Petersson, Patrik Dinne´tz. References 1. Vanwambeke SO, Linard C, Gilbert M. Emerging challenges of infectious diseases as a feature of land systems. Current Opinion in Environmental Sustainability. 2019 Jun 1; 38:31–6 2. Vignoli L, Scirè S, Bologna MA. Rural–urban gradient and land use in a millenary metropolis: how urbanization affects avian functional groups and the role of old villas in bird assemblage patterning. Web Ecology. 2013 Jul 8; 13(1):49–67. 3. Nagy DD, Magura T, Horva´ th R, Debna´r Z, To´ thme´re´sz B. Arthropod assemblages and functional responses along an urbanization gradient: a trait-based multi-taxa approach. Urban Forestry & Urban Greening. 2018 Mar 1; 30:157–68. 4. Sandifer PA, Sutton-Grier AE, Ward BP. Exploring connections among nature, biodiversity, ecosystem services, and human health and well-being: Opportunities to enhance health and biodiversity conserva- tion. Ecosystem services. 2015 Apr 1; 12:1–5. 5. C´ wik A, Wo´jcik T, Ziaja M, Wo´ jcik M, Kluska K, Kasprzyk I. Ecosystem services and disservices of veg- etation in recreational urban blue-green spaces—Some recommendations for greenery shaping. For- ests. 2021 Aug 12; 12(8):1077. 6. Rizzoli A, Silaghi C, Obiegala A, Rudolf I, Huba´ lek Z, Fo¨ ldva´ri G, et al. Ixodes ricinus and its transmitted pathogens in urban and peri-urban areas in Europe: new hazards and relevance for public health. Fron- tiers in public health. 2014 Dec 1; 2:251. 7. Mackenstedt U, Jenkins D, Romig T. The role of wildlife in the transmission of parasitic zoonoses in peri-urban and urban areas. International Journal for Parasitology: Parasites and Wildlife. 2015 Apr 1; 4 (1):71–9. https://doi.org/10.1016/j.ijppaw.2015.01.006 PMID: 25830108 8. Uspensky I. Tick pests and vectors (Acari: Ixodoidea) in European towns: Introduction, persistence and management. Ticks and Tick-borne Diseases. 2014 Feb 1; 5(1):41–7. https://doi.org/10.1016/j.ttbdis. 2013.07.011 PMID: 24183576 9. Hansford KM, Fonville M, Gillingham EL, Coipan EC, Pietzsch ME, Krawczyk AI, et al. Ticks and Borre- lia in urban and peri-urban green space habitats in a city in southern England. Ticks and tick-borne dis- eases. 2017 Mar 1; 8(3):353–61. 10. Borşan SD, Toma-Naic A, Pe´ter A´ , Sa´ndor AD, Peștean C, Mihalca AD. Impact of abiotic factors, habitat type and urban wildlife on the ecology of hard ticks (Acari: Ixodidae) in urban and peri-urban habitats. Par- asites & Vectors. 2020 Dec; 13(1):1–7. https://doi.org/10.1186/s13071-020-04352-3 PMID: 32948235 11. Sormunen JJ, Klemola T, Vesterinen EJ, Vuorinen I, Hyto¨ nen J, Ha¨ nninen J, et al. Assessing the abun- dance, seasonal questing activity, and Borrelia and tick-borne encephalitis virus (TBEV) prevalence of Ixodes ricinus ticks in a Lyme borreliosis endemic area in Southwest Finland. Ticks and tick-borne dis- eases. 2016 Feb 1; 7(1):208–15. 12. Mysterud A, Hu¨ gli C, Viljugrein H. Tick infestation on medium–large-sized mammalian hosts: are all equally suitable to Ixodes ricinus adults?. Parasites & vectors. 2021 Dec; 14(1):1–6. 13. Krawczyk AI, Ro¨ttjers L, Fonville M, Takumi K, Takken W, Faust K, et al. Quantitative microbial popula- tion study reveals geographical differences in bacterial symbionts of Ixodes ricinus. Microbiome. 2022 Aug 4; 10(1):120. 14. Wilhelmsson P, Lindblom P, Fryland L, Nyman D, Jaenson TG, Forsberg P, et al. Ixodes ricinus ticks removed from humans in Northern Europe: seasonal pattern of infestation, attachment sites and dura- tion of feeding. Parasites & vectors. 2013 Dec; 6(1):1-1.Dec;17(1):1–4. 15. Kalma´ r Z, Dumitrache MO, D’Amico G, Matei IA, Ionică AM, Gherman CM, et al. Multiple tick-borne pathogens in Ixodes ricinus ticks collected from humans in Romania. Pathogens. 2020 May 19; 9 (5):390. https://doi.org/10.3390/pathogens9050390 PMID: 32438768 16. Daniel M, Maly´ M, Danielova´ V, Křı´zˇ B, Nuttall P. Abiotic predictors and annual seasonal dynamics of Ixodes ricinus, the major disease vector of Central Europe. Parasites & Vectors. 2015 Dec; 8(1):1–2. https://doi.org/10.1186/s13071-015-1092-y PMID: 26385470 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 16 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance 17. Tomkins JL, Aungier J, Hazel W, Gilbert L. Towards an evolutionary understanding of questing behav- iour in the tick Ixodes ricinus. PLoS One. 2014 Oct 15; 9(10):e110028. 18. Boehnke D, Gebhardt R, Petney T, Norra S. On the complexity of measuring forests microclimate and interpreting its relevance in habitat ecology: the example of Ixodes ricinus ticks. Parasites & vectors. 2017 Dec; 10:1–4. 19. Medlock JM, Pietzsch ME, Rice NV, Jones L, Kerrod E, Avenell D, et al. Investigation of ecological and environmental determinants for the presence of questing Ixodes ricinus (Acari: Ixodidae) on Gower, South Wales. Journal of Medical Entomology. 2014 Oct 28; 45(2):314–25. 20. Van Overbeek L, Gassner F, Van Der Plas CL, Kastelein P, Nunes–da Rocha U, Takken W. Diversity of Ixodes ricinus tick-associated bacterial communities from different forests. FEMS microbiology ecology. 2008 Oct 1; 66(1):72–84. 21. Tack W, Madder M, Baeten L, De Frenne P, Verheyen K. The abundance of Ixodes ricinus ticks depends on tree species composition and shrub cover. Parasitology. 2012 Sep; 139(10):1273–81. 22. Gilbert L, Brunker K, Lande U, Klingen I, Grøva L. Environmental risk factors for Ixodes ricinus ticks and their infestation on lambs in a changing ecosystem: Implications for tick control and the impact of wood- land encroachment on tick-borne disease in livestock. Agriculture, Ecosystems & Environment. 2017 Jan 16; 237:265–73. 23. Ehrmann S, Liira J, Ga¨rtner S, Hansen K, Brunet J, Cousins SA, et al. Environmental drivers of Ixodes rici- nus abundance in forest fragments of rural European landscapes. BMC ecology. 2017 Dec; 17(1):1–4. 24. Estrada-Peña A. The relationships between habitat topology, critical scales of connectivity and tick abundance Ixodes ricinus in a heterogeneous landscape in northern Spain. Ecography. 2003 Oct; 26 (5):661–71. 25. Perez G, Bastian S, Agoulon A, Bouju A, Durand A, Faille F, et al. Effect of landscape features on the relationship between Ixodes ricinus ticks and their small mammal hosts. Parasites & vectors. 2016 Dec; 9(1):1–8. 26. Sonenshine DE, Roe RM, editors. Biology of ticks volume 2. Oxford University Press; 2013. 27. Piedmonte NP, Shaw SB, Prusinski MA, Fierke MK. Landscape features associated with blacklegged tick (Acari: Ixodidae) density and tick-borne pathogen prevalence at multiple spatial scales in central New York state. Journal of medical entomology. 2018 Oct 25; 55(6):1496–508. https://doi.org/10.1093/ jme/tjy111 PMID: 30020499 28. Heylen D, Lasters R, Adriaensen F, Fonville M, Sprong H, Matthysen E. Ticks and tick-borne diseases in the city: Role of landscape connectivity and green space characteristics in a metropolitan area. Sci- ence of the total environment. 2019 Jun 20; 670:941–9. https://doi.org/10.1016/j.scitotenv.2019.03.235 PMID: 30921726 29. Mo¨rtberg UM. Resident bird species in urban forest remnants; landscape and habitat perspectives. Landscape ecology. 2001 Apr; 16:193–203. 30. Furberg D, Ban Y, Mo¨ rtberg U. Monitoring urban green infrastructure changes and impact on habitat connectivity using high-resolution satellite data. Remote Sensing. 2020 Sep 19; 12(18):3072. 31. Swedish Environmental Protection Agency. Svenska Markta¨ ckedata. Edition 2. 2 [Internet]. 2014. https://gpt.vic-metria.nu/data/land/smd_produktbeskrivning_20140325.pdf 32. Gray JS, Dautel H, Estrada-Peña A, Kahl O, Lindgren E. Effects of climate change on ticks and tick- borne diseases in Europe. Interdisciplinary perspectives on infectious diseases. 2009 Jan 4; 2009. https://doi.org/10.1155/2009/593232 PMID: 19277106 33. Mejlon HA, Jaenson TG. Seasonal prevalence of Borrelia burgdorferi in Ixodes ricinus in different vegetation types in Sweden. Scandinavian journal of infectious diseases. 1993 Jan 1; 25(4):449– 56. 34. Asghar N, Petersson M, Johansson M, Dinnetz P. Local landscape effects on population dynamics of Ixodes ricinus. Geospatial health. 2016 Nov 21; 11(3). 35. Tack W, Madder M, De Frenne P, Vanhellemont M, Gruwez R, Verheyen K. The effects of sampling method and vegetation type on the estimated abundance of Ixodes ricinus ticks in forests. Experimental and Applied Acarology. 2011 Jul; 54:285–92. 36. Bitterlich W. The relascope idea. Relative measurements in forestry. Commonwealth Agricultural Bureaux; 1984. 37. Baldwin H, Landesman WJ, Borgmann-Winter B, Allen D. A geographic information system approach to map tick exposure risk at a scale for public health intervention. Journal of Medical Entomology. 2022 Jan; 59(1):162–72. https://doi.org/10.1093/jme/tjab169 PMID: 34642748 38. Jackson HB, Fahrig L. Are ecologists conducting research at the optimal scale?. Global Ecology and Biogeography. 2015 Jan; 24(1):52–63. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 17 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance 39. Swedish Environmental Protection Agency. Svenska Markta¨ ckedata. Edition 2. 2 [Internet]. 2019. https://www.naturvardsverket.se/4a43ca/contentassets/37e8b38528774982b5840554f02a1f81/ produktbeskrivning-nmd-2018-basskikt-v2-2.pdf 40. Li S, Heyman P, Cochez C, Simons L, Vanwambeke SO. A multi-level analysis of the relationship between environmental factors and questing Ixodes ricinus dynamics in Belgium. Parasites & Vectors. 2012 Dec; 5:1–1. 41. McGarigal K. FRAGSTATS: spatial pattern analysis program for quantifying landscape structure. US Department of Agriculture, Forest Service, Pacific Northwest Research Station; 1995. 42. R Core Team (2021). R: A language and environment for statistical computing. R Foundation for Statis- tical Computing, Vienna, Austria. https://www.R-project.org/. 43. Zuur AF, Ieno EN, Walker N, Saveliev AA, Smith GM, Zuur AF, et al. Zero-truncated and zero-inflated models for count data. Mixed effects models and extensions in ecology with R. 2009:261–93. 44. Bouyer Y, Rigot T, Panzacchi M, Van Moorter B, Poncin P, Beudels-Jamar R, et al. Using zero-inflated models to predict the relative distribution and abundance of roe deer over very large spatial scales. In Annales Zoologici Fennici 2015 Apr (Vol. 52, No. 1–2, pp. 66–76). Finnish Zoological and Botanical Publishing Board. 45. Brooks ME, Kristensen K, Van Benthem KJ, Magnusson A, Berg CW, Nielsen A, et al. glmmTMB bal- ances speed and flexibility among packages for zero-inflated generalized linear mixed modeling. The R journal. 2017; 9(2):378–400. 46. Steyerberg EW, Eijkemans MJ, Habbema JD. Stepwise selection in small data sets: a simulation study of bias in logistic regression analysis. Journal of clinical epidemiology. 1999 Oct 1; 52(10):935–42. https://doi.org/10.1016/s0895-4356(99)00103-1 PMID: 10513756 47. Alin A. Multicollinearity. Wiley interdisciplinary reviews: computational statistics. 2010 May; 2(3):370–4. 48. O’brien RM. A caution regarding rules of thumb for variance inflation factors. Quality & quantity. 2007 Oct; 41:673–90. 49. Hartig F. DHARMa: residual diagnostics for hierarchical (multi-level/mixed) regression models. R pack- age. 2020. https://cran.r-project.org/package=DHARMa. Version 0.4.6 50. Nelson C, Banks S, Jeffries CL, Walker T, Logan JG. Tick abundances in South London parks and the potential risk for Lyme borreliosis to the general public. Medical and veterinary entomology. 2015 Dec; 29(4):448–52. https://doi.org/10.1111/mve.12137 PMID: 26400641 51. Oechslin CP, Heutschi D, Lenz N, Tischhauser W, Pe´ ter O, Rais O, et al. Prevalence of tick-borne path- ogens in questing Ixodes ricinus ticks in urban and suburban areas of Switzerland. Parasites & Vectors. 2017 Dec; 10:1–8. 52. Mathews-Martin L, Namèche M, Vourc’h G, Gasser S, Lebert I, Poux V, et al. Questing tick abundance in urban and peri-urban parks in the French city of Lyon. Parasites & Vectors. 2020 Dec; 13:1–9. https:// doi.org/10.1186/s13071-020-04451-1 PMID: 33183354 53. Gray J, Kahl O, Zintl A. What do we still need to know about Ixodes ricinus?. Ticks and Tick-borne Dis- eases. 2021 May 1; 12(3):101682. 54. Qviller L, Grøva L, Viljugrein H, Klingen I, Mysterud A. Temporal pattern of questing tick Ixodes ricinus density at differing elevations in the coastal region of western Norway. Parasites & vectors. 2014 Dec; 7:1–2. 55. Randolph SE, Storey K. Impact of microclimate on immature tick-rodent host interactions (Acari: Ixodi- dae): implications for parasite transmission. Journal of medical entomology. 1999 Nov 1; 36(6):741–8. https://doi.org/10.1093/jmedent/36.6.741 PMID: 10593075 56. Mejlon HA, Jaenson TG. Questing behaviour of Ixodes ricinus ticks (Acari: Ixodidae). Experimental & applied acarology. 1997 Dec; 21:747–54. 57. Gilbert L, Aungier J, Tomkins JL. Climate of origin affects tick (Ixodes ricinus) host-seeking behavior in response to temperature: implications for resilience to climate change?. Ecology and evolution. 2014 Apr; 4(7):1186–98. 58. Dautel H, Ka¨ mmer D, Kahl O. How an extreme weather spell in winter can influence vector tick abun- dance and tick-borne disease incidence. Ecology and prevention of Lyme borreliosis 2016 Nov 30 (p. 362). Wageningen Academic Publishers. 59. Kiewra D, Kryza M, Szymanowski M. Influence of selected meteorological variables on the questing activity of Ixodes ricinus ticks in Lower Silesia, SW Poland. Journal of Vector Ecology. 2014 Jun; 39 (1):138–45. 60. Leal B, Zamora E, Fuentes A, Thomas DB, Dearth RK. Questing by tick larvae (Acari: Ixodidae): a review of the influences that affect off-host survival. Annals of the Entomological Society of America. 2020 Nov; 113(6):425–38. https://doi.org/10.1093/aesa/saaa013 PMID: 33244354 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 18 / 19 PLOS ONEFactors responsible for Ixodes ricinus presence and abundance 61. Carroll JF, Schmidtmann ET. Dispersal of blacklegged tick (Acari: Ixodidae) nymphs and adults at the woods–pasture interface. Journal of Medical Entomology. 1996 Jul 1; 33(4):554–8. https://doi.org/10. 1093/jmedent/33.4.554 PMID: 8699448 62. Pfa¨ ffle M, Littwin N, Muders SV, Petney TN. The ecology of tick-borne diseases. International journal for parasitology. 2013 Nov 1; 43(12–13):1059–77. https://doi.org/10.1016/j.ijpara.2013.06.009 PMID: 23911308 63. Welc-Falęciak R, Kowalec M, Karbowiak G, Bajer A, Behnke JM, Siński E. Rickettsiaceae and Anaplas- mataceae infections in Ixodes ricinus ticks from urban and natural forested areas of Poland. Parasites & Vectors. 2014 Dec; 7:1–3. 64. Kilpatrick AM, Dobson AD, Levi T, Salkeld DJ, Swei A, Ginsberg HS, et al. Lyme disease ecology in a changing world: consensus, uncertainty and critical gaps for improving control. Philosophical Transac- tions of the Royal Society B: Biological Sciences. 2017 Jun 5; 372(1722):20160117. 65. Kowalec M, Szewczyk T, Welc-Falęciak R, Siński E, Karbowiak G, Bajer A. Ticks and the city-are there any differences between city parks and natural forests in terms of tick abundance and prevalence of spi- rochaetes?. Parasites & Vectors. 2017 Dec; 10(1):1–9. https://doi.org/10.1186/s13071-017-2391-2 PMID: 29157278 66. Danell K, Bergstro¨m R. Mer vilt i dag a¨n fo¨r 50 år sedan. Vilt och fisk fakta. 2009. [cited 2021]. https:// pub.epsilon.slu.se/14111/1/bro_2005270_vildsvin_nr409.pdf 67. Estrada-Peñ A A. Distribution, abundance, and habitat preferences of Ixodes ricinus (Acari: Ixodidae) in northern Spain. Journal of Medical Entomology. 2001 May 1; 38(3):361–70. https://doi.org/10.1603/ 0022-2585-38.3.361 PMID: 11372959 68. Van Gestel M, Verheyen K, Matthysen E, Heylen D. Danger on the track? Tick densities near recreation infrastructures in forests. Urban Forestry & Urban Greening. 2021 Apr 1; 59:126994. 69. Jaenson TG, Eisen L, Comstedt P, Mejlon HA, Lindgren E, Bergstro¨m S, et al. Risk indicators for the tick Ixodes ricinus and Borrelia burgdorferi sensu lato in Sweden. Medical and veterinary entomology. 2009 Sep; 23(3):226–37. 70. Gray JS. Review the ecology of ticks transmitting Lyme borreliosis. Experimental & applied acarology. 1998 May; 22:249–58. 71. Walker AR, Alberdi MP, Urquhart KA, Rose H. Risk factors in habitats of the tick Ixodes ricinus influenc- ing human exposure to Ehrlichia phagocytophila bacteria. Medical and Veterinary Entomology. 2001 Mar; 15(1):40–9. 72. Cayol C, Koskela E, Mappes T, Siukkola A, Kallio ER. Temporal dynamics of the tick Ixodes ricinus in northern Europe: epidemiological implications. Parasites & vectors. 2017 Dec; 10:1–1. https://doi.org/ 10.1186/s13071-017-2112-x PMID: 28359294 73. Jones EO, Webb SD, Ruiz-Fons FJ, Albon S, Gilbert L. The effect of landscape heterogeneity and host movement on a tick-borne pathogen. Theoretical Ecology. 2011 Nov; 4:435–48. 74. Ehrmann S. Biodiversity and ecosystem (dis) services of small forest patches in European agricultural landscapes, with emphasis on Ixodes ricinus and Borrelia burgdorferi sensu lato (Doctoral dissertation, Albert-Ludwigs-Universita¨t Freiburg im Breisgau). 75. Rosà R, Andreo V, Tagliapietra V, Bara´ kova´ I, Arnoldi D, Hauffe HC, et al. Effect of climate and land use on the spatio-temporal variability of tick-borne bacteria in Europe. International journal of environ- mental research and public health. 2018 Apr; 15(4):732. https://doi.org/10.3390/ijerph15040732 PMID: 29649132 76. Medlock JM, Hansford KM, Bormane A, Derdakova M, Estrada-Peña A, George JC, et al. Driving forces for changes in geographical distribution of Ixodes ricinus ticks in Europe. Parasites & vectors. 2013 Dec; 6:1–1. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285841 May 17, 2023 19 / 19 PLOS ONE
10.1371_journal.ppat.1011421	RESEARCH ARTICLE Airway proteolytic control of pneumococcal competence Haley Echlin, Amy Iverson, Ugo Sardo, Jason W. RoschID* Department of Infectious Diseases, St Jude Children’s Research Hospital, Memphis, Tennessee, United States of America * Jason.Rosch@stjude.org Abstract Streptococcus pneumoniae is an opportunistic pathogen that colonizes the upper respira- tory tract asymptomatically and, upon invasion, can lead to severe diseases including otitis media, sinusitis, meningitis, bacteremia, and pneumonia. One of the first lines of defense against pneumococcal invasive disease is inflammation, including the recruitment of neutro- phils to the site of infection. The invasive pneumococcus can be cleared through the action of serine proteases generated by neutrophils. It is less clear how serine proteases impact non-invasive pneumococcal colonization, which is the key first step to invasion and trans- mission. One significant aspect of pneumococcal biology and adaptation in the respiratory tract is its natural competence, which is triggered by a small peptide CSP. In this study, we investigate if serine proteases are capable of degrading CSP and the impact this has on pneumococcal competence. We found that CSP has several potential sites for trypsin-like serine protease degradation and that there were preferential cleavage sites recognized by the proteases. Digestion of CSP with two different trypsin-like serine proteases dramatically reduced competence in a dose-dependent manner. Incubation of CSP with mouse lung homogenate also reduced recombination frequency of the pneumococcus. These ex vivo experiments suggested that serine proteases in the lower respiratory tract reduce pneumo- coccal competence. This was subsequently confirmed measuring in vivo recombination fre- quencies after induction of protease production via poly (I:C) stimulation and via co-infection with influenza A virus, which dramatically lowered recombination events. These data shed light on a new mechanism by which the host can modulate pneumococcal behavior and genetic exchange via direct degradation of the competence signaling peptide. Author summary Streptococcus pneumoniae is a major human pathogen that usually colonizes the nasophar- ynx asymptomatically, but can progress to disease causing meningitis, bacteremia, otitis media, and sinusitis. In response to invasive pneumococci, the host produces an inflam- matory response, including the influx of serine proteases. Invasive S. pneumoniae can be cleared through the action of serine proteases, however little is known on how serine pro- teases impact S. pneumoniae processes during colonization. One key modulator of a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Echlin H, Iverson A, Sardo U, Rosch JW (2023) Airway proteolytic control of pneumococcal competence. PLoS Pathog 19(5): e1011421. https://doi.org/10.1371/journal.ppat.1011421 Editor: Carlos J. Orihuela, The University of Alabama at Birmingham, UNITED STATES Received: July 7, 2022 Accepted: May 11, 2023 Published: May 31, 2023 Copyright: © 2023 Echlin et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All data are in the manuscript and/or supporting information files. Funding: JWR is supported by 1R01AI155614, 1R01AI171038, 1R56AI155614, 1U19AI158076, 1U01AI124302 from the Division of Microbiology and Infectious Disease, National Institute of Allergy and Infectious Diseases and by ALSAC. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 1 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence pneumococcal survival in the host environment is its innate competence to uptake DNA. Here, we investigate how serine proteases impact pneumococcal competence through modulation of a key signaling peptide, CSP. We show that two different recombinant ser- ine proteases degrade CSPs and this degradation has a direct impact on pneumococcal competence. We further demonstrate a reduction in pneumococcal competence upon CSP incubation with mouse lung homogenates, which suggests that serine proteases pro- duced in mouse lungs similarly degrade CSP. These data reveal a novel mechanism by which the host can modulate pneumococcal behavior and genetic exchange, including that required for the spread of antibiotic resistance, a growing threat as the number of cases involving multi-drug resistant pneumococci have increased worldwide. Introduction Streptococcus pneumoniae is a major human pathogen causing a multitude of diseases ranging from sinusitis and otitis media to pneumonia, bacteremia, and meningitis [1]. While it is a considerable source of morbidity and mortality primarily in young children and the elderly populations, S. pneumoniae usually colonizes the upper respiratory tract asymptomatically in up to 65% of children and <10% of adults [2,3]. The balance of interactions between the pneu- mococcus and the host can impact the likelihood of transition from commensal to pathogen and the severity of the resulting disease [4]. The hallmark of pneumococcal invasive disease is inflammation and often heralds the influx of neutrophils, which are among the first cells recruited to the site of infection [5,6]. Two prominent mechanisms of neutrophil clearance are oxidative burst and production of antimicrobial molecules including serine proteases [7,8]. Serine proteases hydrolyze peptide bonds via the nucleophilic serine in the enzyme active site. Serine proteases are further classified by their substrate specificity, e.g., trypsin-like serine pro- teases prefer Arg and Lys residues [9–11]. Because of its antiphagocytic capsule, S. pneumoniae is less likely to be killed by neutrophils via oxidative burst and is instead cleared by serine pro- teases (such as elastase, cathepsin G, and protease 3) found in the azurophilic granules of neu- trophils [7]. Besides direct degradation of pathogen proteins, neutrophilic serine proteases also play a role in activation of antimicrobials, including LL-37 [12]. While the impact of serine proteases produced in response to invasive disease has been studied considerably, the impact of host serine proteases on pneumococcal carriage and pro- gression to disease in the respiratory tract is not as well understood. One group of serine prote- ases in the respiratory tract are the human airway trypsin-like proteases (HAT) which are part of the type II transmembrane family of serine proteases (TTSPs), the largest group of mem- brane-anchored serine proteases [13]. Originally identified in the sputum of patients with chronic airway diseases, HAT has been more recently shown to be found specifically on cili- ated epithelial cells, rather than the basal cells, of the epithelial layer in the respiratory tract, suggesting a role within the epithelial layer and on the airway surface [13–15]. This is of partic- ular interest in the context of pneumococcal interactions as asymptomatic colonization of S. pneumoniae requires adherence to the non-inflamed epithelial cells of the respiratory tract [2,16]. Once adhered to the respiratory tract surface, the pneumococcus must often compete with host factors (including serine proteases), changing environmental conditions, and other colonizing bacterial species [17]. Gaining insight into how the complex environment of the upper and lower respiratory tract impacts the strategies deployed by the pneumococcus is criti- cal for understanding pneumococcal biology at the host-pathogen interface. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 2 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence A powerful mechanism that S. pneumoniae utilizes to adapt to its changing environment is natural competence, a physiological state in which it undergoes active uptake of exogenous DNA [18,19]. Pneumococcal natural transformation is triggered via a quorum-sensing cascade induced at low population density of exponentially growing cells [20]. The competence cas- cade is triggered by a small pheromone—competence-stimulating peptide (CSP), which is encoded by comC and then processed and secreted as a mature 17-residue peptide. CSP acti- vates the two-component regulatory system ComD (membrane receptor) and ComE (response regulator), which directly activates early competence genes including the comAB and comCDE operons and comX. ComAB is the membrane transporter that exports CSP to trigger cascades in other cells. Activation of comCDE provides a positive feedback loop to amplify the compe- tence signal, while activation of comX promotes activation of late competence genes including those for DNA uptake [21–25]. Besides genes directly involved in the competence cascade, more than 100 genes are regulated upon activation of competence, including the genes involved in fratricide, in biofilm formation, and in virulence [26–32]. The complexity of how the pneumococcus adapts to changes in nutrients and competing cells, while thriving in the host, is not completely understood. However, it is imperative for us to understand the dynam- ics of this adaptation. Healthy children carrying S. pneumoniae could be colonized simulta- neous by up to six different pneumococcal strains, occurring in up to 65% of young children [33–37]. This provides a high chance for acquisition of new genes, including those of antibiotic resistance. This is of particular concern as there has been a worldwide increase in the number of cases involving multi-drug resistant pneumococci [38]. In our current study, we investigate if the signaling peptide, CSP, can be degraded by tryp- sin-like serine proteases and how this impacts the recombination frequency of Streptococcus pneumoniae. We found that CSP was degraded by two trypsin-like serine proteases, including HAT. This digestion of CSP had a direct, dose-dependent impact on pneumococcal compe- tence, whereby increased concentrations of trypsin-like serine proteases decreased recombina- tion frequency. Modification of the peptide sequence of CSP1 suggested arginine at position nine to be a key modulator for the impact on recombination frequency. We also recapitulated the sensitivity of CSP to digestion by trypsin-like serine proteases in vivo. Taken together, these data shed light on a novel mechanism by which the host environment can modulate pneumococcal adaptation to its niche in the respiratory tract. Results CSPs are cleaved by trypsin-like serine proteases Different strains of S. pneumoniae regulate their competence via competence stimulating pep- tides with strain-specific peptide sequences [39]. To investigate if CSPs could be cleaved by trypsin-like serine proteases, we analyzed the protein sequence of CSP1 and CSP2 [19] for potential cleavage sites. CSP1 contains five and CSP2 contains four potential trypsin cleavage sites (Fig 1A) after arginine or lysine [9]. To determine which sites are cleaved, CSP1 and CSP2 were incubated with either porcine trypsin or human airway trypsin-like protease (HAT; originally purified from patients with chronic airway disease). Mass spectrometry of the digested CSP1 demonstrated that CSP1 was cleaved by porcine trypsin at residues 6, 9, and 15 (Fig 1B) and by HAT at residues 3, 9, and 15 (Fig 1C). Interestingly, HAT digestion of CSP1 preferentially occurred at R9, while porcine trypsin digestion occurred primarily at R15 fol- lowed by K6 and R9. Like CSP1, CSP2 was cleaved by both porcine trypsin (Fig 1D) and HAT (Fig 1E). Unlike CSP1, porcine trypsin and HAT targeted a single site for cleavage—R6 for porcine trypsin and R3 for HAT. These data indicate that pneumococcal CSPs are both recog- nized and cleaved by serine proteases. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 3 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence Fig 1. Serine proteases cleave competence-stimulating peptide of S. pneumoniae. (A) Sequence of competence-stimulating peptide 1 (CSP1) and peptide 2 (CSP2). Red arrows indicate potential cleavage sites by trypsin-like serine proteases. (B) Mass spectrometry analysis of cleaved residues of CSP1 upon incubation with no trypsin (grey) or with porcine trypsin (black) for 30 minutes. (C) Mass spectrometry analysis of cleaved residues of CSP1 upon incubation with no trypsin (grey) or with human airway trypsin (HAT) (black) for 30 minutes. (D) Mass spectrometry analysis of cleaved residues of CSP2 upon incubation with no trypsin (grey) or with porcine trypsin (black) for 30 minutes. (E) Mass spectrometry analysis of cleaved residues of CSP2 upon incubation with no trypsin (grey) or with HAT (black) for 30 minutes. (B-E) The fraction of each peptide detected by mass spectrometry was calculated by dividing the area of the cleaved peptide by the total area of all peptides. In the case of two cleavage sites on the same peptide, the fraction was equally attributed to both cleavage sites. Each bar represents the fraction of identified peptides that contained a cleavage event after the amino acid depicted under the bar. Bars above the final amino acid represent uncleaved CSP1 or CSP2. https://doi.org/10.1371/journal.ppat.1011421.g001 Cleavage of CSPs by trypsin-like serine proteases reduces S. pneumoniae competence As the CSPs were cleaved by trypsin-like serine proteases, we next ascertained if this cleavage impacts S. pneumoniae competence. CSP1 or CSP2 was incubated with increasing concentra- tions of trypsin-like proteases, porcine pancreatic trypsin (PPT) and HAT. The CSP1-trypsin mixture was used to transform D39x with genomic DNA from D39 Tn-seq library and suc- cessful recombinants were determined. The frequency of recombination was reduced by incu- bation of CSP1 with PPT (Fig 2A) and with HAT (Fig 2B) in a dose-dependent manner. To confirm that the observed reduced recombination frequency was due to trypsin cleavage of CSP1, the trypsin-like proteases were incubated with a serine protease inhibitor AEBSF prior to incubation with CSP1. Transformation with CSP1 incubated with inhibited PPT or HAT PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 4 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence Fig 2. Serine proteases reduce recombination frequency of S. pneumoniae in a dose dependent manner. Recombination frequency upon incubation of CSP1 or CSP2 with increasing concentrations of serine proteases with and without 0.1mM of the serine protease inhibitor, AEBSF. D39x transformed with CSP1 incubated with (A) porcine pancreatic trypsin (PPT) or (B) human airway trypsin (HAT). D39x comC- transformed with CSP1 incubated with (C) PPT or (D) HAT. TIGR4 transformed with CSP2 incubated with (E) PPT or (F) HAT. TIGR4 comC- transformed with CSP2 incubated with (G) PPT or (H) HAT. Negative controls included no addition of CSP1 and no addition of gDNA. Lines represent median value. Dotted line represents lowest point of detection. Recombination frequencies of increasing concentrations of protease within each group (0 mM AEBSF or 0.1 mM AEBSF) were compared using Kruskal-Wallis one-way ANOVA; p = 0.05–0.01, p = 0.01–0.001, p = 0.001–0.0001. https://doi.org/10.1371/journal.ppat.1011421.g002 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 5 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence demonstrated similar recombination frequency compared to incubation with no trypsin, sug- gesting that the reduced competence was due to digestion of CSP1 by the serine proteases. In this experimental setup, only exogenously added CSP1 would be exposed to trypsin-like proteases directly. Thus, any natively produced CSP1 could still induce competence. The pres- ence of native CSP1 could explain why little difference was observed between 2.5 μg/mL and 10 ug/mL of PPT and why this recombination frequency was similar to the frequency of the control with no exogenous CSP1 added. To prevent potential induction of competence by native CSP1, a comC deletion strain that can no longer produce native CSP1 was generated, whereby transformation of this mutant relied exclusively on exogenously added peptide [25]. Recombination frequency of D39x comC- was reduced when transformed with CSP1 incu- bated with PPT (Fig 2C) or HAT (Fig 2D). Similar to D39x, the frequency was reduced in a dose-dependent manner; however, the level of reduction was more dramatic in the comC- mutant compared to the wild-type (Fig 2). Recombination frequency of D39x comC- trans- formed with CSP1 incubated with PPT or HAT pre-incubated with inhibitor AEBSF was at a similar level as that of the no trypsin control. To determine if cleavage of CSP2 impacts S. pneumoniae competence, CSP2 was incubated with increasing concentrations of trypsin-like proteases and this mixture was used to trans- form TIGR4 with genomic DNA from TIGR4 Tn-seq library. Like CSP1, the recombination frequency was reduced by incubation of CSP2 with PPT (Fig 2E) and, to a lesser extent, with HAT (Fig 2F) in a dose-dependent manner and this reduction was abrogated by incubating the trypsin-like proteases with inhibitor AEBSF. As observed in D39x, this loss in recombina- tion frequency upon incubation with serine proteases was more pronounced in the TIGR4 comC- strain that lacked the ability to produce native CSP2 (Fig 2G and 2H). To further confirm loss of competence upon CSP1 cleavage by trypsin-like proteases, CSP1 incubated with PPT or HAT was used to induce competence in DLA3, a strain derived from D39, and luminescence was measured (Fig 3). DLA3 contains a firefly luciferase reporter under the promoter of ssbB and activity of this reporter has been demonstrated to reflect com- petence and provides a way to monitor competence development in real time [40–42]. Lumi- nescence of DLA3 was reduced in a dose-dependent manner upon incubation of CSP1 with PPT (Fig 3A) and this effect was abrogated if PPT was pre-incubated with AEBSF (Fig 3B). Incubation of CSP1 with HAT similarly reduced luminescence of DLA3 in a dose-dependent manner (Fig 3C and 3D). This loss of luminescence was not due to loss of growth (S1 Fig). Taken together, these results indicate that CSPs are cleaved by both PPT and HAT and this has a direct impact on the competence of S. pneumoniae. Preferential site of CSP cleavage is dependent on protease As PPT and HAT did not cleave equally at all potential cleavage sites (Fig 1), we investigated whether cleavage at some cleavage sites may have more impact on competence than others. Modified CSPs were synthesized that have altered cleavage sites at R3, K6, R9, and R15 whereby the trypsin-like proteases should no longer recognize the site for cleavage. To ensure that the modified CSP could still induce competence, the recombination frequency of D39x comC- transformed with these modified CSP was assessed. There was no or slight reduction in recom- bination frequency for K6H, R9H, and R15H (Fig 4A). However, modification of R3 resulted in complete loss of competence, as previously observed [43]. To ascertain if modification of CSP alters the impact of trypsin-like protease on competence, the recombination frequency of D39x comC- transformed with CSP1 or its modified variants incubated with PPT (Figs 4B and S2) or HAT (Figs 4C and S2) was determined. For both PPT and HAT, incubation of K6H and R15H with trypsin-like proteases reduced recombination frequency similarly to that of unmodified PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 6 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence Fig 3. Serine proteases reduce expression of CSP induced luciferase in S. pneumoniae. Luminescence (RLU) of DLA3 grown in the presence of CSP1 incubated with increasing concentrations of serine proteases with and without inhibitor AEBSF. Incubation of CSP1 with PPT (A) without AEBSF or (B) with AEBSF. Incubation of CSP1 with HAT (C) without AEBSF or (D) with AEBSF. Experiment was repeated in triplicate. The mean value of RLU of each 30-minute timepoint is reported; error bars are SEM. Luminescence of increasing concentrations of protease within each group (0 mM AEBSF or 0.1 mM AEBSF) were compared using two-way ANOVA; p<0.0001. https://doi.org/10.1371/journal.ppat.1011421.g003 CSP1. Incubation of R9H with trypsin-like proteases reduced recombination frequency of D39x comC- to a greater extent than the unmodified CSP1—in that lower concentrations of PPT or HAT were sufficient to reduce recombination frequency. For PPT, the concentration to reduce the recombination frequency to below 5% of 0 trypsin was 5 times less compared to unmodified CSP1 (0.01 vs 0.05 μg/mL). For HAT, the concentration was 2 times less compared to unmodi- fied CSP1 (0.5 vs 1 μg/mL). To determine if the profile of cleavage sites was altered upon R9 modification, the modified CSP was incubated with serine-like proteases and the peptide profile was analyzed via mass spectrometry. Upon cleavage with PPT, the proportion of cleaved pep- tides after K6 increased, while cleavage after R15 remained the same (Fig 5A). Digestion with HAT shifted the cleavage site from primarily R9 to cleave mainly at R3, followed by R15 (Fig 5B). Taken together, these data suggest that different trypsin-like serine proteases have diverse preferential cleavage sites and the cleavage profile can alter recombination frequency. Mouse serine proteases reduce S. pneumoniae competence To assess if the reduced recombination frequency upon incubation of CSP1 with recombinant trypsin-like proteases could be recapitulated with native airway serine proteases, we PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 7 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence Fig 4. Modified CSP alters impact of protease on recombination frequency. Recombination frequency with modified CSP1. (A) Transformation of D39x comC- with CSP1 with modifications R3H, K6H, R9H, and R15H. Transformation of D39x comC- with modified CSP1 incubated with increasing concentrations of (B) PPT or (C) HAT. Recombination frequency reported as % of 0 trypsin. Negative controls included no addition of CSP1 and no addition of gDNA. Lines represent mean value. Recombination frequencies of increasing concentrations of protease within each modified CSP were compared using one-way ANOVA; p = 0.001–0.0001, *p<0.0001. Changes in the concentrations of protease between modified CSP was compared using two-way ANOVA; ##p = 0.01–0.001, ###p = 0.001–0.0001. https://doi.org/10.1371/journal.ppat.1011421.g004 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 8 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence Fig 5. Modification of R9 of CSP1 alters protease digestion profile. Mass spectrometry analysis of cleaved residues of CSP1 (black) or CSP1 R9A (striped) upon incubation with (A) porcine trypsin or with (B) HAT. The fraction of each peptide detected by mass spectrometry was calculated by dividing the area of the cleaved peptide by the total area of all peptides. In the case of two cleavage sites on the same peptide, the fraction was equally attributed to both cleavage sites. Each bar represents the fraction of identified peptides that contained a cleavage event after the amino acid depicted under the bar. Bars above the final amino acid represent uncleaved CSP1. https://doi.org/10.1371/journal.ppat.1011421.g005 determined the recombination frequency of S. pneumoniae upon incubation of CSP1 with mouse lungs ex vivo (Fig 6). Lung homogenates were incubated with increasing concentrations of AEBSF to inhibit native serine proteases prior to incubation with CSP1. Incubation of the lung homogenates with AEBSF increased the recombination frequency of D39x (Fig 6A). This result suggests that the proteases in the lung homogenates reduced recombination frequency and that inhibition of these proteases allowed for increased recombination. Indeed, levels of serine proteases in the lung homogenates were decreased upon incubation with AEBSF (Fig 6B) and there was a strong correlation between the recombination frequency and the level of protease (Fig 6C). A similar result was observed when D39x comC- was transformed with CSP1 incubated with lung homogenates (Fig 6D–6F). Of note, AEBSF itself did not increase recombination frequency (S3 Fig). PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 9 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence Fig 6. Inhibition of proteases from mouse lungs increase recombination frequency of S. pneumoniae. Recombination frequency and protease levels upon incubation of CSP1 with homogenized mouse lungs with increasing concentrations of inhibitor AEBSF. (A) Recombination frequency of D39x transformed with CSP1 incubated with homogenized mouse lungs. (B) Protease levels in homogenized mouse lungs upon incubation with AEBSF used in D39x transformation; determined by fluorescence of substrate t-Butyloxycarbonyl Phe-Ser-Arg 7-amino-4methyl coumarin (BOC). (C) Correlation of recombination frequency of D39x with the protease levels in the same homogenized mouse lung. (D) Recombination frequency of D39x comC- transformed with CSP1 incubated with homogenized mouse lung. (E) Protease levels in homogenized mouse lungs upon incubation with AEBSF used in D39x comC- transformations; determined by fluorescence of substrate BOC. (F) Correlation of recombination frequency of D39x comC- with the protease levels in the same homogenized mouse lung. (A,D) Line represents median; dotted line represents lowest point of detection; recombination frequency of 1 mM and 2 mM AEBSF were compared to 0 mM AEBSF using Kruskal-Wallis one-way ANOVA; p<0.0001. (B,E) Lines represent mean; protease levels of 1 mM and 2 mM AEBSF were compared to 0 mM AEBSF using one-way ANOVA. (C,F) Correlation was compared using two-tailed spearman; * p = 0.0004, **** p<0.0001. https://doi.org/10.1371/journal.ppat.1011421.g006 To confirm that the proteases from mouse lungs could reduce S. pneumoniae competence, we repeated the ex vivo recombination frequency using lungs from mice that were adminis- tered poly (I:C) and AEBSF in vivo. Poly (I:C) is a synthetic analog of double-stranded RNA that simulates a viral infection and induces the inflammatory response via TLR3 signaling and upregulates serine protease activities [44,45]. Lungs were harvested from mice that were administered poly (I:C) and AEBSF or the corresponding controls intranasally. The homoge- nized lungs were then incubated with 0, 1, or 2 mM AEBSF, incubated with CSP1, and the lung-AEBSF-CSP mixture was used to transform D39x and D39x comC-. D39x recombination frequency was reduced when incubated with the homogenized lungs of mice that were previ- ously administered the stimulant poly (I:C) compared to that of control mice. The frequency was restored when the poly (I:C) stimulated mice were administered the protease inhibitor AEBSF IN along with the poly (I:C) (Fig 7A). For all groups, transformation with lungs from mice that were incubated with AEBSF ex vivo increased the recombination frequency com- pared to the lungs homogenates that were incubated with 0mM AEBSF (Fig 7B). This result was similarly observed in the recombination frequency of D39x comC- (Fig 7C and 7D). These PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 10 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence Fig 7. Stimulation of proteases in vivo reduces ex vivo recombination frequency of S. pneumoniae. Recombination frequency upon incubation of CSP1 with homogenized lungs from mice that were administered poly (I:C) in vivo, treated with inhibitor AEBSF in vivo, and then treated ex vivo with increasing concentrations of inhibitor AEBSF. (A) Recombination frequency of D39x transformed with CSP1 incubated with homogenized lungs from mice that received no stimulant (water) or poly (I:C), and either received no inhibitor (PBS) or inhibitor AEBSF. (B) The same lungs were then treated with either 0, 1, or 2 mM AEBSF ex vivo prior to incubation with CSP1 and recombination frequency was determined. The 0 mM AEBSF are the same data used in Fig 7A and were included here for comparison. (C) Recombination frequency of D39x comC- transformed with CSP1 incubated with homogenized lungs from mice that received no stimulant (water) or poly (I:C), and either received no inhibitor (PBS) or inhibitor AEBSF. (D) The same lungs were then treated with either 0, 1, or 2 mM AEBSF ex vivo prior to incubation with CSP1 and recombination frequency was determined. The 0 mM AEBSF are the same data used in Fig 7C and were included here for comparison. Line represents median; dotted line represents lowest point of detection. (A,C) Recombination frequencies were compared pairwise using nonparametric Mann-Whitney t test; p = 0.05–0.01, p = 0.01–0.001. (B,D) Recombination frequency of 1 mM and 2 mM AEBSF were compared to 0 mM AEBSF of each group using Kruskal-Wallis one-way ANOVA; p = 0.05–0.01, p = 0.01–0.001, p = 0.001–0.0001, *p<0.0001. https://doi.org/10.1371/journal.ppat.1011421.g007 data further indicate that the proteases in the murine lung can reduce S. pneumoniae recombi- nation frequency. To further investigate if host serine proteases could reduce S. pneumoniae competence in vivo, mice were infected with two strains of D39 that each harbor a different resistance cassette. If competence is induced, the strains would become resistant to both antibiotics upon recom- bination. To stimulate production of serine proteases in the host, mice were administered inflammatory stimulant, poly (I:C), and serine protease inhibitor, AEBSF, prior to bacterial challenge. A significant reduction in the recombination frequency of D39 in both the lungs and blood of mice administered poly (I:C) was observed compared to the vehicle control (Fig 8A and 8B). Protease levels in the lungs and sera from mice administered poly (I:C) were PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 11 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence Fig 8. Stimulation of protease production in vivo reduces recombination frequency of S. pneumoniae. Recombination frequency of S. pneumoniae from (A) the lungs and (B) the blood of mice that received no stimulant (water) or poly (I:C), and either received no inhibitor (PBS) or inhibitor AEBSF. Protease levels in the same mouse lungs (C) and blood (sera) (D); determined by fluorescence of substrate BOC. Correlation of recombination frequency in lungs (E) and blood (F) with the protease levels in the same tissue. (A,B) Line represents median; dotted line represents lowest point of detection; recombination frequencies of all groups were compared pairwise for each tissue using nonparametric Mann-Whitney t test; p = 0.01–0.001, *p = 0.001–0.0001. (C,D) Lines represent mean; protease levels of all groups were compared pairwise for each tissue using unpaired t test; p = 0.01–0.001, **p<0.0001. (E,F) Correlation was compared using two-tailed spearman; p = 0.004, *p = 0.0001. https://doi.org/10.1371/journal.ppat.1011421.g008 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 12 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence higher than those from the control mice (Fig 8C and 8D). In mice that were treated with the AEBSF inhibitor prior to poly (I:C) stimulation, recombination occurred at a higher frequency than in mice that were administered poly (I:C) but did not receive the inhibitor (Fig 8A and 8B) but was not completely restored to the frequency observed in mice not administered poly (I:C) (Fig 8A and 8B). The protease levels were reduced in the lungs from mice that were treated with the AEBSF inhibitor prior to poly (I:C) stimulation compared to the lungs from mice that were administered poly (I:C) but did not receive the inhibitor (Fig 8C and 8D). However, this reduced level of protease was still greatly elevated compared to the levels in the lungs from the mice that did not receive poly (I:C). This likely explains why the recombina- tion frequency in the mice administered AEBSF inhibitor prior to poly (I:C) stimulation was not restored to the same frequency of the control mice. The incomplete rescue of the AEBSF inhibitor of the poly (I:C) stimulation is likely an artifact of the in vivo model used as the poly (I:C) greatly increases the proteases levels and the dosage and in vivo pharmaco- kinetics of the inhibitor is not sufficient in this model to reduce the levels to baseline. Over- all, there was a significant correlation between the recombination frequency and the level of protease in the lungs and blood (Fig 8E and 8F). These data suggest that the native mouse proteases in the lung and blood can reduce S. pneumoniae recombination frequency in vivo. Influenza virus infection has been shown to induce inflammation and protease production in the respiratory tract and can synergize with S. pneumoniae secondary infections resulting in exacerbated morbidity and mortality [46–50]. To investigate how coinfection could impact S. pneumoniae recombination frequency, mice were infected with influenza virus A prior to bac- terial challenge. A significant reduction in the recombination frequency of D39 in both the lungs and blood of flu-infected mice was observed (Fig 9), further substantiating the results observed with the poly (I:C) stimulation using a live viral challenge. Taken together, these results indicate that mouse serine proteases can reduce S. pneumoniae competence. Discussion The innate ability of S. pneumoniae to recombine exogenous DNA provides a powerful mecha- nism to adapt to the host environment and this is dependent on CSP [25]. In this study, we have demonstrated that host serine proteases can degrade the pneumococcal competence stim- ulating peptide, CSP. This is supported by previous findings that demonstrate crude cell-free extracts were sensitive to trypsin digestion [20]. We also show that this degradation has a direct impact on pneumococcal recombination frequency. While the competence cascade has been studied extensively, pneumococcal competence is a complex community trait and the exact triggers and modulators for competence in the mammalian host is not as extensively delin- eated. As such, advancing our understanding of host and environmental factors that modulate competence remains an important area of investigation. As a human pathogen, it is possible that the pneumococcus has evolutionarily adapted to exploit the degradation of CSP by the host serine proteases in a manner to thrive in the host. This could explain why the pneumococcus would maintain such an advantageous peptide to be sensitive to proteolytic digestion of the host. As demonstrated here, the CSP sites sensitive to protease degradation can be altered without significant loss of competence, with the notable exception of R3. Many pathogens are known to hijack host serine proteases to promote patho- genesis, including viruses such as influenza and SARS-CoV coronavirus and bacteria such as Neisseria meningitidis and Staphylococcus aureus [11,51–53]. In particular, influenza virus has been shown to target the human airway trypsin-like protease [54], which we show here degrades CSP and reduces pneumococcal competence. Previous studies have demonstrated that S. pneumoniae is capable of responding to other host molecules such as LL-37 [55] and of PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 13 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence Fig 9. Co-infection with influenza in vivo reduces recombination frequency of S. pneumoniae. Recombination frequency of S. pneumoniae from (A) the lungs and (B) the blood of mice infected with and without influenza (Flu). The total number of recombinant colonies per mL used to calculate recombination frequency enumerated from (C) the lungs and (D) the blood of mice infected with and without influenza (Flu). Lines represent median. Dotted line represents lowest point of detection. Recombination frequency and total number of recombinants from mice infected with influenza was compared to that without influenza for each tissue using nonparametric Mann-Whitney t test; ** p<0.0001. https://doi.org/10.1371/journal.ppat.1011421.g009 exploiting host proteases, including plasmin and enolase, to promote adherence to epithelial cells of the respiratory tract and to facilitate colonization and invasion [56,57]. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 14 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence Fig 10. Proposed model of impact of host serine proteases on S. pneumoniae adaptation to host epithelium and invasion. Created with BioRender. com. https://doi.org/10.1371/journal.ppat.1011421.g010 Based on the observations from this study and on previous studies, we propose a framework whereby S. pneumoniae manipulates the host serine protease degradation of CSP to adapt to the dynamic host environment (Fig 10). When the pneumococcus first encounters the naso- pharyngeal epithelial cells, it interacts with multiple host factors during initial colonization. In this study, we were unable to determine the recombination frequency in the nasopharynx of mice as the total number of bacteria recovered from the nasopharyngeal tissue was too low to detect recombinants despite repeated attempts. This is in contrast to previous studies demon- strating efficient recombination during murine colonization, which may be partially explained by differences in the experimental design and resistance genes being modeled between differ- ent studies [32]. We utilized a murine model which focuses on pneumococcal infectivity in the lower respiratory tract as it permitted the use of poly (I:C) stimulant and AEBSF inhibitor, which have been previously investigated in an influenza challenge in the lung [45,58]. To our knowledge, there is no current model that demonstrates the pharmacokinetics of these two compounds in the upper respiratory tract. While demonstration of the impact of protease on recombination in the lower respiratory pathway only is a limitation of this study, previous studies have investigated the relationship between competence and colonization of the upper respiratory tract [59–64]. Deletion of the response regulator, comE, results in reduced coloni- zation and increased capsular production in adult mice [62,63]. Another investigation demon- strated that mutants of comAB, comD, and comE were able to colonize at levels similar to that PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 15 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence of the wild-type when inoculated individually in an infant rat model [64]. However, the comE mutant outcompetes the wild-type strain in a competition assay for colonization, while the comAB mutant demonstrates a reduced ability to colonize unless the competence cascade could be induced with CSP produced by co-infected wild-type cells [64]. It is possible that the differences observed could be related to immune status, bioavailability, or structural differ- ences of the niches in the in vivo models used. In the respiratory tract, as the bacterial cells aggregate and cell population increases, CSP is produced. However, the CSP would be accessi- ble to the serine proteases present in the mucosal layer of the epithelium and can be degraded. In the sputum of patients with inflammatory airway diseases, the concentration of HAT varied in the range of 5–40 nM (0.13 to 1.08 μg/mL) for HAT activity and 2–100 nM (0.05 to 2.7 μg/ mL) for the HAT antigen [13,14,65]. Previous reports have demonstrated influenza infection results in a significant increase in trypsin levels in murine lungs following challenge, with approximately twice of much protease being detected compared to uninfected controls [50]. In this study, we observed reduced recombination frequency in the murine lung upon stimula- tion of serine proteases (Fig 8). Using recombinant HAT to generate a standard for the prote- ase assay, we can equate 35000 RFU to be 0.05 μg/mL recombinant HAT and the protease levels to be 0.079 (±0.010) μg/mL in non-inflamed lungs and 0.168 (±0.012) μg/mL in inflamed lungs, which are similar to those observed in the sputum of patients. In the in vitro recombina- tion experiment, we observed reduced recombination frequency when CSP was incubated with at least 0.5 μg/mL HAT in D39x comC- and with at least 10 μg/mL HAT in D39x (Fig 2). The higher concentrations required to reduced recombination frequency in vitro compared to those observed in human sputum/ murine lungs could be reflective of differences in the HAT protease itself (recombinant vs native), the concentration of CSP used in vitro compared to those found in the host, the likely factor of additional host serine proteases that may also con- tribute to the observed levels of recombination frequency in vivo, and the potential variable distribution of proteolytic activity across the lung. These important caveats should be taken into consideration in the interpretation of the experimental data. In the host, the serine proteases could reduce levels of CSP and, as the threshold required to stimulate the competence cascade is not met [66,67], competence would be reduced. This could promote passage through the mucus layer via reduced capsule or dysregulation of other factors, allowing access to the epithelium [68,69]. Once through the mucus layer, interaction with the epithelium stimulates capsular shedding which can tighten adherence [55,70,71] and potentially enhances competence [72]. Cell density increases and biofilm formation is upregu- lated [73,74]. The biofilm protects the growing population from host immune defenses (such as proteases) [75,76] and pneumococcal cells adapt to their host environment, downregulating factors that can stimulate inflammation [77]. This loss of pressure from host serine proteases combined with the proximity of cells within the biofilm provides a golden opportunity for the pneumococcus to increase competence for genetic exchange to best survive in the host envi- ronment and increase fitness [31,32]. As the biofilm matures, host-adapted cells can be trig- gered by host signals to disperse and either reattach and begin the colonization process anew or invade. These host-adapted cells dispersed from the biofilm are more inflammatory and invasive since they disseminate into lungs/ middle ear more readily than either broth-grown bacteria or bacteria that are prematurely dispersed from a biofilm [74,78,79]. The dispersed cells have also been shown to downregulate genes involved in competence and colonization [80]. Inflammation induced by invasive cells promotes neutrophil recruitment and proteases production, which could further reduce pneumococcal competence. A similar model may be applied to the lower respiratory tract, where the production and response to CSP is required for pneumococcal survival and pathogenesis and biofilm-like structures may be present to modulate exposure of CSP to host proteases [26,43,59,81–83]. While biofilm and colonization PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 16 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence in the nasopharynx are of central importance for genetic exchange [32,36], the pneumococcus can remain competent in the lungs and blood [84] and, as shown here, recombination fre- quency can be reduced in the lungs of mice that have been administered stimulant poly (I:C) or co-infection with influenza (Figs 8 and 9). Prolonged colonization by multiple strains likely increases the probability of successful recombination events compared to the acute invasive model used in this study to interrogate recombination frequency, which is an important con- sideration when putting these findings in a biological context. This underscores the impor- tance of understanding pneumococcal recombination events not only in the upper respiratory, but in the lower respiratory as well—in particular, how the pneumococcus can adapt to host and antibiotic pressures during an infection. Pneumococcal competence is a complex community trait and is likely modulated by a mul- titude of factors in the mammalian host. In this study, we have demonstrated one mechanism by which the host can modulate recombination and how the pneumococcus may adapt to the host serine proteases. It is also possible that other factors besides host serine proteases impact recombination efficiency in the host, including those that induce oxygen exposure, SOS, and DNA replication stress, but the direct impact of these factors on recombination in the host has not been extensively delineated [41,85,86]. Another key consideration is the growth state of the pneumococcus in these tissues, as the transition from biofilm to planktonic growth can also play an important role in recombination frequency. The pneumococcus also encodes sev- eral serine proteases (PrtA, SFP, CbpG, HrtA) that are involved in competence, colonization, lung inflammation, and survival in the blood. In particular, HrtA plays an important role in maintaining redox balances and capsule production upon stimulation of the host inflamma- tory response, for example during an influenza co-infection [87–93]. It is unclear how these bacterial proteases interact, if at all, with host airway proteases or proteases encoded by co-col- onizing bacterial species. Nonetheless, these findings underscore the importance of host prote- ases in the modulation of pneumococcal competence and the complexity of the multi-niche interaction between S. pneumoniae and the host. These data indicate that host proteases can modulate pneumococcal competence through the degradation of a microbial signaling peptide by host proteases, suggesting conditions whereby repression or upregulation of such proteases may impact pneumococcal genetic exchange in opposing ways, underscoring the importance of the host environment on this critical facet of pneumococcal biology. Material and methods Ethics statement All experiments involving animals were performed with prior approval of and in accordance with guidelines of the St. Jude Institutional Animal Care and Use Committee. The St Jude lab- oratory animal facilities have been fully accredited by the American Association for Accredita- tion of Laboratory Animal Care. Laboratory animals were maintained in accordance with the applicable portions of the Animal Welfare Act and the guidelines. All mice were maintained in ABSL2 facilities and all experiments were done while the mice were under inhaled isoflurane (2.5%) anesthesia. Media and growth conditions Streptococcus pneumoniae was grown on tryptic soy agar (EMD Chemicals, NJ, USA) supple- mented with 3% defibrinated sheep blood and 20 μg/mL neomycin (TSA blood plates). Cul- tures were inoculated from newly streaked TSA blood plates into C+Y, a semi-synthetic casein liquid media with 0.5% yeast extract [94] and grown at 37˚C, 5% CO2. Strains used in this study are listed in Table 1. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 17 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence Table 1. Strains used in this study. Strain D39x D39x comC- TIGR4 TIGR4 comC- DLA3 D39 Tn-Seq TIGR4Δcps:: PhunSweet Description Bioluminescent D39 with Tn4001 luxABCDE; KanR Replacement of comC (Spd_2065) with PhunSweetErm cassette in D39x; KanR ErmR TIGR4 wild-type strain Replacement of comC (Sp_2237) with PhunSweetErm cassette in TIGR4; ErmR D39 with luciferase under the promoter of ssbB (PssbB-luc) Tn-Seq library generated in D39; SpecR Source of PhunSweet cassette TIGR4 Tn-Seq Tn-Seq library generated in TIGR4; SpecR TIGR4ΔspxB::Erm Source of erm cassette https://doi.org/10.1371/journal.ppat.1011421.t001 Source Francis, et al [95] This study This study Slager, et al [41] Matthews, et al [96] Echlin, et al [97] Van Opijnen, et al [98] Echlin, et al [99] Generation of comC- deletion mutant To delete comC, Spd_2065 was replaced with a modified PhunSweet cassette [97] via SOE PCR [100]. All PCR products were amplified using exTaq polymerase (TAKARA) following the rec- ommended guidelines and primers are listed in Table 2. gDNA was extracted using the aque- ous/organic extraction protocol, as described previously [97]. To utilize the PhunSweet cassette in the D39x background (which has kanamycin resistance), the kanamycin resistance kan in the PhunSweet cassette was replaced with erythromycin resistance erm, generating PhunSweetErm. The PhunSweet cassette (minus kan) was amplified from TIGR4Δcps::PhunS- weet gDNA using primers PhunSweet_F and PhunSweet-kan_R. The erythromycin resistance cassette erm was amplified from TIGR4ΔspxB::Erm [99] gDNA using primers Erm_F and Erm_R. PhunSweet-kan and Erm PCR products were spliced together using PhunSweet_F and Erm_R primers. To generate the comC- deletion mutant, 1 kb fragments upstream and downstream of comC were amplified from D39 gDNA using primers ComC_Up_F/ Com- C_Up_R and ComC_Down_F/ ComC_Down_R, respectively. These amplicons were spliced with the PhunSweetErm cassette with primers ComC_Up_F and ComC_Down_R to generate comC::PhunSweetErm. For generation of D39x comC-, D39x was grown in 10 mL C+Y until OD620~0.08. 1 mL of culture was incubated with 3 μL of 1 mg/ml CSP1 [39] and the comC:: PhunSweetErm amplicon for three hours at 37˚C, 5% CO2. Transformants were selected on plates containing 1μg/mL erythromycin. Correct deletion of comC was confirmed through lack of growth on counter-selection plates (15mM chlorinated-phenylalanine, 10% sucrose) and through PCR. To generate the comC mutant in TIGR4, the comC::PhunSweetErm Table 2. Primers used to generate D39x comC-. Name PhunSweet_F PhunSweet-kan_R Erm_F Erm_R ComC_Up_F ComC_UP_R ComC_Down_F ComC_Down_R Sequence CAATTAACTTTACAAATTCCCACTATTAAGG GTTTGCTTCTAAGTCTTATTTCCACTTTTGTGCCCGTGCTTATAAGGG GGAAATAAGACTTAGAAGCAAAC CCAAATTTACAAAAGCGACTC GAAAAACTACCCAAGGCTCCACT ATAGTGGGAATTTGTAAAGTTAATTGAATAAAATCTCCTAAAATGTTTTTTCTTG GAGTCGCTTTTGTAAATTTGGTGAAATAAGGGGAAAGAGTAATGGATTTAT TAGCTATCAGCCGATCCTTCG https://doi.org/10.1371/journal.ppat.1011421.t002 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 18 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence amplicon was generated in the same manner as above using the same primers, but TIGR4 gDNA was the source of DNA. TIGR4 was transformed with the comC::PhunSweetErm ampli- con following the same protocol as above, except for using CSP2 instead of CSP1. For PhunSweet-kan_R, ComC_Up_R, and ComC_Down_F, nucleotides in bold are over- lapping sequence of 5’ of erm, 5’ of PhunSweet, and 3’ of erm, respectively. CSP synthesis CSP1, CSP2, and all modified variants of CSP1 were synthesized and HPLC purified by the peptide synthesis core at St. Jude Children’s Research Hospital. CSPs were reconstituted in nuclease-free water at a concentration of 1 mg/mL. For CSP1, the amino acid sequence was EMRLSKFFRDFILQRKK; for CSP2, the sequence was EMRISRIILDFLFLRKK; for R3H, the sequence was EMHLSKFFRDFILQRKK; for K6H, the sequence was EMRLSHFFRDFILQ RKK; for R9H, the sequence was EMRLSKFFHDFILQRKK; for R9A, the sequence was EMRLSKFFADFILQRKK; for R15H, the sequence was EMRLSKFFRDFILQHKK. Mass spectrometry CSPs was incubated with either porcine trypsin gold (Promega #V5280) or human airway trypsin-like protease (HAT; Bio-techne R&D #2695-SE-010) for thirty minutes, room temper- ature. After thirty minutes, 10% acetic acid was added to stop further digestion. Peptide sam- ples were loaded on a nanoscale capillary reverse phase C18 column by a HPLC system (Thermo Ultimate3000) and eluted by a gradient (~30 min). The eluted peptides were ionized by electrospray ionization and detected by an inline mass spectrometer (Thermo Orbitrap Fusion). High resolution MS1 spectra were collected in the orbitrap and followed by MS/MS of the 20 most abundant ions using the ion trap. This process was cycled over the entire liquid chromatography gradient. Database searches were performed using Sequest search engine in an in-house SPIDERS software package. All matched MS/MS spectra were filtered by mass accuracy and matching scores to reduce protein false discovery rate to ~1%. The total number of spectra, namely spectral counts (SC), matching to individual proteins may reflect their rela- tive abundance in one sample after the protein size is normalized. Fractional area of each pep- tide and calculations to generate figures are included as S1 File. in vitro recombination frequency Strains were inoculated in C+Y at OD620~0.03. During culture incubation, porcine pancreatic trypsin (PPT; Sigma #T4799) was reconstituted in Hank’s balanced salt solution (Gibco) and incubated with either water or 4-(2-Aminoethyl) benzenesulfonyl fluoride hydrochloride (AEBSF; Cayman #14321) in a total volume of 50μL 30mM Tris, pH 8.5 and incubated for 15 minutes, room temperature. CSP1 or CSP2 was added, followed by incubation for 30 minutes, room temperature. When the culture reached OD620~0.08, 1 mL of culture was transferred to the PPT-AEBSF-CSP mixture and 2 μg of D39 Tn-seq library gDNA (D39 strains) or 2 μg of TIGR4 Tn-seq library gDNA (TIGR4 strains) was added. For D39x comC-, TIGR4, and TIGR4 comC-, the culture was pelleted at 6,000xg, 5 min and resuspended in fresh C+Y (to increase recombination efficiency; S3 Fig) before the culture was transferred to the mixture. Upon addi- tion of the culture, the final concentration of PPT ranged from 0–10 μg/mL, AEBSF was 0.1 mM, CSP was 3 μg/mL, and gDNA was 2 μg/mL. The transformation was incubated for 3 hours at 37˚C, 5% CO2, followed by serial dilution and plating on two TSA blood plates and two TSA blood plates supplemented with 200 μg/mL spectinomycin. To observe low-fre- quency recombination events, 200 μL of the transformation was also spread plated on two TSA blood plates supplemented with 200 μg/mL spectinomycin. For transformations with PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 19 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence HAT incubated CSP, HAT (Bio-techne R&D #2695-SE-010) was reconstituted in 25mM Tris, 150mM NaCl, pH 7.5. All incubation times and final concentrations were the same as for PPT; however, the volumes were 1/10th as that for PPT. For both PPT and HAT, the experiment was repeated at least three times. Recombination frequencies in increasing concentrations of prote- ase within each group (0 mM AEBSF or 0.1 mM AEBSF) were compared using Kruskal-Wallis one-way ANOVA. Luminescence assay To further investigate the impact of trypsin on competence, changes in luminescence upon incubation of CSP with trypsin in the strain DLA3, generously provided by Jan-Willem Veen- ing [41], were monitored. A similar protocol as the in vitro recombination frequency was fol- lowed. DLA3 was inoculated in C+Y at OD620~0.03. Either PPT or HAT was incubated with water or AEBSF for 15 minutes, room temperature, followed by incubation with CSP1 for thirty minutes in a total volume of 5 μL 30mM Tris, pH 8.5. When the culture reached OD620~0.08, cells were pelleted at 6,000xg, 5 min and resuspended in fresh C+Y. 100 μL of the culture was added to the Trypsin-AEBSF-CSP mixture, followed by transfer to a white-walled, clear-bottom 96-well plate and addition of 5 μL of 10 mg/mL luciferin (Sigma #L6882). Lumi- nescence and OD620 measurements were taken every 30 minutes in a Biotek Cytation 3. Increasing concentrations of protease within each group (0 mM AEBSF or 0.1 mM AEBSF) were compared using two-way ANOVA in Graphpad 6. ex vivo recombination frequency To determine the impact of proteases from the lungs of mice on recombination frequency, a similar experimental setup as the in vitro recombination frequency was performed. D39x and D39x comC- were inoculated in C+Y at OD620~0.03. During culture incubation, the lungs of 10-week-old female Balb/cJ mice (N = 7) were harvested and homogenized in 600 μL PBS. Lung homogenate was incubated with either water or AEBSF for 30 minutes, room tempera- ture. 60 μL of lung homogenate-AEBSF mixture was transferred to a new tube. Addition of CSP1, culture, and gDNA and transformation was performed in the same manner as the in vitro recombination frequency. The final concentration of CSP was 3 μg/mL and gDNA was 2 μg/mL. The final concentration of AEBSF was 0mM, 1 mM, or 2mM. The lung from each mouse was divided into six transformations: D39x with lungs incubated with 0mM, 1 mM, or 2mM AEBSF and D39x comC- with lungs incubated with 0mM, 1 mM, or 2mM AEBSF. As a control, the same culture of D39x and D39x comC- were transformed with gDNA and CSP1 incubated with AEBSF in PBS. Recombination frequency of the transformation with CSP1 incubated with 1 mM and with 2 mM AEBSF were compared to that incubated with 0 mM AEBSF using Kruskal-Wallis one-way ANOVA in Graphpad 6. Of note, the concentration of AEBSF was higher here than for the in vitro recombination frequency experiments due to potential other components in the lung homogenate reducing the efficacy of the inhibitor. For in vivo treatment of mice prior to ex vivo recombination frequency experiment, 7-week-old female Balb/cJ mice were treated with an inflammatory stimulant, poly (I:C), or water control and with inhibitor AEBSF or PBS control. 50 μL of 1 mg/mL poly(I:C) HMW (Invivogen) or 50 μL water control was administered IN daily for four days prior to harvest (N = 10) [45]. In the same mice, either 125 μg/ 25 μL AEBSF or 25 μL PBS control was admin- istered IN daily for three days prior to harvest (N = 5) [58]. Lungs were harvested and homoge- nized in 600 μL PBS. The homogenates from all lungs were subjected to the ex vivo protocol described above and the recombination frequency of D39x and D39x comC- subjected to lungs incubated with 0mM, 1 mM, or 2mM AEBSF and CSP1 was determined. Recombination PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 20 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence frequencies of 0mM AEBSF groups were compared pairwise for each tissue using nonparamet- ric Mann Whitney t test and recombination frequency of 1 mM and 2 mM AEBSF were com- pared to 0 mM AEBSF within each group using Kruskal-Wallis one-way ANOVA in Graphpad 6. Protease levels For ex vivo experiments, the protease levels were measured from the same lung homogenate used in the ex vivo recombination frequency experiment, adapting the protocol from the HAT manufacturer (Bio-techne R&D). After the lungs were incubated with either water or AEBSF, 50 μL of lung homogenate mixture was transferred to a black, 96-well plate. 50 μL of substrate t-Butyloxycarbonyl Phe-Ser-Arg 7-amino-4methyl coumarin (Bachem # I1400) was added directly to the well, incubated for 10 minutes at room temperature, and fluorescence (380 exci- tation; 460 emission) was detected in a Biotek Cytation 3. Protease levels of lung homogenates incubated with 1 mM and 2 mM AEBSF were compared to that incubated with 0 mM AEBSF using one-way ANOVA and correlation between protease levels and recombination frequency was compared using two-tailed spearman in Graphpad 6. For in vivo experiments, homogenized lungs were further clarified after plating on TSA plates via centrifugation at 10,000xg, 5min. Sera was separated from whole blood via centrifu- gation at 10,000xg, 5 min in sera-collection tubes. The clarified homogenates and sera were transferred to a black, 96-well plate and 50 μL of substrate t-Butyloxycarbonyl Phe-Ser-Arg 7-amino-4methyl coumarin was added directly to the well and fluorescence (380 excitation; 460 emission) was detected in a Biotek Cytation 3. Protease levels of all groups were compared pairwise for each tissue using unpaired t test and correlation between protease levels and recombination frequency was compared using two-tailed spearman in Graphpad 6. in vivo recombination frequency To stimulate production of protease in vivo, 7-week-old female Balb/cJ mice were treated with an inflammatory stimulant, poly (I:C), or water control and with serine protease inhibitor, AEBSF, or PBS control. 50 μL of 1 mg/mL poly(I:C) HMW (Invivogen) or 50 μL water control was administered IN daily for four days prior to bacterial challenge. In the same mice, either 125 μg AEBSF or PBS control was administered IN daily for two days prior to poly (I:C) treat- ment, two hours prior to daily poly (I:C) treatment, and two hours prior to bacterial challenge. All mice were challenged with a 90:10 mixture of D39x (kanamycin resistance) and D39 Tn- seq library (spectinomycin resistance) intranasally at 107 CFU in 100 μL. At 20 hours following challenge, lungs and blood were harvested. Lungs were homogenized in PBS and lungs and blood were serially diluted and plated. To detect for recombinants, tissues were plated on TSA blood plates plus 400 μg/mL kanamycin plus 200 μg/mL spectinomycin in addition to plates without antibiotic and plates with either antibiotic alone to quantitate the respective bacterial populations. Plates were incubated overnight at 37˚C, 5% CO2 and recombinant colonies were enumerated. Any sample that had a total bacterial burden below 107 CFU/mL was excluded from calculated recombination frequency as it falls below the threshold to observe recombi- nants (S4 Fig). The calculated recombination frequency (number of recombinant colonies divide by total number of colonies) of all groups were compared for each tissue using nonpara- metric Mann-Whitney t test in Graphpad 6. To investigate impact of protease induction due to influenza infection [49,50] prior to bac- terial challenge in vivo, influenza A/California/04/2009 at 5x102 LD50 was intranasally admin- istered to 8-week-old female Balb/cJ mice (N = 20). Influenza A/California/04/2009 virus was purified as described previously [101]. As a control, PBS was intranasally administered to PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 21 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence 8-week-old female Balb/cJ mice (N = 20). Seven days post-infection, the same mice were infected with a 90:10 mixture of D39x (kanamycin resistance) and D39 Tn-seq library (specti- nomycin resistance) intranasally at 105 CFU in 100 μL. At 8 hours post-bacterial challenge, mice were treated with 2.5 mg/kg spectinomycin (adjusted for potency) to drive recombina- tion from the spectinomycin resistant donor strains to the kanamycin resistant, spectinomycin sensitive strains. At 12 hours following spectinomycin treatment, lungs and blood were har- vested. Lungs were homogenized in PBS and both blood and lungs were serially diluted and plated. Tissues were plated, recombinant colonies enumerated, and recombination frequency calculated as above. Number of recombinant colonies and the calculated recombination fre- quency (number of recombinant colonies divide by total number of colonies) detected in mice infected with influenza was compared to that in mice that received PBS for each tissue using nonparametric Mann-Whitney t test in Graphpad 6. Supporting information S1 Fig. Growth of S. pneumoniae is not reduced by addition of AEBSF or trypsin. Absor- bance at OD620 of DLA3 grown in the presence of CSP1 incubated with increasing concentra- tions of trypsin with and without inhibitor AEBSF. Incubation of CSP1 with PPT (A) without AEBSF or (B) with AEBSF. Incubation of CSP1 with HAT (C) without AEBSF or (D) with AEBSF. Experiment was repeated in triplicate. The mean value of OD620 of each 30-minute timepoint is reported; error bars are SEM. (TIF) S2 Fig. Modified CSP alters impact of protease on recombination frequency. Recombina- tion frequency upon incubation of modified CSP1 with increasing concentrations of serine proteases. Transformation of D39x comC- with modified CSP1 incubated with (A) PPT or (B) HAT. Negative controls included no addition of CSP1 and no addition of gDNA. Lines repre- sent median value. Dotted line represents lowest point of detection. Recombination frequen- cies of increasing concentrations of protease within each modified CSP were compared using Kruskal-Wallis one-way ANOVA; p = 0.05–0.01, p = 0.01–0.001. Changes in the concentra- tions of protease between modified CSP was compared using two-way ANOVA; all groups non-significant. (TIF) S3 Fig. AEBSF alone does not induce recombination and D39x comC- recombination fre- quency is increased upon washing with fresh media. (A) Recombination frequency of D39x and D39x comC- upon incubation of CSP1 in PBS with increasing concentrations of AEBSF. Lines represent median. Dotted line represents lowest point of detection. (B) Recombination frequency of D39x comC- with and without washing cells with fresh media. Lines represent median. Dotted line represents lowest point of detection. (TIF) S4 Fig. Total number of cells and recombinants recovered from lungs and blood during in vivo recombination frequency experiment. Total number of cells per mL recovered from (A) the lungs and (B) the blood of mice that received no stimulant (water) or poly (I:C), and either received no inhibitor (PBS) or inhibitor AEBSF. Lines represent median. Dashed line at 107 represents threshold for detection of recombinants. Any bacterial burden that fell below this threshold was excluded from Fig 8. Dotted line represents lowest point of detection. Total number of recombinants per mL recovered from (C) the lungs and (D) the blood. Lines repre- sent median. Number of recombinants per mL of all groups were compared pairwise for each PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 22 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence tissue using nonparametric Mann-Whitney t test; *p = 0.01–0.001. (TIF) S1 File. Raw mass spectrometry data and calculated proportions for CSP cleavage experi- ments. (XLSX) Author Contributions Conceptualization: Haley Echlin, Jason W. Rosch. Formal analysis: Haley Echlin. Funding acquisition: Jason W. Rosch. Investigation: Haley Echlin, Amy Iverson, Ugo Sardo, Jason W. Rosch. Methodology: Haley Echlin, Amy Iverson, Ugo Sardo. Project administration: Haley Echlin, Jason W. Rosch. Supervision: Jason W. Rosch. Validation: Haley Echlin. Writing – original draft: Haley Echlin, Jason W. Rosch. Writing – review & editing: Haley Echlin, Jason W. Rosch. References 1. Henriques-Normark B, Tuomanen EI. The pneumococcus: epidemiology, microbiology, and patho- genesis. Cold Spring Harb Perspect Med. 2013; 3(7). https://doi.org/10.1101/cshperspect.a010215 PMID: 23818515; PubMed Central PMCID: PMC3685878. 2. Weiser JN, Ferreira DM, Paton JC. Streptococcus pneumoniae: transmission, colonization and inva- sion. Nat Rev Microbiol. 2018; 16(6):355–67. Epub 2018/03/31. https://doi.org/10.1038/s41579-018- 0001-8 PMID: 29599457; PubMed Central PMCID: PMC5949087. 3. Bogaert D, De Groot R, Hermans PW. Streptococcus pneumoniae colonisation: the key to pneumo- coccal disease. Lancet Infect Dis. 2004; 4(3):144–54. https://doi.org/10.1016/S1473-3099(04)00938- 7 PMID: 14998500. 4. Henriques-Normark B, Normark S. Commensal pathogens, with a focus on Streptococcus pneumo- niae, and interactions with the human host. Exp Cell Res. 2010; 316(8):1408–14. Epub 2010/03/17. https://doi.org/10.1016/j.yexcr.2010.03.003 PMID: 20227406. 5. Loughran AJ, Orihuela CJ, Tuomanen EI. Streptococcus pneumoniae: Invasion and Inflammation. Microbiol Spectr. 2019; 7(2). Epub 2019/03/16. https://doi.org/10.1128/microbiolspec.GPP3-0004- 2018 PMID: 30873934; PubMed Central PMCID: PMC6422050. 6. Dallaire F, Ouellet N, Bergeron Y, Turmel V, Gauthier MC, Simard M, et al. Microbiological and inflam- matory factors associated with the development of pneumococcal pneumonia. J Infect Dis. 2001; 184 (3):292–300. Epub 2001/07/10. https://doi.org/10.1086/322021 PMID: 11443554. 7. Standish AJ, Weiser JN. Human neutrophils kill Streptococcus pneumoniae via serine proteases. J Immunol. 2009; 183(4):2602–9. Epub 2009/07/22. https://doi.org/10.4049/jimmunol.0900688 PMID: 19620298. 8. Pham CT. Neutrophil serine proteases: specific regulators of inflammation. Nat Rev Immunol. 2006; 6 (7):541–50. Epub 2006/06/27. https://doi.org/10.1038/nri1841 PMID: 16799473. 9. Di Cera E. Serine proteases. IUBMB Life. 2009; 61(5):510–5. Epub 2009/01/31. https://doi.org/10. 1002/iub.186 PMID: 19180666; PubMed Central PMCID: PMC2675663. 10. Page MJ, Di Cera E. Serine peptidases: classification, structure and function. Cell Mol Life Sci. 2008; 65(7–8):1220–36. Epub 2008/02/09. https://doi.org/10.1007/s00018-008-7565-9 PMID: 18259688. 11. Patel S. A critical review on serine protease: Key immune manipulator and pathology mediator. Aller- gol Immunopathol (Madr). 2017; 45(6):579–91. Epub 2017/02/27. https://doi.org/10.1016/j.aller.2016. 10.011 PMID: 28236540; PubMed Central PMCID: PMC7126602. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 23 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence 12. Sorensen OE, Follin P, Johnsen AH, Calafat J, Tjabringa GS, Hiemstra PS, et al. Human cathelicidin, hCAP-18, is processed to the antimicrobial peptide LL-37 by extracellular cleavage with proteinase 3. Blood. 2001; 97(12):3951–9. Epub 2001/06/05. https://doi.org/10.1182/blood.v97.12.3951 PMID: 11389039. 13. Menou A, Duitman J, Flajolet P, Sallenave JM, Mailleux AA, Crestani B. Human airway trypsin-like pro- tease, a serine protease involved in respiratory diseases. Am J Physiol Lung Cell Mol Physiol. 2017; 312(5):L657–L68. Epub 2017/02/27. https://doi.org/10.1152/ajplung.00509.2016 PMID: 28235951. 14. Takahashi M, Sano T, Yamaoka K, Kamimura T, Umemoto N, Nishitani H, et al. Localization of human airway trypsin-like protease in the airway: an immunohistochemical study. Histochem Cell Biol. 2001; 115(3):181–7. Epub 2001/05/01. https://doi.org/10.1007/s004180000243 PMID: 11326745. 15. Yasuoka S, Ohnishi T, Kawano S, Tsuchihashi S, Ogawara M, Masuda K, et al. Purification, character- ization, and localization of a novel trypsin-like protease found in the human airway. Am J Respir Cell Mol Biol. 1997; 16(3):300–8. Epub 1997/03/01. https://doi.org/10.1165/ajrcmb.16.3.9070615 PMID: 9070615. 16. Simell B, Auranen K, Kayhty H, Goldblatt D, Dagan R, O’Brien KL, et al. The fundamental link between pneumococcal carriage and disease. Expert Rev Vaccines. 2012; 11(7):841–55. Epub 2012/08/24. https://doi.org/10.1586/erv.12.53 PMID: 22913260. 17. Shak JR, Vidal JE, Klugman KP. Influence of bacterial interactions on pneumococcal colonization of the nasopharynx. Trends Microbiol. 2013; 21(3):129–35. Epub 2013/01/01. https://doi.org/10.1016/j. tim.2012.11.005 PMID: 23273566; PubMed Central PMCID: PMC3729046. 18. Lorenz MG, Wackernagel W. Bacterial gene transfer by natural genetic transformation in the environ- ment. Microbiol Rev. 1994; 58(3):563–602. Epub 1994/09/01. https://doi.org/10.1128/mr.58.3.563- 602.1994 PMID: 7968924; PubMed Central PMCID: PMC372978. 19. Claverys JP, Havarstein LS. Extracellular-peptide control of competence for genetic transformation in Streptococcus pneumoniae. Front Biosci. 2002; 7:d1798–814. Epub 2002/07/23. https://doi.org/10. 2741/claverys PMID: 12133809. 20. Tomasz A, Hotchkiss RD. Regulation of the Transformability of Pheumococcal Cultures by Macromo- lecular Cell Products. Proc Natl Acad Sci U S A. 1964; 51:480–7. Epub 1964/03/01. https://doi.org/10. 1073/pnas.51.3.480 PMID: 14171462; PubMed Central PMCID: PMC300099. 21. Pestova EV, Havarstein LS, Morrison DA. Regulation of competence for genetic transformation in Streptococcus pneumoniae by an auto-induced peptide pheromone and a two-component regulatory system. Mol Microbiol. 1996; 21(4):853–62. Epub 1996/08/01. https://doi.org/10.1046/j.1365-2958. 1996.501417.x PMID: 8878046. 22. Martin B, Soulet AL, Mirouze N, Prudhomme M, Mortier-Barriere I, Granadel C, et al. ComE/ComE~P interplay dictates activation or extinction status of pneumococcal X-state (competence). Mol Microbiol. 2013; 87(2):394–411. Epub 2012/12/12. https://doi.org/10.1111/mmi.12104 PMID: 23216914. 23. Weyder M, Prudhomme M, Berge M, Polard P, Fichant G. Dynamic Modeling of Streptococcus pneu- moniae Competence Provides Regulatory Mechanistic Insights Into Its Tight Temporal Regulation. Front Microbiol. 2018; 9:1637. Epub 2018/08/09. https://doi.org/10.3389/fmicb.2018.01637 PMID: 30087661; PubMed Central PMCID: PMC6066662. 24. Peterson S, Cline RT, Tettelin H, Sharov V, Morrison DA. Gene expression analysis of the Streptococ- cus pneumoniae competence regulons by use of DNA microarrays. J Bacteriol. 2000; 182(21):6192– 202. Epub 2000/10/13. https://doi.org/10.1128/JB.182.21.6192-6202.2000 PMID: 11029442; PubMed Central PMCID: PMC94756. 25. Havarstein LS, Coomaraswamy G, Morrison DA. An unmodified heptadecapeptide pheromone induces competence for genetic transformation in Streptococcus pneumoniae. Proc Natl Acad Sci U S A. 1995; 92(24):11140–4. Epub 1995/11/21. https://doi.org/10.1073/pnas.92.24.11140 PMID: 7479953; PubMed Central PMCID: PMC40587. 26. Bartilson M, Marra A, Christine J, Asundi JS, Schneider WP, Hromockyj AE. Differential fluorescence induction reveals Streptococcus pneumoniae loci regulated by competence stimulatory peptide. Mol Microbiol. 2001; 39(1):126–35. Epub 2000/12/21. https://doi.org/10.1046/j.1365-2958.2001.02218.x PMID: 11123694. 27. Peterson SN, Sung CK, Cline R, Desai BV, Snesrud EC, Luo P, et al. Identification of competence pheromone responsive genes in Streptococcus pneumoniae by use of DNA microarrays. Mol Micro- biol. 2004; 51(4):1051–70. Epub 2004/02/07. https://doi.org/10.1046/j.1365-2958.2003.03907.x PMID: 14763980. 28. Zhu L, Lin J, Kuang Z, Vidal JE, Lau GW. Deletion analysis of Streptococcus pneumoniae late compe- tence genes distinguishes virulence determinants that are dependent or independent of competence induction. Mol Microbiol. 2015; 97(1):151–65. Epub 2015/04/08. https://doi.org/10.1111/mmi.13016 PMID: 25846124; PubMed Central PMCID: PMC4536566. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 24 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence 29. Steinmoen H, Teigen A, Havarstein LS. Competence-induced cells of Streptococcus pneumoniae lyse competence-deficient cells of the same strain during cocultivation. J Bacteriol. 2003; 185(24):7176– 83. Epub 2003/12/04. https://doi.org/10.1128/JB.185.24.7176-7183.2003 PMID: 14645278; PubMed Central PMCID: PMC296248. 30. 31. Lin J, Zhu L, Lau GW. Disentangling competence for genetic transformation and virulence in Strepto- coccus pneumoniae. Curr Genet. 2016; 62(1):97–103. Epub 2015/09/26. https://doi.org/10.1007/ s00294-015-0520-z PMID: 26403231. Trappetti C, Gualdi L, Di Meola L, Jain P, Korir CC, Edmonds P, et al. The impact of the competence quorum sensing system on Streptococcus pneumoniae biofilms varies depending on the experimental model. BMC Microbiol. 2011; 11:75. Epub 2011/04/16. https://doi.org/10.1186/1471-2180-11-75 PMID: 21492426; PubMed Central PMCID: PMC3098770. 32. Marks LR, Reddinger RM, Hakansson AP. High levels of genetic recombination during nasopharyn- geal carriage and biofilm formation in Streptococcus pneumoniae. mBio. 2012; 3(5). Epub 2012/09/ 28. https://doi.org/10.1128/mBio.00200-12 PMID: 23015736; PubMed Central PMCID: PMC3448161. 33. Felix S, Henares D, Munoz-Almagro C, Sa-Leao R. Carriage of multiple Streptococcus pneumoniae capsular types is frequent among children with invasive pneumococcal disease. Eur J Clin Microbiol Infect Dis. 2021; 40(11):2397–401. Epub 2021/04/03. https://doi.org/10.1007/s10096-021-04231-4 PMID: 33797644; PubMed Central PMCID: PMC8017099. 34. Wyllie AL, Wijmenga-Monsuur AJ, van Houten MA, Bosch A, Groot JA, van Engelsdorp Gastelaars J, et al. Molecular surveillance of nasopharyngeal carriage of Streptococcus pneumoniae in children vac- cinated with conjugated polysaccharide pneumococcal vaccines. Sci Rep. 2016; 6:23809. Epub 2016/ 04/06. https://doi.org/10.1038/srep23809 PMID: 27046258; PubMed Central PMCID: PMC4820691. 35. Wyllie AL, Chu ML, Schellens MH, van Engelsdorp Gastelaars J, Jansen MD, van der Ende A, et al. Streptococcus pneumoniae in saliva of Dutch primary school children. PLoS One. 2014; 9(7): e102045. Epub 2014/07/12. https://doi.org/10.1371/journal.pone.0102045 PMID: 25013895; PubMed Central PMCID: PMC4094488. 36. Lattar SM, Wu X, Brophy J, Sakai F, Klugman KP, Vidal JE. A Mechanism of Unidirectional Transfor- mation, Leading to Antibiotic Resistance, Occurs within Nasopharyngeal Pneumococcal Biofilm Con- sortia. mBio. 2018; 9(3). Epub 2018/05/17. https://doi.org/10.1128/mBio.00561-18 PMID: 29764945; PubMed Central PMCID: PMC5954218. 37. Kandasamy R, Gurung M, Thapa A, Ndimah S, Adhikari N, Murdoch DR, et al. Multi-serotype pneumo- coccal nasopharyngeal carriage prevalence in vaccine naive Nepalese children, assessed using molecular serotyping. PLoS One. 2015; 10(2):e0114286. Epub 2015/02/03. https://doi.org/10.1371/ journal.pone.0114286 PMID: 25643355; PubMed Central PMCID: PMC4313945. 38. Cherazard R, Epstein M, Doan TL, Salim T, Bharti S, Smith MA. Antimicrobial Resistant Streptococcus pneumoniae: Prevalence, Mechanisms, and Clinical Implications. Am J Ther. 2017; 24(3):e361–e9. Epub 2017/04/22. https://doi.org/10.1097/MJT.0000000000000551 PMID: 28430673. 39. Pozzi G, Masala L, Iannelli F, Manganelli R, Havarstein LS, Piccoli L, et al. Competence for genetic transformation in encapsulated strains of Streptococcus pneumoniae: two allelic variants of the pep- tide pheromone. J Bacteriol. 1996; 178(20):6087–90. Epub 1996/10/01. https://doi.org/10.1128/jb. 178.20.6087-6090.1996 PMID: 8830714; PubMed Central PMCID: PMC178474. 40. Attaiech L, Olivier A, Mortier-Barriere I, Soulet AL, Granadel C, Martin B, et al. Role of the single- stranded DNA-binding protein SsbB in pneumococcal transformation: maintenance of a reservoir for genetic plasticity. PLoS Genet. 2011; 7(6):e1002156. Epub 2011/07/09. https://doi.org/10.1371/ journal.pgen.1002156 PMID: 21738490; PubMed Central PMCID: PMC3128108. 41. Slager J, Kjos M, Attaiech L, Veening JW. Antibiotic-induced replication stress triggers bacterial com- petence by increasing gene dosage near the origin. Cell. 2014; 157(2):395–406. Epub 2014/04/15. https://doi.org/10.1016/j.cell.2014.01.068 PMID: 24725406. 42. Kurushima J, Campo N, van Raaphorst R, Cerckel G, Polard P, Veening JW. Unbiased homeologous recombination during pneumococcal transformation allows for multiple chromosomal integration events. Elife. 2020;9. Epub 2020/09/24. https://doi.org/10.7554/eLife.58771 PMID: 32965219; PubMed Central PMCID: PMC7567608. 43. Zhu L, Lau GW. Inhibition of competence development, horizontal gene transfer and virulence in Streptococcus pneumoniae by a modified competence stimulating peptide. PLoS Pathog. 2011; 7(9): e1002241. Epub 2011/09/13. https://doi.org/10.1371/journal.ppat.1002241 PMID: 21909280; PubMed Central PMCID: PMC3164649. 44. Sugimoto S, Morizane S, Nomura H, Kobashi M, Sugihara S, Iwatsuki K. Toll-like receptor signaling induces the expression of lympho-epithelial Kazal-type inhibitor in epidermal keratinocytes. J Dermatol Sci. 2018; 92(2):181–7. Epub 2018/10/03. https://doi.org/10.1016/j.jdermsci.2018.09.001 PMID: 30270115. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 25 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence 45. Stowell NC, Seideman J, Raymond HA, Smalley KA, Lamb RJ, Egenolf DD, et al. Long-term activation of TLR3 by poly(I:C) induces inflammation and impairs lung function in mice. Respir Res. 2009; 10 (1):43. Epub 2009/06/03. https://doi.org/10.1186/1465-9921-10-43 PMID: 19486528; PubMed Central PMCID: PMC2694181. 46. McCullers JA. The co-pathogenesis of influenza viruses with bacteria in the lung. Nat Rev Microbiol. 2014; 12(4):252–62. Epub 2014/03/05. https://doi.org/10.1038/nrmicro3231 PMID: 24590244. 47. McCullers JA, Rehg JE. Lethal synergism between influenza virus and Streptococcus pneumoniae: characterization of a mouse model and the role of platelet-activating factor receptor. J Infect Dis. 2002; 186(3):341–50. Epub 2002/07/23. https://doi.org/10.1086/341462 PMID: 12134230. 48. Rowe HM, Meliopoulos VA, Iverson A, Bomme P, Schultz-Cherry S, Rosch JW. Direct interactions with influenza promote bacterial adherence during respiratory infections. Nat Microbiol. 2019; 4 (8):1328–36. Epub 2019/05/22. https://doi.org/10.1038/s41564-019-0447-0 PMID: 31110359; PubMed Central PMCID: PMC7069060. 49. Wang S, Le TQ, Kurihara N, Chida J, Cisse Y, Yano M, et al. Influenza virus-cytokine-protease cycle in the pathogenesis of vascular hyperpermeability in severe influenza. J Infect Dis. 2010; 202(7):991– 1001. Epub 2010/08/25. https://doi.org/10.1086/656044 PMID: 20731583; PubMed Central PMCID: PMC7537608. 50. Indalao IL, Sawabuchi T, Takahashi E, Kido H. IL-1beta is a key cytokine that induces trypsin upregu- lation in the influenza virus-cytokine-trypsin cycle. Arch Virol. 2017; 162(1):201–11. Epub 2016/10/08. https://doi.org/10.1007/s00705-016-3093-3 PMID: 27714503; PubMed Central PMCID: PMC5225228. 51. Bottcher-Friebertshauser E, Klenk HD, Garten W. Activation of influenza viruses by proteases from host cells and bacteria in the human airway epithelium. Pathog Dis. 2013; 69(2):87–100. Epub 2013/ 07/04. https://doi.org/10.1111/2049-632X.12053 PMID: 23821437; PubMed Central PMCID: PMC7108517. 52. Bertram S, Glowacka I, Steffen I, Kuhl A, Pohlmann S. Novel insights into proteolytic cleavage of influ- enza virus hemagglutinin. Rev Med Virol. 2010; 20(5):298–310. Epub 2010/07/16. https://doi.org/10. 1002/rmv.657 PMID: 20629046; PubMed Central PMCID: PMC7169116. 53. Bertram S, Glowacka I, Muller MA, Lavender H, Gnirss K, Nehlmeier I, et al. Cleavage and activation of the severe acute respiratory syndrome coronavirus spike protein by human airway trypsin-like prote- ase. J Virol. 2011; 85(24):13363–72. Epub 2011/10/14. https://doi.org/10.1128/JVI.05300-11 PMID: 21994442; PubMed Central PMCID: PMC3233180. 54. Bottcher E, Matrosovich T, Beyerle M, Klenk HD, Garten W, Matrosovich M. Proteolytic activation of influenza viruses by serine proteases TMPRSS2 and HAT from human airway epithelium. J Virol. 2006; 80(19):9896–8. Epub 2006/09/16. https://doi.org/10.1128/JVI.01118-06 PMID: 16973594; PubMed Central PMCID: PMC1617224. 55. Kietzman CC, Gao G, Mann B, Myers L, Tuomanen EI. Dynamic capsule restructuring by the main pneumococcal autolysin LytA in response to the epithelium. Nat Commun. 2016; 7:10859. Epub 2016/ 03/01. https://doi.org/10.1038/ncomms10859 PMID: 26924467; PubMed Central PMCID: PMC4773454. 56. Bergmann S, Schoenen H, Hammerschmidt S. The interaction between bacterial enolase and plas- minogen promotes adherence of Streptococcus pneumoniae to epithelial and endothelial cells. Int J Med Microbiol. 2013; 303(8):452–62. Epub 2013/08/03. https://doi.org/10.1016/j.ijmm.2013.06.002 PMID: 23906818. 57. Bergmann S, Rohde M, Preissner KT, Hammerschmidt S. The nine residue plasminogen-binding motif of the pneumococcal enolase is the major cofactor of plasmin-mediated degradation of extracel- lular matrix, dissolution of fibrin and transmigration. Thromb Haemost. 2005; 94(2):304–11. Epub 2005/08/23. https://doi.org/10.1160/TH05-05-0369 PMID: 16113819. 58. Bahgat MM, Blazejewska P, Schughart K. Inhibition of lung serine proteases in mice: a potentially new approach to control influenza infection. Virol J. 2011; 8:27. Epub 2011/01/22. https://doi.org/10.1186/ 1743-422X-8-27 PMID: 21251300; PubMed Central PMCID: PMC3034701. 59. Oggioni MR, Trappetti C, Kadioglu A, Cassone M, Iannelli F, Ricci S, et al. Switch from planktonic to sessile life: a major event in pneumococcal pathogenesis. Mol Microbiol. 2006; 61(5):1196–210. Epub 2006/08/24. https://doi.org/10.1111/j.1365-2958.2006.05310.x PMID: 16925554; PubMed Central PMCID: PMC1618759. 60. van Opijnen T, Camilli A. A fine scale phenotype-genotype virulence map of a bacterial pathogen. Genome Res. 2012; 22(12):2541–51. Epub 2012/07/25. https://doi.org/10.1101/gr.137430.112 PMID: 22826510; PubMed Central PMCID: PMC3514683. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 26 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence 61. Kadioglu A, Weiser JN, Paton JC, Andrew PW. The role of Streptococcus pneumoniae virulence fac- tors in host respiratory colonization and disease. Nat Rev Microbiol. 2008; 6(4):288–301. Epub 2008/ 03/15. https://doi.org/10.1038/nrmicro1871 PMID: 18340341. 62. Aggarwal SD, Eutsey R, West-Roberts J, Domenech A, Xu W, Abdullah IT, et al. Function of BriC pep- tide in the pneumococcal competence and virulence portfolio. PLoS Pathog. 2018; 14(10):e1007328. Epub 2018/10/12. https://doi.org/10.1371/journal.ppat.1007328 PMID: 30308062; PubMed Central PMCID: PMC6181422. 63. Zheng Y, Zhang X, Wang X, Wang L, Zhang J, Yin Y. ComE, an Essential Response Regulator, Nega- tively Regulates the Expression of the Capsular Polysaccharide Locus and Attenuates the Bacterial Virulence in Streptococcus pneumoniae. Front Microbiol. 2017; 8:277. Epub 2017/03/23. https://doi. org/10.3389/fmicb.2017.00277 PMID: 28326061; PubMed Central PMCID: PMC5339220. 64. Kowalko JE, Sebert ME. The Streptococcus pneumoniae competence regulatory system influences respiratory tract colonization. Infect Immun. 2008; 76(7):3131–40. Epub 2008/04/30. https://doi.org/ 10.1128/IAI.01696-07 PMID: 18443092; PubMed Central PMCID: PMC2446691. 65. Matsushima R, Takahashi A, Nakaya Y, Maezawa H, Miki M, Nakamura Y, et al. Human airway tryp- sin-like protease stimulates human bronchial fibroblast proliferation in a protease-activated receptor- 2-dependent pathway. Am J Physiol Lung Cell Mol Physiol. 2006; 290(2):L385–95. Epub 2005/10/04. https://doi.org/10.1152/ajplung.00098.2005 PMID: 16199437. 66. Johnsborg O, Havarstein LS. Regulation of natural genetic transformation and acquisition of trans- forming DNA in Streptococcus pneumoniae. FEMS Microbiol Rev. 2009; 33(3):627–42. Epub 2009/ 04/28. https://doi.org/10.1111/j.1574-6976.2009.00167.x PMID: 19396959. 67. Moreno-Gamez S, Sorg RA, Domenech A, Kjos M, Weissing FJ, van Doorn GS, et al. Quorum sensing integrates environmental cues, cell density and cell history to control bacterial competence. Nat Com- mun. 2017; 8(1):854. Epub 2017/10/13. https://doi.org/10.1038/s41467-017-00903-y PMID: 29021534; PubMed Central PMCID: PMC5636887. 68. Nelson AL, Roche AM, Gould JM, Chim K, Ratner AJ, Weiser JN. Capsule enhances pneumococcal colonization by limiting mucus-mediated clearance. Infect Immun. 2007; 75(1):83–90. Epub 2006/11/ 08. https://doi.org/10.1128/IAI.01475-06 PMID: 17088346; PubMed Central PMCID: PMC1828419. 69. Kahlert CR, Nigg S, Onder L, Dijkman R, Diener L, Vidal AGJ, et al. The quorum sensing com system regulates pneumococcal colonisation and invasive disease in a pseudo-stratified airway tissue model. Microbiol Res. 2023; 268:127297. Epub 2023/01/07. https://doi.org/10.1016/j.micres.2022.127297 PMID: 36608536; PubMed Central PMCID: PMC9868095. 70. Weiser JN, Austrian R, Sreenivasan PK, Masure HR. Phase variation in pneumococcal opacity: rela- tionship between colonial morphology and nasopharyngeal colonization. Infect Immun. 1994; 62 (6):2582–9. Epub 1994/06/01. https://doi.org/10.1128/iai.62.6.2582-2589.1994 PMID: 8188381; PubMed Central PMCID: PMC186548. 71. Cundell DR, Weiser JN, Shen J, Young A, Tuomanen EI. Relationship between colonial morphology and adherence of Streptococcus pneumoniae. Infect Immun. 1995; 63(3):757–61. Epub 1995/03/01. https://doi.org/10.1128/iai.63.3.757-761.1995 PMID: 7868244; PubMed Central PMCID: PMC173067. 72. Schaffner TO, Hinds J, Gould KA, Wuthrich D, Bruggmann R, Kuffer M, et al. A point mutation in cpsE renders Streptococcus pneumoniae nonencapsulated and enhances its growth, adherence and com- petence. BMC Microbiol. 2014; 14:210. Epub 2014/08/29. https://doi.org/10.1186/s12866-014-0210-x PMID: 25163487; PubMed Central PMCID: PMC4243769. 73. Marks LR, Parameswaran GI, Hakansson AP. Pneumococcal interactions with epithelial cells are cru- cial for optimal biofilm formation and colonization in vitro and in vivo. Infect Immun. 2012; 80(8):2744– 60. Epub 2012/05/31. https://doi.org/10.1128/IAI.00488-12 PMID: 22645283; PubMed Central PMCID: PMC3434590. 74. Blanchette-Cain K, Hinojosa CA, Akula Suresh Babu R, Lizcano A, Gonzalez-Juarbe N, Munoz-Alma- gro C, et al. Streptococcus pneumoniae biofilm formation is strain dependent, multifactorial, and asso- ciated with reduced invasiveness and immunoreactivity during colonization. mBio. 2013; 4(5):e00745– 13. Epub 2013/10/17. https://doi.org/10.1128/mBio.00745-13 PMID: 24129258; PubMed Central PMCID: PMC3812715. 75. Tseng BS, Reichhardt C, Merrihew GE, Araujo-Hernandez SA, Harrison JJ, MacCoss MJ, et al. A Bio- film Matrix-Associated Protease Inhibitor Protects Pseudomonas aeruginosa from Proteolytic Attack. mBio. 2018;9(2). Epub 2018/04/11. https://doi.org/10.1128/mBio.00543-18 PMID: 29636440; PubMed Central PMCID: PMC5893882. 76. Srinivasan R, Santhakumari S, Poonguzhali P, Geetha M, Dyavaiah M, Xiangmin L. Bacterial Biofilm Inhibition: A Focused Review on Recent Therapeutic Strategies for Combating the Biofilm Mediated Infections. Front Microbiol. 2021; 12:676458. Epub 2021/06/01. https://doi.org/10.3389/fmicb.2021. 676458 PMID: 34054785; PubMed Central PMCID: PMC8149761. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 27 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence 77. Chao Y, Marks LR, Pettigrew MM, Hakansson AP. Streptococcus pneumoniae biofilm formation and dispersion during colonization and disease. Front Cell Infect Microbiol. 2014; 4:194. Epub 2015/01/30. https://doi.org/10.3389/fcimb.2014.00194 PMID: 25629011; PubMed Central PMCID: PMC4292784. 78. Marks LR, Davidson BA, Knight PR, Hakansson AP. Interkingdom signaling induces Streptococcus pneumoniae biofilm dispersion and transition from asymptomatic colonization to disease. mBio. 2013; 4(4). Epub 2013/07/25. https://doi.org/10.1128/mBio.00438-13 PMID: 23882016; PubMed Central PMCID: PMC3735180. 79. Gilley RP, Orihuela CJ. Pneumococci in biofilms are non-invasive: implications on nasopharyngeal colonization. Front Cell Infect Microbiol. 2014; 4:163. Epub 2014/11/22. https://doi.org/10.3389/fcimb. 2014.00163 PMID: 25414838; PubMed Central PMCID: PMC4222220. 80. Pettigrew MM, Marks LR, Kong Y, Gent JF, Roche-Hakansson H, Hakansson AP. Dynamic changes in the Streptococcus pneumoniae transcriptome during transition from biofilm formation to invasive disease upon influenza A virus infection. Infect Immun. 2014; 82(11):4607–19. Epub 2014/08/20. https://doi.org/10.1128/IAI.02225-14 PMID: 25135685; PubMed Central PMCID: PMC4249342. 81. Lau GW, Haataja S, Lonetto M, Kensit SE, Marra A, Bryant AP, et al. A functional genomic analysis of type 3 Streptococcus pneumoniae virulence. Mol Microbiol. 2001; 40(3):555–71. Epub 2001/05/22. https://doi.org/10.1046/j.1365-2958.2001.02335.x PMID: 11359563. 82. Sanchez CJ, Shivshankar P, Stol K, Trakhtenbroit S, Sullam PM, Sauer K, et al. The pneumococcal serine-rich repeat protein is an intra-species bacterial adhesin that promotes bacterial aggregation in vivo and in biofilms. PLoS Pathog. 2010; 6(8):e1001044. Epub 2010/08/18. https://doi.org/10.1371/ journal.ppat.1001044 PMID: 20714350; PubMed Central PMCID: PMC2920850. 83. Marsh RL, Binks MJ, Smith-Vaughan HC, Janka M, Clark S, Richmond P, et al. Prevalence and sub- typing of biofilms present in bronchoalveolar lavage from children with protracted bacterial bronchitis or non-cystic fibrosis bronchiectasis: a cross-sectional study. Lancet Microbe. 2022; 3(3):e215–e23. Epub 2022/05/12. https://doi.org/10.1016/S2666-5247(21)00300-1 PMID: 35544075. 84. Lin J, Park P, Li H, Oh MW, Dobrucki IT, Dobrucki W, et al. Streptococcus pneumoniae Elaborates Persistent and Prolonged Competent State during Pneumonia-Derived Sepsis. Infect Immun. 2020;88 (4). Epub 2020/01/29. https://doi.org/10.1128/IAI.00919-19 PMID: 31988172; PubMed Central PMCID: PMC7093142. 85. Echenique JR, Trombe MC. Competence repression under oxygen limitation through the two-compo- nent MicAB signal-transducing system in Streptococcus pneumoniae and involvement of the PAS domain of MicB. J Bacteriol. 2001; 183(15):4599–608. Epub 2001/07/10. https://doi.org/10.1128/JB. 183.15.4599-4608.2001 PMID: 11443095; PubMed Central PMCID: PMC95355. 86. Claverys JP, Prudhomme M, Martin B. Induction of competence regulons as a general response to stress in gram-positive bacteria. Annu Rev Microbiol. 2006; 60:451–75. Epub 2006/06/15. https://doi. org/10.1146/annurev.micro.60.080805.142139 PMID: 16771651. 87. Ali MQ, Kohler TP, Schulig L, Burchhardt G, Hammerschmidt S. Pneumococcal Extracellular Serine Proteases: Molecular Analysis and Impact on Colonization and Disease. Front Cell Infect Microbiol. 2021; 11:763152. Epub 2021/11/19. https://doi.org/10.3389/fcimb.2021.763152 PMID: 34790590; PubMed Central PMCID: PMC8592123. 88. Ali MQ, Kohler TP, Burchhardt G, Wust A, Henck N, Bolsmann R, et al. Extracellular Pneumococcal Serine Proteases Affect Nasopharyngeal Colonization. Front Cell Infect Microbiol. 2020; 10:613467. Epub 2021/03/05. https://doi.org/10.3389/fcimb.2020.613467 PMID: 33659218; PubMed Central PMCID: PMC7917122. 89. Mahdi LK, Van der Hoek MB, Ebrahimie E, Paton JC, Ogunniyi AD. Characterization of Pneumococcal Genes Involved in Bloodstream Invasion in a Mouse Model. PLoS One. 2015; 10(11):e0141816. Epub 2015/11/06. https://doi.org/10.1371/journal.pone.0141816 PMID: 26539717; PubMed Central PMCID: PMC4634996. 90. Sender V, Hentrich K, Henriques-Normark B. Virus-Induced Changes of the Respiratory Tract Envi- ronment Promote Secondary Infections With Streptococcus pneumoniae. Front Cell Infect Microbiol. 2021; 11:643326. Epub 2021/04/09. https://doi.org/10.3389/fcimb.2021.643326 PMID: 33828999; PubMed Central PMCID: PMC8019817. 91. Sender V, Hentrich K, Pathak A, Tan Qian Ler A, Embaie BT, Lundstrom SL, et al. Capillary leakage provides nutrients and antioxidants for rapid pneumococcal proliferation in influenza-infected lower air- ways. Proc Natl Acad Sci U S A. 2020; 117(49):31386–97. Epub 2020/11/25. https://doi.org/10.1073/ pnas.2012265117 PMID: 33229573; PubMed Central PMCID: PMC7733805. 92. Gosink KK, Mann ER, Guglielmo C, Tuomanen EI, Masure HR. Role of novel choline binding proteins in virulence of Streptococcus pneumoniae. Infect Immun. 2000; 68(10):5690–5. Epub 2000/09/19. https://doi.org/10.1128/IAI.68.10.5690-5695.2000 PMID: 10992472; PubMed Central PMCID: PMC101524. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 28 / 29 PLOS PATHOGENSAirway proteolytic control of pneumococcal competence 93. 94. 95. de Stoppelaar SF, Bootsma HJ, Zomer A, Roelofs JJ, Hermans PW, van ’t Veer C, et al. Streptococ- cus pneumoniae serine protease HtrA, but not SFP or PrtA, is a major virulence factor in pneumonia. PLoS One. 2013; 8(11):e80062. Epub 2013/11/19. https://doi.org/10.1371/journal.pone.0080062 PMID: 24244609; PubMed Central PMCID: PMC3823867. Lacks S, Hotchkiss RD. A study of the genetic material determining enzyme in the pneumococcus. Biochim Biophys Acta. 1960; 39:508–17. Francis KP, Yu J, Bellinger-Kawahara C, Joh D, Hawkinson MJ, Xiao G, et al. Visualizing pneumococ- cal infections in the lungs of live mice using bioluminescent Streptococcus pneumoniae transformed with a novel gram-positive lux transposon. Infect Immun. 2001; 69(5):3350–8. Epub 2001/04/09. https://doi.org/10.1128/IAI.69.5.3350-3358.2001 PMID: 11292758; PubMed Central PMCID: PMC98294. 96. Matthews AJ, Rowe HM, Rosch JW, Camilli A. A Tn-seq Screen of Streptococcus pneumoniae Uncov- ers DNA Repair as the Major Pathway for Desiccation Tolerance and Transmission. Infect Immun. 2021; 89(8):e0071320. Epub 2021/05/26. https://doi.org/10.1128/IAI.00713-20 PMID: 34031124; PubMed Central PMCID: PMC8281258. 97. Echlin H, Rosch JW. Advancing Genetic Tools in Streptococcus pneumoniae. Genes (Basel). 2020; 11(9). Epub 2020/08/23. https://doi.org/10.3390/genes11090965 PMID: 32825523; PubMed Central PMCID: PMC7563404. 98. van Opijnen T, Camilli A. Genome-wide fitness and genetic interactions determined by Tn-seq, a high- throughput massively parallel sequencing method for microorganisms. Curr Protoc Microbiol. 2010; Chapter 1:Unit1E 3. Epub 2010/11/06. https://doi.org/10.1002/9780471729259.mc01e03s19 PMID: 21053251; PubMed Central PMCID: PMC3877651. 99. Echlin H, Frank MW, Iverson A, Chang TC, Johnson MD, Rock CO, et al. Pyruvate Oxidase as a Criti- cal Link between Metabolism and Capsule Biosynthesis in Streptococcus pneumoniae. PLoS Pathog. 2016; 12(10):e1005951. Epub 2016/10/21. https://doi.org/10.1371/journal.ppat.1005951 PMID: 27760231; PubMed Central PMCID: PMC5070856. 100. Horton RM, Cai ZL, Ho SN, Pease LR. Gene splicing by overlap extension: tailor-made genes using the polymerase chain reaction. Biotechniques. 1990; 8(5):528–35. Epub 1990/05/01. PMID: 2357375. 101. Carlson CM, Turpin EA, Moser LA, O’Brien KB, Cline TD, Jones JC, et al. Transforming growth factor- beta: activation by neuraminidase and role in highly pathogenic H5N1 influenza pathogenesis. PLoS Pathog. 2010; 6(10):e1001136. Epub 2010/10/16. https://doi.org/10.1371/journal.ppat.1001136 PMID: 20949074; PubMed Central PMCID: PMC2951376. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011421 May 31, 2023 29 / 29 PLOS PATHOGENS
10.1371_journal.pwat.0000071	RESEARCH ARTICLE The enabling environment for citywide water service provision: Insights from six successful cities John T. TrimmerID 1, Haleemah QureshiID 1 1, Miriam Otoo2, Caroline DelaireID a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Trimmer JT, Qureshi H, Otoo M, Delaire C (2023) The enabling environment for citywide water service provision: Insights from six successful cities. PLOS Water 2(6): e0000071. https://doi.org/10.1371/journal.pwat.0000071 Editor: Majid Shafiee-Jood, University of Virginia, UNITED STATES Received: October 31, 2022 Accepted: May 25, 2023 Published: June 26, 2023 Copyright: © 2023 Trimmer et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All data used to develop this study’s findings are included in the manuscript and supporting information. Funding: This work was conducted under the U.S. Agency for International Development (USAID) Urban Resilience by Building and Applying New Evidence in Water, Sanitation, and Hygiene (URBAN WASH) project, contract number GS00Q14OADU138 and order number 7200AA21M00012. All authors (JTT, HQ, MO, CD) received funding through this source. The funder (USAID) provided input on research questions and 1 The Aquaya Institute, Nairobi, Kenya, 2 Tetra Tech, Arlington, Virginia, United States of America john.t@aquaya.org Abstract Equitable access to safe drinking water remains a key challenge in many urban areas of low- and middle-income countries. This study aimed to characterize the enabling environ- ment for inclusive urban water service delivery, and specifically to elucidate the institutional arrangements, policies, regulations, service delivery approaches, financing models, and surrounding contextual factors that influence progress. We identified six cities across Africa, Asia, and South America that offered historical examples of success in inclusive piped water provision, resulting in high levels of access and service quality, including within low-income areas. Using a modified form of the social-ecological systems framework to structure our investigation, we conducted a comparative case study analysis to learn from these cities. Our analysis focused on a review of existing case-specific literature, supplemented by inter- views with 1–3 key informants per case to update or fill gaps in the literature. A variety of characteristics supported safe and inclusive services, with contextually appropriate strate- gies depending on existing institutional arrangements, infrastructure, and the surrounding social, economic, political, and environmental context. Our study cities illustrated three types of progress–utility-driven, regulator-supported, and municipality-driven–each charac- terized by specific features and drivers of success. We also identified 12 characteristics making up the enabling environment across all three types. These characteristics highlighted two broad themes. First, a well-functioning water service provider was often a prerequisite for inclusive, pro-poor service provision. Elements such as clear performance indicators, customer feedback mechanisms, and strategies to sustainably finance operating costs contributed to cities’ success. Second, inclusive water services often required explicit pro-poor policies and strategies, such as the removal of land tenure requirements for piped connections and community mobilization for participatory decision-making. Although the importance of specific characteristics will vary depending on context, our analysis offers a common foundation to guide progress toward universal access to safe water. 1. Introduction Achieving and sustaining equitable access to safe drinking water remains a key challenge in many urban areas of low- and middle-income countries (LMICs), especially in view of PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 1 / 29 manuscript preparation, but the authors alone are responsible for the views expressed in this publication and they do not necessarily represent the decisions or policies of USAID. https://www. globalwaters.org/resources/assets/urban-wash- fact-sheet. Competing interests: The authors have declared that no competing interests exist. The enabling environment for citywide water service provision continued urban expansion and the threats posed by climate change [1]. In 2020, an estimated 42% of urban residents in low-income countries did not have access to safely managed drink- ing water, and these average levels masked wide disparities across household income levels [2]. For example, while 95% of the richest urban residents in Uganda had access to at least basic drinking water service in 2017, only 53% of the poorest residents did [2]. Additionally, urban water access levels in some countries (such as Niger and Pakistan) have stagnated or fallen since 2000, while rural access has improved [2, 3]. Future trends in population growth, urbani- zation, and increasing pollution will likely exacerbate these issues [4]. Sustaining equitable improvements in safe urban water access requires an understanding of the enabling environ- ment and best practices from similar settings. In particular, the enabling environment consists of relevant policies, regulations, institutional arrangements, financing approaches, and addi- tional contextual factors that influence the actors and infrastructure responsible for delivering water to residents [5]. A number of contextual factors may contribute to non-inclusive and inequitable urban water provision that does not adequately serve low-income populations. These include persis- tent colonial-era arrangements, where infrastructure development favors a small, privileged class living in a city’s inner core [6]. The exclusion of low-income populations from formal pri- vate property and rental markets can drive unplanned urban expansion into the peripheral and marginal areas that remain available for informal development [1, 3, 7]. Unstable or flood- prone soils, shallow groundwater tables, unfavorable topography or geology, high-density housing, and uncertain land tenure often characterize these locations, hindering the installa- tion of conventional piped networks and increasing pollution risks [3, 7–9]. Those without access to the network typically rely on small-scale providers, who might operate private bore- holes or resell tap water, often at prices higher than those of piped services–paradoxically requiring the most vulnerable households to pay more than wealthier residents for water that may be of lesser quality [1, 6, 10]. The local enabling environment is foundational to the development, operation, and out- comes of service provision approaches, including their effectiveness, inclusivity, and sustain- ability. Broadly speaking, what may be most useful for today’s decision-makers is an overall understanding of the enabling environment for urban drinking water across a variety of geo- graphic, governance, social, political, economic, and environmental contexts [11]. Previous research has often focused on the challenges and opportunities associated with individual aspects of the enabling environment, such as water policies for informal settlements, cost recovery, or privatization [3, 6, 10, 12]. Among more comprehensive analyses of enabling envi- ronments, few have captured multiple countries [13, 14]. Both approaches leave room to better characterize the enabling environment for urban drinking water and draw parallels across diverse contexts. Specific considerations that require further study include the identification and application of appropriate pro-poor mechanisms across different service provision and regulatory environments and the roles of foreign aid and consumer tariffs in balancing finan- cial sustainability with affordability [1, 3, 6, 10, 15–17]. Accordingly, the objective of this study was to learn from diverse historical examples of improvements in inclusive, citywide delivery of piped water services in LMICs. Specifically, we sought to answer the following research question: What policies, regulations, institutional arrangements, and contextual factors have historically driven inclusive improvements in urban piped water access in lower-income countries? We focused on piped water because it is typically considered to be the highest standard of service [1, 18]. Based on existing literature and knowledge, we hypothesized that inclusive, citywide improvements in safe and affordable water provision require institutional, regulatory, and policy frameworks that specify clear mandates and strategies for reaching the poor. Additionally, we hypothesized that many PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 2 / 29 PLOS WATERThe enabling environment for citywide water service provision factors beyond formal policies–such as community participation, environmental shocks, and champions (i.e., individuals who proactively drive institutional or sector-wide reforms)–influ- ence the effectiveness and sustainability of these improvements. To answer our research ques- tion, we conducted a comparative case study analysis of six successful examples of citywide piped water service from various contexts. While we did hypothesize that the enabling envi- ronment would have certain general characteristics, we employed an inductive approach to allow for the possibility that additional features may emerge from our analysis. 2. Materials and methods 2.1. Conceptual framework for analysis In this study, we used a modified form of the social-ecological system (SES) framework to structure our analysis and characterize the urban water sector [11, 19]. Conceptual frameworks such as the SES framework can aid in categorizing key components and their relationships [11, 20, 21]. The SES framework originally emerged from studying governance of common-pool resource systems such as forests and fisheries and built upon the earlier Institutional Analysis and Development (IAD) framework [20–22]. For this study, we chose to employ the SES framework rather than others such as the IAD framework because the SES framework has a more explicit focus on contextual factors related to social, economic, political, and ecological conditions and resources [21]. This feature was especially useful, as we expected such contex- tual factors in our six cases to substantially influence the enabling environment for water ser- vice provision, regulation, and progress toward greater access and equity. The SES framework has been revised, adapted, and applied across multiple sectors, including by one of this study’s authors for urban water and sanitation [11, 19, 21, 23]. For this work, we modified the frame- work slightly to focus on the characteristics of governance structures, relevant actors, service delivery approaches, socioeconomic and political context, and environmental context (Fig 1). The five categories of the modified SES framework provided the basis for our comparative case study analysis. The Governance category defines rules and conditions under which actors and service delivery approaches operate [11]. At their core, these governance structures involve laws, statutes, policies, and institutional arrangements that organize water and sanitation ser- vice delivery and regulation. Actors participate in numerous ways, including as service provid- ers, regulators, policymakers, and consumers [11]. Relationships across these actors are connected with the roles and responsibilities defined through governance structures. Service delivery approaches designate the infrastructure, technologies, and types of services that deliver water, including piped or non-piped provision [11, 19]. The previous three categories exist within a broader context that helps to define why elements have evolved in a certain way and what opportunities for improvement are feasible [11]. The social, economic, and political context includes political agendas, economic trends, and social norms. Finally, the environ- mental and resource context recognizes that water services are closely linked with local ecosys- tems and global climate change, with factors such as water scarcity, groundwater table depth, soil types, and pollution affecting decisions regarding water sources, treatment and distribu- tion approaches, and conservation measures [11, 19]. 2.2. General literature review on the enabling environment for urban water service improvements To begin our process of understanding the enabling environment for urban water provision, we reviewed existing peer-reviewed and grey literature concerning urban water provision, quality, policies, governance, pro-poor mechanisms, and related topics (e.g., [1, 3, 6, 8–10, 12, PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 3 / 29 PLOS WATERThe enabling environment for citywide water service provision Fig 1. A version of the social-ecological systems (SES) framework modified to focus on governance, actors (e.g., service providers, regulators, policymakers, consumers), and contextual factors, adapted from previous work of one of this study’s authors [11]. https://doi.org/10.1371/journal.pwat.0000071.g001 17, 18, 24–43]). Search terms used to identify literature resources included “urban water,” “water policy,” “urban drinking water quality,” and similar phrases. We also included addi- tional resources referenced in key papers from the searches, as well as those recommended by experts interviewed during the case identification process (see Section 2.3). Generally, we focused on materials published within the past 10 years (over 80% of reviewed references were published after 2012), aiming to develop a picture of the current body of knowledge. We cate- gorized literature findings according to the SES framework, separating well-established knowl- edge from unresolved debates (Tables A and B in S1 Text; these tables also identify relevant second-tier variables from the original SES framework [21]). This literature review provided a lens for our subsequent case study analysis. 2.3. Case study identification and selection For our comparative case study analysis, we selected six cities from low- and middle-income countries that have demonstrated historical success in providing inclusive city water services. This number of cases allowed us to capture wide-ranging contexts while also exploring the details of each city in depth. Our identification and selection process involved three steps, described in detail below: (i) shortlisting candidates based on expert interviews and literature review; (ii) screening shortlisted candidates for eligibility; and (iii) selecting a subset of eligible candidates to capture a diversity of contexts (Fig A in S1 Text). As described below, the first two steps represented a two-pronged approach in which we balanced the nuanced (but subjec- tive) information provided by expert knowledge with the objective (but less nuanced) data used for screening. We shortlisted 18 candidate cities based on seven expert interviews and our review of exist- ing literature (Section 2.2). Experts interviewed were individuals having several years of experi- ence working and conducting research in the international urban water sector, and each individual provided verbal consent to contribute to the study (Table C in S1 Text). Interviews PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 4 / 29 PLOS WATERThe enabling environment for citywide water service provision were unstructured and began with the following prompt: “What cities do you know that have made meaningful progress toward safe water for all?” During each discussion, we intentionally kept the definition of “meaningful progress” broad to allow for various forms of progress and success, though we probed on specific topics such as water service coverage levels, water qual- ity, reliability, non-revenue water, cost recovery, and affordability [1, 10]. We also specified that we were particularly interested in progress with regard to equity and inclusion of low- income urban residents [1], emphasized in the original prompt with the phrase “for all.” Addi- tionally, we asked interviewees about specific cities that were identified in the literature, if they did not surface naturally during the conversation, and we reviewed additional resources rec- ommended by interviewees if we had not previously identified them through our literature searches. After shortlisting, we removed five cities that did not meet our eligibility criteria due to (i) a lack of demonstrated citywide improvements in service provision, or (ii) a relatively high eco- nomic level during the time of improvements. In terms of improved service provision, we focused primarily on coverage of on premises piped water in each city’s subnational region, as reported by the Joint Monitoring Programme (JMP). This dataset provided a consistent indicator associated with a representative sample of the population across time and across contexts. While consider- ations such as water quality, reliability, and affordability are also critical elements of successful ser- vice delivery, we were unable to find consistent data across all 18 shortlisted cities. To avoid the possibility of incomplete datasets affecting our selection, we chose to focus on an indicator that was more broadly available (access to on premises piped water). Accordingly, we required that cities met at least one of the following two conditions, based on available JMP data in subnational regions that contained each city: (i) coverage of piped water on premises was higher than 65%, according to the latest data from 2014–2019 (the spe- cific year varied by country); (ii) average annual improvement in coverage of piped water on premises was above one percentage point per year from the earliest (2000–2006) to the latest available data (2014–2019). We also consulted other literature to confirm the trends observed in the JMP data, as some of the JMP’s subnational regions did not correspond precisely to city boundaries. The first condition employed a coverage threshold of 65% because it was the aver- age across our 18 shortlisted cities, and there was a clear break in the data at this point. Four cities met neither of these conditions (Fig A in S1 Text). With regard to economic level, we aimed to find examples of success with limited economic resources. Accordingly, we excluded cities where: (i) the country was classified by the World Bank [44] as upper-middle income or higher, both presently and during the period when most water service improvements occurred in the city; and (ii) the country had a gross national income (in constant 2015 dollars) [45] higher than present low- and lower-middle income countries during the city’s period of water service improvements (i.e., larger than 3,590 USD per capita, in constant 2015 dollars). One city met these exclusion criteria (Fig A in S1 Text). Finally, from the 13 remaining cities, we selected six to represent a diversity of contexts. We first considered city- and national-level characteristics related to demographic, economic, political, and environmental factors, including: city population [46], growth rate of the city’s population [46], national urban informality [47], national democracy index [48], national income level [44], and national water stress [49]. With the aid of principal component analysis, we visualized the distribution of cities across all of these characteristics and identified four clusters with similar features (Fig B in S1 Text). We selected a final set of six cities pulled from all four clusters, with additional consideration for geographic diversity and water service levels. We did not explicitly aim to identify and highlight different forms of progress or specific char- acteristics associated with distinct modes of service provision and governance. From the clus- ter defined by large city populations (Fig B in S1 Text), we selected Bangkok (Thailand) as the PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 5 / 29 PLOS WATERThe enabling environment for citywide water service provision highest performing city in the cluster with respect to water access, and we also chose Ahmeda- bad (India) to provide geographic representation from South Asia. From a second cluster defined by rapid population growth and high levels of informality (Fig B in S1 Text), we selected Phnom Penh (Cambodia) and Abidjan (Cote d’Ivoire). In particular, we chose Abi- djan over Dakar, another high performer in sub-Saharan Africa, because literature reported that water prices were more affordable in Abidjan [50]. From a third cluster defined by rela- tively higher national income levels and democracy indices, we selected Porto Alegre (Brazil) due to its high levels of water access and because it provided geographic representation from South America. Finally, from a fourth cluster defined by high water stress, we selected Cairo (Egypt) because of its high performance. This selection process provided our six cities for the comparative case study analysis (Fig C in S1 Text). 2.4. Case study characterization For each of the six selected case study cities, our analysis focused on a review of existing case- specific literature, supplemented by interviews with 1–3 key informants (Table C in S1 Text). Literature reviews included 26–39 pieces of peer-reviewed and grey literature per case and aimed to provide as comprehensive a picture as possible of institutional arrangements, contex- tual factors, and other drivers of improvements across all categories of our modified SES framework (Table D in S1 Text). Qualitative interviews with key informants were semi-struc- tured and often focused on filling specific information gaps or confirming and providing updates on the evidence found in existing literature. Key informants were local experts, includ- ing staff of utilities or regulatory agencies, government ministry officials, academics, in-coun- try staff of donor agencies or non-governmental organizations, and consultants (Table C in S1 Text). Generally, we identified key informants by emailing local institutions that the literature showed to be important players in the context. Interviews took place virtually in English, French, or Portuguese (with translation) and lasted 30–90 minutes, with each interviewee pro- viding verbal consent to participate in the study at the start of the interview. From the litera- ture and interviews, we developed for each case: (i) a timeline of key events (Figs D through I in S1 Text); (ii) an institutional map characterizing roles, responsibilities, and relationships among key actors [25] (Fig 2; Fig J in S1 Text); and (iii) a case narrative describing important elements associated with each category of the modified SES framework (Table D in S1 Text). In particular, the institutional maps, constructed using methods described by Rahman et al. [25], were complementary to the SES framework. These maps enabled us to visualize roles, responsibilities, and relationships across key actors. 2.5. Comparative case study analysis Using these three elements (timelines, institutional maps, and case narratives) as inputs, we tabulated case information and performed a comparative analysis, using a process of inductive theory building [51] to identify common or contrasting elements and drivers of success across the six cases (Tables E through I in S1 Text). Within each SES category, we qualitatively char- acterized how each case resonated with key literature themes and knowledge gaps (Tables A and B in S1 Text). We also specified any additional key elements that surfaced from each case, beyond the existing topics described in the general literature. Within each SES category, we also identified those elements that appeared to be especially critical in enabling, supporting, or driving progress in urban water service provision during each case’s period of improvement, with special emphasis on equity and inclusion for low-income residents. We then used these tables (one for each SES category; Tables E through I in S1 Text) to define patterns, particularly focusing on elements that played a similar role across multiple cases, either generally or in PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 6 / 29 PLOS WATERThe enabling environment for citywide water service provision Fig 2. Examples of institutional maps [25] associated with each of the three types of progress: (a) Utility-driven: Phnom Penh; (b) Regulator- supported: Cairo; (c) Municipality-driven: Ahmedabad. Fig J in S1 Text shows institutional maps for the remaining three cases. https://doi.org/10.1371/journal.pwat.0000071.g002 specific types of contexts. Following the example of case study research that contributed to the development of the SES framework [52], we formulated these common elements into charac- teristics of the enabling environment across cases. As noted when discussing the limitations of this study, we cannot state definitively that these characteristics are necessary or sufficient in different contexts, but they provide a foundation for understanding based on the existing liter- ature and additional insights gleaned from our six cases. PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 7 / 29 PLOS WATERThe enabling environment for citywide water service provision 3. Results 3.1. Characteristics of study cities Our six case study cities were Abidjan (Cote d’Ivoire), Ahmedabad (India), Bangkok (Thai- land), Cairo (Egypt), Phnom Penh (Cambodia), and Porto Alegre (Brazil), capturing the conti- nents of Africa, Asia, and South America (Fig C in S1 Text). Four cities were in countries currently classified as lower-middle income (Cambodia, Cote d’Ivoire, Egypt, India). The remaining two were in upper-middle income countries (Brazil, Thailand), though water ser- vice improvements had started in periods with lower income levels. The six cities also exempli- fied a variety of demographic, social, political, and environmental characteristics, such as urban informality and water stress (Table 1). Population estimates in 2018 ranged from Table 1. Summary of selected case study cities. LIA: low-income area. Location Population in 2018a Average annual growth, 2013– 2018a National income levelc National water stress in 2018d National urban informalitye Citywide coverage of household piped water connections (%)f 1996– 2010 Before 1996 2010– 2022 Abidjan, Coˆte d’Ivoire Ahmedabad, India Bangkok, Thailand 4,920,776 7,680,935 2.8% 2.6% 10,156,316 2.6% Cairo, Egypt 20,076,002 2.2% Phnom Penh, Cambodia Porto Alegre, Brazil 1,952,329 3.2% 4,094,398 0.8% Lower- middle Lower- middle Upper- middle Lower- middle Lower- middle Upper- middle 5.1% 66.5% 61.1% 34.8% 23.0% 24.5% 141.2% 3.1% 29% (1989) 65–87% (mid- 1990s) 82% (1995) 63% (1998) 85% (2010) - 64% (1979) 97% (1996) 90.8% (2016) 85% (2014) >90% (2019) 90– 100% (2018) 1.0% 45.6% 25% (1993) 90% (2006) 90.5% (2014) 1.4% 15.2% 94.7% (1989) 99.5% (2001) 99.5 (2022) Time period and primary modes of water service improvementsg 1989–2005 (connection subsidies for LIAs) [50, 53, 54] 1995–2014 (integrated slum upgrading and other LIA schemes) [55, 56] 1985–2003 (LIA delivery schemes, efficiency and reliability reforms) [8, 57–59] 1979–1996 (coverage) 2004–2013 (quality and reliability reforms, provision in LIAs) [60–63] 1993–2006 (citywide expansion and reforms) 1998–2015 (provision in LIAs) [8, 59, 64] 1960–1989 (citywide expansion) 1988–2004 (LIA prioritization) [65–67] a Population and annual growth were taken from statistics on urban agglomerations in the United Nations’ World Urbanization Prospects [46]. Note that the boundaries of urban agglomerations were based on population density thresholds, which may have differed from administrative boundaries. b Unless otherwise noted, access to piped water on premises and sanitation were from the latest available Joint Monitoring Program survey data for the subnational regions in which cities were located (accessed April 2022) [2]. c Income levels correspond to World Bank classifications [44]. d Water stress is defined as the proportion of available freshwater resources being withdrawn. It is the ratio between total freshwater withdrawn and total renewable freshwater resources, after accounting for environmental flow requirements [49]. e The share of urban population living in slum households per country, based on 4 out of 5 household shelter deprivations defined by UN-Habitat as indicators of informality: lack of access to improved water, lack of access to improved sanitation, lack of sufficient living area and quality/durability of structure. Security of tenure is the fifth deprivation that is not included due to data limitations [47]. f Data on household piped water coverage was derived from a variety of sources across cases [8, 9, 50, 53–57, 59, 61, 63–65, 67–83]. Generally, these sources did not specify whether the reported coverage levels included informal or low-income areas. We found limited information on coverage within low-income areas specifically: coverage of piped water on premises was 57% in low-income areas of Abidjan in 2018 [68] and was 60% in Ahmedabad in 2014 [55]. g Time period of improvements was determined based on available information in case-specific literature. The coverage data provided in the preceding columns does not always encompass the entire time period of improvements, because specific coverage data was not always available before and after these periods. Accordingly, these coverage levels should not be seen as definitive “baseline” or “endline” data points. Generally, qualitative information from the literature guided our understanding of the timing, type, and drivers of improvements. https://doi.org/10.1371/journal.pwat.0000071.t001 PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 8 / 29 PLOS WATERThe enabling environment for citywide water service provision approximately 2 million in Phnom Penh to 20 million in Cairo, according to urban agglomera- tion data from the United Nations’ World Urbanization Prospects [46]. By 2014, all study cities had achieved at least 85% coverage of piped water on premises, with some cities attaining these high levels much earlier, such as Porto Alegre in 1989 and Cairo in 1996. Coverage data specific to low-income areas was limited, but we found 57% coverage of piped water on premises in surveyed low-income areas of Abidjan in 2018 [68] and 60% in Ahmedabad in 2014 [55]. Most of the cities also performed well with respect to financial man- agement and reliability of water services. Based on reported data from 2005 to 2018 (depend- ing on the city; Table J in S1 Text), all cities except Bangkok were recovering at least 99% of operating costs, while all except Ahmedabad provided reliable water at least 20 hours per day. Phnom Penh performed especially well with respect to non-revenue water, which decreased from 72% in 1993 to only 6% in 2006 and has remained low since then (e.g., 8% in 2018) [81]. Non-revenue water among the remaining five cities ranged from 20% to 34% (Table J in S1 Text), similar to a recent global estimate of 30% [84]. Our six cases all exhibited unique characteristics with respect to the enabling environment for urban water service provision, including governance, key actors, approaches to service delivery, and the social, economic, political, and environmental context (Table 2, Figs D through I in S1 Text). Below, we describe the findings from our comparative analysis on how these enabling environments contributed to success in these cities. First, we define three over- arching types of progress that emerged from the six cases, representing different ways in which certain types of actors and institutional arrangements drove improvements in water access and service quality. Subsequently, we present characteristics, categorized according to our modi- fied SES framework. 3.2. Types of progress toward inclusive urban water services From our six cities, we identified three overarching types of progress, distinguished by the kinds and roles of actors that drove and supported water service improvements, as well as the mechanisms for oversight and monitoring (Fig 2, Fig J in S1 Text). We have labeled these types utility-driven, regulator-supported, and municipality-driven (Table 2), each character- ized by a distinct set of features and drivers (Fig 3). 3.2.1. Utility-driven. Bangkok, Phnom Penh, and Porto Alegre provided examples of util- ity-driven cases. In these cases, proactive utilities with autonomy and strong leadership drove progress through measures such as institutional reforms, increased metering, appropriate tariff increases, and public engagement [8, 17, 59, 65, 67, 83, 85–88]. A lack of political interference provided these utilities with freedom to pioneer the reforms that improved services [8, 17, 89, 90]. Rather than independent regulators, the utilities’ internal bodies provided oversight and guidance [66, 78, 82, 83, 91, 92]. Despite being internal, these entities had some degree of sepa- ration from operations (Boards of Directors in Bangkok and Phnom Penh, a Deliberative Council in Porto Alegre), with members representing various political and civil society inter- ests [66, 78, 82, 83, 91–93]. For example, in Phnom Penh, the Board of Directors included representation from two national ministries, the municipal government, and utility employees [91, 92], while in Porto Alegre, civil society organizations nominated members of the Delibera- tive Council [66, 82, 83]. The individual cases offered similar but distinctive stories. Bangkok’s story centered on a national financial crisis in 1997, which spurred the utility–the Metropolitan Waterworks Authority (MWA)–to resist political pressure to privatize by instituting a series of reforms that improved efficiency and customer relations [57, 89]. Governor Chuanpit Dhamasiri, who led MWA through this period, later referred to the crisis as a “blessing in disguise” because it PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 9 / 29 PLOS WATERPrimary type of progress Governance Actors Service delivery approaches Social, economic, & political context The enabling environment for citywide water service provision Table 2. Key contextual characteristics and drivers of progress across case study cities, categorized according to the modified SES framework. Empty cells indicate a lack of available data related to the corresponding SES category in a given city. Additional details on historical events and conditions can be found in the city timelines (Figs D through I in S1 Text). NRW: Non-revenue water. IBT: Increasing block tariff. City Abidjan, Coˆte d’Ivoire Ahmedabad, India Bangkok, Thailand Cairo, Egypt Phnom Penh, Cambodia Porto Alegre, Brazil Regulator-supported Municipality-driven Utility-driven Regulator-supported Utility-driven Utility-driven Regulation through performance contracts with public agencies Devolution to local authorities with clear mandates for serving the poor Service provider autonomy, with internal Board of Directors oversight National regulatory agency monitors performance Service provider autonomy, with internal Board of Directors oversight Internal Deliberative Council oversight Private company provides services with public asset ownership Municipal government provides services directly and self- regulates Champion-led institutional reforms at autonomous local public utility Reforms made local utilities subsidiaries of national holding company Champion-led institutional reforms at autonomous local public utility City water department transformed to autonomous local public utility National company operates urban piped systems under a countrywide IBT, with connection subsidies for low- income residents; informal resale of water is common Complex arrangement of urban and peri-urban low- income areas; displaced persons from civil wars increased size of low- income areas Municipality provides piped water services funded through property tax surcharges, with connection subsidies for low-income residents; integrated upgrading programs in low- income areas Municipality guaranteed periods of tenure security and removed tenure requirements for connections; decline of local textile sector impacted low- income industrial workers Groundwater over- extraction; climate will likely increase existing susceptibility to flooding, droughts, and water scarcity Local utility provides piped water services under an IBT; utility-led reforms reduced NRW and improved reliability, while national government upgraded and improved services in low-income areas Financial crisis created push for privatization, which was resisted Local utility operating under the national holding company provides piped water services; national and local government promoted upgrading of low- income areas, where many residents depend on private wells Large foreign investment; external push for privatization was resisted; removed tenure requirements for connections Local utility provides piped water services under an IBT, with a connection subsidy program for low- income residents; utility-led reforms reduced NRW and improved reliability Local utility provides piped water services under an IBT, with services for low- income residents improving through participatory budgeting External funding supported reforms Participatory budgeting has increased equity - Arid climate with high levels of water stress and heavy, nearly exclusive reliance on the Nile; these conditions have factored centrally in national development plans Flooding and pollution have been persistent issues; frequent flooding has created a perception of water as an inexhaustible resource; groundwater use reduced after over- extraction led to land subsidence Historically, faced extreme flooding; more recently, a push toward an integrated water cycle vision aims to control pollution Environmental & resource context Peri-urban communities along local lagoon face issues of pollution https://doi.org/10.1371/journal.pwat.0000071.t002 prompted the restructuring that supported the utility’s continued success [59]. In Phnom Penh, development partners including the Asian Development Bank and the World Bank doc- umented the effective leadership of Ek Sonn Chan, director of the Phnom Penh Water Supply Authority (PPWSA) from 1993 to 2012, largely crediting him with the utility’s success [17, 24, 59, 91, 94, 95]. Similar to MWA in Bangkok, PPWSA implemented several reforms during Director Chan’s tenure to increase efficiency, transparency, and financial sustainability, after a governmental decree in 1996 granted the utility financial and administrative autonomy. It is worth noting that the beginning of Director Chan’s leadership coincided with the end of the country’s civil war and an increase in foreign assistance, two important factors that likely also contributed to success in improving water service provision. In Porto Alegre, the city’s water PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 10 / 29 PLOS WATERThe enabling environment for citywide water service provision Fig 3. Drivers promoting an effective enabling environment for urban water provision specific to certain types of progress, including utility-driven (A), regulator-supported (B), and municipality-driven (C). For each driver, we note the cities that provided key examples. https://doi.org/10.1371/journal.pwat.0000071.g003 department transformed into an autonomous, financially independent, and municipally- owned utility (DMAE) in 1961, which expanded coverage to a high level (>94%) by 1989 [66, 82, 83]. The city’s prominent participatory budgeting process, initiated by the municipal gov- ernment from 1988 to 2004, was also instrumental in prioritizing and addressing water needs in underserved low-income communities [65, 67, 82, 96]. 3.2.2. Regulator-supported. We identified regulator-supported types of progress in Abi- djan and Cairo. Independent regulatory institutions providing strong monitoring and sector PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 11 / 29 PLOS WATERThe enabling environment for citywide water service provision coordination were characteristic of this type. In both cities, the utility annually reported water quality, reliability, and financial performance data, among other information, to national-level regulators [50]. Notably, however, both of these cases also saw substantial expansions in piped water cover- age prior to the introduction of independent regulators. Given these earlier periods of improvement, we could have alternatively designated these cases as central government-sup- ported, with centralized state agencies driving infrastructure expansion primarily through donor funding. In Abidjan, a national-level utility (Socie´te´ de distribution d’eau de la Coˆte d’Ivoire, SODECI) operated under a public-private partnership established when Cote d’Ivoire gained independence in 1960, wherein government agencies were responsible for infrastruc- ture investment and development while the private partner was responsible for operations and maintenance [50, 69]. Eventually, overlapping and unclear responsibilities across state actors led the central government to create the regulator (Office National de l’Eau Potable, ONEP) in 2006. Along with its regulatory role, ONEP also oversaw infrastructure development and asset management, providing an additional degree of public control over SODECI’s private opera- tion of the infrastructure and making the regulator central to the effective functioning of the public-private partnership [97–99]. Until 2004 in Egypt, local water and sewer authorities were responsible for delivering ser- vices, providing high levels of coverage in Cairo but depending heavily on donor funding [60, 61, 100–103]. In that year, the national government restructured the sector to establish a national holding company that managed all water and wastewater assets in the country, with all local authorities converted into subsidiary companies that still provided services. This restructuring also created an independent national regulator (Egyptian Water and Wastewater Regulatory Agency, EWRA) to oversee the companies’ performance, determine standard ser- vice levels, and establish performance indicators.[63, 104]–a reform that international donors had been encouraging as early as the 1980s [100–102]. EWRA’s monitoring of financial indica- tors and new tariffs approved by Egypt’s Cabinet of Ministers have improved the sector’s eco- nomic performance, with all subsidiary companies in the Greater Cairo Region recovering more than 100% of operating costs in 2018 [63, 79]. 3.2.3. Municipality-driven. Finally, Ahmedabad represented a municipality-driven case, in which the Ahmedabad Municipal Corporation (AMC) was directly responsible for provid- ing water services and used its role as the local government to proactively promote water access through various integrated initiatives. In particular, AMC leveraged property taxes–as opposed to consumption-based tariffs–to finance water services and implemented holistic informal set- tlement upgrading programs, which incorporated improvements in water provision alongside other urban services [56, 71, 105]. These programs followed directly from AMC’s established urban governance principles, which included a focus on addressing the needs of the urban poor, and from policies at the national and state levels that gave municipalities clear mandates to engage in pro-poor support to fulfill their constitutional responsibilities [56, 71, 74, 76]. Similar to the internal oversight bodies from our three utility-driven cases, AMC had a sepa- rate elected wing to monitor performance, impose sanctions, and confer awards concerning the services being provided by the council’s administrative wing [106]. The fact that these elected municipal officials likely received electoral benefits from improvements in pro-poor service provision may have increased their focus on such efforts. Additionally, non-govern- mental organizations (NGOs) played a major role in holding the municipality accountable during implementation of the upgrading programs. As key implementation partners, they monitored the work of private contractors and trained community leaders in documenting and reporting problems to reduce occurrences of corruption. Partnership between NGOs and public officials also created a culture of mutual respect between different actors and PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 12 / 29 PLOS WATERThe enabling environment for citywide water service provision strengthened public commitment to pro-poor service provision [107]. Finally, effective tax col- lection systems and financial transparency raised the city’s credit rating and enabled the use of municipal bonds to fund infrastructure improvements, further contributing to AMC’s success [71]. 3.3. Characteristics of enabling environments for inclusive urban water services Across all types of progress demonstrated by the six cases, we identified a total of 12 character- istics that contributed to effective enabling environments for successful and inclusive provision of urban water services (Fig 4). These characteristics captured all five categories from our mod- ified SES framework, which we used to guide the analysis. We also employ these categories here to structure our findings. Fig 4. General characteristics to promote an effective enabling environment for urban water provision across types of progress. We noted the cities that illustrated each characteristic. https://doi.org/10.1371/journal.pwat.0000071.g004 PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 13 / 29 PLOS WATERThe enabling environment for citywide water service provision 3.3.1. Governance. Under the Governance category of our modified SES framework, the use of clear performance indicators was critical to monitoring and incentivizing high levels of performance among service providers, regardless of the specific monitoring arrangements employed across all three types of progress (Fig 4) [8, 59, 64]. Performance indicators employed in our cases focused on topics such as coverage, water quality, reliability, non-reve- nue water, and cost recovery [8, 59, 64]. Service providers typically submitted annual reports to regulators or other oversight bodies, detailing their performance levels relative to the indica- tors. In Egypt, the regulatory agency also produced its own nationwide reports, which included additional water quality sampling and analysis conducted by its own staff [63]. Beyond external performance indicators, internal accountability mechanisms and incen- tives for service provider staff also contributed to improved service delivery. Our utility-driven cases such as Bangkok and Phnom Penh put in place clear codes of action and strict penalties for corruption to increase efficiency and transparency, as well as increased employee salaries [8, 69, 78]. Similarly, the utility in Abidjan (SODECI) offered incentives that helped to increase employee motivation, including high salaries, merit-based promotions, training programs, and funds that paid for social activities and assisted with needs such as housing. Furthermore, SODECI reportedly provided employees with stable jobs during periods when the country’s overall job market was unstable [69]. Along with these measures promoting generally high levels of performance among service providers, additional measures supported inclusive, pro-poor water service provision. These mechanisms included subsidies, explicit mandates to serve low-income residents, and the lift- ing of land tenure requirements for access to piped connections. In our cases, service providers commonly applied consumption subsidies through tariff structures employing some form of social or increasing block tariff, which favored customers with low water consumption [65, 69, 108, 109]. However, existing literature has noted that the poorest residents may not benefit from increasing block tariffs if they are unable to connect to the piped network, household sizes are large, or multiple households share connections [1, 110]. Accordingly, several of our cases have also implemented connection subsidies. In Abidjan, for example, consumption tar- iffs paid by higher-volume customers cross-subsidized connections for low-income residents. From 1986 to 1998, piped coverage expanded considerably in the city, with 87% of new con- nections being subsidized [108, 111]. Similarly, Phnom Penh and Ahmedabad implemented connection subsidy programs and have offered installment plans so that residents could pay the fees over time [58, 112]. Additionally in Ahmedabad, key informants described the approach of funding water services through property taxes instead of consumption-based tar- iffs as inherently pro-poor, since low-income residents typically pay less in property tax. With regard to clear mandates for low-income service provision and lifting of land tenure requirements, Ahmedabad provided a strong example of both. India’s 74th Constitutional Amendment Act of 1992 outlined 18 responsibilities for local authorities, and policies at the national and state levels provided clear mandates for municipalities to engage in pro-poor sup- port to fulfill these responsibilities [56, 71, 74, 76]. Ahmedabad also developed its own urban governance principles, one of which focused on addressing the needs of the urban poor [74], and the city’s boundaries have expanded multiple times to absorb growing peripheral areas [56, 76, 113, 114]. However, even with clear mandates and pro-poor subsidy programs, exist- ing literature has highlighted that ambiguous land tenure can often remain a critical barrier to piped water connections, especially when households must provide proof of land ownership [3, 15, 17]. Ahmedabad instituted a program removing such land tenure requirements in 2002, called the No Objection Certificate Scheme. This program enabled low-income residents to apply for connections if their house was below a certain size and, critically, if they could pro- vide proof of residency–rather than ownership–through an electricity bill, voter identification PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 14 / 29 PLOS WATERThe enabling environment for citywide water service provision card, or other document showing their address [76]. Notably, the Cairo Water Company made a similar shift in 2014, following the city governor’s 2008 statement endorsing the extension of piped networks into areas regardless of legal status [103]. In our cases, these pro-poor measures appear to have been more important in making ser- vices more inclusive than constitutional rights to water. Existing literature has shown that explicit rights to water and sanitation, enshrined in national constitutions or other key policy documents, can sometimes be used in court cases to help improve access, although methods for their application may not always be clear [115, 116]. In most of our cases, constitutional rights to water and sanitation did not exist. Brazil did have a constitutional right to water, which has assisted in some legal cases. However, it also conflicted with the right to private property, allowing utilities to continue requiring proof of land tenure when households applied for piped connections and limiting its impact on water access among the poor [117]. 3.3.2. Actors. Under the Actors category, mechanisms for involving city residents–partic- ularly those living in low-income areas–in service delivery, advocacy, and planning activities were critical in promoting inclusive water provision (Fig 4). Community mobilization efforts in low-income areas can take a variety of forms and occur in partnership with utilities and municipalities. In particular, our cases demonstrated examples of political advocacy, delegated management, and participatory budgeting. Existing literature has highlighted that community groups engaging in collective action can increase recognition and progress for low-income areas [15], and we saw similar patterns in Abidjan, Ahmedabad, and Bangkok. Residents of Abidjan’s informal urban settlements, characterized by uncertain legal status, reportedly engaged in efforts to improve services more effectively than legally secure peri-urban fishing villages, in part due to organization through community leaders to develop connections with local government officials [9]. As part of Ahmedabad’s slum upgrading projects, democrati- cally elected neighborhood associations mobilized the necessary finances from within the com- munity, playing an instrumental role in expanding water infrastructure [72, 75, 113]. In Bangkok, low-income community groups worked with the utility to set up what were essen- tially delegated management models. Locally-elected committees managed the area’s network, collected connection fees and tariffs from residents, paid bulk tariffs to the utility, and used surplus funds to support other community projects [58]. Finally, Porto Alegre’s participatory budgeting processes enabled low-income residents to contribute to the prioritization and plan- ning of improved water services, resulting in better outcomes for areas that had largely been left out of previous water system expansions [67]. Beyond specific community mobilization efforts, broader mechanisms for gathering public feedback have also been important in ensuring that service providers were effectively meeting residents’ needs. The utilities in both Bangkok and Phnom Penh developed such mechanisms, with feedback in Phnom Penh leading PPWSA to establish easily accessible payment centers in low-income areas [95]. Outside of these two utility-driven contexts, other mechanisms for public feedback included a consumer relations center set up in Abidjan by the regulator (ONEP) and specific rights for civil society organizations to mediate between communities and the municipality in Ahmedabad [71, 75]. As in other locations [116, 118], these public engagement and accountability measures helped to ensure that service providers meet needs on the ground, including in low-income areas. 3.3.3. Service delivery approaches. Increased metering of piped networks has contributed to substantial reductions in non-revenue water and improvements in service reliability. The literature identifies non-revenue water and reliability concerns, caused by issues such as leaks and illegal connections, as key challenges for piped water systems [1, 8]. The Bangkok and Phnom Penh cases showed that increases in metering, coupled with utilities’ efficiency reforms and transparent reporting of illegal connections, have successfully addressed non-revenue PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 15 / 29 PLOS WATERThe enabling environment for citywide water service provision water and reliability concerns. From 1993 (the beginning of Director Chan’s leadership in Phnom Penh) to 2006, PPWSA reduced non-revenue water from 72% to 6% and increased water supply durations from 10 to 24 hours per day, while the city’s piped coverage area rose from 25% to 90% and the total number of connections increased by more than a factor of five [64]. Similarly in Bangkok, the reforms arising from the 1997 financial crisis reduced non-rev- enue water from 40% to 30%, with water flowing 24 hours a day [8, 59]. In contrast, Ahmeda- bad has continued to provide intermittent supply, with piped water only available for an average of two hours per day in 2010 [76]. Around the same time, non-revenue water was esti- mated to be 25–31%, though this metric has been difficult to measure due to a lack of metering [76, 119, 120]. The municipality’s use of taxes to fund operations–as opposed to consumption tariffs–may have lessened the perceived need for meters from a direct financial standpoint [74, 119], potentially also contributing to the city’s intermittency problems due to a limited capac- ity to detect leaks or illegal connections. 3.3.4. Social, economic, and political context. A key feature of the social, economic, and political context concerned the sources and methods used to cover the water sector’s capital investments and operational expenses. Consistent with the literature [14, 17, 110], interna- tional support often contributed to capital infrastructure development in our cases, while con- sumption-based tariffs have typically funded operations. Alignment with donor countries’ economic or political priorities have led to strong funding relationships. For example, until the civil wars beginning in 2002, Cote d’Ivoire’s political stability, pragmatic policies toward for- mer colonial powers, and capitalist priorities led to sustained relationships with international organizations and nations such as France [69]. Similarly in Egypt, President Al-Sadat’s Open Door Policy in 1974 led to increased alignment with the United States and Europe and high availability of foreign capital for infrastructure development [60, 100]. In Cambodia, following the Paris Peace Agreement that ended the country’s civil war in 1992, the lifting of trade embargoes made possible international aid to support the country’s reconstruction [121]. Director Chan, whose tenure at PPWSA began in 1993, capitalized on this political and eco- nomic climate by working with development agencies and Cambodia’s newly elected govern- ment to support the utility’s institutional reforms [8, 24, 121]. Concerning operations, most cases have achieved financial sustainability through tariffs. Phnom Penh saw particular success in raising local tariffs to financially sustainable levels through a gradual, stepwise process that involved educational campaigns, political support, and evidence of improved utility efficiency and performance to promote public acceptance [17]. In Cote d’Ivoire, nationally-set tariffs enabled the distribution of revenues across various national actors to fund operations as well as some level of infrastructure investment and debt servicing [69]. Moving beyond financial viability, political practices with strong democratic elements, par- ticularly geared toward citizen participation in broad local planning processes, supported greater equity in service provision (Fig 4). The participatory budgeting processes applied in Porto Alegre provided a key example of these types of practices. Participatory budgeting was introduced by the local Workers Party, which had a strong commitment to democracy and came to power in 1989. In particular, the party aimed to create an arena for low-income resi- dents to contribute to spending decisions that would address inequities, and this political com- mitment has been cited as the key driver of the effort’s success [66, 67]. Participants consistently prioritized water and sanitation, and access to these services increased in histori- cally underserved neighborhoods [67]. In Ahmedabad, India’s longstanding tradition of col- lectivist and movement-based struggles dating back to the country’s independence contributed to the municipality’s leadership and vision in pursuing participatory projects that improved basic services in low-income areas [56, 122]. Finally, Phnom Penh’s utility-driven progress showed the value of politicians, including the city’s governor and the country’s prime PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 16 / 29 PLOS WATERThe enabling environment for citywide water service provision minister, approving of the utility’s increased autonomy and supporting the need for increased tariffs, metering, and identification of illegal connections [17]. Similarly, a growing recognition of the complex realities associated with informal areas [36, 42] have supported shifts in public mindsets toward greater rights and levels of services in these locations. Municipal governments in Cairo and Bangkok moved away from classifying these areas as “illegal” and saw the creation of dedicated programs focused on improving ser- vices [123, 124]. Bangkok provided a particularly early example of these shifting mindsets in the 1970s, when popular views evolved to see low-income residents as having a right to basic services [58]. In Phnom Penh, the municipal government (supported by UN-Habitat) devel- oped an Urban Poverty Reduction Strategy in 1999 that ended the practice of treating squatters as illegitimate residents and that decentralized some decision-making power to communities [125]. Finally, Ahmedabad has also been a pioneer in this area, ensuring service provision for peripheral low-income areas through holistic slum upgrading programs coupled with official expansions of city boundaries [56, 71, 76, 105, 113, 114]. 3.3.5. Environmental and resource context. Water pollution concerns were a common element contributing to advancements in water supply and (waste)water treatment. In Ahme- dabad, pollution of the Sabarmati River led to additional sewage treatment plants [76], while surface water pollution associated with severe flooding in Bangkok led the city to change its water supply and treatment approaches [77]. In Porto Alegre, the presence of blue algae in Guaiba Lake, where waste was discharged, led to the development of an “integrated water cycle vision” covering the sanitation chain from household sanitation infrastructure to dis- charge of effluent wastewater [82], with the goal of better protecting the city’s water resources. Beyond water pollution concerns, the occurrence or possibility of broader environmental shocks also spurred efforts toward increased water resilience and security. Bangkok’s overex- traction of groundwater, which led to land subsidence that damaged the pipe network, com- bined with flood events that were made more severe by land subsidence, spurred changes in the city’s water supply approach. In particular, a 1983 cabinet resolution aimed to phase out groundwater use [89, 126]. While the utility did not fully reach the target, less than 0.4% of its water supply came from groundwater by 2001 [8, 126]. This example is consistent with litera- ture noting that natural disasters can highlight weaknesses in existing systems and focus atten- tion on necessary improvements [6]. Cairo was selected as one of our six cases primarily due to its high levels of water stress. The Nile River provides the main water source in an extremely arid climate, and this fact has driven a number of development decisions within the country [62, 127]. Most recently, the Grand Ethiopian Renaissance Dam–under construction since 2011 –is expected to reduce the Nile’s flow into Egypt by 25%, exacerbating concerns regarding the country’s large water demands [127]. As a result, Egypt’s National Water Resources Plan 2037 and National Vision 2030 both included provisions focused on developing new water supplies, advancing water purification technology, managing demand, and increasing wastewater reuse to improve water security [79, 127]. As of 2020, treated wastewater from Cairo has reportedly been used to irrigate tim- ber and non-food crops [127]. While water reuse for agricultural irrigation does not directly contribute to urban service provision, it does advance overall water security in Cairo by lower- ing demands on Nile water. 4. Discussion While our findings generally did not contradict prior literature, our analysis added three key contributions: PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 17 / 29 PLOS WATERThe enabling environment for citywide water service provision i. it addressed several unresolved debates from the prior literature, such as how to effectively monitor service performance in the absence of an independent regulator, how to ensure progress achieved through individual champions is sustainable, how certain institutional arrangements can promote integrated upgrades for low-income areas, strategies for remov- ing land tenure requirements for piped connections, and the importance of explicit pro- poor measures and mandates as compared to general constitutional rights to water (see Sec- tions 4.1 and 4.2 for more detail, as well as Table B in S1 Text); ii. it consolidated prior knowledge as well as new insights derived from the six study cities into a unified list of key characteristics supporting inclusive improvements in water service provision (Fig 4); and iii. it defined a typology to categorize success pathways according to the institution driving reforms (i.e., utility/service provider, regulator, municipality/local government; Fig 3), which suggests that multiple potential entry points exist to catalyze progress. Generally, our six cases showed that a variety of factors and approaches may effectively sup- port safe and inclusive services, with the most appropriate strategy depending on existing insti- tutional arrangements, infrastructure, and the surrounding social, economic, political, and environmental context. 4.1. Overarching lessons from the three types of progress Overall, the three types of progress identified in our six cases suggest that improvements in urban drinking water access may occur in different ways, depending on existing institutional arrangements and contextual factors, and they offer multiple entry points (e.g., service provid- ers, regulators, governments) through which implementing organizations may promote and support advances. Additionally, they illuminate three lessons that contribute toward address- ing unresolved conversations within the literature (Table B in S1 Text), both within our six cases and beyond. First, performance monitoring can take multiple forms and does not neces- sarily require an independent regulator. The literature commonly recommends approaches involving independent regulators as being most effective and transparent [3, 5, 8, 13, 17, 115, 128, 129]. We observed this type of model in Abidjan and Cairo, where independent regulators tracked performance indicators and produced or reviewed annual reports covering multiple locations or utilities throughout the country [50, 63]. Similar approaches also exist elsewhere in sub-Saharan Africa, where stakeholders have focused on developing strong regulatory envi- ronments for water and sanitation [40]. However, other types of performance monitoring, especially those characterized by proactive utilities or municipalities with internal oversight bodies, can also be effective under the right conditions. In particular, the oversight body should have a degree of separation from operational activities to encourage honest reporting and include members representing various stakeholder interests [8, 59, 64]. Second, champions can drive sustainable institutional reforms. While the efforts of passion- ate individuals can move the needle on reforms and can be particularly important in weak institutional or regulatory environments [130], existing literature has noted that champion-led progress may not be sustainable if key individuals retire or change assignments [6, 14, 17, 24, 115]. For example, the reassignment of a proactive ministry secretary who had effectively pro- moted communication in Kenya led to concern among sector actors that progress would stall and priorities would change [115]. However, our two cases in which champions played a prominent role in utility-driven progress–PPWSA Director Ek Sonn Chan in Phnom Penh and MWA Governor Chuanpit Dhamasiri in Bangkok–showed that the improved perfor- mance levels achieved through champion-led utility improvements can be sustainable beyond PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 18 / 29 PLOS WATERThe enabling environment for citywide water service provision the individual’s tenure [17, 59]. These cases provided examples of institutional reforms, staff incentives, and accountability mechanisms that created a lasting culture of transparency and efficiency. Some specific examples included increasing staff salaries, creating customer data- bases and automated billing systems, and establishing incentives for staff to report illegal con- nections [17]–measures which may translate well to low-performing contexts where utilities play a central role in the sector. Director Chan also reportedly took additional steps to prepare the utility for his retirement by training future management staff and promoting the strong institutional culture he created [81]. Finally, municipalities that directly provide services, such as in Ahmedabad, may integrate water provision with other services to holistically upgrade low-income areas. Previous litera- ture has noted that upgrading multiple services simultaneously has often generated advance- ments in water coverage and service levels within low-income areas, for example by installing piped networks when improving road infrastructure [1, 5, 7, 13, 17], but less information exists regarding the institutional arrangements that can effectively support integrated strategies. The institutional arrangement demonstrated by Ahmedabad offers a notable example of how to achieve such upgrades. In locations where the municipality is directly responsible for deliver- ing multiple services–and especially when a clear mandate exists to serve the poor–it can reduce the need for high levels of coordination across a number of institutional actors that may have different mandates and incentives. Ahmedabad also showed, however, that coordi- nation and partnership with NGOs can enhance accountability and encourage public officials to commit to service improvements [107]. 4.2. High-performing service providers and pro-poor strategies across different types of progress Along with these lessons associated with different types of progress, the 12 characteristics of the enabling environment we identified across our six cases highlighted two broad themes. First, a generally well-functioning water sector is often a prerequisite for inclusive, pro-poor service provision. If utilities, regulators, or municipalities are not operating efficiently and per- forming baseline functions adequately, it is unlikely that they will have the capacity to push for improvements among the poor, who are often the hardest to reach. In particular, elements such as clear performance indicators, institutional reforms for greater efficiency and transpar- ency, customer feedback mechanisms, increased metering to improve reliability, financially sustainable strategies to fund operations and infrastructure investments, and the use of crises to spur progress contributed to the overall success of our cases. Generally, these elements align with existing literature, and international funding institutions often promote many of them (apart from crises) when providing financial assistance [64, 65]. Similar measures could likely also contribute to progress in other, lower-performing settings. Second, realizing more inclusive water services often requires the application of explicitly pro-poor policies and strategies that are contextually appropriate. Effectively serving low- income residents may necessitate dedicated measures that are distinct from conventional approaches used in wealthier parts of the city. These measures might include connection and consumption subsidies, the removal of land tenure requirements for piped connections, com- munity mobilization that enables low-income residents to participate in decision-making and service provision, and democratic practices that support efforts to upgrade low-income areas and validate their residents’ rights to services. However, in our cases, explicit constitutional rights to water were often absent, and when these rights did exist, they tended to be less instru- mental for improving equity than specific pro-poor mechanisms. This point speaks directly to one of the unresolved debates identified from existing literature (see Tables A and B in S1 Text PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 19 / 29 PLOS WATERThe enabling environment for citywide water service provision for additional insights from the case study analysis that relate to elements identified in the liter- ature review). 4.3. Limitations Perhaps the most important limitation of this work is the small (though diverse) number of study cities. Our focus on successful cases limited opportunities to identify direct counterfactuals char- acterized by low performance. Accordingly, while this study provides an important step in identi- fying factors that may contribute to successful provision of urban drinking water, it did not directly test whether the characteristics we identified were necessary or sufficient for promoting success. Future research using fuzzy-set Qualitative Comparative Analysis (fsQCA) [131] and a larger sample size should examine the relative importance of the enabling factors identified here. Additionally, our cases may not have provided insight into the full range of characteristics or types of progress that can drive improvements in water service provision across all settings [132]. Our case selection process explicitly aimed to capture a diversity of contexts, but the process also showed that we could have included a number of other cities, which may have emphasized different elements of the enabling environment. We limited the number of selected cases to six, which we felt enabled us to explore the complexities of each context with a high level of detail while also examining multiple settings. However, each of our cases exhib- ited some degree of path dependence [51], in that institutional arrangements and service char- acteristics were often contingent upon a city’s unique history. Accordingly, future study of additional cities with different historical trajectories may uncover new ingredients for success at work in other contexts. Furthermore, we focused primarily on improvements in piped water access, but successful service arrangements that include non-piped provision may also exist. The literature has highlighted hybrid modes of service delivery, in which various public and private actors pro- vide piped and non-piped services, and scenarios where utilities have been experimenting with approaches drawn from informal providers to expand services in low-income areas [6, 16]. We found some similar examples of hybrid provision in our cases, such as delegated management models in Bangkok and water resellers in Abidjan, but other settings may provide further insight on how utilities and municipalities can work together with private operators to effec- tively serve low-income residents. It is also worth noting that the eligibility screening phase of our case selection process focused on piped water access and did not directly incorporate other important elements of successful service provision such as quality, reliability, and affordability, because we were unable to find data on these indicators across all candidates. Generally, piped water service is viewed as the gold standard, providing the highest quality water at the greatest level of conve- nience [1, 18]. Also, the use of open-ended expert interviews to identify successful cities prior to eligibility screening did provide a qualitative understanding of performance beyond piped coverage. However, not directly including these other elements in the process may have affected our final selections, some of which proved to be less successful with respect to certain indicators. For example, while Ahmedabad has promoted equitable access and inclusion of low-income residents, piped water services across the city were only available for an average of two hours per day in 2010 (Table J in S1 Text). Finally, the limited number of interviews, especially when we were only able to conduct a single interview per case, may have influenced or biased our understanding of that case. We aimed to identify interviewees from institutions with a broad perspective on each context, and we followed up with additional individuals in specific roles when possible. However, it is possi- ble for institutional or personal biases and blind spots to affect perceptions of key drivers and PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 20 / 29 PLOS WATERThe enabling environment for citywide water service provision actors. Conducting extensive literature reviews for each case and framing the interviews as supplemental to the literature helped prepare us to identify and adjust for any potential biases. However, we may still have missed certain influences arising from interviews. 4.4. Possibilities for future research into pro-poor water service provision As stated previously, future research that directly follows from this study could apply fsQCA [131] to evaluate the importance and relationships of the characteristics we identified across a larger number of successful and unsuccessful cases. This type of work could offer insight into the degree to which these characteristics hold true in a broader array of contexts, and how they may work together to contribute to improvements in urban water services. There may also be opportunities to integrate or conduct additional analyses of new, innovative regulatory and service provision models currently being implemented in LMICs, particularly related to city- wide inclusive sanitation and fecal sludge management [5, 110, 116, 118]. Furthermore, a number of additional questions arose from the cases related to the develop- ment or advancement of specific pro-poor measures. First, the electricity sector may offer les- sons on how to lift or bypass land tenure requirements for service connections. Ahmedabad and Cairo both offered clear examples of cases where low-income households with unclear land tenure status have acquired electricity connections ahead of piped water, as they used electricity bills to show proof of residency when applying for water connections [76, 103]. Future research focusing on how the electricity sector has removed land tenure requirements may be useful for the water sector in moving beyond this critical barrier to piped access. Second, better characterizing the incentives that can effectively formalize small-scale pro- viders is important, particularly when piped networks cannot reach certain segments of a city’s population. In Abidjan, the utility began attempting to license water resellers in the early 1980s, but these efforts were largely ineffective due to misaligned or insufficient incentives. Estimates from the late 1990s showed the number of informal resellers to be ten times larger than those with licenses [108, 133]. In such instances, some form of non-piped provision may be necessary, at least in the short term. Finding ways to successfully formalize these actors– into cooperatives, for example–may provide better opportunities for monitoring to ensure safe and affordable services, although literature has noted that pricing control may remain politi- cized in certain contexts [6]. Third, the use of a property tax surcharge to fund water services in Ahmedabad was an interesting alternative to the more common use of consumption-based tariffs [74]. However, it was unclear from our cases how successful this approach would be in LMICs when the munici- pality is not directly providing services, as this scenario would likely require high levels of coor- dination to efficiently and appropriately allocate tax revenues to the service provider. Further research on this topic could be beneficial in increasing the range of options available for achieving financially sustainable operations. Another interesting financial strategy from Ahmedabad involved the use of municipal bonds to enable infrastructure investments [71]. Studying the effectiveness and applicability of municipal bond markets to provide funding for capital improvements in lower-income contexts may also be worthwhile. This area of inquiry may be particularly timely, given recent findings from sub-Saharan Africa related to how con- stitutional and regulatory constraints can limit sub-national governments’ use of municipal bonds [134]. Finally, the adverse impact of unplanned urban expansion on citywide water service provi- sion was clear across our six cases, but we found limited information on cities taking proactive measures towards reducing future unplanned expansion. Ahmedabad was a notable exception: while the city’s administrative boundaries increased periodically to encompass surrounding PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 21 / 29 PLOS WATERThe enabling environment for citywide water service provision low-income areas, the city’s Urban Development Authority also established policies to reduce urban sprawl. It is worth noting that, although the city has successfully promoted equitable access and inclusion of low-income residents, it also provided a citywide average of only two hours of water service per day [56, 114]. Generally, however, the water sector literature on which we focused did not commonly engage with literature from the land use planning sector. Further research may be beneficial in identifying forward-thinking land use strategies to pro- actively address issues of service provision in unplanned areas. 5. Conclusion Achieving equitable access to safe water services remains a key challenge in many urban areas of low- and middle-income countries. Our findings from six cases, each of which has made substantial progress toward this goal, demonstrate that multiple pathways toward improve- ment exist, and the relative importance of the various approaches, measures, and characteris- tics contributing to the enabling environment for urban water provision will vary from place to place. However, key elements identified in this study reflect how different sector players, including utilities, regulators, and local governments, can generate sustainable progress across various contexts, and how interventions in lower-performing cities may target each of these actors as entry points for promoting an effective enabling environment. In particular, utility- focused programming can concentrate on encouraging champions to institutionalize any strategies they use to improve service delivery and efficiency by defining clear policies and reforms, promoting clear key performance indicators for regular monitoring by external or internal bodies, and building credibility through public engagement and performance improvements to pave the way for necessary tariff increases. Similarly, programming to sup- port regulators can emphasize regulatory frameworks defining clear roles and responsibilities within the sector, as well as the development of performance indicators specific to low-income areas. Finally, support for local governments can focus on developing or enforcing clear policy mandates for service provision in low-income areas, and on establishing strong financial man- agement systems to promote infrastructure investments. Overall, these strategies and lessons, derived from various settings and institutional landscapes, can offer insight for decision-mak- ers and underpin efforts in the numerous cities continuing to work toward universal access to safe water. Supporting information S1 Text. (DOCX) Acknowledgments The authors would like to thank Irene Atieno, Faith Chepngeno, and Edinah Samuel for their assistance in identifying relevant literature, developing narratives, and generating graphics for the case studies. We are also grateful to the experts and key informants we interviewed to shortlist cities and explore individual cases. Finally, we thank Ryan Mahoney, Jesse Shapiro, Daniel Smith, Rachel Peletz, and our consortium partners in the Urban Resilience by Building and Applying New Evidence in Water, Sanitation, and Hygiene (URBAN WASH) project for their help at various stages of study design and manuscript preparation. Author Contributions Conceptualization: Miriam Otoo, Caroline Delaire. PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 22 / 29 PLOS WATERThe enabling environment for citywide water service provision Formal analysis: John T. Trimmer, Haleemah Qureshi, Miriam Otoo. Investigation: John T. Trimmer, Haleemah Qureshi, Caroline Delaire. Methodology: John T. Trimmer, Haleemah Qureshi, Caroline Delaire. Project administration: Miriam Otoo, Caroline Delaire. Supervision: Miriam Otoo, Caroline Delaire. Visualization: John T. Trimmer, Haleemah Qureshi. Writing – original draft: John T. Trimmer. Writing – review & editing: John T. Trimmer, Haleemah Qureshi, Miriam Otoo, Caroline Delaire. References 1. Mitlin D, Beard VA, Satterthwaite D, Du J. Unaffordable and Undrinkable: Rethinking Urban Water Access in the Global South. Washington, D.C.: World Resources Institute; 2019 Aug. Available: https://www.wri.org/research/unaffordable-and-undrinkable-rethinking-urban-water-access-global- south 2. WHO/UNICEF. Household WASH Data. In: Join Monitoring Programme for Water Supply, Sanitation, and Hygiene [Internet]. 2022 [cited 4 May 2022]. Available: https://washdata.org/data/household#!/ 3. Sinharoy SS, Pittluck R, Clasen T. Review of drivers and barriers of water and sanitation policies for urban informal settlements in low-income and middle-income countries. Utilities Policy. 2019; 60: 100957. https://doi.org/10.1016/j.jup.2019.100957 PMID: 32214692 4. Dos Santos S, Adams EA, Neville G, Wada Y, de Sherbinin A, Mullin Bernhardt E, et al. Urban growth and water access in sub-Saharan Africa: Progress, challenges, and emerging research directions. Sci- ence of The Total Environment. 2017;607–608: 497–508. https://doi.org/10.1016/j.scitotenv.2017.06. 157 PMID: 28704674 5. Gambrill M, Gilsdorf RJ, Kotwal N. Citywide Inclusive Sanitation—Business as Unusual: Shifting the Paradigm by Shifting Minds. Frontiers in Environmental Science. 2020; 7. Available: https://www. frontiersin.org/article/10.3389/fenvs.2019.00201 6. Post A, Ray I. Hybrid Modes of Urban Water Delivery in Low- and Middle-Income Countries. Oxford Research Encyclopedia of Environmental Science. 2020; 1–29. https://doi.org/10.1093/acrefore/ 9780199389414.013.679 7. Satterthwaite D, Beard VA, Mitlin D, Du J. Untreated and Unsafe: Solving the Urban Sanitation Crisis in the Global South. Washington, D.C.: World Resources Institute; 2019 Dec. Available: https://www. wri.org/research/untreated-and-unsafe-solving-urban-sanitation-crisis-global-south 8. McIntosh A. Asian Water Supplies: Reaching the Urban Poor. Asian Development Bank; 2003. Avail- able: https://www.adb.org/publications/asian-water-supplies-reaching-urban-poor 9. Obrist B, Cisse G, Kone´ B, Dongo K, Granado S, Tanner M. Interconnected Slums: Water, Sanitation and Health in Abidjan, Coˆ te d’Ivoire. Eur J Dev Res. 2006; 18: 319–336. https://doi.org/10.1080/ 09578810600708387 10. Rusca M, Schwartz K. The Paradox of Cost Recovery in Heterogeneous Municipal Water Supply Sys- tems: Ensuring Inclusiveness or Exacerbating Inequalities? Habitat International. 2018; 73: 101–108. https://doi.org/10.1016/j.habitatint.2017.03.002 11. Trimmer JT, Miller DC, Byrne DM, Lohman HAC, Banadda N, Baylis K, et al. Re-Envisioning Sanita- tion As a Human-Derived Resource System. Environ Sci Technol. 2020; 54: 10446–10459. https://doi. org/10.1021/acs.est.0c03318 PMID: 32867485 12. Wu X, Malaluan NA. A Tale of Two Concessionaires: A Natural Experiment of Water Privatisation in Metro Manila. Urban Studies. 2008; 45: 207–229. https://doi.org/10.1177/0042098007085108 13. Northover H, Ryu SK, Brewer T. Achieving total sanitation and hygiene coverage within a generation —lessons from East Asia. WaterAid; 2016. Available: https://www.publicfinanceforwash.org/sites/ default/files/uploads/WaterAid-2016-Achieving_total_sanitation_and_hygiene_coverage_within_a_ generation__lessons_from_East_Asia.pdf PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 23 / 29 PLOS WATERThe enabling environment for citywide water service provision 14. WaterAid. A tale of clean cities: insights for planning urban sanitation from Ghana, India and the Philip- pines. London, UK: WaterAid; 2016. Available: https://washmatters.wateraid.org/publications/a-tale- of-clean-cities-insights-for-planning-urban-sanitation-from-ghana-india-and-the 15. McGranahan G, Mitlin D. Learning from Sustained Success: How Community-Driven Initiatives to Improve Urban Sanitation Can Meet the Challenges. World Development. 2016; 87: 307–317. https:// doi.org/10.1016/j.worlddev.2016.06.019 16. Mitlin D, Walnycki A. Informality as Experimentation: Water Utilities’ Strategies for Cost Recovery and their Consequences for Universal Access. The Journal of Development Studies. 2020; 56: 259–277. https://doi.org/10.1080/00220388.2019.1577383 17. McIntosh AC, editor. Pearls of Wisdom in Asian Water Supplies: Lessons Learned from 1000 Years of Collective Experience. Manila, Philippines: Books on Demand Philippines, Inc.; 2013. Available: http://booksondemand.com.ph/wp-content/uploads/2013/11/Pearls-of-Wisdom-in-Asian-Water- Supplies.pdf 18. Kumpel E, Peletz R, Bonham M, Khush R. Assessing Drinking Water Quality and Water Safety Man- agement in Sub-Saharan Africa Using Regulated Monitoring Data. Environ Sci Technol. 2016; 50: 10869–10876. https://doi.org/10.1021/acs.est.6b02707 PMID: 27559754 19. Trimmer JT, Lohman HAC, Byrne DM, Houser SA, Jjuuko F, Katende D, et al. Navigating Multidimen- sional Social–Ecological System Trade-Offs across Sanitation Alternatives in an Urban Informal Set- tlement. Environ Sci Technol. 2020; 54: 12641–12653. https://doi.org/10.1021/acs.est.0c03296 PMID: 32822180 20. Ostrom E. A General Framework for Analyzing Sustainability of Social-Ecological Systems. Science. 2009; 324: 419–422. https://doi.org/10.1126/science.1172133 PMID: 19628857 21. McGinnis MD, Ostrom E. Social-ecological system framework: initial changes and continuing chal- lenges. Ecology and Society. 2014;19. Available: https://www.jstor.org/stable/26269580 22. Ostrom E. A diagnostic approach for going beyond panaceas. Proc Natl Acad Sci USA. 2007; 104: 15181–15187. https://doi.org/10.1073/pnas.0702288104 PMID: 17881578 23. Hinkel J, Cox M, Schlu¨ter M, Binder C, Falk T. A diagnostic procedure for applying the social-ecologi- cal systems framework in diverse cases. Ecology and Society. 2015;20. https://doi.org/10.5751/ES- 07023-200132 24. Stokstad E. Rebuilt From Ruins, a Water Utility Turns Clean and Pure. Science. 2008; 319: 746–747. https://doi.org/10.1126/science.319.5864.746 PMID: 18258893 25. Rahman Z, Crocker J, Chang K, Khush R, Bartram J. A Comparative Assessment of Institutional Frameworks for Managing Drinking Water Quality. Journal of Water, Sanitation and Hygiene for Devel- opment. 2011; 1: 242–258. https://doi.org/10.2166/washdev.2011.002 26. International Finance Corporation. The Market for Small-Scale Piped Water Systems in Kenya. World Bank Group; 2011. Report No.: Issue:1. Available: https://ppp.worldbank.org/public-private- partnership/sites/ppp.worldbank.org/files/2022-06/115055-BRI-Kenya-SmallPipedWaterMarket-IFC- July-2011.pdf 27. Castro V. Improving Water Utility Services through Delegated Management: Lessons from the utility and small-scale providers in Kisumu, Kenya. Nairobi, Kenya: World Bank Water and Sanitation Pro- gram; 2009. Available: https://www.wsp.org/sites/wsp/files/publications/Af-imp_through_delegated_ mgmt.pdf 28. Peletz R, Kumpel E, Bonham M, Rahman Z, Khush R. To What Extent is Drinking Water Tested in Sub-Saharan Africa? A Comparative Analysis of Regulated Water Quality Monitoring. Int J Environ Res Public Health. 2016; 13: 275. https://doi.org/10.3390/ijerph13030275 PMID: 26950135 29. Barraque´ B. The Three Ages of Engineering for the Water Industry. Anuari de la Societat Catalana d’Economia. 2004; 18: 135–152. 30. Orgill-Meyer J, Jeuland M, Albert J, Cutler N. Comparing Contingent Valuation and Averting Expendi- ture Estimates of the Costs of Irregular Water Supply. Ecological Economics. 2018; 146: 250–264. https://doi.org/10.1016/J.ECOLECON.2017.10.016 31. Cook J, Fuente D, Matichich M, Whittington D. A Global Assessment of Nontariff Customer Assistance Programs in Water Supply and Sanitation. In: Chen Z, Bowen WM, Whittington D, editors. Develop- ment Studies in Regional Science: Essays in Honor of Kingsley E Haynes. Singapore: Springer; 2020. pp. 315–371. https://doi.org/10.1007/978-981-15-1435-7_17 32. WSUP. Institutionalising Pro-Poor Services: A New Key Performance Indicator for Kenyan Utilities. Water & Sanitation for the Urban Poor; 2018 Jul p. 2. Available: https://www.wsup.com/content/ uploads/2018/07/PN036-ENGLISH-Institutionalising-pro-poor-service-delivery-KPI-10-Kenya.pdf PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 24 / 29 PLOS WATERThe enabling environment for citywide water service provision 33. Komives K, Foster V, Halpern J, Wodon Q. Water, Electricity, and the Poor: Who Benefits from Utility Subsidies? Washington, DC: World Bank; 2005 Oct. Available: https://openknowledge.worldbank. org/handle/10986/6361 34. Wutich A, Beresford M, Carvajal C. Can Informal Water Vendors Deliver on the Promise of A Human Right to Water? Results From Cochabamba, Bolivia. World Development. 2016; 79: 14–24. https:// doi.org/10.1016/j.worlddev.2015.10.043 35. Wambu CK, Openda NO, Kinuthia LN. Influence of Land Tenure on Gender Participation in Manage- ment of Water Projects in Njoro Sub-County, Nakuru. 2016 [cited 29 Aug 2022]. Available: http://ir. mksu.ac.ke/handle/123456780/4675 36. Ranganathan M. “Mafias” in the waterscape: urban informality and everyday public authority in Banga- lore. Water Alternatives. 2014; 7: 89–105. https://www.cabdirect.org/cabdirect/abstract/20143336591 37. Obani P. The human rights to water and sanitation in courts worldwide: a selection of national, regional, and international case law, WaterLex, 2014, ISBN: 978-2-940526-00-0. Int Environ Agree- ments. 2015; 15: 237–239. https://doi.org/10.1007/s10784-015-9284-7 38. Lawson J, Chappell H. Latin American Politics and Society: Water and Politics: Clientelism and Reform in Urban Mexico. Latin American Politics and Society. 2018; 60: 139–141. https://doi.org/10. 1017/lap.2018.18 39. Hines S. The Power and Ethics of Vernacular Modernism: The Misicuni Dam Project in Cochabamba, Bolivia, 1944–2017. Hispanic American Historical Review. 2018; 98: 223–256. https://doi.org/10.1215/ 00182168-4376680 40. ESAWAS. The Water Supply and Sanitation Regulatory Landscape Across Africa; Content Wide Syn- thesis Report. Eastern and Southern Africa Water and Sanitation Regulators Association; 2022. Avail- able: https://esawas.org/repository/Esawas_Report_2022.pdf 41. Berg SV, Mugisha S. Pro-poor water service strategies in developing countries: promoting justice in Uganda’s urban project. Water Policy. 2010; 12: 589–601. https://doi.org/10.2166/wp.2010.120 42. Banks N, Lombard M, Mitlin D. Urban Informality as a Site of Critical Analysis. The Journal of Develop- ment Studies. 2020; 56: 223–238. https://doi.org/10.1080/00220388.2019.1577384 43. Allen A, Davila JD, Hoffman P. Governance of Water and Sanitation Services for the Peri-Urban Poor: A Framework for Understanding and Action in Metropolitan Regions. DPU University College London; 2011 Nov p. 7. Available: https://assets.publishing.service.gov.uk/media/ 57a08c8b40f0b649740012a6/R8137-Leaflet2.pdf 44. World Bank. World Bank Country and Lending Groups. In: World Bank Data Help Desk [Internet]. [cited 3 Feb 2022]. Available: https://datahelpdesk.worldbank.org/knowledgebase/articles/906519- world-bank-country-and-lending-groups 45. World Bank. World Development Indicators: GNI per capita (constant 2015 US$). 2021. Available: https://databank.worldbank.org/reports.aspx?source=2&series=NY.ADJ.NNTY.PC.CD&country= BRA 46. United Nations. World Urbanization Prospects: The 2018 Revision, Online Edition. 2018 [cited 3 Feb 2022]. Available: https://population.un.org/wup/Download/ 47. Habitat UN. Housing, slums and informal settlements. In: Urban Indicators Database [Internet]. 2021 [cited 3 Jun 2022]. Available: https://data.unhabitat.org/ 48. The Economist Intelligence Unit. Democracy Index 2020: In Sickness and in Health? The Economist Intelligence Unit Limited; 2021. Available: https://www.eiu.com/n/campaigns/democracy-index-2020/ 49. UN FAO. AQUASTAT—FAO’s Global Information System on Water and Agriculture. In: Food and Agricultural Organization of the United Nations [Internet]. 2021 [cited 3 Feb 2022]. Available: https:// www.fao.org/aquastat/en/ 50. van den Berg C, Danilenko A. Performance of Water Utilities in Africa. Washington, DC: World Bank; 2017. https://doi.org/10.1596/26186 51. Cox M. A basic guide for empirical environmental social science. Ecology and Society. 2015;20. https://doi.org/10.5751/ES-07400-200163 52. Ostrom E. Governing the Commons: The Evolution of Institutions for Collective Action. Cambridge University Press; 1990. 53. Cross P, Morel A. Pro-poor strategies for urban water supply and sanitation services delivery in Africa. Water Science & Technology. 2005; 51: 51–57. https://doi.org/10.2166/wst.2005.0223 PMID: 16007927 54. Marin P, Ouayoro E, Fall M, Verspyck R. Partnering for water in Coˆte d’Ivoire: Lessons from 50 years of successful private operation. Washington, D.C.: The World Bank; 2009 Aug. Report No.: 50. Avail- able: https://openknowledge.worldbank.org/entities/publication/88c279a7-fff3-55ed-9e89- 23c23d5b5276 PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 25 / 29 PLOS WATERThe enabling environment for citywide water service provision 55. CEPT University. Ahmedabad Slum Free City Action Plan. CEPT University; 2014 Sep p. 222. Avail- able: https://pas.org.in/Portal/document/PIP%20Application/Ahmedabad%20Slum%20Free%20City %20Action%20Plan%20RAY.pdf 56. Bhatkal T, Avis W, Nicolai S. Towards a better life? A cautionary tale of progress in Ahmedabad. ODI; 2015 Jun p. 48. 57. McIntosh AC, Yñiguez CE, editors. Second Water Utilities Data Book: Asian and Pacific Region. Manila, Philippines: Asian Development Bank; 1997. Available: https://fr.ircwash.org/sites/default/ files/Mclntosh-1997-Second.pdf 58. Lee Y-SF. Intermediary institutions, community organizations, and urban environmental management: The case of three Bangkok slums. World Development. 1998; 26: 993–1011. https://doi.org/10.1016/ S0305-750X(98)00034-5 59. McIntosh AC. Urban Water Supply and Sanitation in Southeast Asia: A Guide to Good Practice. Man- daluyong City, Metro Manila, Philippines: Asian Development Bank; 2014. Available: https://www.adb. org/publications/urban-water-supply-and-sanitation-southeast-asia-guide-good-practice 60. Daef MA. The Impact of Open Door Policy on Public Service Provisions in Urban Areas of Egypt: A Major Factor Influencing Income Distribution since 1974. 1986 Sep. Report No.: 29. Available: https:// www.ucl.ac.uk/bartlett/development/sites/bartlett/files/migrated-files/WP29_0.pdf 61. Hoehn JP, Krieger DJ. An Economic Analysis of Water and Wastewater Investments in Cairo, Egypt. Eval Rev. 2000; 24: 579–608. https://doi.org/10.1177/0193841X0002400602 62. Attia FAR. Water and Development in Greater Cairo (Egypt). Revista CIDOB d’Afers Internacionals. 1999; 81–102. https://www.jstor.org/stable/40586151 63. EWRA. Egyptian Water and Wastewater Regulatory Agency (EWRA): Third Annual Report 2009– 2010. Egyptian Water and Wastewater Regulatory Agency; 2010 p. 57. 64. World Bank. Cambodia: Phnom Penh Water Supply Authority: An Exemplary Water Utility in Asia. World Bank PPPLRC; 2009 Jul p. 2. Available: https://ppp.worldbank.org/public-private-partnership/ sites/ppp.worldbank.org/files/documents/PhnomPenh_ExemplaryWaterUtilityinAsia_EN.pdf 65. Maltz H. Porto Alegre’s Water: Public and for All. Reclaiming Public Water. 2005. pp. 29–36. 66. Viero OM. Water Supply and Sanitation in Porto Alegre Brazil. 20 Feb 2003 [cited 14 Apr 2022]. Avail- able: https://sswm.info/sites/default/files/reference_attachments/VIERO%202003%20Water% 20Supply%20and%20Sanitation%20in%20Porto%20Alegre%20Brazil.pdf 67. Abers R, Brandão I, King R, Votto D. Porto Alegre: Participatory Budgeting and the Challenge of Sus- taining Transformative Change. Washington, D.C.: World Resources Institute; 2018. 68. Angoua ELE, Dongo K, Templeton MR, Zinsstag J, Bonfoh B. Barriers to access improved water and sanitation in poor peri-urban settlements of Abidjan, Coˆ te d’Ivoire. PLOS ONE. 2018; 13: e0202928. https://doi.org/10.1371/journal.pone.0202928 PMID: 30153297 69. Menard C, Clarke G. Reforming the Water Supply in Abidjan, Coˆ te d’lvoire: Mild Reform in a Turbulent Environment. The World Bank Development Research Group; 2000 Jun p. 52. Report No.: 2377. Available: https://elibrary.worldbank.org/doi/abs/ https://doi.org/10.1596/1813-9450-2377 70. Habitat UN. Access to Basic Services in Cities and Urban Areas. In: Urban Indicators Database [Inter- net]. 2019 [cited 5 Jul 2022]. Available: https://data.unhabitat.org/pages/access-to-basic-services-in- cities-and-urban-areas 71. Bhatt MR. The Case of Ahmedabad, India. London: UN-Habitat; 2003 p. 23. Available: https://www. ucl.ac.uk/dpu-projects/Global_Report/cities/ahmedabad.htm 72. Bhuvan K, Ashram NH. Parivartan and Its Impact: A Partnership Programme of Infrastructure Devel- opment in Slums of Ahmedabad City. Sewa Academy; 2002 Jan p. 58. 73. CEPT University Reaching the Poor, Slum Networking Project Ahmedabad. World Water Week; 2011; Stockholm. Available: https://pas.org.in/Portal/document/ResourcesFiles/Reaching%20the% 20Poor%20Slum%20Networking%20Project-Ahmedabad.pdf 74. Gautam IP. Insight on City Management: Perspective from Ahmedabad, Gujarat, India. Ahmedabad, Gujarat: Ahmedabad Municipal Corporation; 2010 p. 14. Available: https://web.worldbank.org/ archive/website01409/WEB/IMAGES/2010_1_1.PDF 75. UNDP, WSP. Ahmedabad Parivartan. UNDP-World Bank WSP; 1999 p. 22. Available: https:// documents.worldbank.org/en/publication/documents-reports/documentdetail/901321468252673972/ ahmedabad-parivartan 76. Urban Management Centre. Urban Water and Sanitation in Gujarat. Urban Management Centre; 2011 Jun p. 199. Available: https://www.pas.org.in/Portal/document/ResourcesFiles/pdfs/Urban%20Water %20&%20Sanitation%20Gujarat%20UMC.pdf PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 26 / 29 PLOS WATERThe enabling environment for citywide water service provision 77. Daniere AG, Takahashi LM. Poverty and Access: Differences and Commonalties across Slum Com- munities in Bangkok. Habitat International. 1999; 23: 271–288. https://doi.org/10.1016/S0197-3975 (98)00052-6 78. MWA. Annual Report 2019. Metropolitan Waterworks Authority; 2019 p. 156. Available: https://web. mwa.co.th/mobile/download/Annual_Report_2562_eng.pdf 79. EWRA. The 10th Annual Report of the Egyptian Water and Wastewater and Consumer Protection Regulatory Agency 2017–2018. Egyptian Water and Wastewater Regulatory Agency; 2018 p. 65. Available: http://www.ewra.gov.eg/DocumentsLibrary/Library/1/English%20Anual%20Repot% 202017-2018.pdf 80. ADB. Sector Assessment (summary): Water and Other Urban Infrastructure and Services. Asian Development Bank; 2017 Nov. Available: https://www.adb.org/sites/default/files/linked-documents/ 41403-013-ssa.pdf 81. GDF. Development Management and Aid: The Case of Water Supply Sector Reform in Cambodia. Grips Development Forum; 2019. Available: https://www.grips.ac.jp/teacher/oono/hp/course/lec19_ cambodia/cambodia_water1.pdf 82. da Costa SS, Heller L, Moraes LRS, Borja PC, de Melo CH, Sacco D. Successful Experiences in Municipal Public Water and Sanitation Services from Brazil. ASSEMAE; 2006. Available: https://www. tni.org/en/publication/successful-experiences-in-municipal-public-water-and-sanitation-services- from-brazil 83. Hall D, Lobina E, Viero OM, Maltz H. Water in Porto Alegre, Brazil; accountable, effective, sustainable, and democratic. 2002. Available: http://www.psiru.org/sites/default/files/2002-08-W-dmae.pdf 84. 85. Liemberger R, Wyatt A. Quantifying the global non-revenue water problem. Water Supply. 2018; 19: 831–837. https://doi.org/10.2166/ws.2018.129 Jaigwang P, Fuangsawat W, Thupmongkhon D, Bamrungphon W, Saengtaiyarak N, Wongpat N. Geographic Information System for water quality work: “Water quality on your click” by MWA, Thai- land. Seville, Spain; 2017. pp. 155–160. https://doi.org/10.2495/WS170141 86. Kordach A, Chardwattananon C, Wongin K, Chayaput B, Wongpat N. Evaluation on the Quality of Bangkok Tap Water with Other Drinking Purpose Water. E3S Web Conf. 2018; 30: 01011. https://doi. org/10.1051/e3sconf/20183001011 87. MWA. MWA Consumer Confidence Report 2021. 2021. Available: https://web.mwa.co.th/download/ prd01/CCR/2021%20Annual%20Water%20Quality%20Report%20-%20%E0%B8%AD%E0%B8% B1%E0%B8%87%E0%B8%81%E0%B8%A4%E0%B8%A9.pdf 88. ADB. Water in Asian Cities: Utilities’ Performance and Civil Society Views. Asian Development Bank; 2004. 89. Dhamasiri C. Water Supply in Bangkok Metropolis. Water Supply in Fast Growing Cities. Amsterdam; 1988. Available: https://www.ircwash.org/sites/default/files/205.40-88WA-5751.pdf 90. Government of Thailand. Information about the Cabinet–Secretariat of the Cabinet. 2021 [cited 28 Jul 2022]. Available: https://www.soc.go.th/?page_id=182 91. Biswas AK, Tortajada C. Water Supply of Phnom Penh: A Most Remarkable Transformation. Singa- pore: Institute of Water Policy Lee Kuan Yew School of Public Policy; 2009 Sep p. 65. Available: https://thirdworldcentre.org/wp-content/uploads/2020/08/Water-Supply-of-Phnom-Penh-A-Most- Remarkable-Transformation.pdf 92. Phnom Penh Water Supply Authority. Board of Directors. [cited 25 Jul 2022]. Available: http://www. ppwsa.com.kh/en/index.php?page=BOD 93. 94. ILO. Thailand—Standard Qualifications of Directors and State Enterprise Employees Act, 1975 [B.E. 2518]. In: International Labor Organization [Internet]. 2014 [cited 28 Jul 2022]. Available: http://www. ilo.org/dyn/natlex/natlex4.detail?p_lang=en&p_isn=100481&p_classification=09 JICA. Country Study for Japan’s Official Development Assistance to the Kingdom of Cambodia. Japan International Cooperation Agency; 2002 Mar p. 366. Available: https://www.jica.go.jp/jica-ri/IFIC_and_ JBICI-Studies/english/publications/reports/study/country/pdf/cambodia.pdf 95. Spencer JH, Meng B. Resilient urbanization and infrastructure governance: the case of the Phnom Penh Water Supply Authority, 1993–2007. Water Policy. 2019; 21: 848–864. https://doi.org/10.2166/ wp.2019.211 96. Calisto Friant M. Deliberating for sustainability: lessons from the Porto Alegre experiment with partici- patory budgeting. International Journal of Urban Sustainable Development. 2019; 11: 81–99. https:// doi.org/10.1080/19463138.2019.1570219 97. Berta I. Key Informant Interview—ONEP. 2022. 98. World Bank. Additional Financing for the Urban Water Supply and Sanitation Project (formerly Urban Water Supply Project). 2019 p. 87. Report No.: PAD3325. PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 27 / 29 PLOS WATERThe enabling environment for citywide water service provision 99. SODECI. SODECI CADRE INSTITUTIONNEL. In: SODECI [Internet]. 2022 [cited 29 Jul 2022]. Avail- able: https://www.sodeci.ci/notre-contrat-eau-potable/cadre-institutionnel 100. World Bank. Water in the Arab World: Management Perspectives and Innovations. Washington DC: Word Bank; 2009 p. 554. Report No.: 49593. Available: https://web.worldbank.org/archive/ website01418/WEB/IMAGES/WATER_AR.PDF 101. Amer K. Water Supply and Sanitation in Egypt. [cited 1 Jul 2022]. Available: https://www.academia. edu/33307626/Water_Supply_and_Sanitation_in_Egypt 102. Wahby NM. Urban Informality and the State: Repairing Cairo’s Waters Through Gehood Zateya. Envi- ronment and Planning E: Nature and Space. 2021; 4: 696–717. https://doi.org/10.1177/ 25148486211025262 103. Khalil D. The Flexible Governance of Water in Cairo’s Informal Areas. Water. 2019; 11: 1644. https:// doi.org/10.3390/w11081644 104. Allen A, Da´ vila JD, Hofmann P. Governance of Water and Sanitation Services for the Peri-urban Poor. A Framework for Understanding and Action in Metropolitan Regions. London: The Development Planning Unit, University College; 2006. Available: https://discovery.ucl.ac.uk/id/eprint/52345/ 105. Urban Management Centre. City Sanitation Plan, Ahmedabad. Ahmedabad, India: Urban Manage- ment Centre; 2012 p. 185. Available: https://www.susana.org/en/knowledge-hub/resources-and- publications/library/details/3236 106. AMC. About the Corporation: Ahmedabad Municipal Corporation. In: Amdavad Municipal Corporation [Internet]. [cited 9 Jun 2022]. Available: https://ahmedabadcity.gov.in/portal/jsp/Static_pages/about_ amc.jsp 107. Davis J. Corruption in Public Service Delivery: Experience from South Asia’s Water and Sanitation Sector. World Development. 2004; 32: 53–71. https://doi.org/10.1016/j.worlddev.2003.07.003 108. Collignon B, Taisne R, Kouadio J. Water and sanitation for the urban Poor in Coˆte d’Ivoire. WSP and Hydroconseil; 2000 May p. 33. https://doi.org/10.13140/RG.2.2.25718.86081 109. WSP. Water Utilities in Africa: Case Studies of Transformation and Market Access. The World Bank; 2007 Jul p. 100. Available: https://www.wsp.org/sites/wsp/files/publications/Water_Utilities_Africa.pdf 110. 111. Jones O, Kimani A. Citywide Inclusive Sanitation: How can resourcing be managed effectively? ESA- WAS Regulators Association; 2021. Available: https://www.wsup.com/insights/citywide-inclusive- sanitation-how-can-resourcing-be-managed-effectively/ Lauria DT, Hopkins OS, Debomy S. Pro-Poor Subsidies for Water Connections in West Africa. The World Bank; 2005 p. 45. Report No.: Note No. 3. Available: https://ppp.worldbank.org/public-private- partnership/library/pro-poor-subsidies-water-connections-west-africa-preliminary-study 112. Mok S. Key Informant Interview—WaterAid. 2022. 113. Diacon D. Slum Networking: An Innovative Approach to Urban Development. Leicestershire, U.K.: Building and Social Housing Foundation; 1997. Available: https://world-habitat.org/publications/slum- networking-an-innovative-approach-to-urban-development-in-india/ 114. AUDA. Urban Planning and Development Practices of Ahmedabad Urban Development Authority. Ahmedabad, Gujarat: Ahmedabad Urban Development Authority; 2021 Apr p. 86. 115. Mallory A, Mdee A, Agol D, Hyde-Smith L, Kiogora D, Riungu J, et al. The Potential for Scaling up Con- tainer-Based Sanitation in Informal Settlements in Kenya. J Int Dev. 2022;n/a: 1–15. https://doi.org/10. 1002/jid.3639 116. Drabble S, Nzitonda J, Abdumaliki M, Eales K. Citywide Inclusive Sanitation: Who is responsible? ESAWAS Regulators Association; 2021. Available: https://www.wsup.com/insights/citywide-inclusive- sanitation-who-is-responsible/ 117. Narzetti DA, Marques RC. Access to Water and Sanitation Services in Brazilian Vulnerable Areas: The Role of Regulation and Recent Institutional Reform. Water. 2021; 13: 787. https://doi.org/10. 3390/w13060787 118. Norman G, Cheruiyot R. Citywide Inclusive Sanitation: How can accountability be strengthened? ESA- WAS Regulators Association; 2021. Available: https://www.wsup.com/insights/citywide-inclusive- sanitation-how-can-accountability-be-strengthened/ 119. Mehta M, Mehta D, Vavaliya J. Urban Drinking Water Security in Gujarat. Journal of Social and Eco- nomic Development. 2021; 23: S166–S180. https://doi.org/10.1007/s40847-020-00122-0 PMID: 34790002 120. Bhatkal T, Avis W, Nicolai S. Urbanizing Diseases: Contested Institutional Terrain of Water- and Vec- tor-Borne Diseases in Ahmedabad, India. Water International. 2013; 38: 875–887. https://doi.org/10. 1080/02508060.2013.851363 PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 28 / 29 PLOS WATERThe enabling environment for citywide water service provision 121. Biswas A k., Tortajada C. Water Supply of Phnom Penh: An Example of Good Governance. Interna- tional Journal of Water Resources Development. 2010; 26: 157–172. https://doi.org/10.1080/ 07900621003768859 122. Brahmbhatt B. Key Informant Interview—Mahila Housing Trust. 2022. 123. Kipper R, Fischer M, editors. Cairo’s Informal Areas; Between Urban Challenges and Hidden Poten- tials. GTZ Egypt; 2009. Available: https://www.citiesalliance.org/sites/default/files/2019-07/ CairosInformalAreas_fulltext.pdf 124. 125. Zayed MA. The challenge of Greater Cairo governance. IJGET. 2012; 1: 19–34. Fallavier P. Urban Slums report: The case of Phnom Penh, Cambodia. UN-Habitat; 2003. Available: https://urbandatabase.khmerstudies.org/storage/media/214/259d50b1fba7f45dcd245d43f26d8ee7. pdf 126. Babel MS, Gupta AD, Domingo NDSto. Land Subsidence: A Consequence of Groundwater Over- Exploitation in Bangkok, Thailand. IGES. 2006; 6: 307–328. 127. Edo E, Sadayeva G, Hassan N. MENA Megacities Approaching Day Zero: A Comparative Study Between Cairo and Istanbul. 2020; 33. 128. McConville JR, Kvarnstro¨ m E, Ahlstro¨m M, Niwagaba CB. Possibilities for changing to resource recov- ery in Kampala’s on-site sanitation regime. Resources, Conservation and Recycling. 2022; 181: 106275. https://doi.org/10.1016/j.resconrec.2022.106275 129. Blackett I, Hawkins P. FSM innovation case studies—case studies on the business, policy and tech- nology of faecal sludge management (second edition). Seattle, USA: Bill & Melinda Gates Founda- tion; 2017. Available: https://www.susana.org/en/knowledge-hub/resources-and-publications/library/ details/2760 130. Peletz R, Kisiangani J, Bonham M, Ronoh P, Delaire C, Kumpel E, et al. Why Do Water Quality Moni- toring Programs Succeed or Fail? A Qualitative Comparative Analysis of Regulated Testing Systems in Sub-Saharan Africa. International Journal of Hygiene and Environmental Health. 2018; 221: 907– 920. https://doi.org/10.1016/j.ijheh.2018.05.010 PMID: 29861398 131. Tribbe J, Zuin V, Delaire C, Khush R, Peletz R. How Do Rural Communities Sustain Sanitation Gains? Qualitative Comparative Analyses of Community-Led Approaches in Cambodia and Ghana. Sustain- ability. 2021; 13: 5440. https://doi.org/10.3390/su13105440 132. Queiro´ s A, Faria D, Almeida F. Strengths and limitations of qualitative and quantitative research meth- ods. European Journal of Education Studies. 2017 [cited 30 Sep 2022]. https://doi.org/10.46827/ejes. v0i0.1017 133. Water Utility Partnership. Better Water and Sanitation for the Urban Poor: Good Practice from Sub- Saharan Africa. Washington, DC: World Bank; 2003 Jul p. 102. Report No.: 43101. Available: https:// openknowledge.worldbank.org/handle/10986/14746 134. Gorelick J. Supporting the future of municipal bonds in sub-Saharan Africa: the centrality of enabling environments and regulatory frameworks. Environment and Urbanization. 2018; 30: 103–122. https:// doi.org/10.1177/0956247817741853 PLOS Water \| https://doi.org/10.1371/journal.pwat.0000071 June 26, 2023 29 / 29 PLOS WATER
10.1371_journal.ppat.1011456	RESEARCH ARTICLE α-Synuclein seeding activity in duodenum biopsies from Parkinson’s disease patients 1☯, Christina D. Orru` 2☯, Bradley R. Groveman2, Sabiha Parveen2, Sarah VascellariID Giuseppe Fenu3, Giada Pisano3, Giuseppe Piga3, Giulia Serra3, Valentina Oppo3, Daniela Murgia3, Andrea Perra1, Fabrizio Angius1, Andrew G. Hughson2, Cathryn L. Haigh2, Aldo Manzin1, Giovanni Cossu3‡, Byron Caughey2‡ 1 Department of Biomedical Sciences, University of Cagliari, Cagliari, Italy, 2 Laboratory of Neurological Infections and Immunity (LNII), Rocky Mountain Laboratories, National Institute of Allergy and Infectious Diseases (NIAID), National Institute of Health (NIH), Hamilton, Montana, United States, 3 S. C. Neurology and Stroke Unit, AOBrotzu, Cagliari, Italy ☯ These authors contributed equally to this work. ‡ These authors are joint senior authors on this work. svascellari@unica.it Abstract Abnormal deposition of α-synuclein is a key feature and biomarker of Parkinson’s disease. α-Synuclein aggregates can propagate themselves by a prion-like seeding-based mecha- nism within and between tissues and are hypothesized to move between the intestine and brain. α-Synuclein RT-QuIC seed amplification assays have detected Parkinson’s-associ- ated α-synuclein in multiple biospecimens including post-mortem colon samples. Here we show intra vitam detection of seeds in duodenum biopsies from 22/23 Parkinson’s patients, but not in 6 healthy controls by RT-QuICR. In contrast, no tau seeding activity was detected in any of the biopsies. Our seed amplifications provide evidence that the upper intestine con- tains a form(s) of α-synuclein with self-propagating activity. The diagnostic sensitivity and specificity for PD in this biopsy panel were 95.7% and 100% respectively. End-point dilution analysis indicated up to 106 SD50 seeding units per mg of tissue with positivity in two con- temporaneous biopsies from individual patients suggesting widespread distribution within the superior and descending parts of duodenum. Our detection of α-synuclein seeding activ- ity in duodenum biopsies of Parkinson’s disease patients suggests not only that such analy- ses may be useful in ante-mortem diagnosis, but also that the duodenum may be a source or a destination for pathological, self-propagating α-synuclein assemblies. Author summary Misfolded α-Synuclein deposition is a hallmark of Parkinson’s disease. The gastrointesti- nal tract may be an initial site of α-Synuclein aggregation, and its detection might be use- ful in the early diagnosis of Parkinson’s disease. Here, we have used a rapid, ultrasensitive seed amplification assay (RT-QuICR) to show that pathologic α-Syn aggregates with prion-like self-propagating activity are in the upper intestine (duodenum) of Parkinson’s disease patients. Our intra vitam detection of α-synuclein seeding activity in duodenum a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Vascellari S, Orru` CD, Groveman BR, Parveen S, Fenu G, Pisano G, et al. (2023) α- Synuclein seeding activity in duodenum biopsies from Parkinson’s disease patients. PLoS Pathog 19(6): e1011456. https://doi.org/10.1371/journal. ppat.1011456 Editor: Amanda L. Woerman, University of Massachusetts Amherst, UNITED STATES Received: April 20, 2023 Accepted: June 2, 2023 Published: June 30, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.ppat.1011456 Copyright: This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 1 / 16 PLOS PATHOGENSFunding: This work was partially supported by Intramural Research Program of the National Institute for Allergy and Infectious Diseases, National Institutes of Health (BC). The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. CDO, BRG, SP, AGH, CLH and BC received a salary from Intramural Research Program of the National Institute for Allergy and Infectious Diseases, National Institutes of Health. Competing interests: BC, CDO and AH are inventors on patent applications pertaining to aSyn RT-QuIC technology. The other authors have declared that no competing interests exist. α-Synuclein seeding activity in Parkinson’s disease intestinal biopsies biopsies gave high diagnostic accuracy. Quantitation revealed high levels of seeds in duo- denal tissue. Thus, our findings suggest that abnormal α-Synuclein seeds in the upper intestine might be both an early accurate biomarker for Parkinson’s disease and cause of gut dysfunction. Introduction Definitive diagnosis of Parkinson’s disease (PD) relies on postmortem detection of disease- associated α-Synuclein aggregates (α-SynD) in the brain [1]. According to the “dual hit hypothesis”, α-SynD may start accumulating at the site of enteric nerves in the gut or the olfac- tory bulb, and progressively spread in stages along neural tracts to the brain [2]. Others have suggested a reverse path of α-SynD spreading, which could originate from the brain and then reach the gastrointestinal (GI) tract [3]. Clinical and neuropathological evidence indicates that motor symptoms in PD are often preceded by GI dysfunctions [4, 5] leading some researchers to consider PD as a gut-brain disorder. One recent study showed α-SynD in the colon of pre- symptomatic mice expressing human A53T α-Syn before detecting it in the brain [6]. It is also postulated that both the gut and brain first hypotheses are possible, which would lead to a new classification of PD, depending on the different initial site of α-SynD accumulation. These include a “brain-first” subtype, in which pathology originates primarily in the brain, and a “body-first” subtype, characterized by premotor rapid eye movement behavior disorder (RBD) and GI impairment, in which pathological changes begin in the enteric or peripheral auto- nomic nervous systems, and follow a route that involves the brainstem [7–10]. These considerations make α-SynD detection in intestinal tissues an attractive strategy for the early diagnosis of PD. However, conventional immunological-based techniques for detec- tion of α-SynD have produced conflicting results, likely due to the limited sensitivities of these assays [11]. One recent study suggested that immunohistochemistry (IHC) based methods are not suitable for the diagnosis of PD [12]. In contrast, an initial report by Emmi et al [13] showed IHC detection of α-SynD in 100% of duodenum samples tested from a small number of PD patients. Other studies also propose that IHC analysis of layers deeper than mucosa and submucosa, such as muscularis propria or myenteric plexuses, for enteric phosphorylated α- Syn detection could be used to predict the onset of motor symptoms in PD [14]. Recently, ultrasensitive seed amplification assays (SAAs) have been applied to help address these contro- versial questions. Based in principle on the RT-QuIC assay platform developed for PrP prions [15, 16], the RT-QuIC-like SAAs for synucleinopathies are techniques that exploit the ability of α-SynD to self-replicate similarly to prions, and to amplify trace amounts of pathological α- Syn in biological specimens (e.g., brain tissues, cerebrospinal fluid [CSF], submandibular glands, saliva, olfactory mucosa and skin) [17–25]. SAAs, such as α-Syn RT-QuIC [18, 19] and α-Syn PMCA [26], have been shown to detect α-SynD in intestinal tissues. However, these studies detected α-SynD seeds in only 55% of cases [27], or used postmortem samples [28]. In this study, we analyzed duodenum biopsies from PD and control patient cohorts with a rela- tively rapid α-Syn RT-QuIC assay (RT-QuICR), which uses a mutated version of the human α-synuclein protein and achieves faster and more specific detection of α-SynD [29] than prior SAAs [18, 26]. Given reports of tau and α-Synuclein copathologies [30–34], we also checked a subset of the biopsies for tau seeding activity using a tau RT-QuIC assay [35]. We found α- SynD, but not tau, seeding activity in the intestinal mucosa (IM) tissue in 95.7% of PD cases, suggesting both diagnostic and potential pathological implications. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 2 / 16 PLOS PATHOGENSα-Synuclein seeding activity in Parkinson’s disease intestinal biopsies Results Patient demographics and clinical information Demographic and clinical information about the participants recruited for this study is given in Table 1. The mean age at the time of biopsy collection was 67 ± 8 years for PD cases and 58 ± 10 years old for healthy controls (HCs). The number of male and female patients within the PD and HC cohorts was similar. For PD patients, the mean disease duration at the time of intestinal biopsy collection was 14 ± 5 years. Thirteen of the 23 cases showed premotor RBD from 7 years before the onset of motor symptoms. At the time of biopsy collections all PD Table 1. Demographic characteristics of Parkinson’s and healthy control patients. ID Diagnosis Age1 Sex (M/F) Disease duration1 Constipation score premotor RBD2 Parkinson’s disease 67±8 11/12 14±5 11±6 13 (57) UPDRS III score1 25±7 Duodenal biopsies tested by RT-QuICR n = 30 6 25 11 15 10 14 11 15 17 12 15 12 19 14 20 12 8 6 19 13 20 8 13 17 19 3 14 16 19 10 14 18 4 6 14 7 5 6 21 18 15 2 7 12 4 7 yes yes no yes yes yes no no no yes yes yes no yes yes no no yes no yes no yes no 23 19 24 24 20 21 20 13 28 38 23 27 27 39 28 14 19 23 38 26 21 30 23 IM1 IM2 IM3 IM4 IM5 IM6 IM7 IM8 IM9 IM10 IM11 IM12 IM13 IM14 IM15 IM16 IM17 IM18 IM22 IM23 IM24 IM26 IM27 IM19 IM20 IM28 IM29 IM30 IM31 Healthy Controls 73 65 75 65 65 76 58 58 66 70 70 69 68 74 57 69 56 77 56 55 77 58 74 58 ±10 42 59 60 72 53 62 F M M F F F F M M M F F M F F F M M F M M F M 3/3 M F M M F F 2 1 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 n = 12 2 2 2 2 2 2 1At time of biopsy (mean±SD) 2 Number of cases with premotor RBD and related percentages in parenthesis: n (%) Abbreviations: RBD, rapid eye movement behavior disorder; SD, Standard Deviation; UPDRS, Unified Parkinson Disease Rating Scale. https://doi.org/10.1371/journal.ppat.1011456.t001 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 3 / 16 PLOS PATHOGENSα-Synuclein seeding activity in Parkinson’s disease intestinal biopsies cases were on levodopa medication and the mean of Unified Parkinson’s Disease Rating Scale (UPDRS) III motor score was 25 ± 7. Several PDs reported moderate/severe constipation as a common condition of GI impairment with an average score of 11 ± 6, ranging from a mini- mum score of 3 to a maximum of 21. Upper GI biopsies were performed and the mucosa from proximal small intestine (duode- num) was taken from each subject and tested by RT-QuICR assay. Histology and immunohistochemistry of human intestinal mucosa biopsies To visualize the general morphology and cellular composition of IM tissues, selected biopsies were examined after staining with hematoxylin and eosin (H&E). Representative samples from two patients showed that IM biopsies mainly contain the mucosa layer (Fig 1A), or mucosa associated to a thin layer of submucosa (Fig 1B). These findings suggest that standard endo- scopic duodenal biopsies, although variable in size, always include the mucosa and may also include small amounts of the sub-mucosal layer. In both specimens, microscopic analysis did not reveal pathological findings. We also performed IHC for thyrosine hydroxylase (TH) and choline acetyltransferase (ChAT) but it was not informative due to the low sensitivity of the method (see S1 Fig and S1 Text). Assessment of IM tissue matrix inhibition of α-SynD RT-QuICR assay Previous studies had shown matrix inhibition of the RT-QuIC when testing tissues such as brain with higher concentrations of material [19, 36]. Therefore, to investigate the ability of the RT-QuICR to discriminate between PD and HC subjects, we initially tested a subset of IMs at 10−2 and 10−3 tissue dilutions (Fig 2). Our results showed inhibition of RT-QuICR amplifi- cation at 10−2, with most of the wells being negative for each sample. However, when seeding with a 10−3 IM a marked improvement was observed in the ability to discriminate between PD and HC samples. We therefore tested the remaining of the samples at 10−3 or further dilutions. Primary RT-QuICR fluorescence curves for representative PD (n = 3) and HC (n = 3) cases are shown in S2 Fig. Diagnostic sensitivity of RT-QuICR detection of α-SynD in IM biopsies We next performed a blinded RT-QuICR analysis of an additional panel of IMs. A total num- ber of 42 duodenal biopsies from 23 PD cases and 6 HCs were tested. Overall, we observed positive RT-QuICR responses in 22 out of the 23 duodenal biopsies from PD patients, and negative responses for all 6 HCs (Fig 3). The sensitivity and specificity of the RT-QuICR were Fig 1. Tissue morphology in duodenum intestinal mucosa biopsies. (A-B): Representative photomicrographs of duodenum biopsies stained with hematoxylin and eosin. Both samples showed a normal morphology, well represented mucosa (black box) and variable amounts of submucosa (black arrow) depending on the biopsy specimen. Magnification x10. https://doi.org/10.1371/journal.ppat.1011456.g001 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 4 / 16 PLOS PATHOGENSα-Synuclein seeding activity in Parkinson’s disease intestinal biopsies Fig 2. Optimization of IM tissue dilutions used for RT-QuICR detection of α-SynD. Inverse of the time to threshold for reactions seeded with IM tissue dilutions from PD or HC cases with the designated sample codes. Each symbol indicates a single reaction well with each sample analyzed in quadruplicate at both 10−2 (black triangles) and 10−3 (green squares) dilutions. A reaction time cut off of 40 h was used. The dotted line indicates the threshold for a positive reaction. https://doi.org/10.1371/journal.ppat.1011456.g002 95.7% (95% CI: 78.1% to 99.9%) and 100% (95% CI: 54.1% to 100.0%), respectively (Table 2). Positive and negative predictive values (PPV and NPV) were 100.0% and 85.7% (95% CI: 46.9–97.6%), respectively. The value of diagnostic efficacy, expressed as the ratio between the sum of subjects classified as true positives and negatives, and the sum of the subjects classified as false positives and negatives, was 96.6% (95% CI: 82.2–99.9%). Quantitation of α-SynD seeding activities in PD duodenum biopsies To measure the levels of α-SynD seeding activity in PD IM biopsies, end-point dilution analysis was performed [19]. IM samples were serially diluted from 10−3 to 10−6 for RT-QuICR analysis. Tissue concentrations of seeding units giving 50% positive replicate reactions (50% seeding doses or SD50s) were then estimated using a modified Spearman-Ka¨rber algorithm [37]. SD50 concen- trations were calculated for all the PD samples with the exception of samples 6 and 7, which did not have high enough seeding activity for the algorithm to be applied properly (Fig 4). On aver- age, RT-QuICR-positive IMs from PD patients had 4.3 ± 0.8 log SD50 per mg of tissue (range: 2.7 to 5.7). Fig 3. RT-QuICR detection of α-SynD in duodenum IM biopsies. Symbols represent the inverse of time to threshold for individual reaction wells seeded with a 10−3 IM tissue dilution from the designated PD or HC IM sample ID numbers. The dotted line indicates the threshold for a positive reaction. Bars show mean ±SD. Quadruplicate reaction wells are shown for each sample, with the exception of samples 6 and 7. The overall RT-QuICR status for the latter samples were inconclusive based on the initial quadruplicates. Upon repeating 6 and 7, as per the methods, both were determined to be positive overall and results from all 8 wells are shown here. https://doi.org/10.1371/journal.ppat.1011456.g003 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 5 / 16 PLOS PATHOGENSα-Synuclein seeding activity in Parkinson’s disease intestinal biopsies Table 2. Diagnostic performance of IM α-Syn RT-QuICR. Measures Sensitivity Specificity PLR NLR Disease prevalence PPV NPV Youden’s index Accuracy Value 95.7% 100.0% NA 0.04 79.3% 100.0% 85.7% 0.95 96.6% 95% CI 78.1% to 99.9% 54.07% to 100.0% NA 0.01 to 0. 30 60.3% to 92.0% 46.9% to 97.6% 82.2% to 99.9% Abbreviations: PLR, Positive Likelihood Ratio; NLR, Negative Likelihood Ratio; PPV, Positive Predictive Value; NPV, Negative Predictive Value; CI, Confidence interval expressed as percentages; NA: Not available, the value cannot be calculated because the denominator is 0. https://doi.org/10.1371/journal.ppat.1011456.t002 RT-QuICR testing of two IM biopsies from individual patients To address the question of α-SynD distribution within the duodenum intestinal mucosa, we tested two samples collected during the same procedure from the superior and descending part of duodenum from 9 individual PD patients and 6 HC patients (Fig 5A). With the excep- tion of patient 27, who was positive in only one of the two samples, RT-QuICR detected α- synD in both biopsies from each patient. To evaluate whether α-SynD levels were consistent across the duodenum sampling areas, end-point dilution analysis was performed on each spec- imen. Log SD50s per mg of tissue (Fig 5B) varied by 0.5±0.4 between biopsies from the same patient. Lack of correlation between α-SynD seed concentrations and clinical parameters Spearman’s correlation analysis was performed to investigate the relationship between the con- centration of α-SynD (log SD50/mg) and clinical parameters. No significant correlation between SD50 values and UPDRS III motor scores (r = – 0.23; p = 0.25; 95% CI:– 0.57% to 0.18; alpha = 0.05) (Fig 6A), constipation scores (r = 0.27; p = 0.18; 95% CI:– 0.14% to– 0.60; alpha = 0.05) (Fig 6B) and disease duration (r = 0.10; p = 0.64; 95% CI:– 0.31% to 0.47; alpha = 0.05) (Fig 6C) was observed. Lack of 3R or 3R+4R tau seeding activity in the IM biopsies For comparative purposes, we also tested PD (n = 9) and HC (n = 1) IM biopsies for the pres- ence of proteopathic seeds of another cytosolic protein, tau, using a tau RT-QuIC assay that ultrasensitively detects the tau seeds associated with either 3R (Pick’s disease) or 3R+4R (Alz- heimer’s disease and chronic traumatic encephalopathy) tauopathies [35]. This comparison was of interest because of reported evidence of tau and synuclein co-pathologies in PD and Alzheimer’s disease, and the detection of tau in Lewy bodies [30–34]. However, no tau seeding activity was detected in any of the IM specimens (Fig 7). As an indication that the IM tissue matrix did not prevent detection of tau seeds using this K12 tau RT-QuIC assay, we found that dilutions as extreme as 10−5 of sporadic Alzheimer’s disease brain homogenates spiked into IM sample dilutions of 10−2 to 10−4 were readily detected. Simultaneous dilutions of the AD brain homogenates themselves showed positivity out to 10−8 dilutions, confirming the high PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 6 / 16 PLOS PATHOGENSα-Synuclein seeding activity in Parkinson’s disease intestinal biopsies Fig 4. RT-QuICR quantification of seeding activity in PD duodenum biopsies. Orange triangles mark log SD50/mg values from individual IM biopsies. Dark red triangles indicate values for PD IMs that did not completely reach end- point (0/4 positive wells) in our serial dilutions and therefore may be slight underestimations of the seeding activity by Spearman-Ka¨rber analysis. However, each of those cases diluted to 2/4 or 1/4 positive wells which is at, or below, the definition of 1 SD50, respectively, in those diluted aliquots, meaning that values shown in the figure should be close to being accurate. The dotted line indicates the RT-QuICR detection limit. https://doi.org/10.1371/journal.ppat.1011456.g004 sensitivity of the assay for AD tau seeds. These tau RT-QuIC assay results provided evidence that no measurable 3R or 3R+4R (Alzheimer’s-like) tau seeding activity accompanied the duo- denal α-SynD seeds of PD. Discussion In this study, we have detected α-SynD, but not tau, seeding activity in enteric duodenal biop- sies from patients with a clinical diagnosis of Parkinson’s disease. We estimated the average α- SynD seeding activity to be 4.3 log SD50 per mg of tissue, which was similar to that reported for RT-QuIC analysis of both post-mortem colon and skin specimens from PD patients [20, 28]. This means that the IM seeding activity was comparable to levels that we have seen in various PD brain specimens, and >3 logs higher than levels typically seen in antemortem PD CSF samples [19, 25] The diagnostic accuracy of RT-QuICR for PD in our initial duodenum biopsy panel was high, i.e., 95.7% sensitivity and 100% specificity. By comparison, a previous analysis of lower intestinal biopsies from a small cohort of PD and healthy subjects using an alternative SAA PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 7 / 16 PLOS PATHOGENSα-Synuclein seeding activity in Parkinson’s disease intestinal biopsies Fig 5. Comparison of seeding activities in IM samples collected from first and second duodenum segments from individual patients. (A) Inverse of time to fluorescence threshold comparisons for PD (orange) and HC (blue) cases. Each IM was analyzed in quadruplicate at 10−3 dilution, except for sample 27 which shows values from 8 reaction wells, with each symbol representing an individual reaction well from either the first (biopsy 1; triangles) or second (biopsy 2; circles) duodenum segments from the given patient. (B) Log SD50/mg comparisons from end-point dilution analysis. The dark red circle represents a PD IM that did not reach end-point, and thus is likely an underestimate of the actual value. Grey triangle indicates a positive sample for which a log SD50/mg value could not be accurately calculated due to low seeding activity. Log SD50/mg for the first samplings are the same as those reported in Fig 2. The dotted lines indicate the threshold for detection (A) and the assay’s detection limit (B). https://doi.org/10.1371/journal.ppat.1011456.g005 Fig 6. Spearman correlation between α-synD seed concentration (logSD50/mg tissue) and clinical parameters. (A) UPDRSIII score versus IM log SD50/mg while patient was on Levodopa medication. (B) Constipation score and (C) Disease duration versus seed concentration. Spearman’s r and p values are inset. https://doi.org/10.1371/journal.ppat.1011456.g006 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 8 / 16 PLOS PATHOGENSα-Synuclein seeding activity in Parkinson’s disease intestinal biopsies Fig 7. Tau RT-QuIC analysis of PD IM specimens. Colored symbols represent the inverse of the time to threshold for individual reaction wells seeded with a 10−3 (green squares) or 10−4 (blue circles) IM tissue dilution from the designated PD or HC cases. Solid downward triangles indicate control reactions seeded with sAD brain homogenate (BH) of the designated dilution with respect to solid tissue. Open downward triangles display representative data from reaction wells seeded with 1:1 mixtures 10−5 sAD BH and the designated tissue dilution of HC IM30. The dotted line indicates the threshold for a positive reaction. Error bars show mean ±SD. Quadruplicate reaction wells are shown for each sample. https://doi.org/10.1371/journal.ppat.1011456.g007 previously detected α-SynD with a diagnostic sensitivity of 55.6% and a specificity of 90.9% [27]. The reason for the increased sensitivity and specificity that we observed in the duodenum biopsies is unknown, but it could be due to the differences in the section of the intestine tested (rectum vs. duodenum) in the two patient cohorts and/or to the version of SAA used [19, 26]. α-SynD can spread from gut-to-brain and brain-to-gut via vagal pathways in animal models of body-first and brain-first PD (reviewed in [10]). Since vagal innervation is highest in the upper gastrointestinal tract, one might expect to detect more α-SynD in duodenal biopsies, compared to rectal biopsies. Emmi et al. recently detected α-Syn aggregates in duodenum biopsies from 22 patients with PD using IHC [13]. Our findings are consistent with these IHC results, and show additionally that duodenal α-SynD has self-propagating (seeding) activity. We found similar amounts of seeding activity in contemporaneous biopsies from superior and descend- ing sections of duodenum from individual patients, suggestive of widespread α-SynD distribu- tion within these duodenum segments. However, we assume that multiple IM collections, when possible, would improve diagnostic sensitivity. Emmi et al. [12] studied the relationship between the extent of α-SynD and the severity of extrapyramidal symptoms, but failed to find a correlation among α-SynD immunoreactive areas and the severity of motor symptoms, the global cognitive scales’ score, the quality of life, and the UPDRS scale values. Likewise, we found no significant correlation between the α-SynD seed concentrations and UPDRS III motor scores in patients on levodopa treatment. Thus, the link between the distribution of α- SynD in peripheral tissues and the stage of disease remains unclear [38]. The extent to which factors such as bilirubin content and gut inflammation might affect α-SynD seed concentra- tions and confound the interpretation of IM RT-QuICR results remains to be determined. Although the high diagnostic sensitivity and specificity were observed in this study, our results cannot be validated by autopsy as the patients were still alive at the time this study was completed. However, the study participants were carefully chosen based on clinical criteria. Additionally, the number of patients was limited due to the availability of patients undergoing PEG-J tube placement, patient hesitation to undergo an additional, albeit minimally invasive procedure, and restrictions placed on elective medical procedures during the Covid-19 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 9 / 16 PLOS PATHOGENSα-Synuclein seeding activity in Parkinson’s disease intestinal biopsies pandemic. We caution that PEG-J for continuous L-DOPA infusion is generally used in more advanced cases of PD where oral L-DOPA is less consistently effective. Thus, the RT-QuICR results that we have obtained with our cohort of more severe PD cases may not be generaliz- able to patients with more moderate/mild PD that can be controlled well via oral L-DOPA. Another caveat is that the average age of the PD cases was 9 years older than the non-PD con- trols, and, at this time we cannot exclude the possibility that this age differential contributed in some way to our divergent RT-QuICR results for these two cohorts. In our study, 1 of 23 PD cases was RT-QuICR negative (IM16). At the time of IM homoge- nization, we noted that the tissue of this patient had an unusual hardened, thickened, and yel- low appearance. A similar observation was made in one duplicate sample from IM27, which also gave an RT-QuICR-negative result. Because of this appearance, we speculate that these specimens might have been contaminated with bilirubin derived from the breakdown of hemoglobin in senescent red blood cells. Because blood contamination can inhibit some RT-QuIC assays [39], it is possible that residual blood components may have interfered. We also observed that in one PD patient (IM7), the RT-QuICR reaction gave a weak positive sig- nal. This sample was collected from a patient affected by scleroderma, a condition which is known to damage peripheral nerves and mucosa [40] in the GI tract, most frequently in the duodenum [41]. Therefore, we hypothesize that the weak positive result for this subject could be due to vascular ischemia and fibrosis associated to scleroderma, which may have altered α- SynD deposition and distribution. Recent histopathological evidence indicates that enteric α-SynD deposition is more preva- lent in RBD-positive than RBD-negative subjects [42]. Moreover, subjects with premotor RBD and constipation fit the criteria of “body first” phenotype [8, 43]. In our cohort about half of the patients (52.2%) presented both premotor RBD and GI symptoms of constipation. Inter- estingly, the RT-QuICR-negative PD patient was also RBD-negative. However, other than this case, no significant differences in RT-QuICR detection or parameters were observed between RBD positive or negative subjects. In conclusion, RT-QuICR provides a method for testing duodenal biopsies with similar sensitivity and specificity to SAAs of other diagnostically relevant samples [18, 19, 22, 25, 29] and skin [20, 23]. Paired with end-point dilution analysis, this assay can quantitatively assess levels of seeding activity among patients as well as across tissues, or in different sections of the same tissue. This could prove to be valuable for monitoring disease progression or selecting cohorts for clinical trials and monitoring their progress. While further studies are needed to elucidate the feasibility and utility of IM RT-QuICR, the potential for intra vitam monitoring and diagnosis, for example during routine esophagogastroduodenoscopy and colonoscopy screenings, holds promise for the early diagnosis of PD. Materials and methods Ethics statement This study was approved by the Institutional Ethics Committee (Prot.PG/2017/17817) of the Azienda Ospedaliera Universitaria di Cagliari, Italy. All participants signed written informed consent documents. Participants This study was approved by the Institutional Ethics Committee (Prot.PG/2017/17817) of the Azienda Ospedaliera Universitaria di Cagliari, Italy. All participants signed written informed consent documents. A total of 29 participants in this study, 23 PD patients and 6 non-neurode- generative healthy controls (HCs) subjects, were recruited between September 2020 and PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 10 / 16 PLOS PATHOGENSα-Synuclein seeding activity in Parkinson’s disease intestinal biopsies September 2022 at the Neurology and Digestive Endoscopy Units of AO Brotzu Cagliari. Idio- pathic PD patients were diagnosed according to the UK Brain Bank criteria, and patients with atypical parkinsonism were excluded. All PD patients were evaluated by the Movement Disor- der Society UPDRS III and IV and by the Non-Motor Symptom Scale (NMSS). The presence of premotor RBD was assessed retrospectively through the REM sleep behavior disorder screening questionnaire (RBDSQ) [44]. The criteria for evaluation of constipation was based on the Constipation Scoring System [45]. A total score ranging from 0 (no constipation) to 30 with higher scores indicating more severe constipation was assigned retrospectively to each patient in according to eight variables [45], such as frequency of bowel movements, painful evacuation, incomplete evacuation, abdominal pain, length of time per attempt, assistance for defecation, unsuccessful attempts for evacuation per 24 hours, and duration of constipation. Exclusion criteria were intestinal cancer, the concomitant presence of neurological, or unstable psychiatric illness or cognitive impairment. All PD subjects included in this study were given levodopa-carbidopa intestinal gel (LCIG). HCs were matched to PD subjects for sex and age and had no neurological or cog- nitive disorders. Intestinal mucosa collection All PD patients underwent upper GI endoscopy for placement of an administration jejunal extension tube (PEG-J) for continuous levodopa enteral infusion, and intestinal mucosa biopsy. HCs were selected from those undergoing upper GI endoscopy biopsy for diagnostic investigations and surveillance. Two to four IM biopsy specimens from the proximal small intestine (duodenum) were taken for each participant (~20 mg of total tissue). All samples were collected in sterile contain- ers containing 20 to 50 mL of physiological solution and immediately frozen at –80˚C until analysis. Two additional IM biopsies were collected in sterile containers with 10% formalin for histological analysis and immunohistochemistry (IHC) examination. Histology and immunohistochemical analysis Histological examination was done on hematoxylin and eosin stained sections. Four μm thick sections were cut from formalin fixed-paraffin embedded biopsies. After rehydration, sections were incubated for 20 minutes with ready to use Carazzi’s Haematoxylin (Bio-Optica, Milan), soaked in tap water and counter-stained with 1% aqueous eosin Y solution for 20 seconds (Bio-Optica). Tissue examination was performed using a Leica DM6000 light microscope. Homogenization of intestinal mucosa samples Intestinal mucosa samples were homogenized following the protocol described by Bargar et al [19, 28] with some modifications. In brief, IMs were thawed and washed three times in 1× phosphate buffered saline (PBS) to remove any residual blood. Samples were weighed and homogenized at 10% (w/vol) in ice-cold homogenization buffer containing 1 x PBS, 0.1% Tri- ton X-100, 150 mM sodium chloride (NaCl), 5 mM of ethylene-diamino-tetraacetic acid (EDTA) and Complete Protease Inhibitor without EDTA (Roche). IM homogenization was done using zirconia beads (1 mm) in a mini-Beadbeater-16 device (Bead Mill24; Thermo Fisher) at maximum speed. Samples were homogenized with 4 rounds of 1 min each at max speed, with cooling in between rounds. Next, samples were centrifuged at 500xg for 5 min and the supernatant was distributed in single use aliquots to be stored at -80˚C for subsequent RT-QuICR analysis. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 11 / 16 PLOS PATHOGENSα-Synuclein seeding activity in Parkinson’s disease intestinal biopsies Recombinant α-Syn (K23Q) and tau (K12CFh) protein purification Human K23Q rec α-Synuclein was purified as described by Groveman et al [19]. In brief, BL21(DE3) Escherichia coli bacteria were freshly transformed with the plasmid for recombi- nant K23Q α-Syn protein expression. One liter cultures were inoculated with a single colony and grown overnight using auto-induction media [46] containing 50 μg/mL kanamycin. The cells were lysed and the periplasmic fraction was isolated using an osmotic shock and acid purification protocol modified from Paslawski et al. [47]. The protein extract was filtered and loaded for purification into a 5 ml Ni-NTA column (Cytiva) on an A¨ kta Start chromatography system (GE). Peak fractions where then further purified using a 5 ml Q-HP column (Cytiva). The purified protein was filtered, diluted to 1 mg/ml, and dialyzed overnight at 4˚C using a 3 kDa MWCO dialysis membrane. Protein concentration was determined with a UV–VIS spec- trophotometer using a theoretical extinction coefficient at 280 nm of 0.36 (mg/mL) − 1 cm − 1. Lyophilized protein aliquots were stored at − 80˚C. Human K12CFh rec tau substrate was purified as described by Metrick et al [35]. IM α-Syn RT-QuICR and tau RT-QuIC assays For α-Syn RT-QuICR assays 10% IM homogenates were thawed at room temperature, serially diluted in PBS containing N2 media supplement (Gibco) and tested by RT-QuICR using puri- fied recombinant K23Q α-Syn as substrate. In brief, RT-QuIC buffer composition was: 40 mM phosphate buffer pH 8, 170 mM NaCl, 0.1 mg/ml K23Q rec-αSyn, 10 μM Thioflavin T (ThT). A volume of 98 uL of RT-QuIC buffer was loaded into wells of a black 96-wells plate with a clear bottom (Nunc). Plates were preloaded with six silica beads (0.8 mm, molecular biology grade; OPS Diagnostics) and reactions were seeded with 2 μL of IM homogenates at the speci- fied dilutions for a final reaction volume of 100 μL. Next, plates were sealed (Nunc Interna- tional sealer) and incubated in a BMG Fluostar plate reader at 42˚C for 60 h with cycles of 1 min shake (400 rpm double orbital) and 1 min rest. ThT fluorescence measurements (450 +/− 10 nm excitation and 480+/− 10 nm emission; bottom read) were taken every 45 min. All IMs were tested with at least 4 replicate reactions per sample dilution. Single reactions were considered positive when their fluorescence exceeded our threshold of 10% of the maximum value in the plate. IM samples were considered positive when �50% of replicate wells were above our ThT threshold within the established 40-h time cutoff. Samples that had 1 out of the 4 replicate reactions score as positive were deemed inconclusive and their analysis was repeated. Then, if the total percentage of positive reaction wells exceeded 25% then the sample was considered positive. The K12 tau-RT-QuIC assay was performed as described by Metrick et al [35]. Briefly, the reaction buffer composition was: 40 mM HEPES, pH 7.4, 400 mM NaF, 40 μM heparin, 10 μM ThT, 0.1 mg/mL K12 CFh substrate. A volume of 48 μL of reaction buffer was added into wells of a black 384-wells plate and seeded with 2 μL PD IM homogenate dilutions alone or spiked with 10−5 AD brain at 1:1 volumetric ratios with the designated dilution of PD IM. Plates were sealed with sealing tape and placed in an Omega FLUOStar plate reader at 42˚C and subjected to rounds of 1 min shaking, 500 rpm, orbital, and 1 min rest, with ThT fluorescence reads (450 excitation, 480 emission) taken every 15 min. Reaction wells were considered positive on an individual reaction basis when their fluorescence value exceeded a threshold value of the mean fluorescence value plus 100 times the standard deviation from the first 5 h of readings. End-point dilution analysis and 50% seeding dose (SD50) calculation IMs were serially diluted 10-fold for end point analysis. Two μl of each IM dilution (10−3 to 10−6) were used to seed individual reactions. RT-QuICR testing was done as reported above. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 12 / 16 PLOS PATHOGENSα-Synuclein seeding activity in Parkinson’s disease intestinal biopsies The amount of sample giving 50% positive replicate reactions (SD50) was estimated as described by Srivastava et al. [37]. Statistical analysis Student’s test and Fisher’s exact test was used to analyze differences in demographic character- istics between cases and controls. Variability of the data was presented as mean ± standard deviation or as percentages. Measures of diagnostic performance, such as sensitivity, specific- ity, negative predictive value, and positive predictive value, were determined for comparisons between clinical diagnosis and RT-QuICR results. Measures were expressed as percentages with 95% confidence interval (95% CI). Spearman’s rank correlation coefficient was used to determinate the correlation between SD50 values and motor UPDRSIII scores, disease dura- tion and constipation score (95% CI; alpha = 0.05). All statistical analyses were performed and plotted using the Prism software (v.8 GraphPad). Statistical significance was set at p < 0.05. Supporting information S1 Fig. Immunohistochemistry analysis of human duodenum IM biopsies. (A) Immunohis- tochemistry for tyrosine hydroxylase (TH). (B) Immunohistochemistry for choline acetyl- transferase (ChAT). (TIF) S2 Fig. Representative ThT fluorescence curves in IM α-Syn RT-QuICR. (A) Traces from single reactions (n = 4 per case) seeded with 10−3 dilutions of duodenum IM biopsies from PDs cases 4, 5, and 15 (as colored) as a function of reaction time. (B) Traces from healthy con- trol duodenum IM biopsies from cases 29, 30, and 31. (TIF) S1 Data. Fig 2 tab: Inverse time to threshold values (h-1) for individual quadruplicate wells seeded with 10−2 or 10−3 dilutions of IM samples with the IM sample ID numbers shown in the “Patient” column. Fig 3 tab: Inverse time to threshold values (h-1) for individual quadrupli- cate wells seeded with 10−3 dilutions of IM samples with the IM sample ID numbers shown in the “Patient” column. For samples 6 and 7, data from an additional 4 wells are shown. Fig 4 tab: log SD50/mg tissue values calculated for individual IM biopsies from end-point dilution data. Fig 5A tab: Inverse time to fluorescence threshold comparisons between first (biopsy 1) or second (biopsy 2) duodenum segments from a given patient for the IM sample ID numbers shown in the “Patient” column. Data are from quadruplicate wells seeded with 10−3 IM dilu- tions. Fig 5B tab: Log SD50/mg tissue values calculated for two IM biopsies from the same patient from end-point dilution analysis. Fig 6 tab: Individual IM log SD50/mg tissue values, UPDRSIII scores, constipation scores, and disease durations (y) associated with the designated IM sample ID in the “Patient” column. Fig 7 tab: Top 2 rows: Inverse time to threshold values (h-1) for individual quadruplicate wells seeded with 10−3 or 10−4 dilutions of IM samples with the IM sample ID numbers shown in the “Patient” row. sAD group: Analogous data from reac- tions seeded with 10−5 or 10−8 dilutions of sAD brain homogenate. The sAD spiked 10−5 group: data from reaction wells seeded with 1:1 mixtures 10−5 sAD BH and the designated tis- sue dilution of HC IM30. S2A and S2B Figs tabs: Quadruplicate ThT fluorescence readings as function of reaction time (h) for reactions seeded with 10−3 dilutions of IM sample IDs given the top row. (XLSX) S1 Text. Immunohistochemistry analysis of human intestinal mucosa biopsies. (DOCX) PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 13 / 16 PLOS PATHOGENSα-Synuclein seeding activity in Parkinson’s disease intestinal biopsies Acknowledgments The authors kindly acknowledge the Digestive Endoscopy Unit of AO Brotzu Cagliari for the assistance with subject recruitment. Author Contributions Conceptualization: Sarah Vascellari, Christina D. Orru`, Giovanni Cossu, Byron Caughey. Data curation: Sarah Vascellari, Christina D. Orru`, Bradley R. Groveman, Sabiha Parveen, Giuseppe Fenu, Giada Pisano, Giuseppe Piga, Giulia Serra, Valentina Oppo, Daniela Murgia, Andrew G. Hughson, Giovanni Cossu. Formal analysis: Sarah Vascellari, Christina D. Orru`, Bradley R. Groveman, Sabiha Parveen, Andrew G. Hughson. Funding acquisition: Byron Caughey. Investigation: Sarah Vascellari, Christina D. Orru`, Bradley R. Groveman, Sabiha Parveen, Giuseppe Fenu, Giada Pisano, Giuseppe Piga, Giulia Serra, Valentina Oppo, Daniela Murgia, Andrea Perra, Andrew G. Hughson, Giovanni Cossu. Methodology: Sarah Vascellari, Christina D. Orru`. Project administration: Sarah Vascellari, Christina D. Orru`. Resources: Sarah Vascellari, Christina D. Orru`, Giovanni Cossu, Byron Caughey. Supervision: Cathryn L. Haigh, Aldo Manzin, Giovanni Cossu, Byron Caughey. Validation: Sarah Vascellari, Christina D. Orru`. Visualization: Sarah Vascellari, Christina D. Orru`, Bradley R. Groveman, Sabiha Parveen, Andrea Perra, Fabrizio Angius. Writing – original draft: Sarah Vascellari, Christina D. Orru`, Byron Caughey. Writing – review & editing: Sarah Vascellari, Christina D. Orru`, Bradley R. Groveman, Andrea Perra, Fabrizio Angius, Aldo Manzin, Giovanni Cossu, Byron Caughey. References 1. Braak H., et al., Staging of brain pathology related to sporadic Parkinson’s disease. Neurobiol Aging, 2003. 24(2): p. 197–211. https://doi.org/10.1016/s0197-4580(02)00065-9 PMID: 12498954 2. Hawkes C.H., Del Tredici K., and Braak H., Parkinson’s disease: a dual-hit hypothesis. Neuropathol Appl Neurobiol, 2007. 33(6): p. 599–614. https://doi.org/10.1111/j.1365-2990.2007.00874.x PMID: 17961138 3. Ulusoy A., et al., Brain-to-stomach transfer of alpha-synuclein via vagal preganglionic projections. Acta Neuropathol, 2017. 133(3): p. 381–393. 4. Cersosimo M.G. and Benarroch E.E., Pathological correlates of gastrointestinal dysfunction in Parkin- son’s disease. Neurobiol Dis, 2012. 46(3): p. 559–64. https://doi.org/10.1016/j.nbd.2011.10.014 PMID: 22048068 5. Pfeiffer R.F., Gastrointestinal dysfunction in Parkinson’s disease. Parkinsonism Relat Disord, 2011. 17 (1): p. 10–5. https://doi.org/10.1016/j.parkreldis.2010.08.003 PMID: 20829091 6. Han J.Y., Shin C., and Choi Y.P., Preclinical Detection of Alpha-Synuclein Seeding Activity in the Colon of a Transgenic Mouse Model of Synucleinopathy by RT-QuIC. Viruses, 2021. 13(5). https://doi.org/10. 3390/v13050759 PMID: 33926043 7. Borghammer P., et al., Neuropathological evidence of body-first vs. brain-first Lewy body disease. Neu- robiol Dis, 2021. 161: p. 105557. https://doi.org/10.1016/j.nbd.2021.105557 PMID: 34763110 8. Horsager J., et al., Brain-first versus body-first Parkinson’s disease: a multimodal imaging case-control study. Brain, 2020. 143(10): p. 3077–3088. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 14 / 16 PLOS PATHOGENSα-Synuclein seeding activity in Parkinson’s disease intestinal biopsies 9. Horsager J., Knudsen K., and Sommerauer M., Clinical and imaging evidence of brain-first and body- first Parkinson’s disease. Neurobiol Dis, 2022. 164: p. 105626. https://doi.org/10.1016/j.nbd.2022. 105626 PMID: 35031485 10. Van Den Berge N. and Ulusoy A., Animal models of brain-first and body-first Parkinson’s disease. Neu- robiol Dis, 2022. 163: p. 105599. https://doi.org/10.1016/j.nbd.2021.105599 PMID: 34952161 11. Corbille A.G., Neunlist M., and Derkinderen P., Cross-linking for the analysis of alpha-synuclein in the enteric nervous system. J Neurochem, 2016. 139(5): p. 839–847. 12. Ruffmann C., et al., Detection of alpha-synuclein conformational variants from gastro-intestinal biopsy tissue as a potential biomarker for Parkinson’s disease. Neuropathol Appl Neurobiol, 2018. 44(7): p. 722–736. https://doi.org/10.1111/nan.12486 PMID: 29676021 13. Emmi A., et al., Duodenal alpha-Synuclein Pathology and Enteric Gliosis in Advanced Parkinson’s Dis- ease. Mov Disord, 2023. https://doi.org/10.1002/mds.29358 PMID: 36847308 14. Beck G., et al., Detection of Phosphorylated Alpha-Synuclein in the Muscularis Propria of the Gastroin- testinal Tract Is a Sensitive Predictor for Parkinson’s Disease. Parkinsons Dis, 2020. 2020: p. 4687530. https://doi.org/10.1155/2020/4687530 PMID: 33029342 15. Atarashi R., et al., Ultrasensitive human prion detection in cerebrospinal fluid by real-time quaking- induced conversion. Nat Med, 2011. 17(2): p. 175–8. https://doi.org/10.1038/nm.2294 PMID: 21278748 16. Wilham J.M., et al., Rapid end-point quantitation of prion seeding activity with sensitivity comparable to bioassays. PLoS Pathog, 2010. 6(12): p. e1001217. https://doi.org/10.1371/journal.ppat.1001217 PMID: 21152012 17. De Luca C.M.G., et al., Efficient RT-QuIC seeding activity for alpha-synuclein in olfactory mucosa sam- ples of patients with Parkinson’s disease and multiple system atrophy. Transl Neurodegener, 2019. 8: p. 24. 18. Fairfoul G., et al., Alpha-synuclein RT-QuIC in the CSF of patients with alpha-synucleinopathies. Ann Clin Transl Neurol, 2016. 3(10): p. 812–818. https://doi.org/10.1002/acn3.338 PMID: 27752516 19. Groveman B.R., et al., Rapid and ultra-sensitive quantitation of disease-associated alpha-synuclein seeds in brain and cerebrospinal fluid by alphaSyn RT-QuIC. Acta Neuropathol Commun, 2018. 6(1): p. 7. 20. Kuzkina A., et al., Diagnostic value of skin RT-QuIC in Parkinson’s disease: a two-laboratory study. NPJ Parkinsons Dis, 2021. 7(1): p. 99. https://doi.org/10.1038/s41531-021-00242-2 PMID: 34782640 21. Luan M., et al., Diagnostic Value of Salivary Real-Time Quaking-Induced Conversion in Parkinson’s Disease and Multiple System Atrophy. Mov Disord, 2022. 37(5): p. 1059–1063. https://doi.org/10.1002/ mds.28976 PMID: 35278004 22. Manne S., et al., Ultrasensitive Detection of Aggregated alpha-Synuclein in Glial Cells, Human Cerebro- spinal Fluid, and Brain Tissue Using the RT-QuIC Assay: New High-Throughput Neuroimmune Bio- marker Assay for Parkinsonian Disorders. J Neuroimmune Pharmacol, 2019. 14(3): p. 423–435. 23. Manne S., et al., Blinded RT-QuIC Analysis of alpha-Synuclein Biomarker in Skin Tissue From Parkin- son’s Disease Patients. Mov Disord, 2020. 35(12): p. 2230–2239. 24. Rossi M., et al., Ultrasensitive RT-QuIC assay with high sensitivity and specificity for Lewy body-associ- ated synucleinopathies. Acta Neuropathol, 2020. 140(1): p. 49–62. https://doi.org/10.1007/s00401- 020-02160-8 PMID: 32342188 25. Russo M.J., et al., High diagnostic performance of independent alpha-synuclein seed amplification assays for detection of early Parkinson’s disease. Acta Neuropathol Commun, 2021. 9(1): p. 179. https://doi.org/10.1186/s40478-021-01282-8 PMID: 34742348 26. Shahnawaz M., et al., Development of a Biochemical Diagnosis of Parkinson Disease by Detection of alpha-Synuclein Misfolded Aggregates in Cerebrospinal Fluid. JAMA Neurol, 2017. 74(2): p. 163–172. 27. Fenyi A., et al., Detection of alpha-synuclein aggregates in gastrointestinal biopsies by protein misfold- ing cyclic amplification. Neurobiol Dis, 2019. 129: p. 38–43. https://doi.org/10.1016/j.nbd.2019.05.002 PMID: 31078683 28. Bargar C., et al., Streamlined alpha-synuclein RT-QuIC assay for various biospecimens in Parkinson’s disease and dementia with Lewy bodies. Acta Neuropathol Commun, 2021. 9(1): p. 62. https://doi.org/ 10.1186/s40478-021-01175-w PMID: 33827706 29. Orru C.D., et al., A rapid alpha-synuclein seed assay of Parkinson’s disease CSF panel shows high diagnostic accuracy. Ann Clin Transl Neurol, 2021. 8(2): p. 374–384. 30. Giannini G., et al., In vivo assessment of Lewy body and beta-amyloid copathologies in idiopathic nor- mal pressure hydrocephalus: prevalence and associations with clinical features and surgery outcome. Fluids Barriers CNS, 2022. 19(1): p. 71. https://doi.org/10.1186/s12987-022-00368-2 PMID: 36071460 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 15 / 16 PLOS PATHOGENSα-Synuclein seeding activity in Parkinson’s disease intestinal biopsies 31. Hall S., et al., Performance of alphaSynuclein RT-QuIC in relation to neuropathological staging of Lewy body disease. Acta Neuropathol Commun, 2022. 10(1): p. 90. 32. Pan L., et al., Tau in the Pathophysiology of Parkinson’s Disease. J Mol Neurosci, 2021. 71(11): p. 2179–2191. https://doi.org/10.1007/s12031-020-01776-5 PMID: 33459970 33. Rossi M., et al., Diagnostic Value of the CSF alpha-Synuclein Real-Time Quaking-Induced Conversion Assay at the Prodromal MCI Stage of Dementia With Lewy Bodies. Neurology, 2021. 97(9): p. e930– e940. 34. Twohig D. and Nielsen H.M., alpha-synuclein in the pathophysiology of Alzheimer’s disease. Mol Neu- rodegener, 2019. 14(1): p. 23. 35. Metrick M.A., 2nd, et al., A single ultrasensitive assay for detection and discrimination of tau aggregates of Alzheimer and Pick diseases. Acta Neuropathol Commun, 2020. 8(1): p. 22. https://doi.org/10.1186/ s40478-020-0887-z PMID: 32087764 36. Hoover C.E., et al., Endogenous Brain Lipids Inhibit Prion Amyloid Formation In Vitro. J Virol, 2017. 91 (9). https://doi.org/10.1128/JVI.02162-16 PMID: 28202758 37. Srivastava A., Alam P., and Caughey B., RT-QuIC and Related Assays for Detecting and Quantifying Prion-like Pathological Seeds of alpha-Synuclein. Biomolecules, 2022. 12(4). 38. Xia R. and Mao Z.H., Progression of motor symptoms in Parkinson’s disease. Neurosci Bull, 2012. 28 (1): p. 39–48. https://doi.org/10.1007/s12264-012-1050-z PMID: 22233888 39. Vascellari S., Orru C.D., and Caughey B., Real-Time Quaking- Induced Conversion Assays for Prion Diseases, Synucleinopathies, and Tauopathies. Front Aging Neurosci, 2022. 14: p. 853050. https://doi. org/10.3389/fnagi.2022.853050 PMID: 35360213 40. McFarlane I.M., et al., Gastrointestinal Manifestations of Systemic Sclerosis. Rheumatology (Sunny- vale), 2018. 8(1). https://doi.org/10.4172/2161-1149.1000235 PMID: 30057856 41. Marie I., et al., Outcome of small-bowel motor impairment in systemic sclerosis—a prospective mano- metric 5-yr follow-up. Rheumatology (Oxford), 2007. 46(1): p. 150–3. https://doi.org/10.1093/ rheumatology/kel203 PMID: 16782730 42. Leclair-Visonneau L., et al., REM sleep behavior disorder is related to enteric neuropathology in Parkin- son disease. Neurology, 2017. 89(15): p. 1612–1618. https://doi.org/10.1212/WNL. 0000000000004496 PMID: 28887374 43. Chen Y., et al., REM sleep behavior disorder correlates with constipation in de novo Chinese Parkin- son’s disease patients. Neurol Sci, 2023. 44(1): p. 191–197. https://doi.org/10.1007/s10072-022- 06381-5 PMID: 36098886 44. Marelli S., et al., National validation and proposed revision of REM sleep behavior disorder screening questionnaire (RBDSQ). J Neurol, 2016. 263(12): p. 2470–2475. https://doi.org/10.1007/s00415-016- 8285-y PMID: 27671482 45. Agachan F., et al., A constipation scoring system to simplify evaluation and management of constipated patients. Dis Colon Rectum, 1996. 39(6): p. 681–5. https://doi.org/10.1007/BF02056950 PMID: 8646957 46. Fox B.G. and Blommel P.G., Autoinduction of protein expression. Curr Protoc Protein Sci, 2009. Chap- ter 5: p. 5 23 1–5 23 18. https://doi.org/10.1002/0471140864.ps0523s56 PMID: 19365792 47. Paslawski W., Lorenzen N., and Otzen D.E., Formation and Characterization of alpha-Synuclein Oligo- mers. Methods Mol Biol, 2016. 1345: p. 133–50. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011456 June 30, 2023 16 / 16 PLOS PATHOGENS
10.1371_journal.pone.0285473	RESEARCH ARTICLE Cellular apoptosis and cell cycle arrest as potential therapeutic targets for eugenol derivatives in Candida auris Hammad Alam1, Vartika Srivastava1, Windy Sekgele2, Mohmmad Younus WaniID Abdullah Saad Al-Bogami3, Julitha Molepo2, Aijaz Ahmad1,4 3, 1 Faculty of Health Sciences, Department of Clinical Microbiology and Infectious Diseases, School of Pathology, University of the Witwatersrand, Johannesburg, South Africa, 2 Faculty of Health Sciences, Department of Oral Biological Sciences, School of Oral Health Sciences, University of the Witwatersrand, Johannesburg, South Africa, 3 Department of Chemistry, College of Science, University of Jeddah, Jeddah, Saudi Arabia, 4 Infection Control, Charlotte Maxeke Johannesburg Academic Hospital, National Health Laboratory Service, Johannesburg, South Africa * mwani@uj.edu.sa (MYW); Julitha.Molepo@wits.ac.za (JM) Abstract Candida auris, the youngest Candida species, is known to cause candidiasis and candide- mia in humans and has been related to several hospital outbreaks. Moreover, Candida auris infections are largely resistant to the antifungal drugs currently in clinical use, necessitating the development of novel medications and approaches to treat such infections. Following up on our previous studies that demonstrated eugenol tosylate congeners (ETCs) to have anti- fungal activity, several ETCs (C1-C6) were synthesized to find a lead molecule with the req- uisite antifungal activity against C. auris. Preliminary tests, including broth microdilution and the MUSE cell viability assay, identified C5 as the most active derivative, with a MIC value of 0.98 g/mL against all strains tested. Cell count and viability assays further validated the fun- gicidal activity of C5. Apoptotic indicators, such as phosphatidylserine externalization, DNA fragmentation, mitochondrial depolarization, decreased cytochrome c and oxidase activity and cell death confirmed that C5 caused apoptosis in C. auris isolates. The low cytotoxicity of C5 further confirmed the safety of using this derivative in future studies. To support the conclusions drawn in this investigation, additional in vivo experiments demonstrating the antifungal activity of this lead compound in animal models will be needed. Introduction Candida auris was identified as a novel human pathogen in 2009, and it has subsequently caused considerable healthcare problems by producing systemic infections in individuals with underlying diseases [1–3]. Even though C. auris shares a strong evolutionary relationship with other pathogenic Candida species, it differs from them in terms of biology, genetics, epidemi- ology, antifungal resistance, virulence, host adaptation, and transmission [4,5]. It was recently identified as a critically important fungal pathogen due to its inherent resistance to the major- ity of presently used antifungal medications, as well as pan-resistance in certain isolates. [6]. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Alam H, Srivastava V, Sekgele W, Wani MY, Al-Bogami AS, Molepo J, et al. (2023) Cellular apoptosis and cell cycle arrest as potential therapeutic targets for eugenol derivatives in Candida auris. PLoS ONE 18(6): e0285473. https:// doi.org/10.1371/journal.pone.0285473 Editor: Muhammad Yasir, University of New South Wales, AUSTRALIA Received: January 7, 2023 Accepted: April 24, 2023 Published: June 21, 2023 Copyright: © 2023 Alam et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are available at: https://www.ebi.ac.uk/biostudies/ studies/S-BSST1078. Funding: The author(s) received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 1 / 15 PLOS ONEEugenol derivatives as antifungal agents against Candida auris Echinocandins are often used to prevent C. auris infections, particularly in patients in critical care or who have had invasive surgical procedures [7,8]. The high mortality rates associated with this pathogen have been attributed to multidrug resistance, accompanying hospital out- breaks, and invasive infections [9,10]. Another clinically important problem is the misdiagno- sis of C. auris, as laboratory yeast identification methods frequently mistake it with other yeasts [11]. Hence, to combat infections caused by C. auris and other developing fungal dis- eases, it is imperative to develop novel, secure, and effective antifungal drugs, and treatment strategies with a range of pharmacological targets. Natural product-based antifungal medications are typically regarded as being efficient, affordable, and safer with less toxicity [12]. Due to the antibacterial, antiviral, antifungal, anti- cancer, anti-inflammatory, and antioxidant properties associated with eugenol and its deriva- tives, they have long been used in cosmetology, medicine, and pharmacology [13,14]. Eugenol and other monoterpene phenols have been demonstrated to inhibit the production of ergos- terol and, also inhibit efflux pump inhibitors in C. albicans and other non-albicans Candida species, resulting in the reversal of drug resistance among these pathogens [15]. In our previ- ous studies, derivatization of eugenol led to the development of eugenol tosylates (ETC’s) with improved antifungal efficacy and safety profile than the parent compound eugenol [14,16,17]. In this study different ETCs (C1–C6) were synthesized in a quest to find a lead molecule with desired antifungal activity against C. auris. Materials and methods The chemical reagents and solvents were procured from Sigma Aldrich and Merck Germany. TLC plates used were precoated aluminium sheets (silica gel 60 F254, Merck Germany) and visualization was done by UV light in a UV cabinet. Heraeus Vario EL III analyser was used for elemental analysis. Bruker ALPHA FT-IR spectrometer (Eco-ATR) was used for FTIR analysis. Bruker AVANCE 400 spectrometer (400 MHz) was used for 1H and 13C NMR spectra using DMSO-d6/CDCl3 as solvent with TMS (Tetramethylsilane) as standard. ESI-MS positive ion mode was recorded on Micromass Quattro II triple quadrapole mass spectrometer. Synthesis Eugenol, also known as 2-methoxy-4-(prop-2-en-1-yl) phenol, was used as the starting mate- rial for all the compounds C1–C6. For their synthesis, eugenol was treated with various phenyl and substituted phenylsulfonyl chlorides in refluxing pyridine for 18–24 hours. (Completion of reaction was monitored by TLC). After the completion of the reaction, the reaction mass was quenched with distilled water and extracted with dichloromethane. Finally, the combined organic layers were washed with distilled water again and dried over anhydrous Na2SO4. The compounds were further purified using column chromatography using dichloromethane and methanol in the ratio of 7:3 as eluant. The detailed synthesis route has been discussed previ- ously [18]. 2-methoxy-4-(prop-2-en-1-yl)phenyl-4-methylbenzenesulfonate (C1). Yield: 85%; Anal. Calc. for C17H18O4S; C, 64.13; H, 5.70%, found; C, 64.28; H, 5.56%; IRmaxcm-1: 3035 (C-H stretch), 1585 (C = C, Ar), 1385, 1155 (S = O); 1H NMR (DMSO-d6) (ppm): 7.87–7.59 (4H, m, Ar-H), 6.88–6.78 (3H, m, Ar-H), 6.25 (1H, m), 4.86 (1H, dd, J = 15.2 Hz, 6.8 Hz), 4.48 (1H, dd, J = 15.2 Hz, 6.8 Hz), 3.90 (2H, d, CH2), 3.64 (3H, s, OCH3), 2.25 (3H, s, CH3); 13CNMR (DMSO-d6) (ppm): 148.9, 144.6, 138.5, 137.0, 135.4, 133.5, 130.6, 129.3, 128.0, 122.5, 118.5, 117.0, 115.7, 56.4, 47.4, 24.8; ESI-MS m/z [M+H]+ 319.10; [M+Na]+ 341.08. 2-methoxy-4-(prop-2-en-1-yl)phenyl-4-nitrobenzenesulfonate (C2). Yield: 89%; Anal. Calc. for C16H15NO6S; C, 55.01; H, 4.33%; N, 4.01; found; C, 55.28; H, 4.30%, N, 4.20; PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 2 / 15 PLOS ONEEugenol derivatives as antifungal agents against Candida auris IRmaxcm-1: 3035 (C-H stretch), 1582 (C = C, Ar), 1385, 1152 (S = O); 1H NMR (DMSO-d6) (ppm): 8.48–8.36 (4H, m, Ar-H), 6.88–6.78 (3H, m, Ar-H), 6.28 (1H, m), 4.97 (1H, dd, J = 15.2 Hz, 6.8 Hz), 4.46 (1H, dd, J = 15.2 Hz, 6.8 Hz), 3.94 (2H, d, CH2), 3.40 (3H, s, OCH3); 13CNMR (DMSO-d6) (ppm): 152.8, 150.5, 141.7, 139.0, 138.0, 135.3, 128.4, 124.7, 122.1, 120.9, 116.4, 112.6, 56.5, 43.0; ESI-MS m/z [M+H]+ 350.42; [M+Na]+ 372.40. 2-methoxy-4-(prop-2-en-1-yl)phenyl-4-iodobenzenesulfonate (C3). Yield: 82%; Anal. Calc. for C16H15IO4S; C, 44.67; H, 3.51%, found; C, 44.76; H, 3.62%; IRmaxcm-1: 3038 (C-H stretch), 1584 (C = C, Ar), 1384, 1155 (S = O); 1H NMR (DMSO-d6) (ppm): 7.64–7.39 (4H, m, Ar-H), 6.99–6.94 (3H, m, Ar-H), 6.25 (1H, m), 4.79 (1H, dd, J = 15.2 Hz, 6.8 Hz), 4.49 (1H, dd, J = 15.2 Hz, 6.8 Hz), 3.90 (2H, d, CH2), 3.61 (3H, s, OCH3); 13CNMR (DMSO-d6) (ppm): 150.9, 142.5, 138.5, 138.0, 135.1, 134.2, 130.7, 123.8, 121.5, 117.5, 112.1, 101.9, 55.8, 40.3; ESI-MS m/z [M+H]+ 331.30; [M+Na]+ 345.26. 2-methoxy-4-(prop-2-en-1-yl)phenyl-4-bromobenzenesulfonate (C4). Yield: 80%; Anal. Calc. for C16H15BrO4S; C, 50.14; H, 3.95%, found; C, 55.25; H, 4.05%; IRmaxcm-1: 3034 (C-H stretch), 1585 (C = C, Ar), 1386, 1158 (S = O); 1H NMR (DMSO-d6) (ppm): 7.84–7.68 (4H, m, Ar-H), 7.00–6.80 (3H, m, Ar-H), 6.21 (1H, m), 4.83 (1H, dd, J = 15.2 Hz, 6.8 Hz), 4.45 (1H, dd, J = 15.2 Hz, 6.8 Hz), 3.87 (2H, d, CH2), 3.64 (3H, s, OCH3); 13CNMR (DMSO-d6) (ppm): 150.4, 139.6, 137.8, 136.9, 134.8, 133.7, 129.5, 122.8, 119.9, 116.4, 112.6, 56.5, 44.9; ESI-MS m/z [M+H]+ 384.26; [M+Na]+ 406.30. 2-methoxy-4-(prop-2-en-1-yl)phenyl-4-(bromomethyl)benzenesulfonate (C5). Yield: 70%; Anal. Calc. for C17H17O4S; C, 64.13; H, 5.70%, found; C, 64.28; H, 5.56%; IRmaxcm-1: 3035 (C-H stretch), 1585 (C = C, Ar), 1387, 1155 (S = O); 1H NMR (DMSO-d6) (ppm): 7.84– 7.68 (4H, m, Ar-H), 7.00–6.80 (3H, m, Ar-H), 6.21 (1H, m), 4.83 (1H, dd, J = 15.2 Hz, 6.8 Hz), 4.46 (1H, dd, J = 15.2 Hz, 6.8 Hz), 4.20 (2H, s, CH2), 3.87 (2H, d, CH2), 3.64 (3H, s, OCH3); 13CNMR (DMSO-d6) (ppm): 150.6, 143.2, 138.8, 138.0, 136.7, 133.8, 130.6, 127.8, 123.5, 123.0, 117.2, 113.7, 56.4, 42.5, 34.0; ESI-MS m/z [M+H]+ 398.30; [M+Na]+ 420.40. 2-methoxy-4-(prop-2-en-1-yl)phenyl-4-propylbenzenesulfonate (C6). Yield: 75%; Anal. Calc. for C19H22O4S; C, 65.87; H, 6.40%, found; C, 65.95; H, 6.48%; IRmaxcm-1: 3036 (C-H stretch), 1587 (C = C, Ar), 1385, 1156 (S = O); 1H NMR (DMSO-d6) (ppm): 7.80–7.54 (4H, m, Ar-H), 6.68–6.57 (3H, m, Ar-H), 6.25 (1H, m), 4.83 (1H, dd, J = 15.2 Hz, 6.8 Hz), 4.57 (1H, dd, J = 15.2 Hz, 6.8 Hz), 3.91 (2H, d, CH2), 3.61 (3H, s, OCH3), 2.65 (2H, t, CH2), 1.63 (2H, m, CH2), 1.11 (3H, t, CH3); 13CNMR (DMSO-d6) (ppm): 151.0, 146.1, 139.5, 138.0, 135.4, 132.6, 128.8, 126.6, 121.4, 120.1, 115.0, 112.8, 56.5, 42.7, 36.7, 24.3, 13.8; ESI-MS m/z [M+H]+ 347.45; [M+Na]+ 369.46. Ethics statement All the Candida auris isolates (n = 5) were obtained from the Division of Mycology, National Institute of Communicable Diseases (NICD), Johannesburg, South Africa. To use these isolates in this study, an ethics waiver was obtained from the Human Research Ethics Committee of University of the Witwatersrand (M140159) and performed according to guidelines outlined in the Helsinki Declaration. All the isolates were revived on Sabouraud Dextrose Agar (SDA) before the experiments. Antifungal activity of eugenol derivatives (C1–C6) Antifungal activity of the newly synthesized compounds was done against C. auris isolates fol- lowing Clinical and Laboratory Standards Institute (CLSI) recommendations M27-A3 guide- lines [19]. The stock concentrations of all the test congeners were prepared to 5000 μg/ml using 1% DMSO, leading to the final test concentrations ranged from 1250–11.34 μg/ml after PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 3 / 15 PLOS ONEEugenol derivatives as antifungal agents against Candida auris serial dilution. After incubation at 37˚C for 24 hours MIC values were visually observed as the lowermost concentrations of the test compounds (C1–C6) at which no fungal growth was seen. Amphotericin B and 1% DMSO were used as positive and negative controls in the sus- ceptibility assays. To further determine MFC, all wells that showed no growth were sub-cultured on SDA plates. After incubation at 37˚C for 24 hours, MFC were recorded as the least concentrations of test congeners that resulted in total fungal mortality (99.9%). Cell viability assay In presence or absence of the active compound (C5) the cell viability of C. auris 6057 was checked with the Guava1 Muse1 Cell Analyzer following manufacturer’s instructions. Briefly, for cell viability C. auris cells were grown till log-phase and then centrifuged for 5 min- utes at 4000 g. The cells were then washed three times with phosphate buffer saline (PBS), fol- lowed by exposing with the different concentrations (½ MIC, MIC, 2MIC) of C-5 at 37˚C for 4 hours. After incubation cells were washed and resuspended in PBS. Cell viability was quanti- fied by adding 20 μL of cell suspension and 380 μL of count & viability reagents at room tem- perature for 5 minutes in dark. The results were calculated by reading cell count and cell viability using the Guava1 Muse1 Cell Analyzer. Apoptotic studies Protoplast preparation. The apoptotic studies were done by making the protoplast of C. auris healthy or C5 treated cells. Protoplasts were made from C. auris MRL6057 cells using the previously published protocol [18]. Effect of C5 on membrane potential of mitochondria (Δψm). The effect of the most active compound C5 on C. auris mitochondrial membrane potential (MMP Δψm) was quantified by using the JC-10 kit (Abcam, UK), according to the directions of manufacture. Briefly, 90 μL of the prepared protoplasts were added with 50 μL JC-10 dye into vibrant bottom and blackened walled 96-well plates and left in dark at room temperature for 1 hour. After incubation period, 50 μL vol- ume of kit’s buffer-B was added, and then microtiter plate was rotated at 800g for 2 minutes. The excitation-emission maxima ratio (Ex/Em = 490/530nm and 540/590nm) was calculated using a microplate reader (Spectra-Max iD-3 multi-mode, USA). The fluorescence intensity (green fluo- rescence) referred to as X was calculated by using Ex/Em = 490/530nm, while the red fluorescence referred to as Y was calculated by using Ex/Em = 540/590nm. The reduction in mitochondrial membrane potential (MMP) in exposed cells was measured using the aggregate/monomeric (Y- mean/X-mean) ratio of JC10 dye. The decrease in ratio was regarded as depolarization of the mitochondrial membrane. During the tests, negative (untreated cells) and positive (treated with 10 mM H2O2) controls were also included. Effect on C. auris cytochrome c oxidase discharge. Using a previously described proto- col, the effect of a C5 on cytochrome c oxidase release in C. auris was studied [20]. Briefly 1×106 CFU/mL C. auris cells were exposed with different concentrations (½ MIC, MIC and 2 MIC) of compound C5 for 4 hours at 37˚C with shaking. The positive control (10 mM H2O2 treated cells) and the negative control (untreated C. cells) were also included. After exposure to test compounds cells were harvested, rinsed with PBS, and then homogenized in buffer-A (EDTA 1mM, phenylmethylsulfonyl fluoride (PMSF) 1mM, tris base 50mM, 7.5pH). After homogenization cells were subjected to another cycle of centrifugation at 4000g for 10 min- utes. The supernatant was collected separately in microcentrifuge tubes and centrifuged at 15,000 g for 45 minutes. The cell free suspension was pipetted out in microcentrifuge tubes and used to calculate the cytochrome c concentration in the cytoplasm. Meanwhile, the pellet PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 4 / 15 PLOS ONEEugenol derivatives as antifungal agents against Candida auris was dispersed in buffer-B (EDTA 2mM, tris base 50mM; 5.0pH) and used to calculate the quantity of cytochrome c in mitochondria. Before taking absorbance ascorbic acid (500 mg/ mL) was added in 1:1 ratio, to calculate the amount of cytochrome c in mitochondria and cyto- sol by taking absorbance at 550 nm with a UV-1800 SHIMADZU spectrophotometer. DNA disintegration analysis by TUNEL-assay. The TUNEL assay, also referred to as ter- minal deoxy-nucleotidyl transferase dUTP nick end labelling staining, is used to identify DNA breaks or fragmentations produced during the last stages of apoptosis. For the TUNEL assay, overnight grown C. auris cells were collected and resuspended in PBS followed by treatment with ½MIC, MIC, and MFC values of the compound C5. H2O2 treated C. auris cells acts as the positive control and healthy cells serve as the negative control. The cells were rinsed three times with PBS and fixed in fixative solution (4% paraformaldehyde) for 30 minutes. The cells were put in 0.25% triton X-100 for 2 minutes for permeabilization. After permeabilization the C. auris cells were washed with PBS and incubated with Click-iT Plus TUNEL-assay kit (Thermo Fisher Scientific’s) for 30 minutes in the dark. Additionally, Candida cells were stained with 50μl of Hoechst 33342 (1X) dye (Thermo-Fisher Scientific USA) and left-over to sit for 15 minutes in the dark at room temperature, and cell were washed two time with PBS, before going to observe under the Confocal Microscopy (Zeiss Laser-Scanning Confocal- Microscope (LSM) 780 Jena, Germany). The excitation-wavelength of 495nm and an emis- sion-wavelength of 519nm for the Click-iT Plus TUNEL assay dye and for Hoechst-33342 dye had an excitation wavelength of 350nm and an emission wavelength of 461nm was used. Cell cycle arrest. The cell cycle arrest is a point in the cell cycle at which cells no longer par- ticipate in the processes of cell duplication and division. The effect of C5 on the cell cycle was determined by allowing yeast cells to grow for 8 hours, harvested at 4000g followed by suspension of 0.5 McFarland cells in SD broth. Different concentrations of C5 (½ MIC, MIC, and 2MIC) was added to different tubes followed by incubation for 4 hours with shaking. After, incubation the protoplasts were prepared according to protocol as above describe. The fixed cells were then incu- bated with the MuseTM Cell Cycle reagents, which contains ribonuclease and propidium iodide for 30 minutes and cell cycle analysis was determined using Muse1 Cell Analyzer. The percent- age of dead and live cells in each phase of the cell cycle was then visualised. Cytotoxicity of C5 compound. The cytotoxic potential of C5 was assessed using horse red blood cells (purchased from National Health Laboratory Service, Johannesburg, South Africa) following a previously described method. [18]. Briefly, 50ml of sterile falcon tubes con- taining 10 mL of horse blood were spun for 10 minutes at 2000rpm. The resultant red blood cells (RBSs) pellet was thoroughly washed three times with a cooled PBS solution before the cells were suspended once more in a cold PBS solution to produce a 10% RBC suspension. This RBC sample was once more diluted with a PBS solution ten times. As a result, a positive control and three different C5 compound concentrations (½MIC, MIC, and MFC) were added to the resulting 1mL RBC suspension. The treated RBC suspension was then maintained at room temperature for 1 hour and then centrifuged for 10 minutes at 2000rpm. 200μL super- natant was aliquoted in flatbottom 96-well plate (Thermo Fisher Scientific, Germany), and absorbance was measured at 450nm with SpectraMax iD3 multi-mode microplate reader. In this experiment Triton X 100 (1%) was kept as positive control and whereas, the fresh PBS as negative control. According to Lone et al., the percent hemolysis was calculated [18]. Results and discussion Chemistry Eugenol (EUG) or 4-allyl-2-methoxyphenol is a naturally occurring compound that is known to show antifungal properties [21]. Eugenol has been known to scavenge free radicals, inhibit PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 5 / 15 PLOS ONEEugenol derivatives as antifungal agents against Candida auris Fig 1. Scheme 1: Structures of derivatives C1-C6. https://doi.org/10.1371/journal.pone.0285473.g001 the generation of reactive oxygen species and increase cyto-antioxidant potential. Besides it is also known to increase H2O2 and Ca2+ concentrations in the cytoplasm, which links the anti- fungal activity of this compound to plasma membrane damaging and destabilizing properties [22]. Despite having potent antifungal properties, eugenol has also been found to show hepato- toxicity, contact stomatitis, and allergic cheilitis besides other complications. It also displays low chemical stability and is sensitive to oxidation and various chemical interactions [13]. To mitigate these side effects and to improve the biological profile of this molecules, various mod- ifications of the structure of this molecule have been carried out [13,23]. In a previous study, it was reported that eugenol tosylate and its congeners, prepared by the functionalization of the hydroxyl group of eugenol with different sulfonyl chlorides under basic conditions, showed promising antifungal activities compared to eugenol. In this study, more derivatives were syn- thesized and screened to find a suitable molecule with desired antifungal activity as illustrated in Scheme 1 (Fig 1). Elemental analysis, FTIR, 1HNMR, 13CNMR, and MS-ESI+ spectrum studies were used to determine the structure of the compounds. The synthesis of eugenol-tosy- late and its congeners C1-C6 was demonstrated using FTIR spectra. The presence of bands around 1156–1158 cm-1 and 1385–1387 cm-1 corresponding to the sulfonyl group of the respective derivatives and the absence of any free or H-bonded band at/or about 3200–3700 cm-1 corresponding to -OH provides strong proof for the formation of the desired com- pounds. The absence of any signal for the OH proton in 1H and 13CNMR, as well as the emer- gence of distinctive peaks at predicted chemical shifts and integral values for phenyl ring and allyl protons, give strong evidence for the synthesis of the C1-C6 derivatives. All of the deriva- tives (C1-C6) had a [M+H]+ peak in their mass spectra, which corresponded to the chemical formula of the compounds, further verifying their synthesis. The molecular ion peaks in some derivatives were found as [M+Na+]+ (Metal adduct ions). The physical and spectral data are presented in the experimental section. Antifungal activity C. auris is a public health concern due to its resistance and proclivity to create nosocomial epi- demics. Fungal infections caused by this pathogen have high morbidity and mortality rates. C. auris, despite being the youngest Candida species, is resistant to all major antifungal drug clas- ses [24]. Therefore, there is an inevitable need of developing new antifungal drugs with novel approaches to curb the infections caused by multidrug resistant pathogens like C. auris. Euge- nol, a natural compound from clove essential oil, has been thoroughly studied for its antimi- crobial activities [25]. Derivatization of eugenol has also been reported to enhance its antimicrobial properties and safety usage [14]. Modifying natural compounds to improve their PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 6 / 15 PLOS ONEEugenol derivatives as antifungal agents against Candida auris efficacy, solubility, and safe use has been appreciated in the discovery of innovative medica- tions against a variety of infectious disorders, including candidiasis [26,27]. Susceptibility testing The in-vitro susceptibility of six newly synthesized eugenol derivatives (C1–C6) was assessed against 5 different C. auris strains by measuring minimum inhibitory concentrations (MIC) and minimum fungicidal concentrations (MFC) (Table 1). All the test compounds showed anti- fungal activity with varying MIC and MFC values against C. auris isolates with compound C5 being the most potent with MIC and MFC values of 0.98 and 1.95 μg/mL respectively. Regard- ing the selected C. auris strains, all the strains except C. auris 6057 were susceptible to ampho- tericin B (AmB) which served as a positive control in this study. Interestingly C5 showed equally potent antifungal activity against the amphotericin B susceptible and resistant isolates. Based on MIC and MFC results, C5 showed considerably high anticandidal activity out of the six compounds examined and resistant isolate C. auris 6057 carried out for further study. Cell viability Cell count and viability assay was performed to further demonstrate the susceptibility of C. auris 6057 cells against the most active compound C5. Fig 2 depicts the cell viability profile and population profile of C. auris before and after treatment with ½MIC, MIC, and 2MIC con- centrations of C5 compound. The percentage (%) of cell viability at ½MIC, MIC, and 2MIC of C5 compound was 46.5%, 34.5%, and 19.9%, respectively (Fig 2). These findings established that the test compound C5 suppresses the growth and viability of C. auris in a concentration dependent manner and thereby validating the susceptibility results. As expected, negative con- trol contained 86.8% live cells, whereas the positive control contained only 23.0% live cells. Cell cycle arrest Following the susceptibility assays, effect of the C5 compound on the cell cycle of C. auris 6057 cells was determined using MUSE cell analyzer. The percentage of cells distributed among the Table 1. Minimum inhibitory concentrations (MIC) and minimum fungicidal concentrations (MFC) of eugenol tosylate congeners (C1-C6) against different C. auris isolates. C1 C2 C3 C4 C5 C6 AmB Compounds MIC (μg/mL) MFC (μg/mL) MIC (μg/mL) MFC (μg/mL) MIC (μg/mL) MFC (μg/mL) MIC (μg/mL) MFC (μg/mL) MIC (μg/mL) MFC (μg/mL) MIC (μg/mL) MFC (μg/mL) MIC (μg/mL) MFC (μg/mL) C. auris 6005 125 C. auris 6015 125 C. auris 6057 >125 C. auris 6059 125 C. auris 6065 125 >125 31.25 125 3.91 7.81 31.25 62.5 0.98 1.8 125 >125 1 2 >125 15.62 62.5 3.91 7.81 15.62 62.5 0.98 1.8 62.5 >125 0.25 0.5 >125 31.25 62.5 3.91 7.81 31.25 125 0.98 1.8 62.5 >125 4 8 >125 15.62 31.5 3.91 7.81 31.25 31.5 0.98 1.8 125 >125 0.5 1 >125 31.25 62.5 3.91 7.81 31.25 62.5 0.98 1.8 62.5 >125 1 2 amphotericin B (AmB)*. https://doi.org/10.1371/journal.pone.0285473.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 7 / 15 PLOS ONEEugenol derivatives as antifungal agents against Candida auris Fig 2. Cell viability of C. auris 6057. Cell viability of C. auris 6057 cells was recorded when cells were exposed to ½MIC, MIC and 2MIC of C5 compound. Cells without any exposure and with 10 mM H2O2 exposed serve as negative and positive controls. https://doi.org/10.1371/journal.pone.0285473.g002 various stages of the cell cycle in the untreated cells represent the normally developing cells, while as in comparison cells stuck in one phase indicate cell cycle arrest. The cell cycle results of C. auris revealed that healthy cells were uniformly distributed with 56% in G0/G1 phase, 32% in S phase and 11% in G2/M phase. In contrast cells treated with C5 halted in G0/G1, and the cell growth was predominantly limited to G0/G1 phase 76.8%, 91.5%, and 94.0% after the exposer with ½MIC, MIC and 2MIC respectively, while as only 15.6%, 6.5% and 4.6% of cells were arrested in S phase when exposed with ½MIC, MIC and 2MIC respectively (Fig 3). In the positive control where cells were treated with H2O2, then most of the cells (92.3%) were seen arrested in G0/G1 phase and only 4.6% and 3% are arrested in S and G2/M phases respectively. Apoptotic studies Apoptosis is an evolutionarily conserved mechanism used in the regulation of growth and dif- ferentiation by all multicellular organisms [28]. Apoptosis and cell cycle are linked to each other and are important in cell survival, proliferation, and reproduction. Arrest of C. auris cells in G1 phase of cell cycle as seen above could be related to the apoptosis in these cells and thereby their inability to enter S phase. Therefore, the present study determined the effect of C-5 compound against different apoptotic markers in C. auris 6057. Mitochondrial membrane potential (MMP) (Δψm) The membrane potential in mitochondria (Δψm) is an indicator of mitochondrial energetic status. MMP (Δψm) is used to assess the activity of proton-pumps in mitochondria, electron transport systems (ETS) and the activation of mitochondrial permeability due to a variety of PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 8 / 15 PLOS ONEEugenol derivatives as antifungal agents against Candida auris Fig 3. C. auris cell cycle analysis. The figure shows the effect of compound C5 at ½MIC, MIC and 2MIC on the cycle in C. auris 6057. Positive controls were cells exposed to H2O2, and negative controls were healthy untreated C. auris cells. https://doi.org/10.1371/journal.pone.0285473.g003 causes. Loss of MMP is regarded as very crucial for the survival and death of cells and is neces- sary stage for the apoptotic cascade. Therefore, C-5 compound was studied for its effect on MMP in C. auris cells to determine its ability to cause apoptosis. The constant Δψm in living yeast cells allowed the JC-10 to dye to clump together which can be detected by red fluoresce. In contrast apoptotic cells show decreasing Δψm, which restricts the JC-10 dye to monomeric form and thereby resulting in green fluorescence. Δψm was measured by calculating the ratio of JC-10 aggregates to JC-10 monomers; a decrease in values when compared to the untreated control indicated Δψm deprotonation. The results showed that in comparison of untreated cells a significant rise in JC-10 monomer was observed, means fluorescence levels was detected, indicating mitochondrial membrane depolarization (Fig 4). The ratio was 2.32 in the untreated cells or the cells with intact mitochondrial membrane and 1.91 in the positive con- trol cells. In terms of C-5 compound exposure, on increasing the concentration of the com- pound (½MIC, MIC, and 2MIC) membrane potential decreases from 1.92, 1.49, and 1.42 respectively. These results suggested that the C-5 compound causes membrane disintegration PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 9 / 15 PLOS ONEEugenol derivatives as antifungal agents against Candida auris Fig 4. Mitochondrial depolarization in C. auris. Fluorescence ratio (Y mean/X mean) representing Δψm and thereby apoptosis. In comparison to the negative untreated cells, C. auris exposed cells with ½MIC, MIC and 2MIC of C-5 compound showed depolarization of the mitochondrial membrane. Positive control represents C. auris 6057 cells treated with H2O2. https://doi.org/10.1371/journal.pone.0285473.g004 in mitochondria by lowering the mitochondrial membrane potential in C. auris cells. Mito- chondrial membrane depolarization is caused by uncontrolled mitochondrial membrane pores, which induces movement and the activation of various pro-apoptotic proteins [29]. Reactive oxygen species (ROS) that are produced by mitochondria were also recognized to play role in causing apoptosis [30]. Cytochrome c oxidase activity Determination of cytochrome c oxidase activity is an established method for studying apopto- tic cell death. This quantitative method involves the measurement of the amount of cyto- chrome c oozing out from mitochondria to cytosol. This study examined the changes in cytochrome c levels in mitochondria and cytosol in untreated and treated cells with various concentrations of compound C-5. According to the observations, there was a considerable increase of cytochrome c level in cytosol and decrease in mitochondrial cytosol when cells were treated with C-5 in comparison to untreated cells (Fig 5). Untreated negative control cells had relative fluorescence values which were consider 1.0 for cytochrome c in both mitochon- dria and cytosol. In comparison, congener C-5 treatment caused significant ooze out of cyto- chrome c in C. auris cells at ½MIC, MIC, and 2MIC values, with relative fluorescence values of 0.8, 0.78 and 0.47 for mitochondrial and 1.15, 1.24 and 1.35 for cytosolic cytochrome c, corre- spondingly (Fig 5). The cytochrome c in positive control was 0.49 for mitochondrial cyto- chrome c and 1.43 for cytosolic cytochrome c. Cytochrome c is a component of the electron transport chain (ETC) that is loosely connected to the inner mitochondrial membrane. Cas- pases are activated by the release of cytochrome c from the mitochondria to the cytoplasm, which is a critical step in the induction of apoptotic cell death [18]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 10 / 15 PLOS ONEEugenol derivatives as antifungal agents against Candida auris Fig 5. Cytochrome c movement in C. auris. Movement of apoptotic factor cytochrome c from mitochondria to cytosol in C. auris 6057 in absence (negative control), H2O2 treated (positive control) and C-5 treated at ½MIC, MIC and 2MIC. At 550nm, cytochrome c level in the mitochondria (orange line) and the cytosol (blue line) were calculated. https://doi.org/10.1371/journal.pone.0285473.g005 DNA damage by TUNEL assay The phosphatidylserine externalization is another important parameter for apoptosis studies. The terminal deoxynucleotidyl transferase-mediated dUTP nick end labeling (TUNEL) test was used to investigate this hallmark of apoptosis in response to the active drugs. The TUNEL- assay is based on the incorporation of modified dUTP at the 30-OH ends of fragmented DNA. C. auris cells were additionally stained with a Hoechst 33342 dye for differentiation, as it pro- duces blue color for live cells. From the results, C. auris cells exposed to various concentrations (½MIC, MIC, and 2MIC) of the compound C-5, showed considerable increase in the amount of TUNEL positive nuclei (green color fluorescent spots) when compared to the unexposed cell population (Fig 6). The results are suggesting the dose dependent effect, with higher concentrations of the test compound C-5 by increasing the population and intensity of green fluorescence cells in TUNEL positive nuclei. Positive control (10 mM H2O2 treated) also showed enhanced green fluorescence, showing a higher population of late apoptotic cells, in comparison of negative control with more blue fluorescence live cell population. Cytotoxicity The cytotoxicity of C-5 at various doses ½MIC, MIC, and 2MIC was tested against horse blood cells (RBCs). We corelate the Triton X-100 as a positive control, which induces 100% RBSs lysis, test compound C-5 caused hemolysis 0.6%, 3.92%, and 7.3% at ½MIC, MIC, and 2MIC values (Fig 7). PBS was utilized as a negative control, and no RBCs were lysed. The results confirmed that the newly synthesized C-5 compound has lower cytotoxicity, implying that it is potentially safe to use for in-vivo studies and thus provide a potential candidate for antifungal drug development. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 11 / 15 PLOS ONEEugenol derivatives as antifungal agents against Candida auris Fig 6. TUNNEL assay representing apoptosis in C. auris. The confocal microscopy of C. auris 6057 cells treated with different concentrations (½MIC, MIC, and 2MIC) of C-5 compound. Hoechst 33342 dye (blue fluorescence) showing live cells, and Alexa Fluor 488 dye (green fluorescence) represent apoptotic cells. https://doi.org/10.1371/journal.pone.0285473.g006 Fig 7. Hemolytic activity of C-5. Hemolysis of horse red blood cells was done in presence of Triton-X (control) and different concentrations of C-5. https://doi.org/10.1371/journal.pone.0285473.g007 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 12 / 15 PLOS ONEEugenol derivatives as antifungal agents against Candida auris Conclusion Significant antifungal activity caused by apoptosis and cell cycle arrest in C. auris by one of the eugenol derivatives (C5) was observed in this study. This compound showed potent antifungal activity against the tested C. auris isolates, and therefore shows promise as a lead molecule that could be studied further. Furthermore, unlike eugenol this lead compound also displayed low toxicity against red blood cells urging its safe use for in vivo assays. To put together, these results are quite encouraging and therefore further detailed studies of the active compound against different fungal pathogens would reveal more about the potential of this compound as a lead molecule. Statistical analysis All experiments were performed independently in triplicates (n = 3), and data were presented as mean ± standard deviation (SD). The statistical analysis was performed using two-way ANOVA test by GraphPad Prism software, version 8.0.1. Supporting information S1 File. Supporting information contains 1H and 13CNMR data of the compounds C1-C6, and Scheme (S1 Scheme) for the synthesis of the compounds. (DOCX) Author Contributions Conceptualization: Mohmmad Younus Wani, Julitha Molepo, Aijaz Ahmad. Formal analysis: Vartika Srivastava, Mohmmad Younus Wani, Abdullah Saad Al-Bogami, Julitha Molepo, Aijaz Ahmad. Investigation: Mohmmad Younus Wani, Abdullah Saad Al-Bogami, Aijaz Ahmad. Methodology: Hammad Alam, Vartika Srivastava, Windy Sekgele, Mohmmad Younus Wani. Project administration: Vartika Srivastava. Resources: Aijaz Ahmad. Software: Hammad Alam, Mohmmad Younus Wani. Supervision: Abdullah Saad Al-Bogami, Julitha Molepo, Aijaz Ahmad. Validation: Windy Sekgele, Mohmmad Younus Wani, Aijaz Ahmad. Writing – original draft: Hammad Alam, Windy Sekgele, Mohmmad Younus Wani. Writing – review & editing: Vartika Srivastava, Mohmmad Younus Wani, Abdullah Saad Al- Bogami, Aijaz Ahmad. References 1. Satoh K., Makimura K., Hasumi Y., Nishiyama Y., Uchida K. et al. Candida auris sp. nov., a novel asco- mycetous yeast isolated from the external ear canal of an inpatient in a Japanese hospital. Microbiol. Immunol. 2009; 53(1), 41–44. https://doi.org/10.1111/j.1348-0421.2008.00083.x. 2. Lorenz A., Papon N. New tools for the new bug Candida auris. Trends Microbiol. 2022; 30(3), 203–205. https://doi.org/10.1016/j.tim.2022.01.010. 3. Rhodes J., Matthew C. F. "Global epidemiology of emerging Candida auris." Curr. Opin. Microbiol. 2019; 52, 84–89, https://doi.org/10.1016/j.mib.2019.05.008. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 13 / 15 PLOS ONEEugenol derivatives as antifungal agents against Candida auris 4. Du H., Bing J., Hu T., Ennis C. L., Nobile C. J., et al. Candida auris: Epidemiology, biology, antifungal resistance, and virulence. PLoS pathogens, 2020; 16(10), e1008921. https://doi.org/10.1371/journal. ppat.1008921 PMID: 33091071 5. Allert S., Schulz D., Ka¨ mmer P., Großmann P., Wolf T., Scha¨ uble S, et al. From environmental adapta- tion to host survival: Attributes that mediate pathogenicity of Candida auris. Virulence, 2022; 13(1), 191–214. https://doi.org/10.1080/21505594.2022.2026037 PMID: 35142597 6. WHO fungal priority pathogens list to guide research, development, and public health action’ was pub- lished by the World Health Organization on Tuesday 25 October 2022. https://www.who.int/ publications/i/item/9789240060241. 7. Epstein D. J., Seo S. K., Brown J. M., Papanicolaou G. A. Echinocandin prophylaxis in patients under- going haematopoietic cell transplantation and other treatments for haematological malignancies. J. Antimicrob. Chemotherapy. 2018; 73, i60–i72. https://doi.org/10.1093/jac/dkx450 PMID: 29304213 8. Caˆndido E. D. S., Affonseca F., Cardoso M. H., Franco O. L. Echinocandins as biotechnological tools for treating candida auris infections. J Fungi, 2020; 6(3), 185–195, https://doi.org/10.3390/jof6030185 PMID: 32971857 9. Al-Rashdi A., Al-Maani A., Al-Wahaibi A., Alqayoudhi A., Al-Jardani A., Al-Abri S. (2021). Characteris- tics, risk factors, and survival analysis of Candida auris cases: results of one-year national surveillance data from Oman. J Fungi, 2021; 7(1), 31–42. https://doi.org/10.3390/jof7010031. 10. Osei Sekyere J. Candida auris: A systematic review and meta-analysis of current updates on an emerg- ing multidrug-resistant pathogen. Microbiol open, 2018; 7(4), e00578. https://doi.org/10.1002/mbo3. 578. 11. Sikora A., Zahra F. “Candida auris,” in Stat Pearls, Stat Pearls Publishing, Treasure Island (FL), 2022. http://www.ncbi.nlm.nih.gov/books/NBK563297/. 12. El-Baky N. A., Amara A. A. A. F. Recent approaches towards control of fungal diseases in plants: An updated review. Journal of Fungi, 2021; 7(11), 900–916. https://doi.org/10.3390/jof7110900 PMID: 34829188 13. Ulanowska M., Olas B. Biological Properties, and prospects for the application of eugenol, A review. Int. J. Mol. Sci. 2021; 22, 3671–3688, https://doi.org/10.3390/ijms22073671 PMID: 33916044 14. Lone S. A., Ahmad A. Inhibitory effect of novel Eugenol Tosylate Congeners on pathogenicity of Can- dida albicans. BMC Complement Altern Med, 2020; 20, 1–14, https://doi.org/10.1186/s12906-020- 02929-0. 15. Ahmad A., Wani M. Y., Khan A., Manzoor N., Molepo J. Synergistic interactions of eugenol-tosylate and its congeners with fluconazole against Candida albicans. Plos one, 2015; 10(12), e0145053. https:// doi.org/10.1371/journal.pone.0145053 PMID: 26694966 16. Didehdar M., Chegini Z., Shariati A. Eugenol: A novel therapeutic agent for the inhibition of Candida species infection. Front. Pharmacol. 2022; 13–24, https://doi.org/10.3389%2Ffphar.2022.872127. 17. Kaufman T. S., The multiple faces of eugenol. A versatile starting material and building block for organic and bio-organic synthesis and a convenient precursor toward bio-based fine chemicals. J. Braz. Chem. Soc. 2015; 26, 1055–1085, https://doi.org/10.5935/0103-5053.20150086. 18. Lone S. A., Wani M. Y., Fru P., Ahmad A. Cellular apoptosis and necrosis as therapeutic targets for novel eugenol tosylate congeners against Candida albicans. Sci. Rep. 2020; 10(1), 1–15. https://doi. org/10.1038/s41598-020-58256-4. 19. Bertout S., Dunyach C., Drakulovski P., Reynes J., Mallie M. Comparison of the Sensititre YeastOne® dilution method with the Clinical Laboratory Standards Institute (CLSI) M27-A3 microbroth dilution refer- ence method for determining MIC of eight antifungal agents on 102 yeast strains. Pathol. Biol. 2011; 59 (1), 48–51. https://doi.org/10.1016/j.patbio.2010.07.020. 20. Yun DG, Lee DG. Silibinin triggers yeast apoptosis related to mitochondrial Ca2+ influx in Candida albi- cans. Int J Biochem Cell Biol 2016; 80:1–9. https://doi.org/10.1016/j.biocel.2016.09.008 PMID: 27639679 21. Nisar M. F., Khadim M., Rafiq M., Chen J., Yang Y., et al. Pharmacological properties and health bene- fits of eugenol: A comprehensive review. Oxidative Medicine and Cellular Longevity, 2021; (2021). https://doi.org/10.1155/2021/2497354 PMID: 34394824 22. Wang C., Zhang J., Chen H., Fan Y., & Shi Z. Antifungal activity of eugenol against Botrytis cinerea. Tropical Plant Pathology, 2010; 35, 137–143. https://doi.org/10.1590/S1982-56762010000300001. 23. Maximino S. C., Dutra J. A., Rodrigues R. P., Gonc¸alves R. C., Morais P. A., Ventura J. A.S Borges W. et al. Synthesis of Eugenol Derivatives and Evaluation of their Antifungal Activity Against Fusarium solani f. sp. piperis. Current Pharmaceutical Design, 2020; 26(14), 1532–1542. https://doi.org/10.2174/ 1381612826666200403120448. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 14 / 15 PLOS ONEEugenol derivatives as antifungal agents against Candida auris 24. Rybak J. M., Cuomo C. A., Rogers P. D. The molecular and genetic basis of antifungal resistance in the emerging fungal pathogen Candida auris. Curr. Opin. Microbiol. 2022; 70, 1–8, https://doi.org/10.1016/ j.mib.2022.102208. 25. Haro-Gonza´lez J. N., Castillo-Herrera G. A., Martı´nez-Vela´ zquez M., Espinosa-Andrews H. Clove essential oil (Syzygium aromaticum L. Myrtaceae): Extraction, chemical composition, food applications, and essential bioactivity for human health. Molecules, 2021; 26(21), 6387–6411. https://doi.org/10. 3390/molecules26216387. 26. Liu X., Ma Z., Zhang J., Yang L. Antifungal compounds against Candida infections from traditional Chi- nese medicine. Biomed Res. Int. 2017; 2017, https://doi.org/10.1155/2017/4614183. 27. Pan S. Y., Zhou S. F., Gao S. H., Yu Z. L., Zhang S. F., Tang M. K., et al. New perspectives on how to discover drugs from herbal medicines: CAM’s outstanding contribution to modern therapeutics eCAM. 2013; 2013, 1–26 https://doi.org/10.1155/2013/627375. 28. Brown D. A., Yang N., Ray S. D. “Apoptosis”. In: Encyclopedia of Toxicology, Elsevier. 2014; p. 287– 94. https://doi.org/10.1016/B978-0-12-386454-3.00242-6. 29. Kamli M. R., Srivastava V., Hajrah N. H., Sabir J. S., Hakeem K. R., Ahmad A., Malik M. A. Facile bio- fabrication of Ag-Cu-Co trimetallic nanoparticles and its fungicidal activity against Candida auris. J Fungi, 2021; 7(1), 62. https://doi.org/10.3390/jof7010062 PMID: 33477480 30. Sasidharan S., Nishanth K. S., Nair H. J.Ethanolic extract of Caesalpinia bonduc seeds triggers yeast metacaspase-dependent apoptotic pathway mediated by mitochondrial dysfunction through enhanced production of calcium and reactive oxygen species (ROS) in Candida albicans. Front. Cell. Infect. Micro- biol. 2022; 1–22, https://doi.org/10.3389/fcimb.2022.970688 PMID: 36093184 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285473 June 21, 2023 15 / 15 PLOS ONE
10.1371_journal.ppat.1011015	RESEARCH ARTICLE An intact S-layer is advantageous to Clostridioides difficile within the host Michael J. Ormsby1☯¤a, Filipa Vaz1☯¤b, Joseph A. Kirk2, Anna Barwinska-Sendra3, Jennifer C. Hallam1, Paola Lanzoni-Mangutchi3¤c, John Cole1, Roy R. Chaudhuri2, Paula 2, Gillian R DouceID S. Salgado3, Robert P. FaganID 1* 1 School of Infection and Immunity, College of Medical, Veterinary & Life Sciences, University of Glasgow, Scotland, United Kingdom, 2 Molecular Microbiology, School of Biosciences, University of Sheffield, England, United Kingdom, 3 Biosciences Institute, Faculty of Medical Sciences, Newcastle University, England, United Kingdom ☯ These authors contributed equally to this work. ¤a Current address: Current addresses: Biological and Environmental Sciences, Faculty of Natural Sciences, University of Stirling, Stirling, United Kingdom ¤b Current address: Department of Immunology, University of Oslo and Oslo University Hospital, Oslo, Norway ¤c Current address: Universite´ Grenoble Alpes, CNRS, CEA, IBS, Grenoble, France * E-mail: Gillian.Douce@glasgow.ac.uk a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Ormsby MJ, Vaz F, Kirk JA, Barwinska- Sendra A, Hallam JC, Lanzoni-Mangutchi P, et al. (2023) An intact S-layer is advantageous to Clostridioides difficile within the host. PLoS Pathog 19(6): e1011015. https://doi.org/10.1371/journal. ppat.1011015 Editor: Bruce A. McClane, University of Pittsburgh School of Medicine, UNITED STATES Received: November 21, 2022 Accepted: May 31, 2023 Published: June 29, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.ppat.1011015 Copyright: © 2023 Ormsby et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: With respect to data availability statement, please amend to X-ray structural data has been deposited in the PDB repository under PDB ID 8BBY. Both the Raw RNA Abstract Clostridioides difficile is responsible for substantial morbidity and mortality in antibiotically- treated, hospitalised, elderly patients, in which toxin production correlates with diarrhoeal disease. While the function of these toxins has been studied in detail, the contribution of other factors, including the paracrystalline surface layer (S-layer), to disease is less well understood. Here, we highlight the essentiality of the S-layer in vivo by reporting the recov- ery of S-layer variants, following infection with the S-layer-null strain, FM2.5. These variants carry either correction of the original point mutation, or sequence modifications which restored the reading frame, and translation of slpA. Selection of these variant clones was rapid in vivo, and independent of toxin production, with up to 90% of the recovered C. difficile population encoding modified slpA sequence within 24 h post infection. Two variants, subsequently named FM2.5varA and FM2.5varB, were selected for study in greater detail. Structural determination of SlpA from FM2.5varB indicated an alteration in the orientation of protein domains, resulting in a reorganisation of the lattice assembly, and changes in interacting interfaces, which might alter function. Interestingly, variant FM2.5varB displayed an attenuated, FM2.5-like phenotype in vivo compared to FM2.5varA, which caused disease severity more comparable to that of R20291. Comparative RNA sequencing (RNA-Seq) analysis of in vitro grown isolates revealed large changes in gene expression between R20291 and FM2.5. Downregulation of tcdA/tcdB and several genes associated with sporulation and cell wall integrity may account for the reported attenuated phenotype of FM2.5 in vivo. RNA-seq data correlated well with disease severity with the more virulent var- iant, FM2.5varA, showing s similar profile of gene expression to R20291 in vitro, while the attenuated FM2.5varB showed downregulation of many of the same virulence associated traits as FM2.5. Cumulatively, these data add to a growing body of evidence that the S-layer contributes to C. difficile pathogenesis and disease severity. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 1 / 26 PLOS PATHOGENSseq and whole genomic sequencing data have been deposited within Gene Expression Omnibus (GEO) repository and can be accessed using the reference ID GSE205747. https://www.ncbi.nlm. nih.gov/geo/query/acc.cgi?acc=GSE205747. All other relevant data are available within the manuscript and its Supporting information files. Funding: This work was supported by the Wellcome Trust, UK [204877/Z/16/Z) awarded to GRD, PSS, RPF. This grant funded research posts for FV., MJO., ABS., JAK., JH. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. C. difficile requires the S-layer in vivo Author summary The S-layer of C. difficile is a paracrystalline array that covers the outer surface of the bac- terial cell but its contribution to overall disease remains unclear. As previously described, spontaneous slpA-null mutant, FM2.5, with a point mutation in slpA offered an opportu- nity to study the role of the S-layer in disease. Here, we confirm that this strain is less viru- lent in vivo despite effectively colonising the host and producing toxin. We also show in vivo selection for sequence modifications that restore slpA translation and produce an S- layer. While such modifications do not affect the overall 3D structure of individual SlpA (sub)domains, they can lead to altered orientation of the structural domains and subse- quent S-layer assembly. Importantly, RNA-Seq analysis in vitro showed large differences in gene expression between FM2.5 and R20291. Detected differences in transcription of genes involved in toxin expression and sporulation suggests that the S-layer provides a selective survival advantage within the host, which contributes to disease severity. Introduction Clostridioides difficile is the most common cause of hospital acquired diarrhoea globally, with disease linked to disruption of the intestinal microbiota through antibiotic use [1]. The viru- lence of C. difficile has widely been attributed to the production of two toxins; toxins A (TcdA, enterotoxin) and B (TcdB, cytotoxin), responsible for cytoskeletal modifications, epithelial damage, inflammation, and fluid loss [2, 3]. A third toxin, the binary C. difficile toxin (CDT), expressed by only a subset of strains, has been linked to enhanced disease severity [3]. Conse- quently, C. difficile colitis has widely been considered as a toxin-mediated disease. However, the availability of tools to analyse gene expression and improved methods of mutagenesis [4], together with the availability of an accessible murine animal model [5, 6], have offered new opportunities to identify other traits, both bacterial and host-associated, that impact disease severity [7, 8]. The recent use of such approaches has provided clearer understanding of the metabolic flexibility of these organisms, the role of the microbiome in disease progression [8– 10] and established several bacterial factors that influence the host response [11–13]. Of partic- ular interest in this context, is the role of the S-layer in disease. This paracrystalline protein array is the outermost layer of the C. difficile cell envelope, with similar structures found in many bacteria and virtually all archaea [14]. The S-layer has been shown to perform multiple and vital roles including providing protec- tion from environmental factors such as variations in pH, mechanical and osmotic stresses [15–17]. In vivo, it is proposed to play a role in molecular sieving [18] and ion trapping, pro- tecting the organism from antimicrobial peptides and bacteriolytic enzymes produced in response to infection [19, 20]. The S-layer has also been shown to be a key target in bacterio- phage predation [20–22]. In C. difficile, the main component of the S-layer is SlpA, which is post-translationally cleaved by a cell wall protein (CWP), Cwp84, into two functional S-layer proteins (SLPs), SLPL and SLPH [14, 23]. The proteinaceous array is further decorated by other CWPs, which pro- vide additional functionality [14]. Assembly of the paracrystalline array relies on tiling of SLPH triangular prisms on the cell wall, interlocked by SLPL ridges facing the environment [24]. Exposure of SLPL to the environment is consistent with its high sequence variability observed between different C. difficile strains, with 13 different S-layer cassette types (SLCTs) identified PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 2 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo to date [20, 25]. Strikingly, the lattice is very compact compared to other studied S-layers, which have pores of between 30–100 Å compared with only ~10 Å in diameter in C. difficile [24]. This tight packing correlates well with the hypothesis that S-layer acts as a molecular sieve [18], as deletion of the most exposed regions of SLPL results in a strain with increased sensitive to lysozyme, in comparison to the parent strain, R20291 [24]. In C. difficile, the S-layer has also been implicated in host cell adhesion [26], biofilm forma- tion [23, 27, 28] and immunomodulation through cell signalling of the host response [29–31]. SlpA has been shown to induce innate and adaptive immune responses through activation of TLR4 [32]. However, the role of the S-layer in C. difficile pathogenesis and in immune evasion remains poorly understood. Previously, we reported the isolation and characterization of a spontaneous C. difficile strain lacking an S-layer, FM2.5 [20]. In studies using the Golden Syrian hamster as the infection model, FM2.5 caused no symptoms of disease, despite effectively colonising infected animals [20]. However, the acute sensitivity of hamsters to C. difficile toxins and lack of readily avail- able immunological tools limits their usefulness in studying the more nuanced facets of this infection. In contrast, mice are naturally less susceptible to CDI, requiring more extensive anti- biotic treatment to suppress the flora, and higher challenge doses to achieve colonisation [6, 33]. However, mice offer greater opportunities to determine the contributions of other viru- lence-associated traits on disease outcome, including long-term persistence associated with relapsing disease [34]. Here, we sought to elucidate the role of the S-layer as a major virulence determinant in a murine model of infection, to determine whether the loss of virulence observed in the ham- ster model is reciprocal in other hosts. Our results suggest that the S-layer offers a competi- tive colonisation advantage within the mouse intestine and is important for in vivo disease severity. Results The S-layer contributes to severe disease in the murine model of C. difficile In a murine model of infection, loss of body weight offers a strong correlative measure of C. difficile disease severity [6, 33, 35]. Infection of antibiotic pre-treated mice with strain R20291 resulted in significant weight loss of up to 15%, peaking between 24 and 48 h post-infection (hpi). In contrast, mice infected with the S-layer deficient derivative strain FM2.5 showed con- sistently and significantly less weight loss than R20291, peaking at around 6% at 48 hpi, which is not statistically significant different to that determined in the PBS control group (Fig 1a). It is worth noting that, despite this modest weight loss, animals infected with FM2.5 failed to return to their pre-infection weight, even after the R20291-infected mice had fully recovered (96 hpi, Fig 1a). Although measurement of total C. difficile in faecal material showed compara- ble levels of total numbers of bacteria shedding at 24 and 72 hpi (Fig 1b), the recovered num- ber of spores was significantly lower in animals infected with FM2.5 at 24 h. This suggests that while FM2.5 appears to be growing in the lumen, it is not sporulating as efficiently as R20291; as previously reported [20]. In addition, a significant reduction in both total and spore counts were also observed at 24 hpi in both caecal (Fig 1c) and colonic (S1 Fig) luminal contents from mice infected with FM2.5 compared to R20291, indicating that this strain was less prevalent. In contrast, by 48 hpi, comparable numbers of both FM2.5 and R20291 (total and spore counts) were recovered from both luminal contents. Equivalent recoveries were also observed in faecal material assessed at 72 hpi and in luminal contents at 96 hpi. It should be noted that acute disease, typified by diarrhoea at 48 hpi, made comparative analysis of faecal material unfeasible at this time point. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 3 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo Fig 1. SlpA deficient C. difficile FM2.5 causes less severe disease in a murine model of infection. Female C57/Bl6 mice were challenged with spores of R20291 (green) or FM2.5 (purple), or mock infected with sterile PBS (grey). (a) Weight loss was monitored every 24 h for four consecutive days following infection. Each point is the average of several replicate experiments (n>3), with at least 10 animals per time point. Infection with R20291 strain results in statistically significant weight loss at 24 and 48 hpi, compared to both FM2.5 and PBS, while no significant difference was observed between FM2.5 and PBS at these time points. Mice infected with FM2.5 showed a slightly, but statistically significant, lower weight at 96 hpi, compared to the PBS control group. (b) CFU ml-1 of total (no pattern) and spore recovery (pattern) from faecal material collected at 24 hpi (n = 8, R20291; n = 15, FM2.5) and 72 hpi (n = 7, R20291; n = 14, FM2.5). Variation in sample numbers reflects either difficulties associated with collection from infected mice due to diarrheal symptoms, or loss of individual animals due to disease severity. (c) CFU ml-1 of total (no pattern) and spore recovery (pattern) in caecal contents at 24, 48 and 96 hpi (n = 5 at each time point except R20291 at 24 hpi; n = 4). (d) Toxin activity of caecal content at 24, 48 and 96 hpi; through challenge of Vero cells in vitro (n = 5 at each time point). Results displayed indicate the reciprocal of lowest dilution at which toxin activity could be measured. (e) Histological scoring of sections of the caecum from mice challenged with PBS, R20291 or FM2.5. Results displayed are the mean ± SEM of sections from at four mice per group (indicated by different symbols) with three sections scored from each tissue. (f) Histopathological sections representing caecal (i, ii and iii) sections following challenge with PBS (i); R20291 (ii); or FM2.5 (iii). Scale bars represent 100 μm. Statistical tests were conducted using GraphPad Prism software v.12. Statistical significance is indicated: ns—not significant; p < 0.05; p < 0.01; and *p < 0.001. https://doi.org/10.1371/journal.ppat.1011015.g001 Interestingly, mice infected with R20291 that survived infection showed full recovery by 96 hpi, returning to pre-infection weights, equivalent to those of non-infected mice. In contrast, FM2.5-infected mice failed to return to their pre-infection weight even when animals were monitored for a further five days (9 days pi) despite animals remaining asymptomatic, with no evidence of loose faeces. Assessment of in vivo toxin production showed that less toxin was detected in mice infected with FM2.5 in both the caecum (Fig 1d) and in the colon (S1b Fig) at 24 hpi compared to PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 4 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo R20291. However, at 48 and 96 hpi, comparable levels of toxin were recovered from all infected animals. Histological examination at 48 hpi of caecal tissue from R20291 infected mice showed greater tissue damage and inflammation than those infected with FM2.5 or control mice chal- lenged with PBS. Cumulative scoring of caecal and colonic tissue histology are shown in Fig 1e and S1c Fig, indicating changes are significant in caecal and lower colon tissue, and show a non-significant trend in the upper colon. Typical images of tissue damage are also displayed in Fig 1f (caecum) and S1d Fig (lower colon). In vivo pressure drives selection for S-layer variants When grown on selective chromogenic agar (ChromID; BioMerieux), the morphology of R20291 presents with the typical ‘fried egg’ C. difficile colony, visible after approximately 16 h of incubation (Fig 2a). In contrast, FM2.5 produces smaller and smoother colonies, which take ~24 h to emerge (Fig 2b). Following infection of mice with R20291 and FM2.5, faecal material was recovered and plated daily. During examination of resultant colonies, it was noted that, while colonies from R20291 infected mice showed the expected morphology, material retrieved from FM2.5-infected mice showed a mixture of both large (FM2.5large) and small col- ony types (FM2.5small) (Fig 2b). Additionally, FM2.5large, were countable after 16 h incubation, while the expected FM2.5-like colonies, FM2.5small, were only observable from 24 h. Several colonies from both types, FM2.5small and FM2.5large, were streaked from the original plates and sub-cultured twice to ensure clonality. Colony morphology of individual clones remained con- sistent on subsequent subcultures, suggesting this was not a phase variable trait. Individual clones were then stored at -80˚C. Colonies of different sizes were observed in several independent animal experiments, from which several clones were recovered. Amplification of the slpA sequence from these clones revealed that the FM2.5large colonies contained modifications in the genomic sequence upstream of the FM2.5 mutation site (single nucleotide insertion, Fig 2c). The first variant identified FM2.5slpA246delT (subsequently referred to as FM2.5varA) showed a single nucleo- tide deletion (246delT), which restored the original reading frame and rescued translation of the full SlpA; modifying three amino acid residues in the translated protein. An insertion of five-nucleotides was observed in a second variant, FM2.5slpA249insCTTAG (subsequently referred to as FM2.5varB; 249-253insCTTAG), resulting in the modification of 13 amino acids within the mature protein (Fig 2c). Genomic sequencing of individuals clones of these variants confirmed that they were closely related to FM2.5 and were not unrelated strains carried as contaminants by the mice (S2 Fig). To confirm restoration of SlpA expression in these strains, low pH cell surface extracts of FM2.5varA and FM2.5varB were analysed by SDS-PAGE. This confirmed that both SLPH and SLPL proteins were present in R20291, absent in FM2.5 but restored in FM2.5varA and FM2.5varB (Fig 2d). The proteins were confirmed as SLPH and SLPL by western immunoblot analysis using anti-SLPH and anti-SLPL antibodies (Fig 2e and 2f). Interestingly, these and other variants were also identified in subsequent in vivo experi- ments, even when animals were infected with batches of independently prepared spores. This reproducible recovery of S-layer variants in vivo raised the possibility that low numbers of genetic variants, present within the FM2.5 population, were being amplified within the in vivo environment. To test this hypothesis, we undertook amplicon sequencing of slpA in a prepara- tion of FM2.5 spores used to infect mice. and from faecal material recovered from these ani- mals at 24, 48, 72 and 96 hpi (Fig 2g,). Analysis of this unbiased amplification of slpA amplicons from faeces from these mice, revealed a third variant, FM2.5slpA252delA (subse- quently named FM2.5varC), in which the original frameshift mutation was corrected by PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 5 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo Fig 2. Recovery of S-layer variants following in vivo challenge. Following challenge with spores of R20291 and FM2.5, faecal material was recovered and plated on C. difficile selective chromogenic agar (Biomerieux). (a) Colonies of R20291. A dashed red box indicates the area enlarged to show colony morphology in the panel below. (b) The two colony types of FM2.5. An equivalently sized dashed red boxed area has been enlarged and colonies from each plate are shown below. FM2.5 is indicative of the typical FM2.5-like, smaller colony morphology, while the FM2.5variant is representative of the larger colony type. (c) Sequencing of a region of slpA shows the nucleotide and protein sequence of SlpAR20291 (green); the nucleotide insertion (red) in SlpAFM2.5 (light purple), introducing a stop codon that results in premature translational termination, leading to truncation of the SlpA protein. A single nucleotide deletion (light blue) in the SlpAFM2.5 sequence, results in the modification of 3 amino acids PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 6 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo (shown in light blue) in SlpAvarA, (dark blue); a five-nucleotide insertion (light crimson) in SlpAFM2.5 sequence results in modification of 13 amino acids (shown in light crimson) in SlpAvarB (crimson). These modifications in the sequence of SlpAvarA and SlpAvarB, restore the reading frame of slpA, allowing transcription and translation of the entire protein. (d) SDS-PAGE analysis of surface layer proteins extracted by low pH preparation. Lane 1: MW Marker; Lane 2: R20291; Lane 3: FM2.5; Lane 4: FM2.5varA; Lane 5: FM2.5varB. (e) Western immunoblot analysis using an anti-SLPH antibody. (f) Western immunoblot analysis using an anti-SLPL antibody. Bands corresponding to SLPL and SLPH indicated with blue and gold arrowheads, respectively. (g) Relative proportion of SlpAFM2.5 and SlpAvarC sequences in samples analysed in both the spore preparations used for mouse inoculations, and in faecal samples recovered up to 10 days post infection. Raw images are provided in S3 Fig. https://doi.org/10.1371/journal.ppat.1011015.g002 deletion of the extra nucleotide (252delA). This variant was also identified, albeit at a low pro- portion of the population (< 5%, Fig 2g), in the spores used for inoculation of these animals. Neither of the previous variants FM2.5varA or FM2.5varB were detected in either the spore or faecal samples tested, although this does not exclude the possibility that these or others were present in this preparation but that detection was beyond the discretionary power of the sequencing data. Based on our overall observations, each spore preparation is likely to contain low numbers of different variants, clones of which outgrow and outcompete the mutant within the population in vivo. However, isolation of variants as early as 24 hpi (> 94% of the popula- tion) suggests that expression of an intact S-layer provides a competitive advantage in vivo for these clones which express an intact S-layer. Sequence modifications can affect SlpA structure and assembly To understand the effects of the detected variations in amino acid sequence on SlpA struc- ture and S-layer assembly, crystallisation of SlpAR20291, SlpAvarA and SlpAvarB was carried out. Although crystals were obtained for all strains, only SlpAvarB crystals were of sufficient quality for X-ray diffraction data collection and structural determination by molecular replacement, using previous SlpA structures as models. As we have so far been unable to obtain an experimental structure of the complete SlpAR20291, our previously described vari- ant—SlpARΔD2—(PDB ID: 7ACZ), which lacks the most exposed region of SLPL [24], was used in our comparisons as a model of the S-layer protein in R20291. SLPH and the interact- ing domains were easily traceable in the electron density but D1 was only partially built, whilst density for domain D2 was very poor and this region could not be traced in the final SlpAvarB model (Fig 3a, PDB ID: 8BBY and S1 Table). This implies that D2 is flexible and/or unstructured, while the structure of the core domains required for S-layer assembly—SLPH, and, to a lesser extent, D1 and LID/HID [24]—seems to be generally maintained. However, the relative orientation of these domains in the SlpAvarB molecule is altered (Fig 3a), with D1 and the interacting domains rotated towards the SLPH plane by ~30 ˚ (Fig 3a). The 13 altered residues in α2L in SlpAvarB result in disruption of the α-helix secondary structure and intro- duce disorder in the upstream loop that links the preceding β-strand (β3L) and α2L. It is worth noting that SLPH in R20291, which belongs to SLCT 4, has several insertions within the cell wall binding 2 (CWB2) sequence motifs that define CWPs in C. difficile, when com- pared to other SlpA types. These insertions could not be traced in our previous SlpARΔD2 model [24] but were traceable in SlpAvarB and result in several loops protruding above the SLPH plane, towards the environment, partially occluding the CWB2 motifs (Fig 3a, right). Together with the movement of the interacting domains and D1 towards the SLPH tiles, this creates a more compressed arrangement (~66 Å thickness compared to ~76 Å in SlpACD630, PDB ID: 7ACY, Fig 3b, bottom). In the previously described crystallographic models of SlpACD630 (PDB ID: 7ACY), SlpAR7404 (PDB ID: 7ACX) and SlpARΔD2 (PDB ID: 7ACZ), α2L was responsible for closing a PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 7 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo Fig 3. Structure of SlpAvarB shows a different assembly arrangement. (a) Structural model of SlpAvarB (SLPL—pale red, SLPH—slate blue, PDB ID: 8BBY), superimposed on the R20291-derived SlpARΔD2 model (SLPL—gold, SLPH— slate blue, semi-transparent), with the rotation angle of the D1 and LID/HID domains shown by an arrow. Three distinct structural features are observed: SLPH, LID/HID and D1. Cartoon representation of the SLPH/SLPL (H/L) complex, as seen from the environmental side (left) and side view (right). Sequence of α2L, with paler colours indicating differences, is shown schematically. (b) Cartoon representation of the H/L planar array (PDB ID 8BBY, interacting molecules coloured and viewed as in a). (c) 2D schematic of H/L complex crystal packing in SlpAvarB (top), SlpARΔD2 (centre) and SlpACD630 (bottom), indicating the interaction network linking a single H/L (slate blue/crimson or slate blue/gold) complex with neighbouring molecules in a planar arrangement generated by SLPH tiling. The missing D2 in the SlpAvarB model is represented as dashed lines. Notably, D1-D1 interactions seen in other models are missing in SlpAvarB and the SLPH tiles are shifted, with new HID-CWB23 interactions stabilising the lattice. Array is depicted as seen from the extracellular environment, with symbols representing key interaction types in the crystal lattice, detailed in S2 Table. https://doi.org/10.1371/journal.ppat.1011015.g003 gap between neighbouring SlpA molecules via D1-D1 interactions [24]. In the crystallographic model of the R20291-derived SlpARΔD2 variant, D1-D1 interactions are mediated by hydrogen bonds between S50L-S50L and Q70L-A49L from neighbouring molecules. In the SlpAvarB structure, disruption of α2L and reorientation of D1 and LID/HID relative to SLPH leads to changes in the interactions between neighbouring molecules and, PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 8 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo consequently, a rearrangement of the S-layer array. Strikingly, the D1-D1 interactions seen in previous models were not observed here, possibly due to the flexibility of α2L and preceding loop caused by the changes in the variant sequence leading to a different orientation of D1 domains. Unlike in the previously determined structures, neighbouring D1 domains in the SlpAvarB structure are too far apart to mediate contacts (> 12 Å). In the previous models, SLPH tiling creates two wide channels, which are stabilised by interactions mediated by the interact- ing domains and D1 [24]. A different mode of stabilising the SLPH tiling is observed in SlpA- varB, with the interacting domains now partially inserted in those cavities (Fig 3b and S4a Fig). A new interacting interface between HID from one molecule and CWB23 from a neighbouring molecule occludes these gaps and stabilises the S-layer lattice (Fig 3c and S4a Fig). This new arrangement of the crystal lattice is in line with our proposed assembly model, where the S- layer 2D array is maintained mostly by hydrogen bonds and salt bridges across surfaces with complementary charges [24], largely dependent on SLPH-SLPH interactions and stabilised by varying degree of interactions involving SLPL (Fig 3c and S2 Table). The structural model of SlpAvarB confirms that changes in SLPL can be accommodated with minor structural changes to the (sub)domains, by exploring flexible loops and hinges to provide a stable S-layer. As no crystal data was obtainable, we initially used AlphaFold 2 [36] to predict models for SlpAR2021 and SlpAvarA. However, in the absence of a specific template, the predicted models show considerable variations of the relative orientation of SLPL and SLPH or some of the pro- teins sub-domains. The observed orientations are not compatible with our knowledge that SLPH faces the cell wall and SLPL the environment [24, 37]. This misorientation was also observed in AlphaFold 2 predictions of our experimental structures of SlpACD630 and the SlpA- varB and therefore we did not proceed with AlphaFold 2 predicted models. Instead, we used the SWISS-MODEL server, either without a template input, or using SlpARΔD2 [24] or the SlpAvarB structure determined here. Depending on which template was used (automatically selected SlpAR7404, PDB ID: 7ACX; SlpARΔD2 or SlpAvarB), different predicted structures of SlpAvarA were obtained, varying mostly in the orientation of D1 and interacting domains relative to SLPH (S4b Fig). Interestingly, one common feature was that the changes resulting from the altered sequence seem to be accommodated not by altering the α-helix but by varying the length of the upstream loop that links the preceding β-strand (β3L) and α2L (S4b Fig). It is therefore unclear if SlpAvarA is more likely to adopt a R20291-like as observed in the SlpARΔD2 model or SlpAvarB-like S-layer assembly, as both can accommodate the modified sequence. In vivo S-layer selection is independent of toxin expression FM2.5 has previously been reported to show a delay in toxin production [20], consequently we chose to investigate whether recovery of an intact S-layer was a toxin-dependent, or toxin- independent event. To answer this question, mice were infected with FM2.5ΔPaLoc, in which the Pathogenicity Locus (PaLoc), encoding toxins A and B, had been deleted. These animals, in contrast to those infected with FM2.5, showed no weight loss over the 96 hours of infection (Fig 4a) and as expected, no toxin was observed in samples from the caecum or colon (S5a Fig). Total numbers of bacteria (S5b Fig) and spores (S5c Fig) recovered from the caecum and colon of these mice were comparable at 96 hpi to that observed in animals infected with FM2.5. Interestingly, S-layer variants were also recovered from these mice, with sequence modifica- tions in the same region of slpA as previously identified. This variant FM2.5ΔPalocslpA237- delTTAT (subsequently referred to as FM2.5ΔPalocvarD) had four nucleotide deletions (Fig 4b; A237delTTAT). These changes also restore an intact SlpA, indicating that any potential selec- tion advantage is independent of toxin production. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 9 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo Fig 4. In vivo challenge of mice with FM2.5ΔPaLoc. Female C57/Bl6 mice were challenged with spores of R20291 (green), FM2.5 (purple) and FM2.5ΔPaLoc (blue) or PBS (grey). (a) Weight loss was monitored every 24 h for four consecutive days following infection. Each point is the average change of weight, calculated from a minimum of 5 mice per group. (b) Comparison of the sequences of slpA shows the sequence for R20291 (green) and the reported insertion in FM2.5 (red) responsible for the truncation of the SlpA protein; deletion of four nucleotides (orange) leads to changes in four amino acids (light orange) in the sequence in FM2.5ΔPalocvarD (orange). For comparison, the intact sequence of slpA from R20291 is shown in green. Statistical tests were conducted using GraphPad Prism software v.12. Statistical significance is indicated: ns—not significant; p < 0.05; p < 0.01; and p < 0.001. https://doi.org/10.1371/journal.ppat.1011015.g004 The variant strains display differing levels of virulence in mice To assess whether recovery of SlpA by these variants correlated with rescued virulence, mice were infected with spore preparations of FM2.5varA and FM2.5varB, alongside R20291 and FM2.5 (Fig 5a). Interestingly, infection with FM2.5varA resulted in significant weight loss within the first 48 h of infection, similar to that observed in mice infected with R20291. Mice infected with FM2.5varB, in contrast, showed a limited pattern of weight loss, like those animals infected with FM2.5, which stabilized from 48 hpi. To determine whether these differences were associated with changes in toxin production, the variants were cultured in vitro for up to 72 h. Filtered spent growth medium removed from the cultures at 36 and 72 hpi were used to determine the production of toxin B. Both FM2.5varA and FM2.5varB produced levels of toxin comparable to R20291 at these time points (Fig 5b), with toxin-mediated damage to Vero cells resulting in cell rounding, cellular loss and reduced Fig 5. Functional analysis of FM2.5varA and FM2.5varB in vivo and in vitro. (a) Female C57/Bl6 mice were challenged with spores of R20291, FM2.5, FM2.5varA, FM2.5varB or mock infected with sterile PBS. Weight loss was monitored at the same timepoint each day for four consecutive days following infection. Each point shows the average weight change for a minimum of 5 animals per treatment. (b) In vitro toxin activity as measured through challenge of Vero cells. Samples were prepared by filtering supernatant following growth of C. difficile in vitro for 36 or 72 h; activity was measured by the challenge of Vero cells. Supernatants were harvested at the same phase of growth for each strain. Cells treated with PBS or with purified Toxin B were included as negative and positive controls respectively. OD600 represents the optical density of Giemsa stain incorporated and released from intact Vero cells, hence high OD represents limited toxicity. Results displayed are the mean ± SEM of at least three independent replicates. Statistical tests were conducted using GraphPad Prism software v.12. Statistical significance is indicated: ns—not significant; p < 0.05; p < 0.01; and p < 0.001. https://doi.org/10.1371/journal.ppat.1011015.g005 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 10 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo levels of staining with Giemsa. In agreement with previous reports [20], FM2.5 produced sig- nificantly less toxin than R20291 at 36 h, although toxin production levels were comparable in all strains by 72 h. Modification of the S-layer results in large changes in gene expression To gain a greater understanding of the differences in gene expression between R20291, FM2.5 and variant strains, comparative RNA-Seq analysis was conducted following in vitro growth. SlpA was the most highly expressed transcript in all strains (R20291, FM2.5, FM2.5varA and FM2.5varB) which is unsurprising given that the S-layer is comprised of approximately 600,000 copies of SlpA. In FM2.5, slpA is transcribed normally but is not translated to a full-length pro- tein due to the frame-shift early in the open reading frame. Consequently, as transcripts are generated to an equivalent level, this gene does not appear in our list of differentially expressed genes. In contrast, comparative analysis of transcripts generated by R20291 and FM2.5 revealed over 287 differentially expressed genes (DEGs) (Fig 6a), linked to alterations in metabolism, transport, membrane integrity and sporulation (Fig 6b). In contrast, less differ- ences were observed when R20291 was compared to FM2.5varA (44 DEGs), than FM2.5varB (185 DEGs), which showed similar numbers of DEGs to FM2.5. This correlates well with the observed phenotype of these strains within animals—FM2.5varA associated with wild type like disease and FM2.5varB with FM2.5-like attenuation. Analysis of these data suggest that differ- ences in observed disease severity could be linked to changes in transcription of several viru- lence-associated traits, including toxin A and B, and genes associated with sporulation (Fig 6c). While in the case of FM2.5varA, the recovery of an intact S-layer would appear to be sufficient to restore wild type gene transcription, only partial transcription profile restoration, including toxin expression, was observed in FM2.5varB. However, as toxin activity was observed at 36 h in culture, the alterations in transcription control seen in FM2.5 and FM2.5varB would appear to be limited to timing rather than absolute prevention of toxin production. Taken together, these data suggest that expression of the S-layer plays a key role in C. diffi- cile gene expression which contributes to disease severity within the host. Discussion The S-layer of C. difficile has long been considered as integral to its physiology and pathogene- sis, with several roles reported, including adherence to the epithelial barrier [26], immune cell signalling [32, 38], resistance to antimicrobial peptides [20] and sporulation efficiency [20]. Here, we describe the pathogenesis of the S-layer-null mutant FM2.5 within the mouse model of disease and report the recovery of toxin-independent, spontaneous S-layer variants in which SlpA expression is restored. This unexpected but reproducible phenomenon supports the growing evidence that this envelope structure plays a key role in adaptation and survival within the host. FM2.5, a strain originally selected through its resistance to the engineered R-type bacterio- cin Av-CD291.2, was previously reported to be attenuated in the Syrian hamster model of C. difficile [20]. These studies indicated that SlpA was essential for disease, with no diarrhoeal symptoms observed in infected animals, despite the recovery of FM2.5 from the caecum and colon of infected hamsters 14 days pi. While we confirmed that the attenuated phenotype was reproducible in mice, we also report the recovery of SlpA variant clones from infected animals. Interestingly, SlpA variants were not observed during hamster infections [20], despite using the same chromogenic agar for recovery of FM2.5 isolates from infected animals. Although we cannot say for certain that such variants did not exist within this population, the failure to observe such clear phenotypic changes suggests that differences in the local environmental PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 11 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo Fig 6. Global transcriptional differences between isolates of C. difficile following in vitro growth. Analysis of mRNA recovered from in vitro grown cultures of C. difficile isolates R20291, FM2.5, FM2.5varA and FM2.5varB. (a) Total number of differentially expressed genes (DEGs) between experimental groups are highlighted in turquoise (upregulated) and orange (downregulated). (b) DEGs from experimental comparison of R20291 and FM2.5 were categorised based on function. (c) Transcriptional differences in select genes of FM2.5, FM2.5varA and FM2.5varB relative to R20291. Lists of all identified differentially expressed genes are provided (S1 Spreadsheet). https://doi.org/10.1371/journal.ppat.1011015.g006 conditions within the hamster and mouse may influence the amplification and outgrowth of these S-layer variant strains. The observation that FM2.5 is acutely sensitive to LL-37 and lyso- zyme [20] and becomes resistant following S-layer restoration [24] further supports the prem- ise that generation of an intact S-layer provides a competitive advantage in vivo. Although SlpA variants of FM2.5 have not been observed in vitro during sequential growth and recovery, several factors may influence their presence, albeit at low numbers, within the inoculum used to infect mice. As an obligate anaerobe, the sensitivity of C. difficile vegetative cells to oxygen complicates quantification of dosing of animals during oral challenge. In con- trast, the preparation and use of spores correlates with the natural route of infection, and addi- tionally avoids co-administration of toxins expressed during in vitro growth and preparation of vegetative cells. However, as FM2.5 has a known reduced sporulation efficiency [20], it is possible that any variants expressing an intact S-layer may sporulate more efficiently and therefore represent a higher proportion of the inoculum used for infection. Sequence analysis PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 12 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo of slpA of several different batches of FM2.5 spores supported this hypothesis, with a small (< 5%) proportion of the population displaying variations upstream of the original slpA muta- tion. Subsequent selection and amplification of these variants in vivo highlights the competi- tive advantage offered by the S-layer in C. difficile intestinal survival. This also provides an explanation for the significantly lower recovery of FM2.5 in mice, at 24 hpi compared to R20291, which becomes equivalent at later time points; reflecting the time required for the var- iant population to multiply to equivalent numbers within the gut. As germination rates of R20291 and FM2.5 have been previously shown to be equivalent [20], it is unlikely that the dif- ference observed at 24 hpi reflects lower or less efficient rates of germination by FM2.5. Alter- natively, lower rates of FM2.5 at 24 hpi could reflect the dynamics between killing of susceptible SlpA-null clones by anti-microbial peptide activity and low numbers of variants within the gut at this time point. FM2.5 has previously been reported to delay toxin production [20] which, coupled with the apparent delayed growth in vivo, could account for the difference in weight loss between R20291 and FM2.5 infected mice after 24 hpi. Using our FM2.5ΔPaLoc strain, we were able to demonstrate that the weight loss and toxin production are linked. However, this raises the question as to why animals infected with FM2.5, that show high and equivalent levels of tissue colonisation by 48 hpi and which are producing high amounts of toxin, do not show equiva- lent levels of weight loss and tissue inflammation at this and subsequent time points. Several studies have shown that the S-layer is essential in immune activation [39, 40] driving the pro- duction of proinflammatory cytokines via TLR4/MyD88 dependent pathways and enhancing the toxin-activated inflammasome [32, 41, 42]. Together, this implies that the timing or spatial localisation of toxins and the S-layer relative to the epithelial barrier could be crucial to immune activation. This hypothesis is further supported by the observation that FM2.5varB, which showed an equivalent reduction in early (6 h of growth) toxin expression to FM2.5 in the RNA-Seq analysis, also displayed reduced disease severity in the mouse. Regulation of toxin production is highly complex with multiple mechanisms of regulation reported to influence expression in response either directly (CodY, CcpA) or indirectly via interactions with tcdR (CodY, RstA, CcpA) in response to various stimuli including nutrient availability and quorum sensing [43]. More recently, c-di-GMP signalling and phase variation via seven invertible switches have been implicated in modulation of expression of genes associ- ated with motility, toxin production and colony morphology [44, 45]. Currently, it is unclear what mechanisms are involved in the observed altered toxin transcripts expression in FM2.5 and FM2.5varB and reduced toxin production by FM2.5 during early growth in culture. How- ever, this does not appear to be linked to phase variation, as orientation of the flgB switch, which has been shown to influence toxin expression, does not correlate with virulence observed in vivo (S6 Fig). Toxin production was delayed rather than absent in FM2.5 and FM2.5varB, with transcription of tcdB at a lower level in these strains than in R20291 or FM2.5varA. However, equivalent functional activity to R20291 was observed when FM2.5varB was grown for 36 h and FM2.5 for 72 h in vitro. This delay rather than impairment of toxin production supports the hypothesis that disease severity might be linked to the timing and co- ordination of S-layer mediated immune signalling and toxin expression within the host. Mice infected with either FM2.5varA or FM2.5varB showed different disease severity, as indi- cated by differences in weight loss. The low virulence of an FM2.5varB infection may be, in part, explained by the structural differences. Indeed, structural analysis of SlpAvarB revealed a different packing of SlpA molecules in the array, with a rearrangement of the position of SLPH and both interacting domains (Fig 3). This suggests a considerable degree of adaptability of both SLPL, where the changes are located, and SLPH, to accommodate varying interactions between neighbouring molecules. The absence of density to model D2 in SlpAvarB further PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 13 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo illustrates that this domain is dispensable for S-layer assembly, as previously reported [24]. As S-layer assembly is maintained mostly by hydrogen bonds and salt bridges [24], rearrangement of the subdomains to create structures with complementary surface charges seems to enable the different assemblies observed so far. These changes in quaternary structure, with minor changes of secondary and tertiary structure of the subdomains, suggests that the ability to form a paracrystalline array is central to S-layer function and can be achieved in different arrangements. The S-layer must retain a certain degree of flexibility, not only to account for the cell pole curvature and allow cell division, but also for incorporation of minor cell wall pro- teins that enhance functionality. The presence of a more intricate network of interactions and more extensive interface areas between neighbouring molecules seen in the SlpAvarB structure, when compared to the R20291-related SlpARΔD2 structure (S2 Table), suggests a potentially less flexible paracrystalline array. This could also reduce the ability to incorporate specific functions of minor CWPs, which may help explain differences in disease patterns observed between SlpAvarB and the other variant strain, SlpAvarA. Further structural studies of SlpAvarA and SlpAR20291 as well as detailed analysis of S-layer assembly and composition, including the capacity to incorporate other minor cell wall proteins, will help elucidate the role of specific aspects of the S-layer. While identification of SlpA variants was unexpected and adds complexity to the interpre- tation of the data from the mouse disease model, the rapid recovery of these strains highlights the key contribution that the expression of an intact S-layer offers to C. difficile infection in vivo. This work supports previous observations that strains lacking the S-layer are less virulent in vivo, although it remains difficult to identify the specific contribution of the S-layer in the infection process. Instead, this work highlights the potential multifunctional contribution that the S-layer plays in disease as, despite the number of differentially expressed genes observed between R20291, FM2.5 and the variants in vitro, recovery of the intact S-layer was sufficient to restore virulence, at least in one variant. Importantly, isolation and characterisation of these variants, together with greater knowledge of gene regulation and metabolic pathways impacted, offers a new opportunity to better understand the structural and functional role of the S-layer in C. difficile pathogenesis. Materials and methods Bacterial strains and growth conditions The bacterial strains used in this study include C. difficile strain R20291, its derivative FM2.5 [20], FM2.5varA, FM2.5varB, FM2.5varC, FM2.5ΔPaLoc and FM2.5ΔPaLocvarD (this study). Strains were routinely grown under anaerobic conditions on Braziers cycloserine, cefoxitin egg yolk (CCEY) agar (Oxoid, UK); CHROMID C. difficile Chromogenic medium (BioMe´r- ieux); or in Tryptone yeast (TY) broth (Oxoid, UK). Generation of FM2.5ΔPaLoc Homologous recombination was used to generate a derivative of strain FM2.5 that lacked the entire pathogenicity locus (PaLoc). Briefly, 1.2 kb up and downstream of the PaLoc was ampli- fied by PCR using oligonucleotides RF920 and RF921, and RF922 and RF923 (S3 Table), respectively, and cloned by Gibson assembly into plasmid pJAK112 [46] that had been linear- ised by PCR using RF311 and RF312. The resulting plasmid, pJAK143, was then conjugated into C. difficile [47] and mutagenesis to knock out the PaLoc was carried out using standard allele exchange [4]. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 14 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo Murine model of infection All procedures were performed in strict accordance with the Animals (Scientific Procedures) Act 1986 with specific approval granted by the Home Office, UK (PPL 60/8797 and PPL PI440270). These applications were also considered and approved by the University of Glas- gow Animal Welfare Ethical Review Body (AWERB). Food and water were provided ad libi- tum and animals kept at a constant room temperature of 20–22 ˚C with a 12 h light/dark cycle. Groups of up to six C57/bl6 female mice aged 6–8 weeks supplied by Charles River (Edin- burgh) were used in each treatment group. All animals were screened for colonisation with C. difficile in advance of the start of each experiment by plating faeces on ChromID selective media (Biomeuriex). An antibiotic cocktail (kanamycin, 0.40 mg ml-1; metronidazole, 0.215 mg ml-1; colistin, 850 U ml-1; gentamicin, 0.035 mg ml-1 and vancomycin, 0.045 mg ml-1 [all Sigma Aldrich, UK]) was administered ad libitum in the drinking water as previously described [35] with clindamycin sulphate (150 mg Kg-1), administered by oral gavage follow- ing cessation of the antibiotic cocktail. Animals were each challenged with approximately 106 spores of C. difficile 72 h after clindamycin treatment. Mice were monitored closely post-infec- tion and weighed daily to determine the severity of the disease. Animals with a weight loss greater than 10% of pre-challenge weight were given soft food and were culled if weight loss reached 20%. C. difficile shedding and organ colonization Fresh faecal samples collected daily were weighed, serially diluted in phosphate buffered saline (PBS) and cultured on CCEY agar at 37 ˚C for 48 h. At the experimental endpoint, animals were culled, and the caecum and colon harvested. Enumeration of total counts and spore-spe- cific counts in lumen associated material were performed as previously described [35]. In brief, total viable counts were determined by plating serial dilutions on ChromID selective media. Spores were enumerated following standard heat treatment of the samples at 65 ˚C for 20 min [33]. Quantification of toxin expression Quantification of toxin activity was performed using monolayers of Vero cells (kidney epi- thelial cells) as described previously [48]. Briefly, toxin was recovered from the spent filtered TY medium used to support bacterial growth for 36–72 h. Spent medium was recovered at the same stage of the growth cycle and at the same OD600nm. In vivo toxin activity was mea- sured by filtering using a 0.2 μm filter, luminal content collected from the caecum and colon of infected mice. Luminal contents collected from the caecum and colon of uninfected mice served as negative controls. Samples for toxin measurement were tested by the addition of serial dilutions to confluent monolayers within 72 h of collection. Cells and toxin were co- cultured for 24 h before cells were washed with phosphate buffered saline (PBS), fixed with 5% formal saline (Fisher), and stained with Giemsa for 30 min, before thorough washing to remove excessive stain. For data presented in Fig 1d and S1b and S5c Figs, toxin activity was determined as the reciprocal of last dilution in which toxin activity was observed, i.e., show- ing cell destruction. In Fig 5, to quantify the toxin activity more precisely, excess stain was removed by washing with PBS before cells were permeabilised to release internalised stain using 200 μl 1% SDS. 100 μl of the supernatant was transferred to a fresh plate and the OD620nm values determined. In this context, higher values indicate Vero cells are intact and unaffected by the toxin, lower values indicate that toxin mediated damage prevents uptake and retention of the dye. Cells treated with PBS only or with 0.0005 μg ml-1 purified Toxin B (Sigma, UK) were included as negative and positive controls respectively. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 15 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo Histology and immunohistochemistry Tissue samples were harvested from the caecum and colon of antibiotically susceptible animals infected with either R20291 or FM2.5 at post-mortem, 48 hpi. These tissues were gently washed in sterile PBS and immediately fixed in 10% formalin. Embedded tissue sections were cut and stained with Hematoxylin and Eosin [35]. Blind histological scoring of tissue was per- formed on three independent sections of caecal and colonic tissue from each mouse. Each sec- tion was scored out of a total of 15, with a score of 0 indicating no change, 1 slight change, 2 moderate change and 3 significant change, for the following categories: epithelial damage, neu- trophil migration, haemorrhagic congestion, tissue oedema and crypt hyperplasia. Data pre- sented represents the mean scores calculated for four mice for caecal, and three mice for colon assessment for each treatment. slpA sequencing from isolated variant clones Individual clones of bacteria, recovered from faecal or tissue associated material that showed different morphology on ChromID plates, where subject to at least two rounds of clonal selec- tion. Genomic DNA was isolated from a 20 ml culture grown anaerobically for 18 h in tryptic soya broth (TSB). Bacterial cells were initially disrupted enzymatically by resuspending the pel- let in lysis buffer (20 mM Tris-Cl, pH8.0, 2 mM Na EDTA, 1.2% Triton X-100, lysozyme 200 mg ml-1), and incubated at 56 ˚C for 90 min. The DNA was recovered using the DNeasy Blood and Tissue Kit (Qiagen), following manufacturer’s instructions. A 478 bp fragment of slpA, centred on the FM2.5 point mutation [20], was amplified by using oligonucleotides RF110 and RF111 (S3 Table). The resulting fragments were subjected to Sanger sequencing and compared to wild type and FM2.5 sequences. Isolation and sequencing of slpA from faecal extracts Faecal samples were also collected for unbiased sequencing of slpA within the population, by directly extracting DNA from faecal samples using the FastDNA SPIN kit for soil (MP Bio- medicals). Briefly, approximately 200–600 mg of faeces sample were suspended in 978 μl sodium phosphate buffer with 122 μl MT buffer lysis solution. Samples were then homoge- nised in a FastPrep instrument using two 30 second pulses, at speed setting 6.5, in lysing matrix E. Samples were centrifuged for 10 min at 14,000 x g to remove debris. 250 μl protein precipitation solution was added to the lysate supernatant, and the precipitant formed removed by centrifugation at 14,000 x g for 5 min. DNA was then bound to a silica matrix, washed using the kit wash buffer, and eluted with water. For analysis of slpA sequences in spores used for challenge experiments in mice, samples were treated with 0.1% taurocholic acid for 60 min to induce germination and then boiled for 5 min in the presence of Chelex resin to release DNA. DNA extracted from either faeces or germinated spores was used as a template for PCR amplification of a 330 bp fragment of slpA using Phusion polymerase (NEB) and oligonucleotides RF2193 and RF2194 (S3 Table). Resulting DNA fragments were purified and sequenced using the amplicon-EZ service offered by GENEWIZ (Azenta Life Sciences). Extraction and western immunoblot analysis of S-layer and associated proteins Surface layer proteins were extracted using low pH glycine as previously described [49] and analysed by SDS-PAGE using standard methods [50] (Fig 2 and S3 Fig). Proteins were trans- ferred to nitrocellulose membranes via semi-dry transfer (Bio-Rad Trans Blot Turbo; 25 V, 30 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 16 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo min) for western immunoblot analysis. Transfer efficiency was confirmed by PonceauS stain- ing of membrane post transfer, and Coomassie staining of the polyacrylamide gel following transfer. Membranes were blocked for 1 h in Phosphate-buffered Saline containing 0.1% Tween20 (PBS-T) with 5% milk powder. Blots were subsequently incubated in primary anti- body (rabbit anti-SLPH raised against C. difficile 630 1:100,000 dilution; rabbit anti-SLPL raised against C. difficile R20291 1:200,000 dilution) in PBS-T containing 3% milk powder, for 1 h at room temperature. Membranes were washed thoroughly in PBS-T before incubation with sec- ondary antibodies (anti-rabbit horseradish peroxidase, Promega WB401B 1:2,500 dilution) for 1 h at room temperature. Blots were washed in PBS-T before detection by chemiluminescence (Bio-Rad). Molecular weight (MW) markers (Thermo Scientific 26616) were imaged (Bio-Rad ChemiDoc XRS+) simultaneously and overlaid onto the blots to aid visualisation. Original full blots are shown in S3 Fig. Whole genome sequence analysis Genomic DNA recovered from R20291, FM2.5, FM2.5varA, FM2.5varB were sequenced by MicrobesNG (Birmingham UK) using Illumina based technologies. The raw data was then aligned to the reference genome (NC_013316.1) using Bowtie2. Polymorphisms were called using FreeBayes under default settings and VCF file was filtered to include only mutations identified with a read depth of > 50 and in at least 50% of reads. To build the tree (S2 Fig), the filtered mutations across all strains were merged and converted into a bed file. Regions were expanded by 100 bp on each side and these were considered the mutable regions. The phyloge- netic tree was constructed using PhyloTree (https://github.com/samandmac/PhyloTree), which uses BLAST and PHYML to generate a tree in Newick format, and subsequently an R script which allows visualisation of the tree using the ggtree package. The raw sequence data has been deposited to the Gene Expression Omnibus; reference ID GSE205747. Note: file name Rv9 contains the genomic data for FM2.5varA, and RV189 for FM2.5varB respectively. Protein purification and X-ray crystallography C. difficile variant strains were cultured in 400 ml of TYG broth for 16 h. Cultures were then centrifuged at room temperature at 4,696 x g and resulting pellets were washed with 40 ml of 0.01 M HEPES pH 7.4 and 0.15 M sodium chloride (HBS) buffer. S-layer extraction was per- formed by resuspending the washed pellet in 4 ml of 0.2 M glycine-HCl pH 2.2 and centrifuga- tion for 5 min at 21,100 x g. Collected supernatant was then neutralized with 2 M Tris-base. S- layer extract was filtered and resolved onto a Superdex 200 26/600 column using an A¨ KTA Pure FPLC system (Cytiva) in 50 mM Tris-HCl pH 7.5, 150 mM NaCl buffer. Purified SlpAvarB at 10 mg ml-1 was subjected to crystallization using a Mosquito liquid handling robot (TTP Labtech), with the sitting drop vapor-diffusion method, at 20 ˚C. Crystals were obtained in 0.03 M magnesium chloride hexahydrate; 0.03 M calcium chloride dihydrate, 0.12 M ethyleneglycol, 0.05 M Tris (base); 0.05 M bicine pH 8.5, 20% v/v glycerol; 10% w/v PEG 4,000. Data were collected on the I24 (λ = 0.71 Å) beamline at the Diamond Light Source Synchro- tron (Didcot, UK; mx24948-136) at 100 K. Two datasets were collected from one crystal and initially processed by the automatic multi-crystal data-analysis software pipeline xia2.multiplex [51] within the Information System for Protein Crystallography Beamline (ISPyB), re-pro- cessed using Automatic Image Processing with Xia-2 (DIALS [52] and Aimless 3d [53]) and scaled with Aimless within ccp4.cloud of CCP4 [54] software suit. The initial model of the core SLPH was obtained by molecular replacement in Phaser [55], using an SlpAvarB model of CWB2 domains, derived from the SlpARΔD2 model (PDB ID: PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 17 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo 7ACZ) and calculated using SWISS-MODEL [56]. The generated solution model was then subjected to automatic model building with Modelcraft [57], followed by manual building with Coot [58] and refinement in Refmac5 [59]. Final models were obtained after iterative cycles of manual model building with Coot and refinement in phenix_refine [60]. Data collection and refinement statistics are summarized in S1 Table. PDBePISA [61] was used to investigate interdomain and protein-protein interfaces in the crystallographic lattice to identify interacting residues, which were confirmed by manual inspection within COOT. Structural representations were generated using PyMOL Molecular Graphics System (Schro¨dinger, LLC). Protein structure prediction Homology models for SlpAR20291 and SlpAvarA were generated by providing SWISS-MODEL webserver with the SlpAvarB (PDB ID: 8BBY) or SlpARΔD2 (PDB ID: 7ACZ) as user templates, as well as without a template. Structural alignments between predicted models and templates were performed using COOT [58]. As SlpARΔD2 lacks the D2 domain, predicted models based on this experimental model have a disordered D2 domain. Therefore, overall comparison of the three predicted models was based on models calculated in the default mode, which uses SlpAR7404 (PDB ID: 7ACX) as a template, while analysis of the modification-containing region in the D1 domain was done using SlpARΔD2 or SlpAvarB derived models. Recovery of mRNA for RNA sequence analysis RNA was recovered from C. difficile strains R20291, FM2.5, FM2.5varA and FM2.5varB which had been cultured in vitro in TY broth. Briefly, bacterial cells reaching an OD620nm = 0.6 were pelleted (5000 x g, 15 min) and immediately fixed in 1.5 ml RNA-protect (Qiagen) for 10 min before being processed using a PureLink RNA mini kit (Ambion) to extract total RNA. To ensure maximal lysis of bacteria and recovery of RNA, the bacterial pellet was additionally sub- ject to treatment with 100 ml lysozyme solution (10 mg ml-1 in 10 mM Tris-HCl, pH8.0, 0.1 mM EDTA), 0.5 ml 10% SDS solution and 350 ml of Lysis Buffer (Invitrogen PureLink RNA Mini Kit) containing 2-mercaptoethanol. Cells were then homogenised using MP Biomedical beads (0.1 mm) and bead beater (MP Biomedical FastPrep24) with cells subject to 2 cycles of 60 s beating, followed by incubation on ice for 2 min. Total RNA was then extracted using the standard PureLink RNA mini kit protocol, according to the manufacturer’s specifications. Genomic DNA was removed using a TURBO DNase kit (Ambion) and samples were tested for efficient removal of DNA by conventional PCR. Samples for RNA-Seq were prepared in triplicate on two separate occasions (6 samples for each) for all four bacterial strains. Illumina library preparation of mRNA samples for RNA-Seq was prepared using a TruSeq Stranded mRNA library prep kit (Illumina) according to the manufacturer’s instructions. Sequencing was performed on the Illumina NextSeq 500 platform (75 bp length; single-end). Library generation, optimisation of amplification and sequencing were performed at the Uni- versity of Glasgow Polyomics facility. Quality control of sequencing data was performed using FastQC (Babraham Bioinformatics) to assess the minimum Phred threshold of 20 and poten- tial data contamination. The raw data has been deposited to the Gene Expression Omnibus ref- erence ID GSE205747. Note: files with sample numbers 1–6 reflect samples generated and analysed independently for each strain; Rv9 was the original name of FM2.5varA, and RV189 for FM2.5varB, respectively. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 18 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo RNA analysis and identification of differentially expressed genes Raw RNA-Seq datasets were subject to the following pipeline. Firstly, fastQ files were assessed using FastP [62] and then were aligned to the C. difficile R20291 (accession number NC_013316) reference genome using STAR [63] (v2.6) with–quantMode GeneCounts–outFil- terMultimapNmax 1 and–outFilterMatchNmin 35. We used a Star index with a–sjdbOverhang of the maximum read length − 1. Next, read count files were merged and genes with mean of < 1 read per sample were excluded. Finally, the expression and differential expression values were generated using DESeq2 [64] (v1.24). For differential comparisons, we used an A versus B model with no additional covariates. All other parameters were left to default. The processed data was then visualised using Searchlight [65], specifying one differential expression workflow for each comparison, an absolute log2-fold cut-off of 1 and adjusted p of 0.05. All other parameters were left to default. Pathway analysis methods Functional and metabolic pathways were implied by interrogation of the WP numbers assigned using the C. difficile annotated genome (NC_013316.1) and the Uniprot or NCBI Blastp databases. Gene ontology (GO) was assigned based on Biological Process assignment within Uniprot. Comparative analysis of all differentially expressed genes are provided (S1 Spreadsheet). Analysis of phase variable switch orientation To analyse the orientation of the seven C. difficile R20291 phase variable switches, orientation specific qPCR was carried out as previously described [44]. All strains were grown on BHI agar and mixed populations were prepared by pooling 100–200 colonies of each followed by geno- mic DNA purification using standard phenol-chloroform extraction and isopropanol precipi- tation. qPCR was performed using SYBR Green JumpStart Taq following the manufacturer’s instructions with 100 ng gDNA and 100 nM of each oligonucleotide (S3 Table). Reactions were run in a Bio-Rad CFX Connect RT-PCR instrument using the following parameters: 95 ˚C for 3 min, followed by 40 cycles of 95 ˚C for 10 sec, 60 ˚C for 15 sec, 72 ˚C for 15 sec and 64 ˚C for 10 sec, followed by a melting curve: 95 ˚C for 10 sec, 55 ˚C for 5 sec and 95 ˚C for 5 sec. Prod- ucts were proportionally quantified using the rpoA gene as a reference and data presented as the percentage of the switch in the orientation of the published R20291 reference genome [66]. Statistical analysis Statistical analysis was carried out in GraphPad Prism v.12. Tests used included for weight loss a two-way ANOVA with Tukey post test; for bacterial counts and toxin measurements, a two way ANOVA; for tissue histology analysis an ordinary one-way ANOVA. Statistical signifi- cance is indicated: ns—not significant; p < 0.05; p < 0.01; and p < 0.001. Supporting information S1 Fig. Characterisation of slpA deficient C. difficile in the colon of infected mice. Female C57/Bl6 mice were challenged with spores of R20291 (green) or FM2.5 (purple), or mock infected with sterile PBS (grey). (a) CFU ml-1 of total (clear fill) bacterial recovery or spores (fill pattern) in colonic contents at 24, 48 and 96 hpi (n = 5 at each time point except R20291 at 24 hpi; n = 4). (b) Toxin activity within colonic content at 24, 48 and 96 hpi; through challenge of Vero cells in vitro (n = 5 at each time point). Results displayed indicate the reciprocal of low- est dilution at which toxin activity could be measured. (c) Histological scoring of sections of PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 19 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo the upper (clear fill) and lower (fill pattern) colon sections from mice challenged with R20291, FM2.5 and PBS treated animals. Results displayed are the mean ± SEM of assessment of at least three regions of tissue from at least three individual animals. (d) Histopathological sec- tions representing caecal (i, ii and iii) sections following challenge with PBS (i); R20291 (ii); or FM2.5 (iii). Scale bars represent 100 μm. Statistical tests were conducted using GraphPad Prism software v.12. Statistical significance is indicated: ns—not significant; p < 0.05; p < 0.01; and p < 0.001. (TIF) S2 Fig. Confirmation of the close relationship between R20291, FM2.5 and slpA variants, FM2.5varA and FM2.5varB using whole genome sequence analysis. Chromosomal DNA was recovered from each strain and the raw data was aligned to the reference genome (NC_013316.1). (a) The variants were identified as closely related to both FM2.5 and R20291 using whole genomic sequencing to generate phylogenetic trees based on identifi- cation of polymorphisms identified during comparative analysis of sequences. Mutations were only included with a read depth of > 50 and which was present in at least 50% of reads. (b) Evaluation of the slpA sequence from all four strains from genomic data con- firmed mutations were limited to the regions highlighted above. Mutations originally iden- tified by PCR analysis were confirmed by short read genomic sequencing of individual strains. Insertions of the previously identified additional A in FM2.5, FM2.5varA, FM2.5varB are highlighted within the purple box, deletion of T in FM2.5varA by the dark blue box and insertion of CTTAG in FM2.5varB in red. Grey indicates the depth of sequence read cover- age. (TIF) S3 Fig. Raw images of SDS-PAGE gels and western immunoblots shown in Fig 2d, 2e and 2f. (a) SDS-PAGE of S-layer extracts, non-boiled and boiled prior to resolving on a poly- acrylamide gel. Molecular weight marker (M, PageRuler Prestained Protein Ladder, Thermo- Scientific); R20291 (1); FM2.5 (2); FM2.5varA (3); FM2.5varB (4). For Fig 2d, the SDS-PAGE gel was cropped to show only heated samples. (b) To assess efficiency of electrotransfer, the nitrocellulose membrane was stained with PonceauS (Sigma) and (c) polyacrylamide gel post-transfer was stained with Coomassie. The Western immunoblot analysis was performed comparing (d) non-boiled and (e) boiled S-layer extracts. For Fig 2e, the blots were cropped to allow focus on the heated samples only. Throughout all images, the SLPL is indicated with blue arrowhead and SLPH is indicated with yellow arrowhead. The additional band, detected with anti-SLPH CD630 antibodies and highlighted with magenta arrowhead, corresponds to SLPH missing interacting domain HID [24]. The high molecular weight band detected with anti-SLPL R20291, indicated with green arrowhead, could correspond to an apparent SDS- resistant oligomers of SLPL or non-specific binding of the antibody to CwpV, a phase vari- able protein. (TIF) S4 Fig. Tiling of SlpAvarB and comparison with predicted models of SlpAvarA and SlpAR20291. (a) Tiling of SLPH CWB2 motifs is stabilised by interacting domains. Poisson- Boltzmann electrostatic potential calculated for SlpAvarB SLPH, represented as a charge distri- bution (positive—blue; negative—red) on the surface representation of SLPH array. Interacting surfaces between molecules 1–2, defined by pseudo-symmetry related CWB23-CWB21, and between molecules 1–3, defined symmetry-related CWB23 triangular prism faces, are labelled. The cavity between symmetry-related CWB21-CWB22 surfaces, (117 Å, green arrow, left) is partially occluded by SLPL D1 (crimson) and by the insertion of the LID/HID domains PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 20 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo (electrostatic potential surface representation), as shown on the right panel. A long cavity of ~83 Å at the CWB22 vertices represented by purple arrow (left) is also occluded by LID/HID domains and interacting SLPL molecules (bottom). Neighbouring CWB21-CWB23 triangular prism faces form the interaction surface, while neighbouring CWB23 vertices complete the SLPH tiling. Insertion of the interacting domains bridges neighbouring SLPH tiles. Interactions across the lattice are maintained via complementary charged interfaces. (b) Predicted struc- tural models for SlpAR20291 (SLPL—green, SLPH—grey) and SlpAvarA (SLPL—dark blue, SLPH—grey) show the same overall structure of both SLPs as the experimental structural model for SlpAvarB (SLPL—gold, SLPH—slate blue) and SlpARΔD2 (SLPL—gold, SLPH—grey). Left: Cartoon representation of the complex, as seen from the environmental side, with all models superimposed on SLPL from SlpAvarB. The different orientation of SLPL and interac- tion domains observed in SlpAvarB is evidenced by the change in position of SLPH in this struc- ture when compared to SlpARΔD2 and the predicted models, calculated based on SlpAR7404 (PDB ID: 7acx). Right: Zoom view comparing D1 region of SlpAvarB (crimson) with SlpAR20291 (top right, green) and SlpAvarA (bottom right, dark blue). Sequence of α2L in the different vari- ants, with paler colours indicating differences compared to R20291, is shown schematically and corresponding sidechains are represented as sticks in the cartoon representation. Carbon —coloured according to molecule as in (a), oxygen—red, nitrogen—dark blue. (TIFF) S5 Fig. In vivo challenge of mice infected with FM2.5ΔPaLoc. Female C57/Bl6 mice were challenged with spores of R20291 (green), FM2.5 (purple) or FM2.5ΔPaLoc (blue). (a) Toxin activity of caecal and colonic content was determined as the reciprocal of last dilution in which cytoplasmic changes to challenged were observed. Statistical significance is indicated: ns—not significant. (b) At the experimental endpoint (96 hpi), total C. difficile counts were enumerated by colony counts from the caecum and colon contents from individual animals (n = 5 per group). (c) Spores were obtained from caecum and colon material and enumerated by colony counts. Vegetative cells were killed by heating samples at 65 ˚C for 20 min prior to plating to ensure only spores remained. (TIF) S6 Fig. Orientations of the phase variable invertible switches in R20291, FM2.5, FM2.5varA and FM2.5varB. To determine if differences in pathogenesis could be linked to a dominance of switch orientations of the 7 known phase switches in C. difficile, their orientation in all four strains was evaluated using qPCR, using the method described in [44]. This shows that the dif- ferences in virulence observed between FM2.5varA and FM2.5varB does not appear to link to orientation of these switches. Mixed populations representing 100–200 independent colonies were pooled, genomic DNA was extracted and the orientations of the seven switches was determined using a previously described orientation-specific qPCR assay. The gene immedi- ately downstream of the switch and presumed or confirmed to be controlled by the switch, is indicated above each graph. The percentages of each switch found to be in the same orienta- tion as in the published R20291 reference genome are shown. Measurements were performed in duplicate with the mean indicated by the horizontal line. Although variation in orientation between strains is observed for the switches upstream of pdcC, flgB and cwpV these differences do not correlate with observed differences in virulence. (TIF) S1 Table. Data collection and refinement statistics. Data collection and refinement statistics for structure determination of SlpAvarB. Data collected at beamline I24, Diamond Light Source. Values in parenthesis correspond to the highest resolution shell. R-work = S \|\|Fobs\|—\|Fcalc\|\| / PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 21 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo S \|Fobs\| R-free = S \|\|Fobs\|—\|Fcalc\|\| / S \|Fobs\| for 5% reflections excluded from refinement I/ sigmaI = I / σ(I) CC1/2 = S (Fobs1 Fobs2) / ( (DOCX) (S \|Fobs1\|^2) (S \|Fobs2\|^2)). p p S2 Table. Interactions across S-layer tiling. List of interactions between neighbouring mole- cules within the S-layer tiling for SlpAvarB and previously determined SlpACD630 and SlpARΔD2. Interacting atoms and the corresponding residue, together with the bond distance and interac- tion type (H—hydrogen bond; S—salt bridge) are listed, as determined in PISA and verified in each model using Coot (see material and methods for details). Numbers in parenthesis refer to the number of the neighbouring molecule, as defined in Fig 3b. L—SLPL; H—SLPH. Interac- tions seen in SlpAvarB not previously described in other models are highlighted in crimson. (DOCX) S3 Table. Oligonucleotides used in this study. Oligonucleotide primers used for PCR and qPCR. Sequences in lowercase indicate homology regions for Gibson assembly. All oligonucle- otides were purchased from Eurofins Genomics Europe. (DOCX) S1 Spreadsheet. Pairwise comparative analysis of differentially expressed genes (DEG’s) between R20291 and FM2.5, FM2.5varA and FM2.5varB. For differential comparisons, we used an A (R20291) versus B (FM2.5, FM2.5varA, FM2.5varB) model with no additional covari- ates. Differential expression values were generated using DESeq2 [64] (v1.24) and the pro- cessed data visualised using Searchlight [65]. DEG’s were determined as those genes with an absolute log2-fold cut-off of 1 (positive values for relative upregulated genes and negative val- ues for relative downregulated genes) and -adjusted p value of 0.05. Functional and metabolic pathways were implied by interrogation of the WP numbers assigned using the C. difficile annotated genome (NC_013316.1). Data is also shown for comparisons between FM2.5 vs FM2.5varA, FM2.5 vs FM2.5varB and FM2.5varA vs FM2.5varB. (XLSX) Acknowledgments The authors would like to thank Dr Arnaud Basle´, facility manager of the Newcastle Structural Biology Lab, and beamline staff at I24, Diamond Light Source (BAG mx24948) for support on structural data collection and Dr Sam McAllister for help with phylogenetic tree analysis. The contents of this work are solely the responsibilities of the authors and do not reflect the official views of any of the funders, who had no role in study design, data collection, analysis, decision to publish, or preparation of the manuscript. Author Contributions Conceptualization: Michael J. Ormsby, Filipa Vaz, Joseph A. Kirk, Anna Barwinska-Sendra, Paola Lanzoni-Mangutchi, Paula S. Salgado, Robert P. Fagan, Gillian R Douce. Data curation: Michael J. Ormsby, Filipa Vaz, Joseph A. Kirk, Anna Barwinska-Sendra, Paola Lanzoni-Mangutchi, John Cole, Roy R. Chaudhuri, Paula S. Salgado. Formal analysis: Filipa Vaz, Joseph A. Kirk, Anna Barwinska-Sendra, Paola Lanzoni-Man- gutchi, John Cole, Roy R. Chaudhuri, Paula S. Salgado, Robert P. Fagan, Gillian R Douce. Funding acquisition: Paula S. Salgado, Robert P. Fagan, Gillian R Douce. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 22 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo Investigation: Michael J. Ormsby, Filipa Vaz, Joseph A. Kirk, Anna Barwinska-Sendra, Jenni- fer C. Hallam, Paola Lanzoni-Mangutchi, John Cole, Roy R. Chaudhuri, Paula S. Salgado, Robert P. Fagan, Gillian R Douce. Methodology: Michael J. Ormsby, Filipa Vaz, Joseph A. Kirk, Anna Barwinska-Sendra, Jenni- fer C. Hallam, Paola Lanzoni-Mangutchi, John Cole, Roy R. Chaudhuri, Paula S. Salgado, Robert P. Fagan, Gillian R Douce. Project administration: Paula S. Salgado, Robert P. Fagan, Gillian R Douce. Resources: Michael J. Ormsby, Filipa Vaz, Joseph A. Kirk, Anna Barwinska-Sendra, Jennifer C. Hallam, Paola Lanzoni-Mangutchi, Roy R. Chaudhuri, Paula S. Salgado, Robert P. Fagan, Gillian R Douce. Software: John Cole, Roy R. Chaudhuri, Paula S. Salgado, Robert P. Fagan, Gillian R Douce. Supervision: Paula S. Salgado, Robert P. Fagan, Gillian R Douce. Validation: Michael J. Ormsby, Filipa Vaz, Joseph A. Kirk, Anna Barwinska-Sendra, Jennifer C. Hallam, Paola Lanzoni-Mangutchi, John Cole, Roy R. Chaudhuri, Paula S. Salgado, Rob- ert P. Fagan, Gillian R Douce. Visualization: Michael J. Ormsby, Filipa Vaz, Joseph A. Kirk, Anna Barwinska-Sendra, Paola Lanzoni-Mangutchi, John Cole, Paula S. Salgado, Robert P. Fagan, Gillian R Douce. Writing – original draft: Michael J. Ormsby, Filipa Vaz, Joseph A. Kirk, Anna Barwinska- Sendra, Paola Lanzoni-Mangutchi, John Cole, Roy R. Chaudhuri, Paula S. Salgado, Robert P. Fagan, Gillian R Douce. Writing – review & editing: Michael J. Ormsby, Filipa Vaz, Joseph A. Kirk, Anna Barwinska- Sendra, Paola Lanzoni-Mangutchi, John Cole, Paula S. Salgado, Robert P. Fagan, Gillian R Douce. References 1. Smits WK, Lyras D, Lacy DB, Wilcox MH, Kuijper EJ. Clostridium difficile infection. Nat Rev Dis Primers. 2016; 2: 1–20. https://doi.org/10.1038/NRDP.2016.20 PMID: 27158839 2. Braun V, Hundsberger T, Leukel P, Sauerborn M, von Eichel-Streiber C. Definition of the single integra- tion site of the pathogenicity locus in Clostridium difficile. Gene. 1996; 181: 29–38. https://doi.org/10. 1016/S0378-1119(96)00398-8 PMID: 8973304 3. Chandrasekaran R, Lacy DB. The role of toxins in Clostridium difficile infection. FEMS Microbiology Reviews. Oxford University Press; 2017. pp. 723–750. https://doi.org/10.1093/femsre/fux048 PMID: 29048477 4. Cartman ST, Kelly ML, Heeg D, Heap JT, Minton NP. Precise manipulation of the Clostridium difficile chromosome reveals a lack of association between the tcdC genotype and toxin production. Appl Envi- ron Microbiol. 2012; 78: 4683–4690. https://doi.org/10.1128/AEM.00249-12 PMID: 22522680 5. Theriot CM, Koumpouras CC, Carlson PE, Bergin II, Aronoff DM, Young VB. Cefoperazone-treated mice as an experimental platform to assess differential virulence of Clostridium difficile strains. Gut Microbes. 2011; 2: 326–334. https://doi.org/10.4161/GMIC.19142 PMID: 22198617 6. Chen X, Katchar K, Goldsmith JD, Nanthakumar N, Cheknis A, Gerding DN, et al. A mouse model of Clostridium difficile-associated disease. Gastroenterology. 2008; 135: 1984–1992. https://doi.org/10. 1053/J.GASTRO.2008.09.002 PMID: 18848941 7. McDermott AJ, Falkowski NR, McDonald RA, Frank CR, Pandit CR, Young VB, et al. Role of interferon- γ and inflammatory monocytes in driving colonic inflammation during acute Clostridium difficile infection in mice. Immunology. 2017; 150: 468–477. https://doi.org/10.1111/IMM.12700 PMID: 27995603 8. Fletcher JR, Pike CM, Parsons RJ, Rivera AJ, Foley MH, McLaren MR, et al. Clostridioides difficile exploits toxin-mediated inflammation to alter the host nutritional landscape and exclude competitors from the gut microbiota. Nat Commun. 2021; 12. https://doi.org/10.1038/S41467-020-20746-4 PMID: 33469019 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 23 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo 9. Girinathan BP, DiBenedetto N, Worley JN, Peltier J, Arrieta-Ortiz ML, Immanuel SRC, et al. In vivo com- mensal control of Clostridioides difficile virulence. Cell Host Microbe. 2021; 29: 1693–1708.e7. https:// doi.org/10.1016/J.CHOM.2021.09.007 PMID: 34637781 10. Buffie CG, Bucci V, Stein RR, McKenney PT, Ling L, Gobourne A, et al. Precision microbiome reconsti- tution restores bile acid mediated resistance to Clostridium difficile. Nature. 2015; 517: 205–208. https:// doi.org/10.1038/NATURE13828 PMID: 25337874 11. Maldarelli GA, de Masi L, von Rosenvinge EC, Carter M, Donnenberg MS. Identification, immunogenic- ity, and cross-reactivity of type IV pilin and pilin-like proteins from Clostridium difficile. Pathog Dis. 2014; 71: 302–314. https://doi.org/10.1111/2049-632X.12137 PMID: 24550179 12. Batah J, Kobeissy H, Pham PTB, Denève-Larrazet C, Kuehne S, Collignon A, et al. Clostridium difficile flagella induce a pro-inflammatory response in intestinal epithelium of mice in cooperation with toxins. Sci Rep. 2017; 7. https://doi.org/10.1038/s41598-017-03621-z PMID: 28607468 13. Arato V, Gasperini G, Giusti F, Ferlenghi I, Scarselli M, Leuzzi R. Dual role of the colonization factor CD2831 in Clostridium difficile pathogenesis. Sci Rep. 2019; 9. https://doi.org/10.1038/s41598-019- 42000-8 PMID: 30944377 14. Fagan RP, Fairweather NF. Biogenesis and functions of bacterial S-layers. Nature Reviews Microbiol- ogy. 2014. pp. 211–222. https://doi.org/10.1038/nrmicro3213 PMID: 24509785 15. Engelhardt H, Peters J. Structural research on surface layers: A focus on stability, surface layer homol- ogy domains, and surface layer-cell wall interactions. J Struct Biol. 1998; 124: 276–302. https://doi.org/ 10.1006/jsbi.1998.4070 PMID: 10049812 16. Claus H, Aka E, Debaerdemaeker T, Evrard C, Declercq JP, Ko¨ nig H. Primary structure of selected archaeal mesophilic and extremely thermophilic outer surface layer proteins. Systematic and Applied Microbiology. 2002. pp. 3–12. https://doi.org/10.1078/0723-2020-00100 PMID: 12086185 17. Engelhardt H. Are S-layers exoskeletons? The basic function of protein surface layers revisited. Journal of Structural Biology. 2007. pp. 115–124. https://doi.org/10.1016/j.jsb.2007.08.003 PMID: 17889557 18. Sleytr UB, Beveridge TJ. Bacterial S-layers. Trends Microbiol. 1999; 7: 253–260. https://doi.org/10. 1016/s0966-842x(99)01513-9 PMID: 10366863 19. Lortal S, van Heijenoort J, Gruber K, Sleytr UB. S-layer of Lactobacillus helveticus ATCC 12046: Isola- tion, chemical characterization and re-formation after extraction with lithium chloride. J Gen Microbiol. 1992; 138: 611–618. https://doi.org/10.1099/00221287-138-3-611 20. Kirk JA, Gebhart D, Buckley AM, Lok S, Scholl D, Douce GR, et al. New class of precision antimicrobials redefines role of Clostridium difficile S-layer in virulence and viability. Sci Transl Med. 2017; 9. https:// doi.org/10.1126/scitranslmed.aah6813 PMID: 28878013 21. Callegari ML, Riboli B, Sanders JW, Cocconcelli PS, Kok J, Venema G, et al. The Slayer gene of Lacto- bacillus helveticus CNRZ 892: cloning, sequence and heterologous expression. Microbiology (N Y). 1998; 144: 719–726. https://doi.org/10.1099/00221287-144-3-719 22. Royer ALM, Umansky AA, Allen M-M, Garneau JR, Ospina-Bedoya M, Kirk JA, et al. Clostridioides diffi- cile S-Layer Protein A (SlpA) serves as a general phage receptor. Microbiol Spectr. 2023; 11:e03894– 22. https://doi.org/10.1128/spectrum.03894-22 23. Kirby JM, Ahern H, Roberts AK, Kumar V, Freeman Z, Acharya KR, et al. Cwp84, a surface-associated cysteine protease, plays a role in the maturation of the surface layer of Clostridium difficile. Journal of Biological Chemistry. 2009; 284: 34666–34673. https://doi.org/10.1074/jbc.M109.051177 PMID: 19808679 24. Lanzoni-Mangutchi P, Banerji O, Wilson J, Barwinska-Sendra A, Kirk JA, Vaz F, et al. Structure and assembly of the S-layer in C. difficile. Nature Communications 2022 13:1. 2022; 13: 1–13. https://doi. org/10.1038/s41467-022-28196-w PMID: 35217634 25. Dingle KE, Didelot X, Ansari MA, Eyre DW, Vaughan A, Griffiths D, et al. Recombinational switching of the Clostridium difficile S-layer and a novel glycosylation gene cluster revealed by large-scale whole-genome sequencing. J Infect Dis. 2013; 207: 675–686. https://doi.org/10.1093/infdis/jis734 PMID: 23204167 26. Merrigan MM, Venugopal A, Roxas JL, Anwar F, Mallozzi MJ, Roxas BAP, et al. Surface-layer protein A (SlpA) is a major contributor to host-cell adherence of Clostridium difficile. PLoS One. 2013; 8. https:// doi.org/10.1371/JOURNAL.PONE.0078404 PMID: 24265687 27. Dapa T, Leuzzi R, Ng YK, Baban ST, Adamo R, Kuehne SA, et al. Multiple factors modulate biofilm for- mation by the anaerobic pathogen Clostridium difficile. J Bacteriol. 2013; 195: 545–555. https://doi.org/ 10.1128/JB.01980-12 PMID: 23175653 28. Richards E, Bouche´ L, Panico M, Arbeloa A, Vinogradov E, Morris H, et al. The S-layer protein of a Clos- tridium difficile SLCT-11 strain displays a complex glycan required for normal cell growth and morphol- ogy. Journal of Biological Chemistry. 2018; 293: 18123–18137. https://doi.org/10.1074/jbc.RA118. 004530 PMID: 30275012 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 24 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo 29. Ausiello CM, Cerquetti M, Fedele G, Spensieri F, Palazzo R, Nasso M, et al. Surface layer proteins from Clostridium difficile induce inflammatory and regulatory cytokines in human monocytes and dendritic cells. Microbes Infect. 2006; 8: 2640–2646. https://doi.org/10.1016/j.micinf.2006.07.009 PMID: 16935543 30. Sakakibara J, Nagano K, Murakami Y, Higuchi N, Nakamura H, Shimozato K, et al. Loss of adherence ability to human gingival epithelial cells in S-layer protein-deficient mutants of Tannerella forsythensis. Microbiology (N Y). 2007; 153: 866–876. https://doi.org/10.1099/mic.0.29275-0 31. Sekot G, Posch G, Messner P, Matejka M, Rausch-Fan X, Andrukhov O, et al. Potential of the Tanner- ella forsythia S-layer to delay the immune response. J Dent Res. 2011; 90: 109–114. https://doi.org/10. 1177/0022034510384622 PMID: 20929722 32. Ryan A, Lynch M, Smith SM, Amu S, Nel HJ, McCoy CE, et al. A Role for TLR4 in Clostridium difficile Infection and the Recognition of Surface Layer Proteins. PLoS Pathog. 2011; 7: e1002076. https://doi. org/10.1371/JOURNAL.PPAT.1002076 PMID: 21738466 33. Winston JA, Thanissery R, Montgomery SA, Theriot CM. Cefoperazone-treated Mouse Model of Clini- cally-relevant Clostridium difficile Strain R20291. J Vis Exp. 2016; 2016: 54850. https://doi.org/10.3791/ 54850 PMID: 28060346 34. Best EL, Freeman J, Wilcox MH. Models for the study of Clostridium difficile infection. Gut Microbes. Gut Microbes; 2012. pp. 145–167. https://doi.org/10.4161/gmic.19526 PMID: 22555466 35. 36. Jukes CA, Ijaz UZ, Buckley A, Spencer J, Irvine J, Candlish D, et al. Bile salt metabolism is not the only factor contributing to Clostridioides (Clostridium) difficile disease severity in the murine model of dis- ease. Gut Microbes. 2020; 11: 481–496. https://doi.org/10.1080/19490976.2019.1678996 PMID: 31793403 Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly accurate protein struc- ture prediction with AlphaFold. Nature 2021 596:7873. 2021; 596: 583–589. https://doi.org/10.1038/ s41586-021-03819-2 PMID: 34265844 37. Willing SE, Candela T, Shaw HA, Seager Z, Mesnage S, Fagan RP, et al. Clostridium difficile surface proteins are anchored to the cell wall using CWB2 motifs that recognise the anionic polymer PSII. Mol Microbiol. 2015; 96. https://doi.org/10.1111/mmi.12958 PMID: 25649385 38. Chen Y, Huang K, Chen LK, Wu HY, Hsu CY, Tsai YS, et al. Membrane Cholesterol Is Crucial for Clos- tridium difficile Surface Layer Protein Binding and Triggering Inflammasome Activation. Front Immunol. 2020; 11: 1675. https://doi.org/10.3389/FIMMU.2020.01675 39. Jarchum I, Liu M, Shi C, Equinda M, Pamer EG. Critical role for myd88-Mediated Neutrophil recruitment during Clostridium difficile colitis. Infect Immun. 2012; 80: 2989–2996. https://doi.org/10.1128/IAI. 00448-12 PMID: 22689818 40. Mamareli P, Kruse F, Friedrich C, Smit N, Strowig T, Sparwasser T, et al. Epithelium-specific MyD88 signaling, but not DCs or macrophages, control acute intestinal infection with Clostridium difficile. Eur J Immunol. 2019; 49: 747–757. https://doi.org/10.1002/EJI.201848022 PMID: 30802297 41. Cowardin CA, Kuehne SA, Buonomo EL, Marie CS, Minton NP, Petri WA. Inflammasome activation contributes to interleukin-23 production in response to Clostridium difficile. mBio. 2015; 6. https://doi. org/10.1128/mBio.02386-14 PMID: 25626905 42. McDermott AJ, Falkowski NR, McDonald RA, Pandit CR, Young VB, Huffnagle GB. Interleukin-23 (IL- 23), independent of IL-17 and IL-22, drives neutrophil recruitment and innate inflammation during Clos- tridium difficile colitis in mice. Immunology. 2016; 147: 114–124. https://doi.org/10.1111/IMM.12545 PMID: 26455347 43. Majumdar A, Govind R. Regulation of Clostridioides difficile toxin production. Curr Opin Microbiol. 2022; 65: 95–100. https://doi.org/10.1016/J.MIB.2021.10.018 PMID: 34781095 44. Ruiz LMR, King KA, Agosto-Burgos C, Gamez IS, Gadda NC, Garrett EM, et al. Coordinated modula- tion of multiple processes through phase variation of a c-di- GMP phosphodiesterase in Clostridioides difficile. PLoS Pathog. 2022; 18. https://doi.org/10.1371/journal.ppat.1010677 PMID: 35789350 45. Anjuwon-Foster BR, Tamayo R. A genetic switch controls the production of flagella and toxins in Clos- tridium difficile. PLoS Genet. 2017; 13: e1006701. https://doi.org/10.1371/JOURNAL.PGEN.1006701 PMID: 28346491 46. Fuchs M, Lamm-Schmidt V, Sulzer J, Ponath F, Jenniches L, Kirk JA, et al. An RNA-centric global view of Clostridioides difficile reveals broad activity of Hfq in a clinically important gram-positive bacterium. Proc Natl Acad Sci U S A. 2021; 118. https://doi.org/10.1073/PNAS.2103579118 47. Kirk JA, Fagan RP. Heat shock increases conjugation efficiency in Clostridium difficile. Anaerobe. 2016; 42: 1–5. https://doi.org/10.1016/J.ANAEROBE.2016.06.009 PMID: 27377776 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 25 / 26 PLOS PATHOGENSC. difficile requires the S-layer in vivo 48. Buckley AM, Spencer J, Candlish D, Irvine JJ, Douce GR. Infection of hamsters with the UK Clostridium difficile ribotype 027 outbreak strain R20291. J Med Microbiol. 2011; 60: 1174–1180. https://doi.org/10. 1099/jmm.0.028514-0 49. 50. Fagan RP, Albesa-Jove´ D, Qazi O, Svergun DI, Brown KA, Fairweather NF. Structural insights into the molecular organization of the S-layer from Clostridium difficile. Mol Microbiol. 2009; 71: 1308–1322. https://doi.org/10.1111/j.1365-2958.2009.06603.x PMID: 19183279 Laemmli U K. Cleavage of structural proteins during the assembly of the head of bacteriophage T4. Nature. 1970; 227: 680–685. https://doi.org/10.1038/227680a0 PMID: 5432063 51. Gildea RJ, Beilsten-Edmands J, Axford D, Horrell S, Aller P, Sandy J, et al. xia2.multiplex: a multi-crys- tal data analysis pipeline. bioRxiv. 2022; 2022.01.17.476589. https://doi.org/10.1101/2022.01.17. 476589 52. Winter G, Waterman DG, Parkhurst JM, Brewster AS, Gildea RJ, Gerstel M, et al. DIALS: implementa- tion and evaluation of a new integration package. urn:issn:2059–7983. 2018; 74: 85–97. https://doi.org/ 10.1107/S2059798317017235 PMID: 29533234 53. Evans PR, Murshudov GN. How good are my data and what is the resolution? urn:issn:0907–4449. 2013; 69: 1204–1214. https://doi.org/10.1107/S0907444913000061 PMID: 23793146 54. Winn MD, Ballard CC, Cowtan KD, Dodson EJ, Emsley P, Evans PR, et al. Overview of the CCP4 suite and current developments. urn:issn:0907–4449. 2011; 67: 235–242. https://doi.org/10.1107/ S0907444910045749 PMID: 21460441 55. McCoy AJ, Grosse-Kunstleve RW, Adams PD, Winn MD, Storoni LC, Read RJ. Phaser crystallographic software. urn:issn:0021–8898. 2007; 40: 658–674. https://doi.org/10.1107/S0021889807021206 PMID: 19461840 56. Waterhouse A, Bertoni M, Bienert S, Studer G, Tauriello G, Gumienny R, et al. SWISS-MODEL: homol- ogy modelling of protein structures and complexes. Nucleic Acids Res. 2018; 46: W296–W303. https:// doi.org/10.1093/nar/gky427 PMID: 29788355 57. Cowtan K, Metcalfe S, Bond P. Shift-field refinement of macromolecular atomic models. Acta Crystal- logr D Struct Biol. 2020; 76: 1192–1200. https://doi.org/10.1107/S2059798320013170 PMID: 33263325 58. Emsley P, Cowtan K. Coot: model-building tools for molecular graphics. Acta Crystallogr D Biol Crystal- logr. 2004; 60: 2126–2132. https://doi.org/10.1107/S0907444904019158 PMID: 15572765 59. Murshudov GN, Skuba´ k P, Lebedev AA, Pannu NS, Steiner RA, Nicholls RA, et al. REFMAC5 for the refinement of macromolecular crystal structures. urn:issn:0907–4449. 2011; 67: 355–367. https://doi. org/10.1107/S0907444911001314 PMID: 21460454 60. Liebschner D, Afonine P v., Baker ML, Bunkoczi G, Chen VB, Croll TI, et al. Macromolecular structure determination using X-rays, neutrons and electrons: recent developments in Phenix. urn:issn:2059– 7983. 2019; 75: 861–877. https://doi.org/10.1107/S2059798319011471 PMID: 31588918 61. Krissinel E, Henrick K. Inference of macromolecular assemblies from crystalline state. J Mol Biol. 2007; 372: 774–797. https://doi.org/10.1016/j.jmb.2007.05.022 PMID: 17681537 62. Chen S, Zhou Y, Chen Y, Gu J. Fastp: An ultra-fast all-in-one FASTQ preprocessor. Bioinformatics. 2018; 34: i884–i890. https://doi.org/10.1093/bioinformatics/bty560 PMID: 30423086 63. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, et al. STAR: Ultrafast universal RNA- seq aligner. Bioinformatics. 2013; 29: 15–21. https://doi.org/10.1093/bioinformatics/bts635 PMID: 23104886 64. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014; 15. https://doi.org/10.1186/s13059-014-0550-8 PMID: 25516281 65. Cole JJ, Faydaci BA, McGuinness D, Shaw R, Maciewicz RA, Robertson NA, et al. Searchlight: auto- mated bulk RNA-seq exploration and visualisation using dynamically generated R scripts. BMC Bioin- formatics. 2021; 22. https://doi.org/10.1186/s12859-021-04321-2 PMID: 34412594 66. Stabler R, He M, Wren B. Comparative genome and phenotypic analysis of Clostridium difficile 027 strains provides insight into the evolution of a hypervirulent bacterium. Genome Biology. 2009. 10 (9) 1–15 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011015 June 29, 2023 26 / 26 PLOS PATHOGENS
10.1371_journal.ppat.1011346	RESEARCH ARTICLE A modified Agrobacterium-mediated transformation for two oomycete pathogens Luyao Wang1,2, Fei Zhao2, Haohao Liu2, Han Chen2, Fan Zhang2, Suhua Li1, Tongjun Sun1, Vladimir NekrasovID 3, Sanwen Huang1, Suomeng DongID 2 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Wang L, Zhao F, Liu H, Chen H, Zhang F, Li S, et al. (2023) A modified Agrobacterium- mediated transformation for two oomycete pathogens. PLoS Pathog 19(4): e1011346. https:// doi.org/10.1371/journal.ppat.1011346 Editor: Paul Birch, University of Dundee, UNITED KINGDOM Received: June 23, 2022 Accepted: April 6, 2023 Published: April 21, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.ppat.1011346 Copyright: © 2023 Wang et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. Funding: In this research, S.W. and S.D. received funding support from Guangdong Major Project of 1 Shenzhen Branch, Guangdong Laboratory of Lingnan Modern Agriculture, Genome Analysis Laboratory of the Ministry of Agriculture and Rural Affairs, Agricultural Genomics Institute at Shenzhen, Chinese Academy of Agricultural Sciences, Shenzhen, China, 2 Department of Plant Pathology and Key Laboratory of Integrated Management of Crop Disease and Pests (Ministry of Education), Nanjing Agricultural University, Nanjing, China, 3 Plant Sciences and the Bioeconomy, Rothamsted Research, Harpenden, United Kingdom * huangsanwen@caas.cn (SH); smdong@njau.edu.cn (SD) Abstract Oomycetes are a group of filamentous microorganisms that include some of the biggest threats to food security and natural ecosystems. However, much of the molecular basis of the pathogenesis and the development in these organisms remains to be learned, largely due to shortage of efficient genetic manipulation methods. In this study, we developed modi- fied transformation methods for two important oomycete species, Phytophthora infestans and Plasmopara viticola, that bring destructive damage in agricultural production. As part of the study, we established an improved Agrobacterium-mediated transformation (AMT) method by prokaryotic expression in Agrobacterium tumefaciens of AtVIP1 (VirE2-interact- ing protein 1), an Arabidopsis bZIP gene required for AMT but absent in oomycetes genomes. Using the new method, we achieved an increment in transformation efficiency in two P. infestans strains. We further obtained a positive GFP transformant of P. viticola using the modified AMT method. By combining this method with the CRISPR/Cas12a genome editing system, we successfully performed targeted mutagenesis and generated loss-of- function mutations in two P. infestans genes. We edited a MADS-box transcription factor- encoding gene and found that a homozygous mutation in MADS-box results in poor sporula- tion and significantly reduced virulence. Meanwhile, a single-copy avirulence effector- encoding gene Avr8 in P. infestans was targeted and the edited transformants were virulent on potato carrying the cognate resistance gene R8, suggesting that loss of Avr8 led to suc- cessful evasion of the host immune response by the pathogen. In summary, this study reports on a modified genetic transformation and genome editing system, providing a poten- tial tool for accelerating molecular genetic studies not only in oomycetes, but also other microorganisms. Author summary Oomycetes are a unique group of animal and plant pathogens that include some of the biggest threats to food security and natural ecosystems. However, much of the PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 1 / 27 PLOS PATHOGENSModification of AMT for two oomycete species Basic and Applied Basic Research (2021B0301030004); S.D. received funding support from National Natural Science Foundation of China (NSFC, 31721004) and China Agriculture Research System (CARS-09-P20); L.W was supported by NSFC (31900303) and the Agricultural Science and Technology Innovation Program (CAASZDRW202101); V.N. was supported by the Biotechnology and Biological Sciences Research Council (BBSRC) through the Designing Future Wheat (DFW) Institute Strategic Programme (grant number BB/P016855/1). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. pathogenesis and development in these organisms remains to be learned, largely due to shortage of efficient genetic manipulation methods for a long time. In this manuscript, we developed modified Agrobacterium-mediated genetic transformation strategies for Phy- tophthora infestans, a notorious oomycete species that caused Irish Famine in the 19th cen- tury, and Plasmopara viticola, the causal agent of grapevine downy mildew that is known as a highly destructive disease of grapevines in all grape-growing areas of the world. Using prokaryotic expression of a plant protein in A. tumefaciens, we achieved considerable increase in transformation rate in different P. infestans strains, and also acquired one posi- tive transformant of P. viticola, a oomycete species that is extremely hard-to-transform and cannot be grown as an axenic culture. Using our improved transformation protocol, combined with the CRISPR/Cas12a system, we performed genome editing and created loss-of-function alleles in P. infestans. In summary, our study reports on modified genetic transformation methods, for two important oomycete species, that have the potential to accelerate the molecular genetic study of many other microorganisms. Introduction Oomycetes are regarded as an important lineage of eukaryotic organisms that includes a con- siderable number of plant pathogens, which pose a threat to natural and arable ecosystems. Grapevine downy mildew, caused by Plasmopara viticola, is considered to be the most devas- tating disease of grapevines in climates with relatively warm and humid summers. P. viticola is an obligate biotrophic oomycete species that cannot be grown as an axenic culture and is very recalcitrant to genetic transformation as demonstrated in previous studies [1]. The genetic transformation obstacle in the case of P. viticola has severely hampered functional genomics research and studies on screening molecular drug targets. Thus, it is important to set up a workable transformation method for P. viticola and other hard-to-transform biotrophic microorganisms. Potato late blight is another one of the most well-known but not well-understood plant dis- eases in terms of molecular genetics of Phytophthora infestans, its causal pathogenic microor- ganism. P. infestans, the oomycete pathogen responsible for the devastating potato late blight, poses a worldwide threat [2], and plays an essential role in studying plant–microbe interac- tions, epidemiology, and field disease control [3]. Currently, reverse genetics studies in P. infes- tans are also hampered by inefficient genetic transformation methods. Up until now, functional genomic research in P. infestans has relied mostly on transient/stable gene overex- pression or target-gene silencing by RNAi [4, 5]. Although polyethylene glycol (PEG)-medi- ated transformation, Agrobacterium-mediated transformation (AMT), microprojectile bombardment and zoospore electroporation have been already set up in P. infestans for years, we need to be open to new methods as other transformation options are always encouraged [6–9]. Moreover, phenotypes of Phytophthora transformants often vary significantly between different pathogen lines, experiment operators, and individual experiments, possibly because of random gene integration sites and different silencing or overexpression qualities [10]. Despite of multiple attempts to apply CRISPR/Cas9 for the purpose of targeted mutagenesis in P. infestans, no genome editing events, generated using this system, have been reported so far [11]. In a recent study, the CRISPR/Cas12a (Cpf1) system was utilized to produce genome editing events in a P. infestans strain using the PEG-mediated protoplast transformation method [12]. To propel future research, it is of great interest to expand the range of available genetic transformation methods for genome editing in P. infestans. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 2 / 27 PLOS PATHOGENSModification of AMT for two oomycete species Agrobacterium tumefaciens, as a typical plant pathogenic bacterium, causes crown gall dis- ease in a wide range of plants [13], and was developed into an efficient genetic transformation tool since Binns and Thomashaw first demonstrated that A. tumefaciens can integrate exoge- nous gene segments into the plant nucleus [14]. Known as the first example of prokaryote to eukaryote horizontal gene transfer, the capacity of A. tumefaciens to transfer alien genes to host cells has become the basis of one of the most essential technologies in manifold research fields, and plant science in particular [15]. By now, transgenesis, achieved through A. tumefa- ciens-mediated transformation (AMT), has been established in many model plant species, including Arabidopsis, tobacco and rice [16–18]. In addition to plants, Agrobacterium is also able to infect a broad range of other non-plant hosts during the co-cultivation process, includ- ing microorganisms identified as plant pathogens, and AMT protocols have been developed as a transformation system for many fungi, including Aspergilus species, Beauveria species, Botry- tis cinerea, Candida species, Coccidiodes species, Colletotrichum species, Fusarium species, Tri- choderma species and Verticillium species [19–30]. Attempts to improve the efficiency of AMT have been undertaken since the method was first reported. However, in most cases, the efforts were focused on optimization of conditions, such as explant treatment, tissue culture medium composition, temperature or selectable marker, but not the core tool—A. tumefaciens [31]. One successful attempt to improve the plant transformation ability of A. tumefaciens involved introducing a compatible plasmid carrying a virG mutant (virGn45D), which constitutively induced vir gene expression during AMT [32]. The first AMT protocol for P. infestans was established in 2003 by Vijin and Govers; in this study, the A. tumefaciens LBA1100 strain was selected, and as many as 30 transformants per 107 zoospores could be obtained [7]. A recent study has also described an efficient AMT proto- col for another oomycete species, Phytophthora palmivora, which produces large amount of zoospores (2–5 × 106/mL) during in vitro preparation [33]. However, the amount of zoospores produced may vary dramatically in different Phytophthora strains, and some oomycete species produce hardly any zoospores under artificial conditions. Thus, the ability to produce suffi- cient amounts of zoospores under in vitro conditions is likely a serious limiting factor for applying zoospore-dependent transformation methods in oomycetes [34–36]. Previous studies illustrated that ectopic expression of several plant proteins significantly improved AMT rate in a range of plant species [37]. A set of proteins from the host are involved with Agrobacterium-mediated transformation of plant [37]. Overexpressing several of these host proteins, mostly from Arabidopsis, such as AtVIP1, AtRTNLB1, Ku80, histone H2A and SGA1, significantly increases AMT efficiency [37–42]. Arabidopsis VirE2-interact- ing-protein1 (AtVIP1) is known as an important plant protein that contributes to the AMT process [38, 43]. AtVIP1 has been proven to interact with Agrobacterium effector VirE2, and its ability to form homomultimeric protein complexes and interact with histone H2A in host cells is required for Agrobacterium-mediated pathogenicity or stable genetic transformation [44]. AtVIP1-overexpressing plant lines have shown significantly increased susceptibility to A. tumefaciens infection, and AtVIP1 seems to facilitate nuclear transport of VirE2 and the T-DNA complex [42]. It would be more practical if we could directly utilize AtVIP1 protein during the AMT process instead of constructing AtVIP1 transgenic material prior to transfor- mation of a gene of interest. Here we report on a modified AMT method utilizing AtVIP1 as an enhancer in the trans- formation of two important oomycete pathogens. In this study, an increment in the transfor- mation efficiency of Phytophthora infestans was achieved by prokaryotic expression of AtVIP1 in the A. tumefaciens EHA105 strain as compared to AMT without AtVIP1. We then extended application of this modified AMT procedure to Plasmopara viticola. By combining our modi- fied AMT method and the recently established Cas12a-based genome editing system, we PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 3 / 27 PLOS PATHOGENSModification of AMT for two oomycete species successfully generated genome-edited P. infestans transformants and observed the expected virulence phenotypes after inoculating potato leaves with these mutants. Our work is provid- ing a new direction for AMT improvement and a way to potentially accelerate molecular genetic studies of the two devastating plant pathogens and other microorganisms. Results Oomycetes lack homologues of specific plant proteins important for Agrobacterium-mediated transformation To better understand the low AMT efficiency in P. infestans and test whether introducing plant proteins into transformation system would significantly improve Phytophthora transfor- mation efficiency, we first performed a bioinformatic study of proteins with similarities to pre- viously identified plant proteins, required for AMT, in different oomycete species. We used target protein sequences from Arabidopsis to perform local BLAST using publicly available genomic data from each selected plant or oomycete species. Sequence similarity shown in Fig 1A is based on the ratio between the highest alignment score (bit-score) of local BLAST results from a target species and the one from Arabidopsis. In the selected monocotyledonous and dicotyledonous plants, the sequence similarities of the obtained proteins were all above 0.25, except that the sequence similarity of Rad50 in Solanum tuberosum and Vitis vinifera were merely 0.04 and 0.17, respectively (Fig 1A and S3 Table). In oomycete species, similar proteins with sequence similarity greater than 0.45 were detected for IMPA-4, Rab8, SGA1 and histone H2A. For AGP17, PP2C, Ku70, Rad50 and Ku80, sequence similarities of similar proteins were mostly above 0.15 (average values were 0.19, 0.19, 0.20, 0.18 and 0.15, respec- tively) in selected oomycete species. Interestingly, sequence similarity of AtVIP1 or its related Arabidopsis paralogues, including bZIP29, bZIP30, bZIP69, posF21, bZIP18 and bZIP52, were less than 0.08 in all selected oomycete species, indicating that absence of an AtVIP1 homo- logue might have a negative impact on the AMT rate in oomycetes. AtVIP1 enhances Agrobacterium-mediated genetic transformation To further test our hypothesis that introduction of plant proteins, required for AMT, into a transformation system increases AMT efficiency in oomycetes, we constructed a vector to deliver AtVIP1 from Agrobacterium to host cells to potentiate the transformation process. To make AtVIP1 translocatable by Agrobacterium T4SS, we fused the coding sequence of AtVIP1 to virFΔ42N (truncated virF, lacking the sequence encoding the 42 N-terminal amino acids, that functions as a C-terminal transport signal for VirB/D4-translocated proteins in Agrobac- terium [45]), and added green fluorescent protein-encoding sequence (GFP) to the N-terminus (S1A Fig). As shown in S1B Fig, GFP signal was observed in both cytoplasm and nucleus of infected Nicotiana benthaminana cells 3 days post leaf infiltration with A. tumefaciens EHA105 strain carrying p302b-gfp-AtVIP1-virFΔN42. These data suggested that the GFP-At- VIP1-virFΔ42N fusion protein was translocated into infected cells through T4SS during Agro- bacterium infection. To test whether our modified protocol enhances AMT efficiency, we first tested it in wheat, a cereal crop with low AMT efficiency [46]. We introduced the prokaryotic expression plasmid p302b-AtVIP1-virFΔN42 (pV1F) and the binary plasmid pWMB110-GUS carrying the β-glu- curonidase coding gene gus (contains maize adh1 intron) into Agrobacterium strain EHA105 (S2A Fig). By inoculating leaf segments of wheat cv. ‘Fielder’ with A. tumefaciens EHA105 strain containing pV1F and pWMB110-GUS, a significantly higher transient transformation efficiency in leaf tissue was observed as compared to the control treatment inoculated with PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 4 / 27 PLOS PATHOGENSModification of AMT for two oomycete species Fig 1. (a) The protein similarity analysis of plant proteins, important for AMT of Arabidopsis, and their homologues in other plant and oomycete species. Representative plants with an established Agrobacterium-mediated transformation protocol and eight oomycete species were selected for phylogenetic analysis. The left phylogenetic tree was generated by the OrthoFinder software based on variations of single-copy orthologous genes in released genomic data of different species using IQ-tree with JTT + G4 model of evolution. The left bar indicates amino acid substitutions per site. The right heatmap indicates sequence similarity of known plant proteins envolved in Arabidopsis AMT process and their PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 5 / 27 PLOS PATHOGENSModification of AMT for two oomycete species homologues in selected plants and oomycete species. The right bar indicates sequence similarity. (b) Schematic representation of the main idea of our improved transformation strategy. The use of the helper plasmid pV1F results in expression of virFΔN42 tagged AtVIP1. The recombinant AtVIP1 is delivered into oomycete zoospores via VirB/VirD4 T4SS, and thereby facilitates the process of T-DNA complex (T-complex) targeting the nucleus of an infected zoospore or an encysted zoospore with a germ tube. https://doi.org/10.1371/journal.ppat.1011346.g001 EHA105 strain containing the vector lacking the AtVIP1 gene (pEV) (S2B Fig). In addition, gus expression in root segments increased from 1.8% to 11.6% when transformed with the vec- tor containing the AtVIP1 gene (S2C and S2D Fig). These results indicate that AtVIP1 is an effective enhancer of Agrobacterium-mediated genetic transformation in plant tissues. Expressing AtVIP1 in A. tumefaciens induces T-DNA integration in P. infestans To extend our AMT system to a wider range of eukaryotic species, characterized by low trans- formation efficiency, we tested it in one of the most important phytopathogenic oomycetes, P. infestans, in which AMT protocols have already been set up since years ago [7, 47]. A sche- matic illustration of our modified AMT method for P. infestans is presented in Fig 1B. To test if AtVIP1 also facilitates AMT in P. infestans, we first constructed the binary vector pLY40-gfp (Fig 2A), which carries a T-DNA region containing the gfp gene driven by Bremia lactucae Ham34 promoter, and the geneticin (G418) resistance gene nptII driven by Phytophthora sojae Rpl41 promoter. The pLY40-gfp was introduced into A. tumefaciens strain EHA105 together with either pV1F or pEV. A. tumefaciens strains carrying above-mentioned DNA construct combinations (Fig 2A) were co-cultivated with zoospores of P. infestans strains JH19 and HB1501, and subsequently geneticin-resistant transformants were obtained according to the methods presented in the flow diagram in S3 Fig. While G418 resistant colonies were obtained with both tested construct combinations, co-cultivation with A. tumefaciens containing the pV1F plasmid resulted in sig- nificantly more G418 resistant colonies (or colonies with observed GFP-signal) in both HB1501 and JH19 (Fig 2B–2C and Table 1). One of the representative transformants (JH19 background), named as T1, which was obtained with this modified AMT method, expressed a GFP-size protein and showed a strong GFP signal compared to the transformant T3 (JH19 background) that was obtained using the empty pLY40 as negative control (Fig 2D and 2E). The Southern blot analysis of representative transformants of the HB1501 strain background suggested that T-DNA fragments with the nptII gene were integrated into the genomic DNA of all seven of them (Fig 2F). These results indicate that prokaryotic-expressed AtVIP1 in A. tumefaciens considerably promotes T-DNA integration during AMT in P. infestans. Transformation of Plasmopara viticola using the modified AMT procedure In this study, we also extended application of our modified AMT protocol to P. viticola, a diffi- cult-to-transform oomycete that causes grapevine downy mildew. Based on the previously established AMT protocol for the biotrophic fungus Podosphaera xanthii [48], A. tumefaciens EHA105, carrying LY40-gfp and pV1F, was used for co-cultivation with released zoospores of P. viticola isolate BS5. After resistance selection, applied by rinsing P. viticola inoculated grape leaves with G418, we isolated one transformant called T1 (Figs 3 and S5B). We observed a strong GFP signal in both sporangia and mycelia of T1 during infection using confocal micros- copy, and the GFP signal was stable on grape leaves in three sub-inoculation generations (Fig 3A and 3C). Western blot analysis indicated that a GFP-size band could also be detected in protein extracts from the transformant T1 (Fig 3B), and Southern blot analysis, carried out using PCR-amplified nptII gene as a probe, showed that one T-DNA segment was successfully PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 6 / 27 PLOS PATHOGENSModification of AMT for two oomycete species Fig 2. Generating P. infestans transformants expressing gfp using the modified AMT method. (a) Schematic representation of the constructs used in this experiment. A. tumefaciens strain EHA105 with pV1F and pLY40-gfp was used for P. infestans transformation. (b) Quantification of colonies that acquired G418 resistance as a result of transforming gfp into P. infestans strains JH19 and HB1501 using our modified AMT procedure. EHA105 with the helper plasmid pEV was used as control in this experiment. All data represent average values from three independent experiments with the indicated standard deviations. Statistical differences among the samples were analyzed with Sˇı´da´k’s multiple comparisons test (P< 0.0021: , P< 0.0001: *). (c) Two representative plates from the transformation experiment shown in (b). (d) Immunoblot of P. infestans JH19 transformant T1, expressing free gfp, probed with an anti-GFP antibody. Protein extracted from N. benthamiana leaves transiently expressing gfp driven by the CaMV35s promoter was used as positive control in lane 1. (e) The P. infestans JH19 transformant T1 expressing a detectable GFP signal was obtained by AMT using A. tumefaciens carrying constructs described in (a). Scale bars = 40 μm. T3, a randomly selected empty vector transformant was used as negative control. GFP expression in the transformant was analysed by PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 7 / 27 PLOS PATHOGENSconfocal microscopy. The protein blot was stained with Coomassie Blue to confirm equal loading. (f) Southern blot analysis of representative gfp transformants of P. infestans strain HB1501. Genomic DNA (4 μg) was digested with HindIII and all blots were probed with a fragment containing the nptII gene to detect the presence of T-DNA. Numbers on the left indicate the positions of molecular weight markers (kb). https://doi.org/10.1371/journal.ppat.1011346.g002 Modification of AMT for two oomycete species integrated into the genome of T1 (Fig 3D). Consequently, the extensibility of our strategy to P. viticola suggests that AtVIP1 could also be considered for optimizing AMT methods in the case of other oomycete species with poor efficiency of transformation. AtVIP1 facilitates Agrobacterium-mediated delivery of LbCas12a into P. infestans To test if our modified AMT method could also facilitate genome editing in oomycetes, we selected P. infestans for the first attempt as the Cas12a-based method had recently been set up in this species [12]. We first fused the coding sequences of the Phytophthora sojae NLS, human codon-optimized Lachnospiraceae bacterium Cas12a (LbCas12a) and GFP, and inserted the recombinant sequence into the T-DNA region of pLY40 to obtain pLY40-Cas12a-gfp (Fig 4A). A. tumefaciens strain EHA105 carrying pV1F and pLY40-Cas12a-gfp was used to transform zoospores of P. infestans strains JH19 and HB1501. As a result, up to 61 and 22 G418 resistant transformants were obtained for JH19 and HB1501, respectively, in three independent trans- formation experiments (Table 1). The localization of the GFP signal in the transformants T5 (JH19) and T6 (HB1501) was distinctly nuclear as visualized by confocal microscopy (Fig 4B). Table 1. Genetic transformations of P. infestans strains JH19 and HB1501 by A. tumefaciens EHA105 with or without the helper plasmid pV1F. Binary plasmid pLY40-gfp pLY40-gfp pLY40-gfp pLY40-gfp pLY40-Cas12a- gfp pLY40-Cas12a- gfp pLY40-Cas12a- gfp pLY40-Cas12a- gfp Binary plasmid pLY40-PiMADS- KO pLY40-PiMADS -KO pLY40-Avr8-KO pLY40-Avr8-KO Helper plasmida pEV pV1F pEV pV1F pEV pV1F pEV P. infestans strain JH19 JH19 HB1501 HB1501 JH19 JH19 HB1501 pV1F HB1501 Helper plasmid pEV pV1F pEV pV1F P. infestans strain JH19 JH19 HB1501 HB1501 GFP+/G418R colonies in attempt 1b 0/3 11/21 0/0 2/8 1/1 13/22 0/0 3/6 GFP+/G418R colonies in attempt 2 GFP+/G418R colonies in attempt 3 Average GFP/G418R colonies 1/2 10/29 0/1 7/12 0/7 3/19 0/0 3/7 1/3 13/24 0/0 5/12 0/2 6/20 0/0 7/9 0.67/2.67 11.33/24.67* 0/0.33 4.67/10.67 0.33/3.33 7.33/20.33 0/0 4.33/7.33 G418R colonies in attempt 1 G418R colonies in attempt 0 7 2 11 2 0 4 2 9 G418R colonies in attempt 3 Average G418R colonies 1 5 1 8 0.33 5.33* 1.67 9.33* a Helper plasmids used in this experiment are described in detail in S2 Table. b As determined with GFP-observation by confocal microscopy (GFP). As determined 14 days after acquired transformants transferred to another rye-sucrose medium with 5 or 10 μg/L geneticin (G418). The marked values indicate significantly different with the AMT process used same binary plasmid but pEV as the helper plasmid. Student’s t-test was used to determine the differences. https://doi.org/10.1371/journal.ppat.1011346.t001 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 8 / 27 PLOS PATHOGENSModification of AMT for two oomycete species Fig 3. Obtaining Plasmopara viticola BS5 stable transformant T1 using the modified AMT protocol. (a) P. viticola BS5 strain was used for transformation assay and was obtained by AMT using A. tumefaciens carrying constructs pLY40-gfp and pV1F. Scale bars = 40 μm. The confocal microscopy images were taken 7 days post inoculation with the zoospores from the third sub-generation of T1; wild type BS5 was used as negative control. (b) Immunoblot of P. viticola BS5 transformant T1 expressing free GFP, probed with an anti-GFP antibody. Protein extracted from N. benthamiana leaves transiently expressing gfp driven by the CaMV35s promoter was used as positive control in lane 2. (c) Transformant T1 of P. viticola BS5 expresses detectable GFP signal when infecting grapevine leaves (Zitian seedless, A17). Scale bars = 100 μm. Images were taken 7 days post inoculation and wild type BS5 was used as negative control. (d) Southern blot analysis of transformant T1 of P. viticola BS5. Genomic DNA (4 μg) was digested with HindIII and all blots were probed with a fragment containing the nptII gene to detect the presence of T-DNA. Numbers on the left indicate the positions of molecular weight markers (kb). https://doi.org/10.1371/journal.ppat.1011346.g003 As shown in Fig 4C, the hlbCas12a-GFP fusion protein was successfully detected by western blotting in both transformants T5 and T6 with hlbCas12a-gfp expressed in N. benthamiana used as a positive control. These data suggest successful delivery of the Cas12a gene into P. infestans using our modified AMT method, paving a way for an efficient genome editing method for this species. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 9 / 27 PLOS PATHOGENSModification of AMT for two oomycete species Fig 4. Integration of LbCas12a in P. infestans. (a) Schematic representation of the constructs used in the experiment. A. tumefaciens EHA105 carrying pLY40-Cas12a-gfp and either pV1F or pEV was used for P. infestans transformation. (b) Confocal micrograph of T5 and T6 transformants expressing PsNLS-lbCas12-GFP. The nuclear localization pattern of the fusion protein, indicated by white arrows, revealed mycelia and sporangia. T3, an empty vector pLY40 transformed line, is used as negative control. Bright field and GFP channels are presented. Scale bars = 40 μm. (c) Immunoblot of two representative P. infestans transformants expressing PsNLS-lbCas12-GFP. The gfp tagged hlbCas12a driven by CaMV35S promoter was expressed in N. benthamiana as positive control. The expected size of the protein is 176.6 kDa. The protein blot was stained with Coomassie Blue to confirm equal loading. (d) Quantification of positive P. infestans transformants expressing GFP tagged Cas12a (JH19 and HB1501 backgrounds) generated using the modified AMT procedure. A. tumefaciens EHA105 carrying the helper plasmid pEV was used as control in this experiment. Statistical differences among the samples were analyzed with Sˇı´da´k’s multiple comparisons test (P< 0.0021: , P< 0.0001: *). https://doi.org/10.1371/journal.ppat.1011346.g004 Editing a MADS-box-encoding gene in P. infestans To investigate the prospect of using the modified AMT method for CRISPR/Cas12a-mediated P. infestans genome editing, a single-copy gene encoding a MADS-box transcription factor (PITG_07059) was chosen as the first editing target in this study. The homologous MADS-box genes play an essential role in asexual reproduction and zoosporogenesis in both P. infestans and P. sojae [49, 50]. Two gRNAs, MADS-g1 and MADS-g2, targeting the 1332 nt MADS-box- PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 10 / 27 PLOS PATHOGENSModification of AMT for two oomycete species encoding sequence were designed using the procedure described by Ah-Fong et al. (Figs 5A and S6) [12]. In the binary construct pLY40-MADS-box-KO, MADS-g1 and MADS-g2 were flanked by 21-bp short direct repeats (DRs) and inserted in a tandem downstream of LbCas12a and the 73-nt poly-adenine sequence (pA), which was added to promote translation of Cas12a mRNA (S6C Fig). As a 3’ uridine-rich tail positively regulates CRISPR/Cas12a crRNA forma- tion, four thymines were added at the 3’ends of sequences encoding the gRNAs (S6C Fig) [12, 51]. The editing events within the MADS-box gene in the transformants were detected by per- forming PCR, with primers amplifying the full length of the gene, and sequencing the PCR amplicons. In 2 out of 16 G418 resistant transformants (T8 and T16), shorter PCR amplicons were observed suggesting presence of CRISPR/Cas12a-induced deletions (Fig 5B). Sequencing analysis confirmed that T8 and T16 carried identical 993 bp deletions, between MADS-g1 and MADS-g2 loci, resulting in a truncated MADS-box gene coding sequence (Figs 5C and S6E). The MEF2-like domain of the MADS-box protein contributes to the DNA binding activity, which is essential for a transcription factor. The MEF2-like domain was predicted to be located at the 2–72 aa sites by InterproScan (version 5.52–86.0) (Fig 5A). The sequencing data pre- sented in Figs 5C and S6E showed that both mutant MADS-box alleles, present in T8 and T16, lack the majority of the sequence encoding the MEF2-like domain, suggesting loss of function of the MADS-box proteins in these transformants. As shown in Fig 5D and 5F, MADS-box-edited JH19 strains T8 and T16 showed merely no sporangia output after 5 days of cultivation in PEA broth compared with untransformed wild type JH19, which yielded 49.22 sporangia in 10 μL PEA broth culture. Another transformant (T5) was selected as a non-edited control, in which no significant reduction of sporangia out- put was detected (Fig 5D and 5F). To check whether the phenotype observed in mutant T8 and T16 strains was specific to sporangia production, we decided to measure the vegetative mycelia growth rate in them. As a result, both T8 and T16 showed similar vegetative growth rates, comparable to the controls, on the rye-sucrose medium at 18˚C (S8A and S8B Figs). In addition, we performed an in planta test by inoculating potato leaves (Solanum tuberosum cv. De´sire´e) with T5, T8, T16 and wild type JH19. As MADS-box-edited JH19 strains yielded no sporangia, we selected mycelial discs instead of zoospores for inoculation assays in this experi- ment. T5 and wild type JH19 strains caused complete infectious lesions and developed obvious aerial mycelia. In contrast, T8 and T16 only showed partial watery lesions without formation of aerial mycelia (Figs 5E, 5G and S8C). We further examined the disease area by microscopy: both T5 and wild type JH19 strains produced sporangia in heavily diseased potato leaves, while T8 and T16 only developed sparse mycelia without detectable sporangia (S8C Fig). Editing Avr8-encoding gene in P. infestans HB1501 Phytophthora avirulence (Avr) genes are key determinant factors for gene-for-gene interaction with host plants such as potato and soybean (Dong et al., 2011). We selected Avr8 (PITG_07558), known as an avirulence gene recognized by potato late blight resistance gene R8 [52], as the second editing target. We selected P. infestans strain HB1501 for Avr8 editing since JH19 naturally overcomes the R8-mediated resistance, as was previously established using inoculation assays. As shown in S7A Fig, genome-wide single nucleotide polymorphism (SNP) analysis revealed that HB1501 is diploid. Avr8 encodes a 245 aa RxLR effector and was confirmed as a single-copy gene in strain HB1501 based on the read depth analysis (S7B Fig). The Avr8 protein includes 3 LWY motifs (63–107, 105–162, 165–218 aa), which might con- tribute to its novel activities as an RxLR effector during infection progress (Fig 6A) [53]. Cas12a guides RNAs Avr8-g1 and Avr8-g2 were designed to target Avr8 before the first LWY motif to increase chances of introducing a frame-shift mutation before this motif due to PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 11 / 27 PLOS PATHOGENSModification of AMT for two oomycete species PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 12 / 27 PLOS PATHOGENSModification of AMT for two oomycete species Fig 5. Editing a MADS-box transcription factor coding gene in P. infestans strain JH19. (a) Target sites of two gRNAs in the MADS-box coding sequence. The MEF2-like domain was predicted by InterproScan (version 5.52–86.0). The MEF2-like domain locates at 2–77 aa sites. (b) Detecting of editing events in MADS-box in JH19 transformants. The PCR assay on T8 and T16 revealed two homozygous editing events. (c) Analysis of PCR amplicon sequencing results from (b). The PAM sequences are marked in red, the target sequences are marked in blue, and the stop codons are indicated by red arrowheads. Each dash line represents a deleted nucleotide. (d) Culture scrapings from 10 days Pea broth cultures of wild type JH19, T5, T8 and T16. Scale bars = 0.4 mm. (e) Infection phenotypes of wild type JH19 (WT), transformants T5, T8 and T16 on leaves of susceptible cultivar potato cv. De´sire´e. Detached potato leaves were inoculated with mycelia medium discs, and images were recorded 5 days post inoculation. Scale bars = 2 cm. (f) Quantification of sporangia numbers in 10 μL PEA broth culture of wild type JH19, T5, T8 and T16 in (d). (g) Quantification of lesion size in detached leaves of potato cv. De´sire´e inoculated with wild type JH19 (WT), T5, T8 and T16 in (e). All data represent average values from three independent experiments with the indicated standard deviations. Statistical differences among the samples were analyzed with Sˇı´da´k’s multiple comparisons test (P< 0.0021: , P< 0.0002: *). https://doi.org/10.1371/journal.ppat.1011346.g005 editing events (S7D Fig). The cartoon illustrating the design of the Avr8 CRISPR/Cas12a knockout construct is shown in S7C Fig, with Avr8-g1 and Avr8-g2 targets shown in red in S7D Fig. The transformants were PCR-genotyped for edits in Avr8 using primers amplifying the full-length gene, with PCR amplicons being subsequently sequenced. Out of 27 G418 resistant transformants, variant bands were observed in two of them, T3 and T10, suggesting both Avr8 alleles were altered in them (Fig 6B). Sequencing of the PCR amplicons from the shifted T3 and T10 bands showed deletions spanning both Avr8-g1 and Avr8-g2 target sites within Avr8. Unlike the editing events detected in MADS-box, edits in the Avr8 gene in T3 and T10 resulted in frame-shift mutations predicted to cause early termination of protein translation before the first LWY motif (Fig 6C). As a following step, we performed virulence assays by inoculating detached leaves of R8 transgenic potato (De´sire´e R8) with transformants T3, T10, T22 (unedited) and wild type HB1501 strain. We recorded the infection phenotypes 5 days post zoospore inoculation. Com- pared with unedited T22 and wild type HB1501, T3 and T10 caused lesions of significantly larger size upon infection of R8 transgenic potato leaves. Importantly, both T3 and T10 caused infection symptoms similar to T22 and wild type HB1501 when using detached leaves from susceptible wild type potato (De´sire´e WT) lacking the R8 gene (Fig 6D and 6E). We therefore reason that creating a loss-of-function Avr8 allele enables P. infestans to evade R8-mediated host resistance. Discussion As a widely used approach for transient and stable expression of exogenous genes, AMT has been set up for a broad spectrum of biological categories, including plants, microorganisms and even human cells [20, 54–56]. The AMT protocols for oomycete species, the majority of which are classified as destructive plant pathogens, have been reported, including Phy- tophthora palmivora, Phythium ultimum and Phytophthora infestans [57]. Although these AMT methods have been utilized for years, they require optimization due to generally low transformation efficiencies. Arabidopsis bZIP family transcription factor AtVIP1, known as a binding partner of A. tumefaciens effector VirE2, facilitates VirE2 nuclear import and A. tume- faciens infectivity [43, 58]. Moreover, significant improvement of transformation was observed in an A. tumefaciens transformation assay using AtVIP1-overexpressing tobacco plants [42]. Until now, utilization of AtVIP1 as a factor boosting the AMT efficiency required two steps: (1) generating a stable AtVIP1 transgenic line; (2) transforming a gene of interest into the AtVIP1 transgenic line. Overexpressing AtVIP1 in a stable transgenic line might cause an unpredictable pleiotropic phenotype as well as limit the range of selectable markers available for transformation of a gene of interest. In addition, when applying AtVIP1 the way described above, one is presented with a chicken-and-egg problem in the case of plant genotypes, which are not amenable to AMT to start with. A recent study in wheat reported that co- PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 13 / 27 PLOS PATHOGENSModification of AMT for two oomycete species Fig 6. Editing an avirulence gene Avr8 in P. infestans strain HB1501. (a) Target sites of two gRNAs in the Avr8 coding sequence. The LWY motifs were predicted by InterproScan (version 5.52–86.0). The LWY1 domain locates at 63–107 aa sites, the LWY2 domain locates at 105–162 aa sites, and the LWY3 domain locates at 165–218 aa sites. (b) Detecting editing of Avr8 in HB1501 transformants. The PCR assay in T3 and T10 revealed two homozygous editing events. (c) Analysis of PCR amplicon sequencing results from (b). The PAM sequences are marked in red, the target sequences are marked in blue, and stop codons are indicated by red arrowheads. Each dash line represents a deleted nucleotide. (d) Infection phenotypes of wild type HB1501 (WT), T3, T10 and T22 on detached leaves of R8 transgenic potato. Detached potato leaves were inoculated with zoospores of selected strains, and images were taken 5 days post inoculation. Scale bars = 2 cm. (e) Quantification of lesion sizes in (d). All data represent average values from three independent experiments with the indicated standard deviations. Statistical differences among the samples were analyzed with Sˇı´da´k’s multiple comparisons test (P< 0.0002: ). https://doi.org/10.1371/journal.ppat.1011346.g006 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 14 / 27 PLOS PATHOGENSModification of AMT for two oomycete species transformation of the wheat gene TaWOX5 from the WUSCHEL family with a gene of interest dramatically increases transformation efficiency, resulting in a lower genotype dependency, in 29 wheat varieties. The latter was similar to our original idea of modifying AMT for oomycete species [59]. Here, we present an alternative method utilizing AtVIP1 to optimize AMT for two oomycete species using prokaryotic expression of AtVIP1 fused with a sequence encoding a T4SS translocation tag in A. tumefaciens. In addition to utilizing AtVIP1, we have done additional modifications to the procedure while developing our modified AMT protocol for P. infestans (S3 Fig). Murashige and Skoog (MS) medium is commonly used for co-cultivation of A. tumefaciens and the explant material [60, 61]. Considering that A. tumefaciens shows a better growth rate in MS medium than in induction medium (IM), we used MS medium instead of IM for co-culturing A. tumefaciens and P. infestans zoospores. The germination of zoospores and transformation efficiency were not affected by this medium substitution. We used geneticin G418 as a selection agent in this study, while different P. infestans wild-type strains showed variation in G418 tolerance. For the JH19 strain, we used 5 μg/mL of G418 for selection of positive transformants, while for the HB1501 strain—10 μg/mL. In previously reported PEG-mediated P. infestans transformation or genome-editing proce- dures, several different situations might occur upon plasmid introduction: (a) the plasmid might integrate into an unstable site in the transformant’s genome; (b) the plasmid DNA might be later degraded by enzymes inside P. infestans cells [62]. The modified AMT proce- dure presented in this study has the advantages of the previously described AMT protocol, including no need to produce protoplasts, no need for a large amount of high-quality plasmid DNA, and, in addition, the integrated T-DNA fragment in the genomic DNA might be more stable than plasmid DNA [63]. The AMT method also comes with its issues e.g. during the T-DNA integration step: (a) some transgene instability might be observed, probably due to rearrangement of the T-DNA region, and/or due to homologous recombination between cop- ies of the transgene inserted into the DNA in the same nucleus; (b) only part of the T-DNA might integrate into genomic DNA [64]. Importantly, during PEG-mediated transformation, the plasmid DNA might not integrate straightaway in protoplasts but at a later stage, during cell differentiation. During the AMT process, T-DNA integration also potentially occurs in encysted zoospores that is followed by emergence of germ tubes. Both of the described above situations would result in chimeric mycelia in resulting transformants, although there are currently no data to indicate that this is the case with AMT [65]. In this study, each AMT experiment started with approximately 8 × 105 zoospores, and 9 of 18 experiments, conducted using our modified AMT method, produced at least 10 G418-resis- tant colonies; 5 of 18 attempts produced at least 20 G418-resistant colonies (Table 1). However, further characterization of acquired transformants revealed that part of the isolates, trans- formed with gfp, that showed resistance to G418, did not show a GFP signal during confocal microscopy analysis or expressed a GFP-size protein detectable by western blotting (Table 1 and S4 Fig). The Southern blot analysis of representative transformants of the HB1501 back- ground suggested that the T-DNA segments with the nptII gene were integrated into all six selected isolates, which were positive for the GFP signal, and one transformant (T1) with no GFP signal (Fig 2F). The latter could be due to the issue (b) of the AMT protocol that was men- tioned above. During the Southern blot analysis, we observed bands of similar small size (< 4.3 kb) in lanes 2, 3, 4, 7 and 8 (Fig 2F). These results might be caused by non-specific hybridization or, possibly, partial T-DNA integration events in selected HB1501 transformants. The above- mentioned observations are consistent with a similar phenomenon observed during plant PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 15 / 27 PLOS PATHOGENSModification of AMT for two oomycete species transformation [66]. To characterize better the integration events in transformants of oomy- cete species, we would like to perform next generation sequencing (NGS), that represents a highly sensitive approach to detect T-DNA insertions in transgenic isolates. There are few more differences between the PEG-mediated transformation and AMT methods, e.g. the different recipient cell types. Importantly, many protoplast cells used for PEG-mediated transformation are coenocytic, while most zoospores used for AMT are single- nucleated and wall-less [63, 67]. However, we have no sufficient evidence to suggest that AMT of P. infestans zoospores could eliminate the issue of potential heterokaryoteyotic transfor- mants, particularly because it was reported that a certain fraction of P. infestans zoospores had been observed to be multinucleate [68]. Additionally, the conditions used in our modified AMT method could not eliminate the possibility of encystment of the zoospores followed by emergence of a germ tube during co-cultivation of zoospores and Agrobacterium cells. To illus- trate such scenario, we included the potential situation of T-DNA integration in multinucleate cells in Fig 1B. We also performed a comparison between the published PEG-mediated trans- formation and AMT protocol, and our modified AMT protocol for P. infestans (S4 Table). As a result, we would define our modified AMT protocol as an option to be considered, rather than top choice, for P. infestans transformation, as transformation efficiencies associated with different protocols are difficult to compare due to different standards used. In addition to P. infestans, genetic transformation methods have not been set up for many biotrophic oomycetes. Martı´nez-Cruz et al. has reported an AMT method for Podosphaera xanthii, a biotrophic fungus that causes cucurbit powdery mildew [48]. Based on the AMT method for P. xanthii, we extended our modified AMT protocol to Plasmopara viticola and successfully obtained a positive transformant (Figs 3 and S5). There are a few more issues that need to be addressed when it comes to AMT of P. viticola, including (a) transformation effi- ciency might vary in different P. viticola isolates; (b) leaves from different grapevine varieties may contribute differently to screening of resistant transformants; (c) it would be difficult to recover P. vitcola transformants from infected grapevine leaves without aseptic conditions; (d) stable expression of an exogenous gene needs further validation in subsequent generations of obtained transformants [1]. Although we have acquired one positive stable transformant of P. viticola based on our modified AMT method with AtVIP1, we still need to do more transfor- mation attempts and analysis to draw a solid conclusion, in respect to the transformation effi- ciencies, in the future. Surely, we know the importance of acquiring a stable transformant of P. viticola, while our experimental procedure comes with a few shortcomings: (a) the transforma- tion efficiency is not high enough as compared to other oomycete species; (b) we have not per- formed a sufficient number of transformation assays with other P. viticola isolates. Consequently, the extensibility of our strategy to P. viticola suggests that AtVIP1 could be con- sidered for modifying AMT methods for other oomycete species, especially some hard-to- transform biotrophic oomycetes, such as Bremia lactucae. Unlike gene overexpression, integration of a CRISPR/Cas cassette in transformants does not guarantee successful editing of a gene target. Previous reports have demonstrated that edit- ing efficiency varies greatly, from 1% to 100%, among different eukaryotic pathogens [69]. The editing frequency of CRISPR/Cas12a in P. infestans reached 13% by using the PEG transfor- mation method [10, 12]. In this study, editing of the MADS-box gene resulted in two homozy- gous mutants out of 16 transformants, and editing of Avr8 genes resulted in two homozygous mutants out of 28 transformants (Table 1). Although introducing AtVIP1 does not seem to increase the frequency of CRISPR/Cas12a-mediated genome editing events in P. infestans, our modified AMT method still provides a valuable option for future related studies. We selected two target genes for CRISPR/Cas12a-mediated genome editing in this study based on two criteria. The first is avoiding multi-copy target genes to reduce the difficulty of PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 16 / 27 PLOS PATHOGENSModification of AMT for two oomycete species editing and transformant screening, and the second is that transformants with successful edit- ing events should present explicit phenotypes. The cycle of aerial asexual sporangia dispersal plays a crucial role in late blight development [70]. Thus, we selected MADS-box as our first target gene that is reported to express only in sporulating mycelia and spores of P. infestans. MADS-box (PITG_07059) was first identified by Leesutthiphonchai and Judelson in 2018 and MADS-box transcription factors play significant roles in eukaryotes [50]. We utilized modified AMT and CRISPR/Cas12a-mediated genome editing methods and obtained two edited MADS-box mutants (Fig 5). Consistently with the results obtained using the RNAi approach, our edited MADS-box mutants produced no sporangia and showed reduced ability to infect potato leaves (Fig 5). RNAi is triggered by siRNAs whose silencing efficiency is not guaranteed and varies widely in different transformants. In contrast to RNAi methods, genome editing would provide more stable phenotypic data for further research on P. infestans. Since during CRISPR/Cas12a-mediated genome editing, based on the AMT method, the Cas12a expression cassette stably integrates into the P. infestans genome, an additional self-fertilization step could be used to remove Cas12a from the original transformant, similarly to the situation in planta [71]. Disrupting the recognition of phytopathogen avirulence (AVR) genes by plant resistance (R) genes normally produces easily observable phenotypes. Thus, we selected Avr8 (also called AVRSmira2 and PITG_07558), a single-copy AVR gene recognized by potato late blight resis- tance gene R8, as our second genome editing target. We chose P. infestans strain HB1501 for Avr8 gene editing because seemingly the JH19 strain broke down the R8-mediated resistance in potato. Avr8 was identified by analysis of variance (ANOVA) using the average AUDPC val- ues of both its responses to the R gene and field trials [72]. The broad spectrum late blight resistance gene R8 that recognizes Avr8 was cloned from Solanum demissum, based on a previ- ously published coarse map position on the lower arm of chromosome IX, and the correlation between the expression levels of Avr8 and R8-mediated resistance had been proven in a previ- ous study [73]. However, an inoculation assay of an Avr8 knockout P. infestans mutant on potato carrying the R8 gene has not been reported yet. In our study, we produced two genome-edited Avr8 mutants of P. infestans (strain HB1501), and both mutants induced infec- tion lesions in leaves of the R8 transgenic potato line cv. De´sire´e (Fig 6D and 6E). Collectively, generating P. infestans transformants or genome edited mutants via our modi- fied AMT procedure is less laborious and results in an acceptable transformation rate as com- pared with the previously reported AMT protocol and PEG-mediated protoplast transformation method. Successfully acquiring a stable transformant of P. viticola gives a strong hint of gaining a potential advantage by utilizing proteins, such as AtVIP1, that play an important role in host cells during the AMT process. Although AMT has already been estab- lished in plenty of phytopathogen species, it would be particularly interesting to investigate whether our modified AMT protocol would increase the efficiency of transformation or that of CRISPR/Cas-mediated genome editing in them or even extend the effect to other oomycete or plant species. Materials and methods Growth conditions for P. infestans, P. viticola, bacteria and plants P. infestans strains were cultured on rye-sucrose medium (agar 15g/L) at 18˚C. P. infestans strain JH19 was isolated from infected tomato in San Diego Country, California in 1982 and kindly provided by Howard S. Judelson lab [12]. P. infestans strain HB1501 was isolated from Hebei Province in 2015 [74]. Plasmopara viticola isolate BS5 was kindly provided by Dr. Linfei Shangguan at Nanjing Agricultural University and was isolated in Nanjing, China in 2017. A. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 17 / 27 PLOS PATHOGENSModification of AMT for two oomycete species tumefaciens strains and E. coli strains were grown on Luria–Bertani (LB) agar (NaCl 10g/L, Yeast extract 5g/L, Tryptone 5g/L, agar 15g/L) at 28˚C and 37˚C, respectively. Solanum tubero- sum L. cv. De´sire´e (2n = 4x = 48) and trangenic lines were grown in soil or in MS medium (MES 0.5 g/L, sucrose 20 g/L, agar 8 g/L, pH5.8), after seed surface sterilization, and main- tained in vitro. Vitis vinifera (Cultivar grape, Zitian Seedless, A17) was used for culturing of P. viticola isolate BS5. All plants were grown in environment-controlled growth chambers under long-day conditions (16 h light/8 h dark cycle at 140 μE sec-1m-2 light intensity) at 22˚C. Construction of plasmids Primer sequences used in these cloning procedures are described in S1 Table, and plasmids and cloning strategies are summarized in S2 Table. For AtVIP1 gene expression in A. tumefa- ciens, the coding sequences of AtVIP1 (AT1G43700) and virF (NC_003065.2) without the N terminal 42 amino acids (ΔN42virF) were PCR-amplified, using A. thaliana Col-0 cDNA library and A. tumefaciens C58 gDNA, respectively, as templates and cloned into p533BL. A pCB302B backbone construct with the virB1 promoter from A. tumefaciens C58 was used to drive acetosyringone induced gene expression in A. tumefaciens. To construct binary plasmid for P. infestans transformation, Bremia lactucae Ham34 pro- moter and terminator were PCR-amplified and cloned into I-CeuI site of pPZP-RCS2, while nptII expression cassette driven by Phytophthora sojae RPL41 promoter and B. lactucae Hsp70 terminator was inserted into AscI site of pPZP-RCS2, which resulted in pLY40. The coding sequence of gfp was PCR-amplified and cloned into I-CeuI site of pLY40 separately to obtain pLY40-gfp. The sequences of the RPL41 promoter from P. sojae, Ham34 promoter and termi- nator, Hsp70 terminator from B. lactucae were all PCR-amplified from pYF515, a plasmid con- struct used for genome editing in P. sojae [75]. To construct binary plasmids for CRISPR/Cas12a mediated P. infestans genome editing, the coding sequences of NLS derived from a P. sojae bZIP transcription factor [10], human codon-optimized LbCas12a from p33lb [76] and artificial synthetic polyA-crRNA-Ham34 ter- minator segment (S6C and S7C Figs) were fused and cloned into I-CeuI/PacI sites in pLY40, to obtain pLY40-MADS-box-KO and pLY40-Avr8-KO. The plasmid sequences of pLY40, pLY40-gfp and pLY40-Cas12a-gfp are presented in Appendix S1. Transient transformation assays in wheat tissue Transient expression assays in wheat tissue were performed as previously described for N. benthamiana with some modifications [77]. A. tumefaciens overnight culture was diluted in LB liquid medium without antibiotics, grown for 3–4 h and adjusted to OD600 = 0.5. Leaves and roots from 4-weeks old in vitro wheat plant were collected and divided into 5 mm seg- ments, then immersed for 10 min in Agrobacterium suspension, placed on MS medium at 22˚C for 3 days in growth chamber. Leaf and root segments were then rinsed by sterilized water and transferred into GUS staining solution and incubated at 37˚C overnight [78]. Agrobacterium meditated transformation of P. infestans zoospores AMT of P. infestans zoospores followed previous published method with few modifications and summarized in S3 Fig [7]. Briefly, P. infestans strains were ready for zoospore-induction after 14 days of growth on solid rye-sucrose medium (90mm petri dish plates). Approximately 5 mL of ice-cold sterilized water was used to flush and soak P. infestans culture and the culture plates were plated at 4˚C for 2 hours, and zoospores will be released (100 zoospores counts/μL for isolate JH19, 200 zoospores counts/μL for isolate HB1501). A. tumefaciens strains with PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 18 / 27 PLOS PATHOGENSModification of AMT for two oomycete species constructed plasmids for transformation was growing overnight with proper antibiotics (25 μg/mL rifampicin, 50 μg/mL kanamycin, 100 μg/mL spectinomycin), 1 mL of Agrobacter- ium culture was added into 50 mL LB liquid medium and grown for 3–4 h till OD600 reach 1.0. Agrobacterium cells were centrifuged at 4000 rpm for 10 min to collect cells and resuspended by MS liquid medium with 200 μM acetosyringone. Agrobacterium suspension was cultured at room temperature for 1 h and then mixed with harvested fresh P. infestans zoospores (about 8 × 105 zoospores for both isolate JH19 and HB1501). The mixture was cultured at room tem- perature for 45 min and zoospores were collected by centrifuge at 265 g for 5 min, each 200 μL of zoospore culture was spread at a 5 cm × 5 cm Nytran membrane upon MS solid medium plate (8 g/L agar), and the plates were cultured in dark at 22˚C for 4 d. The membrane was then moved upside down on Plich medium (0.5 g/L KH2PO4, 0.25 g/L MgSO4.7H2O, 1 g/L Asparagine, 1 mg/L Thiamine, 0.5 g/L Yeast extract, 10 mg/L β–sitosterol, 25 g/L Glucose, 15 g/L agar) with 1.5 mg/L G418 and 300 mg/L timentin, and plates were cultured at 18˚C for 4 d, the membranes were then removed and germinated mycelia were supposed to be observed, the plates were keep in 18˚C for 4 d. Melt rye-sucrose medium with 3 mg/L of G418 was then covered on Plich medium plates for further selection, the G418 concentration could be increased up to 5 mg/L at this stage. Agrobacterium meditated transformation of P. viticola zoospores AMT of P. viticola zoospores followed previous published method on cucurbit powdery mil- dew pathogen Podosphaera xanthii and summarized in S5A Fig [48]. Briefly, culture of A. tumefaciens strain EHA105 with pV1F and proper T-DNA construct was prepared following the same methods that described in AMT of P. infestans. Separately, the sporangia from P. viti- cola BS5 inoculated grape leaves were harvested by immersion of infected tissues in 30 mL of sterilized water with 0.01% Tween-20 and keep in room temperature for 1 h until zoospores released. About 10 mL (about 1 × 106 zoospores) P. viticola zoospores were gently mixed with same volume of A. tumefaciens suspension and co-cultivated for 1 h at room temperature in the dark in an orbital shaker at 65 rpm. Zoospores were then centrifuged for collection (265 g for 5 min), and zoospores in 1 mL of residual were deposited in young detached grape leaves (must be Zitian Seedless, A17 for the best inoculation rate). Two days after inoculation, the grape leaves were rinsed with 5 mg/L G418 and 300 mg/L Timentin to kill A. tumefaciens and select P. viticola transformants. To be noted, rinsing infection samples with G418 (5 mg/L) and Timentin (300 mg/L) should have no visible negative effect on grape leaves during the above-mentioned procedure. To acquire the stable transformant of P. viticola, the obtained isolate was sub-inoculated at zoospore stage on young grape leaves without G418 treatment. After at least three rounds of sub-inoculation, the P. viticola isolate that showed stable GFP sig- nal, was defined as a stable transformant. Virtualization of GFP CLSM (Leica AF6000 modular microsystems) was used to take pictures of P. infestans. A 489-nm line from an argon ion laser were used to excite green fluorescent protein (GFP). For each assay, six independent leaves were observed for each experiment and each experiment has at least 3 repeats. sgRNA design and cloning CRISPR/Cas12a targets for MADS-box (PITG_07059) and Avr8 (PITG_07558) were designed with overall consideration based on output data from EuPaGDT, CRISPOR and Deep-Cpf1 [79–81]. The used crRNAs were carefully inspected with sequencing data of P. infestans JH19 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 19 / 27 PLOS PATHOGENSModification of AMT for two oomycete species and 1501 to avoid off-target events. DNA oligonucleotides contains crRNAs and direct repeats (DR) were artificial synthesized and cloned into pLY40 based strategy described above and in S6C and S7C Figs. Detection of target gene editing Genomic DNA (gDNA) of P. infestans strain or relative transformants was first isolated from liquid cultural mycelia. Specifically, 5 mycelia discs of each P. infestans strain from rye-sucrose medium plates were cut and placed into 10 mL of PEA broth, and mycelia for gDNA extrac- tion were harvested after 5 days growth in dark at 18˚C. Mycelia were then blot up by filter paper and gDNA was extracted with Omega E.Z.N.A. Plant DNA Kit (HP). To confirm edit- ing, primer pair F: 5’-ATGGGCCGCAAGAAGATCCAG-3’ and R: 5’-TCAAACAGCCACA CGTTGACGCTTG-3’ were for checking MADS-box editing in P. infestans JH19 derived transformants, and primer pair F: 5’-ATGCGCTCAATCCAACTTCTG-3’ and R: 5’-TTAC GATGTTTTCGCTTCTTTAAAAAG-3’ was for checking Avr8 editing in P. infestans 1501 derived transformants. Protein analysis P. infestans protein assay referred to previously described method [12]. Briefly, mycelia were collected from PEA broth culture and total protein was extracted with Beyotime RIPA P0013B Lysis Buffer. Immunoblots were performed as described [78]. Total protein was eluted in sodium dodecylsulfate (SDS) sampling buffer and proceed to western blot analysis. LbCas12a was detected by immunoblotting with anti-LbCas12a (Cpf1) antibody (Sigma/SAB4200777, dilution 1:4000), followed by a secondary antibody conjugated to FITC (ThermoFisher Scien- tific, dilution 1:5000). Southern blot analysis For T-DNA integration analysis, we performed southern blot with genomic DNA obtained from P. infestans transformants or wild-type (WT) isolates. Four μg genomic DNA was digested with restriction enzyme HindIII (Takara), and separated on DNA agarose gel (1%) and then blotted on positively charged nylon membranes. A nptII gene fragment was used as probe. Further preparations of probes DIG-labeling, hybridization and chemiluminescent detection were conducted according to the operation protocol of DIG-High Prime DNA label- ing and Detection Starter Kit (ROCHE/11585614910). Leaf inoculation with P. infestans For zoospores inoculation assays, zoospores of P. infestans strains were collected from plates followed by the same method described above. 10 μL drops with 200,000 to 400,000 zoospores per mL were inoculated on detached leaves from 4 to 6 weeks old potato plants of wild type De´sire´e and R8 trangenic potato lines. For mycelial disc inoculation assays, detached potato leaves were inoculated with mycelial discs (5 mm diameter) taken from the edge of 7 days-old rye-sucrose medium culture of P. infestans. Inoculated leaves were kept in a plastic tray (30 × 44 × 8 cm) covered with polypropylene film at 22 ± 3˚C. Results were recorded 5 days post inoculation. For each inoculation assay, 8 independent leaves were used for each experi- ment and each experiment has at least 3 repeats. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 20 / 27 PLOS PATHOGENSModification of AMT for two oomycete species Supporting information S1 Table. Primers used in this study. (XLSX) S2 Table. Plasmids used in this study. (XLSX) S3 Table. Sequence similarity data used in Fig 1A. (XLSX) S4 Table. Comparisons of modified AMT protocol and previous reported transformation protocols for P. infestans. (XLSX) S1 Fig. Translocating AtVIP1 from A. tumefaciens to host cells. (a) The expression cassette used for translocating AtVIP1 fused with GFP (b) Confocal microscopy observation of N. benthamiana leaves infiltrated with A. tumefaciens EHA105 carrying the construct described in (a). Images were taken at 3 days post infiltration. Images are single confocal sections and are representative of images obtained in three independent experiments. White arrows indicate observed nuclei in tobacco cells. Scale bars = 40 μm. Three independent experiments were per- formed for each assay with similar results. (TIF) S2 Fig. AtVIP1 enhances transient AMT efficiency in wheat tissues. (a) Schematic represen- tation of plasmid constructs used in this experiment. The pWMB110-gus construct contains a β-glucuronidase expression cassette, carrying the maize adh1 intron, in the T-DNA region. (b-c) Transient transformation on wheat leaf and root segments. Dissected wheat tissue seg- ments were inoculated with A. tumefaciens EHA105 carrying the binary plasmid pWMB110- gus and either pV1F or the control plasmid pEV. At 3 days post inoculation, GUS activity was analyzed by histochemical staining. Scale bars = 2 mm. At least 50 leaf or root segments were recorded in each experiment. Each experiment was repeated 3 times and representative results were presented. (d) Quantification of root segments that expressed the gus gene in (c). Statisti- cal differences among the samples were analyzed with Sˇı´da´k’s multiple comparisons test (P< 0.0001: ***). (TIF) S3 Fig. Schematic outline of modified AMT for P. infestans. (TIF) S4 Fig. Western blot characterization of P. infestans HB1501 transformants obtained using pLY40-gfp and pV1F. Total protein samples were purified from 14 transformants with a GFP signal (up) and 14 transformants without a GFP signal (down). All blots were probed with an anti-GFP antibody. The protein blot was stained with Ponceau S to confirm equal loading. (TIF) S5 Fig. Obtaining Plasmopara viticola BS5 transformant T1, expressing gfp, using the opti- mized AMT method. (a) Schematic outline of the optimized AMT method for P. viticola BS5. (b) AMT with only pLY40-gfp produced no G418 resistant transformants of P. viticola BS5 (left), while AMT with pLY40-gfp and pV1F produced the transformant T1 (white arrow) that is resistant to G418 (right). Scale bars = 2 mm. (TIF) PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 21 / 27 PLOS PATHOGENSModification of AMT for two oomycete species S6 Fig. The gRNA target regions used for MADS-box genome editing in P. infestans. (a) Genome-wide allele ratio analysis of P. infestans JH19. (b) Copy number of PITG_07059 (MADS-box) relative to single-copy control gene (= 1.0), determined based on read depth in DNA library of JH19 strain. (c) Schematic representation of the constructs used in this experi- ment. The pLY40-MADS-box-KO with either pV1F or pEV were used for P. infestans transfor- mation in this experiment. Two gRNAs for MADS-box editing are named as g1 and g2. (d) Gene sequence of PITG_07059 (MADS-box). Sequences marked in red are targeted by g1 and g2. (e) Sequencing chromatograms of MADS-box in wild type JH19, T8 and T16. Both T8 and T16 showed single peaks in either the g1 (left) or g2 (right) target sites. The wild type sequences with gRNA targets are shown at the top of the panel; black arrows indicate the 5’ border of the detected deletion. (TIF) S7 Fig. The gRNA regions used for Avr8 editing in P. infestans. (a) Genome-wide allele ratio analysis of P. infestans HB1501. (b) Copy number of PITG_07558 (Avr8) relative to single- copy control gene (= 1.0), determined based on read depth in DNA library of HB1501 strain. (c) Schematic representation of the constructs used in this experiment. (d) Gene sequence of PITG_07558 (Avr8). Two selected gRNA target regions are marked in red. (e) Sequencing chromatograms of Avr8 in T3, T10 and T22 of HB1501. Both T3 and T22 showed single peaks in both g1 and g2 target sites. The wild type sequences with gRNA targets are shown at the top of the panel; black arrows indicate the 5’ border of the detected deletion. (TIF) S8 Fig. Vegetative growth of P. infestans strains. (a) Mycelia cultured on the rye-sucrose medium were photographed at 5 days post inoculation. (b) Quantification of mycelium diame- ter of P. infestans strains in (a). All data represent average values from three independent experiments with the indicated standard deviations. (c) Microscopy images of the opposite side of the inoculated region of detached potato leaves in Fig 5E show the details of mycelia generated during infection. Images were taken at 5 days post inoculation. White arrowheads indicate the observed sporangia. Scale bars = 1 mm. (TIF) S1 Appendix. Sequences of pLY40, pLY40-gfp, pLY40-Cas12a-gfp. (DOCX) Acknowledgments We thank Dr. Howard Judelson (UCR, USA) and Prof. Yuanchao Wang (NAU) for the discus- sion on Phytophthora genome editing. Potato R8 transgenic line was kindly provided by Dr. Jack Vossen (WUR, Netherlands). The Plasmopara viticola BS5 was a kind gift from Prof. Lin- fei Shangguan (NAU). We thank Dr. Vitaly Citovsky from Stonybrook University for support- ive discussion on the original project design. We thank Ms. Ying Zheng (NAU) for confocal microscopy. We thank Ms. Zhao Hu, Dr. Xinyu Liu and Dr. Changling Mou (NAU) for their help with Southern blot analysis and constructive comments on this study. Author Contributions Conceptualization: Luyao Wang, Sanwen Huang, Suomeng Dong. Data curation: Luyao Wang, Fei Zhao, Haohao Liu, Suomeng Dong. Formal analysis: Luyao Wang, Fei Zhao, Haohao Liu, Suomeng Dong. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 22 / 27 PLOS PATHOGENSModification of AMT for two oomycete species Funding acquisition: Luyao Wang, Sanwen Huang, Suomeng Dong. Investigation: Luyao Wang, Fei Zhao, Haohao Liu, Suomeng Dong. Methodology: Luyao Wang, Fei Zhao, Haohao Liu, Han Chen, Fan Zhang, Vladimir Nekrasov, Suomeng Dong. Project administration: Luyao Wang, Sanwen Huang, Suomeng Dong. Resources: Luyao Wang, Sanwen Huang, Suomeng Dong. Software: Luyao Wang, Fan Zhang, Sanwen Huang, Suomeng Dong. Supervision: Luyao Wang, Sanwen Huang, Suomeng Dong. Validation: Luyao Wang, Sanwen Huang, Suomeng Dong. Visualization: Luyao Wang, Suomeng Dong. Writing – original draft: Luyao Wang, Sanwen Huang, Suomeng Dong. Writing – review & editing: Luyao Wang, Han Chen, Fan Zhang, Suhua Li, Tongjun Sun, Vladimir Nekrasov, Sanwen Huang, Suomeng Dong. References 1. Dubresson R, Kravchuk Z, Neuhaus J-M, Mauch-Mani B. Optimisation and comparison of transient expression methods to express the green fluorescent protein in the obligate biotrophic oomycete Plas- mopara viticola. Vitis. 2008; 47(4):235–40. 2. Fry W. Phytophthora infestans: the plant (and R gene) destroyer. Molecular plant pathology. 2008; 9 (3):385–402. https://doi.org/10.1111/j.1364-3703.2007.00465.x PMID: 18705878 3. Chen H, Shu H, Wang L, Zhang F, Li X, Ochola SO, et al. Phytophthora methylomes are modulated by 6mA methyltransferases and associated with adaptive genome regions. Genome biology. 2018; 19 (1):1–16. 4. Sabbadin F, Urresti S, Henrissat B, Avrova AO, Welsh LR, Lindley PJ, et al. Secreted pectin monooxy- genases drive plant infection by pathogenic oomycetes. Science. 2021; 373(6556):774–9. https://doi. org/10.1126/science.abj1342 PMID: 34385392 5. Luo X, Tian T, Feng L, Yang X, Li L, Tan X, et al. Pathogenesis-related protein 1 suppresses oomycete pathogen by targeting against AMPK kinase complex. Journal of Advanced Research. 2022. https://doi. org/10.1016/j.jare.2022.02.002 PMID: 36585103 6. Ah-Fong AM, Bormann-Chung CA, Judelson HS. Optimization of transgene-mediated silencing in Phy- tophthora infestans and its association with small-interfering RNAs. Fungal Genetics and Biology. 2008; 45(8):1197–205. https://doi.org/10.1016/j.fgb.2008.05.009 PMID: 18599326 7. Vijn I, Govers F. Agrobacterium tumefaciens mediated transformation of the oomycete plant pathogen Phytophthora infestans. Molecular Plant Pathology. 2003; 4(6):459–67. https://doi.org/10.1046/j.1364- 3703.2003.00191.x PMID: 20569405 8. Cvitanich C, Judelson HS. Stable transformation of the oomycete, Phytophthora infestans, using micro- projectile bombardment. Current genetics. 2003; 42(4):228–35. https://doi.org/10.1007/s00294-002- 0354-3 PMID: 12589474 9. 10. 11. van West P, de Jong AJ, Judelson HS, Emons AMC, Govers F. TheipiO Gene ofPhytophthora infestan- sIs Highly Expressed in Invading Hyphae during Infection. Fungal Genetics and Biology. 1998; 23 (2):126–38. https://doi.org/10.1006/fgbi.1998.1036 PMID: 9578626 Fang Y, Tyler BM. Efficient disruption and replacement of an effector gene in the oomycete P hytophthora sojae using CRISPR/C as9. Molecular plant pathology. 2016; 17(1):127–39. https://doi. org/10.1111/mpp.12318 PMID: 26507366 van den Hoogen J, Govers F. Attempts to implement CRISPR/Cas9 for genome editing in the oomycete Phytophthora infestans. bioRxiv. 2018:274829. 12. Ah-Fong AM, Boyd AM, Matson ME, Judelson HS. A Cas12a-based gene editing system for Phy- tophthora infestans reveals monoallelic expression of an elicitor. Molecular plant pathology. 2021. https://doi.org/10.1111/mpp.13051 PMID: 33724663 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 23 / 27 PLOS PATHOGENSModification of AMT for two oomycete species 13. Mansfield J, Genin S, Magori S, Citovsky V, Sriariyanum M, Ronald P, et al. Top 10 plant pathogenic bacteria in molecular plant pathology. Molecular plant pathology. 2012; 13(6):614–29. https://doi.org/ 10.1111/j.1364-3703.2012.00804.x PMID: 22672649 14. Binns AN, Thomashow MF. Cell biology of Agrobacterium infection and transformation of plants. Annual Reviews in Microbiology. 1988; 42(1):575–606. 15. Gelvin SB. Integration of Agrobacterium T-DNA into the plant genome. Annual review of genetics. 2017; 51:195–217. https://doi.org/10.1146/annurev-genet-120215-035320 PMID: 28853920 16. Zhang X, Henriques R, Lin S-S, Niu Q-W, Chua N-H. Agrobacterium-mediated transformation of Arabi- dopsis thaliana using the floral dip method. Nature protocols. 2006; 1(2):641–6. https://doi.org/10.1038/ nprot.2006.97 PMID: 17406292 17. Gallois P, Marinho P. Leaf disk transformation using Agrobacterium tumefaciens-expression of heterol- ogous genes in tobacco. Plant gene transfer and expression protocols. 1995:39–48. https://doi.org/10. 1385/0-89603-321-X:39 PMID: 8563823 18. Hiei Y, Komari T, Kubo T. Transformation of rice mediated by Agrobacterium tumefaciens. Plant molec- ular biology. 1997; 35(1):205–18. PMID: 9291974 19. D’spain S, Andrade PI, Brockman NE, Fu J, Wickes BL. Agrobacterium tumefaciens-Mediated Trans- formation of Candida glabrata. Journal of Fungi. 2022; 8(6):596. https://doi.org/10.3390/jof8060596 PMID: 35736079 20. Soltani J, van Heusden GPH, Hooykaas PJ. Agrobacterium-mediated transformation of non-plant organisms. Agrobacterium: from biology to biotechnology. 2008:649–75. 21. Michielse CB J Hooykaas PJ, JJ van den Hondel CA, J Ram AF. Agrobacterium-mediated transforma- tion of the filamentous fungus Aspergillus awamori. Nature protocols. 2008; 3(10):1671–8. https://doi. org/10.1038/nprot.2008.154 PMID: 18833205 22. Cardoza RE, Vizcaino JA, Hermosa MR, Monte E, Gutieˊrrez S. A comparison of the phenotypic and genetic stability of recombinant Trichoderma spp. generated by protoplast-and Agrobacterium-medi- ated transformation. Journal of Microbiology. 2006; 44(4):383–95. PMID: 16953173 23. Michielse CB, Hooykaas PJ, van den Hondel CA, Ram AF. Agrobacterium-mediated transformation as a tool for functional genomics in fungi. Current genetics. 2005; 48:1–17. https://doi.org/10.1007/ s00294-005-0578-0 PMID: 15889258 24. 25. Takahara H, Tsuji G, Kubo Y, Yamamoto M, Toyoda K, Inagaki Y, et al. Agrobacterium tumefaciens- mediated transformation as a tool for random mutagenesis of Colletotrichum trifolii. Journal of General Plant Pathology. 2004; 70:93–6. dos Reis MCl Fungaro MHP, Duarte RTD, Furlaneto L, Furlaneto MC. Agrobacterium tumefaciens- mediated genetic transformation of the entomopathogenic fungus Beauveria bassiana. Journal of microbiological methods. 2004; 58(2):197–202. https://doi.org/10.1016/j.mimet.2004.03.012 PMID: 15234517 26. Dobinson KF, Grant SJ, Kang S. Cloning and targeted disruption, via Agrobacterium tumefaciens-medi- ated transformation, of a trypsin protease gene from the vascular wilt fungus Verticillium dahliae. Cur- rent genetics. 2004; 45:104–10. https://doi.org/10.1007/s00294-003-0464-6 PMID: 14618375 27. Rolland S, Jobic C, Fèvre M, Bruel C. Agrobacterium-mediated transformation of Botrytis cinerea, sim- ple purification of monokaryotic transformants and rapid conidia-based identification of the transfer- DNA host genomic DNA flanking sequences. Current genetics. 2003; 44:164–71. https://doi.org/10. 1007/s00294-003-0438-8 PMID: 12937946 28. Sarah F, Kapoor P, Lee M-h, Briley A, Nairn CJ. Agrobacterium tumefaciens-mediated transformation of Fusarium circinatum. Mycological Research. 2001; 105(3):259–64. 29. Mullins ED, Chen X, Romaine P, Raina R, Geiser D, Kang S. Agrobacterium-mediated transformation of Fusarium oxysporum: an efficient tool for insertional mutagenesis and gene transfer. Phytopathology. 2001; 91(2):173–80. https://doi.org/10.1094/PHYTO.2001.91.2.173 PMID: 18944391 30. Abuodeh RO, Orbach MJ, Mandel MA, Das A, Galgiani JN. Genetic transformation of Coccidioides immi- tis facilitated by Agvobactevium tumefaciens. Journal of Infectious Diseases. 2000; 181(6):2106–10. 31. Opabode JT. Agrobacterium-mediated transformation of plants: emerging factors that influence effi- ciency. Biotechnology and Molecular Biology Reviews. 2006; 1(1):12–20. 32. van der Fits L, Deakin EA, Hoge JHC, Memelink J. The ternary transformation system: constitutive virG on a compatible plasmid dramatically increases Agrobacterium-mediated plant transformation. Plant molecular biology. 2000; 43(4):495–502. https://doi.org/10.1023/a:1006440221718 PMID: 11052201 33. Wu D, Navet N, Liu Y, Uchida J, Tian M. Establishment of a simple and efficient Agrobacterium-medi- ated transformation system for Phytophthora palmivora. BMC microbiology. 2016; 16(1):1–8. https:// doi.org/10.1186/s12866-016-0825-1 PMID: 27599726 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 24 / 27 PLOS PATHOGENSModification of AMT for two oomycete species 34. Walker CA, van West P. Zoospore development in the oomycetes. Fungal biology reviews. 2007; 21 (1):10–8. 35. Ahonsi MO, Banko TJ, Hong C. A simple in-vitro ‘wet-plate’method for mass production of Phytophthora nicotianae zoospores and factors influencing zoospore production. Journal of microbiological methods. 2007; 70(3):557–60. https://doi.org/10.1016/j.mimet.2007.06.012 PMID: 17683817 36. Eye LL, Sneh B, Lockwood J. Factors affecting zoospore production by Phytophthora megasperma var. sojae. Phytopathology. 1978; 68(12):1766. 37. Gelvin SB. Plant proteins involved in Agrobacterium-mediated genetic transformation. Annual review of phytopathology. 2010; 48:45–68. https://doi.org/10.1146/annurev-phyto-080508-081852 PMID: 20337518 38. Djamei A, Pitzschke A, Nakagami H, Rajh I, Hirt H. Trojan horse strategy in Agrobacterium transforma- tion: abusing MAPK defense signaling. Science. 2007; 318(5849):453–6. https://doi.org/10.1126/ science.1148110 PMID: 17947581 39. Hwang H-H, Gelvin SB. Plant proteins that interact with VirB2, the Agrobacterium tumefaciens pilin pro- tein, mediate plant transformation. The Plant Cell. 2004; 16(11):3148–67. https://doi.org/10.1105/tpc. 104.026476 PMID: 15494553 40. 41. 42. 43. 44. Li J, Vaidya M, White C, Vainstein A, Citovsky V, Tzfira T. Involvement of KU80 in T-DNA integration in plant cells. Proceedings of the National Academy of Sciences. 2005; 102(52):19231–6. https://doi.org/ 10.1073/pnas.0506437103 PMID: 16380432 Tenea GN, Spantzel J, Lee L-Y, Zhu Y, Lin K, Johnson SJ, et al. Overexpression of several Arabidopsis histone genes increases Agrobacterium-mediated transformation and transgene expression in plants. The Plant Cell. 2009; 21(10):3350–67. https://doi.org/10.1105/tpc.109.070607 PMID: 19820187 Tzfira T, Vaidya M, Citovsky V. Increasing plant susceptibility to Agrobacterium infection by overexpres- sion of the Arabidopsis nuclear protein VIP1. Proceedings of the National Academy of Sciences. 2002; 99(16):10435–40. https://doi.org/10.1073/pnas.162304099 PMID: 12124400 Tzfira T, Vaidya M, Citovsky V. VIP1, an Arabidopsis protein that interacts with Agrobacterium VirE2, is involved in VirE2 nuclear import and Agrobacterium infectivity. The EMBO Journal. 2001; 20(13):3596– 607. https://doi.org/10.1093/emboj/20.13.3596 PMID: 11432846 Li J, Krichevsky A, Vaidya M, Tzfira T, Citovsky V. Uncoupling of the functions of the Arabidopsis VIP1 protein in transient and stable plant genetic transformation by Agrobacterium. Proceedings of the National Academy of Sciences. 2005; 102(16):5733–8. 45. Vergunst AC, Schrammeijer B, den Dulk-Ras A, de Vlaam CM, Regensburg-Tuïnk TJ, Hooykaas PJ. VirB/D4-dependent protein translocation from Agrobacterium into plant cells. Science. 2000; 290 (5493):979–82. https://doi.org/10.1126/science.290.5493.979 PMID: 11062129 46. Uauy C. Wheat genomics comes of age. Current opinion in plant biology. 2017; 36:142–8. https://doi. org/10.1016/j.pbi.2017.01.007 PMID: 28346895 47. Judelson HS, Michelmore RW. Transient expression of genes in the oomycete Phytophthora infestans using Bremia lactucae regulatory sequences. Current Genetics. 1991; 19(6):453–9. 48. Martı´nez-Cruz J, Romero D, de Vicente A, Pe´ rez-Garcı´a A. Transformation of the cucurbit powdery mil- dew pathogen Podosphaera xanthii by Agrobacterium tumefaciens. New Phytologist. 2017; 213 (4):1961–73. https://doi.org/10.1111/nph.14297 PMID: 27864969 49. 50. Lin L, Ye W, Wu J, Xuan M, Li Y, Gao J, et al. The MADS-box transcription factor PsMAD1 is involved in zoosporogenesis and pathogenesis of Phytophthora sojae. Frontiers in Microbiology. 2018; 9:2259. https://doi.org/10.3389/fmicb.2018.02259 PMID: 30319576 Leesutthiphonchai W, Judelson HS. A MADS-box transcription factor regulates a central step in sporu- lation of the oomycete Phytophthora infestans. Molecular Microbiology. 2018; 110(4):562–75. https:// doi.org/10.1111/mmi.14114 PMID: 30194883 51. Bin Moon S, Lee JM, Kang JG, Lee N-E, Ha D-I, Kim DY, et al. Highly efficient genome editing by CRISPR-Cpf1 using CRISPR RNA with a uridinylate-rich 30-overhang. Nature communications. 2018; 9 (1):1–11. 52. Jiang R, Li J, Tian Z, Du J, Armstrong M, Baker K, et al. Potato late blight field resistance from QTL dPI09c is conferred by the NB-LRR gene R8. Journal of experimental botany. 2018; 69(7):1545–55. https://doi.org/10.1093/jxb/ery021 PMID: 29385612 53. Dong S, Ma W. How to win a tug-of-war: the adaptive evolution of Phytophthora effectors. Current Opin- ion in Plant Biology. 2021; 62:102027. https://doi.org/10.1016/j.pbi.2021.102027 PMID: 33684881 54. Hooykaas PJ, van Heusden GPH, Niu X, Reza Roushan M, Soltani J, Zhang X, et al. Agrobacterium- mediated transformation of yeast and fungi. Agrobacterium Biology: From Basic Science to Biotechnol- ogy. 2018:349–74. https://doi.org/10.1007/82_2018_90 PMID: 29770864 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 25 / 27 PLOS PATHOGENSModification of AMT for two oomycete species 55. Gelvin SB. Agrobacterium-mediated plant transformation: the biology behind the “gene-jockeying” tool. Microbiology and molecular biology reviews. 2003; 67(1):16–37. 56. Kunik T, Tzfira T, Kapulnik Y, Gafni Y, Dingwall C, Citovsky V. Genetic transformation of HeLa cells by Agrobacterium. Proceedings of the National Academy of Sciences. 2001; 98(4):1871–6. https://doi.org/ 10.1073/pnas.98.4.1871 PMID: 11172043 57. Judelson HS, Ah-Fong AM. Progress and challenges in oomycete transformation. Oomycete Genetics and Genomics: Diversity, Interactions, and Research Tools. 2009:435–53. 58. Citovsky V, Kapelnikov A, Oliel S, Zakai N, Rojas MR, Gilbertson RL, et al. Protein interactions involved in nuclear import of the Agrobacterium VirE2 protein in vivo and in vitro. Journal of Biological Chemistry. 2004; 279(28):29528–33. https://doi.org/10.1074/jbc.M403159200 PMID: 15123622 59. Wang K, Shi L, Liang X, Zhao P, Wang W, Liu J, et al. The gene TaWOX5 overcomes genotype depen- dency in wheat genetic transformation. Nature Plants. 2022:1–8. 60. Krishnamurthy K, Suhasini K, Sagare A, Meixner M, De Kathen A, Pickardt T, et al. Agrobacterium mediated transformation of chickpea (Cicer arietinum L.) embryo axes. Plant Cell Reports. 2000; 19 (3):235–40. https://doi.org/10.1007/s002990050005 PMID: 30754901 61. Choi PS, Soh WY, Kim YS, Yoo OJ, Liu JR. Genetic transformation and plant regeneration of water- melon using Agrobacterium tumefaciens. Plant cell reports. 1994; 13(6):344–8. https://doi.org/10.1007/ BF00232634 PMID: 24193834 62. 63. Zhang M, Chen S, Yu X, Vikesland P, Pruden A. Degradation of extracellular genomic, plasmid DNA and specific antibiotic resistance genes by chlorination. Frontiers of Environmental Science & Engineer- ing. 2019; 13(3):1–12. Judelson HS, Tyler BM, Michelmore RW. Transformation of the oomycete pathogen, Phytophthora infestans. Mol Plant-Microbe Interact. 1991; 4:602–7. https://doi.org/10.1094/mpmi-4-602 PMID: 1804404 64. Kohli A, Miro B, Twyman RM. Transgene integration, expression and stability in plants: strategies for improvements. Transgenic crop plants: Springer; 2010. p. 201–37. 65. Fitzgerald AM, Mudge AM, Gleave AP, Plummer KM. Agrobacterium and PEG-mediated transforma- tion of the phytopathogen Venturia inaequalis. Mycological research. 2003; 107(7):803–10. https://doi. org/10.1017/s0953756203008086 PMID: 12967207 66. Schouten HJ, vande Geest H, Papadimitriou S, Bemer M, Schaart JG, Smulders MJ, et al. Re-sequenc- ing transgenic plants revealed rearrangements at T-DNA inserts, and integration of a short T-DNA frag- ment, but no increase of small mutations elsewhere. Plant Cell Reports. 2017; 36:493–504. https://doi. org/10.1007/s00299-017-2098-z PMID: 28155116 67. Savory AI, Grenville-Briggs LJ, Wawra S, van West P, Davidson FA. Auto-aggregation in zoospores of Phytophthora infestans: the cooperative roles of bioconvection and chemotaxis. Journal of the Royal Society Interface. 2014; 11(94):20140017. https://doi.org/10.1098/rsif.2014.0017 PMID: 24598206 68. Oertel B, Jelke E. Formation of multinucleate zoospores inPhytophthora infestans (Mont.) de Bary (Oomycetes, Pythiaceae). Protoplasma. 1986; 135(2):173–9. 69. Shi T-Q, Liu G-N, Ji R-Y, Shi K, Song P, Ren L-J, et al. CRISPR/Cas9-based genome editing of the fila- mentous fungi: the state of the art. Applied microbiology and biotechnology. 2017; 101(20):7435–43. https://doi.org/10.1007/s00253-017-8497-9 PMID: 28887634 70. Skelsey P, Rossing W, Kessel G, Van der Werf W. Scenario approach for assessing the utility of dis- persal information in decision support for aerially spread plant pathogens, applied to Phytophthora infestans. Phytopathology. 2009; 99(7):887–95. https://doi.org/10.1094/PHYTO-99-7-0887 PMID: 19522587 71. Schaeffer SM, Nakata PA. CRISPR/Cas9-mediated genome editing and gene replacement in plants: transitioning from lab to field. Plant Science. 2015; 240:130–42. https://doi.org/10.1016/j.plantsci.2015. 09.011 PMID: 26475194 72. Rietman H, Bijsterbosch G, Cano LM, Lee H-R, Vossen JH, Jacobsen E, et al. Qualitative and quantita- tive late blight resistance in the potato cultivar Sarpo Mira is determined by the perception of five distinct RXLR effectors. Molecular Plant-Microbe Interactions. 2012; 25(7):910–9. https://doi.org/10.1094/ MPMI-01-12-0010-R PMID: 22414442 73. Vossen JH, van Arkel G, Bergervoet M, Jo K-R, Jacobsen E, Visser RG. The Solanum demissum R8 late blight resistance gene is an Sw-5 homologue that has been deployed worldwide in late blight resis- tant varieties. Theoretical and Applied Genetics. 2016; 129(9):1785–96. https://doi.org/10.1007/ s00122-016-2740-0 PMID: 27314264 74. Zhang F, Chen H, Zhang X, Gao C, Huang J, Lu¨ L, et al. Genome analysis of two newly emerged potato late blight isolates sheds light on pathogen adaptation and provides tools for disease management. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 26 / 27 PLOS PATHOGENSModification of AMT for two oomycete species Phytopathology. 2021; 111(1):96–107. https://doi.org/10.1094/PHYTO-05-20-0208-FI PMID: 33026300 75. Cao J, Qiu M, Ye W, Wang Y. Phytophthora sojae Transformation Based on the CRISPR/Cas9 System. Bio-protocol. 2022; 12(6). https://doi.org/10.21769/BioProtoc.4352 PMID: 35434189 76. Ding D, Chen K, Chen Y, Li H, Xie K. Engineering introns to express RNA guides for Cas9-and Cpf1- mediated multiplex genome editing. Molecular plant. 2018; 11(4):542–52. https://doi.org/10.1016/j. molp.2018.02.005 PMID: 29462720 77. Lacroix B, Citovsky V. Extracellular VirB5 enhances T-DNA transfer from Agrobacterium to the host plant. PLoS One. 2011; 6(10):e25578. https://doi.org/10.1371/journal.pone.0025578 PMID: 22028781 78. Wang L, Lacroix B, Guo J, Citovsky V. The Agrobacterium VirE2 effector interacts with multiple mem- bers of the Arabidopsis VIP1 protein family. Molecular plant pathology. 2018; 19(5):1172–83. https:// doi.org/10.1111/mpp.12595 PMID: 28802023 79. Concordet J-P, Haeussler M. CRISPOR: intuitive guide selection for CRISPR/Cas9 genome editing experiments and screens. Nucleic acids research. 2018; 46(W1):W242–W5. https://doi.org/10.1093/ nar/gky354 PMID: 29762716 80. Kim HK, Min S, Song M, Jung S, Choi JW, Kim Y, et al. Deep learning improves prediction of CRISPR– Cpf1 guide RNA activity. Nature biotechnology. 2018; 36(3):239–41. https://doi.org/10.1038/nbt.4061 PMID: 29431740 81. Peng D, Tarleton R. EuPaGDT: a web tool tailored to design CRISPR guide RNAs for eukaryotic patho- gens. Microbial genomics. 2015; 1(4). https://doi.org/10.1099/mgen.0.000033 PMID: 28348817 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011346 April 21, 2023 27 / 27 PLOS PATHOGENS
10.1371_journal.pone.0285697	RESEARCH ARTICLE Evolutionary game study on multi-agent value co-creation of service-oriented digital transformation in the construction industry Shiming Wang1, Hui SuID 1, Qiang Hou2 1 School of Business Administration, Liaoning Technical University, Huludao, China, 2 School of Management, Shenyang University of Technology, Shenyang, China a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 1984723125@qq.com Abstract OPEN ACCESS Citation: Wang S, Su H, Hou Q (2023) Evolutionary game study on multi-agent value co-creation of service-oriented digital transformation in the construction industry. PLoS ONE 18(5): e0285697. https://doi.org/10.1371/journal.pone.0285697 Editor: Simon Grima, University of Malta, MALTA Received: February 2, 2023 Accepted: April 28, 2023 Published: May 16, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0285697 Copyright: © 2023 Wang et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Funding: This work was supported in part by the Basic Scientific Research Project of Education Department of Liaoning Province under grant The service-oriented digital transformation of the construction industry is a development trend of cross-border industrial integration and transformation and upgrading in the digital economy environment, and collaborative value creation among stakeholders is seen as a strategic imperative to promote this process. This study aims to achieve efficient collabora- tive value co-creation and accelerate the digital transformation process of the construction industry by exploring the collaborative strategies and evolution laws of value co-creators in the digital service ecosystem of the construction industry. Based on evolutionary game the- ory and methods, this paper analyzes the evolutionary stability strategies and conditions of each participant in the service-oriented value chain at different stages of the digital transfor- mation of the construction industry. It is found that with the improvement of the level of digi- talization, the degree of cooperation among game players continues to increase until a stable state of full cooperation is achieved. The initial willingness of the game players to cooperate accelerates the speed of the system’s evolution to the stable state of full coopera- tion in the middle stage of digital transformation. Additionally, the improvement of the con- struction process digitalization level can subvert the evolution result of full non-coordination caused by a low initial willingness to cooperate. The research conclusions and correspond- ing countermeasures and suggestions can provide a strategic reference for the service-ori- ented digital transformation of the construction industry. 1. Introduction As the pillar industry of the national economy, the construction industry continues to face the leading problem of “being large but not strong”. Under the traditional project construction mode, the coordination among owners, design units, and construction units has been recog- nized as a hindrance to the development of the construction industry. With the arrival of Con- struction 4.0, emerging digital technologies are inevitable in the development of the construction industry to achieve the comprehensive coordination of approval and decision- making, planning and design, construction operation and maintenance. Digital technology PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 1 / 23 PLOS ONELJKMR20220709 and in part by the Social Science Fund of Liaoning Province under grant L21BGL027. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Evolutionary game study on multi-agent value co-creation can integrate customer value into the process of enterprise product and value creation, which is a prominent feature of digital transformation and an important breakthrough for construc- tion enterprises in achieving the integration of digitalization and service. The 14th Five-Year Plan for the Development of the Digital Economy clearly indicated that digital services are a new trend in the development of China’s digital economy and established the accelerated inte- gration of industrial digital transformation and productive services as the goal during the 14th Five-Year Plan period. Driven by both digital technology and policy, the service-oriented digi- tal transformation of the construction industry has not only injected new vitality into the tradi- tional construction industry but also provided new opportunities to increase the value of various stakeholders by transforming from product construction to service construction with the help of digital technology. This has become the key to the high-quality development of the construction industry. Specifically, in the construction industry, all participants take the needs of owners (customers) as the starting point in the construction process and cultivate a new ecology of the construction industry through the integration of “the construction gene” and “digital gene”. Construction enterprises are the core subject of industrial operations, and digi- tal solution suppliers are the digital enablers. The two subjects work together to promote the full participation of customers by adding digital derivative services in the construction process and integrating decentralized heterogeneous resources to achieve the optimization of the entire production process. However, it is difficult for the independent digital investment of each subject to give play to the advantages of digital technology, and collaboration and com- munication among stakeholders are particularly important in resource integration, operation process adjustment, and other aspects [1, 2]. Xie et al. [3] pointed out that the social facts hid- den behind digital transformation and the essence of digital resources are cooperation and sharing, and collaboration is crucial for enterprises in the transformation to stand and develop in the turbulent market environment. In addition, after the impact of COVID-19, the tradi- tional production mode and development path of construction enterprises have been severely damaged. It is urgent to cooperate to obtain and integrate more external resources and revital- ize the industry market through the new business model. Obviously, exploring the evolution- ary laws of collaborative strategies of value co-creators in the service-oriented digital transformation of the construction industry has become a necessary entry point for promoting the efficient transformation of the construction industry and restoring market vitality. Different from previous studies, this article takes the digital service ecosystem of the con- struction industry as the research object, focusing on the collaborative behavior among value co-creators in the value chain, and incorporating digital level factors that have not been addressed in existing research into the research scope, to preliminary explore the integration of digital and service-oriented in the construction industry. Considering that there are multi- ple strategic combinations of value co-creation behaviors among various entities in realistic sit- uations, it may not be a coordinated state of reaching consensus and cooperation. From the perspective of value co-creation, this study constructs an evolutionary game model for three parties, namely, construction enterprises, digital solution suppliers, and customers. It analyzes the dynamic evolution process of strategy selection of each participant in the service-oriented value supply chain and simulates the evolution results of value co-creation collaborative strat- egy selection under the change of factors. It has certain practical guiding significance for grasp- ing the essential characteristics and operation mechanism of value co-creation in the digital transformation process of the construction industry, and scientifically and accurately imple- menting countermeasures based on it to promote the high-quality development of the con- struction industry. Section 2 reviews the previous studies on service-dominant logic and digital transformation respectively, and summarizes the research gaps. Section 3 analyzes the structure of the digital PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 2 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation service ecosystem in the construction industry and defines the collaborative mechanism and role orientation of value co-creation of the multi-agent in the service-oriented digital transfor- mation value chain, and constructs an evolutionary game model of value co-creation in the digital service ecosystem of the construction industry. Section 4 presents the stability analysis results of the evolutionary game model. Section 5 through digital simulation, reveals the dynamic evolutionary law of the game among value co-creators in the digital service ecosystem of the construction industry and discusses the role mechanism of the digital level in the evolu- tionary stability strategy. Section 6 concludes with a summary of the main results. 2. Literature review In the relevant research on value co-creation, a service-oriented logic occupies a dominant position. Vargo and Lusch [4] understood value co-creation from the perspective of a service- oriented logic and believed that value is defined by consumers and co-created by the joint pro- duction process of enterprises and consumers. This concept shows that the interaction between enterprises and consumers is not only a simple business transaction but also a form of process-oriented cooperation. The binding relationship between enterprises and consumers in the early, middle, and late stages of production is not only an effective means for enterprises to capture competitive advantages but also improves consumers’ satisfaction with service quality [5, 6]. Wang et al. [7] analyzed the problems in the Engineering-Procurement-Construction (EPC) project management of the construction industry from the perspective of a service-ori- ented logic and proposed a specific path to overcome the dilemma of low levels of trust between the contracting parties through value co-creation. With the deepening of the develop- ment of a service-oriented logic to service ecosystem theory, the focus of value co-creation has gradually expanded from the early binary interaction between enterprises and consumers to the diversified interaction within the ecological network system [8]. The service ecosystem combines the research perspective of the ecosystem and takes the service-oriented logic theory as the core. It is a dynamic system with multi-agent service exchange, value co-creation, and self-regulation functions [9]. The collaborative strategy of participants is the key to maintain- ing the stability and sustainable development of the service ecosystem [10]. Pacheco et al. [11] pointed out that the effective tools for studying cooperative behavior are evolutionary game method and replication dynamics mechanism. Throughout the literature on the game relation- ship between participants in the service ecosystem, most studies focus on the multilevel inter- action of participants. The service ecosystem carries out resource integration and service exchange at the micro, meso, and macro levels. At the micro level, the binary interaction between enterprises and consumers has been a focus. For example, Sun et al. [12] built a sym- biotic evolutionary model of three value co-creation units, namely, third-party service plat- forms, service providers, and service consumers, and analyzed their symbiotic conditions to maintain stability in the service ecosystem. At the meso level, the interaction among enter- prises has been emphasized. For example, Zhang et al. [13] further considered the value chain system composed of manufacturers, suppliers, logistics providers, and third-party manufactur- ing platforms and analyzed its trend toward collaborative evolution based on game theory. The macro level focuses on the introduction of social participants, including the state, culture, market, etc., and covers the micro and meso levels. For example, Li and Li [14] brought the government, enterprises, and consumers into the same system to explore the impact of govern- ment subsidy policies and consumer purchase behavior on the intelligent transformation deci- sion-making of manufacturing enterprises. These studies further illustrate that game theory provides ideas for analyzing the logic behind the collaborative behavior of value co-creators. Specifically, game theory can guide the decision-making process of participants, while PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 3 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation replication dynamics can reveal the evolution of collaborative strategies in the process of digi- tal transformation. It has been noted that the digital transformation strategy should not only consider the prin- ciple and value of the coordination between society and the economy but also pay attention to the changes in the creation mechanism of products and services [15]. The development and upgrading of digital technology have reshaped the interaction boundary among the subjects of the service ecosystem and has brought more stakeholders into the process of value creation with low-cost communication and connections around customer needs, promoting the collab- orative sharing of heterogeneous resources among multiple subjects [16–18]. It shows that the rapid development of digital technologies such as big data and the Internet of Things(IoT) has not only improved the complexity of the value co-creation network but also reshaped the value co-creation mode of the subjects of the service ecosystem, leading the service ecosystem to address the challenge of transforming to the digital service ecosystem. Kolagar et al. [19] sys- tematically reviewed the trigger factors, driving factors, transformation stages, and the activi- ties at each stage of the transformation to the digital service ecosystem and provided a specific research agenda for the different stages. At the same time, how to maintain the sustainable and stable development of the digital service ecosystem has become the focus of academic atten- tion. Scholars generally believe that participants can promote a series of value co-creation activities to achieve the feasibility and stability of the system by integrating collaborative resources based on a consistent structure and seeking cooperation based on win-win thinking [20]. Goudarzi et al. [21] studied the behavior strategies of participants in the service ecosys- tem in the digital context, constructed cooperative and noncooperative game models between service providers and consumers in the cloud manufacturing service ecosystem, and found that the cooperative game is a win-win for both parties by comparing the values of service quality and the profit values of participants. The above studies are all focused on the manufacturing industry, while in contrast to the construction industry, digital technologies represented by BIM, the Internet of Things, and cloud computing are changing the construction process and business model of the traditional construction industry, driven by the dual drive of Industry 4.0 and industry digitization. For example, cloud computing enables architectural designers and workers to work and operate remotely, quickly optimizing and restructuring the construction supply chain [22]; In order to further strengthen information models such as BIM, CPS (Cyber Physical System) has been proposed and widely used for the installation control of prefabricated formwork, risk control of blind plate cranes, and optimization of safety management during the construction process [23]. In addition, the service-oriented innovative economy is becoming a new trend in the construction industry. In the service ecosystem, the service-oriented construction industry realizes value co-creation through the reconstruction of the value chain among heterogeneous entities, and the cooperation between the construction enterprises and the upstream and downstream enterprises of the value chain can optimize the efficiency of resource allocation [24]. Therefore, the collaborative behavior of the various entities in the value chain is very important for the value co-creation of the service ecosystem in the construction industry. The emergence of digital technology provides a convenient and effective new means for the coop- eration between various subjects. Prebanić and Vukomanović [25] have found that digital technologies such as BIM and virtual reality can effectively attract owners to participate in the design and development process and improve the communication efficiency and collaboration level between owners and construction stakeholders by systematic review of stakeholder man- agement practices in the construction industry. At the same time, several recent studies also focus on the interaction between digitalization and service in the construction industry and propose that digital technology can promote construction enterprises to provide services for PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 4 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation end customers, such as providing customers with preventive services and more value through data collection. In contrast, the creativity and inspiration of digital technology innovation can also be obtained from market solutions by introducing service-oriented models [26, 27]. In summary, the value co-creation of the service ecosystem in the digital context has attracted extensive attention from the academic community. However, most studies focus on the manufacturing industry. Research on the integration of digitalization and services in the construction industry mainly focuses on promoting the service efficiency of construction enterprises and the entire service ecosystem through digitalization. Few studies focus on the collaboration of stakeholders in the value chain under the digital context. There is still room for breakthroughs in the integration research of digitalization and service in the construction industry. Therefore, in order to make up for the shortcomings of existing research, this paper proposes an evolutionary game model for value co-creation of the digital service ecosystem in the construction industry, exploring the evolutionary process and dynamic development trend of collaborative behavior among various participants in the process of digital service in the construction industry. 3. Multi-agent value co-creation evolutionary game model 3.1 Role recognition With reference to the interactive model of participants in the digital transformation of the con- struction industry by Chen et al. [28], this paper proposes that the digital service ecosystem of the construction industry is a complex network system consisting of a service-oriented value chain at the core of the system and external auxiliary institutions (including universities and scientific research institutions, governments, financial institutions, and industry network plat- forms). The internal value chain, supported by external auxiliary institutions, meets the needs of customers through a clear division of labor and cooperation, integration, and sharing of data resources and aims to achieve value co-creation through systematic collaborative innova- tion [29]. The digital service ecosystem of the construction industry is shown in Fig 1. Fig 1 shows that the main body of value co-creation in the service-oriented digital transfor- mation of the construction industry consists of construction enterprises, digital solution sup- pliers, and customers. Construction enterprises are the key node and leading force of the whole value co-creation system. They use digital technology, digital products, and digital man- agement in the whole life cycle of project construction to transform customers’ needs into enti- ties. Enterprises are not only the demanders of digital transformation but also the implementation terminals, driving and leading the digital transformation and coordinated development of the whole system. Overall, the construction industry is relatively decentralized, which also leads to the fragmentation of data storage in all stages of construction. The intelli- gent construction collaborative management platform based on project construction can effec- tively break the status of isolated data islands. As the developers of digital technology integration, digital software, and hardware products and services, digital solution suppliers are the key node in the value chain. Customers are the source of value creation for construction enterprises and digital solution suppliers in the service-oriented value chain, and their person- alized needs and use of information feedback are the innovative force of the whole value co- creation system. 3.2 Tripartite game analysis and research hypothesis 1. Collaboration between construction enterprises and customers In the context of digital transformation, customers are not the end of transactions but the nodes that create value for enterprises. With the help of digital technology, construction PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 5 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation Fig 1. Digital service ecosystem in the construction industry. https://doi.org/10.1371/journal.pone.0285697.g001 enterprises enable customers to participate in all aspects of the whole life cycle of construction projects from planning, design, and construction to operation and maintenance and create platforms and opportunities for communication and feedback with customers. Through the feedback data provided by customers, construction enterprises can more clearly and directly guide how to meet their needs, thereby enhancing customer perceived value, achieving the goal of maximizing corporate profits, and expanding the profit space and scope. In addition, customers can combine the products or services provided by construction enterprises with their resources to stimulate their creative needs and create value through interaction with con- struction enterprises. However, customers participating in the collaborative process must bear a certain amount of energy and opportunity costs. If customers do not participate in collabora- tion, their own perceived value will be reduced and collaborative benefits will be lost. Mean- while, construction enterprises will suffer information losses. Based on the above analysis, the following theoretical hypotheses are proposed: H1: Under the supervision of construction enterprises, customers will ultimately choose to participate in the collaboration. 2. Collaboration between construction enterprises and digital solution suppliers The collaboration between construction enterprises and digital solution suppliers can improve the ability to collaborate and integrate resources between value co-creators to achieve a rapid response to changes, an accurate connection between supply and demand, a flexible allocation of resources, and more innovation possibilities. The whole value chain uses digital technology to realize efficient connections and optimize research and development design, PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 6 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation construction, operation management, marketing, and after-sales service through a data-ori- ented service model. At the same time, it can also deeply endow the wisdom and value of each element in the operation process of construction enterprises and promote the optimization of the whole service-oriented operation process and operational efficiency. In addition, according to resource-based theory, the integration and full utilization of resources is the main source for enterprises to maintain competitive advantages [30]. In the process of service-oriented digital transformation, service quality is an important factor affecting the core competitiveness of construction enterprises. Therefore, construction enterprises will also outsource their non- core professional digital after-sales service business to digital solution suppliers to provide more professional services. However, after signing the service outsourcing agreement, digital solution suppliers are likely to damage the interests of construction enterprises and customers by reducing the quality of digital software and hardware facilities, leaking information, and lowering the service level to maximize their own interests. To prevent these risks, construction enterprises need to impose certain regulatory measures on digital solution suppliers to improve their service quality. In this process, construction enterprises have to bear the cost of collaborative supervision, and digital solution suppliers have to bear the cost of collaborative innovation. If construction enterprises do not implement collaborative supervision strategies, due to the lack of coordination and communication with digital solution suppliers and cus- tomers, as well as the reduction of service quality of digital solution suppliers, construction enterprises and digital solution providers will lose some customers. If digital solution providers do not participate in collaborative innovation, construction enterprises may incur losses due to the lack of digital services support. Based on the above analysis, the following theoretical hypotheses are proposed: H2: Whether digital solution suppliers participate in collaborative innovation depends on the strategic choices of construction enterprises. 3. Collaboration between digital solution suppliers and customers The collaboration between digital solution suppliers and customers is mainly achieved through feedback data provided by construction enterprises or direct feedback data from cus- tomers. At this point, the digital platform provides a bridge for digital solution suppliers to interact with customers, which helps digital solution suppliers quickly capture current market needs and personalized customer needs and improve their own and even the entire value chain’s service innovation capabilities. In addition, digital solution suppliers can realize data access in the product after-sales service link with the help of digital platforms. Customers con- tinuously provide feedback data to digital solution suppliers or construction enterprises in direct and indirect ways during the full cycle of services provided by digital solution suppliers, enhancing their ability to create value. If digital solution suppliers do not participate in collab- orative innovation, customers will suffer certain losses due to the lack of support from digital services and the decrease in service quality. Similarly, if customers choose not to participate in collaborative innovation, digital solution providers will also lose the motivation for innovation due to a lack of information feedback. Based on the above analysis, the following theoretical hypotheses are proposed: H3: The customers’ strategic choice is synchronized with the digital solution suppliers. In addition, the level of digitalization affects the benefits and costs of construction enter- prises and digital solution suppliers in the game process. The level of digitalization is reflected by the level of enterprise data collection and analysis, the level of the platform, and the digitali- zation level of the construction process [31]. First, data are an important information asset for enterprises to understand and develop demand markets and an important medium for PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 7 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation enterprises to communicate with customers and enter the ecosystem. Second, in the platform stage, with the help of the data sharing and connection of the construction end, the sales end, and the service end, the digital platform has realized the close connection among the value cre- ation entities, greatly reducing the cost of cooperation and transactions among the entities. Finally, using digital technology to improve the efficiency of operations and construction can provide a new driving force for the digital transformation of construction enterprises and the efficient development of the digital service ecosystem of the construction industry. Based on the above analysis, the following theoretical hypotheses are proposed: H4: The digital level can promote the collaborative value co-creation of game players. 3.3 Basic assumptions of the model Based on the above analysis of the synergy mechanism among value co-creators in the service- oriented digital transformation of the construction industry, construction enterprises, digital solution suppliers, and customers, as the main players of value co-creation, are the players in the game. They are all limited rational individuals with the ability to learn and adapt to dynamic environmental changes and adjust and optimize their strategies from their own inter- ests. The strategy set of construction enterprises is {collaborative supervision, no collaborative supervision}, and the probability that construction enterprises choose the strategy of collabora- tive supervision at time t is x; then, the probability that they choose the strategy of no collabo- rative supervision is 1−x. The strategy set of digital solution suppliers and customers is {collaborate, not collaborate}. Where the probability of digital solution suppliers and custom- ers choosing the “participate in collaboration” strategy at time t is y and z, respectively, then the probability of choosing the “not participate in collaboration” strategy is 1−y and 1−z, respectively, x,y,z2[0,1], and both are functions of time t. The initial revenues of construction enterprises, digital solution suppliers, and customers are I1, I2 and I3, respectively. When all three parties participate in value co-creation, additional synergy benefits will be created, and the synergy benefits will be distributed proportionally among the participants [30]. The distribution coefficient of the collaborative income of con- struction enterprises is α, that of digital solution suppliers is β, that of customers is γ = 1−α−β, and α,β,γ2[0,1]. When construction enterprises, digital solution providers, and customers choose to participate in value co-creation, it is assumed that the collaboration cost that the three parties need to bear is C, and the collaboration cost will be distributed proportionally among the participants [32]. The collaborative cost-sharing coefficient of the construction enterprises is λ, the collaborative cost-sharing coefficient of the digital solution suppliers is η, the collaborative cost-sharing coefficient of the customers is σ = 1−λ−η, and λ,η,σ2[0,1]. The losses caused to construction enterprises and digital solution suppliers due to their failure to choose collaborative supervision are L1. The losses caused to construction enterprises and cus- tomers by digital solution suppliers choosing not to participate in collaborative innovation are L2. The losses caused to construction enterprises and digital solution suppliers by customers choosing not to participate in collaborative innovation are L3. The coefficients of the three fac- tors reflecting the level of digitalization affect the benefits and costs of participants. The data collection and analysis coefficient is expressed as μ, the platform coefficient is expressed as ν, and the digitization of the construction process coefficient is expressed as ω, and μ,ν,ω2[0,1]. 3.4 Construction of the evolutionary model Based on the above assumptions, the evolutionary game income matrix of construction enter- prises, digital solution suppliers, and customers can be constructed as shown in Table 1. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 8 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation Table 1. The payoff matrix of the tripartite game. The strategies of subjects Customers participate in collaboration (z) Construction enterprises Collaborative supervision (x) Not collaborative supervision (1−x) Digital solution suppliers participate in collaboration (y) A1 ¼ ð1 þ oÞðI1 þ aDIÞ (cid:0) ð1 (cid:0) m (cid:0) nÞlC B1 ¼ ð1 þ n þ oÞðI2 þ bDIÞ (cid:0) ZC Digital solution suppliers don’t participate in collaboration (1−y) A2 ¼ ð1 þ oÞI1 (cid:0) ð1 (cid:0) m (cid:0) nÞlC (cid:0) L2 B5 = I2 C1 ¼ ½I3 þ ð1 (cid:0) a (cid:0) bÞDI� (cid:0) ð1 (cid:0) m (cid:0) nÞð1 (cid:0) l (cid:0) ZÞC C3 ¼ I3 (cid:0) ð1 (cid:0) m (cid:0) nÞð1 (cid:0) l (cid:0) ZÞC (cid:0) L2 A6 ¼ ð1 þ oÞI1 (cid:0) L1 (cid:0) L2 B6 ¼ I2 (cid:0) L1 C4 ¼ I3 (cid:0) ð1 (cid:0) l (cid:0) ZÞC (cid:0) L2 A5 ¼ ð1 þ oÞI1 (cid:0) L1 B2 ¼ ð1 þ oÞI2 (cid:0) ZC (cid:0) L1 C2 ¼ I3 (cid:0) ð1 (cid:0) l (cid:0) ZÞC The strategies of subjects Customers don’t participate in collaboration (1−z) Construction enterprises Collaborative supervision (x) Not collaborative supervision (1−x) https://doi.org/10.1371/journal.pone.0285697.t001 Digital solution suppliers participate in collaboration (y) A3 ¼ ð1 þ oÞI1 (cid:0) lC (cid:0) L3 B3 ¼ ð1 þ n þ oÞI2 (cid:0) ZC (cid:0) L3 C5 = 0 A7 ¼ ð1 þ oÞI1 (cid:0) L1 (cid:0) L3 B4 ¼ ð1 þ oÞI2 (cid:0) ZC (cid:0) L1 (cid:0) L3 C6 = 0 Digital solution suppliers don’t participate in collaboration (1−y) A4 ¼ ð1 þ oÞI1 (cid:0) lC (cid:0) L2 (cid:0) L3 B7 ¼ I2 (cid:0) L3 C7 = 0 A8 ¼ ð1 þ oÞI1 (cid:0) L1 (cid:0) L2 (cid:0) L3 B8 ¼ I2 (cid:0) L1 (cid:0) L3 C8 = 0 4. Stability analysis of the evolutionary game model 4.1 Model equilibrium analysis According to the income matrix of the evolutionary game, the expected income E1(x) or E1(1 −x) of construction enterprises when they choose the “cooperative supervision” or “no cooper- ative supervision” strategy are, respectively: E1ðxÞ ¼ yzA1 þ zð1 (cid:0) yÞA2 þ yð1 (cid:0) zÞA3 þ ð1 (cid:0) yÞð1 (cid:0) zÞA4 ¼ yzð1 þ oÞaDI þ yL2 þ z½ðm þ nÞlC þ L3� þ ½ð1 þ oÞI1 (cid:0) lC (cid:0) L2 (cid:0) L3� E1ð1 (cid:0) xÞ ¼ yzA5 þ zð1 (cid:0) yÞA6 þ yð1 (cid:0) zÞA7 þ ð1 (cid:0) yÞð1 (cid:0) zÞA8 ¼ yL2 þ zL3 þ ½ð1 þ oÞI1 (cid:0) L1 (cid:0) L2 (cid:0) L3� The average expected income �E1 of construction enterprises is: �E1 ¼ xE1ðxÞ þ ð1 (cid:0) xÞE1ð1 (cid:0) xÞ According to Formulas (1), (2), and (3), we can further obtain the replicator dynamics equation of the strategy selection of construction enterprises as follows: FðxÞ ¼ dx dt ¼ x½E1ðxÞ (cid:0) �E1� ¼ xð1 (cid:0) xÞ½E1ðxÞ (cid:0) E1ð1 (cid:0) xÞ� ¼ xð1 (cid:0) xÞ½yzð1 þ oÞaDI þ zðm þ nÞlC (cid:0) lC þ L1� Similarly, the expected income E2(y) or E2(1−y) of digital solution suppliers when they choose “participating in collaboration” or “not participating in collaboration” strategy are, ð1Þ ð2Þ ð3Þ ð4Þ PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 9 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation respectively: E2ðyÞ ¼ xzB1 þ zð1 (cid:0) xÞB2 þ xð1 (cid:0) zÞB3 þ ð1 (cid:0) xÞð1 (cid:0) zÞB4 ¼ xzð1 þ n þ oÞbDI þ xðnI2 þ L1Þ þ zL3 þ ½ð1 þ oÞI2 (cid:0) ZC (cid:0) L1 (cid:0) L3� E2ð1 (cid:0) yÞ ¼ xzB5 þ zð1 (cid:0) xÞB6 þ xð1 (cid:0) zÞB7 þ ð1 (cid:0) xÞð1 (cid:0) zÞB8 ¼ xL1 þ zL3 þ ðI2 (cid:0) L1 (cid:0) L3Þ The average expected income �E2 of digital solution suppliers is: �E2 ¼ yE2ðyÞ þ ð1 (cid:0) yÞE2ð1 (cid:0) yÞ According to Formulas (5), (6), and (7), we can further obtain the replicator dynamics equation of the strategy selection of digital solution suppliers as follows: FðyÞ ¼ dy dt ¼ y½E2ðyÞ (cid:0) �E2� ¼ yð1 (cid:0) yÞ½E2ðyÞ (cid:0) E2ð1 (cid:0) yÞ� ¼ yð1 (cid:0) yÞ½xzð1 þ v þ wÞbDI þ xvI2 þ oI2 (cid:0) ZC� ð5Þ ð6Þ ð7Þ ð8Þ Similarly, the expected income E3(z) or E3(1−z) of customers when they choose “participat- ing in collaboration” or “not participating in collaboration” strategy are, respectively: E3ðzÞ ¼ xyC1 þ yð1 (cid:0) xÞC2 þ xð1 (cid:0) yÞC3 þ ð1 (cid:0) xÞð1 (cid:0) yÞC4 ¼ xyð1 (cid:0) a (cid:0) bÞDI þ xðm þ nÞð1 (cid:0) l (cid:0) ZÞC þ yL2 þ ½I3(cid:0) ð1 (cid:0) l (cid:0) ZÞC (cid:0) L2� E3ð1 (cid:0) zÞ ¼ xyC5 þ yð1 (cid:0) xÞC6 þ xð1 (cid:0) yÞC7 þ ð1 (cid:0) xÞð1 (cid:0) yÞC8 ¼ 0 The average expected income �E3 of customers is: �E3 ¼ zE3ðzÞ þ ð1 (cid:0) zÞE3ð1 (cid:0) zÞ ð9Þ ð10Þ ð11Þ According to Formulas (9), (10), and (11), we can further obtain the replicator dynamics equation of the strategy selection of customers as follows: FðzÞ ¼ dz dt ¼ z½E3ðzÞ (cid:0) �E3 � ¼ zð1 (cid:0) zÞ½E3ðzÞ (cid:0) E3ð1 (cid:0) zÞ� ¼ zð1 (cid:0) zÞfxyð1 (cid:0) a (cid:0) bÞDI þ xðm þ nÞð1 (cid:0) l (cid:0) ZÞC þ yL2 þ ½I3 (cid:0) ð1 (cid:0) l (cid:0) ZÞC (cid:0) L2�g ð12Þ The three replicator dynamic equations F(x), F(y), and F(z) can be combined to obtain a 3D dynamic system of the dynamic evolution for construction enterprises, digital solution suppli- ers, and customers. Let F(x) = 0, F(y) = 0, and F(z) = 0, the eight local equilibrium points of the dynamic system are (0,0,0), (1,0,0), (0,1,0), (0,1,0), (0,0,1), (1,1,0), (1,0,1), (0,1,1), (1,1,1), and (1,1,1), respectively, which form the equilibrium solution domain N of the three-way evolu- tionary game. That is, N ¼ fðx; y; zÞj0 � x � 1; 0 � y � 1; 0 � z � 1g, and there is also a mixed-strategy solution (x,y,z) in the region. Only the pure strategy Nash equilibrium point is asymptotically stable in the three-way evolutionary game, so it is only necessary to discuss the stability of the eight pure strategy equilibrium points. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 10 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation 4.2 Stability analysis of the equilibrium point The above equilibrium point is not completely an evolutionary stability strategy for replicating a dynamic system. It is necessary to further discuss the stability of the system equilibrium point by using the Jacobian matrix local stability analysis method proposed by Friedman [33]. The Jacobian matrix can be constructed according to the partial derivative of the three-dimen- sional dynamic system with respect to x,y,z. 0 B B B B B B B B @ @FðxÞ @x @FðyÞ @x @FðzÞ @x @FðxÞ @y @FðyÞ @y @FðzÞ @y 1 C C C C C C C C A @FðxÞ @z @FðyÞ @z @FðzÞ @z J ¼ 0 B @ ¼ a11 a21 a31 a12 a22 a32 1 C A a13 a23 a33 ð13Þ where a11 ¼ ð1 (cid:0) 2xÞ½L1 (cid:0) lC þ zðm þ nÞlC þ yzð1 þ oÞaDI� a12 ¼ (cid:0) xðx (cid:0) 1Þzð1 þ oÞaDI a13 ¼ (cid:0) xðx (cid:0) 1Þ½ðm þ nÞlC þ yð1 þ oÞaDI� a21 ¼ (cid:0) yðy (cid:0) 1Þ½nI2 þ zð1 þ n þ oÞbDI� a22 ¼ ð1 (cid:0) 2yÞ½oI2 (cid:0) ZC þ xnI2 þ xzð1 þ n þ oÞbDI� a23 ¼ (cid:0) xyðy (cid:0) 1Þð1 þ n þ oÞbDI a31 ¼ zðz (cid:0) 1Þ½ðm þ nÞðl þ Z (cid:0) 1ÞC (cid:0) yð1 (cid:0) a (cid:0) bÞDI� a32 ¼ (cid:0) zðz (cid:0) 1Þ½L2 þ xð1 (cid:0) a (cid:0) bÞDI� a33 ¼ ð1 (cid:0) 2zÞ½I3 (cid:0) L2 þ yL2 þ ðl þ Z (cid:0) 1ÞC (cid:0) xðm þ nÞðl þ Z (cid:0) 1ÞC þ xyð1 (cid:0) a (cid:0) bÞDI� The eigenvalues of the Jacobian matrix corresponding to the eight equilibrium points are calculated, as shown in Table 2. According to Lyapunov’s method, the stability of the equilibrium point of the system can be judged by the sign of the eigenvalue of the Jacobian matrix. Only when all the eigenvalues of the Jacobian matrix are negative is the equilibrium point the evolutionary stability point ESS of the system [33]. In this paper, the life cycle of the digital transformation of construction enterprises is divided into three stages according to industry cycle theory: early stage, middle Table 2. Each equilibrium point corresponds to the eigenvalue of the Jacobian matrix. Equilibrium point λ1 λ2 (0,0,0) (1,0,0) (0,1,0) (0,0,1) (1,1,0) (1,0,1) (0,1,1) (1,1,1) (x,y,z) L1−λC −(L1−λC) L1−λC L1 (cid:0) ð1 (cid:0) m (cid:0) nÞlC −(L1−λC) (cid:0) ½L1 (cid:0) ð1 (cid:0) m (cid:0) nÞlC� L1 (cid:0) ð1 (cid:0) m (cid:0) nÞlC þ ð1 þ oÞaDI (cid:0) ½L1 (cid:0) ð1 (cid:0) m (cid:0) nÞlC þ ð1 þ oÞaDI� λ1* ωI2−ηC ðo þ nÞI2 (cid:0) ZC −(ωI2−ηC) ωI2−ηC (cid:0) ½ðo þ nÞI2 (cid:0) ZC� ðo þ nÞI2 (cid:0) ZC þ ð1 þ n þ oÞbDI (cid:0) ðoI2 (cid:0) ZCÞ (cid:0) ½ðo þ nÞI2 (cid:0) ZC þ ð1 þ n þ oÞbDI� λ2* https://doi.org/10.1371/journal.pone.0285697.t002 λ3 I3 (cid:0) L2 þ ðl þ Z (cid:0) 1ÞC I3 (cid:0) L2 þ ð1 (cid:0) m (cid:0) nÞðl þ Z (cid:0) 1ÞC I3 þ ðl þ Z (cid:0) 1ÞC (cid:0) ½I3 (cid:0) L2 þ ðl þ Z (cid:0) 1ÞC� I3 þ ð1 (cid:0) m (cid:0) nÞðl þ Z (cid:0) 1ÞC þ ð1 (cid:0) a (cid:0) bÞDI (cid:0) ½I3 (cid:0) L2 þ ð1 (cid:0) m (cid:0) nÞðl þ Z (cid:0) 1ÞC� (cid:0) ½I3 þ ðl þ Z (cid:0) 1ÞC� (cid:0) ½I3 þ ð1 (cid:0) m (cid:0) nÞðl þ Z (cid:0) 1ÞC þ ð1 (cid:0) a (cid:0) bÞDI� λ3* PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 11 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation Table 3. Each equilibrium point corresponds to the eigenvalue symbol judgment. Equilibrium point Case 1 Case 2 Case 3 (0,0,0) (1,0,0) (0,1,0) (0,0,1) (1,1,0) (1,0,1) (0,1,1) (1,1,1) −−− +−− −+� −−+ ++� +−+ �+� �+� −−− +++ −+� +−+ +−+ −+− ++� −−− +++ −++ +−+ ++− −−+ −+− +−− −−− https://doi.org/10.1371/journal.pone.0285697.t003 stage, and mature stage. The collection and analysis of data and the research and development of digital technology are the basis for the digital transformation of enterprises. Enterprises need to invest in digital technology, which is the main task in the early stage of digital transfor- mation. With the accumulation of data, the value of a large amount of data needs to be devel- oped. By building a digital platform to fully activate data and starting to establish contacts with other companies in data sharing to develop value networks, services across company bound- aries will be provided to create value for customers, and digital transformation will enter the middle stage [34]. The continuous expansion of customer groups will lead to a diversity of cus- tomer needs, which will be met by various digital solutions. Digital services will enter the stage of mass customization, and digital transformation will enter the mature stage [35]. Combined with the analysis of the above stages and the Jacobian matrix eigenvalue expression corre- sponding to each equilibrium point in Table 2, the stability analysis of each equilibrium point is carried out under the following three conditions, and the specific analysis results are shown in Table 3. Case 1: Construction enterprises are in the early stage of digital transformation, and the level of digitalization is low. They need to invest considerable manpower, material resources, time, and capital to learn and absorb digital technology and knowledge. Therefore, the cost of implementing the whole value chain collaborative supervision for construction enterprises is high, and the digital profits and benefits are not significant. When ð1 (cid:0) m (cid:0) nÞlC > L1; ðn þ oÞI2 þ ð1 þ n þ oÞbDI < ZC; I3þ ð1 (cid:0) a (cid:0) bÞDI < ð1 (cid:0) m (cid:0) nÞð1 (cid:0) l (cid:0) ZÞC þ L2, it is found that the equilibrium point of evolutionary stability can only be (0,0,0). In the early stage of digital transformation, for construction enterprises, digital solution suppliers, and cus- tomers, when at least one party chooses collaborative supervision or to collaboration, the coor- dination costs paid by the entities involved in value co-creation are greater than the benefits obtained. To achieve their own goal of maximizing benefits, the three parties will give up the cooperation strategy with the growth of the number of evolution periods, and the value co-cre- ation strategy of the three parties will finally evolve into no collaborative supervision, no col- laboration, or no collaboration, respectively. Case 2: When the digital transformation of construction enterprises has been ongoing for a period of time, the enterprises have initially mastered a certain amount of digital resources and technologies. The digital level is in the middle, the cost of collaborative supervision is reduced, and the digital level coefficient is increased. In this case, although λC>L1, ð1 (cid:0) m (cid:0) nÞlC < L1, ðn þ oÞI2 > ZC > oI2; ð1 (cid:0) l (cid:0) ZÞC þ L2 > I3 > ð1 (cid:0) m (cid:0) nÞð1 (cid:0) l (cid:0) ZÞC þ L2, two equi- librium points (0,0,0) and (1,1,1) can be identified to achieve steady equilibrium. It shows that with the growth of evolution periods, game players will eventually choose to participate in value co-creation or not to achieve the goal of maximizing benefits, and the system will eventu- ally evolve into a state of complete non-collaboration or complete collaboration. Compared PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 12 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation with Case 1, it is found that the improvement of the digital level encourages game players to choose the strategy of participating in value co-creation. Case 3: The digital transformation of construction enterprises has entered a mature stage. Enterprises already have rich digital technologies and resources, and the construction of digital platforms is relatively mature. The digital level is higher, the cost of collaborative supervision is lower, and the coefficient of the digital level is higher. At the same time, the importance of cus- tomers in the entire value chain has also increased dramatically. In this case, lC < L1; oI2 > ZC; ð1 (cid:0) m (cid:0) nÞð1 (cid:0) l (cid:0) ZÞC þ L2 < I3; thus, it can be concluded that the equilibrium point of the system evolutionary stability strategy is only (1,1,1). In the mature stage of digital transformation, the benefits gained by all three parties participating in value creation are greater than the collaborative costs paid. With the continuous growth of the evolu- tion period, the strategy of game players participating in value co-creation in the process of maximizing profits will eventually evolve into a fully collaborative state, i.e., collaborative supervision, collaboration, and collaboration, respectively. On the other hand, it also shows that when the digitalization level of enterprises is very high, the choice of collaborative innova- tion is more conducive to maximizing the interests of enterprises and high-quality development. 5. Simulation analysis 5.1 Evolutionary path of tripartite participants at different stages To verify the validity of evolutionary stability analysis and more intuitively describe the dynamic evolution of the collaborative strategy of value co-creators in the early stage, the mid- dle stage, and the mature stage during the service-oriented digital transformation of the con- struction industry, this paper conducts a numerical simulation of the optimal evolutionary stability strategies in the above three stages. At the same time, according to the suggestions of relevant experts and with reference to the analog values of relevant parameters set by Gao [36], Peng [37], and Mei [38] et al., the parameters in the early, middle and mature stages of digital transformation were assigned. In the early stage of digital transformation, the initial values of each parameter are I2 = 20, I3 = 10, α = β = 1/3, ΔI = 30, L1 = 4, L2 = 6, C = 60, λ = η = 1/3, and μ = ν = ω = 0.1. The above parameters meet the conditions of Case 1 and evolve 50 times from different initial strategy combinations. The evolution results are shown in Fig 2. The results show that the system even- tually evolves into a stable strategy combination, i.e., no collaborative supervision, no collabo- ration, and no collaboration, respectively, under different combinations of the initial strategy probability values of the game players, which is consistent with the conclusion of Case 1. It shows that even though there is a certain proportion of construction enterprises, digital solu- tion suppliers and customers that initially choose to participate in value co-creation, the low level of digitalization makes the cooperation costs between construction enterprises and digital solution suppliers higher, the channels for customers to collaborate are fewer, and the three parties do not participate in value co-creation to obtain greater benefits. Driven by interests, the three parties eventually tend to choose not to collaborate. In the middle stage of digital transformation, the initial values of each parameter are I2 = 20, I3 = 10, α = β = 1/3, ΔI = 40, L1 = 7, L2 = 6, C = 30, λ = η = 1/3, and μ = ν = ω = 0.4. The above parameters meet the conditions of Case 2 and evolve 50 times from different initial strat- egy combinations. The evolution results are shown in Fig 3. The results show that there are two stable strategies under different combinations of the initial strategy probability values of the game players, i.e., all three parties collaborate or do not collaborate, which is consistent with the conclusion of case 2. It shows that the construction capacity and service efficiency of PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 13 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation Fig 2. Evolutionary paths in the early stage of digital transformation. https://doi.org/10.1371/journal.pone.0285697.g002 enterprises have improved with the development of digitalization, thus, increasing the total revenue of the three parties. At the same time, the application of digital technology provides a variety of collaboration channels, which promotes effective communication and knowledge sharing among various entities, reduces their collaboration costs, and greatly increases the pos- sibility of choosing to participate in collaborative innovation. In the mature stage of digital transformation, the initial values of each parameter are I2 = 20, I3 = 10, α = β = 1/3, ΔI = 50, L1 = 7, L2 = 7, C = 20, λ = η = 1/3, and μ = ν = ω = 0.85. The above parameters meet the conditions of Case 3 and evolve 50 times from different initial strat- egy combinations. The evolution results are shown in Fig 4. The results show that the system eventually evolves into a stable strategy,i.e., collaborative supervision, collaboration, and col- laboration, respectively, under different combinations of the initial strategy probability values of the game players, which is consistent with the conclusion of Case 3. It shows that even though there is a certain proportion of construction enterprises, digital solution suppliers, and customers who choose not to collaborate at first, high-level enterprise digitalization can bring high benefits and low costs to all entities. Compared with the medium term, the loss of the three parties not cooperating will increase. Driven by interests, the three parties eventually tend to choose to participate in cooperation. According to the constructed Jacobi matrix, it can be seen that the equilibrium point is affected by different parameters of different subjects, which makes the three parties form dif- ferent equilibrium states in different stages of Digital transformation. However, according to the above simulation analysis, from the perspective of the whole Digital transformation cycle, the strategy combination (0, 0, 0) in the early stage will develop into the strategy combination PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 14 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation Fig 3. Evolutionary paths in the middle stage of digital transformation. https://doi.org/10.1371/journal.pone.0285697.g003 (0, 0, 0) or (1, 1, 1) in the middle stage, and finally will reach the equilibrium state of (1, 1, 1) in the mature stage. It indicates that with the improvement of the digitalization level, the value co-creation entities have finally reached the ideal state of (construction enterprises cooperating in supervision, digital solution suppliers participating in cooperation, customers participating in cooperation), and H1 and H4 are accepted. At the same time, the strategy of digital solution suppliers changes with the strategy of construction enterprises, and the strategic choice of cus- tomers is also synchronized with digital solution suppliers, H2 and H3 are accepted. Goldfar and Tucker [39] believe that digital technology has achieved permanent and real-time data sharing, fully reduced the degree of information asymmetry in enterprises, and thus reduced the cost of collaborative innovation between enterprises and customers. At the same time, digi- tal transformation helps enterprises shorten the cycle of product production and innovation, further amplifying the driving effect of interests [40]. Obviously, the higher the level of digitali- zation, the greater the difference between collaborative benefits and costs, and the system will inevitably evolve into an ideal state of complete collaboration eventually. 5.2 Simulation of key parameters There are two distinct stability strategies in the middle stage of digital transformation, and the entire system will eventually develop into a fully collaborative state in the mature stage. There- fore, it is necessary to further study the impact of the main parameters in the middle stage on the evolution of game players’ behavior to promote the service-oriented digital transformation of the construction industry to achieve the ideal state of full collaboration among value co-cre- ators faster. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 15 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation Fig 4. Evolutionary paths in the mature stage of digital transformation. https://doi.org/10.1371/journal.pone.0285697.g004 (1) Influence of initial willingness to collaborate on the system On the basis of the values taken in the middle stage of digital transformation, other variables are assumed to remain unchanged, and the initial values of x,y, and z are taken as variables. We set x = y = z = 0.2, x = y = z = 0.3, and x = y = z = 0.4, and the impact of the initial willing- ness to cooperate of game players on their decision-making behavior is shown in Fig 5. It can be seen from the figure that the initial willingness to cooperate has an impact on the evolution of the behavior strategies of the game players in the middle stage of digital transformation. With the increase in the initial willingness to cooperate, the system gradually evolves from the stable equilibrium point (0,0,0) of complete non-collaboration to the stable equilibrium point (1,1,1) of complete collaboration, and the critical value of the initial collaborative willingness leading to this change is between 0.2 and 0.3. After exceeding the critical value, for the whole system, the higher the initial willingness of the three parties to cooperate, the faster the system will eventually evolve into a stable state of complete cooperation. For a single agent, digital solution suppliers always reach a stable state at a fast speed and choose to collaborate. (2) Influence of the digital level coefficient on the system On the basis of the values taken in the middle stage of digital transformation, the impact of the initial willingness to cooperate of the game players on the system evolution results is elimi- nated, so the probability of the initial collaboration strategy of construction enterprises, digital solution suppliers, and customers is 0.2. We assume that other variables remain unchanged and set μ = 0.3, μ = 0.4, and μ = 0.7. The simulation results are shown in Fig 6. With the increase in data collection and analysis coefficients, the evolution results of the system have PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 16 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation Fig 5. Effect of initial willingness to cooperate on the evolutionary path. https://doi.org/10.1371/journal.pone.0285697.g005 not changed significantly. When the initial collaborative willingness of the game players is low, the changes in data collection and analysis coefficient will not change the final strategy of the game players in the middle stage of the digital transformation. We assume that other variables remain unchanged and set ν = 0.3, ν = 0.4, and ν = 0.7. The simulation results are shown in Fig 7. With the increase in the platform coefficient, the proba- bility of digital solution suppliers collaborating increases, while the probability of construction enterprises’ collaborative supervision decreases. Although the evolutionary process of the sys- tem has changed, the evolutionary stability strategy of the game players is still no collaborative supervision, no participation in collaboration, and no participation in collaboration, respec- tively. In other words, when the initial collaborative willingness of game players is low, the change in the platform coefficient will not affect the evolutionary stability strategy combina- tion of the system. We assume that other variables remain unchanged and set ω = 0.3, ω = 0.4, and ω = 0.7. The simulation results are shown in Fig 8. With the increase in digital coefficients in the con- struction process, the evolutionarily stable equilibrium point of the system gradually tends to (1,1,1) from (0,0,0). It shows that the value co-creation behavior of game players will gradually evolve from a completely uncooperative stable equilibrium point (0,0,0) to a completely coop- erative stable equilibrium point (1,1,1) with the increase in the digital coefficient in the con- struction process even if the initial willingness to cooperate of game players is lower than the critical value. The reason is that the digitalization of the construction process optimizes the business chain and supply chain of construction enterprises and digital solution suppliers in an all-around way, and the digitalization business runs through all the activities of the whole life cycle of product design, production, construction, service, and so on. It makes the business PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 17 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation Fig 6. Effect of data collection and analysis coefficient on the evolutionary path. https://doi.org/10.1371/journal.pone.0285697.g006 activities of both parties more transparent and easier to communicate and reduces the supervi- sion costs of construction enterprises and the collaboration costs of both enterprises. At the same time, the digitalization of the construction process also provides a channel for customers to participate in activities of the entire the whole life cycle, reducing their participation costs and increasing their value perception. The profits obtained by the three parties participating in the collaboration are far higher than the cost they have to pay. The motivation of pursuing profit maximization urges them to choose to participate in the collaboration. Furthermore, by comparing the results of Figs 6–8, it is found that the digitalization level of the construction process is the key factor in promoting the development from the middle stage of the digitalization transformation to the mature stage and is an important catalyst and break- through for promoting the service-oriented digitalization transformation of the construction industry. This conclusion is consistent with the views of other scholars. The collection and analysis of data only provide a framework for the composition of related technologies, and the digital platform provides new channels for the cooperation of value co-creators. However, the embodiment of their value still depends on production activities. Compared with the level of data collection and analysis and platformization, the digitalization of the construction process has a more direct effect on the collaborative intention of value co-creators [41]. The applica- tion of digital technology in the construction process is the focus of construction enterprises. The leaders of government and enterprises should speed up the improvement of the digital infrastructure of construction enterprises, promote the integration and application of advanced digital construction technology in the construction process, and optimize the inter- action and construction capacity of various departments. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 18 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation Fig 7. Effect of the platform coefficient on the evolutionary path. https://doi.org/10.1371/journal.pone.0285697.g007 6. Conclusion In this paper, a tripartite evolutionary game model composed of construction enterprises, digi- tal solution suppliers, and customers is constructed, aiming at the problem of uncertain strat- egy combinations generated by multiple agents in pursuit of the maximization of their own interests in the service-oriented digital transformation process of the construction industry. We reveal the collaborative evolutionary path of various value-creating agents in the process of digital transformation and the impact of relevant elements on the realization mechanism of value co-creation. According to the influence relationship and stability conditions of various factors, we propose relevant countermeasures and suggestions for the digital transformation and upgrading of the construction industry. The main conclusions of the study are as follows: First, the level of digitalization has a direct impact on the synergistic benefits and costs of construction enterprises, digital solution suppliers, and customers. At different stages of digital transformation, the combination of behavior strategies of game players will change. The higher the level of digitalization is, the higher the degree of collaboration among the value co-creators is. Value co-creators in the service-oriented digital ecosystem of the construction industry will eventually choose a stable combination of completely collaborative strategies. The collabora- tive behavior of the subjects has an impact on the realization of their own interests maximiza- tion and the value creation efficiency of the value chain. Second, in the middle stage of the digital transformation, the initial willingness to cooperate of the game players will eventually make the system evolve into two types of completely opposite stable strategies, and the higher the initial willingness to cooperate of the players is, the faster they will develop to the complete collaboration state in the mature stage. Finally, in the middle stage of the digital PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 19 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation Fig 8. Effect of digitization coefficient of the construction process on the evolutionary path. https://doi.org/10.1371/journal.pone.0285697.g008 transformation, when the initial willingness to cooperate of the three parties is lower than the critical value, the improvement of the digital level of the construction process can reverse the final evolution of the system into a completely uncooperative state, which is the key to promot- ing the development of the system and the cooperation of the agents. To promote cooperation among value co-creators and achieve the balance of the service- oriented digital ecosystem of the construction industry, the following suggestions are proposed: (1) At different stages of digital transformation, construction enterprises should adjust the focus of collaborative development according to the level of digital transformation and the actual situation of each participant to reduce the participation cost of value co-creation. Specifically, in the early stage of digital transformation, the development of digital technol- ogy is the focus of the construction industry. Construction enterprises should pay attention to the improvement of digital infrastructure, build a platform together with digital solution suppliers to provide customers with channels to participate in value co-creation, and reduce the difficulty of tripartite participation. In the middle stage of digital transformation, con- struction enterprises should pay attention to the improvement of the digital level in the con- struction process, promote the integration and application of advanced digital construction technology in the construction process, and promote the transition of the industry to the mature stage of digital transformation. In the mature stage of digital transformation, the optimization of the income distribution mechanism should be considered, and cooperation PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 20 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation and trust among participants should be enhanced to maintain the stability of the value co- creation network. (2) Change the concept of all participants in the service-oriented value chain of the construc- tion industry and establish a win-win attitude. First, construction enterprises, digital solu- tion suppliers, and customers themselves need to change their mindset and regard each other as partners for value co-creation. Second, the government should give full play to its guiding role, actively use the new national system, and fully mobilize the collaborative inno- vation and willingness of value co-creators. Although this paper has conducted a useful discussion on the collaborative behavior of value co-creators in the digital service ecosystem of the construction industry, there are still some limitations. Firstly, in terms of research methods, due to the difficulty in obtaining data related to the digital transformation of service-oriented construction enterprises, this paper carries out research based on artificially simulated data. In future research, a specific construc- tion enterprise should be taken as an example and real data should be used for more convinc- ing simulation analysis; Secondly, in terms of research content, this article only considers the behavior strategies of three value co-creators, namely construction enterprises, digital solution suppliers, and customers. However, under the special national conditions of China, the gov- ernment plays a key role in the transformation process of enterprise digitization and service. In the future, it is possible to further explore the four-party evolutionary game considering the government under the national conditions of socialism with Chinese characteristics. Supporting information S1 File. (ZIP) Author Contributions Conceptualization: Shiming Wang. Data curation: Hui Su. Formal analysis: Hui Su. Funding acquisition: Shiming Wang. Investigation: Hui Su. Methodology: Hui Su, Qiang Hou. Project administration: Shiming Wang. Resources: Hui Su. Software: Hui Su. Supervision: Shiming Wang. Validation: Qiang Hou. Visualization: Hui Su. Writing – original draft: Hui Su. Writing – review & editing: Shiming Wang, Qiang Hou. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 21 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation References 1. Teece DJ. Profiting from innovation in the digital economy: Enabling technologies, standards, and licensing models in the wireless world. Res Policy. 2018; 47(8): 1367–1387. 2. Li T, Yan XL, Guo WP, Wang FF. Research on Factors Influencing Intelligent Construction Develop- ment: An Empirical Study in China. Buildings. 2022; 12(4): 478. https://doi.org/10.3390/ buildings12040478. 3. Xie XX. Who Will Become the Digital Leading Country: Digitalization in the View of Industrialization. Theory and Reform. 2021; 34(3): 67–79. 4. Vargo SL, Lusch RF. Evolving to a New Dominant Logic for Marketing. J Mark. 2004; 68(1): 1–17. 5. Kim E, Tang LR, Bosselman R. Customer Perceptions of Innovativeness: An Accelerator for Value Co- Creation. Journal of Hospitality & Tourism Research. 2019; 43(6): 807–838. 6. Parkinson J, Mulcahy RF, Schuster L, Taiminen H. A transformative value co-creation framework for online services. Journal of Service Theory and Practice. 2019; 29(3): 353–374. 7. Wang Y, Li Y, Liu BS, Cao XY. Research on the Path of Value Co-creation in the Whole Process of EPC Project: Based on the Perspective of Service Dominant Logic. Construction Economy. 2022; 43 (9): 89–96. 8. Vargo SL, Lusch RF. Service-dominant logic 2025. International Journal of Research in Marketing. 2017; 34(1): 46–67. 9. Xin BL, Liu YQ. Research on Service Ecosystem Integration Model Based on Institutional Mechanism. Forum on Science and Technology in China. 2021; 37(1): 136–146. 10. Hua ZS, Wei J, Zhou WH, Yang Y, Zhang W. Prospect of Service Science and Innovation Management Under Network Environment. Chinese Journal of Management Science. 2018; 26(2): 186–196. 11. Pacheco JM, Traulsen A, Nowak MA. Coevolution of strategy and structure in complex networks with dynamical linking. Physical review letters. 2006; 97(25): 258103. https://doi.org/10.1103/PhysRevLett. 97.258103 PMID: 17280398 12. Sun FJ, He T, Jin CM, Dong GQ. Symbiotic evolution model of value co-creation unit of service ecosys- tem. Computer Integrated Manufacturing Systems. 2021; 28(5): 1549–1561. 13. 14. Zhang XM, Yuan JF, Dan B, Sui RH, Li WB. The evolution mechanism of the multi-value chain network ecosystem supported by the third-party platform. Journal of Industrial and Management Optimization. 2022; 18(6): 4071–4091. Li WH, Li N. Dynamic Game and Simulation of Intelligent Upgrade of Manufacturing Enterprise Based on Complex Network——Considering the Driving Effect of Government and Consumers. Soft Sci. 2022; 36(3): 39–47. 15. Koscheyev V, Rapgof V, Vinogradova V. Digital transformation of construction organizations. IOP Conf Ser Mater Sci Eng. 2019; 497(1): 012010. https://doi.org/10.1088/1757-899X/497/1/012010 16. Beltagui A, Rosli A, Candi M. Exaptation in a digital innovation ecosystem: The disruptive impacts of 3D printing. Res Policy. 2020; 49(1): 103833. https://doi.org/10.1016/j.respol.2019.103833. 17. Barile S, Bassano C, Piciocchi P, Saviano M, Spohrer JC. Empowering value co-creation in the digital age. Journal of Business & Industrial Marketing. 2021. https://doi.org/10.1108/JBIM-12-2019-0553. 18. Nambisan S, Wright M, Feldman M. The digital transformation of innovation and entrepreneurship: Progress, challenges and key themes. Res Policy. 2019; 48(8): 103773. https://doi.org/10.1016/j. respol.2019.03.018. 19. Kolagar M, Parida V, Sjo¨ din D. Ecosystem transformation for digital servitization: A systematic review, integrative framework, and future research agenda. J Bus Res. 2022; 146: 176–200. 20. Ciasullo MV, Polese F, Montera R, Carrubbo L. A digital servitization framework for viable manufactur- ing companies. Journal of Business & Industrial Marketing. 2021; 36(13): 142–160. 21. Goudarzi EV, Houshmand M, Valilai OF, Ghezavati V, Bamdad S. Equilibrial service composition model in Cloud manufacturing (ESCM) based on non-cooperative and cooperative game theory for healthcare service equipping. PeerJ Comput Sci. 2021; 7: e410. https://doi.org/10.7717/peerj-cs.410 PMID: 33817051 22. Bello S A, Oyedele L O, Akinade O O, et al. Cloud computing in construction industry: Use cases, bene- fits and challenges. Automation in Construction. 2021; 122: 103441. 23. Machado R L, Vilela C. Conceptual framework for integrating BIM and augmented reality in construction management. Journal of civil engineering and management. 2020; 26(1): 83–94. 24. Zhang J, Li H, Li V, Xia B, Skitmore M. Internal relationships of market-oriented EFQM enablers in the Chinese construction industry. Engineering, Construction and Architectural Management. 2020; 28(3): 765–787. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 22 / 23 PLOS ONEEvolutionary game study on multi-agent value co-creation 25. Prebanić K R, Vukomanović M. Realizing the need for digital transformation of stakeholder manage- ment: A systematic review in the construction industry. Sustainability. 2021; 13(22): 12690. 26. Rasmussen NV, Beliatis MJ. IoT based Digitalization and Servitization of Construction Equipment in Concrete Industry. Global IoT Summit (GIoTS); 2019 Jun 17–21; Aarhus, Denmark. IEEE; 2019. p. 1– 4. 27. Galera-Zarco C, Campos JA. Exploring Servitization in Industrial Construction: A Sustainable Approach. Sustainability. 2021; 13(14): 8002. https://doi.org/10.3390/su13148002. 28. Chen K, Du P, Fang LW, Gao LF. Digital Transformation of Construction Industry in China: Connotation, Subject, and Policy Tool. Journal of Civil Engineering and Management. 2021; 38(4): 23–29. 29. Shan ZD, Chen L, Zeng YH, Li HM. Research on symbiotic relationship among flexible populations under digital guidance: Based on the perspective of service ecosystem. Science-Technology and Man- agement. 2021; 23(05): 75–84. 30. Barney JB. Resource-based theories of competitive advantage: A ten-year retrospective on the resource-based view. Journal of Management. 2001; 27(6): 643–650. 31. Meng FS, Zhao G, Xu Y. Research on Intelligent Transformation and Upgrading Evolution Game of High-end Equipment Manufacturing Enterprises Based on Digitalization. Scientific Management Research. 2019; 37(5): 89–97. 32. Wang Q, Lai W, Ding M, Qiu Q. Research on Cooperative Behavior of Green Technology Innovation in Construction Enterprises Based on Evolutionary Game. Buildings. 2022; 12(1): 19. https://doi.org/10. 3390/buildings12010019. 33. 34. Friedman D. Evolutionary Games in Economics. Econometrica. 1991; 59(3): 637–666. Ziemba E, Leyh C, Scha¨ ffer T, Bley K, Forstenha¨usler S. Assessing the IT and Software Landscapes of Industry 4.0-Enterprises: The Maturity Model SIMMI 4.0. Information Technology for Management: New Ideas and Real Solutions. In: Ziemba E. (eds) Information Technology for Management: New Ideas and Real Solutions. ISM AITM 2016 2016. Lecture Notes in Business Information Processing; 2016 Sep 11–14; Cham: Springer; 2017. p. 103–119. 35. Kolagar M, Reim W, Parida V, Sjo¨din D. Digital servitization strategies for SME internationalization: the interplay between digital service maturity and ecosystem involvement. Journal of Service Management. 2021; 33(1): 143–162. 36. Gao J, Zhang W, Guan T, Feng QH. Evolutionary game study on multi-agent collaboration of digital transformation in service-oriented manufacturing value chain. Electronic Commerce Research. 2022 May: 1–22. 37. Peng BH, Gu XF, Lu Q. Analysis of Evolutionary Game of Stakeholders in Service-oriented Manufactur- ing Project Governance. Journal of System Simulation. 2017; 29(3): 595–608+617. 38. Mei GP, He Y, Wan JX. Coordination mechanism research of value co-creation behavior of service sup- ply chain system. Journal of System Simulation. 2022; 37(2): 263–274. 39. Goldfarb A, Tucker C. Digital economics. Journal of Economic Literature. 2019; 57(1): 3–43. 40. 41. Li G, Yu H, Lu M. Low-Carbon Collaboration in the Supply Chain under Digital Transformation: An Evo- lutionary Game-Theoretic Analysis. Processes. 2022; 10(10): 1958. Li XH, Wang YF. Data value chain and the mechanism of value creation. Economic Review Journal. 2020; 420(11): 54–62+2. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285697 May 16, 2023 23 / 23 PLOS ONE
10.1371_journal.pone.0284821	RESEARCH ARTICLE Numeracy skills learning of children in Africa:—Are disabled children lagging behind? Huafeng ZhangID 1 1,2, Stein T. HoldenID 1 School of Economics and Business, Norwegian University of Life Sciences, Ås, Norway, 2 Fafo Institute for Labour and Social Research, Oslo, Norway zhu@fafo.no Abstract Significant progress has been achieved in universal basic education in African countries since the late 1990s. This study provides empirical evidence on the within- and across-coun- try variation in numeracy skills performance among children based on nationally representa- tive data from eight African countries (DR Congo, The Gambia, Ghana, Lesotho, Sierra Leone, Togo, Tunisia, and Zimbabwe). We assess whether and to what extent children with disabilities lag in numeracy skills and how much it depends on their type of disabilities. More specifically, we explore whether disabled children benefit equally from better school system quality. The assessment is analysed as a natural experiment using the performance of non- disabled children as a benchmark and considering the different types of disabilities as ran- dom treatments. We first evaluate the variation in average numeracy skills in the eight Afri- can countries. They can roughly be divided into low- and high-numeracy countries. We apply Instrumental Variable (IV) methods to control the endogeneity of completed school years when assessing subjects’ school performance and heterogeneous disability effects. Children with vision and hearing disabilities are not especially challenged in numeracy skills performance. The low numeracy skills among physically and intellectually disabled children are mainly attributable to their limited school attendance. Children with multiple disabilities are constrained both by low school attendance and by poor numeracy skills return to school- ing. The average differences in school performance across the high- versus low-numeracy skill country groups are larger than the within-group average differences for disabled versus non-disabled kids. This indicates that school enrolment and quality are crucial for children’s learning of numeracy skills, and that disabled children benefit equally from better school quality across these African countries. 1. Introduction The Sustainable Development Goal (SDG) 4 aims at inclusive and equal access to education for all children [1]. Significant progress has been achieved [2] since the adoption of several development frameworks, such as Education for all [3] and the Millennium Development Goals [4]. Data from UNESCO Institute for Statistics suggests that since the late 1990s and a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Zhang H, Holden ST (2023) Numeracy skills learning of children in Africa:—Are disabled children lagging behind? PLoS ONE 18(4): e0284821. https://doi.org/10.1371/journal. pone.0284821 Editor: Verda Salman, NUST: National University of Sciences and Technology, PAKISTAN Received: November 11, 2022 Accepted: April 7, 2023 Published: April 20, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0284821 Copyright: © 2023 Zhang, Holden. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The data is openly available for all at: https://mics.unicef.org/surveys. Funding: The paper has been undertaken as part of the research project “Education outcome variability in children with disabilities: Structure, institution or PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 1 / 23 PLOS ONEagency?” funded by the Research Council of Norway (Project Number 300635). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Numeracy skills learning of children in Africa early 2000s, most African countries have increased the gross enrolment in primary schools from about 75% to almost 100%. Even countries with low school enrolment historically, such as Niger, also witnessed their primary school gross enrolment to grow from 30% in the late 1990s to about 60–70% in recent years [5]. Although universal basic education has achieved great success, recent studies are concerned about poor school performance among children across African countries [6]. Many children did not gain basic skills in reading and mathematics even after many years of schooling [6–8]. Furthermore, the achievement gained in school enrolment has masked problems related to unequal distribution and disparity in school performance, as well as the marginalisation of the most disadvantaged and vulnerable groups of children [9–12]. Children with disabilities are possibly among those exposed to such limitations and risks [13, 14]. This paper aims to investigate the learning outcome in form of numeracy skills for children with and without disabilities in eight African countries. Based on the sixth round of Multiple Indicator Cluster Surveys (MICS), we aim to answer the following research questions: 1) To what extent do the average numeracy skills vary across African countries? 2) To what extent does the average performance differ between children with and without disabilities? 3) To what extent is the numeracy skills return to schooling dependent on children’s disability status and types of disabilities? 4) Are disabled children able to benefit from better school system quality to the same extent as non-disabled children? To answer these questions, we first evalu- ate the variation in the numeracy skills across the eight African countries in our study and esti- mate the disability effects on numeracy skills returns to schooling by using non-disabled children as the counterfactual. Afterwards, we assess the relative performance between dis- abled and non-disabled children in countries with low- and high-numeracy skills. The coun- try-level school quality is measured by the mean numeracy score of non-disabled children in these countries. There is a growing research interest in timely and reliable empirical evidence on school enrolment and learning performance for children with disabilities in developing countries to measure the across-region variation [15, 16]. Several comparative studies based on data from multiple countries provided evidence on disabled children’s overall low school attendance, enrolment, and school completion [15, 17–21]. However, none of these comparative studies has assessed disabled children’s school performance. Earlier studies based on Western experiences have presented evidence for learning chal- lenges among disabled children since they are often limited in cognitive, behavioural, motor, and emotional abilities [22, 23]. However, the evidence on disability gaps in learning skills in the developing context is limited and primarily through simple tests embedded in surveys in individual countries. For example, studies in India [24] and Pakistan [25, 26] suggest a signifi- cant disability gap in numeracy skills. These studies do not indicate whether the low numeracy skills among disabled children have been merely correlated with their low school attendance or have originated from their challenges in learning in school. Takeda and Lamichhane (2018) notice that when the interaction between disability status and school status is included in the model, the disability dummy becomes insignificant [24]. They conclude that once disabled children access education, they become less likely to fall behind in school performance. We suggest that the endogeneity of selection into schooling should be considered when estimating the disability gap in learning. Due to the challenges in sample size for disabled children, many studies in the developing context used the catch-all category for disability. There are a few exceptions. Singal et al. (2018) evaluated Pakistani children’s basic numeracy skills among those with three types of disabilities and varying severity: sensory (walking, seeing and hearing); self-caring (difficulty in dressing and washing all over); and cognitive [26]. They only found significant disability PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 2 / 23 PLOS ONENumeracy skills learning of children in Africa gap in learning outcomes among those with moderate or severe sensory disabilities but not among those with mild disabilities or other disability types. Another study that also differentiates disability types is Bakhshi, Babulal and Trani (2018), who predicted school access and literacy in Western Darfur in Sudan for children with four types of disabilities: sensory (physical, seeing and hearing), mental and cognitive, behavioural and mood, and multiple disabilities [6]. They found no difference in skills performance either with the catch-all disability category or with different disability types. However, the authors further argue that in the conflict setting in Darfur, where all children are exposed to a high risk of being excluded and not taught in school, the differences in school performance might have disappeared. More evidence is needed to understand the heterogeneous effects of disability types and the potential correlation between the disability effect and school quality. To the best of our knowledge, our study is the first comprehensive study evaluating disabled children’s achievement in numeracy skills based on the standardised WG-CFM and numeracy tests in African countries. Our analysis uses the natural experiment framework by using the sample of non-disabled children as a benchmark (counterfactual). When assessing subjects’ numeracy skills returns to schooling and heterogeneous disability effects, we apply Instrumen- tal Variable (IV) methods to control the endogeneity of completed school years, since the dis- ability status may directly affect children’s likelihood of being in school. 2. Conceptual framework This paper explores whether children with disabilities lag in numeracy skills compared to non- disabled children and to what extent such a lag varies with their disability status, school enrol- ment and country-level numeracy performance. Children with disabilities might lag in numer- acy skills if they do not attend school as much as their non-disabled peers. Earlier studies found that children with disabilities are exposed to a higher risk of not attending school, enrolling late, or dropping out of school early [15, 17, 18, 27]. The factors constraining disabled children from school attendance can be diverse due to their varied functional difficulties. On the other hand, disabled children can also be constrained in learning numeracy skills due to diversified challenges in the learning process, even if they are equally enrolled in school. Some literature indicates that children with developmental delay in motor coordination, severe delay in motor skills, and visual-motor integration skills have challenges in learning math [28, 29]. “Embodied cognition” theory argues that the mathematical cognitive process is grounded in the simulations of sensorimotor processes through the interaction of the body with the world [15]. Earlier studies have not supported that children’s seeing or hearing abilities are prerequi- sites for developing essential numeracy competencies. Zarfaty et al. (2004) conclude that deaf children in their early years do not have a problem with representing numbers and are particu- larly good at representing numbers when sets are presented as spatial arrays [30]. Morgan et al. (2011) also find that children with speech-language impairments do not lag behind non- disabled children in their math skills growth [31]. Crollen et al. (2021) have reported that blind children might even outperform their non-blind peers in numeracy abilities [32]. However, Zhang et al. (2019) demonstrate that children with seeing or auditory perception challenges struggle to learn numeracy skills related to visual Arabic or verbal modules [33]. Numerous studies have also presented evidence that children’s development in various abil- ities, such as information processing, cognitive abilities, and attentive behaviours, is critical for their learning process [22, 34]. Children with intellectual disabilities are often characterised by cognitive, behavioural, and emotional difficulties [35], which can constrain children’s ability to learn numeracy skills [34]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 3 / 23 PLOS ONENumeracy skills learning of children in Africa Finally, a lack of teaching materials (such as braille or eyeglasses, hearing aids equipment, walking equipment, and sign language) and proper pedagogical interventions for children with disabilities may also constrain their skill learning. Other potential challenges in school can be stigma and negative perceptions, attributions, and expectations of their teachers [36]. Children with multiple disabilities have higher risks related to all the challenges discussed ear- lier than children with single disabilities. The question is whether or to what extent disabled children’s numeracy skills are influenced by factors other than their school attendance and how these factors correlate with their disability status and disability types. Another concern in investigating children’s learning outcomes is school quality [37, 38]. Heyneman & Loxley (1983) studied 29 high- and low-income countries and concluded that in low-income countries, the effect of school quality on primary school children’s academic achievement was more prominent than the effect of family socioeconomic status [39]. Bakhshi, Babulal, and Trani (2018) report that when the overall school learning is poor in a conflict set- ting, there is no difference in learning performance as everybody may lag in poor-quality schools [6]. So far, little evidence in African countries has indicated whether children with dis- abilities might benefit more or less from high school quality and whether the gap between chil- dren with and without disabilities will expand or stabilise when school quality improves. We suggest testing this by comparing the disability effects on children’s numeracy skills perfor- mance across countries with low- and high-numeracy skills. Our framework is presented in Fig 1. This paper will estimate the heterogeneous disability effect on the return to schooling regarding numeracy skills with IV models. S1 Table presents the sample size for the split samples. We will estimate the disability effect in the low- and high- numeracy skills country groups for the three disability types, respectively. Ideally, we would have included all five disability types. However, the sample is too small for vision and hearing disabilities in the split sample of sub-groups. We aim to test the following hypotheses: H1. There is a considerable variation in average school performance, measured by the average numeracy skills of children across African countries. Fig 1. Framework on numeracy skills performance for children with and without disability. https://doi.org/10.1371/journal.pone.0284821.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 4 / 23 PLOS ONENumeracy skills learning of children in Africa H2. Children with disabilities perform significantly worse than their non-disabled peers of the same age in learning numeracy skills. H3. There are heterogeneous disability effects in numeracy skills return to schooling for children with different disabilities. More specifically, we hypothesise that: H3a. Children with vision and hearing disabilities perform well in numeracy skills return to schooling compared to non-disabled children. This hypothesis is based on earlier evidence [30, 32], suggesting that vision and hearing abilities might not be crucial in developing numeracy skills. Although learning numeracy skills related to visual or verbal modules might be relevant [33], the numeracy tests involved in this survey are pretty basic. H3b. Children with physical disabilities have a lower return to schooling in numeracy skills learning than non-disabled children. This hypothesis is based on the embodied cognition theory [40] that motor skills can constrain children’s numeracy skills learning. H3c. Children with intellectual disabilities have a lower return to schooling in numeracy skills learning than non-disabled children. This hypothesis follows various research findings [22, 34, 35] that children’s cognitive and behavioural abilities development is crucial for their numeracy learning. H3d. Children with multiple disabilities have the lowest numeracy skills return to schooling among all disability types. Children with multiple disabilities are exposed to higher chal- lenges [27] because they have fewer opportunities of substituting across senses and func- tions in their learning processes. H4a. The gap in numeracy skills between non-disabled and disabled children is larger in high- numeracy skills countries. It is based on the assumption that children with disabilities are less capable of benefiting from the better quality of the school system than non-disabled children. Disabled children likely need to give extra effort to the senses and functions that work well to compensate for their disability constraints. More resources and teaching skills are needed to cater for the unique needs of disabled children. H4b. The within-group average differences in the numeracy skills between non-disabled and dis- abled children are smaller than the between-group average differences between the low- and high-numeracy skills country groups. This is based on the assumption that despite the func- tional challenges among disabled children, schooling with good quality may greatly contrib- ute to the school performance of children both with and without disabilities. 3. Data, methods, and estimation strategy The MICS surveys aim to provide internationally comparable data about the education status of children and women. Our sample is a national representative sample from eight African countries (DR Congo, The Gambia, Ghana, Lesotho, Sierra Leone, Togo, Tunisia, and Zimba- bwe) that conducted the sixth round of the MICS surveys in 2017–2019, conducted by United Nations International Children’s Emergency Fund (UNICEF). The survey includes a standard learning assessment test for children aged 7–14 [41]. The MICS survey adopts Washington Group Child Functioning Module (WG-CFM), which aims to capture the most common functional problems related to child development for children aged 6–17 [42, 43]. WG-CFM include 26 questions (related to 13 functional domains) with four severity scales. This paper formulates vision disability as severe difficulty in vision even with glasses or contact lenses; hearing disability as extreme difficulty in hearing even with PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 5 / 23 PLOS ONENumeracy skills learning of children in Africa a hearing aid; physical disability as severe difficulty in self-care or walking 500 metres on level ground without equipment; intellectual disability as severe difficulties in communication, learning, remembering or concentrating on activities that the child enjoys doing; and multiple disabilities as more than one co-occurring severe functional difficulties as prescribed earlier. Here, severe functional difficulty refers to a lot of functional difficulties or no function at all. The study only uses eight (categorised as four disability types) out of the 13 functional domains captured by WG-CFM. The remaining five functional domains (accepting change, controlling behaviour, making friends, anxiety, and depression) are not included since their prevalence rates vary greatly across the eight African countries, possibly indicating a challenge in inter- preting these functional challenges in the local context. The sample includes currently-in-school, dropout, and never-in-school children. Table 1 shows the total sample size is 32,306, including 30,013 non-disabled children as the counterfac- tual and 2,293 disabled treatment sample. School enrolment is lower among disabled children (87.8%) than non-disabled children (91.0%) and lowest among multiple disabled children (70.5%). The response rates to the numeracy test among different groups of Ever-In-School children are generally quite high (about 95% or higher) but much lower among the Never-In- School disabled sample (76.1%). We frame our econometric analysis as a natural experiment, which assumes that the sub- jects are exposed to a random disability treatment determined by nature or factors outside the control of the subjects or researchers [44]. Disability can be considered an exogenous treat- ment variable since it is most likely not determined by the characteristics of the population or geographic, economic or social aspects. Despite the potential correlation between poverty and childhood disability declared by some studies [45], the nature of this connection has been complicated. Two mechanisms coexist: children in poor households can be exposed to a higher risk of being disabled, while families with disabled children might experience social depriva- tion due to the high costs related to their healthcare needs [46]. Moreover, some studies sug- gest that the gaps in socioeconomic characteristics between people with and without disabilities might be limited and are not statistically significant in a poor environment [47, 48]. To critically assess our natural experiment assumption, we further regress each disability type on a set of individual, family, wealth, and geographical variables (S2 Table). It supports our natural experiment assumption if we find no or very low correlation between these. S2 Table shows that the natural experiment assumption is supported. Table 1. Sample size by school status and disability status. A B C D E F G H I Total sample Ever-In-School Children (EISC) % EISC (B/A) EISC took numeracy test % EISC took numeracy test (D/ B) Never-In-School Children (NISC) % NISC (F/A) NISC took numeracy test % NISC took numeracy test (H/ F) Non-disabled 30,013 Disabled Vision disability Hearing disability Physical disability Intellectual disability Multiple disabilities Total 2,293 168 96 422 1,366 241 27,305 2,013 163 91.0 87.8 97.0 87 90.6 357 84.6 1,236 90.5 170 70.5 26,556 1,922 154 81 347 1,194 146 32,306 29,318 90.8 28,478 https://doi.org/10.1371/journal.pone.0284821.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 97.3 95.5 94.5 93.1 97.2 96.6 85.9 97.1 2,708 280 5 9 9.0 12.2 3.0 9.4 65 15.4 130 9.5 71 29.5 2,563 213 5 6 54 114 34 2,988 9.2 2,776 94.6 76.1 100.0 66.7 83.1 87.7 47.9 92.9 6 / 23 PLOS ONENumeracy skills learning of children in Africa Our outcome variable in this study is children’s performance in a numeracy skills test, which is measured as the mean numeracy test score based on four sets of altogether 21 numer- acy test questions on symbols reading, quantity comparison, addition and, logical sequence. Our exogenous “treatment” sample consists of children classified in one of the five severe dis- ability types (seeing, hearing, physical, intellectual, and multiple disabilities). The counterfac- tual sample includes those who did not report severe or moderate disabilities. The disparities in the numeracy test between treatment and control children are assumed to be the treatment impacts or causal disability effects. The majority of our sample consists of non-disabled children; therefore, we can test hypoth- esis H1 by assessing the variation in numeracy skills within and across countries. The non-dis- abled children’s performance also serves as a good benchmark to evaluate the numeracy performance of disabled children that are much fewer in number. The fact that we found size- able across-country variation in numeracy scores among non-disabled children caused us to split our sample into low- and high-numeracy skill countries. We assess the relative perfor- mance of non-disabled children versus disabled children within these country groups. This split also serves as a proxy measurement of school quality across the two groups to evaluate the role of school system quality on numeracy skills for disabled children and the gaps between disabled and non-disabled children. Most studies on disabled children’s education apply bivariate or multivariate logistic or probit models to evaluate their access to education (such as school enrolment, school comple- tion, dropout, highest grade achieved) for children with and without disability [17–19, 26]. Some studies simply use univariate analysis while including the disability status [15, 21]. Some studies dichotomise the indicators (able to read or write) for school performance and use logis- tic or probit models [6, 25]. Takeda and Lamichhane (2018) use an OLS model to estimate school performance as a continuous score [24]. These studies assess the correlation between children’s disability status and their school performance without considering the cause-effect of disability on children’s schooling. A few studies use household fixed-effect models to esti- mate the disability effect by controlling for unobserved and observed household characteristics [17, 18]. However, such kind of studies require a sample of children both with and without dis- abilities from the same household, which may not always be available. Children’s numeracy skills are primarily learnt through school attendance. Disabled chil- dren may fall behind other children in numeracy skills for two reasons. First, they may fall behind because they cannot attend school and complete fewer school years. Second, their dis- ability may limit their numeracy skill learning ability while in school. Children’s educational attainment (completed school years) can be considered as both the outcome of disability and, at the same time, an endogenous treatment on skill learning. Therefore, to estimate the disabil- ity effect of numeracy skills, we suggest using the instrumental variable (IV) method to control for the potential bias associated with endogenous completed school years of disabled versus non-disabled children. In the first set of regressions (Eq (1) below), we test hypothesis H2, which states that chil- dren with disabilities perform significantly worse than their non-disabled peers of the same age in learning numeracy skills. We first test a reduced-form model which ignores the causal mechanisms with a parsimonious specification. The first model includes only age and the treatment variable Dij, indicating children as non-disabled or with disability type j. We then run additional models, first including the country dummies and then gender. Without consid- ering endogenous treatment and possible interaction effects, the first set of regressions allows us to assess the variation in numeracy skills by age and disability types. Numeracyi ¼ b0 þ b1jDij þ b2Agei þ b3Genderi þ b4kCountryik þ uijk ð1Þ PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 7 / 23 PLOS ONENumeracy skills learning of children in Africa Here, the subscript i represents each individual child, j represents a type of disabilities (including children without disability, children with vision, hearing, physical, intellectual, and multiple disabilities), k represents countries, and uijk is the error term. In the models, β0 esti- mates the average score rates of numeracy tests for the 7-year-old non-disabled control chil- dren in DRCongo (the country used as the base). β1j estimates the marginal disability treatment effects of disability type j on children’s performance of numeracy skills. In the second set of regressions, we want to test hypothesis H3, which suggests heteroge- neous disability effects in learning numeracy skills for children with different disabilities. The type of disability may affect each step in the causal mechanisms in different ways; therefore, we run the IV models on the split samples by various disability statuses: Outcome equation : Numeracyij ¼ U1j∗CSYij Selection equation : CSYij ¼ p0j þ p1jlnðAgeijÞ þ p2jGenderij þ εij ð2Þ ð3Þ Here, U1j estimates the average numeracy skills return to each completed school year among the children with disability type j. This is the parameter of interest. We want to test whether the return to education per school year in numeracy skills is homogeneous or depends on disability types. In the first stage of regressions, π1j and π2j capture the effect of age and gen- der on the number of school years completed by children with disability type j. ln(age) is included since it performs best in satisfying the Sargan overidentification test. The constant term π0j is included in the first stage but not the second one since we assume that children learn numeracy skills mainly from school and therefore have no numeracy skills when they start school. We apply the ivregress 2SLS estimator in Stata 15. In the IV model, to satisfy the theoretical validity of our identification strategy, we use age and gender as instruments, as these variables affect completed school years. They do not directly affect numeracy skills learning (exclusion restriction). For children’s age and gender to be strong instruments, they must be strongly correlated with the completed school years. For these instruments to be statistically valid, they must be uncorrelated with the error term in the numeracy skills (outcome) model. These properties are also statistically testable in the overidentified case. We present standard IV instrument tests of endogeneity (Robust Wu- Hausman test), the strength of the instruments (first stage F test), and the overidentification (Sargan IV validity test). We also present results from Ordinary Least Square (OLS) regres- sions if the IV tests are not satisfied. In the third set of regressions, we want to test hypotheses H4a and H4b, which evaluate the role of school system quality on the numeracy skills of disabled children and the gaps between disabled and non-disabled children. We run all IV split-sample models in the low- and high- numeracy skills country groups for the non-disabled children and children with physical, intel- lectual, and multiple disabilities, respectively. 4. Results 4.1 Descriptive analysis The descriptive statistics of outcome and control variables are presented in S3 Table. We calcu- late children’s overall numeracy test scores as the mean value of 21 numeracy questions (0 = wrong, 1 = correct). We show the mean test scores by children’s age (left figure) and by completed school years (right figure) in Fig 2. The figure draws vertical box plots, which show the median, 25th and 75th percentile (upper and lower hinge) and lower and upper adjacent values of the mean test scores in each group. The outside values are plotted as dots. The figure PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 8 / 23 PLOS ONENumeracy skills learning of children in Africa Fig 2. Numeracy test scores by children’s age or by completed school years (median, 25th and 75th percentile). https://doi.org/10.1371/journal.pone.0284821.g002 suggests that children perform better in numeracy skills when they grow older. The disparity in numeracy skills performance by completed school years is higher than the age disparity. It is in line with the earlier assumption that age does not directly affect numeracy skills and only involves exposure to schooling. Table 2 shows the mean numeracy score by countries for non-disabled and disabled chil- dren, respectively. The overall mean numeracy score for the non-disabled is 0.57, which is rela- tively low in DRCongo (0.35), Sierra Leone (0.41) and The Gambia (0.50). In the remaining five countries (Ghana, Lesotho, Togo, Tunisia, and Zimbabwe), the mean numeracy score is between 0.63 and 0.88. The average numeracy skills in these countries are about double those Table 2. Mean numeracy score by countries. Non-disabled Disabled Mean Std. err. Mean Std. err. Non-disabled Sample size Disabled Total DRCongo The Gambia Ghana Lesotho Sierra Leone Togo Tunisia Zimbabwe Total 0.35 0.50 0.70 0.68 0.41 0.63 0.88 0.75 0.57 0.004 0.007 0.005 0.006 0.005 0.007 0.004 0.005 0.002 https://doi.org/10.1371/journal.pone.0284821.t002 0.25 0.37 0.59 0.57 0.36 0.57 0.73 0.63 0.49 0.014 0.033 0.015 0.029 0.019 0.023 0.025 0.025 0.008 6268 3104 4372 2567 4761 2252 2135 3660 395 128 542 141 324 202 168 235 6,663 3,232 4,914 2,708 5,085 2,454 2,303 3,895 29,119 2,135 31,254 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 9 / 23 PLOS ONENumeracy skills learning of children in Africa Fig 3. Numeracy test scores in low- and high-numeracy skills countries (median, 25th and 75th percentile). https://doi.org/10.1371/journal.pone.0284821.g003 in DRCongo. The mean numeracy scores in DRCongo and Sierra Leone are significantly lower than all the five countries with higher scores. Hypothesis H1 on the large variation in average numeracy skills performance among children across African countries is supported. We suggest dividing our sample into two groups: the low-numeracy countries group (DRCongo, Sierra Leone, and The Gambia) and the high-numeracy country group (Ghana, Lesotho, Togo, Tunisia, and Zimbabwe). Table 2 shows that non-disabled children answered 57% of the questions correctly, and the disabled sample answered 49% correctly. The descriptive statistics in S3 Table demonstrate that children with hearing and vision disabilities answered more questions correctly than non- disabled children. In contrast, the correct response rates for children with other disabilities are much lower. We present the test score distributions (median, p25, and p75) for the low- numeracy countries (left figure) versus the high-numeracy countries (right figure) by disability types as vertical box plots in Fig 3. The mean test scores with 95% confidence intervals by dis- ability type are presented in Fig 4. With the split sample, the sample size is too small for reliable statistical analysis for children with vision and hearing disabilities, as shown in S1 Table. Therefore, we restrict our split sample analysis to children with physical, intellectual, and mul- tiple disabilities. Figs 3 and 4 indicate the significant disparities in numeracy tests not only between the two groups of countries but also between children with and without disabilities. Disabled children lag in numeracy skills performance in both groups of countries. However, descriptive data sug- gest that disabled children benefit from improved school quality since disabled children in high-numeracy skills countries perform even better than non-disabled children in low-numer- acy skills countries. The question is whether disabled children gain as much as non-disabled children in learning numeracy skills when the learning environment has improved. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 10 / 23 PLOS ONENumeracy skills learning of children in Africa Fig 4. Mean numeracy test scores with 95% confidence intervals for the means in low- and high-numeracy skills countries. https://doi.org/10.1371/journal.pone.0284821.g004 4.2 Disability effect with age control The first set of regressions aims to test hypothesis H2, which states that children with disabili- ties perform significantly worse than their non-disabled peers of the same age in learning numeracy skills. Without considering the causal mechanisms, we start with a parsimonious specification, including age, country and gender dummy variables stepwise as control vari- ables. The regression results are presented in Table 3. The constant term in Model 1 suggests that the estimated average score is 0.31 for 7-year- old control children. Children’s numeracy skills improve with age, probably related to their access to schooling. Model 2 shows effective numeracy skills variation across countries. To evaluate the numeracy skills gap over countries, we run separate regressions with age dummies for non-disabled children in each country (S4 Table). DRCongo is the country with the lowest numeracy skills, where the average numeracy score is only 0.106 for 7-year-old children, while Tunisia has the highest average numeracy score of 0.77 for 7-year-old children. The country dummy variable parameters and their significance levels illustrate large variations in school quality across countries in their performance in enhancing children’s average numeracy skills. Finally, gender is not significantly correlated with children’s numeracy skills performance. It indicates that girls are not discriminated against in the school systems in a way that affects their basic numeracy skills. The coefficients on the disability status in model 1 in Table 3 show a significant and nega- tive disability effect on children’s numeracy skills for children with physical, intellectual, and multiple disabilities. However, the estimated disability effect for children with physical disabili- ties turns insignificant after controlling for the macro country dummy (models 2 and 3). In PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 11 / 23 PLOS ONENumeracy skills learning of children in Africa Table 3. Regression results for disability effects on the mean numeracy test score. Model 1 Model 2 Model 3 Disability status Vision disability Hearing disability Physical disability Intellectual disability Multiple disabilities Age 8 9 10 11 12 13 14 Country (base category: DRCongo) The Gambia 0.121* (0.024) -0.002 (0.036) -0.068* (0.020) -0.072* (0.010) -0.213* (0.026) 0.128* (0.007) 0.242* (0.007) 0.284* (0.007) 0.355* (0.008) 0.382* (0.007) 0.420* (0.007) 0.439* (0.007) Ghana Lesotho Sierra Leone Togo Tunisia Zimbabwe Constant Sample size Gender (1 = girl, 0 = boy) 0.305* (0.005) 31254 0.028 (0.021) -0.049 (0.031) -0.019 (0.017) -0.109* (0.009) -0.205* (0.024) 0.127* (0.006) 0.231* (0.007) 0.277* (0.006) 0.337* (0.007) 0.371* (0.007) 0.398* (0.006) 0.415* (0.006) 0.147* (0.012) 0.327* (0.010) 0.298* (0.009) 0.066* (0.010) 0.274* (0.011) 0.501* (0.008) 0.396* (0.008) 0.106* (0.007) 31254 0.029 (0.021) -0.049 (0.031) -0.019 (0.017) -0.109* (0.009) -0.205* (0.024) 0.127* (0.006) 0.231* (0.007) 0.277* (0.006) 0.337* (0.007) 0.371* (0.007) 0.398* (0.006) 0.415* (0.006) 0.147* (0.012) 0.327* (0.010) 0.298* (0.009) 0.066* (0.010) 0.274* (0.011) 0.501* (0.008) 0.396* (0.008) 0.003 (0.003) 0.105* (0.007) 31254 (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 12 / 23 PLOS ONENumeracy skills learning of children in Africa Table 3. (Continued) R2 Model 1 0.171 Model 2 0.373 Model 3 0.373 Significance levels: * p<0.05; p<0.01; * p<0.001 https://doi.org/10.1371/journal.pone.0284821.t003 contrast, it becomes larger for children with intellectual disabilities after controlling for the country dummy. The country effect might be important for evaluating the disability effect for children with physical and intellectual disabilities. The first set of regressions supports hypoth- esis H2 that children with physical, intellectual, and multiple disabilities perform significantly worse than their non-disabled peers of the same age in learning numeracy skills. However, hypothesis H2 is not supported for children with vision and hearing disabilities. 4.3 IV models with endogenous completed school years We will now more closely study the causal mechanisms for the links between the exogenous disability (treatment) variables and the outcome. The disability effect on numeracy skills may come from reduced school participation or a lower ability to acquire numeracy skills while in school. To analyse this, we run IV models with completed school years as the endogenous exposure to schooling on the split samples for each disability type. We run IV models with age and gender as instruments. To test the strength of the two instruments and assess the endogeneity of completed school years in the model, we first run a set of regressions, presented in S5 Table. All the models in S5 Table suggest that age and gender are significantly associated with the completed school years. Moreover, the disability effects on completed school years vary a lot across disability types, which suggests potentially high endo- geneity of the completed school years. Furthermore, the regressions in section 4.2 suggest that gender does not directly affect children’s numeracy skills. The regression results and IV tests are shown in Table 4. The OLS model results are pre- sented for the non-disabled when the IV tests are invalid. For the models that satisfy the tests, we find the following results. The first-stage regression indicates that children with vision or hearing disabilities do not lag in completed school years compared to non-disabled children. However, children with physical, intellectual, or multiple disabilities have completed signifi- cantly fewer school years than non-disabled children per year of age. The return to each completed school year in numeracy skills score is estimated at 0.146 units for non-disabled children in the IV model and 0.142 in the OLS model, noting that the overidentification test failed for this IV model. For the other IV models, the statistical validity could not be rejected. For children with vision, hearing, physical, and intellectual disabilities, there is no significant disability effect on numeracy skills returns to completed school years. Hypothesis H3a, which states that children with vision and hearing disabilities perform well in numeracy skills return to schooling compared to non-disabled children, cannot be rejected. However, H3b and H3c, which state that children with physical or intellectual disabilities have a lower return to schooling in numeracy skills than non-disabled children, are not supported. The estimated return to each completed school year is 0.142 (CI:0.140–0.144) for non-dis- abled children and 0.121 (CI:0.105–0.137) for children with multiple disabilities. Significant disability effects of 0.121–0.142 = -0.021 score points for each completed school year are reported for children with multiple disabilities. Hypothesis H3d that children with multiple disabilities have the lowest return to schooling regarding numeracy skills cannot be rejected. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 13 / 23 PLOS ONENumeracy skills learning of children in Africa Table 4. Regressions on the mean numeracy score by disability types. Completed school years (base category: 1) Sample size OLS for non- disabled 0.142* (0.000) 29119 First stage regressions (Dep: Completed school year) Ln(age) Gender (1 = girl, 0 = boy) Constant IV test Robust Wu-Hausman test (p value) Sargan IV validity test (p-value) Strength (First stage F test) IV (separate model for each disability type) Non- disabled 0.146* (0.001) 29119 Vision disabled 0.147* (0.005) 159 7.599* 8.440* (0.044) 0.021 (0.376) 0.108 (0.019) -13.954* (0.193) -15.371* (0.092) (0.838) 0.000 0.000 21324.1 0.000 0.960 381.4 Hearing disabled 0.143* (0.009) 87 7.970* (0.641) 0.340 (0.336) -15.064* (1.444) 0.000 0.989 145.54 Physical disabled 0.151* (0.006) 401 6.523* (0.379) 0.106 (0.136) -11.902* (0.789) 0.000 0.560 302.73 Intellectual disabled 0.145* (0.002) 1308 6.731* (0.209) -0.025 (0.090) -12.268* (0.450) 0.000 0.349 1497.77 Multiple disabled 0.121* (0.008) 180 5.637* (0.637) -0.020 (0.287) -10.203*** (1.374) 0.060 0.126 149.46 Instrumented: Completed school year. Instruments: ln(age) and gender dummy. Significance levels: * p<0.05; p<0.01; * p<0.001, based on the standard errors which allow for intragroup correlation https://doi.org/10.1371/journal.pone.0284821.t004 4.4 IV models for low- and high-numeracy skills countries The results in Table 3 show that there might be a country effect when evaluating the overall disability effect for children with physical and intellectual disabilities. This might indicate het- erogeneous disability effects across the eight African countries. To further explore the disabil- ity effects for different disability types, we run IV regressions after dividing the sample into low- and high-numeracy skills country groups as defined in section 4.1. The sample sizes of the split samples by country numeracy skills level and disability status only allow for the analy- ses of three disability types (physical, intellectual, and multiple disabled). The regressions are run on the split samples of the non-disabled and disabled for each of the three specific disabil- ity statuses in the countries with low and high numeracy skills, respectively. The results are presented in Table 5. We then graph the regression coefficients with 95 per cent confidence intervals to present the first stage estimated completed school year by age (Fig 5) and the second stage estimated numeracy skills return to completed school years (Fig 6) over different disability types in low- and high-numeracy skills country groups. Fig 5 indicates that in both groups of countries, the mean estimated completed school years by age for intellectually disabled children and multiple disabled children are significantly lower than those for non-disabled children. Children with physical disabilities have also completed fewer school years than non-disabled children, but the differences are not significant. The gap in completed school years in the low-numeracy skills country group is higher than in the high-numeracy skills group. Fig 6 suggests that the mean estimated numeracy skills return to each completed school year in low-numeracy skills countries is 0.132 (CI: 0.130–0.134) score points for non-disabled children. In contrast, it is estimated to be 0.152 (CI: 0.150–0.154) score points in the high- numeracy skills country group. Children with physical or intellectual disabilities are not signif- icantly different from non-disabled children in numeracy skills return to schooling. In con- trast, the mean estimated numeracy returns are 0.107 (CI: 0.082–0.132) and 0.129 (CI: 0.111– PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 14 / 23 PLOS ONENumeracy skills learning of children in Africa Table 5. Regressions on the mean numeracy score in low- and high-numeracy skills country group. Low-numeracy skills group High-numeracy skills group OLS for intellectual disabled 0.128* (0.005) IV (separate model for each disability type) Non- disabled Physical disabled Intellectual disabled Multiple disabled OLS for non- disabled IV (separate model for each disability type) Non- disabled Physical disabled Intellectual disabled Multiple disabled 0.132* 0.141* (0.001) (0.009) 0.138* (0.005) 0.107* 0.152* 0.155* 0.166* (0.013) (0.001) (0.001) (0.009) 0.148* (0.003) 0.129* (0.009) Completed school years (base category: 1) Sample size 435 14133 268 435 93 14986 14986 133 873 87 First stage regressions (Dep: Completed school year) Ln(age) Gender (1 = girl, 0 = boy) Constant IV test 6.580* 5.782* 4.738* 3.664* (0.075) -0.042 (0.479) 0.096 (0.364) -0.014 (0.983) -0.030 8.330* (0.046) 0.135* 7.731* 7.376* 6.977* (0.542) 0.349 (0.222) 0.098 (0.778) -0.154 (0.029) -12.183* (0.163) -10.581* (0.156) (0.988) (0.150) -8.523* (0.762) (0.358) -6.530* (2.159) (0.021) -15.142* (0.209) -14.054* (0.098) -13.434* (0.332) -12.378*** (0.097) (1.119) (0.487) (1.647) Robust Wu-Hausman test (p value) Sargan IV validity test (p-value) 0.000 0.656 0.000 0.655 Strength (First stage F test) 5193.02 159.57 0.000 0.004 334.46 0.568 0.694 40.59 0.000 0.000 24230.81 0.000 0.465 191.2 0.000 0.452 1483.36 0.004 0.123 143.04 Instrumented: Completed school year. Instruments: Ln(age) and gender Significance levels: * p<0.05; p<0.01; * p<0.001, based on the cluster-robust standard errors https://doi.org/10.1371/journal.pone.0284821.t005 0.147) for children with multiple disabilities in countries with low- and high-numeracy skills. The gap between non-disabled children and children with multiple disabilities in the low numeracy countries -0.025 (= 0.107–0.132) is marginally higher than that -0.023 (= 0.129– 0.152) in the high numeracy countries. Furthermore, numeracy skills return to schooling for children with physical 0.166 (CI: 0.148–0.184) or intellectual disabilities 0.148 (CI: 0.142– 0.154) in high-numeracy skills countries are significantly higher than that of the non-disabled children 0.132 (CI: 0.130–0.134) in low-numeracy skills countries. It indicates that disabled children benefit as much from higher school quality as non-disabled children do. Finally, the numeracy skills performance is predicted for a 14-year-old child by disability status in both low- and high-numeracy skills groups in Fig 7. The endogenous school year dif- ferences, as well as differences in return to each endogenous school year in both stages, are taken into consideration. The total effects of disability on numeracy skills for 14-year-old chil- dren are negative and significant for both intellectual and multiple disabled children in coun- tries with low- and high-numeracy skills. The predicted mean numeracy skill for children with intellectual disability is 0.547 (CI: 0.504–0.590) in low-numeracy skills countries and 0.899 (CI: 0.869–0.930) in high-numeracy skills countries, which is significantly lower than that for non-disabled children of 0.679 (CI: 0.669–0.688) and 1.073 (CI: 1.065–1.081) in low- and high-numeracy skills countries. The disability effect for children with intellectual disability are -0.13 (= 0.547–0.679) and -0.17 (= 0.899–1.073) in low- and high-numeracy skills countries, and that for children with multiple disabilities are -0.34 and -0.30, respectively. For those with physical disabilities, there is no significant disability effect in low- or high-numeracy skills groups. The cross-country difference in predicted numeracy skills for a 14-year-old non-disabled child is about 0.4 points between low- and high-numeracy skills country groups, which is PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 15 / 23 PLOS ONENumeracy skills learning of children in Africa Fig 5. First-stage regression coefficients of age on completed school years with 95% confidence intervals (IV regression on numeracy skill return to each completed school year, separate IV models for three disability types in low- and high- numeracy skills country groups). https://doi.org/10.1371/journal.pone.0284821.g005 marginally higher than the estimated numeracy skills gaps across disability types, as discussed above. Furthermore, 14-year-old children with intellectual disabilities in high-numeracy skills countries show significantly better numeracy skills performance (0.90) than the non-disabled children in the low-numeracy skills group (0.68). The average score of non-disabled children in the low-numeracy skills countries (0.68) is even below the average numeracy score for the most challenged multiple-disabled children in high-numeracy skills countries (0.77). These findings do not support hypothesis H4a, that children with disabilities are less capa- ble of benefiting from the better quality of the school system than non-disabled children. Dis- abled children do benefit substantially from improved school quality. The gap between non- disabled and disabled children in numeracy skills is smaller than the variation across countries, which supports hypothesis H4b. 5. Discussion We will now summarise our findings for the key hypotheses and discuss our results related to the relevant literature and earlier studies. The first hypothesis (H1) states a considerable varia- tion in average numeracy skills across the eight African countries we have studied. Our analy- ses reveal large variations in average numeracy skills across countries based on nationally representative data; thus, we cannot reject this hypothesis. The large sample sizes provide accu- rate estimates of mean numeracy skill scores by country since they have confidence intervals in the range of 0.01–0.015 around the mean numeracy skills scores, ranging from the lowest PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 16 / 23 PLOS ONENumeracy skills learning of children in Africa Fig 6. Outcome regression coefficients of completed school years on numeracy scores with 95% confidence intervals (IV regression on numeracy skill return to each completed school year, separate IV models for three disability types in low- and high-numeracy skills country groups). https://doi.org/10.1371/journal.pone.0284821.g006 0.35 in DRCongo to the highest 0.88 in Tunisia. It indicates considerable variation in the aver- age quality of school systems across these eight countries regarding their ability to teach chil- dren numeracy skills. Our second hypothesis (H2) that disabled children perform worse than their non-disabled peers in numeracy skills was supported for children with physical, intellectual, and multiple disabilities but not those with vision and hearing disabilities. To our knowledge, almost no similar study has evaluated disabled children’s numeracy skills in the African context. The only exception is the study by Bakhshi, Babulal, and Trani (2018) from Sudan [6]. The other few earlier papers in the developing context are mainly from Asia, with the study of Takeda and Lamichhane (2018) from India [24], Malik et al. (2020) and Singal et al. (2020) from Paki- stan [25, 26]. Most studies have applied the Washington Group definition of disabilities. Bakh- shi, Babulal, and Trani (2018) used a disability screening questionnaire (DSQ-35), and Takeda and Lamichhane (2018) revised the WG module to a large extent. The age range of children included in the learning assessment test also varies. The two studies in Pakistan use the ASER (Annual Status of Education Report) test on reading and math. Takeda and Lamichhane (2018) use reading, math and writing test in the Indian Human Development Survey (IHDS). Bakhshi, Babulal, and Trani (2018) use simple self-reporting assessments. Despite the dispari- ties of these studies, most studies reported a performance gap between disabled and non-dis- abled children, except the study by Bakhshi, Babulal, and Trani (2018). Our findings provide evidence in the African context, suggesting a gap in numeracy skills between disabled and non-disabled children, which varies across disability types. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 17 / 23 PLOS ONENumeracy skills learning of children in Africa Fig 7. Predicted numeracy skills performance by disability status for an average 14-year-old child in both low- and high- numeracy skills groups. https://doi.org/10.1371/journal.pone.0284821.g007 Little empirical evidence has been available for heterogenous disability effects on school performance by disability types in the African context. Our results suggest that children with vision and hearing disabilities do not have lower numeracy skills than non-disabled children, which supports our hypothesis H3a. It is not the case for children with other disabilities. Also, based on the WG definition of disabilities, a study in Pakistan by Singal et al. (2020) is one of the few studies differentiating the disability types, which uses the ASER (Annual Status of Edu- cation Report) test [26]. They report that children with moderate or severe sensory disabilities (walking, seeing and hearing) have the lowest level of basic numeracy skills. Singal et al. (2018) transferred the test scale to a very low threshold dichotomy variable and only evaluated whether children could identify one-digit numbers. It might explain the special challenges for children with sensory disabilities compared to children with other disabilities. Our study does not find challenges for children with vision and hearing disabilities, but it does not mean they are not exposed to additional risks in school performance. The numeracy test embedded in the MICS survey might not fully capture the potential risk for those with vision and hearing dis- abilities to learn more advanced numeracy skills. Earlier studies on the numeracy skills differences have not specifically differentiated the mechanisms behind possible disability effects. Such effects could simply be caused by the lack of school attendance, or they could be related to disabled children’s low returns to schooling in numeracy skills. We separate the two types of disability effects by IV models to control for the endogeneity of completed school years for each disability type. In their study in India, Takeda and Lamichhane (2018) suggest that disabled children are less likely to fall behind in skills once they access education [24]. They made this conclusion because they noticed that when PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 18 / 23 PLOS ONENumeracy skills learning of children in Africa the interaction between disability status and school status is included in the model, the disabil- ity dummy becomes insignificant. However, they did not consider the potential endogeneity of schooling. Our IV model shows that low numeracy scores among the physical- and intellectually dis- abled children are mainly attributable to the low school years they manage to complete but are not constrained by their numeracy skills returns to schooling. Hypotheses H3b and H3c state that children with physical or intellectual disabilities have a lower return to schooling in numeracy skills (after controlling for differences in completed school years) compared to non- disabled children. Our results do support the two hypotheses. These findings suggest that school enrolment is especially crucial for children with disabilities to gain equal access to edu- cation. On the other hand, children with multiple disabilities have not only completed the least school years but also have the lowest numeracy skill returns per completed school year among children with various disability types, which supports hypothesis H3d. Finally, hypothesis H4a states that the gap in numeracy skills between non-disabled and dis- abled children is larger in high-numeracy skills countries. However, our study shows that the overall gap between children with and without disabilities in terms of numeracy skills, consid- ering both effects of endogenous school year differences and differences in school return to each school year, is not significantly different between low- and high-numeracy skills coun- tries. It does not provide evidence of a broader gap in school performance for disabled children when the school quality is improved. Therefore, we reject hypothesis H4a. Bakhshi, Babulal, and Trani (2018) found in their study in West Darfur of Sudan that when all the children are exposed to low-quality schools in a conflict context, there is no difference in numeracy skills between the disabled and non-disabled children [6]. By controlling the endogeneity of completed school years, we find that in low- and high-numeracy skills coun- tries, most children with disabilities (except children with multiple disabilities) do not lag sig- nificantly in gaining numeracy skills if they complete the same schooling as the non-disabled children. Their main challenge is low school enrollment, especially in countries with poor school quality. The estimated numeracy skills return to schooling among children with physical or intellec- tual disabilities in high-numeracy skills countries are significantly higher than that of the non- disabled peers in low-numeracy skills countries. The variation in numeracy skills performance is higher across countries than over disability types. We cannot reject hypothesis H4b, that the average numeracy skills of non-disabled children vary more across countries with different school system quality than the gap between non-disabled and disabled children. The variation across countries can be even higher if more countries are included, which suggests the quality of the school system is the key to improving school performance in Africa. 6. Conclusion Based on large-scale nationally representative samples in the eight African countries, we assess the within- and across-country variation in numeracy skills and the gap between children with and without disabilities. The Washington Group Child Functional Module (WG-CFM) and standard numeracy test are applied to all the countries and ensure the credibility of this com- parison study. We identify two types of disability effects using IV models to control for the endogeneity of completed school years. These models allow us to divide the numeracy skill dif- ferences into the difference in completed school years and the difference in numeracy skill returns per completed school year. Combining these two effects results in systematic variations in the overall numeracy skill performance across disability types. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 19 / 23 PLOS ONENumeracy skills learning of children in Africa First, we find a considerable variation in average numeracy skills across the eight African countries (hypothesis H1 is supported). Second, there is systematic variation in numeracy skills across disability types (hypothesis H2 is supported). More specifically, children with vision and hearing disabilities perform as well as non-disabled children, while children with physical, intellectual and multiple disabilities lag behind. Third, the reason why children with different disability types lag behind varies (hypotheses H3a and H3d are supported; hypotheses H3b and H3c are rejected). Those with physical and intellectual disabilities lag because they, conditional on age, have completed fewer school years. Those with multiple disabilities lag both due to fewer completed school years and due to lower numeracy skill returns per school year. Furthermore, based on average performance, we find that the within-group average differ- ences in numeracy skill returns to school between non-disabled and disabled children are sim- ilar between the low- and high-numeracy country groups (hypothesis H4a is rejected). More importantly, the difference in average performance between high-performing and low-per- forming countries is larger than the within-country group difference in performance between non-disabled and disabled children categories (hypothesis H4b is supported). Disabled chil- dren in the high numeracy skill countries perform even better than the non-disabled children in the low numeracy skill countries. Except for children with multiple disabilities characterised by low enrolment and low numeracy skill returns to completed school years, the main challenge for most children with disabilities is the low school enrolment. This is especially the case for children in low-numer- acy skill countries. This suggests that the priority for the education policy in low-income Afri- can countries is to improve children’s school enrolment, especially for children with disabilities. The fact that the within-group differences between children with and without dis- abilities are similar between the low- and high-numeracy skill country groups suggests that disabled children benefit equally when the school quality improves. It demonstrates substantial room for improvement in the school system, and such enhancements also benefit disabled children. Disability effects in numeracy skills across country groups are more fundamental than the within-group gaps. Therefore, improving overall school quality and promoting school attendance for disabled children are crucial for better school performance among disabled children in the African context. We acknowledge several limitations of this study. First, the study is limited to numeracy skills and may not be correlated with other benefits from schooling. Second, the study assesses fundamental numeracy skills and may have failed to capture substantial variation in more advanced numeracy skills that may vary more, especially among the older children that the test can identify. The test may be more appropriate in low-numeracy countries and for younger children. In high-numeracy countries, about 40% of all children answered over 90% of the questions correctly. It might lead to a potentially underestimated disability effect in high- numeracy countries. Third, while the MICS surveys are nationally representative samples aim- ing to provide data on the general population of children, the incidence of severe disability is very low in the population. Therefore, we have merged some categories to achieve sufficient sample sizes for statistical analysis. Last, the eight African countries included in this study are the countries that recently conducted the sixth round of the MICS survey. We do not know the external validity of the conclusions drawn based on the data from these eight countries. For future research, we recommend studies with a broader range of skills, such as reading and science skills, and tests with more advanced numeracy skills for older children. Covering more countries with large samples may also be possible to do more statistical analyses for more disaggregated samples with rare forms of disabilities. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 20 / 23 PLOS ONENumeracy skills learning of children in Africa Supporting information S1 Table. The sample size of children who have done the numeracy test by disability status and country. (PDF) S2 Table. Regression results for estimating the determinant factors of each disability type. (PDF) S3 Table. Sample characteristics. (PDF) S4 Table. Regressions on the mean numeracy score with age dummy by countries (non-dis- abled children). (PDF) S5 Table. Regression on the completed school years. (PDF) Acknowledgments This paper has been undertaken as part of the research project “Education outcome variability in children with disabilities: Structure, institution or agency?”. Valuable comments were received from the project coordinator Anne Hatløy and other project partners and the partici- pants in the 44th Annual Meeting of the Norwegian Association of Economists, September 25– 26, 2022. The authors take full responsibility for any remaining errors. Author Contributions Conceptualization: Huafeng Zhang, Stein T. Holden. Data curation: Huafeng Zhang. Formal analysis: Huafeng Zhang, Stein T. Holden. Methodology: Huafeng Zhang, Stein T. Holden. Software: Huafeng Zhang. Supervision: Stein T. Holden. Writing – original draft: Huafeng Zhang. Writing – review & editing: Huafeng Zhang, Stein T. Holden. References 1. UN. Sustainable Development Goals: Department of Economic and Social Affairs; 2015 [Available from: https://sdgs.un.org. 2. Lewin KM. Access to education in sub-Saharan Africa: patterns, problems and possibilities. Compara- tive Education. 2009; 45(2):151–74. 3. UNESCO. Education for All 2000–2015: Achievements and Challanges. Paris: UNESCO, 2015. Ke´ pze´ s e´s gyakorlat. 2016; 14(1–2):283–7. 4. UN. Millennium Development Goals. 2000. 5. UNESCO. Global Education Monitoring Report 2020: Inclusion and Education–All Means All. Unesco Paris; 2020. 6. Bakhshi P, Babulal GM, Trani J-F. Education and disability in a conflict affected context: Are children with disabilities less likely to learn and be protected in Darfur? World Development. 2018; 106:248–59. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 21 / 23 PLOS ONENumeracy skills learning of children in Africa 7. UNESCO. Teaching policies and learning outcomes in Sub-Saharan Africa: issues and options; Sum- mary. 2016. 8. Johnson D, editor The changing landscape of education in Africa: Quality, equality and democracy 2008: Symposium Books Ltd. 9. Spaull N, Taylor S. Effective enrolment–Creating a composite measure of educational access and edu- cational quality to accurately describe education system performance in sub-Saharan Africa. Stellen- bosch EconomicWorking Papers. 2012; 21(12):1–25. 10. Unterhalter E. Education targets, indicators and a post-2015 development agenda: Education for All, the MDGs, and human development. The power of numbers: A critical review of MDG targets for human development and human rights. 2013. 11. Bonal X., Education Poverty, and the “Missing Link”. The Handbook of Global Education Policy: John Wiley & Sons, Ltd; 2016. p. 97–110. 12. Ansong D, Ansong EK, Ampomah AO, Afranie S. A spatio-temporal analysis of academic performance at the Basic Education Certificate Examination in Ghana. Applied Geography. 2015; 65:1–12. 13. Adugna M, Ghahari S, Merkley S, Rentz K. Children with disabilities in Eastern Africa face significant barriers to access education: a scoping review. International Journal of Inclusive Education. 2022:1–17. 14. Gregorius S. Exploring narratives of education: disabled young people’s experiences of educational institutions in Ghana. Disability & Society. 2016; 31(3):322–38. 15. UNESCO. Education and Disability: Analysis of Data from 49 Countries. United Nations Educational, Scientific and Cultural Organization. 2018. 16. Stromquist NP. The Learning Generation: Investing in Education for a Changing World; A Report by the International Commission on Financing Global Education Opportunity. September 2016. 176 pp. Free distribution online at http://report.educationcommission.org/report. Comparative Education Review. 2017;61(1):214–7. 17. Filmer D., Disability Poverty, and Schooling in Developing Countries: Results from 14 Household Sur- veys. The World Bank Economic Review. 2008; 22(1):141–63. 18. Mizunoya S, Mitra S, Yamasaki I. Disability and School Attendance in 15 Low-and Middle-Income Countries. World Development. 2018; 104:388–403. 19. Kuper H, Saran A, White H, Kumar ST, Tolin L, Muthuvel T, et al. Rapid Evidence Assessment (REA) of What Works to Improve Educational Outcomes for People with Disabilities in Low-and Middle-Income Countries. International Centre for Evidence in Disability, London School of Hygiene and Campbell Col- laboration. 2018. 20. 21. 22. Lewis E, Mitra S, Yap J. Do Disability Inequalities Grow with Development? Evidence from 40 Coun- tries. Sustainability. 2022; 14(9):5110. Luo Y, Zhou RY, Mizunoya S, Amaro D. How Various Types of Disabilities Impact Children’s School Attendance and Completion-Lessons Learned from Censuses in Eight Developing Countries. Interna- tional Journal of Educational Development. 2020; 77:102222. Tolar TD, Fuchs L, Fletcher JM, Fuchs D, Hamlett CL. Cognitive Profiles of Mathematical Problem Solv- ing Learning Disability for Different Definitions of Disability. J Learn Disabil. 2016; 49(3):240–56. https:// doi.org/10.1177/0022219414538520 PMID: 24939971 23. Murphy MM, Mazzocco MM, Hanich LB, Early MC. Cognitive characteristics of children with mathemat- ics learning disability (MLD) vary as a function of the cutoff criterion used to define MLD. J Learn Disabil. 2007; 40(5):458–78. https://doi.org/10.1177/00222194070400050901 PMID: 17915500 24. Takeda T, Lamichhane K. Determinants of schooling and academic achievements: Comparison between children with and without disabilities in India. International Journal of Educational Develop- ment. 2018; 61:184–95. 25. Malik R, Raza F, Rose P, Singal N. Are children with disabilities in school and learning? Evidence from a household survey in rural Punjab, Pakistan. Compare: A Journal of Comparative and International Education. 2020; 52(2):211–31. 26. Singal N, Sabates R, Aslam M, Saeed S. School enrolment and learning outcomes for children with dis- abilities: findings from a household survey in Pakistan. International Journal of Inclusive Education. 2020; 24(13):1410–30. 27. Zhang H, Holden S. Disability Types and Children’s Schooling in Africa. SSRN Electronic Journal. 2022. 28. Pieters S, Desoete A, Roeyers H, Vanderswalmen R, Van Waelvelde H. Behind mathematical learning disabilities: What about visual perception and motor skills? Learning and Individual Differences. 2012; 22(4):498–504. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 22 / 23 PLOS ONENumeracy skills learning of children in Africa 29. Pieters S, Roeyers H, Rosseel Y, Van Waelvelde H, Desoete A. Identifying Subtypes Among Children With Developmental Coordination Disorder and Mathematical Learning Disabilities, Using Model- Based Clustering. J Learn Disabil. 2013; 48(1):83–95. https://doi.org/10.1177/0022219413491288 PMID: 23757349 30. Zarfaty Y. The Performance of Young Deaf Children in Spatial and Temporal Number Tasks. Journal of Deaf Studies and Deaf Education. 2004; 9(3):315–26. https://doi.org/10.1093/deafed/enh034 PMID: 15304434 31. Morgan PL, Farkas G, Wu Q. Kindergarten children’s growth trajectories in reading and mathematics: who falls increasingly behind? J Learn Disabil. 2011; 44(5):472–88. https://doi.org/10.1177/ 0022219411414010 PMID: 21856991 32. Crollen V, Warusfel H, Noe¨l M-P, Collignon O. Early visual deprivation does not prevent the emergence of basic numerical abilities in blind children. Cognition. 2021; 210:104586. https://doi.org/10.1016/j. cognition.2021.104586 PMID: 33477011 33. Zhang S, Xia X, Li F, Chen C, Zhao L. Study on Visual and Auditory Perception Characteristics of Chil- dren with Different Type of Mathematics Learning Disability. International Journal of Disability, Develop- ment and Education. 2019; 68(1):78–94. 34. Chan LKS, Dally K. Learning disabilities and literacy & numeracy development. Australian Journal of Learning Disabilities. 2001; 6(1):12–9. 35. Jimenez BA, Stanger C. Math Manipulatives for Students with Severe Intellectual Disability: A Survey of Special Education Teachers. Physical Disabilities: Education and Related Services. 2017; 36(1):1–12. 36. Shifrer D. Stigma and stratification limiting the math course progression of adolescents labeled with a learning disability. Learning and Instruction. 2016; 42:47–57. 37. Bernal P, Mittag N, Qureshi JA. Estimating effects of school quality using multiple proxies. Labour Eco- nomics. 2016; 39:1–10. 38. Singh R, Sarkar S. Does teaching quality matter? Students learning outcome related to teaching quality in public and private primary schools in India. International Journal of Educational Development. 2015; 41:153–63. 39. Heyneman SP, Loxley WA. The Effect of Primary-School Quality on Academic Achievement Across Twenty-nine High- and Low-Income Countries. American Journal of Sociology. 1983; 88(6):1162–94. https://doi.org/10.1086/227799 PMID: 6614303 40. Soylu F, editor Mathematical cognition as embodied simulation. Proceedings of the Annual Meeting of the Cognitive Science Society; 2011. 41. UNICEF. Module on Child Functioning concept note. 2017. 42. Groce NE, Mont D. Counting Disability: Emerging Consensus on the Washington Group Questionnaire. The Lancet Global Health. 2017; 5(7):e649–e50. https://doi.org/10.1016/S2214-109X(17)30207-3 PMID: 28619216 43. WG. An Introduction to Washington Group on Disability Statistics Question Sets. 2020. 44. Rosenzweig MR, Wolpin KI. Natural”Natural Experiments” in Economics. Journal of Economic Litera- ture. 2000; 38(4):827–74. 45. Banks LM, Kuper H, Polack S. Poverty and Disability in Low-and Middle-Income Countries: A System- atic Review. PloS One. 2017; 12(12):e0189996. https://doi.org/10.1371/journal.pone.0189996 PMID: 29267388 46. Blackburn CM, Spencer NJ, Read JM. Prevalence of childhood disability and the characteristics and cir- cumstances of disabled children in the UK: secondary analysis of the Family Resources Survey. BMC Pediatrics. 2010; 10(1). https://doi.org/10.1186/1471-2431-10-21 PMID: 20398346 47. Trani J-F, Bah O, Bailey N, Browne J, Groce N, Kett M. Disability in and around Urban Areas of Sierra Leone. 2010. 48. Groce N, Kett M. The Disability and Development Gap. Leonard Cheshire Disability and Inclusive Development Centre Working Paper Series No 21. London, UK: LCDIDC; 2013. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284821 April 20, 2023 23 / 23 PLOS ONE
10.1371_journal.ppat.1011341	RESEARCH ARTICLE The structure of a 12-segmented dsRNA reovirus: New insights into capsid stabilization and organization Qinfen ZhangID Haidong Xu1, Jianguo He1, Jason T. Kaelber4, Shaoping Weng1, Wen Jiang5 1☯, Yuanzhu Gao1☯, Matthew L. Baker2,3☯, Shanshan Liu1☯, Xudong Jia1, 1 State key lab for biocontrol, School of Life Sciences, Sun Yat-sen University, Guangzhou, China, 2 Department of Biochemistry and Molecular Biology, Structural Biology Imaging Center, McGovern Medical School at the University of Texas Health Science Center, Houston, Texas, United States of America, 3 Verna and Marrs McLean Department of Biochemistry and Molecular Biology, Baylor College of Medicine, Houston, Texas, United States of America, 4 Institute for Quantitative Biomedicine, Rutgers, The State University of New Jersey, Piscataway, New Jersey, United States of America, 5 Markey Center for Structural Biology, Department of Biological Sciences, Purdue University, West Lafayette, Indiana, United States of America ☯ These authors contributed equally to this work. lsszqf@mail.sysu.edu.cn (QZ); lsswsp@mail.sysu.edu.cn (SW); jiang12@purdue.edu (WJ) a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Abstract Citation: Zhang Q, Gao Y, Baker ML, Liu S, Jia X, Xu H, et al. (2023) The structure of a 12- segmented dsRNA reovirus: New insights into capsid stabilization and organization. PLoS Pathog 19(4): e1011341. https://doi.org/10.1371/journal. ppat.1011341 Editor: John T. Patton, Indiana University Bloomington, UNITED STATES Received: December 28, 2022 Accepted: April 2, 2023 Published: April 21, 2023 Copyright: © 2023 Zhang et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Density maps of qMCRV and tMCRV with icosahedral symmetry and D5 symmetry have been deposited in the EMDataBank (EMDB) under entries EMD-33403 for icosahedral qMCRV, EMD-33404 for D5 symmetric qMCRV, EMD-33405 and EMD-33406 for icosahedral and D5 symmetric tMCRV, respectively. Atomic models for the asymmetric unit of MCRV for both icosahedral and D5 reconstruction have been deposited in the Protein Data Bank (PDB) under entries 7XR2 and 7XR3 respectively. Infecting a wide range of hosts, members of Reovirales (formerly Reoviridae) consist of a genome with different numbers of segmented double stranded RNAs (dsRNA) encapsu- lated by a proteinaceous shell and carry out genome replication and transcription inside the virion. Several cryo-electron microscopy (cryo-EM) structures of reoviruses with 9, 10 or 11 segmented dsRNA genomes have revealed insights into genome arrangement and tran- scription. However, the structure and genome arrangement of 12-segmented Reovirales members remain poorly understood. Using cryo-EM, we determined the structure of mud crab reovirus (MCRV), a 12-segmented dsRNA virus that is a putative member of Reovir- ales in the non-turreted Sedoreoviridae family, to near-atomic resolutions with icosahedral symmetry (3.1 Å) and without imposing icosahedral symmetry (3.4 Å). These structures revealed the organization of the major capsid proteins in two layers: an outer T = 13 layer consisting of VP12 trimers and unique VP11 clamps, and an inner T = 1 layer consisting of VP3 dimers. Additionally, ten RNA dependent RNA polymerases (RdRp) were well resolved just below the VP3 layer but were offset from the 5-fold axes and arranged with D5 symme- try, which has not previously been seen in other members of Reovirales. The N-termini of VP3 were shown to adopt four unique conformations; two of which anchor the RdRps, while the other two conformations are likely involved in genome organization and capsid stability. Taken together, these structures provide a new level of understanding for capsid stabiliza- tion and genome organization of segmented dsRNA viruses. Author summary Mud crab reovirus (MCRV), a putative member of the Sedoreoviridae family in Reovirales, infects an economically important crab, Scylla serrata, and has resulted in serious PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 1 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus Funding: This work was supported by the National Natural Science Foundation of China to QZ (NSFC No. 31570736 and 31672677), Science and Technology Program of Guangzhou (No. 201707020003) to QZ, the China Agriculture Research System CARS-48 to SW. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. economic loss for the crab aquaculture. In this study, we report the structure of MCRV, which has allowed us to examine the detailed structural elements that might be responsi- ble for capsid stability, as well as those involved in genome packaging and RdRp organiza- tion. The arrangement of viral polymerases inside the capsid differ from what has been seen in other viruses of the same order, as ten of the vertices are occupied by RdRp in a symmetrical organization not seen elsewhere. In addition, the VP11 protein is analogous to the “clamp” protein of turreted reovirads. Introduction Infecting a wide range of hosts including marine protists, fungi, plants, insects, crustaceans, fish, birds, and mammals, members of the order Reovirales consist of 9–12 segments of dsRNA genome encapsulated by a proteinaceous shell [1]. Infection can result in human diseases and mortality, as well as posing a significant impact to the economy. The order Reovirales is divided into two families, Spinareoviridae and Sedoreoviridae, based on whether they encode for turrets (protrusions or spikes) atop the 5-fold vertices of the virion. The family Sedoreoviri- dae, which includes human pathogens, such as rotaviruses, contains “nonturreted” viruses with 2 or 3 layers of relatively smooth protein shell. A non-turreted virus consisting of a 12-segment dsRNA genome, mud crab reovirus (MCRV), which may belong to a new order of Reovirales [2], infects the economically impor- tant crab Scylla serrata [2,3]. Like other members of Reovirales, MCRV packages its dsRNA segments and a number of RNA-dependent RNA polymerase (RdRp) inside a double-layered icosahedral capsid [2,4]. Underneath an outer T = 13 layer, 120 copies of VP3 form the inner shell; two copies of VP3 in slightly different conformations (termed A and B) are found in each asymmetric unit, and often referred to as a “pseudo T = 2” lattice. This organization of the inner shell is conserved not only in the members of Reovirales, but also among members of other dsRNA virus families, such as Totiviridae, Partitiviridae, Picobirnaviridae, and Cystoviri- dae; it also bears resemblance to the Chrysoviridae structure. For unknown reasons, all but the members of Birnaviridae and Polymycoviridae of dsRNA viruses and none of the non-dsRNA viruses have this core arrangement [5–7]. Electron cryo-microscopy (cryo-EM) and X-ray crystallographic structures of several dsRNA viruses, such as mammalian reovirus (MRV) [8] and cytoplasmic polyhedrosis virus (CPV) [9], have indicated a potential role of N-terminal extensions of the VP3-like subunits in connecting inner core subunits, contributing to the stabilization of the capsid. In previous studies of reovirus core particles, large regions of the N-terminus in VP3 homologs were disor- dered [8,10,11]. It was postulated that the N-terminal regions of the inner capsid proteins might also serve as transcriptional regulating factors for RdRps [12,13]. Across members of Reovirales, the RdRps, located on the interior of the inner capsid protein layer, show a conserved finger-palm-thumb core surrounded by N- and C-terminal elabora- tions, creating a cage-like structure, with four tunnels used for RNA template entry, nucleoside triphosphate (NTP) entry, template exit, or RNA transcript exit [14,15]. Recent reports found ten of the twelve vertices occupied in the 10-segmented CPV [16,17], and eleven of the twelve vertices occupied in the 11-segmented Grass Carp Reovirus (GCRV) [13,18], organized with pseudo-D3 symmetry (S1 Table). These previously reported results are in agreement with a model where each genome segment is specifically associated with one RdRp [19]. Thus, a 10-segmented CPV and 11-segmented GCRV would package 10 and 11 copies of RdRp, and accordingly, occupy 10 and 11 of the twelve 5-fold vertices, respectively. Interestingly, 10 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 2 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus vertices were occupied by RdRp density, arranged with pseudo-D3 symmetry, in Fako virus (FAKV) despite having only nine genome segments [20]. Moreover, genome-free particles of Reovirales family members still contain the normal number of RdRp [20–23], raising the ques- tion of whether capsid and steric constraints collaborate in packaging a set number of RdRps at fixed positions regardless of genomic content. For twelve-segmented dsRNA viruses, like MCRV, the organization of the RdRps inside the capsid remains an outstanding question, which necessitates a structural solution to fill the gap in addressing the packaging and organi- zation of Reovirales family members with 9, 10, 11, or 12 dsRNA genome segments. Here, we present the near-atomic resolution structures of both transcriptionally quiescent MCRV (qMCRV) and actively transcribing MCRV (tMCRV) resolved by single particle cryo- EM with and without icosahedral symmetry imposed. We report distinct conformations for the VP3 molecules clustered around the five-fold pore, suggesting how the N-termini of VP3 stabilize the inner core, as well as how these termini hold the RdRp complex in place and pro- vide a structural framework for regulation of RdRp activity. While previous works have shown that these conformationally variable N-termini can break icosahedral symmetry, this study reveals the organization of these extensions in their asymmetric context, shedding new insights into the interactions that guide the structure and function of the viral protein shell and RdRps. Furthermore, this work clearly defines the presence of ten RdRps with a unique D5 symmetric organization in the intact virus, despite having twelve dsRNA segments in MCRV. The atomic model of the in situ RdRps also provides new molecular insights into capsid interaction and genome organization. Results Overall structure of qMCRV The three-dimensional structure of the qMCRV was first determined to 3.1 Å resolution with icosahedral symmetry (Figs 1A and S1). The map clearly shows an outer and inner capsid layer; the inner capsid layer is a thin shell composed of 60 copies of a VP3 dimer (Fig 1B), while the outer capsid, arranged on a T = 13 lattice, is thicker and composed of VP12 trimers (Fig 1B) organized in a similar way to the previously reported low-resolution reconstruction [3]. Interdigitating these trimers are two smaller densities per asymmetric unit and assigned to be VP11 (Fig 1B). The resolution of the map was sufficient to clearly define the subunit bound- aries and protein folds, as well as the majority of the side chain density in each of the protein subunits. RdRps are arranged with D5 symmetry in MCRV In an attempt to reveal the molecular mechanism of capsid stability, dsRNA genome arrange- ment and the relationship between the capsid and the genome replication/transcription machines, an asymmetric reconstruction of MCRV was also performed without imposing any symmetry (S2 Fig) [24,25]. Like the icosahedral reconstruction, the capsid proteins in the two layers of the asymmetric reconstruction could be clearly delineated in the density maps. Additionally, densities near the 5-fold vertices and immediately interior to the inner capsid layer were well resolved and assigned as the RdRps (S2 Fig). Visual inspection of the asymmetric MCRV reconstruction revealed 10 well resolved RdRps around 10 of the 12 icosahedral vertices (henceforth referred as “RdRp vertices”) while the densities around the remaining 2 vertices at the opposite side of the icosahedral center (henceforth referred to as “pole vertices”) were much weaker. To quantitatively evaluate the arrangement of the RdRps, the asymmetric reconstruction was rotated to all 60 icosahedral PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 3 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus Fig 1. Overall structure of mud crab reovirus (MCRV). (A) A radially colored, shaded surface representation of the MCRV icosahedral reconstruction as viewed along a 5-fold axis is shown. The icosahedral asymmetric unit contains 13 copies of VP12 arranged as 5 trimers: P, Q, R, S and T. Trimer T locates around the icosahedral three-fold axis and thus contributes a monomer to the asymmetric unit. The icosahedral asymmetric unit is shown in (B). (B) The left panel shows an outside view, while the right panel shows the interior view. The inner capsid proteins (VP3A and VP3B), clamp protein (VP11) and outer capsid protein (VP12) are colored in cornflower blue, orange, green and medium purple, respectively. 5-fold, 2-fold and 3-fold axes are indicated by black pentagon, ellipse and triangle respectively. (C) The D5 reconstruction of MCRV indicates that there are 10 well-resolved densities corresponding to VP1 (RdRp) located near 10 vertices and arranged with D5 symmetry. Under the pole vertices (arrows indicated), the density of expected VP1 is too weak to be annotated. VP1 is highlighted in dark red. The densities of other proteins are in grey. (D) The density map of one vertex indicates the arrangements of major capsid proteins in the D5 reconstruction are the same as that of icosahedral reconstruction. The interior view (right) illustrates that the only difference between the icosahedral and D5 reconstructions is the dark-red colored VP1 near the 5-fold vertex. The left panel is an outside view while the right panel is an interior view. The color scheme is the same with that of Fig 1B and 1C. https://doi.org/10.1371/journal.ppat.1011341.g001 symmetry related views to position each of the 12 vertices at the north pole in 5 azimuthal views (12×5 = 60 views in total), and the similarities among all possible combinations (60×59/ 2 = 1770) of the 60 views of the north-pole vertex were computed. The similarity matrix was then subjected to MDS (multi-dimensional scaling) manifold analysis to position all 60 views of the vertices according to their collective similarity distribution. The MDS analysis revealed 6 apparent clusters with each cluster consisting of 10 views of the vertices (S3A Fig). Clusters 1–5 were much tighter when compared to cluster 6, suggesting that the 10 views within each of the first 5 clusters were very similar in structure. The 10 views in each of the first 5 clusters cor- responded to the 10”RdRp vertices”, with each vertex contributing 1 of its 5 azimuthal views. Cluster 6 consisted of the 10 views of the 2”pole vertices” with weak densities; its larger spread suggested significant differences in structure among those 10 views. This also suggested that the two”pole vertices” densities in cluster 6 were also distinct from the RdRp densities in the first 5 clusters. If they were not different, these 2 vertices would be structurally similar to the first 5 clusters and included in those clusters, thereby increasing the cluster size from 10 to 12. The large gaps among clusters 1–5 represent significant structural differences between the 5-fold axis related views of the vertex with a single RdRp. Additionally, analysis of the distribu- tion of angles between all pairs (10×9/2 = 45) among the 10 views in each of the 6 clusters PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 4 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus (S3B Fig) revealed that the inter-view relationships were quantitively consistent with D5 sym- metry but not other symmetries, including D3 (S3C Fig). To further validate the RdRp organization, we employed a decoy mapping method [20], whereby synthetic maps containing RdRps in various configurations (decoys) are used to assay the symmetry in a density map. Here, the decoys consisted of random placements of 8 RdRps (only one RdRp per five-fold axis), a CPV-like configuration of 10 RdRps, a D5 configuration of 10 RdRps, a D5 configuration plus one or two polar RdRps, and a D5 configuration of 10 RdRps where each RdRp is rotated 144˚. When compared to these decoys, MCRV particles matched best with the D5 configuration as discussed above. Particles aligned to the decoys with the D5 configuration retained the input D5 configuration; particles aligned to the CPV- like configuration also converged to the same D5 configuration. When particles were matched to each of the 5 possible positions in decoys that contained the ten equatorial RdRps, they do not appear to match better to any one RdRp position. Moreover, when iteratively refined against these five positions in a multi-model refinement, the RdRps still show no preference to any of the five positions. However, when particles are iteratively refined against a D5 decoy with one polar RdRp added (total of 11 RdRps), the output structure has the same 11-RdRp configuration as the input structure; no twelfth RdRp appears at the antipodal vertex. As such, the decoy analysis on the MCRV RdRp arrangement is consistent with D5 symmetry men- tioned above. Taken together, these quantitative analyses suggest that our asymmetric recon- struction, including the RdRps, has a unique D5 symmetry with the 5- and 2-fold axes sharing the corresponding icosahedral axes (S3D Fig), not pseudo-D3 symmetry as reported in CPV [17], FAKV [20] etc. Furthermore, we analyzed ~3300 “empty” MCRV particles (their genome was absent). After symmetry expansion and 3D classification were performed, most particles were classified into 6 classes with 10 obvious RdRp densities arranged with D5 symmetry; densities at the remaining two vertices were substantially weaker (S4A and S4B Fig). To quantitate the occu- pancy at each potential site, we masked out a region of the RdRp that does not overlap with a neighboring 5-fold-related potential RdRp site. For each of the 60 potential RdRp sites, we measured the density enclosed by this mask. The D5 RdRp sites were over 100 times as intense as neighboring, sterically-occluded sites, but the polar sites were intermediate in intensity: 18% as intense as the 10 highly-occupied sites and 22 times more intense than the unoccupied sites. This is most consistent with additional, polar RdRp choosing randomly among the five-fold- related binding sites, although it remains possible that only 10 RdRp are present and polar density is from misaligned particles, or from other proteins. The remaining empty MCRV particles were classified into other additional classes, whose densities corresponding to the expected location of the RdRp’s were blurred. Again, we employed a decoy mapping method to validate the RdRp’s arrangement in the empty capsids with RdRps. The decoy set consisted of the closest neighboring 5 RdRps (only one RdRp per five-fold vertex) with random orientations, and the particles matched best with the D5 configu- ration once more (S4C Fig). Imposing D5 symmetry on the C1 map preserved the RdRp densities, further verifying the D5 symmetry arrangement of the RdRps. Subsequent 3D reconstruction with D5 symmetry imposed resulted in a 3.4 Å resolution map (Figs 1C–1D and S1). All copies of the structural proteins in the D5 reconstruction were almost identical to those in the corresponding icosahe- dral map, with the exception of the well resolved ten RdRps near the ten 5-fold vertices (Figs 1C–1D and S1C). A complete de novo atomic model for the MCRV asymmetric unit (2 copies of VP3, 2 cop- ies of VP11, 13 copies of VP12) and RdRp was constructed directly from the icosahedral and D5 symmetrized maps (S5–S8 Figs and S1 Movie). PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 5 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus Fig 2. Structure of inner capsid protein VP3. (A) Different conformations of the inner capsid protein VP3A (cornflower blue) and VP3B (orange). Both VP3A and VP3B contain apical (red ellipse), carapace (green triangle) and dimerization domains (purple rectangle). VP3B has an extended N-terminus composed of 2 helices and connecting loops. (B) The five “helix-loop-helix-loop” N-termini of VP3B are indicated using balls and highlighted in different colors. They interact with different VP3 dimers and form a “belt” that stabilizes the inner shell. (C) The N-terminus of VP3A differs from that of VP3B. Three VP3A N-termini, highlighted by purple balls, have similar conformations, while the other two VP3A locations have different conformations (in red and cyan respectively) in the D5 reconstruction. These two N-termini brace the polymerase (gray density) and may help to hold the polymerase to the inner surface of the capsid. The major parts of VP3A and VP3B are colored in cornflower blue and orange respectively, same as that in Fig 1B. The alignment of all VP3A show in (D) indicates that all VP3As are nearly identical, except for the different conformations of the N-terminus. (E) The alignment of the N-terminus of VP3A and VP3B clearly show four different possible N-terminal configurations for VP3. In (F), the large separations among adjacent VP3 pentamers can be seen. (G) A zoomed-in view of the boxed area in (F) shows the large separation at the VP3 dimerization domains. The black oval indicates the 2-fold axis. https://doi.org/10.1371/journal.ppat.1011341.g002 Structure and interactions of the inner capsid protein, VP3 A VP3 dimer, containing two chemically identical subunits (A and B) with slightly different conformations in an asymmetric unit, forms the thin inner shell of the MCRV (Figs 1B, 2A and S1C). VP3A and VP3B have nearly identical folds, similar to what is seen in inner capsid proteins of other members of Reovirales, forming a flat crescent like structure, divided into three distinct domains (Figs 2A and S5): an apical domain (residues 280–665) composed pri- marily of α-helices located closest to the 5-fold vertices, a carapace domain (residues 86–279, 666–686 and 819–854), and a dimerization domain (residues 687–818), composed mostly of β-sheets and loops, that interacts with both the neighboring VP3 subunit, as well as VP3 sub- units from neighboring asymmetric units. While the overall fold of the two VP3 subunits are similar, there is considerable difference between the neighboring A and B subunits (~6.9 Å RMSD from residues 86–854). These dif- ferences are mainly located in the dimerization and carapace domains (S6 Fig). Specifically, residues 717–750, 760–778, 789–811 at the distal edge of the dimerization domain and residues 258–271 in the carapace domain are shifted by more than 10 Å when the VP3 subunits are aligned (S6 Fig). An additional major structural difference (>10 Å RMSD) is also seen on the inner surface of the carapace domain at residues 608–633 (S6 Fig). In VP3B, the N-terminus (residues 1–85) consists of a helix-loop-helix-loop motif (Fig 2A) that traverses across the adjacent VP3A, extending towards a neighboring dimer (Fig 2B). Spe- cifically, helix 1 (residues 1–26) of the N-terminus extends across two adjacent VP3 molecules, PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 6 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus while helix 2 (residues 54–73) interdigitates the carapace domain of the adjacent VP3A subunit between two α-helices (residues 614–629 and 128–145). The insertion between these two heli- ces likely causes the aforementioned shift in the carapace domain, specifically near residues 608–629. This also likely influences structural rearrangements in the dimerization domain as the α-helix located at residues 128–145 sits at the boundary of the carapace and dimerization domains. With its elongated structure, one VP3B N-terminus interacts with three other VP3 monomers. Residues 39–74 of one VP3B pass through the adjacent VP3A and form an inter- woven structure, “locking” the dimer together. Residues 26–38 occupy the middle part of a VP3B molecule from the neighboring dimer, while residues 1–26 insert into the groove of the neighboring dimer’s VP3A. In context, the five N-terminal arms of VP3B form a “belt” around the five-fold vertex that function to link and stabilize the pentamer of VP3 dimers (Fig 2B). The VP3A N-terminus does not interact with VP3B in the same manner. Rather, based on the D5 reconstructions and corresponding atomic models, it adopts three distinct conforma- tions (Fig 2C–2E). In three of the five symmetry-related VP3A subunits, an α-helix from resi- dues 61–77 bridges the apical and carapace domains, extending towards the adjacent VP3B subunit (Fig 2C). No detectable density is evident in the maps for residues prior to this helix. In one of five VP3A subunits (VP3A1), the aforementioned α-helix becomes shorter (residues 70–81) (Fig 2C–2E). Residues 61–69 form a loop and turn ~30˚ away, protruding inward and extending down to interact with the RdRp (Fig 2C–2E). The third N-terminal conformation is seen in an adjacent VP3A (VP3A2), where an α-helix extends from residues 50–76 and includes a long loop from residues 40–50 (Fig 2C–2E). It is reasonable to deduce that these var- ied conformations in VP3A1 and VP3A2 may play roles in maintaining the polymerase posi- tion and are further discussed later. A striking feature of the VP3 shell is the large separation around each 2-fold axis (Fig 2F and 2G). This separation is approximately rhomboidal in shape with sides of 65 Å and 20 Å; this is significantly larger than the corresponding region in other Reovirales members. With a gap larger than 8 Å precluding any direct contact, interactions between two dimers of different decamers at the 2-fold axis are exceedingly unlikely. Due to this, it would be nearly impossible for VP3 to maintain stable interactions across two neighboring decamers, and thus there must be additional interactions to bridge neighboring decamers in forming a stable inner capsid (discussed below). As such, the term “dimerization domain” is a misnomer when applied to MCRV VP3. Structure and interactions of the major outer capsid protein, VP12 VP12, the primary outer shell trimer protein (Fig 1B), is comprised of a β-sandwich (two β- sheets of four antiparallel strands) atop an all α-helical base (Figs 3A–3B and S7). In an asym- metric unit of the T = 13 outer capsid layer, VP12 forms 4⅓ unique trimers (Fig 1A and 1B), in which each subunit in the trimer is twisted about the other (Fig 3A). The topology of the β- sandwich domain is virtually identical among MCRV, rotavirus [26], rice dwarf virus [11], bluetongue virus [10] and African horse sickness virus [27], though spinareoviruses, such as aquareovirus (ARV) [28], differ considerably. MCRV VP12’s β-sandwich domain is incom- plete, with strands A”BID and A’CHE as opposed to A”BIDG and A’CHEF in rotavirus [26] (Fig 3B). Interestingly, all four loops of the MCRV β-sandwich are found in representative sedoreoviruses (A’A”, BC, DE, and HI in the rotavirus/bluetongue virus nomenclature) [26,29]. The A’A” and BC exterior loops give rise to a hydrophobic concavity of roughly 8 Å in width, opening to Phe118 and Val135 (Fig 3B). The position of this pocket at the outer extreme of the capsid radius, as well as its shape and hydrophobicity, might relate to cellular interac- tions with MCRV; however, this needs further investigation. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 7 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus Fig 3. Structure of VP12 and interactions within the VP12 trimers. (A) The model of a VP12 trimer is shown. Three monomers are in medium purple, goldenrod and gray respectively. (B) A single VP12 subunit is annotated. The strand nomenclature is indicated using the upper-case letters A to I. The exterior loops of strands A’A” and BC give rise to a roughly 8Å wide hydrophobic concavity indicated by a red asterisk. (C) A zoomed-in view of the orange boxed area in (A) is shown. Gly164 in one VP12 interacts with the Ser223 and Asn226 of a neighboring VP12 by hydrogen bonds. (D) A zoomed-in view of the green boxed area in (A) illustrates that the PPPG motif in the linker loop interacts with Lys96 of one VP12 and the Arg31 of a neighbor VP12 via hydrogen bonding. There are multiple hydrogen bonds, such as the one between Arg273 and Lys32 of an adjacent VP12. https://doi.org/10.1371/journal.ppat.1011341.g003 In the outer capsid, a major contributor to intra-trimer interactions can be found at the edges of the β-sandwich in the VP12 trimers, where a loop containing a di-glycine is stabilized by the hydrogen bonding of Gly164 with Ser223 and/or Asn226 of a neighboring VP12 (Fig 3C). A second interaction between neighboring VP12 trimers is mediated by linkers connect- ing the α-helical base to the β-sandwich (Fig 3D). Each of residues 92–102 is within 5 Å to the neighboring trimer; largely polar contacts are seen between sidechain and mainchain atoms. As an example, the Lys96 amine group is positioned to form a hydrogen bond with the car- bonyl group of Arg31 in the neighboring trimer, whose amine group reaches back to interact with Pro97 in a stretch of three consecutive prolines (Pro97-Pro98-Pro99) (Fig 3D). The PPPGF motif in this linker can assume α-helical (i.e. PDB-4OZV [30]), turn (i.e. PDB-5C33 [31]), or polyproline-II conformations (i.e. PDB-4P45 [32]) depending on the structural con- text [33]. While several other members of Reovirales VP12 homologs also have two or more prolines in this linker region, only MCRV contains a polyproline-II motif here. This secondary structure is known to render the backbone more solvent-exposed, which may be related to the extensive mainchain involvement in trimer-trimer interactions. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 8 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus Structure and interactions of the outer capsid protein VP11 Unique to MCRV, the outer capsid layer contains a second capsid protein, VP11 (S8A Fig). Two VP11 subunits per asymmetric unit are embedded between VP12 trimers and bridge multiple asymmetric units (Figs 1B and 4). One of the two VP11 subunits (VP11A) bridges the adjacent asymmetric units around the same 5-fold vertex and block the narrow opening, previ- ously denoted as a type II channel [34], while the other (VP11B) is located along the icosahe- dral 2-fold axis and block the type III channel (Fig 4A and 4B). In both cases, each VP11 is bound by six VP12 trimers (Fig 4C), three from each asymmetric unit, at very different inter- faces. The two VP11 subunits are almost identical (RMSD ~0.79Å over 203 residues), having some slight conformation difference localized to the peripheral loops and termini, likely a result of their different environments (S8B Fig). VP11 has an overall tetrahedral shape, consisting mostly of α-helices and two small β-sheets (Figs 4D, S8A and S8C). Additionally, VP11 contains a core hydrophobic α-helix (Leu82-- Tyr98) with five aromatic residues (Phe88, 91; Tyr93, 94 and 98) and five leucines (Leu86, 87, 89 and 95), around which several other helices are wrapped (Fig 4E). The two small β-sheets serve as exterior elaborations interacting with the VP12 trimers. Sequence and structural searches revealed no VP11 structural homologs, indicating that this protein contains a novel fold. While the exact function of VP11 remains to be confirmed, some parallels can be drawn to other Reovirales family members in an attempt to infer the role of VP11. In other members of Reovirales with clamps, the analogous VP11 positions are largely occupied by clamp proteins that appear to engage many residues of neighboring VP12-like tri- mers. In λ1, the clamp binds at three distinct locations within each icosahedral asymmetric Fig 4. Locations and structure of VP11 implies that it may function as a clamp protein in MCRV. (A) The outside view of MCRV capsidis shown. In (B), a zoomed-in view of the boxed regions in (A) can be seen; relative orientation of this view can be inferred based on the indicated location of the 5-fold vertex. Type I, II and III channels are labeled and show the potential aqueous channel positions. The color scheme is the same as that of Fig 1B. VP11 (green) not only inter-digitates the VP12 trimers, the two copies of VP11s in an asymmetric unit also block type II (VP11A) and III (VP11B) channels. (C) VP11 sits in the middle of the six VP12 trimers and interacts with the 6 VP12 trimers in different manners, clamping these trimers together. (D) The atomic model of VP11 is shown in rainbow color. The colors from blue to red indicate the N-terminal to C-terminal. (E) VP11 has a core hydrophobic α-helix (L82-Y98) with five aromatic residues (red). Panel (F) shows the same view in (A) after removing all VP12 trimers. Panel (G) shows a zoomed-in view of the boxed region in (F). VP11 rests on top of VP3 dimers from the inner capsid. (H) VP11B is located on the top of the separation between VP3 dimers from adjacent pentamers, interacting with VP3A from one pentamer and VP3B from the other pentamer. Through these interactions, VP11B clamps the two neighbor pentamers together. The color schemes are the same as that in Fig 1B, except for that in (D). https://doi.org/10.1371/journal.ppat.1011341.g004 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 9 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus unit [8] through interactions that are broadly conserved among spinareoviruses. As such, VP11 may in fact be a clamp protein, despite the lack of detectable sequence or structural simi- larity to other viral clamp proteins. There is an abundance of hydrophobic amino acids at the interface between VP11 and VP12, resulting in a number of potential hydrophobic interactions (S9A and S9C Fig). In addi- tion, several hydrogen bonds are also found at the VP11/VP12 interface (S9B Fig). The inter- face area between VP11 and six VP12 is about 2,472 Å2. Through these extensive interactions, VP11A may “clamp” the six adjacent VP12 trimers together. VP11A also interacts with the inner capsid protein, VP3. A group of polar amino acids are seen at the interface between VP11A and VP3 (S9D and S9F Fig), as well as several potential hydrogen bonds (S9E Fig). A striking feature of MCRV is the large separation in the inner capsid underneath VP11B at the type III channel as mentioned above (Fig 4F and 4G). VP11B may provide the “missing” interactions to clamp the neighboring VP3 inner capsid proteins that have no direct interac- tions among different decamers. The two ends of VP11B straddling this separation interact with the equivalent sites in the dimerization domain (an α-helix at residues 734–744 and a loop at 696–702) of VP3 (Figs 4H and S9G), even interacting with some of the same VP3 resi- dues (Asn697 and Asp699) (S9I Fig). These interactions are mediated by hydrophobic interac- tions and hydrogen bonds at the interfaces between VP11B and VP3B (S9H–S9J Fig). Together, the extensive interactions between VP12 and VP3 dimers, as well as VP11’s interac- tions with VP12 likely stabilize the inner VP3 layer despite the large separation in subunits. Both the clamp proteins of GCRV [35] and CPV [9] engage the same elements of the dimer- ization domain, but the interacting regions of the clamp involved show little resemblance to that of MCRV VP11. The clamps of GCRV do not have interactions with the outer capsid. As genes typically evolve with higher conservation of fold topology than protein-protein interac- tions, the history of this orthologous set is unusual and may reflect the unique demands of viral capsids. Structure of VP1, the RdRp A hallmark of the MCRV non-icosahedral (C1 and D5) (S2 and S3 Figs) reconstruction is the ten well-resolved RdRp densities located slightly offset (~30 Å) from ten of the twelve icosahe- dral vertices that are organized with D5 symmetry (Fig 1C and 1D). Along the remaining two icosahedral vertices (coincidental with the 5-fold axis), a small amount of density is seen at sig- nificantly lower thresholds. As with the other protein subunits in the MCRV reconstructions, it was possible to construct atomistic models for the RdRp. The structure of the RdRp revealed a compact, cage-like structure ~75 Å in diameter with an extended protrusion near its innermost surface (Fig 5A). This cage-like structure is a com- mon feature in RdRps seen in λ3 of reovirus [15], VP1 of rotavirus [12,14,36], and the RdRps of cypovirus [16,17], aquareovirus [13,18] and bluetongue virus [37] (S10 Fig). The MCRV RdRp can be divided into three domains: an N-terminal domain (residues 1–617), a central polymerase domain (residues 618–1090), and a C-terminal bracelet domain (residues 1091– 1422) (Fig 5). The central polymerase domain can be further divided into three sub-domains: a finger sub-domain including residues 618–775 and 823–896, a palm sub-domain composed of residues 776–822 and 897–1008, and a thumb subdomain that contains residues 1009–1090 (Fig 5B). As with other reovirus RdRps, the arrangement of these domains forms a relatively hollow center that is presumably the catalytic center of the RdRp. As in all polymerases, the central polymerase domain contains the core catalytic residues. In particular, D819, D824, D947 and D948 form the conserved acidic residues in the active site and residues 1003–1012 form the “primer grip” which was thought to be close to the phosphate PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 10 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus Fig 5. Structure of VP1 (RdRp). (A) The model of the RNA dependent RNA polymerase (RdRp), VP1, is shown. Like other Reovirus polymerases, VP1 contains an N-domain (yellow), C-domain (purple) and polymerase domain. The polymerase domain is further divided into the finger subdomain (blue), palm subdomain (dark red) and thumb subdomain (green). The Polymerase domain of the RdRp is shown in (B). The purple arrow indicates the primer grip, the black arrow indicates the priming loop and the orange arrow indicates the unique long loop blocking the gap between the thumb and finger subdomain. (C) The N-domain has a unique protrusion (black circle). (D) The C- domain is shown. The red arrow indicates the C-terminal plug. The overall view of the inner VP3 layer and the RdRp protein VP1 is shown in (E), while (F) shows a zoomed-in view of the green box in E. The VP3A1 N termini (residues 59–69) (red arrow) interacts with the finger subdomain of the RdRp VP1 (blue). (G) A zoomed-in view of the black box area in E is shown. N-terminal residues (40–48, black arrow) of VP3A2, and residues 1220–1226 (green arrow), 67–74 (blue arrow) and 1390–1395 (red arrow) of VP1 form an anti-parallel β-sheet. (H) The zoomed-in view of the cyan-dashed box in F demonstrates the interactions between the apical domain of VP3B and the VP1. Residues 318– 346 of VP3B apical domain (orange) form a helix-loop-helix structure along the inner surface and interact with residues 642–657 (blue) of finger subdomain of an RdRp. Residues Lys642 and His646 in the RdRp appear to form a salt bridge with Glu318 of VP3B. https://doi.org/10.1371/journal.ppat.1011341.g005 that joins the nucleotides at the primer terminus [38] (Fig 5B). Also, the priming loop, residues 776–796, is seen in the ‘down’ conformation in our MCRV structure (S10 Fig). In rotavirus VP1, the down conformation is incapable of supporting initiation of transcription in the prim- ing site [14]. Also like rotavirus VP1, the finger sub-domain lacks a four-stranded β-sheet which protrudes from the surface of the λ3 of reovirus. However, a long loop from residues 675–719 in the MCRV blocks the gap between the thumb and finger subdomains (Figs 5B and S10), a feature unique to the MCRV RdRp. The N-terminal domain is mainly helical and forms a “closed” continuous surface on one side of the RdRp that bridges the finger and thumb sub-domains (Fig 5C). This surface is almost perpendicular to the inner capsid layer though only two small stretches of residues in this domain (67–74 and 523–545) appear to interact with a single VP3A. A protrusion on the innermost surface of the RdRp (residues 324–374) extends outward, which is a unique feature not found in other Reovirales polymerases (Figs 5C and S10). This protrusion, along with another small portion of the N-terminal domain and finger subdomain forms an open tunnel (S11A Fig), which is likely the RNA template entry tunnel [15]. Nearby is another tunnel, simi- lar to the reported NTP entry tunnel [15], composed of two helices (556–574 and 269–279) and a small β-sheet from this N-terminal domain along with the palm and finger subdomain (S11A Fig). Like the N-terminal domain, the C-terminal domain is composed primarily of α-helices and bridges the thumb and finger subdomains on the opposite side of the N-terminal domain PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 11 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus (Fig 5D). This domain is similar to those reported in other reovirus polymerase structures [14]. In λ3 and rotavirus VP1, the C-terminal domain encircles the dsRNA and -RNA template exit tunnel. Likewise, the C-terminal domain of the MCRV RdRp appears to play a similar role. In the MCRV RdRp, a loop and small α-helix from residues 1330–1339 form a relatively compact plug-like structure in this putative exit tunnel, resulting in a tunnel diameter of ~9– 10 Å (Figs 5D and S11A). In the previously reported rotavirus VP1 structure, a short C-termi- nal α-helix occupies a similar position and extends ~15 Å into the (-)RiNA/dsRNA exit tunnel [14] though no plug is seen in λ3 [15] (S10 Fig). It should be noted however that two helices (residues 1283–1293, 1294–1304), also partially restrict the exit tunnel in MCRV RdRp. λ3 and rotavirus VP1 do not appear to contain analogous structures. Additionally, a transcription product exit tunnel is located along the RdRp surface nearest to the VP3 inner capsid layer (S11A Fig). This tunnel consists primarily of positively charged residues at the intersection of the C-terminal domain and the finger and palm subdomains. Interestingly, this tunnel is in line with a gap in the VP3 layer created by two VP3A and one VP3B. Interactions between the inner capsid and RdRp While the N-termini of VP3B appears to be involved in the stabilization of the VP3 decamer and the inner capsid, the structural features also suggest that the N-termini of VP3A may play critical roles in “holding” the RdRp to the inner capsid; biochemical encapsidation assays also suggested a similar role for the inner capsid protein in rotavirus [39,40]. At the N-terminus of VP3A1, residues 59–69 form a loop that facilitates interactions with the RdRp (Figs 2C, 5E, 5F and 5G). The extended N-terminus of the VP3A2 interacts with the RdRp, with residues 40– 48 forming an anti-parallel β-sheet with the residues 1220–1226 and 1390–1395 of the RdRp (Fig 5G). Together, these two unique conformations of N-termini in VP3A essentially bracket opposite sides of the RdRp and likely help anchor the polymerase to the capsid. While the RdRp is located next to the 5-fold vertices, it is offset such that only three of the five VP3AB dimers contact the RdRp (Figs 2C and 5E). The interface is made up of primarily of the N-terminal and C-terminal RdRp domains and the apical domains of the three VP3As, though a portion of the carapace domain is involved in two of three VP3A subunits (VP3A1 and VP3A2). In the VP3B apical domain, residues 318–346 form a helix-loop-helix structure along the inner surface interacting with the residues 642–657 of the RdRp (Fig 5H). Of particu- lar note, Lys642 and potentially His646 in the RdRp appear to form a salt bridge with Glu318 of VP3B (Fig 5H). Along with the aforementioned N-termini, these interactions likely help to anchor the RdRp in its offset position. Structure of tMCRV In some members of Reovirales, capsids expand while transcribing [41], while others need to remove the outer capsid layer to activate transcription [42]. To better understand the structural changes of MCRV during transcription, we determined the structure of MCRV in a transcrip- tionally active state (tMCRV) with icosahedral symmetry and D5 symmetry imposed at 3.36 and 3.7 Å resolution, respectively. Interestingly, at the current stated resolutions, we could not find any obvious conforma- tional change for capsid proteins, including the inner capsid protein (VP3) and the outer cap- sid proteins (VP11 and VP12), nor was there any obvious change in capsid diameter. Furthermore, there were no gross structural differences in VP1 (the RdRp) despite the change in transcription state. Despite the lack of differences in the structural proteins, we did observe the RNA strands in the background of micrographs (S11C Fig) and extra densities in tMCRV map when PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 12 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus Fig 6. The extra densities in the transcribing MCRV (tMCRV). (A) The difference map between tMCRV and qMCRV shows the extra density (light blue) at the 5-fold vertex that has RdRp density underneath. The color schemes for models VP12 and VP3A are same as that of Fig 1B. The extra density is displayed at a contour level of about 8.3. (B) The extra density together with the RdRp of tMCRV is shown. Part of two copies of VP3A model is shown to indicate the position of the inner capsid. The extra density appears to be an RNA transcription product, together with the noncontinuous, non-template single strand RNA (also see S11 Fig). The RdRp density is displayed at a contour level of about 2.6. The color scheme for RdRp is same as that in Fig 5. In right panel, red dotted line indicates the transcript product exiting the RdRp and then through the gap among five VP3As at the 5-fold vertex. https://doi.org/10.1371/journal.ppat.1011341.g006 compared to our qMCRV map. First, obvious extra densities were observed between the VP3 and VP12 subunits at the ten vertices (Fig 6A). After masking away capsid density, extra densi- ties could be found in and around the RdRp of tMCRV (Figs 6B–6C and S11B). The density near the template entry site is consistent with double-strand RNA (dsRNA) with a diameter of about 30Å diameter, followed by a density of smaller diameter—likely a single RNA strand (Figs 6B and S11B)—inside the RdRp. It is reasonable to annotate this density inside the RdRp of tMCRV as the template strand. Continuing further from the template strand, the density becomes stronger and out of the RdRp and again has features of dsRNA. It could be annotated as the upstream dsRNA (Figs 6B, S11B and S11D–S11E). Opposite of the putative template strand and pointing to the transcript exit, the extra density becomes smaller in diameter again. This slim density travels along the surface of RdRp and through a gap between the VP3A sub- units at the 5-fold vertex (Figs 6C, S11D and S11E). We annotated this density as the transcrip- tion product. Similar features have been observed in the transcribing CPV and [16,43], supporting the annotations of these densities and their relative roles in transcription. These features are similar to the reported elongation state of the CPV’s transcription [43], and may indicate that most particles we observed in the reconstruction could be in the transcription elongation state as well. Arrangement of the dsRNA genome Like all the members in the Reovirales, the dsRNA genome of MCRV is highly organized and packaged inside the inner capsid around the RdRps (S12 Fig). In cross-sectional views, the D5 map clearly depicts at least six concentric genome layers organized around the ten RdRps with the outermost layer appearing to interact with the inner capsid. The gap between the adjacent layers is ~32 Å, similar to the 30 Å spacing previously reported in bluetongue virus [44] and cypovirus [16]. Strands in the outermost layers and closest to the RdRp are well defined, clearly showing major and minor grooves consistent with dsRNA; the diameter of the strands is ~23 Å with ~34 Å spacing between the grooves. The distance between the two adjacent strands in the same layer is about ~30 Å. Discussion Our near-atomic resolution structure of MCRV has revealed the complete organization of the major capsid proteins, as well as ten RdRps, and the viral genome. While these structures PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 13 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus certainly provide new insights into the capsid proteins and genomic interactions of dsRNA viruses, they also raise a number of fundamental questions. Reovirus origins The relationship of MCRV to members of Reovirales has been difficult to resolve [2]. All extant members of Reovirales are divided into two clear families: the members of Spinareoviridae, having turrets and clamp proteins to hold together the inner shell; and the members of Sedor- eoviridae, having internal capping enzymes and no turret or clamp. There has been perfect concordance between classification on these morphological features and the phylogenies of the polymerase sequence [21,45,46]; the polymerase is the only phylogenetically informative gene of members of the Reovirales over long timescales. MCRV has been suggested to be classified within the non-turreted Sedoreoviridae as it lacks a capping turret [3] and because it expresses a protein with some homology to the internal cap- ping enzyme of Seadornavirus [2]. Furthermore, MCRV VP1, VP3, and VP12 exhibit more structural similarity to rotavirus homologs than to any turreted reovirus. This agrees with the previous evidence placing MCRV in the non-turreted family [2,3]. However, we have now identified VP11 as a potential clamp protein (Fig 4), analogous to the cypovirus (CPV) LPP clamp [9], Fako virus clamp [21], orthoreovirus σ2 clamp [8], and aquareovirus VP6 clamp [35]. In the absence of detectable structural homology, it is unclear whether VP11 is a highly- divergent homolog of those clamp proteins or arose by convergent evolution due to specific constraints around stability. The large amount of divergence between the members of two families of Reovirales have made it unclear whether the turreted and non-turreted virus are sis- ter clades or whether one is paraphyletic to the other. The presence of a clamp may have emerged after the divergence between Spinareoviridae and Sedoviridae. RdRp and its arrangement Studies of RdRp positions inside the virion of reovirads have revealed some commonalities and also some enigmas. For both cypovirus and MRV, symmetry-mismatch reconstructions revealed 10 RdRps in a pseudo-D3 symmetric organization along with its 10 RNA segments; potentially each genome segment is specifically associated with one RdRp [16,17,19,47]. In GCRV and ARV, 11 RdRps along with their 11 RNA segments are arranged with a pseudo-D3 symmetric organization [13]. In Fako virus with 9 RNA segments, the RdRps are distributed at 10 locations with pseudo-D3 symmetry [20]. All the above-mentioned viruses belong to the Spinareoviridae family of Reovirales, which have turrets at their 5-fold vertices. Additionally, all their TECs are arranged with pseudo-D3 symmetry inside the capsid, though the locations of the unoccupied vertices within the capsid are different [47] (S1 Table). For Sedoreoviridae, although the high-resolution structures of Rotavirus and Bluetongue virus TECs have been determined [12,37], a defined arrangement of the TECs in the virus particles have not been reported. In rotavirus, it was reported that VP1 arrangement (the RdRp) has no symmetry and binds stochastically at one of the five possible positions at each of the 5-fold vertices [36]. MCRV, which has no turrets atop its 5-fold vertices, has 12 genome segments [2] but our results revealed 10 well-resolved RdRps. Additionally, these 10 RdRps are consistent with D5 symmetry, distinct from the pseudo-D3 symmetry found in Fako virus, cypovirus, MRV and ARV (S1 Table). Only faint, blurred density can be found at the two pole vertices that do not have obvious RdRp density in either the D5- or C1-symmetry reconstructions. Our exhaustive pair-wise sim- ilarity analysis of the 60 views of the C1 map vertices related by icosahedral symmetry (S3 Fig) clearly demonstrated the D5 symmetry organization for the 10 vertices, each having clear PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 14 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus density corresponding to a RdRp, while the remaining two pole vertices are distinct and have little density at a putative RdRp position. These results show that 10 well-resolved RdRp are arranged with D5 symmetry, not D3 as in other members of Reovirales. MCRV may have 11 or 12 vertex-associated RdRps, but if 12, the orientations of the polar RdRp are not correlated. This marked difference from all other characterized Reovirales members could be attributable to having 12 genome segments or to unique properties of this clade; distinguishing between these two hypotheses will require investigation of more species. Also, these techniques will only reveal capsid-associated RdRp; RdRp that do not have a consistent position/orientation with respect to the capsid require alternative approaches for detection [48]. In addition to the distinctive arrangement, the RdRp of MCRV has two conspicuous struc- tural features: a specific protrusion on the innermost surface of the RdRp (Fig 5C), and a long loop that blocks the gap between the thumb and finger subdomains (Figs 5B and S10). In blue- tongue virus, the RdRp has a “fingernail” motif that sits atop the finger subdomain and it pushes the terminal RNA to approach the template entry tunnel in a slightly different orienta- tion [37]. The unique protrusion of MCRV RdRp is not linked to the finger subdomain, though the long loop is linked to the finger subdomain. The position of the protrusion is close to the template entry tunnel (Figs 5, S10 and S13), while the long loop points to different orien- tation; we hypothesize that the unique protrusion, despite not being directly linked to the fin- ger domain, serves a similar function in MCRV as that of the fingernail domain in bluetongue virus. Multiple conformations and functions of VP3 As seen in the MCRV structure, the N-terminus of VP3 has several unique conformations and interacts with other VP3s, as well as the viral genome and the RdRp. In the VP3 dimer, the long “helix-loop-helix-loop” N-terminus of VP3B forms an interwoven network that stabilizes the shell (Fig 2B). The N-terminus in other dsRNA viruses with a pseudo T = 2 shell also have extend structures of the N-terminus to create a stable capsid shell (i.e. the N-anchor of capsid shell protein B (CSP-B) in cypovirus [9] and the N-terminus subdomain III of aquareovirus [28]). However, the long “helix-loop-helix-loop” interwoven structure observed in VP3B of MCRV is much more complex and, potentially, more stable. The N-terminal fragments of three neighboring VP3As around the five-fold axes interact with the genome (Fig 2C); N-terminal fragments have been shown to be involved in RNA organization and movement around the RdRp [49,50]. In bluetongue virus, evidence suggests that the smallest genome segment triggers RNA–RNA interaction and further recruits the medium to larger size ssRNA genome segments to be packaged into the capsid, with the inner capsid protein playing some correcting roles in the whole process [51]. This result gives clues that the packaging and organizations of genome segments together with the inner capsid may direct the RdRps organization. It has been widely reported that the inner capsid proteins of Reovirales members, especially the N-terminal region, are involved in the regulation of the polymerase activity, genome repli- cation, and mRNA transcription [12,49,52,53,54]. In cypovirus, the N-terminus of the inner capsid protein also interacts with the RdRp; the N-terminal helices of two CSPs interact with the RdRp to affect its activities [17]. In bluetongue virus, the N-termini of five VP3As have five slightly different conformations that interact with the RdRp. Similarly, N-terminal fragments from two VP3As (VP3A1, VP3A2) interact with the RdRp (Fig 5F and 5G). In VP3A2, not only does the α-helix (residues 51–76) insert into the RdRp in a manner similar to that of cypovirus, but residues 40–48 form a strand and interact with two strands of the bracelet domain of the RdRp to form an anti-parallel β-sheet (Fig 5G). Forming one strand of this β- PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 15 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus sheet, residues 1220–1204 of the RdRp directly link to “module A” of the bracelet domain, as described in cypovirus. In the transcription and quiescent stages, module A and this region of the capsid protein undergo conformational changes [17]. Secondly, in the same VP3A, apical domain residues 309–348 are proximal to residues 1350–1364 of RdRp, which directly link to the “α-helix plug”. Therefore, it is justifiable to surmise that the aforementioned apical domain fragment can regulate the “plug” as its conformation is changed. Additional studies have shown that not only can the N-terminus of the inner capsid protein be involved in the regula- tion, but also that the residues at or beyond the apical domain are critical for the RdRp activi- ties [52]. Our results provide clear structural evidence that the residues at the apical domain might be critical to regulate the “plug” of RdRp. The N-terminal fragment of one VP3B is found to interact with the finger subdomain of the RdRp (Fig 5F and 5H). In cypovirus, this area is the interface between the NTPase of VP4 and the finger subdomain [17]. In MCRV, no obvious extra density can be found here. How- ever, the interface is close to the entries of two RdRp tunnels: the NTP entry and the template entry tunnels. Previous studies have demonstrated that VP2 (the inner capsid shell protein) in rotavirus can strongly bind to the mRNA template during genome replication [54]. Further studies have shown more structural evidence that the N-terminal regions of the inner capsid shell protein homologs serve as transcriptional regulating factors for RdRp [12]. In summary, our results reveal several unique features of the unusual, 12-segmented MCRV. Despite its classification into the turrret-less/clamp-less Sedoreoviridae, MCRV does contain “clamp”-like proteins (VP11) in its outer capsid. These clamp proteins appear to be crucial in maintaining capsid stability. Additionally, the N-termini of the inner capsid pro- teins, VP3, have multiple conformations with vastly different functions. The N-terminus of VP3B has a unique “helix-loop-helix-loop” motif, suggesting it may play roles in capsid assem- bly and stable, while the N-termini of VP3A may help to anchor the RdRp in its offset location on the inner surface of the capsid. The RdRps are positioned off the 5-fold axes, and their arrangement in the capsid obeys D5 symmetry. Ten RdRp arrange in a correlated, D5 configu- ration within the capsid, while the remaining two fivefold vertices could bind zero, one, or two RdRps. Further differentiating MCRV from sister viruses, the RdRp has two unique features: a protrusion of the N-domain and long loop on the top of the finger domain which may play a role in transcription. The extra densities in tMCRV RdRps demonstrate its transcriptional ability. Together these findings help to establish a more complete understanding of the organi- zation of the members of Reovirales, the structure and function for individual proteins, and the evolutionary relationships among Reovirales members. Methods Virus isolation and purification The mud crab Scylla serrata was infected by MCRV. Virus purification was performed accord- ing to the previous method [4] with slight modification as described below. The gills of the mud crab infected by MCRV were homogenized in PBS (2.7mM KCl, 137mM NaCl, 10mM Na2HPO4, 2mM KH2PO4, pH 7.4) at 4˚C. The homogenized samples were centrifuged for 1h at 10,000g to discard the tissue and cell debris. The supernatant was then centrifuged at 200,000g for 2 hours. The pellet was re-suspended in PBS buffer and then loaded on a 15~45% (w/w) CsCl gradient and centrifuged at 200,000g for 8 hours. The recovered fractions were diluted in PBS buffer and centrifuged for 2 hours at 200,000g. The pellet was re-suspended in PBS buffer and was checked by negative stain TEM (JEM 2010) to make sure the concentration and purity was adequate for further cryo-EM imaging. For negative stain EM, 3 μl of sample was applied to a glow-discharged carbon-film coated grid for 1 min. After removal of excess PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 16 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus buffer, the grid was stained with 3% phosphotungstic acid (PTA) (w/v) for 1 min. The excess PTA was then removed by filter paper and the grid was dried in the air. Cell-free transcription reaction For the transcription reaction, purified MCRV particles were incubated in transcription buffer (70 mM Tris–HCl, pH 8, 4 mM rATP; 2 mM rGTP; 2 mM rCTP and 2 mM rUTP; 1 mM S- adenosylmethionine; 1 mM recombinant Rnase Inhibitor; 10 mM MgAc2; 10 mM NaAc) for 10 min at 31˚C prior to plunge-freezing for cryo-EM [41,55]. RNA transcription was con- firmed with a control experiment using [α-32P] UTP in the transcription reaction mixture before the sample was used for cryo-EM. Cryo-EM sample grid preparation and data collection The R1.2/1.3 copper Quantifoil holey grids were coated a fresh thin layer of continuous carbon film just before cryo-EM sample freezing. 2.5 μl MCRV sample was applied to the grids, blot- ted, and then flash-frozen in precooled liquid ethane using a FEI Vitrobot Mark IV at 100% humidity. The frozen grids were stored in liquid nitrogen for subsequent cryo-EM data acquisition. The frozen grids were loaded into an FEI Titan Krios electron microscope operated at 300 kV. Cryo-EM micrographs were recorded on an FEI Falcon II direct electron detection camera at 75000 × nominal magnification with the calibrated pixel size of 1.09 Å at the sample level. The intended defocuses ranged from 0.6 to 3 μm. The dose rate on the sample was 23 eÅ-2s-1. The data were recorded as movies of 16 frames with total dose of 25 eÅ-2 and exposure time of 1.1 s. 3595 were used for final image processing and 3D reconstruction for the qMCRV; while for tMCRV, 2323 movie were used for final reconsctruction. Image processing and 3D reconstruction The movies were aligned and drift-corrected using the GPU-accelerated program Motioncorr V2.0 [56]. The EMAN2 program e2boxer.py [57] was used to select 58,095 particles from 3,595 micrographs (Fig 1A) and 2,1945 particles from 2,323 movies for qMCRV and tMCRV, respectively. Some of the selected particles were “empty” without encpasidated genome. It is not known whether these particles were assembled without RNA or if they lost RNA after cap- sid assembly (either before or after second-strand synthesis). These empty particles were man- ually separated from other genome-containing particles for further data analysis. The particles were extracted directly from the movie frames with dose/radiation damage dependent weight- ing using the batchboxer.py program in JSPR [58]. Contrast transfer function parameters were determined automatically using fitctf2.py and then visually validated using EMAN ctfit pro- gram. The dataset was divided into two halves and all subsequent image processing, including construction of initial de novo models and iterative refinements, was performed on each of the two subsets independently using the JSPR software. All resolutions were assessed with “gold standard” Fourier Shell Correlation using 0.143 criterion [59,60] (S1 Fig). The particles were first reconstructed as icosahedral particles as described previously [58]. Briefly, the initial mod- els were built by iteratively refining randomly assigned initial orientations using 4x binned particles. The particle parameters were then transferred to 2× binned particles and particles without binning for high resolution refinements. The icosahedral reconstructions were then used as the starting point to perform C1 asymmetric reconstructions of the MCRV particles using the symmetry relaxation method that we previously developed [24,25]. Without impos- ing any symmetry, the 3D reconstructions quickly converged to reveal well-resolved RdRp protein densities under the 5-fold vertices (S2 Fig). By visual inspection and quantitative PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 17 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus evaluation of the similarities of the RdRp densities and their relative orientations (S2 and S3 Figs), we found RdRp densities at 10 of the 12 vertices that were well resolved and related by D5 symmetry, while the remaining 2 vertices had only weak RdRp densities. To further improve the RdRp densities, we imposed D5 symmetry in subsequent reconstructions (S2 and S3 Figs). Symmetry assessment: decoy method Particle position, scale, defocus, and astigmatism were fixed throughout refinement and the orientation was allowed to vary only among icosahedrally-equivalent Euler angles. A RdRp was first segmented from the D5-symmetrized map with UCSF Chimera. Decoys were then constructed by summing RdRp and capsid maps together. The icosahedrally-sym- metrized capsid map was masked to remove density at each of the 60 possible RdRp sites. The intensity of the RdRp and capsid were matched by multiplying the RdRp map by a constant. Next, a number of RdRp maps were added to the capsid at specified icosahedrally-related loca- tions. This decoy map was then filtered to 8Å resolution. Various decoys were generated with RdRp placements as follows: random placements of 8 RdRps (subject to the constraint of only one RdRp per five-fold axis), a CPV-like configuration of 10 RdRps, a D5 configuration of 10 RdRps, a D5 configuration plus one or two polar RdRps in each possible configuration, and a D5 configuration of 10 RdRps, where each RdRp is turned 144˚ compared to the structure reported herein. Particles were subjected to iterative multi-reference alignment and reconstruction starting with the decoys as models. Per-particle correlations to the various decoys were evaluated to assess which decoy best reflects the particle population. Model building and assessment Models for the capsid proteins and RdRp were constructed using de novo methods as previ- ously described [61,62]. Briefly, UCSF Chimera [63] was used to manually segment out indi- vidual capsid proteins based on visual inspection. Once segmented, individual subunits were aligned using a combination of the e2foldhunter.py program [64,65] from EMAN2 and the “Fit in map” function in Chimera. From this alignment, average maps for each of RdRp, VP3A, VP3B, VP12 and VP11 were constructed. Initial backbone models were constructed directly from the density map with Pathwalking (e2pathwalker.py in EMAN2) [64]; only a density threshold and the corresponding number of residues in a subunit were provided as additional inputs. Pathwalking produces a non-directional Cα backbone trace and visual inspections of the models in the corresponding maps were used to determine directionality. The primary sequence of the corresponding protein was then threaded onto the model and converted to an all-atom model with Remo [66]. Iterative real-space refinement and visual optimization were done using Phenix [67] and Coot [68], respectively. Model quality was monitored by examin- ing fit-to-density, Ramachandran outliers, clashes and rotamers. For models from the asym- metric map, individual subunit models were first fit to the corresponding regions of the map and the refined with Phenix real space refinement. For regions with weak density, extra density or where there were large structural rearrangements, manual model building was done with Coot, followed by additional real-space refinement steps. Once models for a complete asym- metric unit were refined, five neighboring asymmetric unit models were placed in adjacent asymmetric unit positions of the density map in order to capture all possible interactions. A final round of Phenix real-space refinement was used to minimize clashes between subunits in neighboring asymmetric units. The final asymmetric units were then transformed to construct a complete capsid model. Final models were assessed using molprobity [69] for assessing PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 18 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus model quality (S2 Table). Figures and model analysis were done in UCSF Chimera and Coot; subunit interactions were examined using PDBe Pisa [70]. Supporting information S1 Table. Comparison of reoviruses with different numbers of genome segments/RdRps. (DOCX) S2 Table. Model quality statistics. (DOCX) S1 Script. Python script symAngleDifferences.py that was used to compute the count num- bers in S1C Fig. (PY) S1 Movie. The icosahedral structure of MCRV. (MOV) S1 Fig. Cryo-EM of MCRV. (A) A representative cryo-EM image of MCRV particles embed- ded in vitreous ice. (B) The Fourier shell correlation curves of the icosahedral and D5 recon- structions of MCRV and tMCRV. Resolutions are measured at 3.1 Å and 3.4 Å for MCRV with icosahedral and D5 symmetry, respectively, and 3.36 Å and 3.7 Å for tMCRV with icosa- hedral and D5 symmetry, respectively. Resolution measurements were based on the “gold- standard” FSC = 0.143 criterion. (C) Cartoon view of the MCRV. The polymerase (VP1), inner capsid proteins (VP3A and VP3B), clamp protein (VP11) and outer capsid protein (VP12) are colored in red, cornflower blue, orange, light green and medium purple, respec- tively. (TIF) S2 Fig. Convergence process of ab initio asymmetry reconstruction of MCRV. XC row: cen- tral section perpendicular to icosahedral 2-fold axis (i.e. X axis). Z1 to Z4 rows: sections per- pendicular to icosahedral 5-fold axis (i.e. Z axis) at locations indicated by the dashed lines in XC row. The icosahedral reconstruction was used as starting model to initiate the iterative symmetry relaxation alignment and asymmetric reconstruction process until convergence. The XC and Z sections of the map for each iteration are shown in a column indicated by the iteration numbers (0 to 8). The black arrows indicate the RdRp densities are clear resolved by iteration 4. (TIF) S3 Fig. Discovery of the D5 symmetry arrangement of MCRV’s RdRps. (A) MDS clustering of the C1 reconstruction vertices in all 60 icosahedral related views. The C1 map was rotated to all 60 icosahedral related views and similarities among all possible combinations of the vertices were computed. The 6 apparent clusters were labeled 1 to 6. Each cluster consisted of 10 views. (B) Analysis of the distribution of the angles between all pairs (10*9/2 = 45) among the 10 views in each of the 6 clusters. (C) List of the expected distribution of angles between all pairs of symmetry operations for the listed symmetries. All the distributions in (B) matched that of D5 symmetry (bold) suggesting that the RdRps located at the 10 vertices were arranged in D5 symmetry, a subset of icosahedral symmetry. (D) Central section view of C1 (left) and D5 (right) symmetry reconstructions. The middle column (C1—D5 view) shows the same C1 map (left) but re-oriented in the same D5 view as that in the right column (D5 symmetry map) then imposed with D5 symmetry. XC: central section perpendicular to icosahedral 2-fold axis (i.e. X axis). Z1 to Z4: sections perpendicular to icosahedral 5-fold axis (i.e. Z axis) at locations PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 19 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus indicated by the dashed lines in XC row. The numbers in table (C) were calculated using the python script symAngleDifferences.py (S1 Script). (TIF) S4 Fig. The arrangement of RdRps inside empty MCRV. (A) 3D classification of empty MCRV reveals 6 classes with 10 well resolved RdRp densities arranged with D5 symmetry, after masking away the capsid density and symmetry expansion. The densities at the remaining two vertices are substantially weaker. The dot lines show the potential axes although we didn’t impose the symmetry during the refinement. (B)10 well resolved RdRp densities can be seen in the reconstruction from empty particles. An icosahedron is displayed in orange as a visual guide, and two icosahedral vertices contain weaker densities. Numbers 1–10 mark the 10 RdRps. (C) Thirteen decoys were constructed by placing 5 closest neighboring RdRps, (such as, if we select RdRp1 marked in (B), then the other closest RdRps will include RdRp2, 5, 6 and 10) with random orientation. Decoy 13 is in agreement with that of D5 symmetry. For each of the thirteen, the fraction of particles aligning to the corresponding decoy is shown in gray bar. The RdRps of each decoy are displayed and linked to the corresponding bar by dot lines. One of the thirteen random decoys reflected the underlying data better than did the other decoys. The results demonstrates that empty MCRV particles matched best with D5 symmetry. (TIF) S5 Fig. The structure of VP3. (A) Density map (gray) of VP3A superimposed with its atomic model (cornflower blue). (B) Zoom-in regions of the density map (black net) superimposed with atomic model, demonstrating the quality of cryo-EM map VP3A and that of the refined atomic model. (TIF) S6 Fig. Comparison between VP3A and VP3B. The atomic models for VP3A (cornflower blue) and VP3B (orange) are shown overlaid. There is considerable difference between the VP3A and VP3B subunits. Specifically, the biggest differences are at the distal edge of the dimerization domain (red circle), in the carapace domain (green circle) and on the inner sur- face of the carapace domain (black circle), and the extended N-terminus of VP3B. (TIF) S7 Fig. The structure of VP12. (A) Density map (gray) of VP12 superimposed with its atomic model (medium purple). (B) Zoom-in regions of the density map (black nets) superimposed with atomic model (medium purple), demonstrating the quality of cryo-EM map VP12 and that of the refined atomic model. (TIF) S8 Fig. The structure of VP11. (A) Density map (gray) of VP11 superimposed with its atomic model (green). Right panel shows that zoom-in regions of the density map superimposed with atomic model, demonstrating the quality of cryo-EM map VP11 and that of the refined atomic model. (B) Alignment of VP11A and VP11B (cyan) reveals few conformation differences between them. The main differences are at the termini (blue arrows) and two peripheral loops (black arrows). (C) The sequence and secondary structural elements are indicated. The color schemes for secondary structure are the same as the model in Fig 4D. (TIF) S9 Fig. Interactions between the VP11A and its neighbor proteins. (A) VP11A interacts with the surrounding VP12s. (B) The hydrogen bonds between the VP11A and two VP12 pro- teins. (C) The interacting amino acids at the interface between VP11 and VP12 are indicated. (D) VP11A also interacts with the VP3s. (E) The hydrogen bond between the VP11A and PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 20 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus VP3A. (F) The interacting amino acids at the interface between VP11A and VP3 are indicated. (G) VP11B clamps the two neighbor VP3 pentamers together. (H) The hydrogen bond between the VP11B and VP3B. (I) The interacting amino acids at the interface between VP11B and VP3 are listed. (J) The hydrophobic interactions between VP11B and VP3B, made by Phe130 in VP11 and Val106, Val 821, Leu822 in VP3. The model of VP11B is colored with cyan, while the model of VP3B is colored with orange. (TIF) S10 Fig. A Comparison of RdRps of several Reovirales family members. The color scheme is identical to that in Fig 5. The structure inside the black circle is the unique protrusion of the N-domain in MCRV. Black arrows indicate the priming loop; red arrows indicate the C-termi- nal plug. The orange arrow indicates the unique long loop blocking the gap between the thumb and finger subdomain of MCRV RdRp. The unique “fingernail motif” of bluetongue virus is colored in cyan. (TIF) S11 Fig. The tunnels of RdRp and major differences in RdRp between MCRV and tMCRV. (A) indicates the four tunnels of RdRp. The asterisks indicate the positions of tunnels. The color schemes of RdRp and capsid are same as that in Figs 1 and 5. (B) indicates the RdRp map together with the extra densities (colored with light blue) in tMCRV. The orientation of each image view, together with the color schemes of RdRp and capsid are the same as that in A. (C) shows a typical cryo-EM image of tMCRV particles. Black arrows show the RNA strands. (D) The difference map (light blue) shows the extra density in the center of the RdRp. The color scheme of the RdRp model is same as that of Fig 5. The inner capsid position is shown by partial models of VP3A. (E), The extra density has features concordant with downstream dsRNA, the ssRNA template, the upstream dsRNA together with the noncontinuous non-template single strand RNA, and the transcript product. The density of non-template ssRNA is not continuous and is indicated by dot lines. The difference map is shown at a contour level of roughly 8.3. (TIF) S12 Fig. The organization of the RdRps and dsRNA genome segments of MCRV. Density maps without the two capsid shells viewed along the 5-fold axis reveal that there are two kinds of icosahedral 5-fold vertices (A, B). Two of the twelve vertices have less density around the 5-fold axes (A), while at the remaining 10 vertices there is obvious RdRp density (orange) slightly offset from the 5-fold axes (B). Panels (C) and (D) show the maps along the 3- and 2-fold axis, respectively. 5-, 3-, and 2-fold axes are marked in pentagon, triangle and ellipse respectively. The map, except for the densities of polymerase, is radially colored. (TIF) S13 Fig. Comparison of the RdRps of MCRV and Bluetongue virus. The color scheme is identical to that in S9 Fig. The left column and middle column are the polymerase of Blue- tongue and MCRV respectively. The right column displays the alignment of tow polymerases. The unique protrusion (yellow) of MCRV’s N-terminal domain is displayed. The alignment of the two RdRps reveals that the unique protrusion’s position is close to that of the fingerail motif of bluetongue virus RdRp. (TIF) Acknowledgments We are grateful to the members at Bio-Electron Microscopy facility at the School of Life Sci- ences in Sun Yat-sen University, members at Center of Biological Imaging in IBP,china, and PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 21 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus members at the NCPSS, china, for instrument support and technical assistance; We also thank the Purdue Rosen Center for Advanced Computing and the Office of Advanced Research Computing at Rutgers, the State University of New Jersey for computing resources. Author Contributions Conceptualization: Qinfen Zhang, Matthew L. Baker, Jianguo He, Jason T. Kaelber, Wen Jiang. Data curation: Qinfen Zhang, Yuanzhu Gao, Matthew L. Baker, Shanshan Liu, Xudong Jia, Haidong Xu, Wen Jiang. Formal analysis: Qinfen Zhang, Yuanzhu Gao, Shanshan Liu, Xudong Jia, Haidong Xu, Jason T. Kaelber, Wen Jiang. Funding acquisition: Qinfen Zhang, Jianguo He, Shaoping Weng. Investigation: Qinfen Zhang, Yuanzhu Gao, Matthew L. Baker, Shanshan Liu, Xudong Jia, Haidong Xu, Jianguo He, Jason T. Kaelber, Shaoping Weng, Wen Jiang. Methodology: Yuanzhu Gao, Matthew L. Baker, Xudong Jia, Jason T. Kaelber, Wen Jiang. Project administration: Qinfen Zhang, Jianguo He, Shaoping Weng, Wen Jiang. Resources: Haidong Xu, Jianguo He, Shaoping Weng. Software: Wen Jiang. Supervision: Qinfen Zhang, Shaoping Weng, Wen Jiang. Validation: Qinfen Zhang, Yuanzhu Gao, Matthew L. Baker, Shanshan Liu, Xudong Jia, Jason T. Kaelber. Visualization: Qinfen Zhang, Matthew L. Baker, Xudong Jia, Jason T. Kaelber. Writing – original draft: Qinfen Zhang, Yuanzhu Gao, Shanshan Liu. Writing – review & editing: Qinfen Zhang, Yuanzhu Gao, Matthew L. Baker, Jason T. Kaelber, Shaoping Weng, Wen Jiang. References 1. Attoui H M P.P.C., Becnel J., Belaganahalli S., Bergoin M., Brussard C.P., Chappell J.D.Ciarlet M., del Vas M., & other authors. Family Reoviridae. Virus taxonomy: ninth report of the International Committee on Taxonomy of Viruses: Elsevier; 2012. 2. Deng XX, Lu L, Ou YJ, Su HJ, Li G, Guo ZX, et al. Sequence analysis of 12 genome segments of mud crab reovirus (MCRV). Virology. 2012; 422(2):185–94. Epub 2011/11/18. https://doi.org/10.1016/j.virol. 2011.09.029 PMID: 22088215. 3. Huang Z, Deng X, Li Y, Su H, Li K, Guo Z, et al. Structural insights into the classification of Mud Crab Reovirus. Virus research. 2012; 166(1–2):116–20. Epub 2012/03/17. https://doi.org/10.1016/j.virusres. 2012.02.025 PMID: 22421382. 4. Weng SP, Guo ZX, Sun JJ, Chan SM, He JG. A reovirus disease in cultured mud crab, Scylla serrata, in southern China. Journal of fish diseases. 2007; 30(3):133–9. Epub 2007/03/14. https://doi.org/10.1111/ j.1365-2761.2007.00794.x PMID: 17352788. 5. Dunn SE, Li H, Cardone G, Nibert ML, Ghabrial SA, Baker TS. Three-dimensional structure of victori- virus HvV190S suggests coat proteins in most totiviruses share a conserved core. PLoS pathogens. 2013; 9(3):e1003225. Epub 2013/03/22. https://doi.org/10.1371/journal.ppat.1003225 PMID: 23516364; PubMed Central PMCID: PMC3597494. 6. Luque D, Gonzalez JM, Garriga D, Ghabrial SA, Havens WM, Trus B, et al. The T = 1 capsid protein of Penicillium chrysogenum virus is formed by a repeated helix-rich core indicative of gene duplication. Journal of virology. 2010; 84(14):7256–66. Epub 2010/05/14. https://doi.org/10.1128/JVI.00432-10 PMID: 20463071; PubMed Central PMCID: PMC2898261. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 22 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus 7. Coulibaly F, Chevalier C, Gutsche I, Pous J, Navaza J, Bressanelli S, et al. The birnavirus crystal struc- ture reveals structural relationships among icosahedral viruses. Cell. 2005; 120(6):761–72. https://doi. org/10.1016/j.cell.2005.01.009 WOS:000228067500007. PMID: 15797378 8. Reinisch KM, Nibert ML, Harrison SC. Structure of the reovirus core at 3.6 A resolution. Nature. 2000; 404(6781):960–7. Epub 2000/05/09. https://doi.org/10.1038/35010041 PMID: 10801118. 9. Yu X, Ge P, Jiang J, Atanasov I, Zhou ZH. Atomic model of CPV reveals the mechanism used by this single-shelled virus to economically carry out functions conserved in multishelled reoviruses. Structure. 2011; 19(5):652–61. Epub 2011/05/14. https://doi.org/10.1016/j.str.2011.03.003 PMID: 21565700; PubMed Central PMCID: PMC3405907. 10. Grimes JM, Burroughs JN, Gouet P, Diprose JM, Malby R, Zientara S, et al. The atomic structure of the bluetongue virus core. Nature. 1998; 395(6701):470–8. Epub 1998/10/17. https://doi.org/10.1038/ 26694 PMID: 9774103. 11. Nakagawa A, Miyazaki N, Taka J, Naitow H, Ogawa A, Fujimoto Z, et al. The atomic structure of rice dwarf virus reveals the self-assembly mechanism of component proteins. Structure. 2003; 11 (10):1227–38. Epub 2003/10/07. https://doi.org/10.1016/j.str.2003.08.012 PMID: 14527391. 12. Ding K, Celma CC, Zhang X, Chang T, Shen W, Atanasov I, et al. In situ structures of rotavirus polymer- ase in action and mechanism of mRNA transcription and release. Nat Commun. 2019; 10(1):2216. https://doi.org/10.1038/s41467-019-10236-7 PMID: 31101900; PubMed Central PMCID: PMC6525196. 13. Wang XR, Zhang FX, Su R, Li XW, Chen WY, Chen QX, et al. Structure of RNA polymerase complex and genome within a dsRNA virus provides insights into the mechanisms of transcription and assembly. Proceedings of the National Academy of Sciences of the United States of America. 2018; 115 (28):7344–9. https://doi.org/10.1073/pnas.1803885115 WOS:000438050900070. PMID: 29941585 14. 15. 16. 17. Lu X, McDonald SM, Tortorici MA, Tao YJ, Vasquez-Del Carpio R, Nibert ML, et al. Mechanism for coor- dinated RNA packaging and genome replication by rotavirus polymerase VP1. Structure. 2008; 16 (11):1678–88. Epub 2008/11/13. https://doi.org/10.1016/j.str.2008.09.006 PMID: 19000820; PubMed Central PMCID: PMC2602806. Tao Y, Farsetta DL, Nibert ML, Harrison SC. RNA synthesis in a cage—structural studies of reovirus polymerase lambda3. Cell. 2002; 111(5):733–45. Epub 2002/12/05. https://doi.org/10.1016/s0092- 8674(02)01110-8 PMID: 12464184. Liu H, Cheng L. Cryo-EM shows the polymerase structures and a nonspooled genome within a dsRNA virus. Science. 2015; 349(6254):1347–50. Epub 2015/09/19. https://doi.org/10.1126/science.aaa4938 PMID: 26383954. Zhang X, Ding K, Yu X, Chang W, Sun J, Hong Zhou Z. In situ structures of the segmented genome and RNA polymerase complex inside a dsRNA virus. Nature. 2015. Epub 2015/10/28. https://doi.org/10. 1038/nature15767 PMID: 26503045. 18. Ding K, Nguyen L, Zhou ZH. In Situ Structures of the Polymerase Complex and RNA Genome Show How Aquareovirus Transcription Machineries Respond to Uncoating. Journal of virology. 2018; 92(21). ARTN e00774-1810.1128/JVI.00774-18. WOS:000447139100006. https://doi.org/10.1128/JVI.00774- 18 PMID: 30068643 19. Yazaki K, Miura K. Relation of the structure of cytoplasmic polyhedrosis virus and the synthesis of its messenger RNA. Virology. 1980; 105(2):467–79. Epub 1980/09/01. https://doi.org/10.1016/0042-6822 (80)90047-1 PMID: 18631679. 20. Kaelber JT, Jiang W, Weaver SC, Auguste AJ, Chiu W. Arrangement of the Polymerase Complexes inside a Nine-Segmented dsRNA Virus. Structure. 2020; 28(6):604–12 e3. https://doi.org/10.1016/j.str. 2020.01.011 PMID: 32049031; PubMed Central PMCID: PMC7289189. 21. Auguste AJ, Kaelber JT, Fokam EB, Guzman H, Carrington CV, Erasmus JH, et al. A newly isolated reovirus has the simplest genomic and structural organization of any reovirus. Journal of virology. 2015; 89(1):676–87. Epub 2014/10/31. https://doi.org/10.1128/JVI.02264-14 PMID: 25355879; PubMed Cen- tral PMCID: PMC4301156. 22. Dryden KA, Farsetta DL, Wang G, Keegan JM, Fields BN, Baker TS, et al. Internal/structures containing transcriptase-related proteins in top component particles of mammalian orthoreovirus. Virology. 1998; 245(1):33–46. Epub 1998/06/06. https://doi.org/10.1006/viro.1998.9146 PMID: 9614865. 23. Xia Q, Jakana J, Zhang JQ, Zhou ZH. Structural comparisons of empty and full cytoplasmic polyhedro- sis virus. Protein-RNA interactions and implications for endogenous RNA transcription mechanism. The Journal of biological chemistry. 2003; 278(2):1094–100. Epub 2002/10/29. https://doi.org/10.1074/jbc. M205964200 PMID: 12401805. 24. Jiang W, Chang J, Jakana J, Weigele P, King J, Chiu W. Structure of epsilon15 bacteriophage reveals genome organization and DNA packaging/injection apparatus. Nature. 2006; 439(7076):612–6. Epub PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 23 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus 2006/02/03. https://doi.org/10.1038/nature04487 PMID: 16452981; PubMed Central PMCID: PMC1559657. 25. Guo F, Liu Z, Vago F, Ren Y, Wu W, Wright ET, et al. Visualization of uncorrelated, tandem symmetry mismatches in the internal genome packaging apparatus of bacteriophage T7. Proceedings of the National Academy of Sciences of the United States of America. 2013; 110(17):6811–6. https://doi.org/ 10.1073/pnas.1215563110 PMID: 23580619; PubMed Central PMCID: PMC3637776. 26. Mathieu M, Petitpas I, Navaza J, Lepault J, Kohli E, Pothier P, et al. Atomic structure of the major capsid protein of rotavirus: implications for the architecture of the virion. The EMBO journal. 2001; 20(7):1485– 97. https://doi.org/10.1093/emboj/20.7.1485 PMID: 11285213; PubMed Central PMCID: PMC145492. 27. Basak AK, Gouet P, Grimes J, Roy P, Stuart D. Crystal structure of the top domain of African horse sick- ness virus VP7: comparisons with bluetongue virus VP7. Journal of virology. 1996; 70(6):3797–806. https://doi.org/10.1128/JVI.70.6.3797-3806.1996 PMID: 8648715; PubMed Central PMCID: PMC190256. 28. Cheng L, Zhu J, Hui WH, Zhang X, Honig B, Fang Q, et al. Backbone model of an aquareovirus virion by cryo-electron microscopy and bioinformatics. Journal of molecular biology. 2010; 397(3):852–63. Epub 2009/12/29. https://doi.org/10.1016/j.jmb.2009.12.027 PMID: 20036256; PubMed Central PMCID: PMC2900198. 29. Grimes J, Basak AK, Roy P, Stuart D. The crystal structure of bluetongue virus VP7. Nature. 1995; 373 (6510):167–70. https://doi.org/10.1038/373167a0 PMID: 7816101. 30. Wolfram F, Arora K, Robinson H, Neculai AM, Yip P, Howell PL. Expression, purification, crystallization and preliminary X-ray analysis of Pseudomonas aeruginosa AlgL. Acta Crystallogr Sect F Struct Biol Cryst Commun. 2012; 68(Pt 5):584–7. https://doi.org/10.1107/S1744309112012808 PMID: 22691793; PubMed Central PMCID: PMC3374518. 31. Yuchi Z, Yuen SM, Lau K, Underhill AQ, Cornea RL, Fessenden JD, et al. Crystal structures of ryano- dine receptor SPRY1 and tandem-repeat domains reveal a critical FKBP12 binding determinant. Nat Commun. 2015; 6:7947. https://doi.org/10.1038/ncomms8947 PMID: 26245150; PubMed Central PMCID: PMC4530471. 32. Novakova Z, Cerny J, Choy CJ, Nedrow JR, Choi JK, Lubkowski J, et al. Design of composite inhibitors targeting glutamate carboxypeptidase II: the importance of effector functionalities. FEBS J. 2016; 283 (1):130–43. https://doi.org/10.1111/febs.13557 PMID: 26460595; PubMed Central PMCID: PMC4712110. 33. Adzhubei AA, Sternberg MJE, Makarov AA. Polyproline-II Helix in Proteins: Structure and Function. Journal of molecular biology. 2013; 425(12):2100–32. https://doi.org/10.1016/j.jmb.2013.03.018 WOS:000320414300002. PMID: 23507311 34. 35. 36. Li Z, Baker ML, Jiang W, Estes MK, Prasad BV. Rotavirus architecture at subnanometer resolution. Journal of virology. 2009; 83(4):1754–66. Epub 2008/11/28. https://doi.org/10.1128/JVI.01855-08 PMID: 19036817; PubMed Central PMCID: PMC2643745. Zhang X, Jin L, Fang Q, Hui WH, Zhou ZH. 3.3 A cryo-EM structure of a nonenveloped virus reveals a priming mechanism for cell entry. Cell. 2010; 141(3):472–82. Epub 2010/04/20. https://doi.org/10.1016/ j.cell.2010.03.041 PMID: 20398923; PubMed Central PMCID: PMC3422562. Jenni S, Salgado EN, Herrmann T, Li ZL, Grant T, Grigorieff N, et al. In situ Structure of Rotavirus VP1 RNA-Dependent RNA Polymerase. Journal of molecular biology. 2019; 431(17):3124–38. https://doi. org/10.1016/j.jmb.2019.06.016 WOS:000482872100008. PMID: 31233764 37. He Y, Shivakoti S, Ding K, Cui YX, Roy P, Zhou ZH. In situ structures of RNA-dependent RNA polymer- ase inside bluetongue virus before and after uncoating. Proceedings of the National Academy of Sci- ences of the United States of America. 2019;116(33):16535–40. https://doi.org/10.1073/pnas. 1905849116 WOS:000481404300060. 38. Jacobo-Molina A, Ding J, Nanni RG, Clark AD, Jr., Lu X, Tantillo C, et al. Crystal structure of human immunodeficiency virus type 1 reverse transcriptase complexed with double-stranded DNA at 3.0 A res- olution shows bent DNA. Proceedings of the National Academy of Sciences of the United States of America. 1993;90(13):6320–4. https://doi.org/10.1073/pnas.90.13.6320 PMID: 7687065; PubMed Cen- tral PMCID: PMC46920. 39. Boudreaux CE, Vile DC, Gilmore BL, Tanner JR, Kelly DF, McDonald SM. Rotavirus core shell subdo- mains involved in polymerase encapsidation into virus-like particles. Journal of General Virology. 2013; 94:1818–26. https://doi.org/10.1099/vir.0.052951-0 WOS:000322850200013. PMID: 23596269 40. Zeng CQ-Y, Estes MK, Charpilenne A, Cohen J. The N Terminus of Rotavirus VP2 Is Necessary for Encapsidation of VP1 and VP3. Journal of virology. 1998; 72(1):201–8. https://doi.org/10.1128/JVI.72. 1.201-208.1998 PMID: 9420216 41. Yang C, Ji G, Liu H, Zhang K, Liu G, Sun F, et al. Cryo-EM structure of a transcribing cypovirus. Pro- ceedings of the National Academy of Sciences of the United States of America. 2012;109(16):6118–23. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 24 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus Epub 2012/04/12. https://doi.org/10.1073/pnas.1200206109 PMID: 22492979; PubMed Central PMCID: PMC3341035. 42. Danthi P, Guglielmi KM, Kirchner E, Mainou B, Stehle T, Dermody TS. From touchdown to transcription: the reovirus cell entry pathway. Curr Top Microbiol Immunol. 2010; 343:91–119. https://doi.org/10. 1007/82_2010_32 PMID: 20397070; PubMed Central PMCID: PMC4714703. 43. Cui Y, Zhang Y, Zhou K, Sun J, Zhou ZH. Conservative transcription in three steps visualized in a dou- ble-stranded RNA virus. Nature structural & molecular biology. 2019; 26(11):1023–34. Epub 20191106. https://doi.org/10.1038/s41594-019-0320-0 PMID: 31695188; PubMed Central PMCID: PMC7526681. 44. Gouet P, Diprose JM, Grimes JM, Malby R, Burroughs JN, Zientara S, et al. The highly ordered double- stranded RNA genome of bluetongue virus revealed by crystallography. Cell. 1999; 97(4):481–90. Epub 1999/05/25. https://doi.org/10.1016/s0092-8674(00)80758-8 PMID: 10338212. 45. Attoui H, Jaafar FM, Belhouchet M, Biagini P, Cantaloube JF, de Micco P, et al. Expansion of family Reoviridae to include nine-segmented dsRNA viruses: Isolation and characterization of a new virus des- ignated aedes pseudoscutellaris reovirus assigned to a proposed genus (Dinovernavirus). Virology. 2005; 343(2):212–23. https://doi.org/10.1016/j.virol.2005.08.028 WOS:000234262100007. PMID: 16171838 46. Stenger DC, Sisterson MS, Krugner R, Backus EA, Hunter WB. A new Phytoreovirus infecting the glassy-winged sharpshooter (Homalodisca vitripennis). Virology. 2009; 386(2):469–77. https://doi.org/ 10.1016/j.virol.2009.01.037 WOS:000267310200026. PMID: 19239972 47. Pan M, Alvarez-Cabrera AL, Kang JS, Wang L, Fan C, Zhou ZH. Asymmetric reconstruction of mam- malian reovirus reveals interactions among RNA, transcriptional factor micro2 and capsid proteins. Nat Commun. 2021; 12(1):4176. Epub 20210707. https://doi.org/10.1038/s41467-021-24455-4 PMID: 34234134; PubMed Central PMCID: PMC8263624. 48. Ilca SL, Sun X, El Omari K, Kotecha A, de Haas F, DiMaio F, et al. Multiple liquid crystalline geometries of highly compacted nucleic acid in a dsRNA virus. Nature. 2019; 570(7760):252–6. Epub 20190529. https://doi.org/10.1038/s41586-019-1229-9 PMID: 31142835. 49. Gridley CL, Patton JT. Regulation of rotavirus polymerase activity by inner capsid proteins. Current opinion in virology. 2014; 9:31–8. Epub 2014/09/23. https://doi.org/10.1016/j.coviro.2014.08.008 PMID: 25243800; PubMed Central PMCID: PMC4268341. 50. Labbe M, Baudoux P, Charpilienne A, Poncet D, Cohen J. Identification of the nucleic acid binding domain of the rotavirus VP2 protein. The Journal of general virology. 1994; 75 (Pt 12):3423–30. Epub 1994/12/01. https://doi.org/10.1099/0022-1317-75-12-3423 PMID: 7996135. 51. Sung PY, Roy P. Sequential packaging of RNA genomic segments during the assembly of Bluetongue virus. Nucleic acids research. 2014; 42(22):13824–38. Epub 2014/11/28. https://doi.org/10.1093/nar/ gku1171 PMID: 25428366; PubMed Central PMCID: PMC4267631. 52. McDonald SM, Patton JT. Rotavirus VP2 core shell regions critical for viral polymerase activation. Jour- nal of virology. 2011; 85(7):3095–105. Epub 2011/01/21. https://doi.org/10.1128/JVI.02360-10 PMID: 21248043; PubMed Central PMCID: PMC3067889. 53. Mansell EA, Patton JT. Rotavirus RNA replication: VP2, but not VP6, is necessary for viral replicase activity. Journal of virology. 1990; 64(10):4988–96. Epub 1990/10/01. https://doi.org/10.1128/JVI.64. 10.4988-4996.1990 PMID: 2168982; PubMed Central PMCID: PMC247990. 54. Patton JT, Jones MT, Kalbach AN, He YW, Xiaobo J. Rotavirus RNA polymerase requires the core shell protein to synthesize the double-stranded RNA genome. Journal of virology. 1997; 71(12):9618– 26. Epub 1997/11/26. https://doi.org/10.1128/JVI.71.12.9618-9626.1997 PMID: 9371626; PubMed Central PMCID: PMC230270. 55. Smith RE, Furuichi Y. Gene mapping of cytoplasmic polyhedrosis virus of silkworm by the full-length mRNA prepared under optimized conditions of transcription in vitro. Virology. 1980; 103(2):279–90. https://doi.org/10.1016/0042-6822(80)90187-7 PMID: 18631652. 56. 57. Li X, Mooney P, Zheng S, Booth CR, Braunfeld MB, Gubbens S, et al. Electron counting and beam- induced motion correction enable near-atomic-resolution single-particle cryo-EM. Nature methods. 2013; 10(6):584–90. Epub 2013/05/07. https://doi.org/10.1038/nmeth.2472 PMID: 23644547; PubMed Central PMCID: PMC3684049. Tang G, Peng L, Baldwin PR, Mann DS, Jiang W, Rees I, et al. EMAN2: an extensible image processing suite for electron microscopy. Journal of structural biology. 2007; 157(1):38–46. Epub 2006/07/25. https://doi.org/10.1016/j.jsb.2006.05.009 PMID: 16859925. 58. Guo F, Jiang W. Single particle cryo-electron microscopy and 3-D reconstruction of viruses. Methods Mol Biol. 2014; 1117:401–43. Epub 2013/12/21. https://doi.org/10.1007/978-1-62703-776-1_19 PMID: 24357374; PubMed Central PMCID: PMC4020923. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 25 / 26 PLOS PATHOGENSThe structure of a 12-segmented dsRNA reovirus 59. Henderson R, Sali A, Baker ML, Carragher B, Devkota B, Downing KH, et al. Outcome of the First Elec- tron Microscopy Validation Task Force Meeting. Structure. 2012; 20(2):205–14. https://doi.org/10.1016/ j.str.2011.12.014 WOS:000300388000005. PMID: 22325770 60. Scheres SH, Chen S. Prevention of overfitting in cryo-EM structure determination. Nature methods. 2012; 9(9):853–4. Epub 2012/07/31. https://doi.org/10.1038/nmeth.2115 PMID: 22842542. 61. Baker ML, Hryc CF, Zhang Q, Wu W, Jakana J, Haase-Pettingell C, et al. Validated near-atomic resolu- tion structure of bacteriophage epsilon15 derived from cryo-EM and modeling. Proceedings of the National Academy of Sciences of the United States of America. 2013;110(30):12301–6. https://doi.org/ 10.1073/pnas.1309947110 PMID: 23840063; PubMed Central PMCID: PMC3725109. 62. Baker ML, Baker MR, Hryc CF, Ju T, Chiu W. Gorgon and pathwalking: macromolecular modeling tools for subnanometer resolution density maps. Biopolymers. 2012; 97(9):655–68. https://doi.org/10.1002/ bip.22065 PMID: 22696403; PubMed Central PMCID: PMC3899894. 63. Pettersen EF, Goddard TD, Huang CC, Couch GS, Greenblatt DM, Meng EC, et al. UCSF chimera—A visualization system for exploratory research and analysis. J Comput Chem. 2004; 25(13):1605–12. https://doi.org/10.1002/jcc.20084 WOS:000223379100005. PMID: 15264254 64. Chen MY, Baker ML. Automation and assessment of de novo modeling with Pathwalking in near atomic resolution cryoEM density maps. Journal of structural biology. 2018; 204(3):555–63. https://doi.org/10. 1016/j.jsb.2018.09.005 WOS:000454373000021. PMID: 30237066 65. Chen MY, Baldwin PR, Ludtke SJ, Baker ML. De Novo modeling in cryo-EM density maps with Pathwalking. Journal of structural biology. 2016; 196(3):289–98. https://doi.org/10.1016/j.jsb.2016.06. 004 WOS:000389295000001. PMID: 27436409 66. Li YQ, Zhang Y. REMO: A new protocol to refine full atomic protein models from C-alpha traces by opti- mizing hydrogen-bonding networks. Proteins-Structure Function and Bioinformatics. 2009; 76(3):665– 74. https://doi.org/10.1002/prot.22380 WOS:000267963900012. PMID: 19274737 67. Adams PD, Afonine PV, Bunkoczi G, Chen VB, Davis IW, Echols N, et al. PHENIX: a comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr D. 2010; 66:213–21. https://doi.org/10.1107/S0907444909052925 WOS:000273820800013. PMID: 20124702 68. Emsley P, Cowtan K. Coot: model-building tools for molecular graphics. Acta Crystallogr D. 2004; 60:2126–32. https://doi.org/10.1107/S0907444904019158 WOS:000225360500002. PMID: 15572765 69. Chen VB, Arendall WB, Headd JJ, Keedy DA, Immormino RM, Kapral GJ, et al. MolProbity: all-atom structure validation for macromolecular crystallography. Acta Crystallographica Section D-Structural Biology. 2010; 66:12–21. https://doi.org/10.1107/S0907444909042073 WOS:000273758800003. PMID: 20057044 70. Krissinel E, Henrick K. Inference of macromolecular assemblies from crystalline state. Journal of molec- ular biology. 2007; 372(3):774–97. https://doi.org/10.1016/j.jmb.2007.05.022 WOS:000249494800018. PMID: 17681537 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011341 April 21, 2023 26 / 26 PLOS PATHOGENS
10.1371_journal.pone.0284934	RESEARCH ARTICLE Rotavirus infections and their genotype distribution in Rwanda before and after the introduction of rotavirus vaccination Jean-Claude Kabayiza1,2☯, Staffan Nilsson3☯, Maria AnderssonID 4,5☯* 1 Department of Paediatrics, School of Medicine and Pharmacy, University of Rwanda, Kigali, Rwanda, 2 Department of Paediatrics, University Teaching Hospital of Kigali, Kigali, Rwanda, 3 Department of Laboratory Medicine, Institute of Biomedicine, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden, 4 Department of Infectious Diseases, Institute of Biomedicine, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden, 5 Department of Clinical Microbiology, Sahlgrenska University Hospital, Gothenburg, Sweden ☯ These authors contributed equally to this work. * maria.andersson.3@gu.se Abstract Rotavirus vaccination has reduced mortality and hospital admissions due to rotavirus diar- rhoea, but its effect on rotavirus infections and the impact of rotavirus genotypes are still unclear. Real-time PCR was used to detect rotavirus and other pathogens in faeces sam- ples from children below five years of age with acute diarrhoea, collected before (n = 827) and after (n = 807, 92% vaccinated) the introduction of vaccination in Rwanda in 2012. Rota- virus was genotyped by targeting VP7 to identify G1, G2, G3, G4, G9 and G12 and VP4 to identify P[4], P[6] and P[8]. In vaccinated children, rotavirus infections were rarer (34% vs. 47%) below 12 months of age, severe dehydration was less frequent, and rotavirus was more often found as a co-infecting agent. (79% vs 67%, p = 0.004). Norovirus genogroup II, astrovirus, and sapovirus were significantly more often detected in vaccinated children. The predominant rotavirus genotypes were G2P[4] and G12P[6] in 2009–2010 (50% and 12%), G9P[8] and G1P[8] in 2011–2012 (51% and 22%), and G12P[8] in 2014–2015 (63%). Rota- virus vaccination in Rwanda has reduced the severity of rotavirus gastroenteritis and rotavi- rus infection frequency during the first year of life. Rotavirus infections were frequent in vaccinated children with diarrhoea, often as co-pathogen. Rotavirus genotype changes might be unrelated to vaccination because shifts were observed also before its introduction. Introduction Acute gastroenteritis is a major cause of disease and death among young children and infants in low-income countries. A wide range of viruses, bacteria and protozoa can induce infectious diarrhoea. Rotavirus is one of the most important aetiologies, which prior to vaccination caused an estimated 500,000 deaths every year. Two rotavirus vaccines, the monovalent Rotarix (GlaxoSmithKline) and the pentavalent RotaTeq (Merck) have been available several years, and more recently ROTAVAC (Bharat Biotech) and ROTASIIL (Serum Institute of a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Kabayiza J-C, Nilsson S, Andersson M (2023) Rotavirus infections and their genotype distribution in Rwanda before and after the introduction of rotavirus vaccination. PLoS ONE 18(4): e0284934. https://doi.org/10.1371/journal. pone.0284934 Editor: Mrinmoy Sanyal, Stanford University School of Medicine, UNITED STATES Received: December 15, 2022 Accepted: April 11, 2023 Published: April 25, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0284934 Copyright: © 2023 Kabayiza et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284934 April 25, 2023 1 / 12 PLOS ONEFunding: The number I previously provided was a project number (2008-333) while the number 2006-006238 is a decision number for the collaboration. The funders of this project, SIDA, is the Sweden’s government agency for development cooperation. Money used in these studies has been paid in a collaboration between the University of Gothenburg and the University of Rwanda for higher education. This means that the funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Rotavirus infections in Rwanda before and after the introduction of rotavirus vaccination India) have been available and WHO-prequalified. Rotavirus vaccination of all children was recommended by WHO in 2009 [1], and in Africa vaccination has been implemented in the national immunization program in over 70% of countries. In Rwanda, rotavirus vaccine was implemented in May 2012 as three doses of RotaTeq vaccine given at 6, 10, and 14 weeks of age, and the vaccination coverage today reached > 97%. Studies in Latin America, the United States and Europe have shown that rotavirus vaccina- tion reduces severe rotavirus diarrhoea by 69–90% and hospitalization due to rotavirus by 82– 98% [2–5]. A meta-analysis from studies in 24 countries between 2006 and 2016 established that vaccination reduces both mortality and severe rotavirus diarrhoea, in particular in chil- dren below 2 years of age [6]. Studies from East Africa report 40–70% reduction of hospital admissions [7–10] and a reduction (39–61%) of severe diarrhoea due to rotavirus [11–14]. A recent merge of data from the sub-Saharan Africa region shows an reduction, with internal deviations, of rotavirus positive cases from 42% during pre-vaccination period to 21% in post- vaccination period [15]. Most of the rotavirus infections in humans are caused by rotavirus genogroup A, in which several genotypes have been identified on the basis of variability in the VP7 glycoprotein (G types) and the protease-sensitive VP4 protein (P types), which surround the outer capsid. The most prevalent rotavirus genotypes, identified in more than 80% of human infections during last decades, are G1P[8], G2P[4], G3P[8], G4P[8] and G9P[8] [16]. Additionally, G12P[8] has become frequently detected in recent years [17, 18]. Rotavirus genotypes have been assessed before and after vaccine introduction in Europe, USA, Latin America and Australia [19–22]. These studies report differences in genotype distri- bution after vaccine introduction, which to a large extent might represent normal shifts induced by acquisition of immunity to a circulating strain, but the results from Australia sug- gest that the vaccination indeed may influence the spectrum of circulating genotypes. Studies on the potential impact of vaccination on circulating rotavirus genotypes in African countries are not conclusive. In some studies genotype G2P[4] and G12P[8] were frequently observed after vaccine introduction and possibly associated with hospital admission in vaccinated chil- dren [23, 24]. A report from South Africa showed a temporal association between vaccination and more complex changes in genotype distribution [25]. Polymicrobial enteric infections are frequent in low-income countries [26–28], and it is often difficult to identify the causative agent in individual cases. However, some agents, in par- ticular rotavirus, Shigella, Cryptosporidium and Escherichia coli with heat stable toxin (ETEC- estA) have been more strongly associated with diarrhoea, and of those, rotavirus has had the highest odds ratio reflecting that it has been rarely detected among healthy controls [26–28]. If rotavirus vaccination reduces the number of rotavirus infections, or diarrhoea due to rotavi- rus, then one would expect, as a mathematical effect, a relative increase of the frequency of infections or diarrhoea by other diarrheagenic pathogens, but to our knowledge data on this subject are lacking. In this work we have analysed rotavirus frequency and genotype distribution before and after the introduction of vaccine in Rwanda, and related the findings to the frequency of other enteric pathogens and to the degree of dehydration in vaccinated and unvaccinated children. Materials and methods Patients and samples In total 1634 children with diarrhoea were included during two time periods, before and after the introduction of vaccination using the RotaTec vaccine, which according to the National Institute of Statistics of Rwanda had coverage of 98% during 2013–2015. Recruitment of PLOS ONE \| https://doi.org/10.1371/journal.pone.0284934 April 25, 2023 2 / 12 PLOS ONERotavirus infections in Rwanda before and after the introduction of rotavirus vaccination patients took place at study centres which were chosen to allow inclusion of both out-patients (health centres) or in-patients (district hospitals and university hospitals) in and around Kigali and Butare. Between November 2009 and June 2012, 827 children were recruited at five Health Centres (n = 444), three District Hospitals (n = 343) and two University Hospitals (n = 40), and between June 2014 and December 2015, 807 children were recruited at six District Hospi- tals (n = 736) and two University Hospitals (n = 71). The inclusion criteria during both periods were age below 5.0 years and diarrhoea, defined as passage of 3 or more loose or watery stools per day, with duration of <96 hours (with or without vomiting or fever). The samples were collected as faeces (when possible) or as rectal swabs. Demographic and clinical data (includ- ing the degree of dehydration) were recorded by a study nurse. Severe dehydration was identi- fied according to the criteria in the WHO classification of dehydration. Nucleic acid extraction and real-time PCR Faeces (approximately 250 μL) was dissolved in 4.5 mL of saline and centrifuged 5 minutes at 750x g. Then, 250 μL of the dissolved faeces or 250 μL of the rectal swab solution were mixed with 2 mL of lysis buffer, and this volume was used for extraction of total nucleic acid in an EasyMag instrument (Biomerieux, Marcy l’E´toile, France). The nucleic acids were eluted in 110 μL of which 5 μL were used for each of the 9 multiplex real-time PCR reactions targeting astrovirus, norovirus genotype I (GI) or genotype II (GII), rotavirus, sapovirus, Campylobacter jejuni, Cryptosporidium parvum/hominis, enterotoxin-producing Escherichia coli (ETEC) cod- ing for heat labile toxin (eltB) or heat stable toxin (estA), enteropathogenic Escherichia coli (EPEC) coding for intimin (eae) or bundle forming pilus (bfpA), Salmonella and Shigella. The target genes for each pathogen and oligonucleotide sequences have been previously describe [26, 27]. Real-time PCR was performed in an ABI 7900 384-well system (Applied Biosystems, Foster City, CA) in 20 μL-reactions containing oligonucleotides and TaqMan Fast Virus 1-step Mas- ter mix (ABI, for RNA targets) or Universal Master mix (ABI, for DNA targets). After a reverse transcription step at 46˚C for 30 min followed by 10 min of denaturation at 95˚C, 45 cycles of two-step PCR was performed (15 s at 95˚C, 60 s at 56˚C). In each run, plasmids containing the target regions for all agents were amplified in parallel with patient specimens to verify the per- formance of each target PCR. Genotyping of rotavirus All rotavirus positive samples were analysed further with a genotype specific real-time PCR targeting the VP7 (G1, G2, G3, G4, G9 and G12) and VP4 (P[4], P[6] and P[8]). This amplifi- cation was carried out in 3 parallel multiplex reactions in a Quant Studio 6 instrument (Applied Biosystems, Carlsbad, CA), as previously described [29]. Each 50 μL reaction mixture contained 10 μL of extracted sample, TaqMan Fast Virus 1-step Master mix (Applied Biosys- tems), and oligonucleotides specific for each genotype. After a reverse transcription step at 46˚C for 30 min followed by 10 min of denaturation at 95˚C, 45 cycles of two-step PCR was performed (15 s at 95˚C, 60 s at 58˚C) with the ramp rate adjusted to 1˚C/s. Sanger sequencing If genotyping was incomplete because only either the VP7 or the VP4 genotyping PCR was reactive, Sanger sequencing of the VP7 and VP4 regions was performed as previously described [30]. Sequencing was also performed in order to verify rare genotype combination, or to identify the genotype in samples that could not be genotyped at all by the genotyping real-time PCR. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284934 April 25, 2023 3 / 12 PLOS ONERotavirus infections in Rwanda before and after the introduction of rotavirus vaccination Statistics The frequencies of rotavirus and other pathogens between unvaccinated and vaccinated were compared, for all children and after stratification into children � 12 months and age 12–36 months of age. To further compare the presence of pathogens other than rotavirus, the chil- dren were stratified by presence and absence of rotavirus. If vaccination protected against rota- virus infection, and if other conditions were equal, one would in children with diarrhoea expect, merely on mathematical basis that the frequencies of other pathogens would be higher in vaccinated than in unvaccinated children. If vaccination protected against rotavirus induced diarrhoea but not against rotavirus infection, one would mathematically expect higher fre- quencies of other pathogens in combination with rotavirus (whereas the rates in children with- out rotavirus should not be influenced by rotavirus vaccination). Fisher’s exact test was used to compare groups as regards categorical data and Mann-Whit- ney U test was used to compare groups as regards numerical data. Logistic regressions with rotavirus as outcome was used to adjust for care level (health centre, hospital) and location. Ethics The study was approved by the ethical committee at the National University in Rwanda and by the regional ethical review board in Gothenburg (ID:052–08), Sweden. The study was approved by the ethical committee at the University of Rwanda and by the regional ethical review board in Gothenburg (ID:052–08), Sweden. After verbal information about the study, written informed consent was obtained from a caregiver for each child included in the study. Results Age and vaccination Children included 2009–2012 had a mean age of 18.6 months, and those included 2014–2015, after the rotavirus vaccine introduction, a mean age of 15.9 months (14 months in vaccinated and 35.1 months in unvaccinated). Table 1 presents the number of included patients during each period, vaccinated or not vaccinated, separated in the age groups <12 months, 12–36 months and >36 months. Rotavirus infection rates in vaccinated and unvaccinated children Rotavirus was detected in the same proportion of all children seeking care because of diarrhoea after (34%) as compared with before (34%) the introduction of rotavirus vaccination. There was no significant difference between vaccinated (34%) and unvaccinated (29%) children in the period after the rotavirus vaccine had been introduced either. The rotavirus frequency was however significantly lower in vaccinated (34%) than in unvaccinated (47%) children below 12 months of age (Table 2). Table 1. Age distribution for children that had or had not received rotavirus vaccination. Total Not vaccinated Period 1 Period 2 Vaccinated (Period 2) n = 1634 n = 901(55%) n = 827 (48%) n = 74 (7%) n = 733 (45%) Period 1, November 2009 to June 2012. Period 2, June 2014 to December 2015. https://doi.org/10.1371/journal.pone.0284934.t001 <12 months 12–36 months �36 months n = 679 330 (49%) 326 (48%) 4 (0.6%) 349 (51%) n = 805 430 (53%) 396 (49%) 34 (4%) 375 (47%) n = 150 141 (94%) 105 (70%) 36 (24%) 9 (6%) PLOS ONE \| https://doi.org/10.1371/journal.pone.0284934 April 25, 2023 4 / 12 PLOS ONERotavirus infections in Rwanda before and after the introduction of rotavirus vaccination Table 2. The frequency of rotavirus infection in vaccinated and unvaccinated children by age group. Age<12 Rotavirus Age 12–36 Rotavirus Unvaccinated 330 155 (47%) 430 128 (30%) Vaccinated 349 117 (34%) 375 133 (35%) OR 0.57 1.30 P ORadj 0.0004 0.1 0.87 0.99 Padj 0.003 0.93 OR, odds ratio; ORadj and Padj, adjusted for health care level and location. https://doi.org/10.1371/journal.pone.0284934.t002 Severe dehydration was more common in unvaccinated children who had as compared with those who did not have rotavirus infection (18.3% vs. 8.1%, p<0.01). By contrast, in vacci- nated children, severe dehydration was observed at similar rates in rotavirus positive (3.6%) and rotavirus negative (4.1%) children (Table 3). The rotavirus concentration in faeces was not influenced by vaccination: the median threshold cycle value for rotavirus was identical (21.6) in unvaccinated and vaccinated children. Infections with other pathogens As shown in Fig 1A and 1B and S1 Table, astrovirus, norovirus GII and sapovirus infections were significantly more common, and Cryptosporidium infections less common, in children that were rotavirus vaccinated than in those who were not, in particular in the group less than 12 month. This relative increase of some viral infections in vaccinated children was mainly seen in children below 12 months of age co-infected with rotavirus, but to some extent seen also in those that did not also have rotavirus. Co-infections between rotavirus and other pathogens Rotavirus was significantly more often detected together with co-infecting agents (79% vs. 67%, p = 0.003), and the mean number of co-infecting pathogens was higher (mean 1.75 vs. 1.40, p<0.0001), in vaccinated as compared with unvaccinated children (Fig 2). When only unvaccinated children with rotavirus infections were compared, a smaller (non-significant) difference in the frequency and number of co-infecting pathogens after as compared with before the introduction of vaccination was seen (73% vs. 67%; mean 1.47 vs. 1.39, p = 0.55). Severe dehydration tended to be more frequent when rotavirus was present alone as com- pared with as co-pathogen (24/151 [16%] vs. 40/401 [9.1%]; OR = 1.71, p = 0.07), and this Table 3. Degree of dehydration in children with diarrhea who had or did not have rotavirus detected in faeces, and who had or had not been rotavirus vaccinated. Rotavirus + Rotavirus – All Vaccinated Not vaccinated OR p Vaccinated Not vaccinated OR p Severe Dehydration (n = 133) n = 251 9 (3.6%) No Severe Dehydration (n = 1501) 242 (96.4%) n = 301 55 (18.3%) 246 (81.7%) 0.17 <0.0001 20 (4.1%) n = 482 n = 600 49 (8.2%) 0.49 0.0082 462 (95.9%) 551 (91.8%) <12 month Vaccinated Not vaccinated OR p Vaccinated Not vaccinated OR p Severe Dehydration (n = 66) No Severe Dehydration (n = 613) 12–36 month Severe Dehydration (n = 63) n = 117 5 (4.3%) 112 (95.7%) ‘Vaccinated n = 133 4 (3%) No Severe Dehydration (n = 740) 129 (97%) https://doi.org/10.1371/journal.pone.0284934.t003 n = 155 34 (22%) 121 (78%) 0.16 <0.0001 n = 232 6 (2.6%) 226 (97.4) n = 175 21(12%) 154 (88%) 0.19 0.0002 Not vaccinated OR P Vaccinated Not vaccinated OR p n = 127 21 (16.5%) 106 (83.4%) 0,16 0.0002 14 (5.8%) n = 242 n = 301 24 (8%) 0.71 0.3980 228 (94.2%) 277 (92%) PLOS ONE \| https://doi.org/10.1371/journal.pone.0284934 April 25, 2023 5 / 12 PLOS ONERotavirus infections in Rwanda before and after the introduction of rotavirus vaccination Fig 1. The frequency of other pathogens in vaccinated and unvaccinated children shown as odds ratios (OR) and their 95% confidence intervals, for all children (black), and separated in children with (red) or without (blue) rotavirus infection. 1A represent children below 12 months of age and 1B children between 12 and 36 months of age. https://doi.org/10.1371/journal.pone.0284934.g001 trend was seen both in vaccinated (3/53 [5.7%] vs. 6/198 [3.0%]; OR = 1.92, p = 0.40) and unvaccinated children (21/98 [21%] vs. 34/203 [17%]; OR = 1.35, p = 0.34). Genotype distribution of rotavirus before and after vaccine introduction Out of all 552 rotavirus positive samples, 505 could be genotyped (91% and 92% of the rotavi- rus positive samples from the periods before and after vaccine introduction). As shown in Fig 3, the genotype distribution changed considerably over time. The most common genotypes were G2P[4] (46%) during 2009–2010, G9P[8] (50%) during 2011–2012, and G12P[8] (58%) during 2014–2015, after the rotavirus vaccination was introduced. The prevalence of minor genotypes also fluctuated: G12P[6] was present only before introduction of the vaccine in 2009–2012 (12%), G1P[8] was relatively frequent in 2011 (29%) and 2015 (26%) but rare the other years, and G4P[8] and G8P[4] were present in 2014 (15% and 3%), but essentially absent the other years. Rare genotypes, genotype mixtures, or only either a G or a P type, were observed in 4–14% of the samples over the years. Among the very rare genotypes, G4P[4], G4P[6], G8P[8], G9P [6], G12P[4] were detected in one case each, and G8P[6] in two cases during the whole study period. In total, there were 23 mixed infections with several G and/or P types present in the same sample, and in 27 samples only a G or a P type was detected. In general, detection of only either P or G, as well as the failure to detect any type in 45 samples, was explained by a low viral load, but there were samples from 2011 and 2014, in which a G type was not identified despite detection of P types (6 P[6] and 4 P[4]) with relatively low Ct values. Discussion Rotavirus vaccination was introduced in Rwanda in 2012. This study shows that among chil- dren seeking care because of diarrhoea, rotavirus vaccination has reduced the frequency of rotavirus infections in those less than 12 months of age, but not in the older children. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284934 April 25, 2023 6 / 12 PLOS ONERotavirus infections in Rwanda before and after the introduction of rotavirus vaccination Fig 2. The percentage of samples without or with one or several pathogens detected together with rotavirus before (2009–2010) and after (2014–2015) the introduction of rotavirus vaccination. https://doi.org/10.1371/journal.pone.0284934.g002 Moreover, in vaccinated children, those with rotavirus infection less often had severe dehydra- tion and rotavirus was more often presented as a co-infection together with other pathogens than in unvaccinated children. The finding that the frequency of rotavirus infections was not generally lower in vaccinated children agrees with a recent study in Malawi in which rotavirus was detected in 32% of chil- dren with diarrhoea before and in 29% after vaccination started [8]. These rates are higher than seen in Latin America, where rotavirus has been detected in 11–20% of children with diarrheal infections after vaccination commenced [5]. High rotavirus rates subsequent years after the introduction of vaccination have also been observed in Kenya (31.5%), Ghana (26%) and Togo (36%), but in the latter two studies even higher frequencies (50% and 53%) were observed before vaccination [31–33]. Despite these reports of a prevailing high prevalence of rotavirus infections shortly after vaccine introduction several African studies have shown high vaccine efficiency against hospital admission (54–75%) [7, 32, 34–36]. In the present study, the impact on hospitalization could not be evaluated, but severe dehydration was rarer in vacci- nated than in unvaccinated children also when age differences were considered. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284934 April 25, 2023 7 / 12 PLOS ONERotavirus infections in Rwanda before and after the introduction of rotavirus vaccination Fig 3. A. Genotype distribution before (2009–2010) and after (2014–2015) the introduction of rotavirus vaccination. RotaTeq+, vaccinated; RotaTeq–, not vaccinated. B. Rotavirus genotype distribution in all samples taken between 2009 and 2015. C. Number of samples with different P (VP4) and G (VP7) type combinations. https://doi.org/10.1371/journal.pone.0284934.g003 Our results show that in vaccinated children, other pathogens have become relatively more frequent, in particular norovirus GII (18% vs. 10%), astrovirus (10% vs. 4%) and sapovirus (13% vs. 4%). Similarly, previous studies have reported higher rates of norovirus GII infections in children with acute gastroenteritis after rotavirus vaccine implementation [37–39]. These increased rates in vaccinated children likely reflect a relative increase due to a decline of rotavi- rus diarrhoea rather than an absolute increase in number of infections. Cryptosporidium was less frequent after as compared with before the introduction of vaccination, possibly reflecting seasonal variation of this pathogen. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284934 April 25, 2023 8 / 12 PLOS ONERotavirus infections in Rwanda before and after the introduction of rotavirus vaccination In our previous studies, rotavirus infections were rare among healthy controls, and the association with diarrhoea was stronger than for other pathogens, suggesting that rotavirus was usually the cause of diarrhoea also when other pathogens were present [26, 27]. In the present study, rotavirus was more often detected as a co-infection together with other patho- gens in vaccinated as compared with unvaccinated children. In many of these cases, rotavirus was probably not the cause of diarrhoea, or at least not the main cause. The finding that rotavi- rus often is a co-infecting pathogen in vaccinated children with diarrhoea is an important observation of our study. It points at a risk, that the effect of rotavirus vaccination may be underestimated if other pathogens are not analysed in studies that evaluate the effect of vacci- nation. A resent meta-analysis of enteric pathogens among children under 5 years in sub-Saha- ran Africa highlighted the importance of analysing a wide range of potentially causative agents, in order to identify the true aetiologies of diarrheal disease and plan for prevention and treatment [40]. To investigate to what extent rotavirus infections in vaccinated children con- tribute to symptoms, additional studies including also healthy controls are required. Even if severe disease was rare in vaccinated children, the persistent circulation of rotavirus in the population is problematic. First, it remains a threat to children who respond poorly to vaccination or who are not vaccinated. Second, circulating rotaviruses might evolve, by re- assortment with each other or with animal strains, into novel and more virulent strains. There- fore, surveillance of rotavirus infections and their genotype distribution should continue, to be able to detect possible vaccine breakthroughs. A potential advantage of a continuing circula- tion of rotavirus might be that vaccinated persons would be boosted, and therefore maintain protective immunity against rotavirus disease. In the present study, the spectrum of rotavirus genotype changed radically between 2009 and 2015. After rotavirus vaccination was introduced in 2012, there was a shift from predomi- nance of G9P[8] in 2011–2012 to predominance of G12[P8] in 2014–2015 paralleled by an emergence of G4[P8] and G8[P4] and re-emergence of G1[P8]. These changes might be related to genotype differences in the protection of the RotaTeq vaccine, but it is likely that they to a large extent reflect natural genotype shifts, which have been observed in several previ- ous studies and also were seen in the present study during the pre-vaccination period. In summary, we report that after the introduction of rotavirus vaccination the frequency of rotavirus infections in Rwanda overall has remained high, but has decreased significantly in children less than 12 month. In vaccinated children, rotavirus infections were more rarely pre- sented with severe dehydration and were more often accompanied by another pathogens, which likely was the cause of diarrhoea in many of these cases. Supporting information S1 Table. Pathogen detection rates in children who had or had not received rotavirus vac- cine, and did or did not have rotavirus infection. (DOCX) Acknowledgments We thank the nurses and laboratory personnel in Rwanda for their dedicated work with inclu- sion of patients and collection of samples in this project. Author Contributions Conceptualization: Jean-Claude Kabayiza, Staffan Nilsson, Maria Andersson. Data curation: Jean-Claude Kabayiza, Maria Andersson. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284934 April 25, 2023 9 / 12 PLOS ONERotavirus infections in Rwanda before and after the introduction of rotavirus vaccination Formal analysis: Maria Andersson. Methodology: Maria Andersson. Project administration: Jean-Claude Kabayiza. Software: Staffan Nilsson, Maria Andersson. Visualization: Staffan Nilsson, Maria Andersson. Writing – original draft: Jean-Claude Kabayiza, Staffan Nilsson, Maria Andersson. Writing – review & editing: Jean-Claude Kabayiza, Staffan Nilsson, Maria Andersson. References 1. Rotavirus vaccines:an update. Wkly Epidemiol Rec. 2009; 84(50):533–40. PMID: 20034143. 2. Jiang V, Jiang B, Tate J, Parashar UD, Patel MM. Performance of rotavirus vaccines in developed and developing countries. Hum Vaccin. 2010; 6(7):532–42. https://doi.org/10.4161/hv.6.7.11278 PMID: 20622508; PubMed Central PMCID: PMC3322519. 3. Karafillakis E, Hassounah S, Atchison C. Effectiveness and impact of rotavirus vaccines in Europe, 2006–2014. Vaccine. 2015; 33(18):2097–107. Epub 20150318. https://doi.org/10.1016/j.vaccine.2015. 03.016 PMID: 25795258. 4. Rha B, Tate JE, Payne DC, Cortese MM, Lopman BA, Curns AT, et al. Effectiveness and impact of rota- virus vaccines in the United States—2006–2012. Expert Rev Vaccines. 2014; 13(3):365–76. Epub 20140106. https://doi.org/10.1586/14760584.2014.877846 PMID: 24392657. 5. Santos VS, Marques DP, Martins-Filho PR, Cuevas LE, Gurgel RQ. Effectiveness of rotavirus vaccines against rotavirus infection and hospitalization in Latin America: systematic review and meta-analysis. Infect Dis Poverty. 2016; 5(1):83. Epub 20160812. https://doi.org/10.1186/s40249-016-0173-2 PMID: 27514855; PubMed Central PMCID: PMC4982225. 6. Jonesteller CL, Burnett E, Yen C, Tate JE, Parashar UD. Effectiveness of Rotavirus Vaccination: A Sys- tematic Review of the First Decade of Global Postlicensure Data, 2006–2016. Clin Infect Dis. 2017; 65 (5):840–50. https://doi.org/10.1093/cid/cix369 PMID: 28444323. 7. Abeid KA, Jani B, Cortese MM, Kamugisha C, Mwenda JM, Pandu AS, et al. Monovalent Rotavirus Vaccine Effectiveness and Impact on Rotavirus Hospitalizations in Zanzibar, Tanzania: Data From the First 3 Years After Introduction. J Infect Dis. 2017; 215(2):183–91. https://doi.org/10.1093/infdis/jiw524 PMID: 27815381. 8. Bar-Zeev N, Kapanda L, Tate JE, Jere KC, Iturriza-Gomara M, Nakagomi O, et al. Effectiveness of a monovalent rotavirus vaccine in infants in Malawi after programmatic roll-out: an observational and case-control study. Lancet Infect Dis. 2015; 15(4):422–8. Epub 20150129. https://doi.org/10.1016/ S1473-3099(14)71060-6 PMID: 25638521; PubMed Central PMCID: PMC4374102. 9. Mujuru HA, Burnett E, Nathoo KJ, Ticklay I, Gonah NA, Mukaratirwa A, et al. Monovalent Rotavirus Vaccine Effectiveness Against Rotavirus Hospitalizations Among Children in Zimbabwe. Clin Infect Dis. 2019; 69(8):1339–44. https://doi.org/10.1093/cid/ciy1096 PMID: 30590488; PubMed Central PMCID: PMC8851377. 10. Ngabo F, Tate JE, Gatera M, Rugambwa C, Donnen P, Lepage P, et al. Effect of pentavalent rotavirus vaccine introduction on hospital admissions for diarrhoea and rotavirus in children in Rwanda: a time- series analysis. Lancet Glob Health. 2016; 4(2):e129–36. https://doi.org/10.1016/S2214-109X(15) 00270-3 PMID: 26823214. 11. Armah GE, Sow SO, Breiman RF, Dallas MJ, Tapia MD, Feikin DR, et al. Efficacy of pentavalent rotavi- rus vaccine against severe rotavirus gastroenteritis in infants in developing countries in sub-Saharan Africa: a randomised, double-blind, placebo-controlled trial. Lancet. 2010; 376(9741):606–14. Epub 20100806. https://doi.org/10.1016/S0140-6736(10)60889-6 PMID: 20692030. 12. Cunliffe NA, Witte D, Ngwira BM, Todd S, Bostock NJ, Turner AM, et al. Efficacy of human rotavirus vaccine against severe gastroenteritis in Malawian children in the first two years of life: a randomized, double-blind, placebo controlled trial. Vaccine. 2012; 30 Suppl 1:A36–43. https://doi.org/10.1016/j. vaccine.2011.09.120 PMID: 22520135; PubMed Central PMCID: PMC3982044. 13. Lamberti LM, Ashraf S, Walker CL, Black RE. A Systematic Review of the Effect of Rotavirus Vaccina- tion on Diarrhea Outcomes Among Children Younger Than 5 Years. Pediatr Infect Dis J. 2016; 35 (9):992–8. https://doi.org/10.1097/INF.0000000000001232 PMID: 27254030. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284934 April 25, 2023 10 / 12 PLOS ONERotavirus infections in Rwanda before and after the introduction of rotavirus vaccination 14. Madhi SA, Kirsten M, Louw C, Bos P, Aspinall S, Bouckenooghe A, et al. Efficacy and immunogenicity of two or three dose rotavirus-vaccine regimen in South African children over two consecutive rotavirus- seasons: a randomized, double-blind, placebo-controlled trial. Vaccine. 2012; 30 Suppl 1:A44–51. https://doi.org/10.1016/j.vaccine.2011.08.080 PMID: 22520136. 15. Godfrey O, Zhang W, Amponsem-Boateng C, Bonney Oppong T, Zhao Q, Li D. Evidence of rotavirus vaccine impact in sub-Saharan Africa: Systematic review and meta-analysis. PLoS One. 2020; 15(4): e0232113. Epub 20200427. https://doi.org/10.1371/journal.pone.0232113 PMID: 32339187; PubMed Central PMCID: PMC7185587. 16. Banyai K, Laszlo B, Duque J, Steele AD, Nelson EA, Gentsch JR, et al. Systematic review of regional and temporal trends in global rotavirus strain diversity in the pre rotavirus vaccine era: insights for understanding the impact of rotavirus vaccination programs. Vaccine. 2012; 30 Suppl 1:A122–30. https://doi.org/10.1016/j.vaccine.2011.09.111 PMID: 22520121. 17. Matthijnssens J, Bilcke J, Ciarlet M, Martella V, Banyai K, Rahman M, et al. Rotavirus disease and vac- cination: impact on genotype diversity. Future Microbiol. 2009; 4(10):1303–16. https://doi.org/10.2217/ fmb.09.96 PMID: 19995190. 18. Rahman M, Matthijnssens J, Yang X, Delbeke T, Arijs I, Taniguchi K, et al. Evolutionary history and global spread of the emerging g12 human rotaviruses. J Virol. 2007; 81(5):2382–90. Epub 20061213. https://doi.org/10.1128/JVI.01622-06 PMID: 17166908; PubMed Central PMCID: PMC1865926. 19. Guerra SF, Linhares AC, Mascarenhas JD, Oliveira A, Justino MC, Soares LS, et al. Rotavirus strain surveillance for three years following the introduction of rotavirus vaccine into Belem, Brazil. J Med Virol. 2015; 87(8):1303–10. Epub 20150416. https://doi.org/10.1002/jmv.24183 PMID: 25879653. 20. Hull JJ, Teel EN, Kerin TK, Freeman MM, Esona MD, Gentsch JR, et al. United States rotavirus strain surveillance from 2005 to 2008: genotype prevalence before and after vaccine introduction. Pediatr Infect Dis J. 2011; 30(1 Suppl):S42–7. https://doi.org/10.1097/INF.0b013e3181fefd78 PMID: 21183839. 21. Roczo-Farkas S, Kirkwood CD, Cowley D, Barnes GL, Bishop RF, Bogdanovic-Sakran N, et al. The Impact of Rotavirus Vaccines on Genotype Diversity: A Comprehensive Analysis of 2 Decades of Aus- tralian Surveillance Data. J Infect Dis. 2018; 218(4):546–54. https://doi.org/10.1093/infdis/jiy197 PMID: 29790933. 22. Usonis V, Ivaskeviciene I, Desselberger U, Rodrigo C, Pediatric REC. The unpredictable diversity of co- circulating rotavirus types in Europe and the possible impact of universal mass vaccination programmes on rotavirus genotype incidence. Vaccine. 2012; 30(31):4596–605. Epub 20120510. https://doi.org/10. 1016/j.vaccine.2012.04.097 PMID: 22579864. 23. Matthijnssens J, Zeller M, Heylen E, De Coster S, Vercauteren J, Braeckman T, et al. Higher proportion of G2P[4] rotaviruses in vaccinated hospitalized cases compared with unvaccinated hospitalized cases, despite high vaccine effectiveness against heterotypic G2P[4] rotaviruses. Clin Microbiol Infect. 2014; 20(10):O702–10. Epub 20140329. https://doi.org/10.1111/1469-0691.12612 PMID: 24580887. 24. Neves MA, Pinheiro HH, Silva RS, Linhares AC, Silva LD, Gabbay YB, et al. High prevalence of G12P [8] rotavirus strains in Rio Branco, Acre, Western Amazon, in the post-rotavirus vaccine introduction period. J Med Virol. 2016; 88(5):782–9. Epub 20151103. https://doi.org/10.1002/jmv.24404 PMID: 26466923. 25. Page NA, Seheri LM, Groome MJ, Moyes J, Walaza S, Mphahlele J, et al. Temporal association of rota- virus vaccination and genotype circulation in South Africa: Observations from 2002 to 2014. Vaccine. 2018; 36(47):7231–7. Epub 20171027. https://doi.org/10.1016/j.vaccine.2017.10.062 PMID: 29110933; PubMed Central PMCID: PMC6497815. 26. Elfving K, Andersson M, Msellem MI, Welinder-Olsson C, Petzold M, Bjorkman A, et al. Real-time PCR threshold cycle cutoffs help to identify agents causing acute childhood diarrhea in Zanzibar. J Clin Microbiol. 2014; 52(3):916–23. Epub 20140108. https://doi.org/10.1128/JCM.02697-13 PMID: 24403298; PubMed Central PMCID: PMC3957757. 27. Kabayiza JC, Andersson ME, Nilsson S, Bergstrom T, Muhirwa G, Lindh M. Real-time PCR identifica- tion of agents causing diarrhea in Rwandan children less than 5 years of age. Pediatr Infect Dis J. 2014; 33(10):1037–42. https://doi.org/10.1097/INF.0000000000000448 PMID: 25037040. 28. Kotloff KL, Nataro JP, Blackwelder WC, Nasrin D, Farag TH, Panchalingam S, et al. Burden and aetiol- ogy of diarrhoeal disease in infants and young children in developing countries (the Global Enteric Multi- center Study, GEMS): a prospective, case-control study. Lancet. 2013; 382(9888):209–22. Epub 20130514. https://doi.org/10.1016/S0140-6736(13)60844-2 PMID: 23680352. 29. Andersson M, Lindh M. Rotavirus genotype shifts among Swedish children and adults-Application of a real-time PCR genotyping. J Clin Virol. 2017; 96:1–6. Epub 20170909. https://doi.org/10.1016/j.jcv. 2017.09.005 PMID: 28915451. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284934 April 25, 2023 11 / 12 PLOS ONERotavirus infections in Rwanda before and after the introduction of rotavirus vaccination 30. van Doorn LJ, Kleter B, Hoefnagel E, Stainier I, Poliszczak A, Colau B, et al. Detection and genotyping of human rotavirus VP4 and VP7 genes by reverse transcriptase PCR and reverse hybridization. J Clin Microbiol. 2009; 47(9):2704–12. Epub 20090624. https://doi.org/10.1128/JCM.00378-09 PMID: 19553575; PubMed Central PMCID: PMC2738096. 31. Agutu MT, Ongus J, Kombich J, Kamenwa R, Nyangao J, Kagira J, et al. Prevalence and genetic diver- sity of rotavirus infection in children with acute gastroenteritis in a hospital setting, Nairobi Kenya in post vaccination era: a cross-sectional study. Pan Afr Med J. 2017; 26:38. Epub 20170124. https://doi.org/ 10.11604/pamj.2017.26.38.10312 PMID: 28451016; PubMed Central PMCID: PMC5398238. 32. Armah G, Pringle K, Enweronu-Laryea CC, Ansong D, Mwenda JM, Diamenu SK, et al. Impact and Effectiveness of Monovalent Rotavirus Vaccine Against Severe Rotavirus Diarrhea in Ghana. Clin Infect Dis. 2016; 62 Suppl 2:S200–7. https://doi.org/10.1093/cid/ciw014 PMID: 27059357; PubMed Central PMCID: PMC7962386. 33. Tsolenyanu E, Mwenda JM, Dagnra A, Leshem E, Godonou M, Nassoury I, et al. Early Evidence of Impact of Monovalent Rotavirus Vaccine in Togo. Clin Infect Dis. 2016; 62 Suppl 2:S196–9. https://doi. org/10.1093/cid/civ1182 PMID: 27059356. 34. Bar-Zeev N, Jere KC, Bennett A, Pollock L, Tate JE, Nakagomi O, et al. Population Impact and Effec- tiveness of Monovalent Rotavirus Vaccination in Urban Malawian Children 3 Years After Vaccine Intro- duction: Ecological and Case-Control Analyses. Clin Infect Dis. 2016; 62 Suppl 2:S213–9. https://doi. org/10.1093/cid/civ1183 PMID: 27059359; PubMed Central PMCID: PMC4825885. 35. Beres LK, Tate JE, Njobvu L, Chibwe B, Rudd C, Guffey MB, et al. A Preliminary Assessment of Rotavi- rus Vaccine Effectiveness in Zambia. Clin Infect Dis. 2016; 62 Suppl 2:S175–82. https://doi.org/10. 1093/cid/civ1206 PMID: 27059353. 36. Tate JE, Ngabo F, Donnen P, Gatera M, Uwimana J, Rugambwa C, et al. Effectiveness of Pentavalent Rotavirus Vaccine Under Conditions of Routine Use in Rwanda. Clin Infect Dis. 2016; 62 Suppl 2: S208–12. https://doi.org/10.1093/cid/civ1016 PMID: 27059358. 37. Doll MK, Gagneur A, Tapiero B, Charest H, Gonzales M, Buckeridge DL, et al. Temporal Changes in Pediatric Gastroenteritis after Rotavirus Vaccination in Quebec. Pediatr Infect Dis J. 2016; 35(5):555– 60. https://doi.org/10.1097/INF.0000000000001077 PMID: 26855410. 38. Lopman BA, Steele D, Kirkwood CD, Parashar UD. The Vast and Varied Global Burden of Norovirus: Prospects for Prevention and Control. PLoS Med. 2016; 13(4):e1001999. Epub 20160426. https://doi. org/10.1371/journal.pmed.1001999 PMID: 27115709; PubMed Central PMCID: PMC4846155. 39. Santos VS, Gurgel RQ, Cavalcante SM, Kirby A, Cafe LP, Souto MJ, et al. Acute norovirus gastroenteri- tis in children in a highly rotavirus-vaccinated population in Northeast Brazil. J Clin Virol. 2017; 88:33–8. Epub 20161029. https://doi.org/10.1016/j.jcv.2016.10.015 PMID: 28160726. 40. Oppong TB, Yang H, Amponsem-Boateng C, Kyere EKD, Abdulai T, Duan G, et al. Enteric pathogens associated with gastroenteritis among children under 5 years in sub-Saharan Africa: a systematic review and meta-analysis. Epidemiol Infect. 2020; 148:e64. Epub 20200302. https://doi.org/10.1017/ S0950268820000618 PMID: 32115003; PubMed Central PMCID: PMC7118358. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284934 April 25, 2023 12 / 12 PLOS ONE
10.1371_journal.ppat.1010103	RESEARCH ARTICLE γδ T cell IFNγ production is directly subverted by Yersinia pseudotuberculosis outer protein YopJ in mice and humans Timothy H. ChuID Yue ZhangID Vincent W. Yang3, James B. Bliska6, Brian S. SheridanID 1,2, Camille KhairallahID 1,2, Onur Eskiocak4, David G. Thanassi1,2, Mark H. KaplanID 1,2* 1,2, Jason Shieh3, Rhea Cho1,2, Zhijuan QiuID 1,2, 5, Semir Beyaz4, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 1 Department of Microbiology and Immunology, Renaissance School of Medicine, Stony Brook University, Stony Brook, New York, United States of America, 2 Center for Infectious Diseases, Renaissance School of Medicine, Stony Brook University, Stony Brook, New York, United States of America, 3 Department of Medicine, Renaissance School of Medicine, Stony Brook University, Stony Brook, New York, United States of America, 4 Cold Spring Harbor Laboratory, Cold Spring Harbor, New York, United States of America, 5 Department of Microbiology and Immunology, School of Medicine, Indiana University, Indianapolis, Indiana, United States of America, 6 Department of Microbiology and Immunology, Geisel School of Medicine, Dartmouth College, Dartmouth, New Hampshire, United States of America OPEN ACCESS Citation: Chu TH, Khairallah C, Shieh J, Cho R, Qiu Z, Zhang Y, et al. (2021) γδ T cell IFNγ production is directly subverted by Yersinia pseudotuberculosis outer protein YopJ in mice and humans. PLoS Pathog 17(12): e1010103. https:// doi.org/10.1371/journal.ppat.1010103 Editor: Denise M. Monack, Stanford University School of Medicine, UNITED STATES Received: July 29, 2021 Accepted: November 9, 2021 Published: December 6, 2021 Copyright: © 2021 Chu et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The RNA-seq datasets generated during this study are available at Bioproject accession number PRJNA742496 in the NCBI Bioproject database (http://www.ncbi. nlm.nih.gov/bioproject/742496). Funding: This work was supported by The G. Harold and Leila Y. Mathers Foundation grant MF- 1901-00210 (B.S.S.), the NIH grants T32 AI007539 (T.H.C.), R01 AI099222 (J.B.B.), K12 GM102778 (Z.Q.), and R01 AI141633 (D.G.T.), and funds provided by The Research Foundation for the State * brian.sheridan@stonybrook.edu Abstract Yersinia pseudotuberculosis is a foodborne pathogen that subverts immune function by translocation of Yersinia outer protein (Yop) effectors into host cells. As adaptive γδ T cells protect the intestinal mucosa from pathogen invasion, we assessed whether Y. pseudotu- berculosis subverts these cells in mice and humans. Tracking Yop translocation revealed that the preferential delivery of Yop effectors directly into murine Vγ4 and human Vδ2+ T cells inhibited anti-microbial IFNγ production. Subversion was mediated by the adhesin YadA, injectisome component YopB, and translocated YopJ effector. A broad anti-pathogen gene signature and STAT4 phosphorylation levels were inhibited by translocated YopJ. Thus, Y. pseudotuberculosis attachment and translocation of YopJ directly into adaptive γδ T cells is a major mechanism of immune subversion in mice and humans. This study uncov- ered a conserved Y. pseudotuberculosis pathway that subverts adaptive γδ T cell function to promote pathogenicity. Author summary Unconventional γδ T cells are a dynamic immune population important for mucosal pro- tection of the intestine against invading pathogens. We determined that the foodborne pathogen Y. pseudotuberculosis preferentially targets an adaptive subset of these cells to subvert immune function. We found that direct injection of Yersinia outer proteins (Yop) into adaptive γδ T cells inhibited their anti-pathogen functions. We screened all Yop effec- tors and identified YopJ as the sole effector to inhibit adaptive γδ T cell production of IFNγ. We determined that adaptive γδ T cell subversion occurred by limiting activation of the transcription factor STAT4. When we infected mice with Y. pseudotuberculosis PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 1 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function University of New York (B.S.S.) and Stony Brook University (B.S.S.). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. expressing an inactive YopJ, this enhanced the adaptive γδ T cell response and led to greater cytokine production from this subset of cells to aid mouse recovery. This mecha- nism of immune evasion appears conserved in humans as direct injection of Y. pseudotu- berculosis YopJ into human γδ T cells inhibited cytokine production. This suggested to us that Y. pseudotuberculosis actively inhibits the adaptive γδ T cell response through YopJ as a mechanism to evade immune surveillance at the site of pathogen invasion. Introduction Pathogens in the genus Yersinia include three species (Y. pestis, Y. pseudotuberculosis, and Y. enterocolitica) that can cause human disease. Y. pseudotuberculosis and Y. enterocolitica cause enteric infections [1,2] while Y. pestis is the causative agent of bubonic, septicemic, and pneu- monic plague that has claimed over 200 million human lives [3,4]. Bubonic and septicemic plague is transmitted by blood sucking fleas while aerosols spread the pneumonic plague. Despite vaccine availability [5], and sensitivity to antibiotic treatment, pneumonic plague com- monly results in fatality in part due to the rapid course of the infection [6]. Pathogenic Yersinia spp. harbor a virulence plasmid that encodes numerous virulence fac- tors to subvert host immune responses, including IFNγ production [7–9]. Immune cell subver- sion requires Yersinia adherence to host cells through bacterial adhesins and translocation of Yersinia outer proteins (Yop) effectors into the host cell cytoplasm by a type III secretion sys- tem (T3SS). Yersinia spp. predominately target host phagocytes like macrophages, dendritic cells (DC), neutrophils, and B cells to subvert immune function during infection, but injection into other immune populations like conventional T cells has been reported, albeit to a lesser degree than their phagocytic counterparts [10–12]. Yersinia virulence factors include compo- nents of the T3SS (e.g., YopB) and translocated effectors (e.g., YopJ and YopH). YopB forms a pore in the host cell membrane necessary for translocation of Yop effectors [13,14]. Numerous Yop effectors translocate into host cells to inhibit immune responses and promote Yersinia spp. pathogenesis. One notable example is YopJ, an acetyl transferase and a possible cysteine protease that inhibits the mitogen-activated protein kinase (MAPK) pathway and tumor necrosis factor receptor-associated factor (TRAF) ubiquitination [15–19]. YopJ is the major Yop effector responsible for the induction of pyroptosis in macrophages during infection [20] and limits toll-like receptor 4 (TLR4) dependent signaling pathways [21]. While YopJ has no known direct effects on conventional T cell activation, YopP (a YopJ homolog in Y. enterocoli- tica) indirectly inhibits T cell priming via DC subversion [22]. YopH has been reported to have direct effects on conventional T cells in vitro. Transfection of a YopH expression plasmid into Jurkat or human T cells inhibited T cell receptor (TCR) signaling and promoted T cell apoptosis [23,24]. Additionally, stimulation of Jurkat cells with a YopH deficient Y. pseudotu- berculosis restored T cell signaling and IL-2 production [25,26]. Even in this context, it is nota- ble that many of the downstream targets in the αβ TCR signaling pathway were inhibited at an excessively high (>50) multiplicity of infection (MOI) and in vivo relevance is unclear [23,26]. Thus, the role of direct subversion of T cell function, especially among unconventional T cells, by Yersinia spp. remains largely unexplored. γδ T cells make up a large proportion of lymphocytes at barrier surfaces and mucosal tissues including the intestines of mice and humans [27,28]. This is particularly pertinent to infections caused by Y. pseudotuberculosis, which has evolved to invade the intestinal barrier. The activity of γδ T cells can be modulated by numerous cell-intrinsic and environmental factors like the γδ TCR, cytokines, and co-stimulatory or inhibitory receptors [29]. For example, IL-12 and PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 2 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function IL-18 may promote IFNγ production from some γδ T cell subsets whereas IL-1β and IL-23 predominantly drive IL-17A production from other γδ T cell subsets [30–35]. Vγ4Vδ1 (Gar- man nomenclature [36]) T cells have traditionally been considered an innate-like cell. How- ever, our group recently characterized a long-lived CD27- CD44hi Vγ4Vδ1 T cell memory population in the context of foodborne Listeria monocytogenes infection [37,38]. While Vγ4 T cells are typically programmed for IL-17A production, this subset has the multifunctional capacity to produce both IL-17A and IFNγ [37]. Similar observations of IFNγ production were made in clonally expanded Vγ4 T cells in response to Staphylococcus aureus in the skin [39]. IFNγ activates macrophages to kill intracellular pathogens or phagocytosed bacteria and induces chemokines that attract immune cells to the site of infection. IFNγ is a critical cytokine in protection from Y. enterocolitica infection [2,40], Y. pestis intranasal challenge [41], and associated with protection from Y. pseudotuberculosis [42]. Interestingly, IFNγ but not IL-17A production from type-3 innate lymphoid cells is critical for the control of foodborne Y. entero- colitica infection [43]. As such, unconventional T cells like Vγ4 T cells that are ideally placed to provide protection against pathogen invasion at mucosal sites may be particularly relevant to Yersinia infections that invade mucosal barriers of the lungs (pneumonic Y. pestis) and gut (Y. pseudotuberculosis and Y. enterocolitica). Despite a foundational understanding of Yersinia pathogenesis, physiologically robust evi- dence linking Yersinia pathogenesis to direct subversion of T cell function is lacking. Here, we uncovered a novel YopJ-dependent immunomodulatory pathway used by Y. pseudotuberculo- sis to directly subvert a murine Vγ4Vδ1 anti-microbial response to aid Y. pseudotuberculosis pathogenesis. Y. pseudotuberculosis also directly subverted a human Vδ2+ T cell IFNγ response, suggesting that this pathway may function similarly in human infection to aid Y. pseudotuberculosis pathogenesis. Results Viable Y. pseudotuberculosis inhibits IFNγ production by adaptive γδ T cells in a YopB- and YadA-dependent manner Initial experiments were carried out to determine if Y. pseudotuberculosis inhibits adaptive γδ T cell function ex vivo. To overcome the extremely low number of Vγ4 T cells in gut-associated lymphoid tissues of naïve specific pathogen free (SPF) mice, a previously established in vivo methodology was utilized to generate a sizable population of adaptive Vγ4 T cells for in vitro manipulation. As such, naïve Balb/c mice were exposed to foodborne L. monocytogenes and MLN enriched in adaptive γδ T cells were isolated 9 days after infection [37], several days after mice typically clear L. monocytogenes [44]. MLN single cell suspensions were infected directly ex vivo with heat-killed or live wild-type (WT) Y. pseudotuberculosis (Yptb) 32777 (Table 1) at a multiplicity of infection (MOI) of 10 for 2 hours. Antibiotics were then added to prevent overgrowth of the live bacteria, and the cultures were incubated an additional 22 hours. Flow cytometry in conjunction with intracellular cytokine staining was used to assess IFNγ produc- tion from Vγ1.1/2- CD44hi CD27- γδ T cells (identifying the adaptive Vγ4 T cell subset [37,38]). Heat-killed Y. pseudotuberculosis elicited a significantly higher IFNγ response from adaptive Vγ4 T cells than was detectable after stimulation with live Y. pseudotuberculosis (Fig 1A). This observation suggests that live Y. pseudotuberculosis subverts adaptive Vγ4 T cell function. The virulence activity of Y. pseudotuberculosis relies substantially on its T3SS and translocation of Yop effectors into host cells. To determine if the T3SS is required for live Y. pseudotuberculosis inhibition of γδ T cell function, MLN single cell suspensions were infected with WT Y. pseudotuberculosis or Y. pseudotuberculosis that were unable to translocate Yop effectors (ΔYopB) or lacked the virulence plasmid that encodes the T3SS (32777c) (Table 1) PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 3 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function Table 1. Y. pseudotuberculosis strains and mutants used in this study. Y. pseudotuberculosis Notation Relevant Characteristics 32777 32777c 32777 YopJC172A 32777 YopHR409A 32777 ΔYopB 32777 YopER144A 32777 YopTC139A 32777 ΔYopM 32777 ΔYpkA 32777 ΔYopK 32777 YopE/β-lac 32777 ΔYopB YopE/β-lac IP2666 IP40 IP2666 ΔInv IP2666 ΔYadA IP2666 ΔInv ΔYadA IP40 ΔInv ΔYadA WT WT2777c YopJC172A YopHR409A ΔYopB YopER144A YopTC139A ΔYopM ΔYpkA ΔYopK WT Yptb-βla ΔYopB Yptb-βla WT ΔYopB ΔInv ΔYadA ΔInv ΔYadA Yptb wild-type serogroup O:1 strain Virulence pYV-cured derivative of 32777 that lacks the T3SS Catalytically inactive YopJ Catalytically inactive YopH Deletion of YopB Catalytically inactive YopE Catalytically inactive YopT Deletion of YopM Deletion of YpkA Frameshift mutation in YopK YopE TME-1 β-lactamase fusion protein Deletion of YopB in the YopE/β-lac Yptb wild-type serogroup O:3 strain IP2666 yopB40 (a stop codon at codon 8 of YopB followed by a frameshift) Deletion of adhesin and invasin Deletion of adhesin and YadA Deletion of adhesin, invasion, and YadA ΔYopB ΔInv ΔYadA Deletion of invasion and YadA in IP40 and pMMB207 mCherry https://doi.org/10.1371/journal.ppat.1010103.t001 References [47] [47] [45] [45,48,49] [45,46] [45] [45] [50] [51] [52] [53–55] [53–55] [47] [56] [12,57] [12,57] [12,57] [12,57] [45–47]. γδ T cell function was assessed 24 hours later as described above. Stimulation with live Y. pseudotuberculosis strains ΔYopB or 32777c restored the IFNγ response of Vγ1.1/2- CD44hi CD27- γδ T cells (Fig 1B), similar to levels seen after stimulation with heat-killed WT Y. pseudotuberculosis (Fig 1A). These data indicate that Y. pseudotuberculosis inhibits IFNγ production by Vγ4 T cells in a manner that requires the T3SS and translocation of Yop effectors. Y. pseudotuberculosis adheres to host cells with the bacterial adhesins invasin (Inv) and YadA to translocate effectors through the T3SS [58–60]. For Y. enterocolitica, both Inv and YadA bind β1-integrin either directly or indirectly through the extracellular matrix, respec- tively [61]. Additionally, β1-integrin expressed on host cells is a known adhesion target for Y. pseudotuberculosis [60,62]. To evaluate the role of these adhesins in the inhibition of γδ T cell function, live Y. pseudotuberculosis with a deletion of Inv (ΔInv), YadA (ΔYadA), or both (ΔInv ΔYadA) (Table 1) were utilized to infect MLN cell suspensions. Vγ1.1/2- CD44hi CD27- γδ T cells stimulated with ΔInv bacteria produced only minimal IFNγ, comparable to unstimu- lated cells or cells stimulated with WT (Fig 1C). In contrast, ΔYadA or ΔInv ΔYadA stimula- tion led to partial restoration of IFNγ production, and stimulation with ΔYopB or ΔYopB ΔInv ΔYadA bacteria led to full restoration of IFNγ production (Fig 1C). Thus, YadA but not Inv contributes to translocation dependent inhibition of IFNγ production by Vγ4 T cells. Translocation of Yop effectors into adaptive γδ T cells by Y. pseudotuberculosis is associated with IFNγ inhibition To determine if Y. pseudotuberculosis can translocate Yop effectors into adaptive Vγ4 T cells, a WT strain expressing a YopE-β-lactamase fusion protein (Yptb-βla) in conjunction with a FRET-based β-lactamase reporter assay was used [63]. YopE translocation into target cells can be readily assessed by a change in fluorescence using flow cytometry. Thus, translocation of the YopE-β-lactamase fusion protein reports Yop effector translocation by emission in the blue range (Yop+) or lack thereof by emission in the green range (Yop-) [10,64,65]. A YopB PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 4 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function Fig 1. Y. pseudotuberculosis inhibition of Vγ1.1/2- CD44hi CD27- γδ T cell function is YopB- and YadA- dependent. MLN cell suspensions from L. monocytogenes infected Balb/c mice were left unstimulated or stimulated with 10 MOI of the indicated Y. pseudotuberculosis for 24 hours. Antibiotics were given 2 hours post-stimulation and brefeldin A was added for the last 5 hours of stimulation. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for IFNγ production. Representative histograms are displayed. (A) Cells were stimulated with live or heat-killed (HK) wild-type (WT) Y. pseudotuberculosis. The graph depicts the mean ± SEM and represents at least two independent experiments with 4 mice/group/experiment. (B) Cells were stimulated with live WT, ΔYopB, or 32777c Y. pseudotuberculosis. The graph depicts the mean ± SEM and represents at least two independent experiments with 4 mice/group/experiment. (C) Cells were stimulated with WT, ΔYopB, ΔYadA, ΔInv, ΔInv ΔYadA, or ΔYopB ΔInv ΔYadA Y. pseudotuberculosis. The graph depicts the mean ± SEM pooled from two independent experiments with 3 mice/group/experiment. ��p < 0.0001, ��p < 0.01, and �p < 0.05. An unpaired t-test was used for (A) and a repeated measures one-way ANOVA was used for (B) and (C). Experimental groups were compared to live WT Y. pseudotuberculosis in (A) and WT Y. pseudotuberculosis in (B) and (C). https://doi.org/10.1371/journal.ppat.1010103.g001 deficient β-lactamase Y. pseudotuberculosis reporter (ΔYopB Yptb-βla) was used as a transloca- tion deficient control. Stimulation of MLN cell suspensions with WT or ΔYopB Yptb-βla con- firmed reporter activity at various MOI (S1A and S1B Fig). Two hours post stimulation with WT Yptb-βla, the majority of Vγ1.1/2- CD44hi CD27- γδ T cells were positive for Yop effector PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 5 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function translocation (Fig 2A). Yop translocation into Vγ1.1/2- CD44hi CD27- γδ T cells was compara- ble to known DC and macrophage targets (Fig 2A). Additionally, Yop translocation was more efficient into Vγ1.1/2- CD44hi CD27- γδ T cells than CD4 or CD8 T cells (Fig 2A). Y. pseudotu- berculosis also preferentially targeted Vγ1.1/2- CD44hi CD27- γδ T cells over CD44- γδ T cells and activated phenotype CD4 or CD8 T cells for Yop translocation (S1C Fig). YadA and Inv promote Yersinia adherence by direct or indirect interactions with the β1-integrin [66–69]. Analysis of β1-integrin expression on Vγ1.1/2- CD44hi CD27- γδ T cells, CD4 T cells, and CD8 T cell revealed that most Vγ1.1/2- CD44hi CD27- γδ T cells expressed the β1-integrin (S2A Fig). In contrast, most conventional CD4 and CD8 T cells did not express the β1-integrin. In addition, use of the WT Yptb-βla reporter for Yop translocation demonstrated that Yop trans- location was associated with higher β1-integrin expression among γδ T cells (S2B Fig). Thus, Y. pseudotuberculosis selectively targets adaptive γδ T cells for Yop translocation among a diverse group of immune populations assessed in an ex vivo culture system. As adaptive Vγ4 T cells were directly targeted with Yop effector translocation, WT Yptb-βla was utilized to determine whether Vγ1.1/2- CD44hi CD27- γδ T cells that contained Yop effec- tors were functionally impaired. An MOI of 1 was used as it provided similarly sized popula- tions of Vγ1.1/2- CD44hi CD27- γδ T cells that did or did not contain translocated effectors from the same culture conditions (S1B Fig). Among WT Yptb-βla stimulated cells, Yop+ Vγ1.1/2- CD44hi CD27- γδ T cells had reduced IFNγ production as compared to their Yop- counterparts (Fig 2B). To extend these results, the ability of Y. pseudotuberculosis to translocate Yop effectors into human γδ T cells and inhibit IFNγ production was assessed in peripheral blood mononuclear cells (PBMC) cultures stimulated with the WT Yptb-βla reporter. Approx- imately 8% of human Vδ2+ T cells were Yop+ and these cells had significantly reduced IFNγ production as compared to the Yop- counterparts (Fig 2C and 2D). These data indicate that Y. pseudotuberculosis is capable of translocating Yop effectors into γδ T cell subsets and inhibiting IFNγ production in mice and humans. YopJ is necessary for Y. pseudotuberculosis to inhibit IFNγ production in adaptive γδ T cells As multiple effectors are translocated into target cells, a panel of yop mutant Y. pseudotubercu- losis (Table 1) [45] was screened to determine if individual Yop effectors inhibit IFNγ produc- tion. Similar to the ΔYopB mutant, stimulation with a catalytically inactive YopJ (YopJC172A) mutant that lacks acetyl transferase activity, but not other mutant Y. pseudotuberculosis, restored IFNγ production in Vγ1.1/2- CD44hi CD27- γδ T cells (Fig 3A). The C172A mutation in YopJ prevents YopJ mediated inhibition of MAPK and NF-κB signaling pathways by abol- ishing its serine and threonine acetylation activity [70]. A similar restoration of IFNγ produc- tion was observed in human Vδ2+ T cells from PBMC of healthy donors upon YopJC172A Y. pseudotuberculosis stimulation (Fig 3B). Thus, the YopJ effector is responsible for inhibition of IFNγ production from murine Vγ4 and human Vδ2+ T cells. YopJ inhibits expression of multiple genes, including ifng, in adaptive γδ T cells To uncover mechanisms by which YopJ inhibits IFNγ production, the transcriptome of cell sorter-purified Vγ1.1/2- CD44hi CD27- γδ T cells after WT or YopJC172A Y. pseudotuberculosis stimulation of MLN cells was assessed by RNA-Seq. Principal component analysis revealed unique gene expression clustering, and approximately 900 genes were expressed at higher lev- els in the YopJC172A stimulation as compared to WT Y. pseudotuberculosis (Fig 4A and 4B). These differentially expressed genes may include genes that are directly inhibited by YopJ PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 6 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function Fig 2. Direct translocation of Yop effectors inhibits the function of murine Vγ4 and human Vδ2+ T cells. MLN suspensions from L. monocytogenes infected mice (A and B) or human PBMC (C and D) were left unstimulated or stimulated with WT or ΔYopB Yptb-βla as indicated. Cells were loaded with CCF4-AM dye prior to stimulation to measure β-lactamase activity. FITC indicates CCF4-AM loaded cells without translocation (Yop-) and BV421 indicates CCF4-AM loaded cells with Yop translocation (Yop+). (A) Adaptive γδ T cells (Vγ1.1/2- CD44hi CD27- γδ T cells), DC (CD11chi MHCIIhi), Macrophages (F4/80+ CD11b+), and CD4 and CD8 T cells were analyzed for Yop translocation 2 hours post stimulation at an MOI of 10. Representative contour plots are displayed. Yop translocation among the indicated populations is depicted as mean ± SEM and is pooled from 2 experiments with a total of 4–8 mice per group. (B) Antibiotics were given 2 hours post-stimulation and brefeldin A was added for the last 5–6 hours of stimulation. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for Yop translocation and IFNγ production 24 hours after stimulation. Representative contour plots and histograms are shown. IFNγ production among the indicated populations is depicted as mean ± SEM and is pooled from 3 experiments with a total of 8 mice per group. (C and D) Antibiotics were given 2 hours post-stimulation and brefeldin A was added for the last 5–6 hours of stimulation. Vδ2+ T cells were analyzed for Yop translocation and IFNγ production post stimulation. Representative contour plots are displayed and IFNγ production is quantified among Yop+ or Yop- Vδ2+ T cells. The graph depicts mean ± SEM and is pooled from 3 experiments with 5 healthy donors per group. ��p < 0.0001, ��p < 0.001, ��p < 0.01, and �p < 0.05. An ordinary one-way ANOVA was used for (A), a repeated measures one-way ANOVA was used for (B), and a paired t-test was used for (D). Comparisons were performed to adaptive γδ T cells in (A), to unstimulated or as depicted in figure in (B), and to Yop+ in (C). https://doi.org/10.1371/journal.ppat.1010103.g002 activity in Vγ1.1/2- CD44hi CD27- γδ T cells or indirectly inhibited by YopJ activity in other cells such as DC or macrophages. To resolve this, the WT Yptb-βla reporter provided an opportunity to evaluate the molecular changes elicited by the activity of translocated Yop in adaptive γδ T cells. The transcriptome of sort purified Yop- and Yop+ Vγ1.1/2- CD44hi CD27- γδ T cells after WT Yptb-βla stimulation was assessed by RNA-Seq. Principal component anal- ysis revealed unique gene expression clustering, and approximately 900 genes were more highly expressed in Yop- vs Yop+ Vγ1.1/2- CD44hi CD27- γδ T cells after WT Yptb-βla stimula- tion (Fig 4C and 4D). Overlapping gene expression profiles from the two datasets were assessed to narrow the analysis to direct YopJ effects on Vγ1.1/2- CD44hi CD27- γδ T cells. This comparison revealed 130 genes that were differentially expressed in both datasets, sug- gesting they are regulated directly by translocated YopJ in adaptive Vγ4 T cells (Fig 4E). These genes were categorized into different groups depending on their known functions. Some dif- ferentially expressed genes play a particular role in anti-infection functions (3.9%), stress sens- ing (1.6%), and lymphocyte activation/regulation (7.9%), genes that may be important for protective T cell responses (Fig 4E and 4F). Among these genes, IFNγ was the single most sig- nificant differentially expressed gene suggesting it is a major target of direct YopJ-mediated inhibition of adaptive γδ T cell function (Fig 4F). Differentially expressed genes among those that promote antimicrobial function included several that are important in augmenting type-1 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 7 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function Fig 3. YopJ is necessary for inhibition of IFNγ production in murine Vγ4 and human Vδ2+ T cells. (A) MLN from L. monocytogenes infected mice were left unstimulated or stimulated with 10 MOI of the indicated Y. pseudotuberculosis for 24 hours. Antibiotics were given 2 hours post- stimulation and brefeldin A was added for the last 5–6 hours. Vγ1/2- CD44hi CD27- γδ T cells were analyzed for IFNγ post stimulation. Representative histograms are displayed. The graph depicts mean ± SEM and represents at least two independent experiments with 2–4 mice per group. (B) Human PBMC were stimulated with 1 MOI of WT or YopJC172A Y. pseudotuberculosis for 24 hours. Antibiotics were given 2 hours post- stimulation. Brefeldin A was added for the last 5–6 hours of stimulation. Vδ2+ γδ T cells were analyzed for IFNγ production post stimulation. Representative flow plots gated on Vδ2+ T cells are displayed. The graph depicts mean ± SEM and is pooled from 2 experiments with 4 healthy donors. ��p < 0.01 and �p < 0.05. A repeated measures one-way ANOVA was used for (A) and a paired t-test was used for (B). Experimental groups were compared to WT Y. pseudotuberculosis. https://doi.org/10.1371/journal.ppat.1010103.g003 and -3 inflammation in T cells. For example, Ptgs2 (encodes cyclooxygenase-2, COX2), Nkg7 (natural killer cell granule protein 7), Prf1 (perforin-1), and Il17a (IL-17A) appear to be regu- lated directly by translocated YopJ in adaptive Vγ4 T cells (Fig 4F) [71–73]. However, analysis of IL-17A protein after stimulation of MLN cell suspensions with YopJC172A, WT, and ΔYopB Y. pseudotuberculosis demonstrated that YopJ did not regulate IL-17A production from Vγ1.1/ 2- CD44hi CD27- γδ T cells (S4 Fig). Some of the observed differentially expressed genes are important in the activation status of T cells (e.g., Il2ra, Ctla4, and Cd69) and suggest that trans- located YopJ may limit the activation of adaptive Vγ4 T cells. There was also a notable impact (48.0%) on genes associated with cell proliferation, metabolism and energy, mitosis and cell cycle, RNA/DNA processing, and ER/Golgi processing (Figs S3A and S3B and 4E), suggesting that YopJ influences the adaptive γδ T cell transcriptional profile more broadly than just tar- geting the IFNγ pathway. Genes that were differentially expressed upon WT Y. pseudotubercu- losis stimulation or among Yop+ cells also suggest that many of the processes associated with immune responses and cellular activity were regulated by YopJ (S3C–S3E Fig). Collectively, YopJ appears to regulate the expression of many genes associated with T cell function in Vγ1.1/2- CD44hi CD27- γδ T cells, suggesting that adaptive Vγ4 T cells are broadly constrained in their immune functions by Y. pseudotuberculosis. YopJ inhibits the IL-12p40-mediated STAT4 pathway in adaptive γδ T cells To gain potential mechanistic insights into YopJ inhibition of IFNγ production and other Vγ1.1/2- CD44hi CD27- γδ T cell functions, a motif discovery algorithm designed for regula- tory element analysis was utilized to assess our RNA sequencing results [74]. Several transcrip- tion factor binding motifs related to IFNγ signaling were differentially expressed after YopJC172A Y. pseudotuberculosis but not WT Y. pseudotuberculosis stimulation including mem- bers of the E twenty-six (ETS)-domain family, Kru¨ppel-like factor and specificity protein (KLF/SP) transcription factor gene family, and the interferon regulatory factors (IRF) family PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 8 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function Fig 4. YopJ translocation leads to the inhibition of a broad anti-microbial gene response from Vγ4 T cells. (A and B) MLN suspensions from L. monocytogenes infected mice were stimulated with 10 MOI of WT or YopJC172A Y. pseudotuberculosis for 24 hours. Antibiotics were given 2 hours post- stimulation. Five hundred Vγ1.1/2- CD44hi CD27- γδ T cells from each stimulation were flow sorted and processed for RNA sequencing. (A) PCA plots are depicted for similarity of groups YopJC172A and WT Y. pseudotuberculosis stimulated Vγ1.1/2- CD44hi CD27- γδ T cells. (B) Heat maps are depicted for differentially expressed genes of YopJC172A or WT Y. pseudotuberculosis stimulated Vγ1.1/2- CD44hi CD27- γδ T cells. (C and D) MLN suspension from L. monocytogenes infected mice were stimulated with 1 MOI of WT Yptb-βla. Five hundred Yop+ or Yop- Vγ1.1/2- CD44hi CD27- γδ T cells were flow sorted and processed for RNA sequencing. (C) PCA plots are depicted for similarity of Yop+ or Yop- Vγ1.1/2- CD44hi CD27- γδ T cells. (D) Heat maps are depicted for differentially expressed genes of Yop- or Yop+ stimulated Vγ1.1/2- CD44hi CD27- γδ T cells. (E-G) A Venn diagram of differentially expressed genes (higher) that overlapped between RNA sequencing analyses favoring YopJC172A Y. pseudotuberculosis stimulation or Yop- cells is displayed. Shared genes were categorized by gene function. (F) The heat map highlights differentially expressed genes among Vγ1.1/2- CD44hi CD27- γδ T cells from the indicated stimulations and categories. (G) Homer motif analysis was performed on the RNA sequencing dataset. Motifs and associated genes to YopJC172A stimulated Vγ1.1/2- CD44hi CD27- γδ T cells are highlighted. Each experiment was performed with 3 biologic samples per group. Cutoffs for significant genes are p < 0.05 and FDR < 0.10. https://doi.org/10.1371/journal.ppat.1010103.g004 of transcription factors (S5A Fig). IRF8 protein was validated after WT, ΔYopB, and YopJC172A Y. pseudotuberculosis stimulation. Indeed, a higher percentage of Vγ4 T cells expressed IRF8 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 9 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function protein after stimulation with ΔYopB compared to WT Y. pseudotuberculosis stimulation (S5B Fig). Stimulation with YopJC172A Y. pseudotuberculosis was also able to partially restore IRF8 levels to those seen after ΔYopB Y. pseudotuberculosis stimulation (S5B Fig). IRF8 was also impacted by IL-12p40 blockade, which signals through signal transducer and activator of tran- scription 4 (STAT4) (S5B Fig). Interestingly, a number of transcription factor binding motifs downstream of STAT4 signaling were enriched including Etv5, Runx3, and Tead1 (Fig 4G) [75–78]. The RNAseq and homer motif analyses were also compared to an existing STAT4 ChIP-on-chip [79]. 7 genes identified from our main analyses (Figs 4F and 4G and S5B) were STAT4 target genes (S5C Fig). In summary, transcriptional profiling revealed a global subver- sion of anti-pathogen immune functions that may be associated with YopJ subversion of STAT4 activity. IL-12 signaling leads to STAT4 phosphorylation and formation of STAT4-STAT4 homodi- mers that re-localize to the nucleus where they directly bind to the Ifng promoter to induce IFNγ expression [79–81]. To determine whether YopJ inhibits IFNγ production by interfering with the STAT4 pathway, STAT4 protein and phosphorylation were assessed by flow cytome- try of MLN cells stimulated with Y. pseudotuberculosis. STAT4 phosphorylation was analyzed 6 hours after stimulation with WT, ΔYopB, or YopJC172A Y. pseudotuberculosis. Consistent with suppression of IFNγ production and the RNA-Seq analysis, WT Y. pseudotuberculosis sig- nificantly reduced the percentage of pSTAT4+ CD44hi CD27- γδ T cells as compared to ΔYopB and YopJC172A Y. pseudotuberculosis (Fig 5A). However, STAT4 protein levels were the same in all three infection conditions (WT, ΔYopB, and YopJC172A Y. pseudotuberculosis) (Fig 5B). Flow cytometry antibodies for STAT4 protein were validated by comparing STAT4 from WT and STAT4 KO splenocytes (S5D Fig). These data suggest that STAT4 phosphorylation but not protein is decreased upon YopJ translocation. STAT4 phosphorylation was also evaluated using the Yptb-βla reporter system described above. Among CD44hi CD27- γδ T cells, Yop- cells had a higher percentage of pSTAT4+ cells compared to Yop+ cells suggesting intrinsic Yop mediated inhibition of STAT4 phosphorylation levels (Fig 5C). Additionally, as STAT4 phosphorylation is downstream of IL-12 signaling, an anti-IL-12/23p40 subunit antibody (anti-p40) was used to determine whether IL-12 signals in the environment regulated STAT4 phosphorylation after Y. pseudotuberculosis stimulation. Indeed, IL-12/23p40 neutralization abrogated STAT4 phosphorylation levels regardless of Yop translocation (Fig 5C). As IL-12/ 23p40 was required to elicit IFNγ production from adaptive Vγ4 T cells in the culture condi- tions, we assessed whether YopJC172A Y. pseudotuberculosis stimulation modulated IL-12p70. The concentration of IL-12p70 was comparable between WT and YopJC172A Y. pseudotubercu- losis stimulated cultures (Fig 5D). Thus, changes in IL-12 were unlikely to contribute to adap- tive Vγ4 T cell subversion in vitro. To understand the role of YopJ and IL-12 on Vγ1.1/2- CD44hi CD27- γδ T cells in a more simplified system, purified γδ T cells were stimulated with YopJC172A Y. pseudotuberculosis in the presence of excessive IL-12p70. Adaptive Vγ4 T cells were unable to produce IFNγ in response to YopJC172A Y. pseudotuberculosis and IL-12p70 (S6A Fig). Finally, we assessed whether the addition of IL-12p70 could overcome the YopJ mediated inhibition of IFNγ production after WT Y. pseudotuberculosis stimulation. While a supraphysiologic level of IL-12p70 (50 ng/ml) was able to partially overcome YopJ mediated inhibition, lower levels of IL-12p70 addition (2 and 10 ng/ml) were unable to overcome YopJ mediated inhibition (S6B Fig). Importantly, these latter concentrations were orders of magni- tude higher than those detected in our culture conditions. Thus, IL-12 is not sufficient to induce IFNγ production from adaptive Vγ4 T cells. Collectively, these results suggest that Y. pseudotuberculosis stimulation elicits IL-12 production to promote adaptive Vγ4 T cell IFNγ responses, and that YopJ translocation into adaptive Vγ4 T cells inhibits IL-12 mediated STAT4 phosphorylation. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 10 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function Fig 5. YopJ inhibits IL-12p40 mediated STAT4 phosphorylation. (A) MLN cell suspensions from L. monocytogenes infected mice were stimulated with 10 MOI of WT, YopJC172A, or ΔYopB Y. pseudotuberculosis for 6 hours. Antibiotics were given 2 hours after stimulation. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for pSTAT4 after stimulation. Representative contour plots are displayed. The graph depicts mean ± SEM and represents at least two independent experiments with 2–4 mice per group. (B) The same experimental setup was used as in (A), but Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for STAT4 protein after stimulation. Representative plots for mean fluorescent intensity (MFI) are displayed. The graph depicts mean ± SEM and represents two independent experiments with 2 mice per group. (C) MLN suspensions from L. monocytogenes infected mice were either treated or untreated with IL-12/23p40 neutralizing antibody prior to stimulation with 1 MOI of WT Yptb-βla for 6 hours. Antibiotics were given 2 hours post-stimulation. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for pSTAT4 after stimulation. Representative contour plots are displayed. The graph depicts mean ± SEM and represents at least two independent experiments with 2–4 mice per group. (D) MLN cell suspensions from L. monocytogenes infected mice were stimulated with 10 MOI of WT or YopJC172A Y. pseudotuberculosis for 24 hours. Antibiotics were given 2 hours after stimulation. Supernatants were collected 24 hours post stimulation and IL-12p70 concentration was determined via ELISA. ��p < 0.0001, ��p < 0.001, ��p < 0.01, and �p < 0.05. A repeated measures one-way ANOVA was used for (A-C). Comparisons were performed to WT Y. pseudotuberculosis in (A) and as depicted in (B-D). https://doi.org/10.1371/journal.ppat.1010103.g005 Foodborne infection with YopJC172A Y. pseudotuberculosis induces IFNγ production in adaptive Vγ4 T cells To determine whether YopJ subverts adaptive γδ T cell function in vivo, foodborne infection with WT and YopJC172A Y. pseudotuberculosis was performed on naïve Balb/c mice. As YopJC172A Y. pseudotuberculosis is attenuated in vivo [82], a one log higher (2-4x108 CFU) infection dose of YopJC172A Y. pseudotuberculosis was administered to normalize the internal bacteria burdens in the MLN between infection groups (S7A Fig). While mice infected with WT and YopJC172A Y. pseudotuberculosis lost a similar amount of weight, mice infected with YopJC172A Y. pseudotuberculosis recovered slightly faster (Fig 6A). MLN were isolated 9 days after infection to evaluate adaptive γδ T cell function. Consistent with the ex vivo stimulation of L. monocytogenes-elicited Vγ4 T cells with Y. pseudotuberculosis, Vγ1.1/2- CD44hi CD27- γδ T cells from the MLN of YopJC172A Y. pseudotuberculosis infected mice displayed enhanced IFNγ production when stimulated ex vivo compared to their WT Y. pseudotuberculosis infected counterparts (Fig 6B and 6C). When the same infectious doses were used for both WT and YopJC172A Y. pseudotuberculosis (5x107 CFU), Vγ1.1/2- CD44hi CD27- γδ T cells from the MLN of YopJC172A Y. pseudotuberculosis infected mice also displayed enhanced IFNγ PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 11 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function Fig 6. Foodborne YopJC172A Y. pseudotuberculosis infection elicits an IFNγ response from Vγ4 T cells. (A-E) Balb/c mice were foodborne infected with WT (2-4x107 CFU) or YopJC172A Y. pseudotuberculosis (2-4x108 CFU). (A) Mouse weight was assessed daily after infection. (B and C) Nine days post infection, MLN were processed into single cell suspensions and stimulated with PMA/ionomycin in the presence of brefeldin A for 4 hours. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for IFNγ production. Representative histograms are displayed. Graphs represent mean ± SEM and are pooled from 3 experiments with a total of 5–8 mice per group. (D) Mouse survival was assessed daily after infection. anti-IL-12p40 antibody was administered at 0.2 mg/mouse on 0, 2, 4, and 6 days post infection. A Kaplan-Meier survival plot is shown. Study endpoint was 9 days post infection. The data represent 2 independent experiments with a total of 9 mice per group. (E and F) Balb/c mice were foodborne infected with 2x109 CFU L. monocytogenes to elicit a Vγ1.1/2- CD44hi CD27- γδ T cell population in vivo. 30 days post infection, immunized mice were foodborne infected with WT Yptb-βla (2-4x109 CFU) or left uninfected. (E) CFU of WT Yptb-βla were enumerated in MLN 3 days post WT Yptb-βla infection. (F) Three days post foodborne WT Yptb-βla infection, Vγ1.1/2- CD44hi CD27- γδ T cells from the MLN were analyzed for Yop translocation using the CCF4-AM assay. Representative contour plots are shown. Yop translocation (Yop+) among the indicated populations is depicted as mean ± SEM and is pooled from 2 experiments with a total of 4 mice per group. ��p < 0.001, ��p < 0.01, and �p < 0.05. A repeated measures one-way ANOVA was used for (A and F) and an unpaired t- test was used for (B, C, and E). Experimental groups were compared to WT Y. pseudotuberculosis in (A-C), uninfected controls in (E), and Vγ1.1/2- CD44hi CD27- γδ T cells in (F). A logrank test was used for survival curves in (D). https://doi.org/10.1371/journal.ppat.1010103.g006 production compared to their WT Y. pseudotuberculosis infected counterparts (S7B Fig). These experiments demonstrate that the increased IFNγ observed was not a result of a higher infectious dose nor of a higher bacteria burden at the time of analysis. As IL-12/23p40 was required for adaptive Vγ4 T cell IFNγ production in vitro (Fig 5), the impact of IL-12/23p40 was assessed in vivo. Naïve Balb/c mice were infected with WT or YopJC172A Y. pseudotubercu- losis and treated with an IL-12/23p40 neutralizing antibody or isotype control. All YopJC172A Y. pseudotuberculosis infected mice treated with anti-IL-12/23p40 succumbed by day 8 post PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 12 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function infection (Fig 6D). This data suggests that IL-12 promotes the protective capacity of catalyti- cally inactive YopJ. Similarly, IL-12 contributed to the protection of mice infected with WT Y. pseudotuberculosis. Serum was also collected on days 2, 4 and 6 after infection to assess circu- lating IL-12p70 levels. IL-12p70 was detectable 6 days after YopJC172A Y. pseudotuberculosis infection but was mostly below the limit of detection after WT Y. pseudotuberculosis infection (Fig 6E). Collectively, these data suggest that IL-12 is important in the protection of mice infected with WT or YopJC172A Y. pseudotuberculosis. Foodborne infection with WT Yptb-βla was performed to determine whether Y. pseudotu- berculosis could target adaptive Vγ4 T cells with Yop translocation in vivo. Balb/c mice were foodborne infected with L. monocytogenes to elicit a population of Vγ1.1/2- CD44hi CD27- γδ T cells as described previously [37,83]. After a return to homeostasis at 30 days post L. monocy- togenes infection, mice were foodborne infected with WT Yptb-βla. Y. pseudotuberculosis bur- den was assessed in the MLN 3 days post foodborne infection to determine whether WT Yptb- βla could establish a productive infection where Vγ4 T cells reside. Indeed, infected mice con- tained detectable Y. pseudotuberculosis in the MLN 3 days after foodborne infection (Fig 6F). Analysis of the translocation of Yop into Vγ1.1/2- CD44hi CD27- γδ T cells, myeloid cells (CD11b+), and CD4 and CD8 T cells was performed. Consistent with our in vitro observations, Vγ1.1/2- CD44hi CD27- γδ T cells and myeloid cells contained translocated Yop in vivo (Fig 6G). Y. pseudotuberculosis was relatively inefficient at Yop translocation into CD4 and CD8 T cells (Figs 6G and S7C). Collectively, these data show that foodborne YopJC172A Y. pseudotu- berculosis infection of naïve mice elicits IFNγ production in adaptive γδ T cells and that Yop can be translocated into adaptive Vγ4 T cells in vivo. Discussion In this study, we assessed the subversion of an adaptive subset of γδ T cells specialized in the promotion of pathogen resistance at the intestinal mucosa through the production of anti- infective cytokines like IFNγ and IL-17A [37]. While limited evidence suggests that Yop effec- tors directly target T cells to subvert their function, we identified that the Y. pseudotuberculosis effector molecule YopJ directly inhibits IFNγ production from adaptive CD44hi CD27- γδ T cells to subvert host immunity in mice. Additionally, we demonstrate that circulating human Vδ2+ T cells are similarly inhibited by direct translocation of YopJ, demonstrating that the direct targeting of γδ T cells by Y. pseudotuberculosis to inhibit IFNγ production is a conserved pathway of immune evasion in humans. Thus, Y. pseudotuberculosis Yop effectors translocated into murine Vγ4 T cells and human Vδ2+ T cells directly subvert their anti-microbial functions and host immunity by limiting IFNγ production. While Yersinia mediated inhibition of conventional T cells has been previously reported, studies have largely focused on the indirect subversion of T cells that is mediated by transloca- tion of Yop effectors into myeloid cells [55,82]. For example, YopJ/P appears to primarily sub- vert conventional T cell function through indirect mechanisms associated with inhibiting DC [22,26]. On the contrary, the study of γδ T cells in the context of Y. pseudotuberculosis infection has been primarily limited to the potential antigens that drive γδ T cell recognition of infection [84–89]. After phagocytosis of pathogens, activated DC migrate to lymph nodes and present antigen to T cells. APC-derived IL-12 further shapes T cell responses by providing a critical signal dur- ing T cell activation [90]. Interestingly, Y. pestis can limit both the migratory capacity of DC and the production of IL-12 [91], and Y. enterocolitica can induce programmed death of DC and inhibit antigen presentation [22]. Y. pseudotuberculosis YopJ can also indirectly limit NK cell function by interfering with DC TLR4 signaling pathways and YopP in Y. enterocolitica PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 13 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function can limit NK cell function through STAT4 inhibition [21,92]. Given that IL-12 signaling pro- motes STAT4 phosphorylation and IFNγ production in γδ T cells [29,93] and Yersinia spp. can inhibit DC, a potential extrinsic mechanism emerges for Y. pseudotuberculosis to inhibit γδ T cell responses by suppressing DC functions. In line with these observations, IL-12p40 was criti- cal for the induction of phospho-STAT4 in Vγ4 T cells after stimulation with Y. pseudotuber- culosis in in vitro cultures. In contrast, Y. pseudotuberculosis and IL-12 were unable to directly elicit IFNγ production from a highly enriched population of in vitro expanded γδ T cells sug- gesting that IL-12 is required but not sufficient for Vγ4 T cell IFNγ production. However, stimulation of MLN cell suspensions with ΔYopB or YopJC127A Y. pseudotuberculosis elicited STAT4 phosphorylation among Vγ4 T cells suggesting that translocation of Yop effectors and YopJ in particular subverts Vγ4 T cell function. Tracking Yop translocation in vitro revealed that Vγ4 T cells that contain Yop had reduced pSTAT4 levels and inhibited IFNγ production. Importantly, Vγ4 T cells that did not contain Yop effectors from the same cultures expressed higher pSTAT4 levels and comparable IFNγ production as Vγ4 T cells stimulated with ΔYopB-βla Yptb. Additionally, IL-12 levels were comparable between cultures stimulated with WT and YopJC172A Y. pseudotuberculosis. Collectively, these data suggest that functional impairment of Vγ4 T cells was mediated by direct translocation of YopJ into Vγ4 T cells and cell intrinsic mechanisms in vitro. The YopJ effector family has been increasingly described by their acetyltransferase function on serine, threonine, and lysine amino acid residues [19,94]. Serine and threonine are com- mon targets of phosphorylation to propagate signaling cascades or elicit functional conse- quences. For example, phosphorylation of STAT4 leads to dimerization and transport to the nucleus to promote transcription of STAT4 target genes. Acetylation of these target residues may inhibit phosphorylation and downstream signaling events [70]. Thus, a potential mecha- nism of YopJ subversion of Vγ4 T cells is through acetylation of STAT4 to inhibit the phos- phorylation or dimerization of STAT4. Other potential targets of YopJ acetyltransferase activity are the IL-12R and Janus kinases upstream of STAT4 activation. YopJ has also been reported to have cysteine protease, lysine acetyltransferase, ubiquitin-like protein protease, and deubiquitinase activity that may provide other potential avenues for YopJ to modulate the function of Vγ4 T cells through STAT4 [94–96]. An important aspect of Y. pseudotuberculosis pathogenesis unveiled by this work is the pref- erential targeting of a specialized subset of γδ T cells for delivery of inhibitory Yop effector molecules. Y. pseudotuberculosis injected Yop effectors into adaptive γδ T cells in a similar pro- portion as macrophages and DC and to a much greater extent than conventional CD4 or CD8 T cells. Of note, Y. pseudotuberculosis has been reported to translocate Yop effectors more effi- ciently into Treg cells than conventional CD4 T cells at high MOI to modulate their function [65]. The preferential targeting of Vγ4 T cells in this system is associated with expression of the β1-integrin by adaptive Vγ4 T cells. Additionally, the majority of adaptive Vγ4 T cells are anatomically segregated from conventional T cells in the paracortex by their localization in the interfollicular and medullary areas of the gut draining lymph nodes [38]. The distinct localiza- tion of adaptive Vγ4 T cells may facilitate interactions with Y. pseudotuberculosis in vivo. Loss of the adhesin YadA but not Inv abrogated YopJ mediated γδ T cell inhibition, suggesting that Y. pseudotuberculosis utilizes YadA to target adaptive γδ T cells for Yop translocation, consis- tent with previous studies suggesting that the adhesin Inv is largely dispensable for Y. enteroco- litica virulence [61,97]. Interestingly, this appears distinct from the targeting of conventional CD4 T cells that relies on Inv and is enhanced in the absence of YadA [11,65]. While our data demonstrates that STAT4 phosphorylation is inhibited by YopJ, our RNA- Seq analysis suggests that other targets may also be affected. One of YopJ’s known targets is the MAPK family of proteins that can have broad effects on cell proliferation, differentiation, PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 14 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function survival, and apoptosis [18]. The MAPK pathway in CD4 T cells and NK cells may also pro- mote STAT4 activity and downstream IFNγ mRNA stabilization, respectively [98,99]. Our data demonstrate a broad impact of YopJ on adaptive γδ T cell proliferation, metabolism, cell cycle, RNA/DNA processing, and ER/Golgi processing gene expression networks. These path- ways may be regulated by MAPK family member activity [100–105]. Homer motif analysis identified other potential means by which YopJ may regulate IFNγ production. For example, Ets-1 is a T-bet cofactor and necessary for Th1 IFNγ responses [106]. Increases in ETS-domain family of transcription factor motifs were associated with type 3 innate lymphoid cells (ILC3) but not Th17 cells [107], which may suggest that one of the potential mechanisms of YopJ mediated inhibition may target conserved pathways in unconventional lymphocyte popula- tions. Many NK cell receptors are also expressed on γδ T cells and may facilitate TCR indepen- dent effector functions [108–110]. In line with this, our profiling demonstrates that YopJ limits gene expression of Nkg7, which has recently been reported to promote cytotoxic granule release and inflammation during infection and cancer [72], and Prf1, which encodes the pore forming molecule perforin necessary to deliver lytic machinery into target cells [111]. Finally, interactions with Y. pseudotuberculosis YopJ led to the upregulation of Ulbp1, which encodes a stress-induced NKG2D ligand, and Idi1, which encodes an enzyme in the mevalonate pathway. As human γδ T cells respond to phospho-antigens derived from the non-mevalonate pathway in bacteria and mammalian mevalonate pathway in humans [112], this may limit the removal of translocated cells through NK or γδ T cell sensing mechanisms and suggests a broad mecha- nism to subvert human immunity. Thus, our findings suggest that adaptive Vγ4 T cells provide dynamic anti-infectious immunity that is subverted by direct translocation of YopJ. A number of studies have highlighted the importance of IFNγ production from conven- tional CD4 and CD8 T cells, NK cells, and ILC3 for Yersinia resistance [40,43,113,114], although the in vivo relevance of Yop inhibition of conventional T cells has not been addressed. Foodborne infection with YopJC172A Y. pseudotuberculosis led to an enhanced response from Vγ4 T cells, including increased IFNγ production, that was associated with a more rapid recovery of weight. As IL-12 was critical for Vγ4 T cell derived IFNγ production in vitro, the role of IL-12 after foodborne infection of Balb/c mice was assessed. Consistent with previous studies [22,40,115,116], IL-12 appeared critical for protection against foodborne WT and YopJC172A Y. pseudotuberculosis infection. As serum IL-12 was only readily detectable after YopJC172A Y. pseudotuberculosis infection, increased IL-12 may also contribute to the enhanced IFNγ response from Vγ4 T cells in vivo. Finally, assessment of cell populations tar- geted for Yop translocation in vivo was comparable to the results from the ex vivo MLN cul- tures. The highest percentage of Yop+ cells were among CD11b+ cells and Vγ4 T cells. Given the low MOI used in our ex vivo studies with the WT Yptb-βla reporter and the lack of inhibi- tion observed in Vγ4 T cells that lack Yop effectors from the same culture conditions, it is likely that intrinsic Yop effects on Vγ4 T cells is a mechanism of inhibiting Vγ4 T cell function in vivo. Thus, while Y. pseudotuberculosis can use γδ T cell intrinsic mechanisms to subvert the γδ T cell IFNγ response, multiple mechanisms may be available for Yersinia spp. to subvert γδ T cell functions to aid pathogenesis in vivo. As we previously demonstrated that foodborne but not i.v. infection led to adaptive Vγ4 T cell responses [37], our physiologic foodborne Y. pseudotuberculosis infection model revealed novel aspects of Yersinia pathogenesis and adap- tive Vγ4 T cell biology. In summary, the Y. pseudotuberculosis effector YopJ directly inhibits essential anti-effective functions of murine Vγ4 T cells and human Vδ2+ T cells. Y. pseudotuberculosis targeted Vγ4 T cells in a T3SS- and YadA-dependent process to deliver Yop effectors directly into Vγ4 T cells. Ex vivo whole tissue cultures revealed that direct inhibition of Vγ4 T cell function was the major mechanism of Vγ4 T cell subversion. YopJ translocation led to a dramatic reduction in PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 15 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function STAT4 phosphorylation levels and IFNγ production, which is important for protection from Yersinia. YopJ also inhibited a broad anti-infective gene signature. Thus, these findings add substantial insight into YopJ effector functions on murine and human γδ T cells and the patho- genesis of foodborne Y. pseudotuberculosis infection. Materials and methods Ethics statement All animal experiments were conducted in accordance with the Stony Brook University Insti- tutional Animal Care and Use Committee and National Institutes of Health guidelines. Blood collection from healthy human donors was approved by the Institutional Review Board at Stony Brook University. Mice Female 8–12 week old BALB/cJ mice were purchased from the Jackson Laboratory. Mice were euthanized by CO2 inhalation. Human studies Blood was sampled from a total of 6 adult healthy human donors of either gender between the ages of 20 and 40. Studies were designed so no randomization to experimental groups was nec- essary. Donors provided written informed consent. Bacteria Bacteria strains used in this study include: Y. pseudotuberculosis on the 32777 background WT strain, WT32777c, YopJC172A, YopHR409A, ΔYopB, YopER144A, YopTC139A, ΔYopM, ΔYpkA, ΔYopK, WT Yptb-βla, and ΔYopB Yptb-βla. Y. pseudotuberculosis on the IP2666 background WT strain, ΔYopB, ΔInv, ΔYadA, ΔInv ΔYadA, and ΔYopB ΔInv ΔYadA. See Table 1 for details. All strains were stored in 25% glycerol stocks at -80˚C. For stimulations, Y. pseudotu- berculosis strains were cultured overnight at 28˚C and 220 RPM in LB media. The following morning, Y. pseudotuberculosis was sub-cultured 1:10 in LB and 50 mM CaCl2 at 37˚C and 220 RPM for approximately 2 hours. Stimulation doses were based on the OD600. Foodborne L. monocytogenes immunization L. monocytogenes (EGDe strain) expressing a mutation in the internalin A gene (InlAM) was used for foodborne infection to facilitate epithelial cell invasion [117]. InlAM L. monocytogenes was cultured overnight at 37˚C and 220 RPM in BHI media. The following morning, L. mono- cytogenes was sub-cultured 1:10 in BHI at 37˚C and 220 RPM for approximately 2 hours. Infec- tion doses were based on the OD600. Mice were food and water deprived for 4 hours. Approximately 0.5 cm3 bread pieces were inoculated with 2x109 CFU L. monocytogenes in 50 μL. Mice were monitored to ensure the inoculated bread was consumed within 1 hour. Mice that did not fully consume bread were removed from the study. Bacterial infection doses were confirmed by plating inoculum on BHI. Foodborne Y. pseudotuberculosis infection Y. pseudotuberculosis strains (Table 1) were cultured overnight at 28˚C and 220 RPM in LB media. Infection doses were based on the OD600. Mice were food and water deprived for 4 hours. Approximately 0.5 cm3 bread pieces were inoculated with 2-4x107 CFU for WT32777 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 16 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function Y. pseudotuberculosis, 2-4x107–2-4x108 CFU for YopJC172A Y. pseudotuberculosis, or 2x109 CFU for WT Yptb-βla infection in 50 μL. Mice were monitored to ensure the inoculated bread was consumed within 1 hour. Mice that did not fully consume bread were removed from the study. Bacterial infection doses were confirmed by plating inoculum on LB. Single cell preparations, Y. pseudotuberculosis stimulations, and flow cytometry MLN from L. monocytogenes immunized mice were harvested 9 days after immunization and mechanically dissociated using a syringe plunger through a 70 μm cell strainer into a single cell suspension. Cells were resuspended in IMDM (Gibco) supplemented with 10% fetal bovine serum, 0.01 M HEPES, 100 μM non-essential amino acids (Gibco), 2 mM L-alanyl-L- glutamine dipeptide in 0.85% NaCl or 1x Glutamax (Gibco), and 1 mM sodium pyruvate. Cells were counted using a Vi-CELL Viability Analyzer (Beckman Coulter). Cells were stimu- lated in 96 well round-bottom tissue culture treated plates with various strains of Y. pseudotu- berculosis at 1 or 10 MOI (1 MOI for WT or ΔYopB Yptb-βla and 10 MOI for all other Y. pseudotuberculosis stimulations, unless otherwise indicated) at 37˚C/5% CO2. 100 U/mL of penicillin and 100 μg/mL of streptomycin were added to cells 2 hours post-stimulation. Cells were stimulated for a total of 24 hours or as indicated. BD GolgiPlug (BD Biosciences) was added 5 hours prior to the end of stimulation. If translocation was assessed, β-lactamase Load- ing Solutions kit (Invitrogen) was used to load CCF4-AM by incubating CCF4-AM at RT with cells for 1 hour in the dark. Cells were then processed for surface staining via incubation with live/dead stain, antibody, and Fc block (BioXcell) for 20 min in the dark at 4˚C. Antibodies used included antibodies specific to CD45, CD3, TCRδ, CD8, CD4, Vγ1.1, Vγ2, CD44, CD27, F4/80, CD11b, MHCII, CD11c, and CD29 (BioLegend). Cells were fixed, permeabilized, and stained with anti-IFNγ, anti-IRF8, or anti-STAT4 using BD Cytofix/Cytoperm kit (BD Biosci- ences) for intracellular cytokine staining. Functional γδ T cell analysis was done by stimulation with BD leukocyte activation cocktail (containing PMA, ionomycin, and brefeldin A; BD Pharmingen) for 4 hours prior to staining. Flow cytometry data were acquired using a BD LSRFortessa and analyzed by FlowJo software (BD Biosciences). Cell culture supernatant was analyzed for IL-12p70 using the BioLegend ELISA MAX Deluxe Set Mouse IL-12 (p70) kit per manufacturer instructions. Human γδ T cell response Blood was drawn and collected from healthy human donors in BD Vacutainer sodium heparin tubes (BD Biosciences). Blood was diluted 1:1 with 1x PBS at room temperature. Peripheral blood mononuclear cells (PBMC) were isolated from the buffy coat of Ficoll-paque PLUS gra- dient centrifugation (GE Healthcare) for 20 min at 1,400 × g without a brake. PBMC were washed with 1x PBS at room temperature and resuspended in IMDM supplemented with 10% fetal bovine serum, 0.01 M HEPES, 100 μM Non-essential amino acids (Gibco), 2 mM L-ala- nyl-L-glutamine dipeptide in 0.85% NaCl or 1x Glutamax (Gibco), and 1 mM sodium pyru- vate. Y. pseudotuberculosis strains were cultured overnight at 28˚C and 220 RPM in LB media the night prior. The following morning, Y. pseudotuberculosis was sub-cultured 1:10 in LB and 50 mM CaCl2 at 37˚C and 220 RPM for approximately 2 hours. Stimulation doses were based on the OD600. Cells were counted using a Vi-CELL Viability Analyzer (Beckman Coulter). Cells were stimulated in 96 round-bottom tissue culture treated plates with various strains of Y. pseudotuberculosis at 1 MOI at 37˚C/5% CO2. 1x penicillin and streptomycin (100 U/mL penicillin and 100 μg/mL streptomycin) were added to cells 2 hours post-stimulation. Cells were stimulated for a total of 24 hours or as indicated. BD GolgiPlug (BD Biosciences) was PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 17 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function added 5 hours prior to the end of stimulation. If translocation was assessed, β-lactamase Load- ing Solutions kit (Invitrogen) was used to load CCF4-AM by incubating CCF4-AM at RT with cells for 1 hour in the dark. Cells were then processed for surface staining via incubation with live/dead stain, antibody, and Fc block (BioXcell) for 20 min in the dark at 4˚C. Antibodies used included antibodies specific to Vδ2, CD3, TCRδ, (BioLegend). Cells were fixed, permea- bilized, and stained with anti-IFNγ using BD Cytofix/Cytoperm kit (BD Biosciences) for intra- cellular cytokine staining. Flow cytometry data were acquired using a BD LSRFortessa and analyzed by FlowJo software (BD Biosciences). Phospho-flow cytometry After surface staining for flow cytometry, cells were washed and stained for pSTAT4 using a methanol-based approach. Cells were fixed in 4% PFA/1.5% methanol for 30 minutes in the dark at 4˚C. Cells were then washed and incubated in methanol in the dark at 20˚C for 45 min- utes. After washing, cells were stained with anti-pSTAT4 (Y693)-PE (BD Biosciences) and washed once more. Flow cytometry data were acquired using a BD LSRFortessa and analyzed by FlowJo software (BD Biosciences). Enumeration of Y. pseudotuberculosis burden MLN were crushed and diluted in media prior to plating on LB agar. Total Y. pseudotuberculo- sis burden per organ was calculated. Sequencing and analysis Samples were prepared after Y. pseudotuberculosis stimulation as described above. Cell prepa- rations were stimulated with 10 MOI of WT or YopJC172A Y. pseudotuberculosis or 1 MOI of WT Yptb-βla for 24 hours. 500 Vγ1.1/2- CD44hi CD27- γδ T cells were flow sorted directly into a tube with NEBNext Cell Lysis Buffer and Murine RNase Inhibitor and processed for RNA sequencing using NEBNext Single Cell/Low Input RNA Library Prep Kit (Illumina). Sequenc- ing was performed at the Cold Spring Harbor Laboratory sequencing core on a NextSeq500. Fastq files were produced as an output of the sequencing files. Fastq were run through FastQC to perform quality control of transcripts prior to alignment. Fastq files were pair-ended aligned to GRCm38/mm10 by way of HISAT2 and output as .BAM files [118]. Raw counts of aligned transcripts were quantified with FeatureCounts [119]. Dimensionality reduction was per- formed with PCA analysis with the axes PC1 and PC2 in R-studio [120]. To determine differ- ential expression between samples, FeatureCounts raw count matrix was analyzed through DESeq2 with a parametric fitting normalized to the geometric mean of each individual gene across samples [121]. Cutoff values for significance and quality control were a p-value of <0.05 and FDR-value of <0.10, respectively. Significantly differentially expressed genes were visual- ized on a heatmap with a dendrogram that was clustered through average linkage. The distance measurement on the dendrogram used was through the Euclidean method. Overlapping expressions between gene differential expression sets were filtered with R-studio. Upregulated and downregulated genes from the differential expression analysis were separated with R-stu- dio and these Gene IDs were used for HOMER motif analysis [74]. Parameters of analysis of each gene used were 400bp preceding the initiation site and 100bp after the initiation site. The length of the motifs analyzed was set between 8 and 10. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 18 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function γδ T cell purification γδ T cells were expanded in vitro according to published protocols [122]. The MLN and spleen were isolated and processed into a single cell suspension 9 days post foodborne L. monocyto- genes infection of Balb/c mice. Red blood cells were lysed with red blood cell lysis buffer or ammonium chloride for 1 minute and cells from the MLN and spleen were combined. γδ T cells were enriched by negative selection using the following rat IgG primary antibodies from BioLegend: anti-CD4 (clone GK1.5), anti-CD8α (clone 53–6.7), anti-B220 (clone RA3-6B2), anti-MHC-II (clone M5/114.15.2), and anti-CD11b (clone M1/70). The MACS goat anti-rat IgG kit (Miltenyi Biotec) was used per manufacturer instructions with MACS LD columns and a QuadroMACS magnet. Enriched cells were cultured in 48-well plates coated overnight with 5 ug/ml anti-TCRδ (clone GL3). Cells were cultured in RPMI 1640 supplemented with 25 mM HEPES (Gibco), 1x glucose (Gibco), 10 g/ml folate (Sigma Aldrich), 1x sodium pyruvate (Gibco), 5x105 M 2 beta-mercaptoethanol (Sigma Aldrich), 1x Glutamax (Gibco), 1x penicil- lin-streptomycin (Gibco), and 10% FBS with 100 U/ml recombinant human IL-2. After 2 days of culture, cells were transferred into new wells with the same culture media to rest for 5 days. Cells were then stimulated with 10 MOI YopJC172A Y. pseudotuberculosis with 0.1, 1, or 10 ng/ ml of recombinant murine IL-12p70 (Peprotech). In vivo anti-IL-12p40 antibody treatment and serum collection On the day of foodborne WT or YopJC172A Y. pseudotuberculosis infection and on day 2, 4, and 6 after infection, mice were treated with 0.2 mg of anti-IL-12p40 (clone C17.8; BioLegend) by i.p. injection. Blood was collected via tail vein on day 2, 4, and 6 after infection. Blood was incubated at ambient temperature for 30 minutes before being spun down at 1500G for 10 minutes at 4˚C. Serum was isolated and analyzed for IL-12p70 with the BioLegend ELISA MAX Deluxe Set Mouse IL-12 (p70) kit. Ex vivo anti-IL-12p40 and recombinant IL-12p70 treatments At the start of Y. pseudotuberculosis stimulation of MLN cell suspensions, cultures were treated with 10 μg/ml anti-IL-12p40 (clone C17.8; BioLegend) for neutralization. In other conditions, recombinant murine IL-12p70 (Peprotech) was added at the start of Y. pseudotuberculosis stimulation of MLN cell suspensions at 2, 10, or 50 ng/ml. Statistical analysis GraphPad Prism 6 software (GraphPad Software Inc.) was used for statistical analysis. The dif- ferences between the means were compared using the statistical analysis described in the asso- ciated figure legends. All the data are presented as mean ± SEM and p < 0.05 was considered significant. �p < 0.05, ��p < 0.01, ��p < 0.001, ��p < 0.0001. Supporting information S1 Data. Excel spreadsheet containing the underlying numerical data for Figs 1A–1C, 2A and 2B, 3A, 5A–5D and 6A–6F. (XLSX) S1 Fig. Use of the Yptb-βla reporter to track Yop translocation. (A) MLN suspensions from L. monocytogenes infected mice were stimulated with WT or ΔYopB Y. pseudotuberculosis con- taining a β-lactamase translocation reporter (Yptb-βla) for 2 hours and given antibiotics. Cells were loaded with CCF4-AM dye to measure β-lactamase activity. FITC indicates CCF4-AM PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 19 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function loaded cells without translocation (Yop-) and BV421 indicates CCF4-AM loaded cells with Yop translocation (Yop+). Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for Yop transloca- tion 2 hours post stimulation at an MOI of 10. Representative contour plots are displayed. (B) MLN from L. monocytogenes infected mice were stimulated with Yptb-βla for 2 hours and given antibiotics. Yop translocation was detected as described above. The indicated cell popu- lations were analyzed for Yop translocation 2 hours after stimulation. Representative contour plots are displayed. (C) MLN from L. monocytogenes infected mice were stimulated with Yptb- βla for 2 hours and given antibiotics. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for Yop translocation 2 hours post stimulation at the indicated MOI and quantified for Yop transloca- tion. Data consists of one experiment with 2–10 mice/group and the graphs depict the mean ± SEM in (A-C). ��p < 0.0001, ��p < 0.001, and ��p < 0.01. A t-test was used for (A), and a repeated measures one-way ANOVA was used for (C). Comparisons were performed to ΔYopB Y. pseudotuberculosis in (A) and as depicted in (C). (TIF) S2 Fig. The majority of Vγ1.1/2- CD44hi CD27- γδ T cells and γδ T cells containing Yop express β1-integrin. (A) MLN from L. monocytogenes infected mice were isolated and pro- cessed into single cell suspensions. Vγ1.1/2- CD44hi CD27- γδ T cells, CD4 T cells, and CD8 T cells were analyzed for β1-integrin expression. (B) MLN suspensions from L. monocytogenes infected mice were loaded with CCF4-AM dye and stimulated with 10 MOI WT Y. pseudotu- berculosis containing a β-lactamase translocation reporter. CCF4-AM dye reports the occur- rence of β-lactamase activity and Yop translocation. γδ T cells that contain Yop (Yop+) or do not contain Yop (Yop-) were analyzed for β1-integrin expression 2 hours after stimulation. Representative histogram plots are displayed. Data is pooled from two experiments with a total of 7 mice/group and the graphs depict the mean ± SEM in (A-C). ��p < 0.0001 and ��p < 0.001. A repeated measures one-way ANOVA was used for (A) and a t-test was used for (B), and. Comparisons were done to adaptive γδ T cells in (A) and as depicted in (B). (TIF) S3 Fig. YopJ regulates the transcriptional profile of Vγ1.1/2- CD44hi CD27- γδ T cells. (A and B) MLN from L. monocytogenes infected mice were stimulated with 10 MOI of WT Y. pseudotuberculosis (WT) or mutant YopJ Y. pseudotuberculosis (YopJC172A) for 24 hours. Anti- biotics were given 2 hours post-stimulation. Five hundred Vγ1.1/2- CD44hi CD27- γδ T cells from each stimulation were flow sorted and processed for RNA sequencing. The heat map depicts upregulated genes in Vγ1.1/2- CD44hi CD27- γδ T cells after YopJC172A Y. pseudotuber- culosis stimulation and individual genes are listed. (C-E) Genes differentially expressed (down- regulated) that overlapped between RNA sequencing analyses as displayed in the Venn diagram in (C) to select for direct effects of YopJ on Vγ1.1/2- CD44hi CD27- γδ T cells. The heat map depicts downregulated genes in Vγ1.1/2- CD44hi CD27- γδ T cells from the analysis in (C). Individual genes are listed in (E). Each experiment was performed once with biologic replicates. The cutoff for gene significance was p < 0.05 and FDR < 0.10. (TIF) S4 Fig. YopJ does not inhibit IL-17A production in Vγ1.1/2- CD44hi CD27- γδ T cells. MLN cell suspensions from L. monocytogenes infected mice were stimulated with 10 MOI of WT, YopJC172A, or ΔYopB Y. pseudotuberculosis for 24 hours. Antibiotics were given 2 hours after stimulation and brefeldin A was added for the last 5–6 hours. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for IL-17A production after stimulation. Representative histograms are displayed. The graph depicts mean ± SEM and represents two independent experiments with PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 20 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function 4 mice per group. A repeated measures one-way ANOVA was used. � p < 0.05. (TIF) S5 Fig. YopJ impacts IFNγ related transcription factor motifs but not STAT4 protein. (A) Homer motif analysis was performed on the RNA sequencing results for the YopJC172A and WT Y. pseudotuberculosis comparison from Fig 5. The panel highlights the top transcription factor motifs of the ETS, SP/KLF, and IRF family of proteins identified in YopJC172A stimulated Vγ1.1/2- CD44hi CD27- γδ T cells. (B) MLN from L. monocytogenes infected mice were stimu- lated with 10 MOI of WT, YopJC172A, or ΔYopB Y. pseudotuberculosis for 6 hours. Antibiotics were given 2 hours post-stimulation. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for IRF8 levels with or without anti-IL12p40 neutralizing antibody. The graph depicts the percentage of IRF8 protein expression among Vγ1.1/2- CD44hi CD27- γδ T cells after WT, YopJC172A, or ΔYopB Y. pseudotuberculosis stimulation. Data depict two pooled experiments with a total of 8 mice/group and represents the mean ± SEM. (C) The genes from the RNAseq and Homer motif analysis in Figs 4F and 4G and S3B and S5A were compared to an existing STAT4 ChIP- on-chip dataset to identify common genes. Genes from our dataset that were represented in the top 1000 genes of the Chip-on-chip dataset are displayed. (D) STAT4 KO spleens are shown in maroon and WT spleens are shown in black in representative histograms. The graph depicts the MFI of STAT4 protein expression in bulk γδ T cells. Data depicts one experiment with 4 mice/group and represents the mean ± SEM. ��p < 0.0001, ��p < 0.001, ��p < 0.01, �p < 0.05. A repeated measures one-way ANOVA was used for (B), and a t-test was used for (D). Comparisons were performed as depicted in (B) and to Naïve WT in (D). (TIF) S6 Fig. IL-12 is insufficient to induce IFNγ and does not readily overcome the inhibition of YopJ. (A) γδ T cells enriched from the MLN and spleen of L. monocytogenes infected mice were expanded with plate bound γδTCR antibody for 2 days and rested for 5 days. After expansion, ~ 50% of cells were γδ T cells, and the majority of those were Vγ4 T cells. The enrichment summary reflects the mean enrichment from 4 samples. Afterwards, γδ T cells were isolated from cultures and stimulated with YopJC172A Y. pseudotuberculosis with 0.1, 1, or 10 ng/ml IL-12p70 for 24 hours. Antibiotics were added 2 hours after stimulation and brefel- din A was added for the last 5–6 hours. Histograms display IFNγ production from Vγ1.1/2- CD44hi CD27- γδ T cells under different culture conditions. Data depicts one experiment with 4 mice pooled and split into the indicated stimulation conditions. (B) MLN cell suspensions from L. monocytogenes infected mice were stimulated with 10 MOI of WT Y. pseudotuberculo- sis in the presence of 2, 10, or 50 ng/ml IL-12p70 or 10 MOI of YopJC172A Y. pseudotuberculosis for 24 hours. Antibiotics were given 2 hours after stimulation and brefeldin A was added for the last 5–6 hours. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for IFNγ production. Rep- resentative histograms of IFNγ production from Vγ1.1/2- CD44hi CD27- γδ T cells are dis- played. The graph depicts mean ± SEM from one experiment with 4 mice per group �p < 0.05. A repeated measures one-way ANOVA was used for comparisons to YopJC172A Y. pseudotu- berculosis in (B). (TIF) S7 Fig. The impact of foodborne infection of mice with Y. pseudotuberculosis. (A) Balb/c mice were foodborne infected with the indicated doses of WT or mutant YopJC172A Y. pseudo- tuberculosis and tissues were analyzed 9 days post-infection. Bacteria burden was quantified from the MLN. Data reflect 3–5 mice per group pooled from 3 independent experiments and the graphs depict the mean ± SEM. (B) Balb/c mice were foodborne infected with the indicated doses of WT or mutant YopJC172A Y. pseudotuberculosis. Nine days post infection, MLN PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 21 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function suspensions from WT or YopJC172A Y. pseudotuberculosis infected mice were stimulated with PMA/ionomycin and brefeldin A for 4 hours. Vγ1.1/2- CD44hi CD27- γδ T cells were analyzed for IFNγ production. Representative histograms are displayed and quantified. Data depicts one experiment with 3 mice per group. (C) Balb/c mice were foodborne infected with WT (2- 4x107 CFU) or YopJC172A Y. pseudotuberculosis (2-4x108 CFU) and treated with 0.2 mg/mouse of anti-IL12p40 on days 0, 2, 4, and 6 post infection. IL-12p70 concentrations were determined from serum at days 2, 4, and 6 post infection. Data represent 2 independent experiments with a total of 9 mice per group. Serum samples were pooled into groups of 3 per experimental con- dition. (D) Balb/c mice were foodborne infected with 2x109 CFU L. monocytogenes to elicit a Vγ1.1/2- CD44hi CD27- γδ T cell population in vivo. 30 days post infection, adaptive Vγ1.1/2- CD44hi CD27- γδ T cells from the MLN of immune mice were analyzed for Yop translocation using the CCF4-AM assay. Representative contour plots are shown. Yop translocation (Yop+) among the indicated populations represents background staining as a negative control for Fig 6F. The graph depicts the mean ± SEM and is pooled from 2 experiments with a total of 4 mice per group. ��p < 0.0001, �p < 0.05. A one-way ANOVA was used for (A), and an unpaired t-test was used for (B). Comparisons were performed to uninfected in (A) and to 5x107 WT Y. pseudotuberculosis in (B). (TIF) Author Contributions Conceptualization: Timothy H. Chu, Camille Khairallah, James B. Bliska, Brian S. Sheridan. Data curation: Timothy H. Chu, Camille Khairallah, Jason Shieh. Formal analysis: Timothy H. Chu, Camille Khairallah, Jason Shieh. Funding acquisition: Brian S. Sheridan. Investigation: Timothy H. Chu, Camille Khairallah, Rhea Cho, Zhijuan Qiu. Methodology: Timothy H. Chu, Camille Khairallah, Yue Zhang, Onur Eskiocak, Semir Beyaz, Brian S. Sheridan. Project administration: Brian S. Sheridan. Resources: David G. Thanassi, Mark H. Kaplan, Semir Beyaz, Vincent W. Yang, James B. Bliska. Supervision: Brian S. Sheridan. Validation: Timothy H. Chu. Visualization: Timothy H. Chu, Brian S. Sheridan. Writing – original draft: Timothy H. Chu. Writing – review & editing: Timothy H. Chu, Camille Khairallah, David G. Thanassi, James B. Bliska, Brian S. Sheridan. References 1. Barnes PD, Bergman MA, Mecsas J, Isberg RR. Yersinia pseudotuberculosis disseminates directly from a replicating bacterial pool in the intestine. J Exp Med. 2006; 203(6):1591–601. Epub 2006/06/07. https://doi.org/10.1084/jem.20060905 PMID: 16754724; PubMed Central PMCID: PMC2118325. 2. Autenrieth IB, Kempf V, Sprinz T, Preger S, Schnell A. Defense mechanisms in Peyer’s patches and mesenteric lymph nodes against Yersinia enterocolitica involve integrins and cytokines. Infect Immun. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 22 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function 1996; 64(4):1357–68. Epub 1996/04/01. https://doi.org/10.1128/iai.64.4.1357-1368.1996 PMID: 8606101; PubMed Central PMCID: PMC173926. 3. Titball RW, Leary SE. Plague. Br Med Bull. 1998; 54(3):625–33. Epub 1999/05/18. https://doi.org/10. 1093/oxfordjournals.bmb.a011715 PMID: 10326289. 4. Hinnebusch BJ. Bubonic plague: a molecular genetic case history of the emergence of an infectious disease. J Mol Med (Berl). 1997; 75(9):645–52. Epub 1997/11/14. https://doi.org/10.1007/ s001090050148 PMID: 9351703. 5. 6. Titball RW, Williamson ED. Vaccination against bubonic and pneumonic plague. Vaccine. 2001; 19 (30):4175–84. Epub 2001/07/18. https://doi.org/10.1016/s0264-410x(01)00163-3 PMID: 11457543. Inglesby TV, Dennis DT, Henderson DA, Bartlett JG, Ascher MS, Eitzen E, et al. Plague as a biological weapon: medical and public health management. Working Group on Civilian Biodefense. JAMA. 2000; 283(17):2281–90. Epub 2000/05/12. https://doi.org/10.1001/jama.283.17.2281 PMID: 10807389. 7. Nakajima R, Brubaker RR. Association between virulence of Yersinia pestis and suppression of gamma interferon and tumor necrosis factor alpha. Infect Immun. 1993; 61(1):23–31. Epub 1993/01/01. https://doi.org/10.1128/iai.61.1.23-31.1993 PMID: 8418045; PubMed Central PMCID: PMC302683. 8. Cornelis GR, Boland A, Boyd AP, Geuijen C, Iriarte M, Neyt C, et al. The virulence plasmid of Yersinia, an antihost genome. Microbiol Mol Biol Rev. 1998; 62(4):1315–52. Epub 1998/12/05. https://doi.org/ 10.1128/MMBR.62.4.1315-1352.1998 PMID: 9841674; PubMed Central PMCID: PMC98948. 9. Cornelis GR, Wolf-Watz H. The Yersinia Yop virulon: a bacterial system for subverting eukaryotic cells. Mol Microbiol. 1997; 23(5):861–7. Epub 1997/03/01. https://doi.org/10.1046/j.1365-2958.1997. 2731623.x PMID: 9076724. 10. Koberle M, Klein-Gunther A, Schutz M, Fritz M, Berchtold S, Tolosa E, et al. Yersinia enterocolitica tar- gets cells of the innate and adaptive immune system by injection of Yops in a mouse infection model. PLoS Pathog. 2009; 5(8):e1000551. Epub 2009/08/15. https://doi.org/10.1371/journal.ppat.1000551 PMID: 19680448; PubMed Central PMCID: PMC2718809. 11. Maldonado-Arocho FJ, Green C, Fisher ML, Paczosa MK, Mecsas J. Adhesins and host serum factors drive Yop translocation by yersinia into professional phagocytes during animal infection. PLoS Pathog. 2013; 9(6):e1003415. Epub 2013/07/03. https://doi.org/10.1371/journal.ppat.1003415 PMID: 23818844; PubMed Central PMCID: PMC3688556. 12. Durand EA, Maldonado-Arocho FJ, Castillo C, Walsh RL, Mecsas J. The presence of professional phagocytes dictates the number of host cells targeted for Yop translocation during infection. Cell Microbiol. 2010; 12(8):1064–82. Epub 2010/02/13. https://doi.org/10.1111/j.1462-5822.2010.01451.x PMID: 20148898; PubMed Central PMCID: PMC2906667. 13. Viboud GI, Bliska JB. Yersinia outer proteins: role in modulation of host cell signaling responses and pathogenesis. Annu Rev Microbiol. 2005; 59:69–89. Epub 2005/04/26. https://doi.org/10.1146/ annurev.micro.59.030804.121320 PMID: 15847602. 14. Cornelis GR. Yersinia type III secretion: send in the effectors. J Cell Biol. 2002; 158(3):401–8. Epub 2002/ 08/07. https://doi.org/10.1083/jcb.200205077 PMID: 12163464; PubMed Central PMCID: PMC2173816. 15. Bliska JB. Yersinia inhibits host signaling by acetylating MAPK kinases. ACS Chem Biol. 2006; 1 (6):349–51. Epub 2006/12/14. https://doi.org/10.1021/cb600261k PMID: 17163770. 16. Sweet CR, Conlon J, Golenbock DT, Goguen J, Silverman N. YopJ targets TRAF proteins to inhibit TLR-mediated NF-kappaB, MAPK and IRF3 signal transduction. Cell Microbiol. 2007; 9(11):2700–15. Epub 2007/07/05. https://doi.org/10.1111/j.1462-5822.2007.00990.x PMID: 17608743. 17. Cao Y, Guan K, He X, Wei C, Zheng Z, Zhang Y, et al. Yersinia YopJ negatively regulates IRF3-medi- ated antibacterial response through disruption of STING-mediated cytosolic DNA signaling. Biochim Biophys Acta. 2016; 1863(12):3148–59. Epub 2016/11/05. https://doi.org/10.1016/j.bbamcr.2016.10. 004 PMID: 27742471. 18. Mukherjee S, Orth K. In vitro signaling by MAPK and NFkappaB pathways inhibited by Yersinia YopJ. Methods Enzymol. 2008; 438:343–53. Epub 2008/04/17. https://doi.org/10.1016/S0076-6879(07) 38024-5 PMID: 18413260; PubMed Central PMCID: PMC3432501. 19. Ma KW, Ma W. YopJ Family Effectors Promote Bacterial Infection through a Unique Acetyltransferase Activity. Microbiol Mol Biol Rev. 2016; 80(4):1011–27. Epub 2016/10/28. https://doi.org/10.1128/ MMBR.00032-16 PMID: 27784797; PubMed Central PMCID: PMC5116873. 20. Orning P, Weng D, Starheim K, Ratner D, Best Z, Lee B, et al. Pathogen blockade of TAK1 triggers caspase-8-dependent cleavage of gasdermin D and cell death. Science. 2018; 362(6418):1064–9. Epub 2018/10/27. https://doi.org/10.1126/science.aau2818 PMID: 30361383; PubMed Central PMCID: PMC6522129. 21. Rosadini CV, Zanoni I, Odendall C, Green ER, Paczosa MK, Philip NH, et al. A Single Bacterial Immune Evasion Strategy Dismantles Both MyD88 and TRIF Signaling Pathways Downstream of PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 23 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function TLR4. Cell Host Microbe. 2015; 18(6):682–93. Epub 2015/12/15. https://doi.org/10.1016/j.chom.2015. 11.006 PMID: 26651944; PubMed Central PMCID: PMC4685476. 22. Trulzsch K, Geginat G, Sporleder T, Ruckdeschel K, Hoffmann R, Heesemann J, et al. Yersinia outer protein P inhibits CD8 T cell priming in the mouse infection model. J Immunol. 2005; 174(7):4244–51. Epub 2005/03/22. https://doi.org/10.4049/jimmunol.174.7.4244 PMID: 15778387. 23. Alonso A, Bottini N, Bruckner S, Rahmouni S, Williams S, Schoenberger SP, et al. Lck dephosphoryla- tion at Tyr-394 and inhibition of T cell antigen receptor signaling by Yersinia phosphatase YopH. J Biol Chem. 2004; 279(6):4922–8. Epub 2003/11/19. https://doi.org/10.1074/jbc.M308978200 PMID: 14623872. 24. Bruckner S, Rhamouni S, Tautz L, Denault JB, Alonso A, Becattini B, et al. Yersinia phosphatase induces mitochondrially dependent apoptosis of T cells. J Biol Chem. 2005; 280(11):10388–94. Epub 2005/01/06. https://doi.org/10.1074/jbc.M408829200 PMID: 15632192. 25. Yao T, Mecsas J, Healy JI, Falkow S, Chien Y. Suppression of T and B lymphocyte activation by a Yer- sinia pseudotuberculosis virulence factor, yopH. J Exp Med. 1999; 190(9):1343–50. Epub 1999/11/02. https://doi.org/10.1084/jem.190.9.1343 PMID: 10544205; PubMed Central PMCID: PMC2195683. 26. Gerke C, Falkow S, Chien YH. The adaptor molecules LAT and SLP-76 are specifically targeted by Yersinia to inhibit T cell activation. J Exp Med. 2005; 201(3):361–71. Epub 2005/02/09. https://doi.org/ 10.1084/jem.20041120 PMID: 15699071; PubMed Central PMCID: PMC2213036. 27. Viney J, MacDonald TT, Spencer J. Gamma/delta T cells in the gut epithelium. Gut. 1990; 31(8):841– 4. Epub 1990/08/01. https://doi.org/10.1136/gut.31.8.841 PMID: 2143741; PubMed Central PMCID: PMC1378605. 28. Goodman T, Lefrancois L. Expression of the gamma-delta T-cell receptor on intestinal CD8+ intrae- pithelial lymphocytes. Nature. 1988; 333(6176):855–8. Epub 1988/06/30. https://doi.org/10.1038/ 333855a0 PMID: 2968521. 29. Ribeiro ST, Ribot JC, Silva-Santos B. Five Layers of Receptor Signaling in gammadelta T-Cell Differ- entiation and Activation. Front Immunol. 2015; 6:15. Epub 2015/02/13. https://doi.org/10.3389/fimmu. 2015.00015 PMID: 25674089; PubMed Central PMCID: PMC4306313. 30. Yin Z, Zhang DH, Welte T, Bahtiyar G, Jung S, Liu L, et al. Dominance of IL-12 over IL-4 in gamma delta T cell differentiation leads to default production of IFN-gamma: failure to down-regulate IL-12 receptor beta 2-chain expression. J Immunol. 2000; 164(6):3056–64. Epub 2000/03/08. https://doi. org/10.4049/jimmunol.164.6.3056 PMID: 10706694. 31. Li W, Kubo S, Okuda A, Yamamoto H, Ueda H, Tanaka T, et al. Effect of IL-18 on expansion of gam- madelta T cells stimulated by zoledronate and IL-2. J Immunother. 2010; 33(3):287–96. Epub 2010/ 05/07. https://doi.org/10.1097/CJI.0b013e3181c80ffa PMID: 20445349. 32. Moens E, Brouwer M, Dimova T, Goldman M, Willems F, Vermijlen D. IL-23R and TCR signaling drives the generation of neonatal Vgamma9Vdelta2 T cells expressing high levels of cytotoxic media- tors and producing IFN-gamma and IL-17. J Leukoc Biol. 2011; 89(5):743–52. Epub 2011/02/19. https://doi.org/10.1189/jlb.0910501 PMID: 21330350. 33. Sutton CE, Lalor SJ, Sweeney CM, Brereton CF, Lavelle EC, Mills KH. Interleukin-1 and IL-23 induce innate IL-17 production from gammadelta T cells, amplifying Th17 responses and autoimmunity. Immunity. 2009; 31(2):331–41. Epub 2009/08/18. https://doi.org/10.1016/j.immuni.2009.08.001 PMID: 19682929. 34. Lockhart E, Green AM, Flynn JL. IL-17 production is dominated by gammadelta T cells rather than CD4 T cells during Mycobacterium tuberculosis infection. J Immunol. 2006; 177(7):4662–9. Epub 2006/09/20. https://doi.org/10.4049/jimmunol.177.7.4662 PMID: 16982905. 35. Ribot JC, Chaves-Ferreira M, d’Orey F, Wencker M, Goncalves-Sousa N, Decalf J, et al. Cutting edge: adaptive versus innate receptor signals selectively control the pool sizes of murine IFN-gamma- or IL- 17-producing gammadelta T cells upon infection. J Immunol. 2010; 185(11):6421–5. Epub 2010/11/03. https://doi.org/10.4049/jimmunol.1002283 PMID: 21037088; PubMed Central PMCID: PMC3915338. 36. Garman RD, Doherty PJ, Raulet DH. Diversity, rearrangement, and expression of murine T cell gamma genes. Cell. 1986; 45(5):733–42. Epub 1986/06/06. https://doi.org/10.1016/0092-8674(86) 90787-7 PMID: 3486721. 37. Sheridan BS, Romagnoli PA, Pham QM, Fu HH, Alonzo F, 3rd, Schubert WD, et al. gammadelta T cells exhibit multifunctional and protective memory in intestinal tissues. Immunity. 2013; 39(1):184–95. Epub 2013/07/31. https://doi.org/10.1016/j.immuni.2013.06.015 PMID: 23890071; PubMed Central PMCID: PMC3749916. 38. Romagnoli PA, Sheridan BS, Pham QM, Lefrancois L, Khanna KM. IL-17A-producing resident mem- ory gammadelta T cells orchestrate the innate immune response to secondary oral Listeria monocyto- genes infection. Proc Natl Acad Sci U S A. 2016; 113(30):8502–7. Epub 2016/07/13. https://doi.org/ 10.1073/pnas.1600713113 PMID: 27402748; PubMed Central PMCID: PMC4968747. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 24 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function 39. Dillen CA, Pinsker BL, Marusina AI, Merleev AA, Farber ON, Liu H, et al. Clonally expanded gamma- delta T cells protect against Staphylococcus aureus skin reinfection. J Clin Invest. 2018; 128(3):1026– 42. Epub 2018/02/06. https://doi.org/10.1172/JCI96481 PMID: 29400698; PubMed Central PMCID: PMC5824877. 40. Bohn E, Autenrieth IB. IL-12 is essential for resistance against Yersinia enterocolitica by triggering IFN-gamma production in NK cells and CD4+ T cells. J Immunol. 1996; 156(4):1458–68. Epub 1996/ 02/15. PMID: 8568248. 41. Szaba FM, Kummer LW, Duso DK, Koroleva EP, Tumanov AV, Cooper AM, et al. TNFalpha and IFN- gamma but not perforin are critical for CD8 T cell-mediated protection against pulmonary Yersinia pes- tis infection. PLoS Pathog. 2014; 10(5):e1004142. Epub 2014/05/24. https://doi.org/10.1371/journal. ppat.1004142 PMID: 24854422; PubMed Central PMCID: PMC4031182. 42. Logsdon LK, Mecsas J. The proinflammatory response induced by wild-type Yersinia pseudotubercu- losis infection inhibits survival of yop mutants in the gastrointestinal tract and Peyer’s patches. Infect Immun. 2006; 74(3):1516–27. Epub 2006/02/24. https://doi.org/10.1128/IAI.74.3.1516-1527.2006 PMID: 16495522; PubMed Central PMCID: PMC1418670. 43. Seo GY, Shui JW, Takahashi D, Song C, Wang Q, Kim K, et al. LIGHT-HVEM Signaling in Innate Lym- phoid Cell Subsets Protects Against Enteric Bacterial Infection. Cell Host Microbe. 2018; 24(2):249– 60 e4. Epub 2018/08/10. https://doi.org/10.1016/j.chom.2018.07.008 PMID: 30092201; PubMed Cen- tral PMCID: PMC6132068. 44. 45. Imperato JN, Xu D, Romagnoli PA, Qiu Z, Perez P, Khairallah C, et al. Mucosal CD8 T Cell Responses Are Shaped by Batf3-DC After Foodborne Listeria monocytogenes Infection. Front Immunol. 2020; 11:575967. Epub 2020/10/13. https://doi.org/10.3389/fimmu.2020.575967 PMID: 33042159; PubMed Central PMCID: PMC7518468. Zhang Y, Murtha J, Roberts MA, Siegel RM, Bliska JB. Type III secretion decreases bacterial and host survival following phagocytosis of Yersinia pseudotuberculosis by macrophages. Infect Immun. 2008; 76(9):4299–310. Epub 2008/07/02. https://doi.org/10.1128/IAI.00183-08 PMID: 18591234; PubMed Central PMCID: PMC2519449. 46. Palmer LE, Hobbie S, Galan JE, Bliska JB. YopJ of Yersinia pseudotuberculosis is required for the inhibition of macrophage TNF-alpha production and downregulation of the MAP kinases p38 and JNK. Mol Microbiol. 1998; 27(5):953–65. Epub 1998/04/16. https://doi.org/10.1046/j.1365-2958.1998. 00740.x PMID: 9535085. 47. Simonet M, Falkow S. Invasin expression in Yersinia pseudotuberculosis. Infect Immun. 1992; 60 (10):4414–7. Epub 1992/10/01. https://doi.org/10.1128/iai.60.10.4414-4417.1992 PMID: 1398952; PubMed Central PMCID: PMC257481. 48. Ivanov MI, Stuckey JA, Schubert HL, Saper MA, Bliska JB. Two substrate-targeting sites in the Yersinia protein tyrosine phosphatase co-operate to promote bacterial virulence. Mol Microbiol. 2005; 55(5):1346–56. Epub 2005/02/22. https://doi.org/10.1111/j.1365-2958.2005.04477.x PMID: 15720545. 49. Bliska JB, Guan KL, Dixon JE, Falkow S. Tyrosine phosphate hydrolysis of host proteins by an essen- tial Yersinia virulence determinant. Proc Natl Acad Sci U S A. 1991; 88(4):1187–91. Epub 1991/02/15. https://doi.org/10.1073/pnas.88.4.1187 PMID: 1705028; PubMed Central PMCID: PMC50982. 50. McPhee JB, Mena P, Bliska JB. Delineation of regions of the Yersinia YopM protein required for inter- action with the RSK1 and PRK2 host kinases and their requirement for interleukin-10 production and virulence. Infect Immun. 2010; 78(8):3529–39. Epub 2010/06/03. https://doi.org/10.1128/IAI.00269-10 PMID: 20515922; PubMed Central PMCID: PMC2916259. 51. Schoberle TJ, Chung LK, McPhee JB, Bogin B, Bliska JB. Uncovering an Important Role for YopJ in the Inhibition of Caspase-1 in Activated Macrophages and Promoting Yersinia pseudotuberculosis Vir- ulence. Infect Immun. 2016; 84(4):1062–72. Epub 2016/01/27. https://doi.org/10.1128/IAI.00843-15 PMID: 26810037; PubMed Central PMCID: PMC4807483. 52. Chung LK, Philip NH, Schmidt VA, Koller A, Strowig T, Flavell RA, et al. IQGAP1 is important for acti- vation of caspase-1 in macrophages and is targeted by Yersinia pestis type III effector YopM. mBio. 2014; 5(4):e01402–14. Epub 2014/07/06. https://doi.org/10.1128/mBio.01402-14 PMID: 24987096; PubMed Central PMCID: PMC4161239. 53. Harmon DE, Davis AJ, Castillo C, Mecsas J. Identification and characterization of small-molecule inhibitors of Yop translocation in Yersinia pseudotuberculosis. Antimicrob Agents Chemother. 2010; 54(8):3241–54. Epub 2010/05/26. https://doi.org/10.1128/AAC.00364-10 PMID: 20498321; PubMed Central PMCID: PMC2916352. 54. Schweer J, Kulkarni D, Kochut A, Pezoldt J, Pisano F, Pils MC, et al. The cytotoxic necrotizing factor of Yersinia pseudotuberculosis (CNFY) enhances inflammation and Yop delivery during infection by activation of Rho GTPases. PLoS Pathog. 2013; 9(11):e1003746. Epub 2013/11/19. https://doi.org/ 10.1371/journal.ppat.1003746 PMID: 24244167; PubMed Central PMCID: PMC3820761. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 25 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function 55. 56. Zhang Y, Tam JW, Mena P, van der Velden AW, Bliska JB. CCR2+ Inflammatory Dendritic Cells and Translocation of Antigen by Type III Secretion Are Required for the Exceptionally Large CD8+ T Cell Response to the Protective YopE69-77 Epitope during Yersinia Infection. PLoS Pathog. 2015; 11(10): e1005167. Epub 2015/10/16. https://doi.org/10.1371/journal.ppat.1005167 PMID: 26468944; PubMed Central PMCID: PMC4607306. Zhang Y, Romanov G, Bliska JB. Type III secretion system-dependent translocation of ectopically expressed Yop effectors into macrophages by intracellular Yersinia pseudotuberculosis. Infect Immun. 2011; 79(11):4322–31. Epub 2011/08/17. https://doi.org/10.1128/IAI.05396-11 PMID: 21844228; PubMed Central PMCID: PMC3257923. 57. Kim J, Fukuto HS, Brown DA, Bliska JB, London E. Effects of host cell sterol composition upon inter- nalization of Yersinia pseudotuberculosis and clustered beta1 integrin. J Biol Chem. 2018; 293 (4):1466–79. Epub 2017/12/05. https://doi.org/10.1074/jbc.M117.811224 PMID: 29197826; PubMed Central PMCID: PMC5787820. 58. Bohn E, Sonnabend M, Klein K, Autenrieth IB. Bacterial adhesion and host cell factors leading to effec- tor protein injection by type III secretion system. Int J Med Microbiol. 2019; 309(5):344–50. Epub 2019/ 06/11. https://doi.org/10.1016/j.ijmm.2019.05.008 PMID: 31178419. 59. Isberg RR, Voorhis DL, Falkow S. Identification of invasin: a protein that allows enteric bacteria to pen- etrate cultured mammalian cells. Cell. 1987; 50(5):769–78. Epub 1987/08/28. https://doi.org/10.1016/ 0092-8674(87)90335-7 PMID: 3304658. 60. Schulze-Koops H, Burkhardt H, Heesemann J, von der Mark K, Emmrich F. Plasmid-encoded outer membrane protein YadA mediates specific binding of enteropathogenic yersiniae to various types of collagen. Infect Immun. 1992; 60(6):2153–9. Epub 1992/06/01. https://doi.org/10.1128/iai.60.6.2153- 2159.1992 PMID: 1587583; PubMed Central PMCID: PMC257137. 61. Deuschle E, Keller B, Siegfried A, Manncke B, Spaeth T, Koberle M, et al. Role of beta1 integrins and bacterial adhesins for Yop injection into leukocytes in Yersinia enterocolitica systemic mouse infection. Int J Med Microbiol. 2016; 306(2):77–88. Epub 2016/01/01. https://doi.org/10.1016/j.ijmm.2015.12. 001 PMID: 26718660. 62. Bliska JB, Copass MC, Falkow S. The Yersinia pseudotuberculosis adhesin YadA mediates intimate bac- terial attachment to and entry into HEp-2 cells. Infect Immun. 1993; 61(9):3914–21. Epub 1993/09/01. https://doi.org/10.1128/iai.61.9.3914-3921.1993 PMID: 7689542; PubMed Central PMCID: PMC281094. 63. Nguyen GT, McCabe AL, Fasciano AC, Mecsas J. Detection of Cells Translocated with Yersinia Yops in Infected Tissues Using beta-Lactamase Fusions. Methods Mol Biol. 2019; 2010:117–39. Epub 2019/06/10. https://doi.org/10.1007/978-1-4939-9541-7_9 PMID: 31177435; PubMed Central PMCID: PMC6733027. 64. Pasztoi M, Bonifacius A, Pezoldt J, Kulkarni D, Niemz J, Yang J, et al. Yersinia pseudotuberculosis supports Th17 differentiation and limits de novo regulatory T cell induction by directly interfering with T cell receptor signaling. Cell Mol Life Sci. 2017; 74(15):2839–50. Epub 2017/04/06. https://doi.org/10. 1007/s00018-017-2516-y PMID: 28378044; PubMed Central PMCID: PMC5491567. 65. Elfiky A, Bonifacius A, Pezoldt J, Pasztoi M, Chaoprasid P, Sadana P, et al. Yersinia Pseudotuberculo- sis Modulates Regulatory T Cell Stability via Injection of Yersinia Outer Proteins in a Type III Secretion System-Dependent Manner. Eur J Microbiol Immunol (Bp). 2018; 8(4):101–6. Epub 2019/02/06. https://doi.org/10.1556/1886.2018.00015 PMID: 30719325; PubMed Central PMCID: PMC6348704. 66. Clark MA, Hirst BH, Jepson MA. M-cell surface beta1 integrin expression and invasin-mediated target- ing of Yersinia pseudotuberculosis to mouse Peyer’s patch M cells. Infect Immun. 1998; 66(3):1237– 43. Epub 1998/03/06. https://doi.org/10.1128/IAI.66.3.1237-1243.1998 PMID: 9488419; PubMed Central PMCID: PMC108039. 67. El Tahir Y, Skurnik M. YadA, the multifaceted Yersinia adhesin. Int J Med Microbiol. 2001; 291(3):209– 18. Epub 2001/09/14. https://doi.org/10.1078/1438-4221-00119 PMID: 11554561. 68. 69. Tertti R, Skurnik M, Vartio T, Kuusela P. Adhesion protein YadA of Yersinia species mediates binding of bacteria to fibronectin. Infect Immun. 1992; 60(7):3021–4. Epub 1992/07/01. https://doi.org/10. 1128/iai.60.7.3021-3024.1992 PMID: 1612772; PubMed Central PMCID: PMC257272. Flugel A, Schulze-Koops H, Heesemann J, Kuhn K, Sorokin L, Burkhardt H, et al. Interaction of entero- pathogenic Yersinia enterocolitica with complex basement membranes and the extracellular matrix proteins collagen type IV, laminin-1 and -2, and nidogen/entactin. J Biol Chem. 1994; 269(47):29732– 8. Epub 1994/11/25. PMID: 7961965. 70. Mukherjee S, Keitany G, Li Y, Wang Y, Ball HL, Goldsmith EJ, et al. Yersinia YopJ acetylates and inhibits kinase activation by blocking phosphorylation. Science. 2006; 312(5777):1211–4. Epub 2006/ 05/27. https://doi.org/10.1126/science.1126867 PMID: 16728640. 71. Lee J, Aoki T, Thumkeo D, Siriwach R, Yao C, Narumiya S. T cell-intrinsic prostaglandin E2-EP2/EP4 signaling is critical in pathogenic TH17 cell-driven inflammation. J Allergy Clin Immunol. 2019; 143 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 26 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function (2):631–43. Epub 2018/06/24. https://doi.org/10.1016/j.jaci.2018.05.036 PMID: 29935220; PubMed Central PMCID: PMC6354914. 72. Ng SS, De Labastida Rivera F, Yan J, Corvino D, Das I, Zhang P, et al. The NK cell granule protein NKG7 regulates cytotoxic granule exocytosis and inflammation. Nat Immunol. 2020; 21(10):1205–18. Epub 2020/08/26. https://doi.org/10.1038/s41590-020-0758-6 PMID: 32839608. 73. Malarkannan S. NKG7 makes a better killer. Nat Immunol. 2020; 21(10):1139–40. Epub 2020/08/26. https://doi.org/10.1038/s41590-020-0767-5 PMID: 32839609. 74. Heinz S, Benner C, Spann N, Bertolino E, Lin YC, Laslo P, et al. Simple combinations of lineage-deter- mining transcription factors prime cis-regulatory elements required for macrophage and B cell identi- ties. Mol Cell. 2010; 38(4):576–89. Epub 2010/06/02. https://doi.org/10.1016/j.molcel.2010.05.004 PMID: 20513432; PubMed Central PMCID: PMC2898526. 75. Pham D, Sehra S, Sun X, Kaplan MH. The transcription factor Etv5 controls TH17 cell development and allergic airway inflammation. J Allergy Clin Immunol. 2014; 134(1):204–14. Epub 2014/02/04. https://doi.org/10.1016/j.jaci.2013.12.021 PMID: 24486067; PubMed Central PMCID: PMC4209254. 76. Yin L, Li W, Xu A, Shi H, Wang K, Yang H, et al. SH3BGRL2 inhibits growth and metastasis in clear cell renal cell carcinoma via activating hippo/TEAD1-Twist1 pathway. EBioMedicine. 2020; 51:102596. Epub 2020/01/09. https://doi.org/10.1016/j.ebiom.2019.12.005 PMID: 31911271; PubMed Central PMCID: PMC7000347. 77. Pham D, Vincentz JW, Firulli AB, Kaplan MH. Twist1 regulates Ifng expression in Th1 cells by interfer- ing with Runx3 function. J Immunol. 2012; 189(2):832–40. Epub 2012/06/12. https://doi.org/10.4049/ jimmunol.1200854 PMID: 22685315; PubMed Central PMCID: PMC3392532. 78. Wang Y, Godec J, Ben-Aissa K, Cui K, Zhao K, Pucsek AB, et al. The transcription factors T-bet and Runx are required for the ontogeny of pathogenic interferon-gamma-producing T helper 17 cells. Immunity. 2014; 40(3):355–66. Epub 2014/02/18. https://doi.org/10.1016/j.immuni.2014.01.002 PMID: 24530058; PubMed Central PMCID: PMC3965587. 79. Good SR, Thieu VT, Mathur AN, Yu Q, Stritesky GL, Yeh N, et al. Temporal induction pattern of STAT4 target genes defines potential for Th1 lineage-specific programming. J Immunol. 2009; 183 (6):3839–47. Epub 2009/08/28. https://doi.org/10.4049/jimmunol.0901411 PMID: 19710469; PubMed Central PMCID: PMC2748807. 80. Xu X, Sun YL, Hoey T. Cooperative DNA binding and sequence-selective recognition conferred by the STAT amino-terminal domain. Science. 1996; 273(5276):794–7. Epub 1996/08/09. https://doi.org/10. 1126/science.273.5276.794 PMID: 8670419. 81. Yamamoto K, Miura O, Hirosawa S, Miyasaka N. Binding sequence of STAT4: STAT4 complex recog- nizes the IFN-gamma activation site (GAS)-like sequence (T/A)TTCC(C/G)GGAA(T/A). Biochem Bio- phys Res Commun. 1997; 233(1):126–32. Epub 1997/04/07. https://doi.org/10.1006/bbrc.1997.6415 PMID: 9144409. 82. Zhang Y, Ting AT, Marcu KB, Bliska JB. Inhibition of MAPK and NF-kappa B pathways is necessary for rapid apoptosis in macrophages infected with Yersinia. J Immunol. 2005; 174(12):7939–49. Epub 2005/06/10. https://doi.org/10.4049/jimmunol.174.12.7939 PMID: 15944300. 83. Khairallah C, Bettke JA, Gorbatsevych O, Qiu Z, Zhang Y, Cho K, et al. A blend of broadly-reactive and pathogen-selected Vgamma4 Vdelta1 T cell receptors confer broad bacterial reactivity of resident memory gammadelta T cells. Mucosal Immunol. 2021. Epub 2021/09/01. https://doi.org/10.1038/ s41385-021-00447-x PMID: 34462572. 84. Young JL, Goodall JC, Beacock-Sharp H, Gaston JS. Human gamma delta T-cell recognition of Yersi- nia enterocolitica. Immunology. 1997; 91(4):503–10. Epub 1997/08/01. https://doi.org/10.1046/j.1365- 2567.1997.00289.x PMID: 9378487; PubMed Central PMCID: PMC1363868. 85. Hermann E, Ackermann B, Duchmann R, Meyer zum Buschenfelde KH. Synovial fluid MHC- unrestricted gamma delta-T lymphocytes contribute to antibacterial and anti-self cytotoxicity in the spondylarthropathies. Clin Exp Rheumatol. 1995; 13(2):187–91. Epub 1995/03/01. PMID: 7656464. 86. Huang D, Chen CY, Ali Z, Shao L, Shen L, Lockman HA, et al. Antigen-specific Vgamma2Vdelta2 T effector cells confer homeostatic protection against pneumonic plaque lesions. Proc Natl Acad Sci U S A. 2009; 106(18):7553–8. Epub 2009/04/23. https://doi.org/10.1073/pnas.0811250106 PMID: 19383786; PubMed Central PMCID: PMC2678605. 87. Puan KJ, Jin C, Wang H, Sarikonda G, Raker AM, Lee HK, et al. Preferential recognition of a microbial metabolite by human Vgamma2Vdelta2 T cells. Int Immunol. 2007; 19(5):657–73. Epub 2007/04/21. https://doi.org/10.1093/intimm/dxm031 PMID: 17446209. 88. Hermann E, Lohse AW, Mayet WJ, van der Zee R, Van Eden W, Probst P, et al. Stimulation of synovial fluid mononuclear cells with the human 65-kD heat shock protein or with live enterobacteria leads to preferential expansion of TCR-gamma delta+ lymphocytes. Clin Exp Immunol. 1992; 89(3):427–33. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 27 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function Epub 1992/09/01. https://doi.org/10.1111/j.1365-2249.1992.tb06975.x PMID: 1387595; PubMed Cen- tral PMCID: PMC1554482. 89. Deseke M, Prinz I. Ligand recognition by the gammadelta TCR and discrimination between homeosta- sis and stress conditions. Cell Mol Immunol. 2020; 17(9):914–24. Epub 2020/07/28. https://doi.org/10. 1038/s41423-020-0503-y PMID: 32709926. 90. Cui W, Joshi NS, Jiang A, Kaech SM. Effects of Signal 3 during CD8 T cell priming: Bystander produc- tion of IL-12 enhances effector T cell expansion but promotes terminal differentiation. Vaccine. 2009; 27(15):2177–87. Epub 2009/02/10. https://doi.org/10.1016/j.vaccine.2009.01.088 PMID: 19201385; PubMed Central PMCID: PMC2803112. 91. Robinson RT, Khader SA, Locksley RM, Lien E, Smiley ST, Cooper AM. Yersinia pestis evades TLR4- dependent induction of IL-12(p40)2 by dendritic cells and subsequent cell migration. J Immunol. 2008; 181(8):5560–7. Epub 2008/10/04. https://doi.org/10.4049/jimmunol.181.8.5560 PMID: 18832714; PubMed Central PMCID: PMC2640496. 92. Koch I, Dach K, Heesemann J, Hoffmann R. Yersinia enterocolitica inactivates NK cells. Int J Med Microbiol. 2013; 303(8):433–42. Epub 2013/07/03. https://doi.org/10.1016/j.ijmm.2013.05.004 PMID: 23810728. 93. Garcia VE, Jullien D, Song M, Uyemura K, Shuai K, Morita CT, et al. IL-15 enhances the response of human gamma delta T cells to nonpeptide [correction of nonpetide] microbial antigens. J Immunol. 1998; 160(9):4322–9. Epub 1998/05/09. PMID: 9574535. 94. Orth K. Function of the Yersinia effector YopJ. Curr Opin Microbiol. 2002; 5(1):38–43. Epub 2002/02/ 09. https://doi.org/10.1016/s1369-5274(02)00283-7 PMID: 11834367. 95. 96. Li C, Wang D, Lv X, Jing R, Bi B, Chen X, et al. Yersinia pestis acetyltransferase-mediated dual acety- lation at the serine and lysine residues enhances the auto-ubiquitination of ubiquitin ligase MARCH8 in human cells. Cell Cycle. 2017; 16(7):649–59. Epub 2017/01/20. https://doi.org/10.1080/15384101. 2017.1281481 PMID: 28103160; PubMed Central PMCID: PMC5397269. Zhou H, Monack DM, Kayagaki N, Wertz I, Yin J, Wolf B, et al. Yersinia virulence factor YopJ acts as a deubiquitinase to inhibit NF-kappa B activation. J Exp Med. 2005; 202(10):1327–32. Epub 2005/11/23. https://doi.org/10.1084/jem.20051194 PMID: 16301742; PubMed Central PMCID: PMC2212976. 97. Keller B, Muhlenkamp M, Deuschle E, Siegfried A, Mossner S, Schade J, et al. Yersinia enterocolitica exploits different pathways to accomplish adhesion and toxin injection into host cells. Cell Microbiol. 2015; 17(8):1179–204. Epub 2015/02/14. https://doi.org/10.1111/cmi.12429 PMID: 25678064. 98. Morinobu A, Gadina M, Strober W, Visconti R, Fornace A, Montagna C, et al. STAT4 serine phosphor- ylation is critical for IL-12-induced IFN-gamma production but not for cell proliferation. Proc Natl Acad Sci U S A. 2002; 99(19):12281–6. Epub 2002/09/06. https://doi.org/10.1073/pnas.182618999 PMID: 12213961; PubMed Central PMCID: PMC129436. 99. Mavropoulos A, Sully G, Cope AP, Clark AR. Stabilization of IFN-gamma mRNA by MAPK p38 in IL- 12- and IL-18-stimulated human NK cells. Blood. 2005; 105(1):282–8. Epub 2004/09/04. https://doi. org/10.1182/blood-2004-07-2782 PMID: 15345584. 100. Zhang W, Liu HT. MAPK signal pathways in the regulation of cell proliferation in mammalian cells. Cell Res. 2002; 12(1):9–18. Epub 2002/04/11. https://doi.org/10.1038/sj.cr.7290105 PMID: 11942415. 101. Gehart H, Kumpf S, Ittner A, Ricci R. MAPK signalling in cellular metabolism: stress or wellness? EMBO Rep. 2010; 11(11):834–40. Epub 2010/10/12. https://doi.org/10.1038/embor.2010.160 PMID: 20930846; PubMed Central PMCID: PMC2966959. 102. Wilkinson MG, Millar JB. Control of the eukaryotic cell cycle by MAP kinase signaling pathways. FASEB J. 2000; 14(14):2147–57. Epub 2000/10/29. https://doi.org/10.1096/fj.00-0102rev PMID: 11053235. 103. Sugiura R, Satoh R, Ishiwata S, Umeda N, Kita A. Role of RNA-Binding Proteins in MAPK Signal Transduction Pathway. J Signal Transduct. 2011; 2011:109746. Epub 2011/07/22. https://doi.org/10. 1155/2011/109746 PMID: 21776382; PubMed Central PMCID: PMC3135068. 104. Kopper F, Bierwirth C, Schon M, Kunze M, Elvers I, Kranz D, et al. Damage-induced DNA replication stalling relies on MAPK-activated protein kinase 2 activity. Proc Natl Acad Sci U S A. 2013; 110 (42):16856–61. Epub 2013/10/02. https://doi.org/10.1073/pnas.1304355110 PMID: 24082115; PubMed Central PMCID: PMC3801042. 105. Wei JH, Seemann J. Remodeling of the Golgi structure by ERK signaling. Commun Integr Biol. 2009; 2(1):35–6. Epub 2009/08/26. https://doi.org/10.4161/cib.2.1.7421 PMID: 19704864; PubMed Central PMCID: PMC2649298. 106. Grenningloh R, Kang BY, Ho IC. Ets-1, a functional cofactor of T-bet, is essential for Th1 inflammatory responses. J Exp Med. 2005; 201(4):615–26. Epub 2005/02/25. https://doi.org/10.1084/jem. 20041330 PMID: 15728239; PubMed Central PMCID: PMC2213045. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 28 / 29 PLOS PATHOGENSYersinia pseudotuberculosis subversion of γδ T cell function 107. Koues OI, Collins PL, Cella M, Robinette ML, Porter SI, Pyfrom SC, et al. Distinct Gene Regulatory Pathways for Human Innate versus Adaptive Lymphoid Cells. Cell. 2016; 165(5):1134–46. Epub 2016/ 05/10. https://doi.org/10.1016/j.cell.2016.04.014 PMID: 27156452; PubMed Central PMCID: PMC4874868. 108. Dandekar AA, O’Malley K, Perlman S. Important roles for gamma interferon and NKG2D in gamma- delta T-cell-induced demyelination in T-cell receptor beta-deficient mice infected with a coronavirus. J Virol. 2005; 79(15):9388–96. Epub 2005/07/15. https://doi.org/10.1128/JVI.79.15.9388-9396.2005 PMID: 16014902; PubMed Central PMCID: PMC1181615. 109. Nitahara A, Shimura H, Ito A, Tomiyama K, Ito M, Kawai K. NKG2D ligation without T cell receptor engagement triggers both cytotoxicity and cytokine production in dendritic epidermal T cells. J Invest Dermatol. 2006; 126(5):1052–8. Epub 2006/02/18. https://doi.org/10.1038/sj.jid.5700112 PMID: 16484989. 110. Girardi M, Oppenheim DE, Steele CR, Lewis JM, Glusac E, Filler R, et al. Regulation of cutaneous malignancy by gammadelta T cells. Science. 2001; 294(5542):605–9. Epub 2001/09/22. https://doi. org/10.1126/science.1063916 PMID: 11567106 111. Osinska I, Popko K, Demkow U. Perforin: an important player in immune response. Cent Eur J Immu- nol. 2014; 39(1):109–15. Epub 2014/01/01. https://doi.org/10.5114/ceji.2014.42135 PMID: 26155110; PubMed Central PMCID: PMC4439970. 112. Gogoi D, Chiplunkar SV. Targeting gamma delta T cells for cancer immunotherapy: bench to bedside. Indian J Med Res. 2013; 138(5):755–61. Epub 2014/01/18. PMID: 24434328; PubMed Central PMCID: PMC3928706. 113. Autenrieth IB, Beer M, Bohn E, Kaufmann SH, Heesemann J. Immune responses to Yersinia entero- colitica in susceptible BALB/c and resistant C57BL/6 mice: an essential role for gamma interferon. Infect Immun. 1994; 62(6):2590–9. Epub 1994/06/01. https://doi.org/10.1128/iai.62.6.2590-2599.1994 PMID: 8188382; PubMed Central PMCID: PMC186549. 114. Bohn E, Heesemann J, Ehlers S, Autenrieth IB. Early gamma interferon mRNA expression is associ- ated with resistance of mice against Yersinia enterocolitica. Infect Immun. 1994; 62(7):3027–32. Epub 1994/07/01. https://doi.org/10.1128/iai.62.7.3027-3032.1994 PMID: 8005693; PubMed Central PMCID: PMC302917. 115. Hein J, Sing A, Di Genaro MS, Autenrieth IB. Interleukin-12 and interleukin-18 are indispensable for protective immunity against enteropathogenic Yersinia. Microb Pathog. 2001; 31(4):195–9. Epub 2001/09/20. https://doi.org/10.1006/mpat.2001.0458 PMID: 11562172. 116. Bohn E, Schmitt E, Bielfeldt C, Noll A, Schulte R, Autenrieth IB. Ambiguous role of interleukin-12 in Yersinia enterocolitica infection in susceptible and resistant mouse strains. Infect Immun. 1998; 66 (5):2213–20. Epub 1998/05/09. https://doi.org/10.1128/IAI.66.5.2213-2220.1998 PMID: 9573110; PubMed Central PMCID: PMC108184. 117. Wollert T, Pasche B, Rochon M, Deppenmeier S, van den Heuvel J, Gruber AD, et al. Extending the host range of Listeria monocytogenes by rational protein design. Cell. 2007; 129(5):891–902. Epub 2007/06/02. https://doi.org/10.1016/j.cell.2007.03.049 PMID: 17540170. 118. Kim D, Langmead B, Salzberg SL. HISAT: a fast spliced aligner with low memory requirements. Nat Methods. 2015; 12(4):357–60. Epub 2015/03/10. https://doi.org/10.1038/nmeth.3317 PMID: 25751142; PubMed Central PMCID: PMC4655817. 119. Liao Y, Smyth GK, Shi W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics. 2014; 30(7):923–30. Epub 2013/11/15. https://doi.org/10. 1093/bioinformatics/btt656 PMID: 24227677. 120. Stacklies W, Redestig H, Scholz M, Walther D, Selbig J. pcaMethods—a bioconductor package pro- viding PCA methods for incomplete data. Bioinformatics. 2007; 23(9):1164–7. Epub 2007/03/09. https://doi.org/10.1093/bioinformatics/btm069 PMID: 17344241. 121. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014; 15(12):550. Epub 2014/12/18. https://doi.org/10.1186/s13059-014- 0550-8 PMID: 25516281; PubMed Central PMCID: PMC4302049. 122. Collins C, Shi C, Russell JQ, Fortner KA, Budd RC. Activation of gamma delta T cells by Borrelia burg- dorferi is indirect via a TLR- and caspase-dependent pathway. J Immunol. 2008; 181(4):2392–8. Epub 2008/08/08. https://doi.org/10.4049/jimmunol.181.4.2392 PMID: 18684928; PubMed Central PMCID: PMC2832482. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010103 December 6, 2021 29 / 29 PLOS PATHOGENS
10.1371_journal.pone.0285197	RESEARCH ARTICLE Weight loss efficiency and safety of tirzepatide: A Systematic review Fei LinID Xiaoling Jie1,2, Jing Chen1,2, Yan Li1,2 1,2, Bin YuID 3, Baodong Ling4, Guangyao Lv5, Huijun Shang6, Xia Zhao1,2, 1 Department of Pharmacy, The First Affiliated Hospital of Chengdu Medical College, Chengdu, China, 2 Clinical Medical College, Chengdu Medical College, Chengdu, China, 3 Department of Pharmacy, Mianyang Central Hospital, Mianyang, China, 4 School of Pharmacy, Chengdu Medical College, Chengdu, China, 5 Department of Pharmacy, Binzhou Medical University Hospital, Binzhou, China, 6 School of Pharmacy, Collaborative Innovation Center of Advanced Drug Delivery System and Biotech Drugs in Universities of Shandong, Key Laboratory of Molecular Pharmacology and Drug Evaluation, Ministry of Education, Yantai University, Yantai, China a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 medicine2134@163.com Abstract Objective OPEN ACCESS Citation: Lin F, Yu B, Ling B, Lv G, Shang H, Zhao X, et al. (2023) Weight loss efficiency and safety of tirzepatide: A Systematic review. PLoS ONE 18(5): e0285197. https://doi.org/10.1371/journal. pone.0285197 Editor: Inge Roggen, Universitair Kinderziekenhuis Koningin Fabiola: Hopital Universitaire des Enfants Reine Fabiola, BELGIUM Received: February 14, 2023 Accepted: April 18, 2023 Published: May 4, 2023 Copyright: © 2023 Lin et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Funding: The authors received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. Tirzeptide is a novel glucagon-like peptide-1 receptor (GLP-1) and glucose-dependent insu- linotropic polypeptide (GIP) drug, which shows good efficiency for weight loss. Therefore, we aim to investigate the efficacy and safety of tirzepatide for weight loss in type 2 diabetes mellitus (T2DM) and obesity patients in this meta-analysis study. Methods Cochrane Library, PubMed, Embase, Clinical Trials, and Web of Science were searched from inception to October 5, 2022. All randomized controlled trials (RCTs) were included. The odds ratio (OR) was calculated using fixed-effects or random-effects models by Review Manager 5.3 software. Results In total, ten studies (12 reports) involving 9,873 patients were identified. A significant loss body weight in the tirzepatide group versus the placebo by -9.81 kg (95% CI (-12.09, -7.52), GLP-1 RAs by -1.05 kg (95% CI (-1.48, -0.63), and insulin by -1.93 kg (95% CI (-2.81, -1.05), respectively. In sub-analysis, the body weight of patients was significantly reduced in three tirzepatide doses (5 mg, 10 mg, and 15 mg) when compared with those of the pla- cebo/GLP-1 RA/insulin. In terms of safety, the incidence of any adverse events and adverse events leading to study drug discontinuation was higher in the tirzepatide group, but the inci- dence of serious adverse events and hypoglycaemia was lower. Additionally, the gastroin- testinal adverse events (including diarrhea, nausea, vomiting and decreased appetite) of tirzepatide were higher than those of placebo/basal insulin, but similar to GLP-1 RAs. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 1 / 16 PLOS ONEWeight loss of tirzepatide Conclusion In conclusion, tirzeptide can significantly reduce the weight of T2DM and patient with obe- sity, and it is a potential therapeutic regimen for weight-loss, but we need to be vigilant about its gastrointestinal reaction. Introduction Obesity is a metabolic disease, which is related to a variety of chronic diseases in addition to affecting the quality of life [1]. Recent statistics indicate that overweight/obesity and its relent- less global rise, with the number of people with excess body weight reaching > 2 billion, approximately 30% of the world population [2]. Some researchers reckon that overweight and obesity are major risk factors for cardiovascular disease [3]. Thus, weight loss can reduce the incidence of cardiovascular events and all-cause mortality in cardiovascular patients [3, 4], and lessen the incidence of diabetes [5, 6]. Currently, a growing number of drugs are used for weight loss, such as glucagon-like peptide-1 receptor agonists (GLP-1 RAs). Liraglutide was the first GLP-1RAs to be approved by the U.S. Food and Drug Administration (FDA) and European Medicines Agency (EMA) for the treatment of obesity [7]. Additionally, more evi- dence supports the use of the GLP-1RAs semaglutide in people with obesity without type 2 dia- betes mellitus (T2DM) [8]. As time goes on, an increasing number of drugs have been developed for the treatment of T2DM or obesity. In recent years, Glucagon-like peptide-1 receptor (GLP-1) and glucose- dependent insulinotropic polypeptide (GIP) are known as incretins among the many hor- mones in the body that has attracted the attention of researchers, which can promote insulin release after meals, lowering blood sugar and making the body more sensitive to insulin [9, 10]. Moreover, it also contributed to weight loss by slowing gastric emptying. GLP-1RAs are now considered the choice of injectable therapy for many people with T2DM and obesity, with several members of the class having weight loss efficacy [11–13]. Building on that concept, the combined GIP and GLP-1 RAs have been proposed as a novel therapeutic option for T2DM and obesity. Tirzepatide (LY3298176, Mounjaro) is the first dual GIP and GLP-1 RAs for the treatment of T2DM, obesity, and nonalcoholic steatohepatitis [14]. It is a first-in-class GLP-1/GIP recep- tor agonists that FDA approved on May 13, 2022, to improve blood sugar control in adults with T2DM as an adjunct to diet and exercise [15]. Tirzepatide can lower the hemoglobin A1C level more than other medications to which it was compared [16, 17]. At the same time, there is growing evidence that tirzepatide plays a role in the weight loss of T2DM patients. Further- more, another study showed that tirzepatide did not increase the risk of major cardiovascular events in participants with T2DM versus controls [18]. Tirzepatide also supported substantial weight loss in a recent clinical trial, potentially supporting its use as an obesity treatment [19]. In this paper, we performed a comprehensive systematic review and meta-analysis of all currently available randomized controlled trials (RCTs) of tirzepatide in individuals with T2DM and obesity to evaluate weight loss and adverse events when they were treated with tirzepatide. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 2 / 16 PLOS ONEWeight loss of tirzepatide Methods Study search and selection To conduct our study, we systematically searched PubMed, EMBASE, Cochrane library, Web of Science, and Clinical Trails databases from their inception to October 5, 2022, in the English language. "Tirzepatide" [MeSH] OR "LY3298176" OR "Mounjaro" were among the search phrases used. According to the inclusion and exclusion criteria, two researchers independently read the title and abstract of the literature for preliminary screening and also read the full text of literature that potentially met the inclusion criteria. Any disagreement was discussed and decided by the third researcher. Studies were included for this meta-analysis if they met the following criteria: only RCT; adults of obesity patients with or without T2DM; tirzepatide is the intervention drug; compari- son is placebo or antidiabetic; and outcome of efficacy and safety. Authorship; year of publica- tion; randomization; intervention; and patient number; study design; study duration; study site; study population; therapy duration; body weight; and risk of AEs were extracted from all included studies. Outcome indicators and the risk of bias assessment The primary outcome indicators included body weight, glycosylated hemoglobin, type A1C (HbA1c) and the incidence of any AEs. The secondary outcome indicators included the inci- dence of SAEs, AE leading to study drug discontinuation, hypoglycemia, and other AEs. The Cochrane Collaboration bias assessment tool was used to evaluate the risk bias of the included studies by two researchers independently [20]. According to the tool the risk was categorized as “high risk”, “low risk”, or “unclear”. Review Manager 5.3 was used to carry out quality assessment and an investigation of publication bias. Statistical analysis Review Manager 5.3 was utilized to perform statistical analysis. The mean difference (MD) was used as the effect analysis statistic for continuous measurement data; Oddi ratio (OR) was used as the effect analysis statistic for dichotomous variables, and 95%CI was considered for each effect. Statistical heterogeneity between the results was analyzed by Chi-square (χ2) test, and the heterogeneity was quantitatively judged by I2. When I2 � 50% and P > 0.1, the fixed effect model was applied, and when I2 > 50% and P < 0.1, the random effect model was applied. Additionally, we also investigated the source of heterogeneity with a sensitivity analy- sis when I2 was higher than 50%. The meta-analysis level was set as 0.05. Results Searching results and study characteristics The initial 401 articles were searched, including Cochrane library (n = 38), PubMed (n = 74), Embase (n = 156), Clinical Trails (n = 25) and Web of Science (n = 108). The duplicate litera- ture (n = 140) was first removed with EndNote X8 software, then the rest literature was further read for screening, and finally, the 10 studies that conformed to the inclusion criteria were included. A total of 9873 T2DM patients were involved. All studies were published in English. In this study, three tirzepatide doses has been giving (5 mg, 10 mg and 15 mg, subcutaneous injection, once a week), and a comparator, including placebo (two study by Frı´as [21, 22], SURPASS-1 [23], SURPASS-5 [24], SURMOUNT-1 [19], study by Heise [25], basal insulin [10 U/day insulin degludec (SURPASS–3) [26], and 10 U/day insulin glargine (SURPASS-4) [27]], PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 3 / 16 PLOS ONEWeight loss of tirzepatide GLP-1 RAs [1mg semaglutide (SURPASS-2) [28], and 1.5mg dulaglutide [21], 0.75 mg dula- glutide (SURPASS J–mono) [29]]. The study by Frias [21] with five groups, including tirzepa- tide (5 mg, 10 mg, 15 mg), placebo, and dulaglutide (1.5 mg) groups. Another one study by Frı´as [22] had four groups, we just including tirzepatide (15−2 mg) and placebo. The 15−2 mg tirzepatide dose-escalation regimens were 2.5 mg weeks 0–3; 7.5 mg weeks 4–7; and 15 mg weeks 8–11. Meanwhile, the duration of intervention in 4 studies was 40 weeks, 4 studies was 52 weeks, one study was 26 weeks, and another duration was 12 weeks. Ten studies were pub- lished from 2018 to 2022. The literature screening process and results are shown in Fig 1. Table 1 shows the baseline characteristics of the selected studies. Quality assessment The results of the quality assessment of 10 studies are furnished in Fig 2. Five RCTs described the detailed randomization methods, allocation concealment, blinding of participants and per- sonnel, incomplete outcome data, and other biases. Three RCTs did not have detail randomi- zation methods and allocation concealment. Two RCTs are open-label and have a high bias risk for research. The risks of study design bias was shown in Fig 3. Efficacy analysis In this meta-analysis, the included 10 RCT studies displayed varying degrees of weight loss effi- cacy. Over all, meta-analysis showed a significant reduction in body weight in the tirzepatide group versus the placebo group by -9.81 kg (95% CI (-12.09, -7.52). There were three doses investigated compared to the placebo group were affected significantly reduced the body weight of patients [5 mg: MD = -7.52 kg, 95% CI (-10.86, -4.18), P < 0.0001; I2 = 94%; 10 mg: MD = -10.48 kg, 95% CI (-15.34, -5.62), P < 0.0001; I2 = 97%; 15 mg: MD = -10.91 kg, 95% CI (-14.81, -7.01), P < 0.00001; I2 = 96%] (Fig 4). The sensitivity analysis excluding the SUR- MOUNT-1 [19] trial showed that statistical heterogeneity decreased from 94% to 43%, 97% to 35%, and 96% to 78%, respectively. The body weight of patients was significantly reduced 1.05 kg (95% CI (-1.48, -0.63) when compared with GLP-1 RAs group. There were three doses investigated [5 mg: MD = -0.53, 95% CI (-1.10, -0.05), P = 0.07; I2 = 95%; 10 mg: MD = -0.97, 95% CI (-1.80, -0.1), P = 0.02; I2 = 97%; 15 mg: MD = -1.53, 95% CI (-2.61, -0.45), P = 0.005; I2 = 98%] (Fig 5). The sensitivity analysis removing SURPASS J-mono [30] trial showed that statistical heterogeneity decreased from 95% to 0%, 97% to 90%, and 98% to 90%, respectively. The body weight of patients was significantly decreased 1.93 kg (95% CI (-2.81, -1.05) when compared with insulin group. Three doses were tested [5 mg: MD = -1.09, 95% CI (-1.87, -0.30), P = 0.007; I2 = 98%; 10 mg: MD = -1.50, 95% CI (-2.26, -0.73), P = 0.0001; I2 = 98%; 15 mg: MD = -3.21, 95% CI (-5.64, -0.78), P = 0.01; I2 = 100%] significantly decreased the body weight of patients when compared with insulin group (Fig 6). Initially, the heterogeneities of three tirzepatide doses were observed to be high, but when we removed any one study, the het- erogeneities in both groups did not decrease remarkably. Consistently, compared with pla- cebo, GLP-1 RAs and insulin, more participants receiving any of the three tirzepatide doses had reductions in body weight of at least 5%, 10%, or 15% (Table 2). The changes of HbA1c of patients was also collected. When compared with placebo, tirze- patide can significantly reduce the HbA1c of patients [5 mg: MD = -1.55%, 95% CI (-1.72, -1.39), P < 0.00001; I2 = 85%; 10 mg: MD = -1.75%, 95% CI (-1.92, -1.58), P < 0.00001; I2 = 71%; 15 mg: MD = -1.87%, 95% CI (-2.03, -1.70), P < 0.00001; I2 = 86%]. The same results were found in comparing tirzepatide with GLP-1 RAs group and insulin group, the level of HbA1c of all patients was significantly reduced [GLP-1 RAs: 5 mg: MD = -0.51%, 95% CI PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 4 / 16 PLOS ONEWeight loss of tirzepatide Fig 1. Flow diagram of studies searched in this meta-analysis. https://doi.org/10.1371/journal.pone.0285197.g001 (-0.62–0.39), P < 0.00001; I2 = 96%; 10 mg: MD = -0.73%, 95% CI (-0.85, -0.62), P < 0.00001; I2 = 96%; 15 mg: MD = -0.89%, 95% CI (-1.00, -0.77), P < 0.00001; I2 = 97%; Insulin: 5 mg: MD = -0.71%, 95% CI (-0.80, -0.63), P < 0.00001; I2 = 81%; 10 mg: MD = -0.94%, 95% CI (-1.03, -0.85), P < 0.00001; I2 = 50%; 15 mg: MD = -1.10%, 95% CI (-1.18, -1.01), P < 0.00001; I2 = 30%]. Safety analysis For the safety, meta-analysis showed a significant difference the incidence of any adverse events between tirzepatide group and placebo group [OR = 1.59, 95% CI (1.29, 1.95), P < 0.00001, I2 = 53%], GLP-1 RAs group [OR = 1.15, 95% CI (1.00, 1.32), P = 0.05, I2 = 0%], and basal insulin [OR = 1.55, 95% CI (1.25, 1.91), P < 0.0001, I2 = 69%], respectively. In the sub-analysis, the inci- dence of any adverse events was lower in the tirzepatide group than in the placebo group and basal insulin. But when compared to GLP-1 RAs, there was no significant difference in the tirze- patide 5 mg [OR = 1.01, 95% CI (0.80, 1.28), P = 0.92] and 10mg [OR = 1.17, 95% CI (0.92, 1.48), P = 0.2] groups (Table 3). Additionally, there was a statistically significant difference in the incidence of adverse events leading to study drug discontinuation between tirzepatide and placebo, between tirzepatide (15mg) and GLP-1 RAs, and between tirzepatide (5mg) and basal insulin. However, no significant statistics were found between tirzepatide and GLP-1 RAs (5mg and 10 mg) or between tirzepatide (10mg and 15 mg) and basal insulin. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 5 / 16 PLOS ONETable 1. General baseline of included studies. Study, year published Frı´as JP, et al. 2018 Intervention Patient number Study duration Therapy duration 26–week Weight loss of tirzepatide Study population Study design Study site Male (%) Mean ±S.D. age 18–75 years with 2 type diabetes for at least 6 months, HbA1c (7.0–10.5), BMI 23–50 kg/m2. phase 2b 47 sites in 4 countries Between May, 2017 and March, 2018 Between November, 2017 and April, 2018 Between June, 2019 and Oct, 2020 5mg 10mg 15mg 1.5mg dulaglutide placebo Frı´as JP, et al. 2020 15mg Rosenstock J, et al. 2021 (SURPASS–1) Frı´as JP, et al. 2021 (SURPASS–2) Ludvik B, et al. 2021 (SURPASS–3) Prato SD, et al. 2021 (SURPASS–4) placebo 5mg 10mg 15mg placebo 5mg 10mg 15mg 1 mg semaglutide 5mg 10mg 15mg degludec 5mg 10mg 15mg 55 51 53 54 51 28 26 121 121 121 115 470 469 470 469 358 360 359 360 329 328 338 glargine 1000 12–week Type 2 diabetes for at least 6 months HbA1c 7.0–10.5, BMI 23–45 kg/m2. phase 2 40–week �18 years with type 2 diabetes. HbA1c phase 3 7.0–9.5, BMI�23 kg/m2, and stable weight (±5) during the previous 3 months 13 sites in United States 52 sites in 4 countries Between July, 2019 and February, 2021 40–week �18 years with type 2 diabetes, metformin� 1500 mg/d. HbA1c 7.0– 10.5, BMI �25 kg/m2, stable weight (±5) during the previous 3 months. open– label, phase 3 128 sites in 8 countries Between April, 2019 and Jan, 2021 52–week �18 years and type 2 diabetes, HbA1c 7.0–10.5, metformin alone or combination with an SGLT2 inhibitor for at least 3 months, BMI � 25 kg/m2, and stable weight (±5) during the previous3 months. open label, phase 3 122 sites in 13 countries Between Nov, 2018 and April, 2021 52–week �18 years with type 2 diabetes, HbA1c 7.5–10.5, three oral glucose–lowering medications either alone or in any combination, BMI � 25 kg/m2, and stable weight (�5) during the previous 3 months. open– label, phase 3 187 sites in 14 countries 34 (62) 30 (59) 22 (42) 24 (44) 29 (57) 23 (82.1) 12 (46.2) 56 (46) 72 (60) 63 (52) 56 (49) 205 (43.6) 238 (50.7) 214 (45.5) 225 (48.0) 200 (56) 195 (54) 194 (54) 213 (59) 198 (60) 209 (64) 203 (60) 636 (64) 57.9 ± 8.2 56.5 ± 9.9 56.0 ± 7.6 58.7 ± 7.8 56.6 ± 8.9 56.6 ± 9.21 56.0 ± 10.13 54.1 ± 11.9 55.8 ± 10.4 52.9 ± 12.3 53.6 ± 12.8 56.3 ±10.0 57.2 ±10.5 55.9 ±10.4 56.9 ±10.8 57.2 ± 10.1 57.4 ± 9.7 57.5 ± 10.2 57.5 ± 10.1 62.9 ± 8.6 63.7 ± 8.7 63.7 ± 8.6 63.8 ± 8.5 (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 6 / 16 PLOS ONETable 1. (Continued) Study, year published Intervention Patient number Study duration Between August, 2019 and January, 2021 Between May, 2019 and March, 2021 Therapy duration 40–week Study population adults with type 2 diabetes, HbA1c 7.0–10.5, BMI � 23 kg/m2, insulin glargine (>20 IU/d or >0.25 IU/kg/d) with or without metformin (�1500 mg/d). Study design phase 3 45 sites in in 8 countries 52–week �20 Years with type 2 diabetes, HbA1c phase 3 Japan 7.0–10.0, BMI �23 kg/m2, stable weight (±5) during 3 months preceding Between June 28, 2019, and April 8, 2021, 28-week 20–74 years, type 2 diabetes for at least 6 months, and were being treated with lifestyle advice and stable doses of metformin, with or without one additional stable dose of another oral anti-hyperglycaemic medicine phase 1 2 sites in Germany Between December 2019 and April 2022 72-week �18 years, BMI � 30 kg/m2 or BMI � 27 kg/m2or more and at least one weight-related complication. phase 3 119 sites in 9 countries Weight loss of tirzepatide Study site Male (%) Mean ±S.D. age 60 (10) 72 (61) 65 (54) 66 (55) 113 (71.1) 119 (75.3) 132 (82.5) 117 (73.6) 31 (69.0) 34 (77.0) 21 (75.0) 204 (32.3) 209 (32.9) 205 (32.5) 207 (32.2) 62 ± 10 60 ± 10 61 ± 10 60 ± 10 56.8 ± 10.1 56.2 ± 10.3 56.0 ± 10.7 57.5 ± 10.2 61.1 ± 7.1 63.7 ± 5.9 60.4 ± 7.6 45.6±12.7 44.7±12.4 44.9±12.3 44.4±12.5 Dahl D, et al. 2022 (SURPASS–5) Inagaki N, et al. 2022 (SURPASS J– mono) 5mg 10mg 15mg placebo 5mg 10mg 15mg dulaglutide Heise T, et al. 2022 15 mg Jastreboff, AM, et al. 2022 (SURMOUNT-1) Semaglutide 1 mg Placebo 5mg 10mg 15mg placebo 116 119 120 120 159 158 160 159 45 44 28 630 636 630 643 https://doi.org/10.1371/journal.pone.0285197.t001 In this study, there was no significant difference in the incidence of serious adverse events between tirzepatide and placebo, and basal insulin, but significant difference to GLP-1 RAs. Hypoglycemia is the major SAEs, hypoglycemia definition as blood glucose < 70 mg/dL. Across all trials, the hypoglycemia risk of tirzepatide did not differ compared with placebo and GLP-1 RAs, and was lower with tirzepatide compared with basal insulin. After consuming tirzepatide, most of the patients experienced diarrhea, nausea, vomiting, decreased appetite, constipation, injection site reactions, and nasopharyngitis. Compared with basal insulin and placebo, more frequent gastrointestinal adverse events occurred, including diarrhea, nausea, vomiting, decreased appetite, and constipation in all tirzepatide groups. When compared with the GLP-1 RAs, the tirzepatide group showed a similar risk of nausea, diarrhea, vomiting, and constipation. While tirzepatide 5 mg and 10 mg were also associated with a higher incidence of decreased appetite. Meanwhile, there were no statistically significant differences in the incidence of injection site reactions between tirzepatide and GLP-1 RAs, and basal insulin. When compared with placebo, the incidence of injection site reactions was lower in tirzepatide (5 mg and 10 mg), but no significant difference in tirzepatide (15 mg). Besides, there was no significant difference in the incidence of nasopharyngitis was noticed between tirzepatide and placebo, GLP-1 RAs, and basal insulin. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 7 / 16 PLOS ONEWeight loss of tirzepatide Fig 2. Graphs of risk of bias for studies. https://doi.org/10.1371/journal.pone.0285197.g002 Discussions Tirzepatide as the first dual GIP and GLP-1 RA drug, which shown effects on hypoglycemia, body weight and cardiovascular indicators in previous studies [31–33]. Its effect on body weight could make it useful as a weight loss drug. Thus, in this meta-analysis, a systematic review to assess the weight loss efficacy and safety of tirzepatide is conducted. Based on our Fig 3. Quality assessment for risk of bias for studies. https://doi.org/10.1371/journal.pone.0285197.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 8 / 16 PLOS ONEWeight loss of tirzepatide Fig 4. Effect of tirzepatide vs placebo on body weight. https://doi.org/10.1371/journal.pone.0285197.g004 findings, both doses of tirzepatide (5 mg, 10 mg and 15 mg) were more effective than other drugs in reducing body weight. With the increase of obese people, the drugs for obesity treatment has been on the rise, and most of the drugs currently used to treat obesity started out as treatments for diabetes [34]. Based on our findings, both doses of tirzepatide were more effective in reducing bodyweight compared with other drugs. In this study, we included 10 RCT (12 reports) and compared the weight loss effects of tirzepatide with placebo, insulin, and GLP-1 RAs. The results showed that tirzepatide could reduce the weight of T2DM and obese patients. This is the same as the previous study showed [32, 35]. At the same time, the network meta-analysis results of Alkhezi et al. also showed that compared with liragrutide and semaglutide, tirzepatide had better weight reduction in non-diabetes patients, and the safety results were similar without differ- ence [36]. Furthermore, some studies found that meaningful weight loss is achieve 5 ~ 10% [37]. In our study, tirzepatide were more effective than other drugs in reducing bodyweight of Fig 5. Effect of tirzepatide vs GLP-1 RAs (semaglutide and dulaglutide) on body weight. https://doi.org/10.1371/journal.pone.0285197.g005 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 9 / 16 PLOS ONEWeight loss of tirzepatide Fig 6. Effect of tirzepatide vs insulin (insulin degludec and insulin glargine) on body weight. https://doi.org/10.1371/journal.pone.0285197.g006 5%, 10%, or 15%. More importantly, compared with placebo, GLP-1 RAs and insulin, tirzepa- tide can significantly reduce their HbA1c level of patients, and the results were same with other studies [32, 38]. For the safety, a significant difference the incidence of any adverse events between tirzepa- tide group and placebo/GLP-1 RAs/basal insulin. This is contrary to the results of a study by Bhagavathula et al. the results shown that no significant difference the incidence of any adverse events [32]. In addition, no significant difference in the incidence of SAEs between tirzepatide and placebo, and basal insulin, but significant difference to GLP-1 RAs. Across all trials, the risk of hypoglycemia with tirzepatide did not differ compared to placebo and GLP-1 RAs, but was lower with tirzepatide than with basal insulin. In the same results has been found in the study by Karagiannis et al. that incidence of serious adverse events did not differ between any of the tirzepatide doses and any comparator [35]. Gastrointestinal adverse events were the most common adverse events in all groups. In this study, the incidence of gastrointestinal adverse events, including diarrhea, nausea, vomiting, and decreased appetite, and the incidence of diarrhea, nausea, vomiting, and constipation were similar when comparing tirzepatide with GLP-1 RAs. However, in comparison with pla- cebo or basal insulin, tirzepatide increased the odds of diarrhea, nausea, vomiting, decreased appetite, and constipation. The results were the same as this study [35]. The clinical trials of SURPASS reported on the gastrointestinal system, and nausea, diarrhea and vomiting were the most common AEs [17]. The results from other studies found that GLP-1 receptor agonists are often accompanied by nausea, emesis, and undesired anorexia. Importantly, the hypopha- gic and emetic effects of GLP-1 receptor agonists are caused by activation of central GLP-1 receptors [39]. Gastrointestinal side effects of high-dose GLP-1 RAs and co-agonists occurred in 30% ~ 70% of patients, mostly arising within the first 2 weeks of the first dose, being mild or moderate in severity, and transient [40]. A study found the incidence of gastrointestinal bleed- ing occurred most frequently 0 ~ 4 weeks after the first dose and was higher for the 15 mg tir- zepatide group [41]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 10 / 16 PLOS ONETable 2. Meta-analysis results for tirzepatide vs placebo, GLP-1 RAs (semaglutide and dulaglutide) and basal insulin (insulin degludec and insulin glargine) for weigh loss. Intervention Comparator No. of participants with outcome/participants analysed OR (95% CI) Tirzepatide arm Comparator arm I2 (%) P value Weight loss of tirzepatide �5% weight loss Tirzepatide 5 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 10 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 15 mg Placebo GLP-1 RAs Basal insulin �10% weight loss Tirzepatide 5 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 10 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 15 mg Placebo GLP-1 RAs Basal insulin �15% weight loss Tirzepatide 5 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 10 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 15 mg Placebo GLP-1 RAs Basal insulin https://doi.org/10.1371/journal.pone.0285197.t002 699/922 428/684 438/687 761/927 522/678 542/688 780/924 552/683 595/697 502/922 169/525 249/687 613/927 240/520 365/688 649/924 288/523 464/697 329/922 73/525 89/687 483/927 124/520 176/688 516/924 182/523 272/697 245/929 282/682 100/1360 245/929 282/682 100/1360 245/929 282/682 100/1360 214/929 118/523 25/1360 214/929 118/523 25/1360 214/929 118/523 25/1360 57/929 38/523 5/1360 57/929 38/523 5/1360 57/929 38/523 5/1360 11.93 [9.39, 15.15] 3.97 [0.97, 16.26] 22.66 [15.64, 32.84] 17.33 [13.43, 22.35] 9.43 [1.58, 56.18] 48.54 [27.20, 86.65] 21.41 [16.37, 28.00] 11.02 [1.66, 73.26] 75.38 [50.05, 113.54] 15.82 [3.29, 76.10] 1.64 [1.24, 2.16] 27.27 [17.59, 42.25] 33.97 [5.62, 205.33] 2.95 [2.25, 3.86] 54.19 [35.06, 83.75] 37.79 [7.68, 186.01] 4.28 [3.27, 5.61] 96.07 [62.15, 148.50] 9.95 [7.30, 13.55] 2.07 [1.37, 3.14] 43.49 [16.98, 111.43] 22.31 [16.25, 30.64] 3.99 [2.71, 5.88] 95.35 [37.61, 241.73] 27.14 [19.66, 37.47] 6.88 [4.72, 10.04] 174.26 [69.74, 435.43] 0 95 42 34 97 72 38 97 40 73 0 42 79 52 34 74 0 15 0 0 0 0 38 32 0 0 33 < 0.00001 0.06 < 0.00001 < 0.00001 0.01 < 0.00001 < 0.00001 0.01 < 0.00001 < 0.00001 0.0006 0.0004 < 0.00001 0.0001 < 0.00001 < 0.00001 < 0.00001 < 0.00001 < 0.00001 < 0.00001 0.0006 < 0.00001 < 0.00001 < 0.00001 < 0.00001 < 0.00001 < 0.00001 < 0.00001 In addition, all the included studies were RCT, and some studies have a high bias risk for research because they did not have detailed randomization methods and allocation conceal- ment or open-label. Of these, this may affect the positive outcomes. Moreover, this study could have a higher effect on weight loss due to participants of the SURMOUNT-1 study not having T2DM and the SURPASS J-Combo study with only Japanese populations. Sensitive analysis results showed that statistical heterogeneity was decreased after removing the SURMOUNT-1 study and the SURPASS J-Combo study, but the statistical effect did not change. There are limitations in the current study. First, as all the RCTs involved the pharmaceutical industry, the positive outcomes should be interpreted cautiously [42, 43]. Second, only one study focused on the obesity patients in all 10 RCTs (12 reports), and RCTs have generally car- ried out in selected populations of T2DM or obesity patients. More research is suggested in the available guidance. Fortunately, 20 additional trials [17, 44, 45] aiming to investigate the PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 11 / 16 PLOS ONETable 3. The results of safety in meta-analysis. Intervention Comparator Studies (n) Tirzepatide arm (n) Comparator arm (n) I2 (%) Effect Estimate P value Weight loss of tirzepatide Any Adverse events Tirzepatide 5 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 10 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 15 mg Placebo GLP-1 RAs Basal insulin Serious adverse events Tirzepatide 5 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 10 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 15 mg Placebo GLP-1 RAs Basal insulin 4 3 2 4 3 2 6 4 2 4 3 2 4 3 2 5 4 2 Adverse event leading to study drug discontinuation Tirzepatide 5 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 10 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 15 mg Placebo GLP-1 RAs Basal insulin Hypoglycemia (blood glucose <70 mg/dL) Tirzepatide 5 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 10 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 15 mg Placebo GLP-1 RAs Basal insulin Nausea Tirzepatide 5 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 10 mg Placebo GLP-1 RAs Basal insulin 4 2 2 4 2 2 6 3 2 3 2 2 3 2 2 5 4 2 4 3 2 4 3 2 718/922 478/684 451/687 722/927 493/678 489/688 780/997 553/728 522/697 55/922 59/684 77/687 62/927 53/678 74/688 45/969 55/728 67/697 43/922 33/525 62/687 64/927 43/520 65/688 74/997 54/568 75/697 81/292 7/525 142/687 88/291 6/520 156/688 92/367 17/728 178/697 195/922 112/684 80/687 260/927 132/678 134/688 647/929 475/682 872/1360 647/929 475/682 872/1360 682/983 518/726 872/1360 59/929 30/682 215/1360 59/929 30/682 215/1360 61/957 30/726 215/1360 25/929 25/523 59/1360 25/929 25/523 59/1360 29/983 24/567 59/1360 76/286 4/523 811/1360 76/286 4/523 811/1360 81/340 5/726 811/1360 74/929 112/682 29/1360 74/929 112/682 29/1360 0 0 0 62 0 69 71 0 74 0 37 71 0 0 0 0 0 77 0 0 63 0 72 89 48 0 87 44 0 94 51 25 72 24 0 86 0 35 0 0 77 58 1.55 [1.25, 1.91] 1.01 [0.80, 1.28] 1.23 [1.01, 1.50] 1.48 [0.98, 2.23] 1.17 [0.92, 1.48] 1.58 [1.09, 2.29] 2.06 [1.24, 3.43] 1.28 [1.01, 1.63] 1.91 [1.26, 2.89] 0.94 [0.64, 1.38] 2.08 [1.32, 3.28] 0.94 [0.51, 1.75] 1.06 [0.73, 1.54] 1.87 [1.17, 2.98] 0.84 [0.63, 1.13] 0.72 [0.49, 1.07] 1.91 [1.20, 3.01] 0.80 [0.39, 1.64] 1.77 [1.07, 2.92] 1.34 [0.78, 2.29] 3.09 [1.33, 7.18] 2.68 [1.68, 4.30] 1.23 [0.30, 5.09] 3.46 [0.69, 17.43] 2.57 [1.67, 3.96] 2.40 [1.46, 3.95] 4.00 [0.96, 16.79] 1.22 [0.76, 1.96] 1.76 [0.51, 6.13] 0.17 [0.06, 0.48] 1.42 [0.89, 2.27] 1.59 [0.44, 5.73] 0.22 [0.15, 0.34] 1.25 [0.81, 1.93] 3.29 [1.25, 8.69] 0.25 [0.14, 0.46] 3.15 [2.34, 4.17] 1.00 [0.65, 1.56] 6.38 [4.02, 10.11] 4.59 [3.46, 6.07] 1.31 [0.62, 2.75] < 0.0001 0.92 0.04 0.06 0.2 0.02 0.005 0.04 0.002 0.76 0.002 0.85 0.74 0.008 0.24 0.11 0.006 0.54 0.03 0.29 0.009 < 0.0001 0.78 0.13 < 0.0001 0.0005 0.06 0.4 0.37 0.0008 0.14 0.48 < 0.00001 0.3 0.02 < 0.00001 < 0.00001 0.99 < 0.00001 < 0.00001 0.48 11.02 [5.21, 23.31] < 0.00001 (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 12 / 16 PLOS ONETable 3. (Continued) Intervention Comparator Studies (n) Tirzepatide arm (n) Comparator arm (n) I2 (%) Effect Estimate Tirzepatide 15 mg Placebo GLP-1 RAs Basal insulin Diarrhea Tirzepatide 5 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 10 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 15 mg Placebo GLP-1 RAs Basal insulin Vomiting Tirzepatide 5 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 10 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 15 mg Placebo GLP-1 RAs Basal insulin Decreased appetite Tirzepatide 5 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 10 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 15 mg Placebo GLP-1 RAs Basal insulin Injection site reactions Tirzepatide 5 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 10 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 15 mg Placebo GLP-1 RAs Basal insulin Nasopharyngitis Tirzepatide 5 mg Placebo GLP-1 RAs Basal insulin 6 4 2 4 3 2 4 3 2 6 4 2 4 3 2 4 3 2 6 4 2 4 3 2 4 3 2 6 4 2 4 3 2 4 3 2 5 4 2 3 2 2 281/997 168/728 161/697 153/922 96/684 96/687 180/927 104/678 125/688 214/997 104/728 130/697 59/922 42/684 37/687 77/927 52/678 61/688 102/997 76/728 65/697 83/922 68/684 51/687 109/927 68/678 73/688 126/997 114/728 78/697 29/922 14/684 2/687 47/927 22/678 8/688 53/969 34/728 9/697 28/292 32/214 21/687 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 83/983 125/726 29/1360 70/929 74/682 58/1360 70/929 74/682 58/1360 78/983 87/726 58/1360 17/929 46/682 20/1360 17/929 46/682 20/1360 19/983 51/726 20/1360 25/929 35/682 7/1360 25/929 35/682 7/1360 32/983 66/726 7/1360 5/929 19/682 10/1360 5/929 19/682 10/1360 17/957 31/726 10/1360 39/289 32/213 87/1360 53 54 0 39 63 0 59 0 0 50 0 0 50 79 0 60 35 0 37 69 0 0 54 0 1 64 0 25 79 0 0 80 0 36 80 0 81 83 0 0 23 0 Weight loss of tirzepatide 4.19 [2.37, 7.39] 1.51 [0.94, 2.44] P value < 0.00001 0.09 14.34 [9.30, 22.10] < 0.00001 2.13 [1.30, 3.48] 1.34 [0.97, 1.85] 3.63 [2.53, 5.19] 2.59 [1.41, 4.76] 1.50 [1.09, 2.06] 5.20 [3.69, 7.32] 2.65 [1.58, 4.44] 1.23 [0.90, 1.67] 5.47 [3.90, 7.68] 3.67 [2.13, 6.33] 1.16 [0.25, 5.34] 3.94 [2.19, 7.09] 4.86 [2.86, 8.27] 1.25 [0.60, 2.59] 6.77 [3.91, 11.71] 5.76 [3.51, 9.45] 1.75 [0.66, 4.61] 0.003 0.08 < 0.00001 0.002 0.01 < 0.00001 0.0002 0.19 < 0.00001 < 0.00001 0.85 < 0.00001 < 0.00001 0.55 < 0.00001 < 0.00001 0.26 7.13 [4.15, 12.26] < 0.00001 3.60 [2.28, 5.68] 2.39 [1.15, 4.96] 15.83 [6.93, 36.18] 4.85 [3.11, 7.56] 2.58 [1.11, 5.99] 22.72 [9.99, 51.68] 4.27 [2.84, 6.43] 2.18 [0.86, 5.50] < 0.00001 0.02 < 0.00001 < 0.00001 0.03 < 0.00001 < 0.00001 0.1 23.59 [10.38, 53.61] < 0.00001 4.78 [1.83, 12.49] 0.78 [0.09, 6.65] 0.31 [0.07, 1.40] 6.18 [1.80, 21.17] 1.29 [0.20, 8.15] 1.13 [0.44, 2.94] 2.78 [0.45, 17.40] 0.68 [0.07, 6.36] 1.19 [0.46, 3.06] 0.69 [0.41, 1.16] 1.00 [0.58, 1.70] 0.47 [0.28, 0.77] 0.001 0.82 0.13 0.004 0.79 0.8 0.27 0.74 0.72 0.16 0.99 0.003 (Continued ) 13 / 16 PLOS ONEWeight loss of tirzepatide Table 3. (Continued) Intervention Comparator Studies (n) Tirzepatide arm (n) Comparator arm (n) I2 (%) Effect Estimate Tirzepatide 10 mg Placebo GLP-1 RAs Basal insulin Tirzepatide 15 mg Placebo GLP-1 RAs Basal insulin 3 2 2 4 3 2 18/291 27/209 30/688 33/339 32/258 31/697 39/289 32/213 87/1360 42/317 40/257 87/1360 0 31 0 0 0 0 0.42 [0.24, 0.76] 0.83 [0.48, 1.45] 0.69 [0.45, 1.07] 0.70 [0.43, 1.14] 0.76 [0.46, 1.26] 0.70 [0.45, 1.07] P value 0.004 0.52 0.1 0.15 0.29 0.1 GLP-1 Ras: semaglutide and dulaglutide; Basal insulin: insulin degludec and insulin glargine. https://doi.org/10.1371/journal.pone.0285197.t003 efficacy of tirzepatide in the clinical setting of T2DM or obesity are ongoing. We can obtain more data to analyze after these trials are completed in the near future. Conclusion To sum up, this meta-analysis indicated that tirzepatide could significantly decrease body weight in T2DM and obesity patients, and it is a potential therapeutic regimen for weight-loss, but we need to be vigilant about its gastrointestinal reaction. Supporting information S1 Table. The data of meta-analysis. (XLSX) S1 Checklist. PRISMA 2020 checklist. (DOCX) Author Contributions Data curation: Baodong Ling, Xia Zhao, Xiaoling Jie, Yan Li. Formal analysis: Baodong Ling, Jing Chen. Methodology: Fei Lin. Writing – original draft: Fei Lin, Huijun Shang. Writing – review & editing: Fei Lin, Bin Yu, Guangyao Lv. References 1. Gadde K.M., et al., Obesity: Pathophysiology and Management. J Am Coll Cardiol, 2018. 71(1): 69–84. 2. Caballero B., Humans against Obesity: Who Will Win? Adv Nutr, 2019. 10(suppl_1): S4–s9. https://doi. org/10.1093/advances/nmy055 PMID: 30721956 3. Otto C.M., Heartbeat: weight loss interventions in patients with cardiovascular disease. Heart, 2021. 107(19): 1521–1523. https://doi.org/10.1136/heartjnl-2021-320238 PMID: 34521734 4. Yannakoulia M. and Panagiotakos D., Weight loss through lifestyle changes: impact in the primary pre- vention of cardiovascular diseases. Heart, 2021. 107(17): 1429–1434. https://doi.org/10.1136/heartjnl- 2019-316376 PMID: 33219107 5. Cosentino F., et al., 2019 ESC Guidelines on diabetes, pre-diabetes, and cardiovascular diseases developed in collaboration with the EASD. Eur Heart J, 2020. 41(2): 255–323. https://doi.org/10.1093/ eurheartj/ehz486 PMID: 31497854 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 14 / 16 PLOS ONEWeight loss of tirzepatide 6. 7. Joseph J.J., et al., Comprehensive Management of Cardiovascular Risk Factors for Adults With Type 2 Diabetes: A Scientific Statement From the American Heart Association. Circulation, 2022. 145(9): e722–e759. https://doi.org/10.1161/CIR.0000000000001040 PMID: 35000404 Jepsen M.M. and Christensen M.B., Emerging glucagon-like peptide 1 receptor agonists for the treat- ment of obesity. Expert Opin Emerg Drugs, 2021. 26(3): 231–243. https://doi.org/10.1080/14728214. 2021.1947240 PMID: 34176426 8. Williams D.M., et al., Glucagon-like Peptide-1 Receptor Analogues for the Treatment of Obesity. tou- chREV Endocrinol, 2022. 18(1): 43–48. https://doi.org/10.17925/EE.2022.18.1.43 PMID: 35975210 9. Baggio L.L. and Drucker D.J., Biology of incretins: GLP-1 and GIP. Gastroenterology, 2007. 132(6): 2131–57. https://doi.org/10.1053/j.gastro.2007.03.054 PMID: 17498508 10. Mathiesen D.S., et al., The Effects of Dual GLP-1/GIP Receptor Agonism on Glucagon Secretion-A Review. Int J Mol Sci, 2019. 20(17). https://doi.org/10.3390/ijms20174092 PMID: 31443356 11. Perez-MontesD.E.O.A., Pellitero S., and Puig-Domingo M., Obesity and GLP-1. Minerva Endocrinol (Torino), 2021. 46(2): 168–176. https://doi.org/10.23736/S2724-6507.20.03369-6 PMID: 33213122 12. Chadda K.R., Cheng T.S., and Ong K.K., GLP-1 agonists for obesity and type 2 diabetes in children: Systematic review and meta-analysis. Obes Rev, 2021. 22(6): e13177. https://doi.org/10.1111/obr. 13177 PMID: 33354917 13. Nauck M.A., et al., GLP-1 receptor agonists in the treatment of type 2 diabetes—state-of-the-art. Mol Metab, 2021. 46: 101102. https://doi.org/10.1016/j.molmet.2020.101102 PMID: 33068776 14. Willard F.S., et al., Tirzepatide is an imbalanced and biased dual GIP and GLP-1 receptor agonist. JCI Insight, 2020. 5(17). https://doi.org/10.1172/jci.insight.140532 PMID: 32730231 15. Administration, U.S.F.a.D., FDA Approves Novel, Dual-Targeted Treatment for Type 2 Diabetes. https://www.fda.gov/news-events/press-announcements/fda-approves-novel-dual-targeted-treatment- type-2-diabetes, 2018. 20(9): 2113–2120. 16. Aroda V.R., et al., Comparative efficacy, safety, and cardiovascular outcomes with once-weekly subcu- taneous semaglutide in the treatment of type 2 diabetes: Insights from the SUSTAIN 1–7 trials. Diabetes Metab, 2019. 45(5): 409–418. https://doi.org/10.1016/j.diabet.2018.12.001 PMID: 30615985 17. Min T. and Bain S.C., The Role of Tirzepatide, Dual GIP and GLP-1 Receptor Agonist, in the Manage- ment of Type 2 Diabetes: The SURPASS Clinical Trials. Diabetes Ther, 2021. 12(1): 143–157. 18. Sattar N., et al., Tirzepatide cardiovascular event risk assessment: a pre-specified meta-analysis. Nat Med, 2022. 28(3): 591–598. https://doi.org/10.1038/s41591-022-01707-4 PMID: 35210595 19. 20. 21. 22. Jastreboff A.M., et al., Tirzepatide Once Weekly for the Treatment of Obesity. New England Journal of Medicine, 2022. Zeng J., et al., The effect of soy intervention on insulin-like growth factor 1 levels: A meta-analysis of clinical trials. Phytother Res, 2020. 34(7): 1570–1577. https://doi.org/10.1002/ptr.6630 PMID: 32072706 Frı´as J.P., et al., Efficacy and safety of LY3298176, a novel dual GIP and GLP-1 receptor agonist, in patients with type 2 diabetes: a randomised, placebo-controlled and active comparator-controlled phase 2 trial. Lancet, 2018. 392(10160): 2180–2193. https://doi.org/10.1016/S0140-6736(18)32260-8 PMID: 30293770 Frias J.P., et al., Efficacy and tolerability of tirzepatide, a dual glucose-dependent insulinotropic peptide and glucagon-like peptide-1 receptor agonist in patients with type 2 diabetes: A 12-week, randomized, double-blind, placebo-controlled study to evaluate different dose-escalation regimens. Diabetes Obes Metab, 2020. 22(6): 938–946. https://doi.org/10.1111/dom.13979 PMID: 31984598 23. Rosenstock J., et al., Efficacy and safety of a novel dual GIP and GLP-1 receptor agonist tirzepatide in patients with type 2 diabetes (SURPASS-1): a double-blind, randomised, phase 3 trial. Lancet, 2021. 398(10295): 143–155. https://doi.org/10.1016/S0140-6736(21)01324-6 PMID: 34186022 24. Dahl D., et al., Effect of Subcutaneous Tirzepatide vs Placebo Added to Titrated Insulin Glargine on Gly- cemic Control in Patients With Type 2 Diabetes: The SURPASS-5 Randomized Clinical Trial. Jama, 2022. 327(6): 534–545. https://doi.org/10.1001/jama.2022.0078 PMID: 35133415 25. Heise T., et al., Effects of subcutaneous tirzepatide versus placebo or semaglutide on pancreatic islet function and insulin sensitivity in adults with type 2 diabetes: a multicentre, randomised, double-blind, parallel-arm, phase 1 clinical trial. Lancet Diabetes Endocrinol, 2022. https://doi.org/10.1016/S2213- 8587(22)00085-7 PMID: 35468322 26. Ludvik B., et al., Once-weekly tirzepatide versus once-daily insulin degludec as add-on to metformin with or without SGLT2 inhibitors in patients with type 2 diabetes (SURPASS-3): a randomised, open- label, parallel-group, phase 3 trial. Lancet, 2021. 398(10300): 583–598. https://doi.org/10.1016/S0140- 6736(21)01443-4 PMID: 34370970 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 15 / 16 PLOS ONEWeight loss of tirzepatide 27. Del Prato S., et al., Tirzepatide versus insulin glargine in type 2 diabetes and increased cardiovascular risk (SURPASS-4): a randomised, open-label, parallel-group, multicentre, phase 3 trial. Lancet, 2021. 398(10313): 1811–1824. https://doi.org/10.1016/S0140-6736(21)02188-7 PMID: 34672967 28. 29. Frı´as J.P., et al., Tirzepatide versus Semaglutide Once Weekly in Patients with Type 2 Diabetes. N Engl J Med, 2021. 385(6): 503–515. https://doi.org/10.1056/NEJMoa2107519 PMID: 34170647 Inagaki N., et al., Efficacy and safety of tirzepatide monotherapy compared with dulaglutide in Japanese patients with type 2 diabetes (SURPASS J-mono): a double-blind, multicentre, randomised, phase 3 trial. Lancet Diabetes Endocrinol, 2022. 10(9): 623–633. https://doi.org/10.1016/S2213-8587(22) 00188-7 PMID: 35914543 30. ClinicalTrials.gov, A Study of Tirzepatide (LY3298176) Compared to Dulaglutide in Participants With Type 2 Diabetes (SURPASS J-mono). https://clinicaltrials.gov/ct2/show/NCT03861052, 2021. 31. Bastin M. and Andreelli F., Dual GIP–GLP1-receptor agonists in the treatment of type 2 diabetes: A short review on emerging data and therapeutic potential. Diabetes, Metabolic Syndrome and Obesity: Targets and Therapy, 2019. 12: 1973–1985. https://doi.org/10.2147/DMSO.S191438 PMID: 31686879 32. Bhagavathula A.S., Vidyasagar K., and Tesfaye W., Efficacy and Safety of Tirzepatide in Patients with Type 2 Diabetes Mellitus: A Systematic Review and Meta-Analysis of Randomized Phase II/III Trials. Pharmaceuticals (Basel), 2021. 14(10). https://doi.org/10.3390/ph14100991 PMID: 34681215 33. Conlon J.M., F.P.M. O’Harte, and Flatt P.R., Dual-agonist incretin peptides from fish with potential for obesity-related Type 2 diabetes therapy–A review. Peptides, 2022. 147. https://doi.org/10.1016/j. peptides.2021.170706 PMID: 34861327 34. Apovian C.M., et al., Pharmacological management of obesity: an endocrine Society clinical practice guideline. J Clin Endocrinol Metab, 2015. 100(2): 342–62. https://doi.org/10.1210/jc.2014-3415 PMID: 25590212 35. Karagiannis T., et al., Management of type 2 diabetes with the dual GIP/GLP-1 receptor agonist tirzepa- tide: a systematic review and meta-analysis. Diabetologia, 2022. https://doi.org/10.1007/s00125-022- 05715-4 PMID: 35579691 36. Alkhezi O.S., et al., Comparative effectiveness of glucagon-like peptide-1 receptor agonists for the man- agement of obesity in adults without diabetes: A network meta-analysis of randomized clinical trials. Obes Rev, 2023. 24(3): e13543. https://doi.org/10.1111/obr.13543 PMID: 36579723 37. Bays H.E., et al., Anti-Obesity Medications and Investigational Agents: An Obesity Medicine Associa- tion (OMA) Clinical Practice Statement (CPS) 2022. Obesity Pillars, 2022. 38. Guan R., et al., Efficacy and safety of tirzepatide in patients with type 2 diabetes mellitus: A bayesian network meta-analysis. Front Pharmacol, 2022. 13: 998816. https://doi.org/10.3389/fphar.2022. 998816 PMID: 36313305 39. Borner T., et al., Glucagon-like peptide-1 in diabetes care: Can glycaemic control be achieved without nausea and vomiting? Br J Pharmacol, 2022. 179(4): 542–556. https://doi.org/10.1111/bph.15647 PMID: 34363224 40. De Block C.E.M., et al., Efficacy and safety of high-dose glucagon-like peptide-1, glucagon-like peptide- 1/glucose-dependent insulinotropic peptide, and glucagon-like peptide-1/glucagon receptor agonists in type 2 diabetes. Diabetes Obes Metab, 2022. 24(5): 788–805. https://doi.org/10.1111/dom.14640 PMID: 34984793 41. Kadowaki T., et al., Safety and efficacy of tirzepatide as an add-on to single oral antihyperglycaemic medication in patients with type 2 diabetes in Japan (SURPASS J-combo): a multicentre, randomised, open-label, parallel-group, phase 3 trial. Lancet Diabetes Endocrinol, 2022. 10(9): 634–644. https://doi. org/10.1016/S2213-8587(22)00187-5 PMID: 35914542 42. 43. Fundytus A., et al., Industry Funding of Oncology Randomised Controlled Trials: Implications for Design, Results and Interpretation. Clin Oncol (R Coll Radiol), 2022. 34(1): 28–35. https://doi.org/10. 1016/j.clon.2021.08.003 PMID: 34479769 Janiaud P., Cristea I.A., and Ioannidis J.P.A., Industry-funded versus non-profit-funded critical care research: a meta-epidemiological overview. Intensive Care Med, 2018. 44(10): 1613–1627. https://doi. org/10.1007/s00134-018-5325-3 PMID: 30151688 44. Kaneko S., Tirzepatide: A Novel, Once-weekly Dual GIP and GLP-1 Receptor Agonist for the Treatment of Type 2 Diabetes. touchREV Endocrinol, 2022. 18(1): 10–19. https://doi.org/10.17925/EE.2022.18.1. 10 PMID: 35949358 45. Frı´as J.P., Tirzepatide: a glucose-dependent insulinotropic polypeptide (GIP) and glucagon-like pep- tide-1 (GLP-1) dual agonist in development for the treatment of type 2 diabetes. Expert Rev Endocrinol Metab, 2020. 15(6): 379–394. https://doi.org/10.1080/17446651.2020.1830759 PMID: 33030356 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285197 May 4, 2023 16 / 16 PLOS ONE
10.1371_journal.pone.0285097	RESEARCH ARTICLE Socio-demographic, maternal, and infant characteristics associated with pacifier use among six-months old infants in Clark County, Nevada Kaelia Lynn Saniatan1, Smriti Neupane1, Chad CrossID 2, Gabriela BucciniID 1* a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 1 Department of Social and Behavioral Health, School of Public Health, University of Nevada, Las Vegas, Nevada, United States of America, 2 Department of Epidemiology & Biostatistics, School of Public Health, University of Nevada, Las Vegas, Nevada, United States of America * gabriela.buccini@unlv.edu Abstract OPEN ACCESS Background Citation: Saniatan KL, Neupane S, Cross C, Buccini G (2023) Socio-demographic, maternal, and infant characteristics associated with pacifier use among six-months old infants in Clark County, Nevada. PLoS ONE 18(4): e0285097. https://doi.org/ 10.1371/journal.pone.0285097 Editor: Mona Nabulsi, American University of Beirut Medical Center, LEBANON Received: December 22, 2022 Accepted: April 14, 2023 Published: April 27, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0285097 Copyright: © 2023 Saniatan et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: A de-identified data set is not possible to provide due to ethical and legal considerations. These sharing restrictions are imposed by the UNLV Institutional Board Review Pacifier use can interfere with nurturing care practices such as breastfeeding, soothing, and sleeping. Due to contradicting beliefs, recommendations, and the high frequency of pacifier use, understanding its associations may support shaping equitable public health recommen- dations. This study explored the socio-demographic, maternal, and infant characteristics associated with pacifier use among six-months old infants in Clark County, Nevada. Method Cross-sectional survey was conducted in 2021 with a sample of mothers (n = 276) of infants under six months old in Clark County, Nevada. Participants were recruited through adver- tisements in birth, lactation, pediatric care centers, and social media. We used binomial and multinomial logistic models to assess the association between pacifier use and the age of pacifier introduction, respectively, with household, maternal, infant, healthcare characteris- tics, and feeding and sleeping practices. Results More than half of the participants offered pacifiers (60.5%). Pacifier use was higher among low-income households (OR (95% CI) 2.06 (0.99–4.27)), mothers who identified as non-His- panic (OR (95% CI) 2.09 (1.22–3.59)), non-first-time mothers (OR (95% CI) 2.09 (1.11– 3.05)), and bottle-feeding infants (OR (95% CI) 2.76 (1.35–5.65)). Compared to those who did not introduce a pacifier, non-Hispanic mothers (RRR (95% CI) 2.34 (1.30–4.21)) and bottle-fed infants (RRR (95% CI) 2.71 (1.29–5.69)) had a higher risk of introducing pacifier within two weeks. Likewise, infants living in food insecure households (RRR (95% CI) 2.53 (0.97–6.58)) and mothers who have more than one child (RRR (95% CI) 2.44 (1.11–5.34)) had a higher risk of introducing a pacifier after two weeks. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 1 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada (IRB). The authors declare that a de-identified data set from this study are available upon request directly to Dr. Buccini, assistant professor at UNLV (gabriela.buccini@unlv.edu) and/or to the UNLV IRB (irb@unlv.edu). Funding: The authors received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. Conclusion Pacifier use is independently associated with maternal income, ethnicity, parity, and bottle feeding among six-month-old infants living in Clark County, Nevada. Household food inse- curity increased the relative risk of introducing a pacifier after two weeks. Qualitative research on pacifier use among families with diverse ethnic/racial backgrounds is needed to improve equitable interventions. Background Nurturing care is characterized as an environment that is responsive, emotionally supportive, sensitive to children’s health and nutritional needs, and developmentally appropriate and stim- ulating, with opportunities for play and exploration, protecting children from adversities [1, 2]. Nurturing mutually responsive care practices such as breastfeeding, soothing, and safe sleeping are critical public health approaches that will influence young children’s ability to self-regulate their emotions and actions, shape early childhood development outcomes, and reduce early disparities [3, 4]. Pacifiers have been used globally by at least two-thirds of parents during their infants’ first year as an aide to modulate nurturing care practices assisting with soothing, sleeping, and feed- ing [5]. Despite the immediate perceived benefits of pacifier use, there is strong evidence of the negative consequences of its use in the short- and long-term for the child. Pacifier use is related to atypical child development of oral functions [6], increased incidence of acute otitis media and other infections [7–9], speech problems [10], malocclusion [11, 12], lower levels of intelli- gence [13–15], and more recently, with early life weight outcomes [16] and the development of unhealthy lifestyles in adulthood including smoking, overeating, and other compulsive disor- ders [17, 18]. Further, the use of pacifiers has been associated with shorter exclusive breastfeed- ing duration [19, 20]. However, the causal relationship has not been proven due to methodological shortcomings in trials investigating this relationship [21]. On the other hand, there is evidence that reducing pacifier use may improve exclusive breastfeeding rates [22]. Recommendations for pacifier use vary across different cultures and purposes. The Ameri- can Academy of Pediatrics recommends breastfeeding [23] and using pacifiers to prevent Sud- den Infant Death Syndrome (SIDS). Pacifiers may be introduced after breastfeeding is firmly established [24] and should be used for the first six months of an infant’s life; after this age, the pacifier should be removed [9, 25]. By contrast, the American Academy of Pediatrics recom- mends implementing breastfeeding-supportive hospital practices, including avoiding pacifiers [23]. The World Health Organization recommends counseling mothers on the use and risks of pacifiers as part of the “Ten Steps to Successful Breastfeeding,” upon which the Baby-Friendly Hospital Initiative (BFHI) has been based [26]. Thus, there is no straightforward recommenda- tion for pacifier use and age of introduction as it needs to consider the potential benefits (e.g., SIDS) and risks (e.g., interference with exclusive breastfeeding). Due to these contradicting belief systems, recommendations, and the high frequency of use, understanding the associations of pacifier use may support shaping public health recommen- dations and interventions that are mutually nurturing, responsive, and culturally appropriate. Evidence from low and middle-income countries shows that infant characteristics linked with pacifier use include low birth weight, under six months of age, not breastfed within the first hour, nor in the maternity ward, nor on-demand, and tea consumed on the first day at home [20, 27]. Caregiver’s characteristics included younger caregivers, primiparous, lower education PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 2 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada level, low socioeconomic status, smoking history, working outside the home, maternal stress, and depression [18, 25]. However, all studies acknowledged were performed in countries out- side of the U.S. To date, as far as we know, no study has investigated the determinants of paci- fier use using a representative sample of infants. Therefore, this study aims to identify the socio-demographic, maternal, and infant characteristics associated with of pacifier use and age of pacifier introduction among six-months older infants living in Clark County, Nevada. Methods Study design This secondary data analysis is based on cross-sectional survey data from the 2021 Early Responsive Nurturing Care (EARN) survey, which investigated a wide range of factors impact- ing nurturing care practices (i.e., breastfeeding, soothing, and sleeping) among infants under six months old living in Clark County (Boulder City, Henderson, Las Vegas, North Las Vegas, and Mesquite), Nevada. The study protocol was approved by the Institutional Review Board of the University of Nevada Las Vegas (protocol 1767759–2). Participation in the study was voluntary, and written informed consent was obtained from all mothers for themselves and on behalf of their partici- pating children. No incentives were provided for participation in this survey. Setting As of 2022, Clark County, Nevada, has a total population of 2,350,206, with more than half of this population (54.0%) being White non-Hispanic, followed by those who identify as His- panic/Latino (32.8%) and non-Hispanic Black (12.8%) [26]. In 2020, 18.5% of non-Hispanic Black, 14.5% of Hispanic/Latino, and 5.5% of non-Hispanic (i.e., Asian, Native Hawaiian/ Pacific Islander, Native American, Bi-Racial) families lived below the poverty level [28]. 35.7% of households in Clark County have an average household income under $49,999 and an aver- age household size of 2.74 persons [26]. In the Centers for Disease Control and Prevention (CDC) 2022 Breastfeeding Report Card, the prevalence of exclusive breastfeeding in Nevada through six months was 22.3%, slightly lower than the national prevalence of 24.9% [29]. However, a median value of 75.4% of moth- ers in Clark County received early prenatal care, slightly lower than the national average of 77% [30]. Socio-cultural barriers present disadvantages within breastfeeding families through the lack of lactation services. Clark County only has two hospitals accredited in the Baby Friendly Hospital Initiative [31]. Sampling and data collection The sample size was designed to represent live births in Clark County. Therefore, the Southern Nevada Health District’s vital records birth certificate file was used as a source for the sampling frame, where infants who were born alive to mothers’ residents in Clark County in the past year (2020) were considered the sampling unit. In 2020, live births were estimated at 25,586 per year. Assuming a confidence level of 90% and an error of 5% and considering 50% comple- tion, the minimum sample size calculated was 268 mother-infant dyads. Eligible participants were mothers 18 years or older with an infant under the age of six months living within the Clark County district, including the city of Las Vegas, North Las Vegas, Henderson, Mesquite, and Boulder City. If mothers were found not to meet the eligibility criteria, they were not allowed to move forward with the survey. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 3 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada A convenience sampling technique was employed. Participants were recruited through advertisements across birth, lactation, and pediatric care centers throughout Clark County, social media (i.e., Facebook) posts in groups of mothers living in the selected area, and paid advertisements. The 2021 EARN surveys were available in English and Spanish and were dis- tributed only online due to COVID-19 safety measures. Consent was received before the start of the survey, and participants could stop the study at any time. Data collection started in August 2021 and ended in October 2021. There were 323 respondents, out of which 47 [14.6%] did not answer the survey questions regarding pacifier use and were excluded. This yielded an analytical sample of 276 participants to explore the associations of pacifier use. Measurements Outcomes. Two dependent variables were defined: (1) Pacifier use and (2) Age of Pacifier Introduction. 1. Pacifier use. The key dependent variable was pacifier use in the previous 24 hours to com- plete the survey. The status of pacifier use is aimed at minimizing possible biases resulting from the informant’s memory, which is recommended by the World Health Organization (WHO) when collecting information on nurturing care practices [32]. Pacifier use status was determined by the question, “In the last 24 hours, has your baby used a pacifier?” The response options were “yes” or “no.” 2. Age of Pacifier Introduction. The second dependent variable was the age of pacifier intro- duction. The age of introduction was determined by the question, “When did you start giv- ing pacifiers for your baby?” The response options were “No Pacifier Use,” “Within Two Weeks of Life,” or “After Two Weeks of Life.” Co-variables. Covariate selection was guided by the hierarchical framework [33] devel- oped based on a literature review supporting associations with pacifier use [17, 18] and data available on the 2021 EARN survey. These determinations informed the hierarchical frame- work illustrating potential associations of pacifier use organized across five categories of vari- ables: household characteristics (model 1), maternal characteristics (model 2), infant characteristics (model 3), healthcare characteristics (model 4), and infant feeding and sleeping practices (model 5) (Fig 1). Household characteristics included household food insecurity screening (yes/no) and household income (�$50,000/$50–79,999/$80–99,999/�$100,000). For the household food insecurity screening, the Hunger Vital Sign™, a validated two-question screening tool based on the U.S. Household Food Security Survey Module, was used [33, 34]. Households were identi- fied as being at risk for food insecurity if they answered either the two-question screening as ‘often true’ or ‘sometimes true’ (vs. ‘never true’). The maternal characteristics included mater- nal age (18-24/25-44), maternal education (no college degree/college degree), maternal ethnic- ity (Hispanic/non-Hispanic; due to limited sample for logistic regression, the non-Hispanic category included non-Hispanic White/non-Hispanic Black /non-Hispanic Other (i.e., Asian, Native Hawaiian/Pacific Islander, Native American, Bi-Racial)), first-time mother (yes, pri- mipara/no, multipara), and depression screening (low/high). For the depression screening, the modified two-item patient health questionnaire (PHQ-2) used [35]. Participants responding “nearly every day” or “3” to the questions were classified as high risk for depression. The infant characteristics included sex (male/female), low birth weight (yes/no), type of delivery (vaginal/ c-section), and age of pacifier introduction (no pacifier use/within two weeks of life/after two weeks of life). The healthcare characteristics included whether the infant was delivered in a PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 4 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada Fig 1. Framework displaying the models used and their proximal relationship to the outcome. https://doi.org/10.1371/journal.pone.0285097.g001 Baby-Friendly hospital accredited through the Baby Friendly Hospital Initiative (yes/no) and if the baby was breastfed within the first hour of life (yes/no). The infant feeding and sleeping practices included bottle-feeding (yes/no), exclusive breastfeeding (yes/no), and night-time family sleeping arrangements (bed sharing/no bed-sharing). As recommended by the WHO, infant feeding practices were determined by the status, i.e., on the day before the survey, to minimize possible biases resulting from the informant’s memory. Exclusive breastfeeding was defined as recommended by the WHO [32]—children younger than six months who had received breast milk as their only source of nutrition and hydration, without any solid or liquid supplement, including water and tea. Exclusive breastfeeding information was confirmed with questions about intake in the previous 24 hours of tea, juice, water, or other milk/infant for- mula and about intake of other foods like fruit or savory food. Bottle-feeding was defined as the children who are fed with any liquid (including breast milk) or semi-solid food from a bot- tle with nipple/teat, as recommended by the WHO [36]. Data analysis Analyses were conducted using the Statistical Package for Social Sciences Version 28 (SPSS) and Stata Version 17. Descriptive analyses of the outcomes and covariates were conducted. Then, bivariate analyses were conducted to examine the associations between pacifier use out- comes and covariates. Covariates were selected for inclusion in a multivariable model when the association had a p-value <0.20 in the bivariate analyses. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 5 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada To identify the associations of pacifier use, a hierarchical binomial logistic regression with robust variance [37] was performed to generate an adjusted odds ratio (AOR) and correspond- ing 95% confidence intervals (CIs). Similarly, to identify the associations of the age of pacifier introduction, a hierarchical multinomial logistic regression with robust variance to generate relative risk ratios (RRR) comparing the introduction of pacifier “Within Two Weeks,” “After Two Weeks” to the reference “No Pacifier Use.” The theoretical model presented in Fig 1 guided the hierarchical modelling data analysis approach [33, 38], which are particularly appropriate for use in studying determinants of childhood health outcomes such as breastfeed- ing [39] and pacifier use [27] as in the case of our analysis. The distal level consisted of the household characteristics variables and was the first to be included in the analysis (level 1 model) and remained as the control for the more proximal hierarchical models. The first intermediate level of maternal characteristics variables was included the model (level 2 model), which was adjusted by the distal level model and remained as the control for the subsequent models. The second intermediate level of infant characteris- tics variables was included in the subsequent model (level 3 model), which was adjusted for variables in the previous two more distal models and remained as the control for the subse- quent models. The third intermediate level consisted of healthcare characteristics variables and was included in the subsequent model (level l 4 model), which was adjusted for variables in the more distal three levels and remained as the control for the subsequent models. Lastly, the proximal level consisted of infant feeding and sleeping practices variables was included in the final model (level 5 model), which was adjusted for variables in the more distal levels. At each level model, a p value <0.05 was used as a statistical significance criterion to assess the correlation between variables and outcomes [33]. All variables that had a p-value <0.20 in the bivariate analyses were included in both binomial and multinomial hierarchical models and maintained in the models regardless of losing significance, as these data provide important adjustments to the parameter estimates in the final model [33]. Results The analytical sample consisted of 276 children under six months old from mothers in Clark County. Of the total participants, 60.5% of the respondents reported the use of a pacifier and were offered a pacifier within the first two weeks of life (n = 128, 46.4%). Around 14.9% of respondents were at risk for food insecurity. The majority of mothers were aged 25–44 (n = 248, 89.9%), had a college degree (n = 202, 73.2%), were non-Hispanic (n = 194, 70.3%), were not first-time mothers (n = 151, 54.7%), and had a low score for depression screening (n = 260, 94.2%). Among infants, most were female (n = 150, 54.3%) and were born from vagi- nal deliveries (n = 194, 70.3%). Although fewer babies were born in a Baby Friendly Hospital (n = 55, 19.9%), many were still placed on the breast within the first hour of delivery (n = 215, 77.9%). More than half of the mothers were exclusively breastfeeding (n = 148, 53.6%), most mothers bottle-fed their infants (n = 187, 67.8%) at the time of the survey, and 19.3% reported bed-sharing as their sleeping arrangement (Table 1). Pacifier use was more frequent among infants from households with a median income of less than $50,000 (n = 36, 70.6%) compared to those with incomes greater than $100,000 (n = 60, 55.1%). Mothers identified as non-Hispanic (n = 127, 65.5%), first-time mothers (n = 86, 65.5%), as well as infants that were delivered via cesarean section (n = 57, 69.5%), and were not breastfed within the first hour of delivery (n = 43, 66.4%) had higher frequencies of pacifier use compared to the comparison groups. Lastly, infants that are exclusively breastfed (n = 82, 55.4%), not bottle fed (n = 39, 43.8%), and share a bed with their caregiver (n = 27, 50.9%) have lower frequencies of pacifier use (Table 2). Regarding the age of pacifier PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 6 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada Table 1. Descriptive analysis of pacifier use, household, maternal, infant, and health care characteristics, and infant feeding and sleeping practices, 2021. Study Variables Pacifier Use No Yes Pacifier Introduction Age No Pacifier Use Within Two Weeks After Two Weeks Household Income �$100,000 $50,000–99,999 �$50,000 Household Food Insecurity (n = 266) Food Secured Food Insecure Maternal Age 18–24 25–44 Mother’s Education College Degree No College Degree Maternal Ethnicity Hispanic Non-Hispanic First Time Mother Multiparous Primiparous Depression Screening Low Risk High Risk Child’s Sex Female Male Type of Delivery Vaginal C-Section Baby Friendly Hospital Yes No Breastfed in First Hour Yes No Exclusive Breastfeeding Yes No Bottle Feeding No Frequency (n = 276) Percent (%) 109 167 109 128 39 109 116 51 226 40 28 248 202 74 82 194 151 125 260 16 150 126 194 82 55 221 215 61 148 128 89 39.5 60.5 39.5 46.4 14.1 39.5 42.0 18.5 85.0 15.0 10.1 89.9 73.2 26.8 29.7 70.3 54.7 45.3 94.2 5.80 54.3 45.7 70.3 29.7 19.9 80.1 77.9 22.1 53.6 46.4 32.2 (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 7 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada Table 1. (Continued) Study Variables Yes Sleeping Arrangements (n = 274) No Bed Sharing Bed Sharing https://doi.org/10.1371/journal.pone.0285097.t001 Frequency (n = 276) Percent (%) 187 221 53 67.8 80.7 19.3 introduction, non-Hispanic mothers (n = 100, 51.5%) introduced pacifiers within two weeks, while more Hispanic mothers (n = 12, 14.6%) introduced them after two weeks. Families living in food-insecure households (n = 9, 22.5%) were more likely to introduce pacifiers after two weeks compared to food-secure households. Infants that were breastfed within the first hour of delivery had a lower prevalence of introduction both within two weeks (n = 94, 43.7%) and after two weeks (n = 30, 13.9%) compared to those who were not breastfed within the first hour. Lastly, infants who were not bottle-fed infants were less likely to be introduced to a paci- fier within two weeks (n = 30, 33.7%) and after two weeks (n = 9, 10.1%) (Table 3). Pacifier use was independently associated with mothers who identified as non-Hispanic (OR (95% CI) 2.09 (1.22–3.59)) and mothers who have more than one child (OR (95% CI) 2.09 (1.11–3.05])) in level 2, and with bottle-feeding infants (OR (95% CI) 2.76 (1.35–5.65)) in level 5 (Table 4). Compared with those who did not introduce a pacifier, infants living in food insecure households (RRR (95% CI) 2.53 (1.00–6.58)) in level 1 and mothers who have more than one child (RRR (95% CI) 2.44 (1.11–5.34)) in level 2 had a higher risk of introducing a pacifier after two weeks. Likewise, non-Hispanic mothers (RRR (95% CI) 2.34 (1.30–4.21)) in level 2 and bottle-fed infants (RRR (95% CI) 2.71 (1.29–5.69)) in level 5 had a higher risk of introduc- ing pacifier within two weeks (Table 5). Discussion By taking a hierarchal modelling data analysis approach our study also identified that pacifier use is associated with maternal ethnicity and parity (level 2 model) and bottle feeding (level 5 model) among six-month-old living in Clark County, Nevada. In addition to these known fac- tors, our study found that the age of pacifier introduction was associated with household food insecurity (level 1 model). There is limited research on the relationship between pacifier use, age of introduction, and food insecurity. Food insecurity, defined as “a household-level eco- nomic and social condition of limited or uncertain access to adequate food” [40], can cause nutrient deficiencies that can affect the health of both the caregiver and the child [41]. In turn, food insecurity is a chronic stressor that can lead to mental health issues for the caregiver, such as anxiety and depression, which can affect parenting skills and the ability to provide nurturing care to their infant [42]. An emotionally distressed caregiver may be more likely to introduce a pacifier [42, 43] rather than finding alternative soothing methods to calm a fussy baby. Although maternal depression did not sustain an association in the multivariate analysis, it is essential to consider when developing pacifier-use interventions [17, 20, 43, 44], especially among food-insecure families, which due to COVID is a group that has increased exponen- tially in recent years [45]. Our study highlights the role of maternal ethnicity in pacifier use practices. While mothers of Hispanic ethnicity were more likely not to offer a pacifier, mothers of non-Hispanic ethnic- ity were found to be more likely to offer a pacifier. Prior studies have linked maternal ethnicity with disparities in child health and nutrition outcomes, such as low exclusive breastfeeding PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 8 / 19 PLOS ONETable 2. Bivariate analysis of pacifier use by household, maternal, infant, and health care characteristics, and infant feeding and sleeping practices, 2021. Variables Pacifier Use No Pacifier Use P-Value Factors associated with pacifier use among six-months old infants in Nevada Household Income �$100,000 $50,000–99,999 �$50,000 Household Food Insecurity (n = 266) Food Secured Food Insecure Maternal Age 18–24 25–44 Mother’s Education College Degree No College Degree Maternal Ethnicity Hispanic Non-Hispanic First Time Mother Multiparous Primiparous Depression Screening Low Risk High Risk Child’s Sex Female Male Type of Delivery Vaginal C-Section Baby Friendly Hospital Yes No Breastfed in First Hour Yes No Exclusive Breastfeeding Yes No Bottle Feeding No Yes Sleeping Arrangements (n = 274) No Bed Sharing Bed Sharing p<0.20 https://doi.org/10.1371/journal.pone.0285097.t002 n 60 71 36 131 27 18 149 118 49 40 127 81 86 156 11 88 79 110 57 34 133 124 43 82 85 39 128 139 27 % 55.1 61.2 70.6 58.0 67.5 64.3 60.1 58.4 66.2 48.8 65.5 53.6 68.8 60.0 68.8 58.7 62.7 56.7 69.5 61.8 60.2 57.7 70.5 55.4 66.4 43.8 68.4 62.9 50.9 n 49 45 15 95 13 10 99 84 25 42 67 70 39 104 5 62 47 84 25 21 88 91 18 66 43 50 59 82 26 % 44.9 38.8 29.4 42.0 32.5 35.7 39.9 41.6 33.8 51.2 34.5 46.4 31.2 40.0 31.2 41.3 37.3 43.3 30.5 38.2 39.8 42.3 29.5 44.6 33.6 56.2 31.6 37.1 49.1 .169 .258 .666 .240 .010 .010 .487 .495 .047 .824 .071 .062 < .001 .110 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 9 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada Table 3. Bivariate analysis of the age of pacifier use introduction by household, maternal, infant, and health care characteristics, and infant feeding and sleeping practices, 2021. Variables Household Income �$100,000 $50,000–99,999 �$50,000 Household Food Insecurity (n = 266) Food Secured Food Insecure Maternal Age 18–24 25–44 Mother’s Education College Degree No College Degree Maternal Ethnicity Hispanic Non- Hispanic First Time Mother Multiparous Primiparous Depression Screening Low Risk High Risk Child’s Sex Female Male Type of Delivery Vaginal C-Section Baby Friendly Hospital Yes No Breastfed in First Hour Yes No Exclusive Breastfeeding Yes No Bottle Feeding No Yes Sleeping Arrangements (n = 274) No Bed Sharing Bed Sharing p<0.20 https://doi.org/10.1371/journal.pone.0285097.t003 Pacifier Introduced Within 2 Weeks Pacifier Introduced After 2 Weeks P-Value n 48 55 25 105 18 13 115 90 38 28 100 65 63 120 8 35 63 87 41 22 106 94 34 65 63 30 98 104 23 % 44.0 47.4 49.0 46.5 45.0 46.4 46.4 44.6 51.3 34.1 51.5 43.0 50.4 46.1 50.0 43.3 50.0 44.8 50.0 40.0 48.0 43.7 55.7 43.9 49.2 33.7 52.4 47.1 43.4 n 12 16 11 26 9 5 34 28 11 12 27 16 23 36 3 23 16 23 16 12 27 30 9 17 22 9 30 35 4 % 11.0 13.8 21.6 11.5 22.5 17.9 13.7 13.9 14.9 14.6 13.9 10.6 18.4 13.8 18.7 15.3 12.7 11.9 19.5 21.8 12.2 13.9 14.7 11.5 17.2 10.1 16.0 15.8 07.5 .273 .142 .811 .493 .020 .021 .742 .529 .077 .173 .176 .129 < .000 .154** PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 10 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada Table 4. Hierarchical logistic regression on the association between pacifier use and household, maternal, infant, and health care characteristics as well as infant feeding and sleeping practices, 2021. Each model level represents the addition of blocks of model variables as specified. A value of “1” across rows indicates the compari- son group in the model categorical predictors. Level 1 Model Level 2 Model Level 3 Model Level 4 Model Level 5 Model Pacifier use Adjusted OR (95% CI) Level 1 Variables: Household Characteristics Household Income �$100,000 $50,000–99,999 �$50,000 1 1.29 (0.76–2.19) 1.96 (0.96–3.99) 1 1.34 (0.77–2.33) 2.06 (0.99–4.27) 1 1.29 (0.74–2.25) 1.99 (0.96–4.15) 1 1.28 (0.74–2.23) 1.95 (0.94–4.07) 1 1.52 (0.86–2.69) 2.73 (1.22–6.10) Level 2 Variables: Maternal Characteristics Maternal Ethnicity Hispanic Non-Hispanic Parity Primiparous Multiparous Level 3 Variables: Infant Characteristics Type of Delivery Vaginal C-section Level 4 Variables: Healthcare Characteristics Breastfed in First Hour Yes No - - - - - - - - Level 5 Variables: Infant Feeding and Sleeping Practices Exclusive Breastfeeding Yes No Bottle Feeding No Yes Sleeping Arrangements No Bed Sharing Bed Sharing p<0.05 - - - - - - https://doi.org/10.1371/journal.pone.0285097.t004 1 2.09 (1.22–3.59) 1 1.84* (1.11–3.05) 1 1 1 2.04 (1.18–3.50) 2.02 (1.18–3.49) 2.16 (1.23–3.77) 1 1 1 1.78 (1.07–2.96) 1.72 (1.02–2.90) 1.37 (0.77–2.43) - - - - - - - - - - 1 1 1 1.52 (0.86–2.70) 1.45 (0.82–2.58) 1.40 (0.76–2.56) - - - - - - - - 1 1 1.23 (0.64–2.37) 1.24 (0.62–2.50) - - - - - - 1 0.83 (0.43–1.59) 1 2.76* (1.35–5.65) 1 0.68 (0.35–1.34) rates [46–48], high infant mortality [49], and lower usage of emergency department services [50]. However, there is a lack of data exploring the different ways maternal ethnicity influences pacifier use across diverse settings, which may contribute to the lack of culturally appropriate nurturing care support amplifying disparities among communities of color [48, 51]. We acknowledge that our data is limited to compare maternal ethnicity and future studies should analyze the influence of maternal race on pacifier outcomes. Maternal parity throughout research has been shown to influence an increase in pacifier use and early cessation of breast- feeding [27, 52]. The research aligns with the data displayed in this study. A study in Australia investigated why a first-time caregiver may offer their infant a pacifier. It was found that an area of opportunity to educate first-time mothers on pacifier use is with their families. Many first-time mothers received pacifier use advice from their own mothers and/or mothers-in-law [52]. Therefore, when developing an intervention to reduce pacifier use, involving other PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 11 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada Table 5. Multinomial logistic regression on the association between age of pacifier use introduction and household, maternal, infant, and health care characteristics as well as infant feeding and sleeping practices, 2021. A value of “1” across rows indicates the comparison group in the model categorical predictors. Age of pacifier use introduction RRR (95% CI) Level 1 Model Level 2 Model Level 3 Model Level 4 Model Level 5 Model Pacifier Introduced Within 2 Weeks Pacifier Introduced After 2 Weeks Pacifier Introduced Within 2 Weeks Pacifier Introduced After 2 Weeks Pacifier Introduced Within 2 Weeks Pacifier Introduced After 2 Weeks Pacifier Introduced Within 2 Weeks Pacifier Introduced After 2 Weeks Pacifier Introduced Within 2 Weeks Pacifier Introduced After 2 Weeks Level 1 Variables: Household Characteristics Household Food Insecurity Food Secured Food Insecure 1 1.25 (0.58– 2.70) 1 2.53* (1.00– 6.58) 1 1 1 1 1 1 1 1 1.37 (0.64– 2.93) 2.80 (1.06– 7.42) 1.39 (0.64– 2.99) 2.95 (1.10– 7.87) 1.33 [0.61– 2.90] 3.25 (1.19– 8.88) 1.46 (0.65– 3.28) 3.45 (1.22– 9.71) Level 2 Variables: Maternal Characteristics Maternal Ethnicity Hispanic Non-Hispanic Parity Primiparous Multiparous - - - - Level 3 Variables: Infant Characteristics Type of Delivery Vaginal C-section - - Level 4 Variables: Healthcare Characteristics Baby Friendly Hospital Yes No Breastfed in First Hour Yes No - - - - - - - - - - Level 5 Variables: Infant Feeding and Sleeping Practices Exclusive Breastfeeding Yes No Bottle Feeding No Yes Sleeping Arrangements - - - - - - - - 1 2.34* (1.30– 4.21) 1 1 1 1 1 1 1 1.56 (0.66– 3.67) 2.28 (1.27– 4.12) 1.44 (0.61– 3.41) 2.32 (1.27– 4.23) 1.32 (0.55– 3.18) 2.42 (1.31– 4.48) 1.27 (0.51– 3.13) 1 1.65 (0.96– 2.84) 1 2.44* (1.11– 5.34) 1 1 1 1 1 1 1.62 (0.94– 2.78) 2.33 (1.06– 5.14) 1.56 (0.90– 2.71) 2.36 (1.04– 5.33) 1.31 (0.72– 2.39) 1.96 (0.87– 4.42) - - - - - - - - - - 1 1 1 1 1 1 1.38 (0.76– 2.53) 2.10 (0.91– 4.87) 1.28 (0.69– 2.38) 2.33 (0.97– 5.57) 1.27 (0.66– 2.43) 2.29 (0.96– 5.42) - - - - - - - - - - - - - - - - 1 1 1 1 1.31 (0.62– 2.73) 0.51 (0.21– 1.25) 1.31 (0.63– 2.74) 0.47 (0.18– 1.20) 1 1 1 1 1.34 (0.65– 2.64) 0.74 (0.25– 2.17) 1.38 (0.68– 2.80) 0.79 (0.26– 2.36) - - - - - - - - 1 1 0.75 (0.38– 1.46) 0.99 (0.38– 2.60) 1 2.71* (1.29– 5.69) 1 1.98 (0.69– 5.65) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 (Continued ) 12 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada Table 5. (Continued) Age of pacifier use introduction RRR (95% CI) Level 1 Model Level 2 Model Level 3 Model Level 4 Model Level 5 Model Pacifier Introduced Within 2 Weeks Pacifier Introduced After 2 Weeks Pacifier Introduced Within 2 Weeks Pacifier Introduced After 2 Weeks Pacifier Introduced Within 2 Weeks Pacifier Introduced After 2 Weeks Pacifier Introduced Within 2 Weeks Pacifier Introduced After 2 Weeks Pacifier Introduced Within 2 Weeks Pacifier Introduced After 2 Weeks No Bed Sharing Bed Sharing - - - - - - - - - - - - - - 1 1 0.90 (0.45– 1.82) 0.50 (0.15– 1.73) *p<0.05; reference category is no pacifier use. https://doi.org/10.1371/journal.pone.0285097.t005 members of the family such as fathers and grandmothers has shown to be effective [53], further demonstrating why culturally appropriate approaches are imperative in creating equitable solutions. Corroborating our findings, bottle feeding has been associated with increased pacifier use in infants [54]. Increased use of bottle feeding has been associated with low birth weight, whether this was the first child, delivery by cesarean section, and male sex [25]. Bottle feeding along with pacifier use can create difficulties with breastfeeding compared to infants not offered an artificial nipple [54]. Introducing an artificial nipple, by either bottle or pacifier use, can create “nipple confusion,” which refers to an infant’s inability to establish proper oral con- figuration, latching techniques, and suck patterns to extract milk from the breast after being exposed to an artificial nipple [55]. Nipple confusion can occur due to the inability of some infants to adapt to different oral configurations, the “imprinting” in latching learning that occurs in the neonatal period, and initial difficulties in latching on to the breast, who are more susceptible to nipple confusion. A literature review found robust evidence of nipple confusion due to bottle use since it releases milk faster than sucking the breast and the use of both pacifi- ers and bottles [55]. Pacifier use is still controversial due to its implied benefits and drawbacks. In contrast, the use of a pacifier is recommended for specific outcomes, such as the protection against Sudden Infant Death Syndrome (SIDS) [24], stimulation of non-nutritive sucking [43], pain manage- ment in the newborn [56], and regulation of fussy behavior [57], the extended use of a pacifier can lead to further health disparities. Disparities include but are not limited to the poor devel- opment of oral functions [6], increased incidence of acute otitis media, and other infections [7–9], malocclusion [11, 12], early life weight outcomes [16], the development of unhealthy lifestyles in adulthood including smoking, overeating, and other compulsive disorders [17, 18] as well as pacifier use beyond six months, have been associated with accelerated infant growth and toddler obesity [14]. Moreover, a pacifier use has been associated to speech problems [10], lower social interactions [58], and lower levels of intelligence. A recent analysis of a birth cohort study found a strong association between pacifier use and lower intelligence quotient at six years [14]. One possible hypothesis is that children using a pacifier, especially those who use it more intensely, are less stimulated affecting social interaction skills and may lose a criti- cal window for optimal early childhood development exacerbating disparities early in life. Addressing these disparities is one of the overarching goals of all areas of public health to create a healthier population [59]. In addition, pacifier use may contribute to disparities in exclusive breastfeeding. However, the causal relationship has not been fully elucidated [20], it is known that promoting exclusive breastfeeding can reduce pacifier use [22]. In the context of PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 13 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada our study, the prevalence of exclusive breastfeeding in Nevada through six months was 22.3% [29], which is slightly lower than the national prevalence, but far beyond the Global Maternal, Infant, and Young Child Nutrition Target to reach at least 70.0% of EBF by 2030 [59]. There- fore, the exclusive breastfeeding goal in Nevada will not be met unless key modifiable risk fac- tors such as pacifier use are culturally appropriately addressed [20]. Some evidence-based initiatives that could embed culturally appropriate messages to promote breastfeeding and reduce pacifier use may include (i) providing education programs for parents and caregivers through various channels such as prenatal classes, postnatal support groups, and online resources [60], (ii) increasing access to community-based initiatives that offer perinatal sup- port through peer counselor [61] and lactation consultants [62], (iii) training of health provid- ers can help ensure that parents receive consistent messages on pros and cons of pacifier use, and (iv) promoting regulatory policies such as restrictions of predatory marketing of pacifiers [63, 64]. Many of these recommendations are embedded into the Baby Friendly Hospital Ini- tiative which follows the “Ten Steps to Successful Breastfeeding”. Steps 8 & 9 of the “Ten Steps. . .” focus on educating parents to recognize and respond to their infant’s feeding cues and the pros and cons associated with pacifier use, respectively. While our study did not find an association between place of birth and pacifier use, potentially due to the low coverage of births occurred in Baby Friendly Hospitals in Clark County (19.9% vs 27.0% in the U.S–[31]); it is important to note that scaling up Baby Friendly Hospital Initiative has been associated with increased rates of exclusive breastfeeding and decreased rates of pacifier use [65]. Our study has some limitations to be considered when generalizing the findings. We sur- veyed a convenience sample of infants under six months old across Clark County, Nevada. Several data collection efforts across birth, lactation, and pediatric care centers throughout Clark County were made to recruit a diverse population of mothers; however, due to the COVID-19 pandemic, the majority of the 2021 EARN surveys’ sample was recruited through paid advertisement on social media. We acknowledge that this may have limited the diversity of the sample size; however, our data were similar compared to the demographic data (i.e., eth- nicity, educational attainment, and household income) in Clark County [66]. Nevertheless, our findings may be generalized to U.S. areas with a similarly high proportion of urban popu- lations as Clark County. To prevent recall bias questions pertaining to pacifier use were formatted to capture the sta- tus of the use on the 24-h prior to the survey as recommended by WHO [32]. In doing so, we understand that this may have limited our results by excluding mother-infant dyads that used a pacifier until certain age in infancy. On the other hand, it minimizes the participant to pro- vide false or inaccurate answers. Another limitation is that we did not collect the age of the infant at the time of the survey. We acknowledge that practices are influenced by the infant’s age, and some associations may not be found or weakened due to lack of age detail. The inten- sity of pacifier uses and psychosocial factors that may influence the breastfeeding process, including the infant’s behavior (e.g., temperament and the mother’s breastfeeding intentions), were not collected. Due to the cross-sectional design, the temporal sequence of events between pacifier use and the associated factors cannot be established; while reverse causality cannot be ruled out, the use of a conceptual hierarchical approach considering social, biological, and temporal relationships may leverage complex inter-relationships between these determinants [33]. Further, cross-sectional designs can generate hypotheses for the development of longitu- dinal studies [67]. The selection of a hierarchical approach to analyze the data come with both advantages and limitations throughout this study. The use of the hierarchical model was selected for its ability to measure significance across multiple variables in relation to their level influence to the out- comes [33]. Many of the selected variables in this study are interrelated to each other as they PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 14 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada can influence the overall significance to the outcomes. The hierarchical display of variables showcases these complex inter-relationships [33]. Due to this, some variables would display significance in one level of analysis while losing significance when other levels are added to the model. Regardless, variables were maintained in the entire model despite having lost statistical significance after the inclusion of inferior variables as they provide insight to the adjustments made in the analysis [27]. In this context, our findings support the importance to promote evidence-based initiatives that embed culturally appropriate messages to reduce pacifier use, including the scaling up of Baby Friendly Hospitals and community-based education programs for parents and caregiv- ers. Hence, recommendations regarding pacifier use should use a counseling approach to sup- port conscious decisions considering each family’s context, culture, and goals regarding nurturing care practices, including exclusive breastfeeding goals [19–22]. Messages should be delivered by taking a welcoming and listening approach that encompasses both standardized information regarding the pros and cons of pacifier use, early childhood developmental expected behaviors, soothing methods to calm a fussy baby can help parents to understand and respond assertively to their child’s needs [57]; by favoring dialogic communication, emo- tions that generate guilt, pain, social pressure, and obligation, which are not infrequent, can be welcomed and contribute to reflecting on culturally appropriate nurturing care resources to help reduce disparities in breastfeeding and pacifier use practices. Lastly, tailored messages for families who choose to offer the pacifier, including options to limited by one-year-old or restricted use at critical times and, once the habit of using a pacifier is established, support those families for withdrawal are critical [68]. Further qualitative research into cultural aspects of pacifier use among families from diverse racial backgrounds in the U.S. is needed. Author Contributions Conceptualization: Kaelia Lynn Saniatan, Smriti Neupane, Gabriela Buccini. Data curation: Chad Cross. Formal analysis: Smriti Neupane, Chad Cross, Gabriela Buccini. Funding acquisition: Gabriela Buccini. Methodology: Kaelia Lynn Saniatan, Smriti Neupane. Software: Chad Cross. Supervision: Chad Cross, Gabriela Buccini. Writing – original draft: Kaelia Lynn Saniatan, Gabriela Buccini. Writing – review & editing: Gabriela Buccini. References 1. Black MM, Walker SP, Fernald LCH, Andersen CT, DiGirolamo AM, Lu C, et al. Early childhood devel- opment coming of age: science through the life course. The Lancet. 2016; 389[10064]:77–90. https:// doi.org/10.1016/S0140-6736(16)31389-7 PMID: 27717614 2. Britto PR, Lye SJ, Proulx K, Yousafzai AK, Matthews SG, Vaivada T, et al. Nurturing care: promoting early childhood development. The Lancet. 2017 Jan 7; 389[10064]:91–102. https://doi.org/10.1016/ S0140-6736(16)31390-3 PMID: 27717615 3. Paul IM, Savage JS, Anzman SL, Beiler JS, Marini ME, Stokes JL, et al. Preventing obesity during infancy: a pilot study. Obesity [Silver Spring]. 2011 Feb; 19[2]:353–61. https://doi.org/10.1038/oby. 2010.182 PMID: 20725058 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 15 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada 4. Skouteris H, Bergmeier HJ, Berns SD, Betancourt J, Boynton-Jarrett R, Davis MB, et al. Reframing the early childhood obesity prevention narrative through an equitable nurturing approach. Matern Child Nutr. 2021 Jan; 17[1]:e13094. https://doi.org/10.1111/mcn.13094 PMID: 33067918 5. Vogel A, Hutchison B, Mitchell E. The impact of pacifier use on breastfeeding: A prospective cohort study. Journal of Paediatrics and Child Health. 2001; 37[1]:58–63. https://doi.org/10.1046/j.1440-1754. 2001.00581.x PMID: 11168872 6. Silveira LM da Prade LS, Ruedell AM Haeffner LSB, Weinmann ARM. Influence of breastfeeding on children’s oral skills. Rev Saude Publica. 2013 Feb; 47[1]:37–43. https://doi.org/10.1590/s0034- 89102013000100006 PMID: 23703128 7. Rovers MM, Numans ME, Langenbach E, Grobbee DE, Verheij TJ, Schilder AG. Is pacifier use a risk factor for acute otitis media? A dynamic cohort study. Fam Pract. 2008 Aug; 25[4]:233–6. https://doi. org/10.1093/fampra/cmn030 PMID: 18562333 8. North K, Fleming P, Golding J. Pacifier use and morbidity in the first six months of life. Pediatrics. 1999 Mar; 103[3]:E34. https://doi.org/10.1542/peds.103.3.e34 PMID: 10049990 9. Niemela¨ M, Pihakari O, Pokka T, Uhari M. Pacifier as a risk factor for acute otitis media: A randomized, controlled trial of parental counseling. Pediatrics. 2000 Sep; 106[3]:483–8. https://doi.org/10.1542/ peds.106.3.483 PMID: 10969091 10. Shotts LL, McDaniel M, Neeley RA. The Impact of Prolonged Pacifier Use on Speech Articulation: A Preliminary Investigation. Contemporary Issues in Communication Science and Disorders. 2008 Mar; 35[Spring]:72–5. 11. Warren JJ, Bishara SE. Duration of nutritive and nonnutritive sucking behaviors and their effects on the dental arches in the primary dentition. American Journal of Orthodontics and Dentofacial Orthopedics. 2002 Apr 1; 121[4]:347–56. https://doi.org/10.1067/mod.2002.121445 PMID: 11997758 12. Peres KG, Barros AJD, Peres MA, Victora CG. Effects of breastfeeding and sucking habits on malocclu- sion in a birth cohort study. Rev Sau´de Pu´blica. 2007 Jun; 41:343–50. https://doi.org/10.1590/s0034- 89102007000300004 PMID: 17515986 13. Gale CR, Martyn CN. Breastfeeding, dummy use, and adult intelligence. The Lancet. 1996 Apr 20; 347 [9008]:1072–5. https://doi.org/10.1016/s0140-6736(96)90278-0 PMID: 8602057 14. Giugliani ERJ, Gomes E, Santos IS, Matijasevich A, Camargo-Figuera FA, Barros AJD. All day-long pacifier use and intelligence quotient in childhood: A birth cohort study. Paediatr Perinat Epidemiol. 2021 Jul; 35[4]:511–8. https://doi.org/10.1111/ppe.12752 PMID: 33570810 15. Giugliani ERJ, Gomes E, dos Santos I da S, Matijasevich A, Camargo-Figuera FA, Barros A. Daily Use of Pacifier in the First Years of Life Is Associated with Lower IQ at Age Six [Internet]. Rochester, NY; 2019 [cited 2022 Dec 22]. Available from: https://papers.ssrn.com/abstract=3346507 16. Hohman EE, Savage JS, Birch LL, Beiler JS, Paul IM. Pacifier Use and Early Life Weight Outcomes in the Intervention Nurses Start Infants Growing on Healthy Trajectories Study. Child Obes. 2018 Jan; 14 [1]:58–66. https://doi.org/10.1089/chi.2017.0177 PMID: 28976781 17. Ferreira HR, Rosa EF, Antunes JLF, Duarte DA, Imparato JCP, Pannuti CM, et al. Prolonged Pacifier Use during Infancy and Smoking Initiation in Adolescence: Evidence from a Historical Cohort Study. EAR. 2015; 21[1]:33–8. https://doi.org/10.1159/000365351 PMID: 25358513 18. Arenz S, Ru¨ckerl R, Koletzko B, von Kries R. Breast-feeding and childhood obesity—a systematic review. Int J Obes. 2004 Oct; 28[10]:1247–56. https://doi.org/10.1038/sj.ijo.0802758 PMID: 15314625 19. Buccini G, Pe´ rez-Escamilla R, Venancio S. Pacifier Use and Exclusive Breastfeeding in Brazil. Journal of human lactation: official journal of International Lactation Consultant Association. 2015 Oct 8; 32. https://doi.org/10.1177/0890334415609611 PMID: 26446096 20. Buccini G, Perez-Escamilla R, Paulino L, Araujo C, Venancio S. Pacifier use and interruption of exclu- sive breastfeeding: Systematic review and meta-analysis. 2016 [cited 2022 Jul 10]; Available from: https://onlinelibrary.wiley.com/doi/10.1111/mcn.12384 PMID: 27863027 21. Buccini G, Pe´ rez-Escamilla R, Venancio SI. Routine pacifier use in infants: pros and cons. J Pediatr [Rio J]. 2019 Sep 1; 95[5]:619–21. https://doi.org/10.1016/j.jped.2019.06.001 PMID: 31185193 22. Buccini G, Pe´ rez-Escamilla R, Benicio MHD, Giugliani ERJ, Venancio SI. Exclusive breastfeeding changes in Brazil attributable to pacifier use. PLOS ONE. 2018 Dec 19; 13[12]:e0208261. https://doi. org/10.1371/journal.pone.0208261 PMID: 30566449 23. Meek JY, Noble L, Section on Breastfeeding. Policy Statement: Breastfeeding and the Use of Human Milk. Pediatrics. 2022 Jun 27; 150[1]:e2022057988. 24. Moon RY, Carlin RF, Hand I. Sleep-Related Infant Deaths: Updated 2022 Recommendations for Reducing Infant Deaths in the Sleep Environment \| Pediatrics \| American Academy of Pediatrics [Inter- net]. [cited 2022 Jul 14]. Available from: https://publications.aap.org/pediatrics/article/150/1/ e2022057990/188304/Sleep-Related-Infant-Deaths-Updated-2022?autologincheck=redirected PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 16 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada 25. Task Force on Sudden Infant Death Syndrome. The Changing Concept of Sudden Infant Death Syn- drome: Diagnostic Coding Shifts, Controversies Regarding the Sleeping Environment, and New Vari- ables to Consider in Reducing Risk. Pediatrics. 2005 Nov 1; 116[5]:124n˙5–55. https://doi.org/10.1542/ peds.2005-1499 PMID: 16216901 26. WHO. Ten steps to successful breastfeeding [Internet]. [cited 2022 Jul 14]. Available from: https://www. who.int/teams/nutrition-and-food-safety/food-and-nutrition-actions-in-health-systems/ten-steps-to- successful-breastfeeding 27. Buccini G dos S, Benı´cio MHD, Venancio SI. Determinants of using pacifier and bottle feeding. Rev Saude Publica. 2014 Aug; 48[4]:571–82. https://doi.org/10.1590/s0034-8910.2014048005128 PMID: 25210816 28. Healthy Southern Nevada. Southern Nevada:: Indicators:: Families Living Below Poverty Level:: County: Clark [Internet]. [cited 2022 Aug 3]. Available from: https://www.healthysouthernnevada.org/ indicators/index/view?indicatorId=240&localeId=1800 29. CDC. 2022 Breastfeeding Report Card [Internet]. Centers for Disease Control and Prevention. 2022 [cited 2022 Nov 15]. Available from: https://www.cdc.gov/breastfeeding/data/reportcard.htm 30. Healthy Southern Nevada. Southern Nevada:: Indicators:: Mothers who Received Early Prenatal Care [Internet]. [cited 2022 Aug 3]. Available from: https://www.healthysouthernnevada.org/indicators/index/ view?indicatorId=325&localeTypeId=3&comparisonId=6639 31. Wright R. Baby-Friendly USA—About [Internet]. Baby-Friendly USA. [cited 2022 Nov 15]. Available from: https://www.babyfriendlyusa.org/about/ 32. WHO/UNICEF. WHO/UNICEF discussion paper: The extension of the 2025 maternal, infant and young child nutrition targets to 2030 [Internet]. UNICEF DATA. 2019 [cited 2022 Aug 3]. Available from: https://data.unicef.org/resources/who-unicef-discussion-paper-nutrition-targets/ 33. Victora CG, Huttly SR, Fuchs SC, Olinto MT. The role of conceptual frameworks in epidemiological analysis: a hierarchical approach. International Journal of Epidemiology. 1997 Feb 1; 26[1]:224–7. https://doi.org/10.1093/ije/26.1.224 PMID: 9126524 34. Hager ER, Quigg AM, Black MM, Coleman SM, Heeren T, Rose-Jacobs R, et al. Development and Validity of a 2-Item Screen to Identify Families at Risk for Food Insecurity. Pediatrics. 2010 Jul 1; 126[1]: e26–32. https://doi.org/10.1542/peds.2009-3146 PMID: 20595453 35. Kroenke K, Spitzer RL, Williams JBW. The Patient Health Questionnaire-2: Validity of a Two-Item Depression Screener. Medical Care. 2003; 41[11]:1284–92. https://doi.org/10.1097/01.MLR. 0000093487.78664.3C PMID: 14583691 36. Indicator Metadata Registry Details [Internet]. [cited 2023 Mar 16]. Available from: https://www.who.int/ data/gho/indicator-metadata-registry/imr-details/359 37. Coutinho LMS, Scazufca M, Menezes PR. Methods for estimating prevalence ratios in cross-sectional studies. Rev Sau´ de Pu´ blica. 2008 Dec; 42:992–8. PMID: 19009156 38. Hermanson Å, Åstrand LL. The effects of early pacifier use on breastfeeding: A randomised controlled trial. Women and Birth. 2020 Sep 1; 33[5]:e473–82. https://doi.org/10.1016/j.wombi.2019.10.001 PMID: 31704126 39. Nascimento EN, Leone C, Abreu LC de, Buccini G. Determinants of exclusive breast-feeding discontin- uation in southeastern Brazil, 2008–2013: a pooled data analysis. Public Health Nutrition. 2021 Jul; 24 [10]:3116–23. https://doi.org/10.1017/S1368980020003110 PMID: 32924912 40. USDA ERS—Definitions of Food Security [Internet]. [cited 2022 Dec 21]. Available from: https://www. ers.usda.gov/topics/food-nutrition-assistance/food-security-in-the-u-s/definitions-of-food-security/ 41. de Oliveira KHD, de Almeida GM, Gubert MB, Moura AS, Spaniol AM, Hernandez DC, et al. Household food insecurity and early childhood development: Systematic review and meta-analysis. Matern Child Nutr. 2020 Jul; 16[3]:e12967. https://doi.org/10.1111/mcn.12967 PMID: 32052571 42. Pedroso J, Herkrath FJ, Buccini G, Venancio SI, Pe´ rez-Escamilla R, Gubert MB. Maternal Mental Health Mediates the Relationship Between Household Food Insecurity and Early Child Development in Brazil. Food Nutr Bull. 2021 Mar 1; 42[1]:77–90. https://doi.org/10.1177/0379572120987815 PMID: 33878902 43. Batista CLC, Rodrigues VP, Ribeiro VS, Nascimento MDSB. Nutritive and non-nutritive sucking pat- terns associated with pacifier use and bottle-feeding in full-term infants. Early Human Development. 2019 May; 132:18–23. https://doi.org/10.1016/j.earlhumdev.2019.03.007 PMID: 30928831 44. Sipsma HL, Kornfeind K, Kair LR. Pacifiers and Exclusive Breastfeeding: Does Risk for Postpartum Depression Modify the Association? J Hum Lact. 2017 Nov 1; 33[4]:692–700. https://doi.org/10.1177/ 0890334417725033 PMID: 28841401 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 17 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada 45. Pe´ rez-Escamilla R, Cunningham K, Moran VH. COVID-19 and maternal and child food and nutrition insecurity: a complex syndemic. Matern Child Nutr. 2020 Jun 16; 16[3]:e13036. https://doi.org/10.1111/ mcn.13036 PMID: 32458574 46. Jones KM, Power ML, Queenan JT, Schulkin J. Racial and Ethnic Disparities in Breastfeeding. Breast- feed Med. 2015 May 1; 10[4]:186–96. https://doi.org/10.1089/bfm.2014.0152 PMID: 25831234 47. Holdren S, Fair C, Lehtonen L. A qualitative cross-cultural analysis of NICU care culture and infant feed- ing in Finland and the U.S. BMC Pregnancy and Childbirth. 2019 Oct 10; 19[1]:345. 48. Segura-Pe´rez S, Hromi-Fiedler A, Adnew M, Nyhan K, Pe´ rez-Escamilla R. Impact of breastfeeding interventions among United States minority women on breastfeeding outcomes: a systematic review. Int J Equity Health. 2021 Mar 6; 20[1]:72. https://doi.org/10.1186/s12939-021-01388-4 PMID: 33676506 49. Anstey EH, Chen J, Elam-Evans LD, Perrine CG. Racial and Geographic Differences in Breastfeeding —United States, 2011–2015. MMWR Morb Mortal Wkly Rep. 2017 Jul 14; 66:723–7. https://doi.org/10. 15585/mmwr.mm6627a3 PMID: 28704352 50. Cherian T, Crookes DM, Suglia SF. Association of Maternal Nativity Status and Race/Ethnicity on Emergency Department Utilization Among Children in Vulnerable Families. Pediatr Emerg Care. 2021 Dec 1; 37[12]:e1549–54. https://doi.org/10.1097/PEC.0000000000002113 PMID: 32398598 51. Ford CL, Airhihenbuwa CO. Critical Race Theory, Race Equity, and Public Health: Toward Antiracism Praxis. Am J Public Health. 2010 Apr; 100[Suppl 1]:S30–5. https://doi.org/10.2105/AJPH.2009.171058 PMID: 20147679 52. Mauch CE, Scott JA, Magarey AM, Daniels LA. Predictors of and reasons for pacifier use in first-time mothers: an observational study. BMC Pediatrics. 2012 Jan 19; 12[1]:7. https://doi.org/10.1186/1471- 2431-12-7 PMID: 22257532 53. Giugliani ERJ, Nunes LM, Issler RMS, Santo LC do E, de Oliveira LD. Involvement of maternal grand- mother and teenage mother in intervention to reduce pacifier use: a randomized clinical trial. Jornal de Pediatria [Versão em Português]. 2019 Mar 1; 95[2]:166–72. https://doi.org/10.1016/j.jped.2017.12.011 PMID: 29447832 54. Batista CLC, Ribeiro VS, Nascimento M do DSB, Rodrigues VP. Association between pacifier use and bottle-feeding and unfavorable behaviors during breastfeeding. Jornal de Pediatria. 2018 Nov; 94 [6]:596–601. https://doi.org/10.1016/j.jped.2017.10.005 PMID: 29136496 55. Zimmerman E, Thompson K. Clarifying nipple confusion. Journal of Perinatology. 2015 Nov 1; 35 [11]:895–900. https://doi.org/10.1038/jp.2015.83 PMID: 26181720 56. Harrison D, Reszel J, Bueno M, Sampson M, Shah VS, Taddio A, et al. Breastfeeding for procedural pain in infants beyond the neonatal period. Cochrane Database Syst Rev. 2016 Oct 28; 10:CD011248. https://doi.org/10.1002/14651858.CD011248.pub2 PMID: 27792244 57. Kramer MS, Barr RG, Dagenais S, Yang H, Jones P, Ciofani L, et al. Pacifier Use, Early Weaning, and Cry/Fuss BehaviorA Randomized Controlled Trial. JAMA. 2001 Jul 18; 286[3]:322–6. https://doi.org/10. 1001/jama.286.3.322 PMID: 11466098 58. Barca L. Toward a speech-motor account of the effect of Age of Pacifier Withdrawal. J Commun Disord. 2021; 90:106085. https://doi.org/10.1016/j.jcomdis.2021.106085 PMID: 33550069 59. Office of Disease Prevention and Health Promotion. Health Equity in Healthy People 2030—Healthy People 2030 \| health.gov [Internet]. [cited 2022 Aug 4]. Available from: https://health.gov/ healthypeople/priority-areas/health-equity-healthy-people-2030 60. Rocha CR, Verga KE, Sipsma HL, Larson IA, Phillipi CA, Kair LR. Pacifier Use and Breastfeeding: A Qualitative Study of Postpartum Mothers. Breastfeeding Medicine. 2020 Jan; 15[1]:24–8. https://doi. org/10.1089/bfm.2019.0174 PMID: 31859530 61. Chapman DJ, Morel K, Anderson AK, Damio G, Pe´rez-Escamilla R. Breastfeeding peer counseling: from efficacy through scale-up. J Hum Lact. 2010 Aug; 26[3]:314–26. https://doi.org/10.1177/ 0890334410369481 PMID: 20715336 62. Haroon S, Das JK, Salam RA, Imdad A, Bhutta ZA. Breastfeeding promotion interventions and breast- feeding practices: a systematic review. BMC Public Health. 2013;13 Suppl 3[Suppl 3]:S20. https://doi. org/10.1186/1471-2458-13-S3-S20 PMID: 24564836 63. Pe´ rez-Escamilla R, Tomori C, Herna´ndez-Cordero S, Baker P, Barros AJD, Be´ gin F, et al. Breastfeed- ing: crucially important, but increasingly challenged in a market-driven world. The Lancet. 2023 Feb 11; 401[10375]:472–85. 64. Rollins N, Piwoz E, Baker P, Kingston G, Mabaso KM, McCoy D, et al. Marketing of commercial milk for- mula: a system to capture parents, communities, science, and policy. The Lancet. 2023 Feb 11; 401 [10375]:486–502. https://doi.org/10.1016/S0140-6736(22)01931-6 PMID: 36764314 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 18 / 19 PLOS ONEFactors associated with pacifier use among six-months old infants in Nevada 65. Venancio SI, Saldiva SRDM, Escuder MML, Giugliani ERJ. The Baby-Friendly Hospital Initiative shows positive effects on breastfeeding indicators in Brazil. J Epidemiol Community Health. 2012 Oct; 66 [10]:914–8. https://doi.org/10.1136/jech-2011-200332 PMID: 22080818 66. Nevada S. Southern Nevada:: Demographics:: County:: Clark [Internet]. [cited 2022 Dec 21]. Available from: https://www.healthysouthernnevada.org/index.php?module=DemographicData&controller= index&action=index 67. Singh S, Sagar R. A critical look at online survey or questionnaire-based research studies during COVID-19. Asian J Psychiatr. 2021 Nov; 65:102850. https://doi.org/10.1016/j.ajp.2021.102850 PMID: 34534919 68. Pacifiers and Breastfeeding. JAMA. 2001 Jul 18; 286[3]:374. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285097 April 27, 2023 19 / 19 PLOS ONE
10.1371_journal.ppat.1010767	RESEARCH ARTICLE TNF licenses macrophages to undergo rapid caspase-1, -11, and -8-mediated cell death that restricts Legionella pneumophila infection Tzvi Y. Pollock1, Vı´ctor R. Va´ zquez Marrero1, Igor E. Brodsky2, Sunny ShinID 1* 1 Department of Microbiology, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania, United States of America, 2 Department of Pathobiology, University of Pennsylvania School of Veterinary Medicine, Philadelphia, Pennsylvania, United States of America * sunshin@pennmedicine.upenn.edu Abstract The inflammatory cytokine tumor necrosis factor (TNF) is necessary for host defense against many intracellular pathogens, including Legionella pneumophila. Legionella causes the severe pneumonia Legionnaires’ disease and predominantly affects individuals with a suppressed immune system, including those receiving therapeutic TNF blockade to treat autoinflammatory disorders. TNF induces pro-inflammatory gene expression, cellular prolif- eration, and survival signals in certain contexts, but can also trigger programmed cell death in others. It remains unclear, however, which of the pleiotropic functions of TNF mediate control of intracellular bacterial pathogens like Legionella. In this study, we demonstrate that TNF signaling licenses macrophages to die rapidly in response to Legionella infection. We find that TNF-licensed cells undergo rapid gasdermin-dependent, pyroptotic death down- stream of inflammasome activation. We also find that TNF signaling upregulates compo- nents of the inflammasome response, and that the caspase-11-mediated non-canonical inflammasome is the first inflammasome to be activated, with caspase-1 and caspase-8 mediating delayed pyroptotic death. We find that all three caspases are collectively required for optimal TNF-mediated restriction of bacterial replication in macrophages. Furthermore, caspase-8 is required for control of pulmonary Legionella infection. These findings reveal a TNF-dependent mechanism in macrophages for activating rapid cell death that is collec- tively mediated by caspases-1, -8, and -11 and subsequent restriction of Legionella infection. Author summary The immune signaling molecule tumor necrosis factor (TNF) is critical for defense against many bacterial infections. In the absence of TNF, opportunistic bacterial pathogens such as Legionella pneumophila, which causes the severe pneumonia Legionnaires’ disease, readily invade and replicate inside innate immune cells known as macrophages. While it is clear that TNF signaling is required for protection of macrophages against pathogens like L. pneumophila, the mechanisms by which this signaling protects cells from infection a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Pollock TY, Va´zquez Marrero VR, Brodsky IE, Shin S (2023) TNF licenses macrophages to undergo rapid caspase-1, -11, and -8-mediated cell death that restricts Legionella pneumophila infection. PLoS Pathog 19(6): e1010767. https:// doi.org/10.1371/journal.ppat.1010767 Editor: Raphael H. Valdivia, Duke University, UNITED STATES Received: July 28, 2022 Accepted: May 25, 2023 Published: June 6, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.ppat.1010767 Copyright: © 2023 Pollock et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting information files. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 1 / 26 PLOS PATHOGENSFunding: This work was supported by National Institutes of Health (NIH)/National Institute of Allergy and Infectious Diseases (NIAID) grants: AI118861 (S.S.), AI123243 (S.S.), and AI151476 (S.S. and T.Y.P.); AI128530 (I.E.B.), AI135421 (I.E. B), and AI139102 (I.E.B.); AI140508 (T.Y.P.) and AI141393 (T.Y.P.); the Linda Pechenik Montague Investigator Award from the University of Pennsylvania Perelman School of Medicine (S.S.), the Burroughs-Wellcome Fund Investigators in the Pathogenesis of Infectious Disease Award (S.S. and I.E.B.), and the National Science Foundation Graduate Research Fellowship DGE-1650114 (V.R. V.M). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors declare that no competing interests exist. TNF licenses macrophages to die and restrict Legionella replication remain unclear. Here, we show that one of the methods by which TNF protects macro- phages from L. pneumophila infection is by promoting the death of infected cells. TNF sig- naling can lead to the production of anti-microbial factors and inflammation, but it can also prompt cells to engage programmed forms of cell death. We demonstrate that in L. pneumophila infection, TNF specifically licenses cells to more rapidly and robustly acti- vate a kind of cell death called pyroptosis, which causes inflammation and warns nearby uninfected cells. We additionally demonstrate several cooperative pathways downstream of TNF by which this death occurs. Our work shows that TNF protects macrophage popu- lations from L. pneumophila by licensing cells to rapidly self-sacrifice upon infection and limit the replicative niche of the bacteria. Introduction The innate immune system is generally capable of preventing and controlling infection. How- ever, pathogens are able to establish infection when key immune factors are evaded or deficient [1]. Inflammatory cytokine signaling constitutes one such category of immune factors. The loss or inhibition of cytokine signaling often results in greatly increased risk of infection. Tumor Necrosis Factor (TNF) is one such critical mediator of host defense against intracellular pathogens [2–6]. Therapeutic blockade of TNF in the context of autoinflammatory diseases results in greatly increased rates of infection [7,8]. While the downstream effects of TNF sig- naling are well-established in regulating cell survival, proliferation, and death in many sterile contexts, it remains unclear how TNF controls intracellular pathogens. Among the bacteria which require TNF for efficient immune clearance is the gram-nega- tive, facultative intracellular bacterial pathogen Legionella pneumophila. The etiologic agent of the severe pneumonia Legionnaire’s disease, L. pneumophila is an opportunistic pathogen that causes disease predominantly in immunocompromised hosts [9,10]. L. pneumophila replicates within host macrophages by using a Dot/Icm type IV secretion system (T4SS) to translocate more than 300 bacterial effectors [11–15]. T4SS effectors modulate numerous host processes, including membrane trafficking and protein translation [16–19]. L. pneumophila infection is typically well-controlled by the innate immune system [20,21]. This immune protection, how- ever, is severely attenuated in the absence of TNF signaling in macrophages and mice or in autoimmune patients receiving TNF blockade [7,8,20,22–28]. A key aspect of the innate immune response to L. pneumophila involves inflammasomes, multi-protein cytosolic complexes which assemble in response to pathogenic insult and medi- ate downstream inflammatory signaling [20,29–32]. Delivery of Legionella flagellin into the host cytosol triggers the NAIP5/NLRC4 inflammasome in mice, activating the cysteine prote- ase caspase-1 to induce an inflammatory form of cell death known as pyroptosis, processing and release of interleukin-1 family cytokines, and restriction of L. pneumophila replication [33–35]. L. pneumophila T4SS activity further activates the NLRP3 inflammasome, while cyto- solic detection of L. pneumophila LPS activates caspase-11 to mediate the “non-canonical” inflammasome [36,37]. Inflammation and cell death mediated by inflammasomes restrict bac- terial infection, in part by limiting the replicative niche of the pathogen and promoting cyto- kine production by bystander immune cells [38–40]. TNF is required for optimal NAIP5/ NLRC4-mediated restriction of L. pneumophila replication within macrophages. However, TNF can also mediate restriction of L. pneumophila in the absence of NAIP5/NLRC4 inflam- masome activation via a proposed cell death-independent mechanism that invokes the activity of caspases other than caspases-1 and -11 [27]. Thus, it is unclear whether TNF-mediated PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 2 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication control of L. pneumophila replication observed in the absence of the NAIP5/NLRC4 inflam- masome is due to caspase-mediated cell death or other cellular fates. TNF signaling in infectious and non-infectious settings can lead to multiple outcomes. Upon binding of TNF to its cognate receptors, TNF Receptor 1 (TNFR1) or TNF Receptor 2 (TNFR2), the adaptor protein TRADD is recruited to the cytosolic domain of the receptor to allow formation of functionally distinct signaling complexes [3,4,41]. Complex I mediates immune activation, inflammation, and production of anti-apoptotic survival factors through downstream NF-κB and MAPK signaling [4,42]. In contrast, Complex II can engage two dis- tinct forms of programmed cell death: extrinsic apoptosis mediated by the cysteine protease caspase-8 cleaving and activating the executioner caspases-3 and -7 under situations when NF- κB and MAPK signaling are inhibited [4,5,43], or necroptosis mediated by receptor-interact- ing protein kinase 3 (RIPK3) when caspase-8 activity is blocked or absent [44–46]. Addition- ally, caspase-8 promotes inflammatory gene expression independently of its cell death function [43,47–49], as well as promotes caspase-1 activation and compensates for its absence in inflammasome activation to process gasdermin-D and IL-1 family cytokines [50–54]. In this study, we sought to define the mediators of bacterial restriction that function inde- pendently of the NAIP/NLRC4 inflammasome in the context of L. pneumophila infection. Our data demonstrate that TNF signaling through TNFR1 licenses macrophages to undergo more rapid and robust cell death in response to L. pneumophila infection. We found that this cell death was associated with gasdermin-dependent loss of membrane integrity, indicating that this cell death is pyroptotic. Our findings indicate that TNF licenses cells to upregulate and rapidly activate the non-canonical caspase-11 inflammasome in response to L. pneumophila infection. We further found that caspase-1 and caspase-8 contributed to delayed cell death in the absence of caspase-11. In addition, we found that caspase-8 contributed to cell death inde- pendently of its ability to execute extrinsic apoptosis. Moreover, caspase-8 activity was required for clearance of pulmonary L. pneumophila infection. These data together indicate that TNF signaling during L. pneumophila infection restricts bacterial replication by licensing macrophages to rapidly undergo caspase-1, -8, and -11 inflammasome activation and pyropto- sis, thereby eliminating the replicative niche of the bacteria. Results TNFR1 signaling is required for restriction of L. pneumophila infection and licenses cells to rapidly undergo cell death Macrophages infected with L. pneumophila rapidly respond to injected bacterial flagellin via the NAIP5/NLRC4 inflammasome, initiating pyroptotic cell death and eliminating the replica- tive niche of the bacteria [33,55]. Maximal NAIP5/NLRC4-dependent restriction requires TNF signaling [20]. However, TNF also restricts intracellular replication of L. pneumophila within macrophages independently of flagellin and the NAIP inflammasome [27], implying that TNF-dependent control of L. pneumophila likely involves multiple downstream innate responses. For the entirety of this study, we infected bone marrow-derived macrophages (BMDMs) with mutant L. pneumophila deficient for flagellin (ΔflaA) in order to bypass the NAIP5 inflammasome response and specifically examine the NAIP5/NLRC4-independent role of TNF signaling in control of infection. Consistent with previous studies, we observed that BMDMs isolated from wild-type (WT) C57BL/6 mice harbor between 10-fold and 100-fold replication of flagellin-deficient bacteria over the course of 72 hours, while BMDMs from Tnf-/- mice demonstrated significantly higher levels of bacteria at 48 and 72 hours post- infection (Fig 1A). Of note, TNF-dependent restriction was apparent only after 24 hours post- infection, indicating that endogenous TNF produced in response to the initial infection was PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 3 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication Fig 1. TNFR1 signaling is required for restriction of L. pneumophila replication in macrophages and licenses cells to rapidly undergo cell death. (A,C-E) WT, Tnf-/-, (A, C) and Tnfr1-/- (D-E) BMDMs were infected with ΔflaA L. pneumophila at MOI = 1. (A,D) The fold change in CFUs were quantified at 24, 48, and 72 hours post-infection. (C,E) BMDMs were either primed with 10 ng/mL recombinant murine TNF or PBS mock control for 16 hours and infected for 48 hours. (B) WT and Tnf-/- BMDMs were infected with non-replicating ΔflaA L. pneumophila at MOI = 10 for 12 hours. Cytokine release was measured every 2 hours using ELISA. (F-H) WT, Tnf-/- (G, H), and Tnfr1-/- (H) BMDMs were infected with non-replicating ΔflaA L. pneumophila at MOI = 10 (E) or 50 (F-H) for 8 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 4 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication hours. Cytotoxicity was measured using LDH release assay (F,H) or PI uptake assay (G). BMDMs were primed with 10 ng/mL recombinant murine rTNF or PBS mock control for 16 hours. Graphs show the mean ± SEM of triplicate wells. * is p<0.05, is p<0.01, * is p<0.001, and **** is p<0.0001 by 2-way ANOVA with Sˇı´da´k’s post-test (A-D, F-H) or Student’s t-test (E). NS is not significant. Data shown are representative of at least three independent experiments. https://doi.org/10.1371/journal.ppat.1010767.g001 responsible for mediating restriction of L. pneumophila at later timepoints. Indeed, infection of BMDMs with ΔflaA L. pneumophila resulted in gradual secretion of TNF, plateauing at about 10 to 12 hours post-infection (Fig 1B). To mimic this endogenous TNF produced follow- ing initial infection and subsequent restriction, we primed cells with recombinant TNF (rTNF) for 16 hours prior to infection. The bacterial replication observed in WT BMDMs was completely abrogated when cells were primed with rTNF (Fig 1C). Likewise, rTNF priming limited bacterial replication in Tnf-/- BMDMs. TNF can signal through either TNF receptor 1 (TNFR1) or 2 (TNFR2), and each are known to contribute positively to host defense during pulmonary L. pneumophila infection, with TNFR1 promoting an inflammatory environment and restriction of L. pneumophila infection within the lung and TNFR2 limiting immunopathology [27,56]. We infected Tnfr1-/- and Tnfr1-/-Tnfr2-/- BMDMs, as well as WT and Tnf-/- BMDMs as controls. Bacterial replication was greatly increased in the absence of TNFR1, consistent with our observation in Tnf-/- BMDMs (Fig 1D). Crucially, priming with rTNF was able to restrict bacterial replication in Tnf-/- BMDMs, but did not restrict bacterial replication in Tnfr1-/- BMDMs (Fig 1E). Doubly deficient Tnfr1-/-Tnfr2-/- BMDMs demonstrated no additional defect in bacterial control rela- tive to singly deficient Tnfr1-/- BMDMs (S1A Fig). Thus, TNF signals chiefly through TNFR1 to mediate restriction of L. pneumophila replication in BMDMs, in agreement with previous findings [27]. We then sought to determine whether TNF signaling restricts L. pneumophila infection via increased production of anti-microbial molecules. TNF signaling through NF-κB and MAPK pathways upregulates production of anti-microbial molecules such as reactive nitrogen and oxygen species (RNS, ROS) [24,57]. BMDMs lacking inducible nitric oxide synthase (Nos2-/-), the enzyme responsible for reactive nitrogen production, showed no defect in control of L. pneumophila replication relative to WT BMDMs (S1B Fig). We also observed that BMDMs deficient in NADPH oxidase 2 (Cybb-/-), the enzyme responsible for reactive oxygen produc- tion in the lysosome, did not show a defect in controlling L. pneumophila replication, in agree- ment with previous findings [27]. We additionally observed no defect in TNF secretion in Nos2-/- or Cybb-/- cells (S1C Fig). While we saw no basal defect in control of L. pneumophila replication in the absence of NADPH oxidase-derived ROS, we did observe a decreased ability of Cybb-/- cells to restrict L. pneumophila following exogenous TNF priming (S1D Fig). This is in line with studies that have shown a role for ROS and TNF in control of L. pneumophila infection [27]. Even in the absence of NADPH oxidase, however, we observed a significant decrease in bacterial growth in the context of TNF priming (S1D Fig). These data suggested that neither RNS nor NADPH oxidase-derived ROS are entirely responsible for TNFR1-me- diated control of L. pneumophila replication and prompted us to investigate alternative mecha- nisms of TNF-mediated restriction. Cell-extrinsic TNF signaling frequently mediates cell death in response to foreign insults [58]. Cell death in response to infection can restrict intracellular bacterial replication by limit- ing the replicative niche of the pathogen [38]. We therefore sought to determine whether TNF promotes cell death during L. pneumophila infection. To avoid any confounding effect of bac- terial replication, we infected cells with thymidine auxotrophic bacteria incapable of replicat- ing in the absence of exogenous thymidine. We measured cell death by monitoring both release of the cytosolic protein lactate dehydrogenase (LDH) into the extracellular space, as PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 5 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication well as uptake of the membrane-impermeable dye propidium iodide (PI). We observed LDH release (Fig 1F) and PI uptake (Fig 1G) in L. pneumophila-infected WT and Tnf-/- BMDMs by 6–8 hours following infection. WT and Tnf-/- BMDMs, when primed with rTNF prior to L. pneumophila infection, instead demonstrated hallmarks of death as early as 2–4 hours post infection, a significant acceleration relative to unprimed controls (Fig 1F and 1G). In accor- dance with our data indicating a requirement for TNFR1 signaling in control of bacterial restriction, we additionally observed that accelerated death in rTNF-primed cells was reliant on TNFR1 signaling (Fig 1H). These data indicate that TNF priming through TNFR1 licenses cells to rapidly undergo cell death and limit bacterial replication in response to L. pneumophila infection. The Legionella type IV secretion system triggers TNF-licensed activation of the caspase-1 and -11 inflammasomes Infection of macrophages by L. pneumophila and subsequent injection via the T4SS results in assembly and activation of the inflammasome in the host cell cytosol [20,33,55,59]. Cytosolic recognition of L. pneumophila flagellin specifically is mediated by the NAIP5/NLRC4 inflam- masome, and results in efficient restriction of intracellular replication [33,55]. Among the downstream effects of inflammasome activation are the processing of IL-1 family cytokines and the assembly of gasdermin pores to execute an inflammatory form of cell death known as pyroptosis [60]. Even during infection with flagellin-deficient L. pneumophila, which does not induce NAIP5/NLRC4 activation, we observed significant induction of L. pneumophila-trig- gered cell death following TNF priming (Fig 1F–1H). Importantly, using a ΔflaAΔdotA strain, we found that PI uptake and LDH release following L. pneumophila infection is entirely depen- dent on the T4SS, indicating that rapid cell death following TNF priming requires detection of T4SS activity (Fig 2A and S2 Fig). We therefore sought to determine what form of cell death is potentiated by TNF priming upon L. pneumophila infection. Cell death downstream of inflammasome detection of bacterial ligands is potentiated by inflammatory caspases. In mice, caspase-1 is activated within a canon- ical inflammasome that typically contains an NLR and an adaptor protein [30–32]. Caspase- 11, meanwhile, is activated within a noncanonical inflammasome in response to direct sensing of cytosolic LPS [37,61]. Notably, we observed that, within the first 8 hours of infection, TNF- licensed cell death as measured by PI uptake and LDH release depended on caspases-1 and -11 (Fig 2B and 2C). Likewise, we observed that release of IL-1 family cytokines following L. pneu- mophila infection was significantly increased in the context of TNF priming (Fig 2D). How- ever, in the absence of both caspases-1 and -11, we found that IL-1 release was severely attenuated (Fig 2D). These data suggest that TNF licenses cells to undergo pyroptosis and secrete IL-1 family cytokines rapidly in response to L. pneumophila T4SS activity in a caspase- 1/11 dependent manner. We then examined whether caspase-1 and -11 are required for TNF-mediated restriction of bacterial replication in BMDMs. We observed that, similarly to WT and Tnf-/- BMDMs, Casp1-/-Casp11-/- BMDMs more robustly restrict bacterial replication when primed with rTNF (Fig 2E). It is worth noting that TNF-mediated restriction is slightly less robust in Casp1-/-- Casp11-/- BMDMs relative to WT or Tnf-/- BMDMs, though not to the point of statistical sig- nificance. In addition, we observed that at later timepoints, Casp1-/-Casp11-/- BMDMs still underwent cell death, as measured by PI uptake (Fig 2C). These data suggest that, while TNF licenses cells to rapidly engage caspase-1/11-mediated death, IL-1 release, and restriction of L. pneumophila infection, other caspase-1/11-independent factors contribute to TNF-mediated cell death and restriction of bacterial replication. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 6 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication Fig 2. Legionella type IV secretion system activity triggers TNF-licensed activation of the caspase-1 and -11 inflammasomes. (A) WT and Tnf-/- BMDMs were primed with either 10 ng/mL rTNF or PBS mock control for 16 hours prior to infection with non-replicating ΔflaA L. pneumophila or ΔdotA bacteria lacking a functional T4SS at MOI = 10. (B-C) WT and Casp1-/-Casp11-/- BMDMs were primed with either 10 ng/mL rTNF or PBS mock control for 16 hours and then infected for 8 hours with non-replicating ΔflaA L. pneumophila at MOI = 50 (B) or 10 (C). Cytotoxicity was measured by PI uptake assay (A,C) or LDH release assay (B). (D) WT and Casp1-/-Casp11-/- BMDMs were primed with 10 ng/mL rTNF for 16 hours then left uninfected or infected with non- replicating ΔflaA L. pneumophila at MOI = 50. Cytokine release measured by ELISA. (E) WT, Tnf-/-, and Casp1-/-Casp11-/- BMDMs were primed with either 10 ng/mL rTNF or PBS mock control for 16 hours prior to infection with replicating ΔflaA L. pneumophila at MOI = 1. Fold change in CFUs was quantified at 72 hours post-infection. Graphs show the mean ± SEM of triplicate wells. * is p<0.05, is p<0.01, * is p<0.001, and **** is p<0.0001 by 2-way ANOVA with Sˇı´da´k’s post-test (A-D) or Tukey HSD (E). NS is not significant. Data shown are representative of at least three independent experiments. https://doi.org/10.1371/journal.ppat.1010767.g002 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 7 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication TNF priming licenses cells to activate the non-canonical inflammasome during L. pneumophila infection Given the pyroptotic nature of TNF-licensed cell death we observed in response to non-flagel- lated L. pneumophila, we sought to determine which inflammasomes are responsible. The NLRP3 inflammasome oligomerizes in response to diverse cytosolic triggers, including reac- tive oxygen species, potassium efflux, and lysosomal damage signals [38,62], and contributes to the inflammasome response to L. pneumophila in the absence of TNF priming [36,37]. We observed significant attenuation of IL-1β release in MCC950-treated and TNF-primed BMDMs infected with L. pneumophila, indicating that the NLRP3 inflammasome is required for IL-1β release in TNF-primed cells (Fig 3A). Notably, we observed no decrease in IL-1α release or in cell death in TNF-primed BMDMs treated with the NLRP3-specific inhibitor, MCC950, compared to mock-treated cells (Fig 3A and 3B). These data suggest that the NLRP3 inflammasome contributes to IL-1β release but is not required for cell death and IL-1α release in TNF-licensed cells, indicating that another inflammasome is activated in TNF-primed BMDMs following L. pneumophila infection. The non-canonical inflammasome, formed by caspase-11, is able to directly mediate pyrop- tosis and release of IL-1α, but cannot independently process IL-1β, and thus requires second- ary NLRP3 activation to mediate release of active IL-1β during L. pneumophila infection following LPS-priming [36,37]. We thus tested whether the non-canonical inflammasome is involved in the TNF-licensed response to L. pneumophila. Indeed, we observed a significant decrease in TNF-licensed cell death early during infection, as well as significant attenuation of both IL-1α and IL-1β release, in Casp11-/- BMDMs relative to wild-type controls (Fig 3C–3E). These data implicate the caspase-11 non-canonical inflammasome as a downstream target of TNF licensing in L. pneumophila infection. We next sought to determine how TNF may be enhancing caspase-11 inflammasome acti- vation during infection. Activation of NF-κB and MAP kinase signaling and subsequent gene expression, as well as caspase-8-mediated inflammatory gene expression, are mediated by upstream TNF receptor signaling [3,6,41,63,64]. We hypothesized that TNF licensing of non- canonical inflammasome activation involves upregulated expression of non-canonical inflam- masome components such as caspase-11, gasdermin D, and IL-1 family cytokines. We used quantitative RT-PCR to determine the effect of TNF priming on transcription of inflamma- some factors prior to and during L. pneumophila infection. As early as 2 hours following TNF treatment, we observed statistically significant increases in Il1a, Il1b, and Il18 mRNA levels (Fig 3F). We additionally observed increased expression of Casp11, though this increase did not achieve statistical significance (Fig 3F). TNF-mediated increase in Casp11 transcription can still be observed at 16 hours following priming (at the time of infection), while Il1a and Il1b transcription fades by that time (S3A Fig). Additional inflammasome-related genes, such as Casp8 and Casp1, did not show an increase in mRNA expression in TNF-primed cells (S3B Fig). We observed a substantial increase in the level of caspase-11 protein in TNF-primed cel- lular lysates prior to L. pneumophila infection, suggesting that TNF licenses cells to react more rapidly to infection by upregulating caspase-11 levels (Fig 3G). Accordingly, TNF priming also resulted in release of cleaved caspase-11 into the supernatant of infected cells (Fig 3G). These data collectively indicate that TNF priming leads to increased caspase-11 expression and acti- vation during infection. As the non-canonical inflammasome detects cytosolic LPS, we aimed to elucidate how L. pneumophila is being detected by the non-canonical inflammasome, and whether this detec- tion itself is enhanced by TNF priming. Guanylate Binding Proteins (GBPs) are activated downstream of IFN signaling, subsequently localizing to and permeabilizing pathogen- PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 8 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication Fig 3. TNF priming licenses cells to activate the non-canonical inflammasome during L. pneumophila infection. (A-B) Tnf-/- BMDMs were either primed with 10 ng/mL rTNF or PBS mock control for 16 hours prior to infection with non-replicating ΔflaA L. pneumophila at MOI = 50 for 6 hours. Cells were additionally either treated with DMSO vehicle or 1 uM MCC950 for 1 hour prior to infection. Cytokine release was measured by ELISA and cytotoxicity was determined by LDH release assay. (C-D) WT and Casp11-/- BMDMs primed with either 10 ng/mL rTNF or PBS mock control for 16 hours prior to infection with non-replicating ΔflaA L. pneumophila at MOI = 50 (C) or 10 (D). (E) WT and Casp11-/- BMDMs were primed with 10 ng/mL PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 9 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication rTNF for 16 hours prior to infection with non-replicating ΔflaA L. pneumophila at MOI = 50. Cytokine release was measured by ELISA. (F) WT BMDMs were treated with either 10 ng/mL rTNF or PBS for 2 hours, then supernatants were collected and cells lysed for immunoblot and qPCR. (G) WT BMDMs were primed with either 10 ng/mL rTNF or PBS mock control for 16 hours prior to infection with non-replicating ΔflaA L. pneumophila at MOI = 50. Supernatants were collected and BMDDMs lysed at times indicated. Graphs show the mean ± SEM of triplicate wells. * is p<0.05, is p<0.01, * is p<0.001, and **** is p<0.0001 by 2-way ANOVA with Sˇı´da´k’s post-test (A-E) or Student’s t-test (F). NS is not significant. Data shown are representative of at least two independent experiments. https://doi.org/10.1371/journal.ppat.1010767.g003 containing vacuoles in order to introduce bacterial LPS to the cytosol for caspase-11 detection, as well as binding bacterial surfaces directly [65,66]. GBPs promote inflammasome responses to L. pneumophila in mouse macrophages downstream of IFN signaling [67,68]. However, BMDMs from GbpChr3-/- mice deficient for the six GBPs found on murine chromosome 3 did not exhibit a significant loss of TNF-licensed cell death (S4A Fig) or bacterial restriction (S4B Fig). This suggests that in the context of TNF priming, L. pneumophila activates caspase-11 independently of the GBPs on chromosome 3. Altogether, these data indicate that TNF prim- ing licenses BMDMs to engage the caspase-11-mediated non-canonical inflammasome by upregulating inflammasome components in advance of infection. Caspase-8 contributes to cell death independently of its autocleavage downstream of TNF signaling While TNF-licensed cell death is significantly attenuated in the absence of caspase-1 and cas- pase-11, we observed that, even in the absence of these caspases, TNF-primed macrophages exhibited cell death at later timepoints (Fig 2C). Prior studies revealed that other inflamma- some stimuli can induce cells to undergo delayed pyroptosis that requires caspase-8 in the absence of canonical pyroptotic mediators [50,51,69,70]. We therefore investigated the possi- ble role of caspase-8 in compensatory or redundant mechanisms of cell death downstream of TNF signaling in L. pneumophila infection. TNF triggers caspase-8 mediated cell death in the absence or inhibition of NF-κB-mediated survival signals [3–6]. Thus, we sought to determine whether TNF priming leads to caspase-8-dependent cell death following L. pneumophila infec- tion. Mice deficient in caspase-8 signaling experience uncontrolled activation of RIPK3-me- diated necroptosis, a form of cell death that eliminates cells lacking caspase-8 activity, and thus are embryonically lethal [44–46,71]. To address the role of caspase-8, we therefore used BMDMs lacking both RIPK3 and CASP8. BMDMs from Ripk3-/-Casp8-/- mice exhibited signif- icantly less TNF-licensed cell death following L. pneumophila-infection relative to Ripk3-/- or WT controls (Fig 4A). We then aimed to determine whether the caspase-8-mediated death we observe is necessary for restriction of L. pneumophila replication. Much like in caspase-1/11 deficiency, we still observed robust TNF priming-mediated restriction of bacterial replication in the absence of caspase-8 (Fig 4B). Similarly to Casp1-/-Casp11-/- BMDMs, we observed less robust TNF-mediated restriction in Ripk3-/-Casp8-/- BMDMs relative to controls, though not to the point of statistical significance. These data suggest that TNF signaling poises cells to engage caspase-8-mediated cell death in response to L. pneumophila infection, and that this pathway contributes to but is not required for bacterial restriction. We then sought to dissect which downstream pathways are engaged by caspase-8 in TNF- primed BMDMs to mediate cell death following infection. During extrinsic apoptosis, caspase- 8 is recruited by RIPK1, at which point caspase-8 dimerizes and auto-cleaves before cleaving downstream apoptotic substrates caspase-3 and caspase-7 [72]. Independently of auto-cleav- age, caspase-8 also regulates inflammatory gene expression downstream of TLR signaling [43,47–49]. Caspase-8 can additionally compensate for caspase-1 in the NAIP/NLRC4 and NLRP3 inflammasomes, though it is unknown whether this is dependent on auto-cleavage PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 10 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication Fig 4. Caspase-8 mediates optimal cell death independently of its autocleavage downstream of TNF signaling. (A-B) WT (B), Ripk3-/-, Ripk3-/-Casp8-/-, and Ripk3-/-Casp8DA BMDMs were primed with either 10 ng/mL rTNF or PBS mock control for 16 hours prior to infection with non-replicating (A) or replicating (B) ΔflaA L. pneumophila at MOI = 10 (A) or 1 (B) for 8 (A) or 72 (B) hours. (C) Ripk3-/-, Ripk3-/-Casp8-/-, and Ripk3-/-Casp8DA BMDMs were primed with 10 ng/mL rTNF for 16 hours prior to infection with non-replicating ΔflaA L. pneumophila at MOI = 10 for 8 hours. (D) BMDMs were primed with either 10 ng/mL rTNF or PBS mock control for 16 hours prior to infection with non-replicating ΔflaA L. pneumophila at MOI = 50 for 6 hours. Caspase-3/7 activity was measured with the caspase GLO assay. Graphs show the mean ± SEM of triplicate wells. * is p<0.05, is p<0.01, * is p<0.001, and **** is p<0.0001 by Student’s t-test. NS is not significant. Data shown are representative of at least three independent experiments. https://doi.org/10.1371/journal.ppat.1010767.g004 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 11 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication [49,70,73]. We observed a significant defect in IL-1 family cytokine release in the absence of caspase-8 in TNF-primed BMDMs (S4B and S5A Figs), suggesting a role for caspase-8 in TNF-licensed inflammatory cytokine release. To investigate the contribution of caspase- 8-mediated inflammatory gene expression downstream of TNF, we assessed expression of cas- pase-11 in the absence of caspase-8. Caspase-11 remained more highly expressed in TNF- primed Ripk3-/-Casp8-/- BMDMs relative to unprimed BMDMs, suggesting that caspase-8 is not required for TNF-mediated upregulation of caspase-11 (S5B Fig). To test the specific contribution of caspase-8 autoprocessing in TNF-mediated cell death and restriction of L. pneumophila replication, we used knock-in mutant mice bearing a D387A mutation in caspase-8 (Ripk3-/-Casp8DA), which eliminates caspase-8 autoprocessing and apo- ptotic activity [43,71,74]. Infection of TNF-primed Ripk3-/-Casp8DA BMDMs resulted in sub- stantial increase in cell death, as measured by PI uptake, relative to unprimed Ripk3-/-Casp8DA BMDMs. Furthermore, the levels of cell death were comparable to TNF-primed Ripk3-/- cells, whereas we observed substantially lower cell death in both TNF-primed and unprimed Ripk3-/-Casp8-/- cells (Fig 4C and S6 Fig). These data suggest that caspase-8 is required for opti- mal cell death during L. pneumophila infection, but that caspase-8 autoprocessing, and thus the classical apoptotic function of caspase-8, is not required. Indeed, using a fluorogenic substrate assay to interrogate downstream caspase-3/7 enzymatic activity during infection, we found that TNF-licensed cells demonstrated significantly lower caspase-3 and -7 (Fig 4D) activity rel- ative to unprimed cells. These data indicate that an additional, non-apoptotic mechanism of TNF-licensed death is therefore downstream of caspase-8 in L. pneumophila infection. TNF priming licenses gasdermin-dependent cell death downstream of caspase-1, caspase-8, and caspase-11 following infection Observing a requirement for multiple caspases in TNF-licensed death following L. pneumo- phila infection, we next investigated the role of terminal gasdermins in this death. Inflamma- some activation triggers caspase-mediated cleavage of the cytosolic pore-forming protein gasdermin D. This frees the N-terminal pore-forming domain from the autoinhibitory C-ter- minal domain, allowing oligomerization and formation of a pyroptotic pore from which the cell can release inflammatory cytokines and intracellular damage-associated molecular pat- terns [53,60]. To dissect the role of gasdermins in TNF-licensed death, we used BMDMs defi- cient in either gasdermin D (Gsdmd-/-) or the closely related protein, gasdermin E (Gsdme-/-), which is known to be activated by caspase-8 downstream of TNF to cause pyroptosis in certain contexts and cell types [75]. We additionally infected BMDMs lacking both gasdermin D and gasdermin E (Gsdmd-/-Gsdme-/-), as gasdermin E has been suggested to mediate secondary or non-canonical pyroptosis in the absence of gasdermin D [51,75]. We observed no decrease in TNF-licensed cell death in Gsdme-/- BMDMs, confirming that gasdermin E does not bear a pri- mary role in cell death following L. pneumophila infection (Fig 5A). We did, however, observe a significant defect in cell death in both Gsdmd-/- and Gsdmd-/-Gsdme-/- macrophages, suggest- ing that gasdermins are required for TNF-licensed cell death during L. pneumophila infection (Fig 5A and 5B). While we did observe a significant attenuation of cell death solely in the absence of gasdermin D, we observed a further significant decrease in the absence of both gas- dermins D and E, suggesting some compensation by gasdermin E in the absence of gasdermin D (Fig 5A). This is in keeping with a recent study that found a compensatory role for gasder- min E in pyroptosis downstream of caspase-8, specifically in the context of gasdermin D defi- ciency [75]. Later, at 8 hours post-infection, we observed a delayed return of cell death, which may be reflective of compensatory apoptosis in the absence of pyroptotic death (Fig 5B). We also observed significant abrogation of IL-1α and IL-1β release in the absence of gasdermins D PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 12 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication Fig 5. TNF priming licenses a gasdermin-dependent cell death, requiring caspase-1, caspase-8, and caspase-11 following infection. (A) WT, Gsdme-/-, Gsdmd-/-, and Gsdmd-/-Gsdme-/- BMDMs were primed with either 10 ng/mL rTNF or PBS mock control for 16 hours prior to infection with non-replicating ΔflaA L. pneumophila at MOI = 10. Cytotoxicity was measured by PI uptake assay. (B) WT and Gsdmd-/-Gsdme-/- BMDMs were primed with either 10 ng/mL rTNF or PBS mock control for 16 hours prior to infection with non-replicating ΔflaA L. pneumophila at MOI = 50, and cells. Cytotoxicity was measured by LDH release assay. (C) WT and Gsdmd-/-Gsdme-/- BMDMS were primed with 10 ng/mL rTNF for 16 hours prior to infection with non-replicating ΔflaA L. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 13 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication pneumophila at MOI = 50 for 6 hours. Cytokine release was measured by ELISA. (D) WT and Gsdmd-/-Gsdme-/- BMDMs were primed with either 10 ng/mL rTNF or PBS mock control for 16 hours prior to infection with replicating ΔflaA L. pneumophila at MOI = 1 for 72 hours. (E-F) Ripk3-/-, Ripk3-/-Casp8-/-, Ripk3-/-Casp1-/-Casp11-/-, and Ripk3-/-Casp1-/-Casp11-/-Casp8-/- BMDMs were primed with either 10 ng/mL rTNF or PBS mock control for 16 hours prior to infection with non-replicating (E) or replicating (F) ΔflaA L. pneumophila at MOI = 10 (E) or 1 (F). Graphs show the mean ± SEM of triplicate wells. * is p<0.05, is p<0.01, * is p<0.001, and **** is p<0.0001 by 2-way ANOVA with Sˇı´da´k’s post-test (B,C) or Student’s t-test (D,E). NS is not significant. Data shown are representative of at least three independent experiments. https://doi.org/10.1371/journal.ppat.1010767.g005 and E following infection during TNF priming, indicating that TNF licenses cells to undergo gasdermin-mediated pyroptosis to release IL-1 cytokines (Fig 5C). Of note, cells deficient for both gasdermin D and E still exhibited complete abrogation of bacterial growth in the context of TNF priming, suggesting additional TNF-mediated protective mechanisms independent of gasdermin D and E (Fig 5D). These data provoked the hypothesis that TNF-licensed caspase-1, caspase-8, and caspase-11 may all be participating in inflammasome activation to mediate subsequent pyroptosis in L. pneumophila-infected macrophages. To further define the relative contributions of caspases-1, -8, and -11, we generated BMDMs from Ripk3-/-, Ripk3-/-Casp8-/-, Ripk3-/-Casp1-/-Casp11-/-, and Ripk3-/-Casp8-/-Casp1-/-Casp11-/- mice. Using these cells, we observed that TNF-licensed cell death was partially attenuated in the absence of either caspase-8 or caspases-1 and -11 (Fig 5E). However, TNF-licensed cell death was fully abrogated in the absence of all three caspases (Fig 5E). We likewise assessed restriction of bacterial replication in these cell populations and found that, while the absences of caspase-8 or caspases-1 and -11 still allowed for a moderate but significant TNF-mediated attenuation of bacterial growth, the absence of all three caspases rendered cells unable to restrict L. pneumophila replication even in the context of TNF prim- ing (Fig 5F). These data together suggest that in TNF-licensed BMDMs infected with L. pneu- mophila, caspases-1, -8, and -11 collectively facilitate gasdermin-dependent, pyroptotic cell death and control of intracellular bacterial replication. TNFR1 and caspase-8 are required for control of pulmonary L. pneumophila infection The TNF-mediated multi-caspase control of L. pneumophila we observed in BMDMs may contribute to the effective TNF-dependent control of bacterial replication in vivo. We therefore investigated whether the TNF and caspase-8-mediated restriction of flagellin-deficient L. pneu- mophila in primary BMDMs in vitro could also be observed during active pulmonary infec- tion. In keeping with our findings in cell culture, we observed that Tnf-/- mice were unable to control bacterial loads in the lung, unlike their wild type C57BL/6 counterparts (Fig 6A). Like- wise, we observed a significant defect in bacterial control in Tnfr1-/- mice compared to WT, suggesting a role for TNF signaling through TNFR1 in control of pulmonary L. pneumophila infection (Fig 6B). These data agree with our findings in vitro, as well as other in vivo studies [8,13,20,56]. Moreover, Ripk3-/-Casp8-/- mice also exhibited a significant defect in restriction of bacterial replication relative to WT and Ripk3-/- control mice, indicating a requirement for caspase-8 in control of L. pneumophila infection in vivo (Fig 6C). Notably, we observed that Ripk3-/-- Casp8DA mice unable to undergo caspase-8 autocleavage exhibited similar levels of bacterial replication compared to Ripk3-/- mice or Ripk3-/-Casp8DA/+ littermate controls (Fig 6D), in agreement with our in vitro findings indicating that caspase-8 autocleavage is not necessary for TNF-mediated restriction of L. pneumophila replication (Fig 4C). These data together indicate that signaling through TNFR1 and non-apoptotic caspase-8 activity mediate control of pulmo- nary L. pneumophila infection. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 14 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication Fig 6. TNFR1 and caspase-8 are required for control of pulmonary L. pneumophila infection. WT, Tnf-/- (A, D), Tnfr1-/- (B), Ripk3-/-(C, D), Ripk3-/-Casp8-/- (C), Ripk3-/-Casp8DA/+ (D), and Ripk3-/-Casp8DA/DA mice were infected intranasally with 1x106 ΔflaA L. pneumophila in 40 μL sterile PBS. Lung CFUs were quantified after 48 hours of infection. Graph shows the ± SEM of three or four infected mice per group. * is p<0.05, is p<0.01, and * is p<0.001 by Student’s t-test (A,B) or one-way ANOVA with Tukey HSD post-test (C,D). Data shown are representative of at least two independent experiments. https://doi.org/10.1371/journal.ppat.1010767.g006 Discussion The inflammatory response generated by the innate immune system determines the ability of the mammalian host to efficiently clear bacterial infection. While the relative contributions of immune mechanisms such as inflammasomes and cytokines have been well characterized, the cell-intrinsic effects of cytokines such as TNF in restricting intracellular bacterial pathogens have remained mechanistically obscure. In this study, we set out to determine how TNF con- trols L. pneumophila infection. Specifically, we used priming of macrophages with recombi- nant TNF prior to infection in order to mimic the kinetics of initially infected cells inducing PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 15 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication TNF production in uninfected neighboring cells to defend against the second wave of infec- tion. We also used L. pneumophila lacking flagellin to avoid triggering the NAIP/NLRC4 inflammasome, as our findings and previous studies indicated that TNF can mediate restric- tion of flagellin-deficient L. pneumophila (Fig 1) [27]. Using this priming and infection model alongside genetic tools, we demonstrate that TNF signaling through TNFR1 licenses L. pneu- mophila-infected cells to rapidly and robustly undergo cell death independently of flagellin and the NAIP/NLRC4 inflammasome. We find that TNF promotes accelerated inflammasome activation, gasdermin-dependent pyroptosis, and release of IL-1 family cytokines. TNF is able to license this pyroptotic death in part by upregulating components of the caspase-11 non- canonical inflammasome ahead of infection. Furthermore, we find that following TNF prim- ing, NLRP3 inflammasome activation and caspase-8 non-apoptotic activity in parallel mediate maximal cytokine release and death of infected cells. We find that caspases-1, -11, and -8 are all required for maximal TNF-dependent cell death and control of L. pneumophila replication within macrophages. We also observe a defect in control of pulmonary L. pneumophila infec- tion in the absence of caspase-8. Thus, our findings indicate that TNF licenses the host to respond to flagellin-deficient L. pneumophila infection by rapidly activating cell death path- ways downstream of caspases-1, -11, and -8, thereby restricting bacterial replication. Our find- ings also indicate a critical role for caspase-8 in control of pulmonary L. pneumophila infection independent of the NAIP5/NLRC4 inflammasome. Intracellular pathogens elicit inflammasome responses as the sanctity of the intracellular compartment is violated by infectious insult [76]. The exact nature of the inflammasome acti- vated is determined by the type of ligand detected, be it an injected bacterial protein, effector or toxin activity, foreign nucleic acid, or bacterial cell wall components [77]. Additionally, the inflammatory context in which these foreign signals are detected will orient the sensitivity, magnitude, and character of the inflammatory response. L. pneumophila robustly activates the NAIP5/NLRC4 inflammasome within infected macrophages by injecting flagellin into the host cytosol [20,33,55,59], and this process requires TNF signaling for optimal restriction of L. pneumophila replication [20]. Our study uses flagellin-deficient L. pneumophila to highlight that TNF also licenses rapid activation of multiple inflammasomes to restrict bacterial infec- tion independently of flagellin and NAIP5/NLRC4 inflammasome activation. Our data dem- onstrate that caspase-11, which detects cytosolic LPS, is crucial for the early TNF-licensed response to L. pneumophila infection. This caspase-11 activation results in the formation of gasdermin D (GSDMD) pores that lead to cell lysis and IL-1α release. Importantly, TNF prim- ing also enhances NLRP3 inflammasome-dependent release of IL-1β. It is possible that this NLRP3 inflammasome activation may be downstream of caspase-11 activation, likely by virtue of caspase-11-dependent gasdermin pores facilitating K+ efflux [36,37]. Furthermore, it is pos- sible that both caspase-1 and caspase-8 are being recruited to the NLRP3 inflammasome [50,73,78]. Regardless, we demonstrate that this TNF-mediated inflammasome activation is dependent on the action of the L. pneumophila T4SS. While our data suggest that the activating signal is independent of the GBPs on chromosome 3, it is possible that one of the other 7 murine GBPs is mediating vacuolar permeabilization in response to the T4SS [68]. Alterna- tively, it is possible that either an uncharacterized L. pneumophila effector molecule or even the process of secretion system injection alerts the promiscuous NLRP3 inflammasome, trig- gering further feedforward loops among the other inflammasomes. While we observe a requirement for GSDMD in optimal TNF-licensed cell death in L. pneumophila-infected macrophages, we see further attenuation of cell death in the absence of both GSDMD and the related protein gasdermin E (GSDME), indicating that both GSDMD and GSDME contribute to TNF-licensed cell death in response to L. pneumophila infection. Notably, TNF-primed Gsdme-/- BMDMs exhibited no defect in cell death, indicating that PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 16 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication GSDMD is primarily responsible for cell lysis in this context. GSDME is cleaved by caspase-3 downstream of caspase-8 in cancer cells, intestinal epithelial cells, and macrophages [75,79,80]. This cleavage results in both pyroptosis and permeabilization of mitochondria, enhancing intrinsic apoptosis [81]. In THP-1 macrophages, GSDME mediates IL-1β release and limited pyroptosis in response to nigericin treatment and Salmonella infection, as well as pyroptosis in the absence of GSDMD [75]. GSDME does not universally contribute to pyrop- tosis, however, as in the setting of pathogen-induced NF-κB blockade GSDME is activated by a caspase-8/3 pathway yet exhibits no role in macrophage lysis [82]. Additionally, a secondary form of pyroptosis can occur downstream of caspase-8 in the absence of GSDMD [51]. It therefore appears likely that the residual, GSDMD-independent death we observe in our sys- tem is mediated by GSDME compensation downstream of caspase-8 and caspase-3. Intrigu- ingly, we observed that although TNF-primed cell death was attenuated in the absence of GSDMD and GSDME, there still remained delayed cell death and restriction of L. pneumo- phila replication, indicating that additional host factors downstream of caspases-1, -11, and -8 contribute to cell death and restriction of L. pneumophila. These additional host factors may include accelerated phagolysosomal fusion with the Legionella-containing vacuole, which has been observed as a downstream consequence of TNF signaling by other groups [27]; or TNF- induced macrophage necrosis, which has been shown to be mediated by mitochondrial ROS in the context of Mycobacterium infection [83]. Alternatively, gasdermin-independent restric- tion of L. pneumophila may be a result of compensatory caspase-8-mediated apoptosis [84], which is in line with the gasdermin-independent cell death we observe occurring late during infection. This is additionally congruous with multiple studies which have shown that, espe- cially in vivo, apoptotic death can compensate for the lack of caspase-1-mediated death late during infection [50,84,85]. The potential compensation by GSDME indicates an additional node of caspase-8-medi- ated compensation in the TNF-primed inflammasome response against L. pneumophila. Our study demonstrates that in TNF-primed cells, caspases-1, -11, and -8 collectively mediate cell death and control of L. pneumophila infection. While caspase-8 is thought to predominantly initiate extrinsic apoptosis, caspase-8 has been shown to compensate in the absence of caspase- 1 to cleave shared substrates, including IL-1β and GSDMD [50–53,86]. Notably, during L. pneumophila infection in the absence of TNF priming, NAIP5/NLRC4 inflammasome activa- tion leads to caspase-1 and -8-mediated activation of GSDMD and caspase-7, respectively [70]. In contrast, caspases-1, -8, and -11 are dispensable for restriction of flagellin-deficient L. pneu- mophila in the absence of TNF priming [50]. However, in TNF-primed macrophages infected with flagellin-deficient L. pneumophila, we see involvement of all three caspases, as we find that in the absence of caspases-1 and -11, caspase-8 contributes to restriction of L. pneumo- phila infection within BMDMs. Our data also indicate that caspase-8 autocleavage is not involved in control of in vitro or in vivo infection, suggesting that caspase-8 is promoting cell death and bacterial restriction independently of its apoptotic activity. It is possible that in addi- tion to caspase-8’s role in cell death, it may contribute to transcription of inflammatory cyto- kine genes such as Il1a, Il1b, and Il12b downstream of TNF signaling, which may also contribute to protection [43,47–49]. Taken as a whole, our study deepens our understanding of the mechanisms by which TNF is able to position cells to better control intracellular bacterial infection. We find that cells which have been licensed by TNF rapidly undergo pyroptosis and robustly respond to flagel- lin-deficient L. pneumophila infection. We further characterize this death as being mediated not only by caspase-11, but additionally involving caspases-1 and -8, which together contribute to control of L. pneumophila replication in TNF-primed macrophages. We finally demonstrate that caspase-8 is required for bacterial control in a mouse model of pulmonary L. pneumophila PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 17 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication infection. Altogether, our findings highlight the multiple mechanisms by which TNF triggers protective death in cells through the activation of multiple caspases, and provide new insight into the function of TNF in host defense against intracellular L. pneumophila infection. Materials and methods Ethics statement All animal experiments were carried out in accordance with the Federal regulations set forth in the Animal Welfare Act (AWA), recommendations in the NIH Guide for the Care and Use of Laboratory Animals, and the guidelines of the University of Pennsylvania Institutional Ani- mal Use and Care Committee. All protocols used in this study were approved by the IACUC at the University of Pennsylvania (Protocol #804928, Protocol #804523). Bacterial culture Legionella pneumophila Philadelphia 1 strains derived from the JR32 background or the LP02 thyA background [87] were cultured on charcoal yeast extract (CYE) agar containing strepto- mycin, as well as thymidine for the LP02 background strains, at 37˚C for 48 hours prior to infection. Wild-type strains as well as flagellin-deficient ΔflaA and Dot/Icm type IV secretion system-deficient ΔdotA mutant strains were used on both JR32 and LP02 genetic backgrounds [33,88]. Mice C57BL/6 mice were purchased from Jackson Laboratories. Tnf-/- mice, Tnfr1-/- mice, Ripk3-/- mice [43], Ripk3-/-Casp8-/- [43,54], Ripk3-/-Casp8DA [71], Gsdmd-/- [89], and Gsdme-/- [82] mice have been previously described and maintained as breeding lines in-house. Gsdme-/- mice were crossbred with Gsdmd-/- mice to produce Gsdmd-/-Gsdme-/- mice. Ripk3-/-Casp8-/-- Casp1-/-Casp11-/- bone marrow was previously described [70] and graciously provided by Dr. Russell Vance. All animals were housed and bred in specific-pathogen-free conditions in accordance with the Animal Welfare Act (AWA) and the guidelines of the University of Penn- sylvania Institutional Animal Use and Care Committee. Mouse infection Mice between the ages of 8 and 12 weeks were anesthetized via intraperitoneal injection of 100 mg/kg ketamine and 10 mg/kg xylazine in PBS. Following confirmation of anesthetization, mice were infected through the intranasal route with 40 μL PBS carrying 1x106 JR32 ΔflaA bac- teria. Quantification of bacterial growth following infection was conducted by excision, weigh- ing, and homogenization of lung tissue at the indicated timepoints using gentleMACS tissue dissociator (Miltenyi Biotec). CFUs were enumerated via plating of lung homogenates on CYE agar containing streptomycin. Mouse bone marrow-derived macrophage culture Bone marrow was harvested from femurs, tibiae, and pelvises of mice described above. Bone marrow was suspended at 1x107 cells/mL in 90% FBS, 10% DMSO solution for freezing in liq- uid nitrogen storage. Bone marrow cells were thawed and differentiated into macrophages by culture at 37˚C in media comprising RPMI, 30% L929 cell supernatant, 20% FBS, 100 IU/mL penicillin, and 100 μg/mL streptomycin. One day prior to infection, cells were plated in media comprising RPMI, 15% L929 cell supernatant, and 10% FBS. Macrophages were plated at 2x105 cells per well in 24-well plates, 5x104 cells per well in 48-well plates, or 1x105 cells per PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 18 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication well in 96-well plates. Recombinant murine TNF-primed wells were plated in media contain- ing 10 ng/mL rTNF for 16 hours prior to infection. Bacterial growth curves For experiments analyzing bacterial growth restriction, infection of BMDMs with JR32 ΔflaA bacteria was carried out in 24-well plates at MOI = 1 in 500 μL macrophage plating media. At 1 hour following infection, cells were washed with warm RPMI to remove extracellular bacte- ria. Macrophages were lysed with sterile diH2O and lysates were serially diluted, then plated on CYE agar plates containing streptomycin. Bacterial CFUs were quantified following 4–5 days of incubation at 37˚C and normalized relative to CFUs isolated at 1 hour post infection. Cell death assays To measure cytotoxicity by way of lactate dehydrogenase (LDH) release, BMDMs were infected with LP02 ΔflaA bacteria in 48-well tissue culture plates. Release of LDH into the cul- ture supernatant was quantified after infection using an LDH Cytotoxicity Detection Kit (Clontech). LDH release was normalized to mock-infected cells and cells treated with 1% Tri- ton to establish maximum LDH release. To measure cytotoxicity by uptake of propidium iodide (PI), BMDMs were infected in 96-well black-walled tissue culture plates. At the time of infection, 5 μM PI was added to plate reader media (20 mM HEPES buffer and 10% FBS in Hank’s Balanced Salt Solution). Cells were then allowed to equilibrate to 37˚C for 10 minutes before being spun to the bottom of the plate at 1200 rpm for 5 minutes. PI uptake into cells was then measured at an excitation wavelength of 530 nm and an emission wavelength of 617 nm. PI uptake was normalized to mock-infected cells and 1% Triton-treated cells. Immunoblotting To analyze protein expression and processing, cells were lysed directly with 1x SDS/PAGE sample buffer. Secreted proteins were isolated from cell supernatants by centrifugation at 2000 rpm for 10 minutes to remove cellular debris, followed by precipitation using trichloroacetic acid (TCA) overnight. Precipitated protein was pelleted by spinning at 13,000 rpm for 15 min- utes at 4˚C, then washed with ice-cold acetone, centrifuged at 13,000 rpm again for 10 minutes, before finally being suspended in 1x SDS/PAGE sample buffer. Samples were heated at 100˚C for 5 minutes and then separated by SDS/PAGE and transferred to PVDF membranes (Milli- pore). Membranes were then probed with primary antibodies specific for murine caspase-11 (#C1354; Sigma-Aldrich), caspase-3 (#9662; Cell Signaling), caspase-8 (#4798; Cell Signaling), gasdermin D (#G7422; Sigma-Aldrich), IL-1β (12242S; Cell Signaling), and β-actin (#4967; Cell Signaling). Membranes were then probed with secondary antibodies anti-rat IgG (7077S; Cell Signaling), anti-mouse IgG (7076S; Cell Signaling), or anti-rabbit IgG (7074S; Cell Signal- ing). ECL Western Blotting Substrate and SuperSignal West Femto Substrate (Thermo Scien- tific) were used. ELISAS Harvested cell supernatants were assayed using ELISA kits for murine IL-1α, IL-1β, and IL-6 (BD Biosciences). Caspase activity assay Activity of caspases-3/7 and -8 were assessed using the corresponding Caspase-Glo Assays in white-walled 96-well plates (Promega). PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 19 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication Statistical analysis Graphing and statistical analysis were carried out in GraphPad Prism 7.0. In comparisons between two groups, unpaired Student’s t-test was utilized to determine significance. In com- parisons between more than two groups, two-way ANOVA was utilized to determine signifi- cance, with Tukey HSD test following up. Difference considered significant when the P value is < 0.05. Supporting information S1 Fig. TNFR1, but not NADPH oxidase or iNOS, is required to restrict intracellular L. pneumophila replication. (A) WT, Tnfr1-/-, and Tnfr1-/-Tnfr2-/- BMDMs were infected with replicating ΔflaA L. pneumophila at MOI = 1. The fold change in CFUs was quantified at 72 hours post-infection. (B-D) WT, Nos2-/-, and Cybb-/- BMDMs were infected with replicating (B, D) or non-replicating (C) ΔflaA L. pneumophila at MOI = 1 (B,D) or 10 (C). The fold change in CFUs was quantified at 72 hours post-infection, while TNF secretion was assessed at 16 hours post-infection. * is p<0.05 is p<0.01, and * is p<0.001 by one-way ANOVA with Tukey HSD post-test (A,B) or Student’s t-test (D), ns is not significant. (TIF) S2 Fig. The L. pneumophila type-IV secretion system is required for induction of BMDM cell death. WT BMDMs were uninfected or infected with non-replicating L. pneumophila ΔflaA or ΔdotA mutant strains at MOI = 10 for 16 hours. Cells were primed with either 10 ng/ mL rTNF or PBS mock control for 16 hours prior to infection. (TIF) S3 Fig. Inflammasome-related gene expression following TNF priming and L. pneumo- phila infection. WT cells were primed with either 10 ng/mL rTNF or PBS for 16 hours, then infected with non-replicating ΔflaA L. pneumophila at MOI = 50. Supernatants and cell lysates were collected for immunoblot and RT-qPCR analysis. (TIF) S4 Fig. GBPs on chromosome 3 are dispensable for TNF-licensed cell death and bacterial restriction. WT and GBPchr3 BMDMs were primed with either 10 ng/mL rTNF or PBS mock control for 16 hours prior to infection with non-replicating (A) or replicating (B) ΔflaA L. pneumophila at MOI = 10 (A) or 1 (B). (A) Cytotoxicity was measured by LDH release assay. (B) The fold change in CFUs was determined at 48 hours post-infection. Data shown are rep- resentative of two independent experiments. (TIF) S5 Fig. Caspase-8 contributes to TNF-licensed release of IL-1 family cytokines, but not expression of caspase-11, following L. pneumophila infection. Ripk3-/- and Ripk3-/-Casp8-/- BMDMs were primed with either 10 ng/mL rTNF or PBS mock control for 16 hours prior to infection with non-replicating ΔflaA L. pneumophila at MOI = 50. (A) Supernatants were ana- lyzed by ELISA at 24 hpi. (B) BMDMs were lysed and analyzed by immunoblot for caspase-11 protein at indicated timepoints. Graphs show the mean ± SEM of triplicate wells. * is p<0.05, is p<0.01, and * is p<0.001 by 2-way ANOVA with Tukey HSD post-test. (TIF) S6 Fig. Caspase-8 is required for maximal cell death following L. pneumophila infection. Ripk3-/-, Ripk3-/-Casp8-/-, and Ripk3-/-Casp8DA cells were infected with non-replicating L. pneu- mophila ΔflaA mutant strain at MOI = 10 for 8 hours. Cells were primed with either 10 ng/mL PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 20 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication rTNF or PBS mock control for 16 hours prior to infection. (TIF) Acknowledgments We would like to acknowledge the labs of Sunny Shin and Igor Brodsky for frequent scientific insight and sharing of reagents. We also thank James Grayczyk for Gsdmd-/-Gsdme-/- mice. We thank Jessica Doerner for help with pulmonary infection. We thank members of the Brodsky lab for mouse bone marrow, as well as Russell Vance for Ripk3-/-Casp1-/-Casp11-/- and Ripk3-/-- Casp1-/-Casp11-/-Casp8-/- triple and quadruple knockout bone marrow. Author Contributions Conceptualization: Tzvi Y. Pollock. Data curation: Tzvi Y. Pollock. Formal analysis: Tzvi Y. Pollock, Vı´ctor R. Va´zquez Marrero. Funding acquisition: Tzvi Y. Pollock, Sunny Shin. Investigation: Tzvi Y. Pollock, Vı´ctor R. Va´zquez Marrero. Methodology: Tzvi Y. Pollock. Project administration: Sunny Shin. Resources: Igor E. Brodsky, Sunny Shin. Supervision: Igor E. Brodsky, Sunny Shin. Validation: Tzvi Y. Pollock, Vı´ctor R. Va´zquez Marrero. Visualization: Tzvi Y. Pollock. Writing – original draft: Tzvi Y. Pollock. Writing – review & editing: Tzvi Y. Pollock, Igor E. Brodsky, Sunny Shin. References 1. Roy CR, Mocarski ES. Pathogen subversion of cell-intrinsic innate immunity. Nat Immunol. 2007; 8 (11):1179–87. https://doi.org/10.1038/ni1528 PMID: 17952043 2. Flynn JL, Goldstein MM, Chan J, Triebold KJ, Pfeffer K, Lowenstein CJ, et al. Tumor necrosis fac- tor-alpha is required in the protective immune response against Mycobacterium tuberculosis in mice. Immunity. 1995; 2(6):561–72. https://doi.org/10.1016/1074-7613(95)90001-2 PMID: 7540941 3. Hsu H, Xiong J, Goeddel DV. The TNF receptor 1-associated protein TRADD signals cell death and NF-kappa B activation. Cell. 1995; 81(4):495–504. https://doi.org/10.1016/0092-8674(95)90070-5 PMID: 7758105 4. Micheau O, Tschopp J. Induction of TNF receptor I-mediated apoptosis via two sequential signaling complexes. Cell. 2003; 114(2):181–90. https://doi.org/10.1016/s0092-8674(03)00521-x PMID: 12887920 5. Wang L, Du F, Wang X. TNF-alpha induces two distinct caspase-8 activation pathways. Cell. 2008; 133 (4):693–703. https://doi.org/10.1016/j.cell.2008.03.036 PMID: 18485876 6. Parameswaran N, Patial S. Tumor necrosis factor-alpha signaling in macrophages. Crit Rev Eukaryot Gene Expr. 2010; 20(2):87–103. 7. Ali T, Kaitha S, Mahmood S, Ftesi A, Stone J, Bronze MS. Clinical use of anti-TNF therapy and increased risk of infections. Drug Healthc Patient Saf. 2013; 5:79–99. https://doi.org/10.2147/DHPS. S28801 PMID: 23569399 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 21 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication 8. Lanternier F, Tubach F, Ravaud P, Salmon D, Dellamonica P, Bretagne S, et al. Incidence and risk fac- tors of Legionella pneumophila pneumonia during anti-tumor necrosis factor therapy: a prospective French study. Chest. 2013; 144(3):990–8. https://doi.org/10.1378/chest.12-2820 PMID: 23744173 9. McDade JE, Shepard CC, Fraser DW, Tsai TR, Redus MA, Dowdle WR. Legionnaires’ disease: isola- tion of a bacterium and demonstration of its role in other respiratory disease. N Engl J Med. 1977; 297 (22):1197–203. https://doi.org/10.1056/NEJM197712012972202 PMID: 335245 10. 11. Fraser DW, Tsai TR, Orenstein W, Parkin WE, Beecham HJ, Sharrar RG, et al. Legionnaires’ disease: description of an epidemic of pneumonia. N Engl J Med. 1977; 297(22):1189–97. https://doi.org/10. 1056/NEJM197712012972201 PMID: 335244 Isberg RR, O’Connor TJ, Heidtman M. The Legionella pneumophila replication vacuole: making a cosy niche inside host cells. Nat Rev Microbiol. 2009; 7(1):13–24. https://doi.org/10.1038/nrmicro1967 PMID: 19011659 12. Hubber A, Roy CR. Modulation of host cell function by Legionella pneumophila type IV effectors. Annu Rev Cell Dev Biol. 2010; 26:261–83. https://doi.org/10.1146/annurev-cellbio-100109-104034 PMID: 20929312 13. Horwitz MA. Formation of a novel phagosome by the Legionnaires’ disease bacterium (Legionella pneu- mophila) in human monocytes. J Exp Med. 1983; 158(4):1319–31. https://doi.org/10.1084/jem.158.4. 1319 PMID: 6619736 14. Vogel JP, Andrews HL, Wong SK, Isberg RR. Conjugative transfer by the virulence system of Legionella pneumophila. Science. 1998; 279(5352):873–6. https://doi.org/10.1126/science.279.5352.873 PMID: 9452389 15. Segal G, Purcell M, Shuman HA. Host cell killing and bacterial conjugation require overlapping sets of genes within a 22-kb region of the Legionella pneumophila genome. Proc Natl Acad Sci U S A. 1998; 95 (4):1669–74. https://doi.org/10.1073/pnas.95.4.1669 PMID: 9465074 16. McCusker KT, Braaten BA, Cho MW, Low DA. Legionella pneumophila inhibits protein synthesis in Chi- nese hamster ovary cells. Infect Immun. 1991; 59(1):240–6. https://doi.org/10.1128/iai.59.1.240-246. 1991 PMID: 1987039 17. Tilney LG, Harb OS, Connelly PS, Robinson CG, Roy CR. How the parasitic bacterium Legionella pneu- mophila modifies its phagosome and transforms it into rough ER: implications for conversion of plasma membrane to the ER membrane. J Cell Sci. 2001; 114(Pt 24):4637–50. https://doi.org/10.1242/jcs.114. 24.4637 PMID: 11792828 18. Belyi Y, Niggeweg R, Opitz B, Vogelsgesang M, Hippenstiel S, Wilm M, et al. Legionella pneumophila glucosyltransferase inhibits host elongation factor 1A. Proc Natl Acad Sci U S A. 2006; 103(45):16953– 8. https://doi.org/10.1073/pnas.0601562103 PMID: 17068130 19. Coers J, Kagan JC, Matthews M, Nagai H, Zuckman DM, Roy CR. Identification of Icm protein com- plexes that play distinct roles in the biogenesis of an organelle permissive for Legionella pneumophila intracellular growth. Mol Microbiol. 2000; 38(4):719–36. https://doi.org/10.1046/j.1365-2958.2000. 02176.x PMID: 11115108 20. Coers J, Vance RE, Fontana MF, Dietrich WF. Restriction of Legionella pneumophila growth in macro- phages requires the concerted action of cytokine and Naip5/Ipaf signalling pathways. Cell Microbiol. 2007; 9(10):2344–57. https://doi.org/10.1111/j.1462-5822.2007.00963.x PMID: 17506816 21. Akhter A, Gavrilin MA, Frantz L, Washington S, Ditty C, Limoli D, et al. Caspase-7 activation by the Nlrc4/Ipaf inflammasome restricts Legionella pneumophila infection. PLoS Pathog. 2009; 5(4): e1000361. https://doi.org/10.1371/journal.ppat.1000361 PMID: 19343209 22. Matsiota-Bernard P, Lefebre C, Sedqui M, Cornillet P, Guenounou M. Involvement of tumor necrosis factor alpha in intracellular multiplication of Legionella pneumophila in human monocytes. Infect Immun. 1993; 61(12):4980–3. https://doi.org/10.1128/iai.61.12.4980-4983.1993 PMID: 8225572 23. Brieland JK, Remick DG, Freeman PT, Hurley MC, Fantone JC, Engleberg NC. In vivo regulation of rep- licative Legionella pneumophila lung infection by endogenous tumor necrosis factor alpha and nitric oxide. Infect Immun. 1995; 63(9):3253–8. https://doi.org/10.1128/iai.63.9.3253-3258.1995 PMID: 7642253 24. Skerrett SJ, Martin TR. Roles for tumor necrosis factor alpha and nitric oxide in resistance of rat alveolar macrophages to Legionella pneumophila. Infect Immun. 1996; 64(8):3236–43. https://doi.org/10.1128/ iai.64.8.3236-3243.1996 PMID: 8757859 25. McHugh S, Yamamoto Y, Klein TW, Friedman H. Differential expression of IL-1 and TNF receptors in murine macrophages infected with virulent vs. avirulent Legionella pneumophila. Can J Microbiol. 2000; 46(10):885–91. PMID: 11068674 26. McHugh SL, Newton CA, Yamamoto Y, Klein TW, Friedman H. Tumor necrosis factor induces resis- tance of macrophages to Legionella pneumophila infection. Proc Soc Exp Biol Med. 2000; 224(3):191– 6. https://doi.org/10.1046/j.1525-1373.2000.22420.x PMID: 10865236 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 22 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication 27. Ziltener P, Reinheckel T, Oxenius A. Neutrophil and Alveolar Macrophage-Mediated Innate Immune Control of Legionella pneumophila Lung Infection via TNF and ROS. PLoS Pathog. 2016; 12(4): e1005591. https://doi.org/10.1371/journal.ppat.1005591 PMID: 27105352 28. Koo S, Marty FM, Baden LR. Infectious complications associated with immunomodulating biologic agents. Hematol Oncol Clin North Am. 2011; 25(1):117–38. https://doi.org/10.1016/j.hoc.2010.11.009 PMID: 21236394 29. Shin S, Case CL, Archer KA, Nogueira CV, Kobayashi KS, Flavell RA, et al. Type IV secretion-depen- dent activation of host MAP kinases induces an increased proinflammatory cytokine response to Legio- nella pneumophila. PLoS Pathog. 2008; 4(11):e1000220. https://doi.org/10.1371/journal.ppat.1000220 PMID: 19043549 30. Martinon F, Burns K, Tschopp J. The inflammasome: a molecular platform triggering activation of inflammatory caspases and processing of proIL-beta. Mol Cell. 2002; 10(2):417–26. https://doi.org/10. 1016/s1097-2765(02)00599-3 PMID: 12191486 31. Mariathasan S, Newton K, Monack DM, Vucic D, French DM, Lee WP, et al. Differential activation of the inflammasome by caspase-1 adaptors ASC and Ipaf. Nature. 2004; 430(6996):213–8. https://doi.org/ 10.1038/nature02664 PMID: 15190255 32. Tsuji NM, Tsutsui H, Seki E, Kuida K, Okamura H, Nakanishi K, et al. Roles of caspase-1 in Listeria infec- tion in mice. Int Immunol. 2004; 16(2):335–43. https://doi.org/10.1093/intimm/dxh041 PMID: 14734619 33. Ren T, Zamboni DS, Roy CR, Dietrich WF, Vance RE. Flagellin-deficient Legionella mutants evade cas- pase-1- and Naip5-mediated macrophage immunity. PLoS Pathog. 2006; 2(3):e18. https://doi.org/10. 1371/journal.ppat.0020018 PMID: 16552444 34. Whitfield NN, Byrne BG, Swanson MS. Mouse macrophages are permissive to motile Legionella spe- cies that fail to trigger pyroptosis. Infect Immun. 2010; 78(1):423–32. https://doi.org/10.1128/IAI.00070- 09 PMID: 19841075 35. Case CL, Shin S, Roy CR. Asc and Ipaf Inflammasomes direct distinct pathways for caspase-1 activa- tion in response to Legionella pneumophila. Infect Immun. 2009; 77(5):1981–91. https://doi.org/10. 1128/IAI.01382-08 PMID: 19237518 36. Casson CN, Copenhaver AM, Zwack EE, Nguyen HT, Strowig T, Javdan B, et al. Caspase-11 activation in response to bacterial secretion systems that access the host cytosol. PLoS Pathog. 2013; 9(6): e1003400. https://doi.org/10.1371/journal.ppat.1003400 PMID: 23762026 37. Case CL, Kohler LJ, Lima JB, Strowig T, de Zoete MR, Flavell RA, et al. Caspase-11 stimulates rapid flagellin-independent pyroptosis in response to Legionella pneumophila. Proc Natl Acad Sci U S A. 2013; 110(5):1851–6. https://doi.org/10.1073/pnas.1211521110 PMID: 23307811 38. Lamkanfi M, Dixit VM. Mechanisms and functions of inflammasomes. Cell. 2014; 157(5):1013–22. https://doi.org/10.1016/j.cell.2014.04.007 PMID: 24855941 39. Copenhaver AM, Casson CN, Nguyen HT, Duda MM, Shin S. IL-1R signaling enables bystander cells to overcome bacterial blockade of host protein synthesis. Proc Natl Acad Sci U S A. 2015; 112 (24):7557–62. https://doi.org/10.1073/pnas.1501289112 PMID: 26034289 40. Liu X, Boyer MA, Holmgren AM, Shin S. Legionella-Infected Macrophages Engage the Alveolar Epithe- lium to Metabolically Reprogram Myeloid Cells and Promote Antibacterial Inflammation. Cell Host Microbe. 2020; 28(5):683–98 e6. https://doi.org/10.1016/j.chom.2020.07.019 PMID: 32841604 41. Hsu H, Huang J, Shu HB, Baichwal V, Goeddel DV. TNF-dependent recruitment of the protein kinase RIP to the TNF receptor-1 signaling complex. Immunity. 1996; 4(4):387–96. https://doi.org/10.1016/ s1074-7613(00)80252-6 PMID: 8612133 42. Ting AT, Pimentel-Muinos FX, Seed B. RIP mediates tumor necrosis factor receptor 1 activation of NF- kappaB but not Fas/APO-1-initiated apoptosis. EMBO J. 1996; 15(22):6189–96. PMID: 8947041 43. Philip NH, DeLaney A, Peterson LW, Santos-Marrero M, Grier JT, Sun Y, et al. Activity of Uncleaved Caspase-8 Controls Anti-bacterial Immune Defense and TLR-Induced Cytokine Production Indepen- dent of Cell Death. PLoS Pathog. 2016; 12(10):e1005910. https://doi.org/10.1371/journal.ppat. 1005910 PMID: 27737018 44. Sun L, Wang H, Wang Z, He S, Chen S, Liao D, et al. Mixed lineage kinase domain-like protein medi- ates necrosis signaling downstream of RIP3 kinase. Cell. 2012; 148(1–2):213–27. https://doi.org/10. 1016/j.cell.2011.11.031 PMID: 22265413 45. Zhao J, Jitkaew S, Cai Z, Choksi S, Li Q, Luo J, et al. Mixed lineage kinase domain-like is a key receptor interacting protein 3 downstream component of TNF-induced necrosis. Proc Natl Acad Sci U S A. 2012; 109(14):5322–7. https://doi.org/10.1073/pnas.1200012109 PMID: 22421439 46. Wang H, Sun L, Su L, Rizo J, Liu L, Wang LF, et al. Mixed lineage kinase domain-like protein MLKL causes necrotic membrane disruption upon phosphorylation by RIP3. Mol Cell. 2014; 54(1):133–46. https://doi.org/10.1016/j.molcel.2014.03.003 PMID: 24703947 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 23 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication 47. DeLaney AA, Berry CT, Christian DA, Hart A, Bjanes E, Wynosky-Dolfi MA, et al. Caspase-8 promotes c-Rel-dependent inflammatory cytokine expression and resistance against Toxoplasma gondii. Proc Natl Acad Sci U S A. 2019; 116(24):11926–35. https://doi.org/10.1073/pnas.1820529116 PMID: 31147458 48. Gitlin AD, Heger K, Schubert AF, Reja R, Yan D, Pham VC, et al. Integration of innate immune signalling by caspase-8 cleavage of N4BP1. Nature. 2020; 587(7833):275–80. https://doi.org/10.1038/s41586- 020-2796-5 PMID: 32971525 49. Allam R, Lawlor KE, Yu EC, Mildenhall AL, Moujalled DM, Lewis RS, et al. Mitochondrial apoptosis is dispensable for NLRP3 inflammasome activation but non-apoptotic caspase-8 is required for inflamma- some priming. EMBO Rep. 2014; 15(9):982–90. https://doi.org/10.15252/embr.201438463 PMID: 24990442 50. Mascarenhas DPA, Cerqueira DM, Pereira MSF, Castanheira FVS, Fernandes TD, Manin GZ, et al. Inhibition of caspase-1 or gasdermin-D enable caspase-8 activation in the Naip5/NLRC4/ASC inflam- masome. PLoS Pathog. 2017; 13(8):e1006502. https://doi.org/10.1371/journal.ppat.1006502 PMID: 28771586 51. Schneider KS, Gross CJ, Dreier RF, Saller BS, Mishra R, Gorka O, et al. The Inflammasome Drives GSDMD-Independent Secondary Pyroptosis and IL-1 Release in the Absence of Caspase-1 Protease Activity. Cell Rep. 2017; 21(13):3846–59. https://doi.org/10.1016/j.celrep.2017.12.018 PMID: 29281832 52. Doerflinger M, Deng Y, Whitney P, Salvamoser R, Engel S, Kueh AJ, et al. Flexible Usage and Intercon- nectivity of Diverse Cell Death Pathways Protect against Intracellular Infection. Immunity. 2020; 53 (3):533–47 e7. https://doi.org/10.1016/j.immuni.2020.07.004 PMID: 32735843 53. Demarco B, Grayczyk JP, Bjanes E, Le Roy D, Tonnus W, Assenmacher CA, et al. Caspase-8-depen- dent gasdermin D cleavage promotes antimicrobial defense but confers susceptibility to TNF-induced lethality. Sci Adv. 2020; 6(47). https://doi.org/10.1126/sciadv.abc3465 PMID: 33208362 54. Philip NH, Dillon CP, Snyder AG, Fitzgerald P, Wynosky-Dolfi MA, Zwack EE, et al. Caspase-8 medi- ates caspase-1 processing and innate immune defense in response to bacterial blockade of NF-kappaB and MAPK signaling. Proc Natl Acad Sci U S A. 2014; 111(20):7385–90. 55. Molofsky AB, Byrne BG, Whitfield NN, Madigan CA, Fuse ET, Tateda K, et al. Cytosolic recognition of flagellin by mouse macrophages restricts Legionella pneumophila infection. J Exp Med. 2006; 203 (4):1093–104. https://doi.org/10.1084/jem.20051659 PMID: 16606669 56. Fujita M, Ikegame S, Harada E, Ouchi H, Inoshima I, Watanabe K, et al. TNF receptor 1 and 2 contribute in different ways to resistance to Legionella pneumophila-induced mortality in mice. Cytokine. 2008; 44 (2):298–303. https://doi.org/10.1016/j.cyto.2008.08.015 PMID: 18838275 57. Hammermann R, Dreissig MD, Mossner J, Fuhrmann M, Berrino L, Gothert M, et al. Nuclear factor-kap- paB mediates simultaneous induction of inducible nitric-oxide synthase and Up-regulation of the cat- ionic amino acid transporter CAT-2B in rat alveolar macrophages. Mol Pharmacol. 2000; 58(6):1294– 302. PMID: 11093766 58. Peterson LW, Philip NH, Dillon CP, Bertin J, Gough PJ, Green DR, et al. Cell-Extrinsic TNF Collabo- rates with TRIF Signaling To Promote Yersinia-Induced Apoptosis. J Immunol. 2016; 197(10):4110–7. https://doi.org/10.4049/jimmunol.1601294 PMID: 27733552 59. Lightfield KL, Persson J, Brubaker SW, Witte CE, von Moltke J, Dunipace EA, et al. Critical function for Naip5 in inflammasome activation by a conserved carboxy-terminal domain of flagellin. Nat Immunol. 2008; 9(10):1171–8. https://doi.org/10.1038/ni.1646 PMID: 18724372 60. Kovacs SB, Miao EA. Gasdermins: Effectors of Pyroptosis. Trends Cell Biol. 2017; 27(9):673–84. https://doi.org/10.1016/j.tcb.2017.05.005 PMID: 28619472 61. Kayagaki N, Warming S, Lamkanfi M, Vande Walle L, Louie S, Dong J, et al. Non-canonical inflamma- some activation targets caspase-11. Nature. 2011; 479(7371):117–21. https://doi.org/10.1038/ nature10558 PMID: 22002608 62. Swanson KV, Deng M, Ting JP. The NLRP3 inflammasome: molecular activation and regulation to ther- apeutics. Nat Rev Immunol. 2019; 19(8):477–89. https://doi.org/10.1038/s41577-019-0165-0 PMID: 31036962 63. Ea CK, Deng L, Xia ZP, Pineda G, Chen ZJ. Activation of IKK by TNFalpha requires site-specific ubiqui- tination of RIP1 and polyubiquitin binding by NEMO. Mol Cell. 2006; 22(2):245–57. https://doi.org/10. 1016/j.molcel.2006.03.026 PMID: 16603398 64. Li H, Kobayashi M, Blonska M, You Y, Lin X. Ubiquitination of RIP is required for tumor necrosis factor alpha-induced NF-kappaB activation. J Biol Chem. 2006; 281(19):13636–43. https://doi.org/10.1074/ jbc.M600620200 PMID: 16543241 65. Santos JC, Boucher D, Schneider LK, Demarco B, Dilucca M, Shkarina K, et al. Human GBP1 binds LPS to initiate assembly of a caspase-4 activating platform on cytosolic bacteria. Nat Commun. 2020; 11(1):3276. https://doi.org/10.1038/s41467-020-16889-z PMID: 32581219 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 24 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication 66. Finethy R, Jorgensen I, Haldar AK, de Zoete MR, Strowig T, Flavell RA, et al. Guanylate binding pro- teins enable rapid activation of canonical and noncanonical inflammasomes in Chlamydia-infected macrophages. Infect Immun. 2015; 83(12):4740–9. https://doi.org/10.1128/IAI.00856-15 PMID: 26416908 67. Price JV, Russo D, Ji DX, Chavez RA, DiPeso L, Lee AY, et al. IRG1 and Inducible Nitric Oxide Synthase Act Redundantly with Other Interferon-Gamma-Induced Factors To Restrict Intracellular Rep- lication of Legionella pneumophila. mBio. 2019; 10(6). https://doi.org/10.1128/mBio.02629-19 PMID: 31719183 68. Pilla DM, Hagar JA, Haldar AK, Mason AK, Degrandi D, Pfeffer K, et al. Guanylate binding proteins pro- mote caspase-11-dependent pyroptosis in response to cytoplasmic LPS. Proc Natl Acad Sci U S A. 2014; 111(16):6046–51. https://doi.org/10.1073/pnas.1321700111 PMID: 24715728 69. Taabazuing CY, Okondo MC, Bachovchin DA. Pyroptosis and Apoptosis Pathways Engage in Bidirec- tional Crosstalk in Monocytes and Macrophages. Cell Chem Biol. 2017; 24(4):507–14 e4. https://doi. org/10.1016/j.chembiol.2017.03.009 PMID: 28392147 70. Goncalves AV, Margolis SR, Quirino GFS, Mascarenhas DPA, Rauch I, Nichols RD, et al. Gasdermin- D and Caspase-7 are the key Caspase-1/8 substrates downstream of the NAIP5/NLRC4 inflamma- some required for restriction of Legionella pneumophila. PLoS Pathog. 2019; 15(6):e1007886. https:// doi.org/10.1371/journal.ppat.1007886 PMID: 31251782 71. Tummers B, Mari L, Guy CS, Heckmann BL, Rodriguez DA, Ruhl S, et al. Caspase-8-Dependent Inflammatory Responses Are Controlled by Its Adaptor, FADD, and Necroptosis. Immunity. 2020; 52 (6):994–1006 e8. https://doi.org/10.1016/j.immuni.2020.04.010 PMID: 32428502 72. Oberst A, Pop C, Tremblay AG, Blais V, Denault JB, Salvesen GS, et al. Inducible dimerization and inducible cleavage reveal a requirement for both processes in caspase-8 activation. J Biol Chem. 2010; 285(22):16632–42. https://doi.org/10.1074/jbc.M109.095083 PMID: 20308068 73. Antonopoulos C, Russo HM, El Sanadi C, Martin BN, Li X, Kaiser WJ, et al. Caspase-8 as an Effector and Regulator of NLRP3 Inflammasome Signaling. J Biol Chem. 2015; 290(33):20167–84. https://doi. org/10.1074/jbc.M115.652321 PMID: 26100631 74. Kang TB, Oh GS, Scandella E, Bolinger B, Ludewig B, Kovalenko A, et al. Mutation of a self-processing site in caspase-8 compromises its apoptotic but not its nonapoptotic functions in bacterial artificial chro- mosome-transgenic mice. J Immunol. 2008; 181(4):2522–32. https://doi.org/10.4049/jimmunol.181.4. 2522 PMID: 18684943 75. 76. Zhou B, Abbott DW. Gasdermin E permits interleukin-1 beta release in distinct sublytic and pyroptotic phases. Cell Rep. 2021; 35(2):108998. https://doi.org/10.1016/j.celrep.2021.108998 PMID: 33852854 Lamkanfi M, Dixit VM. Inflammasomes: guardians of cytosolic sanctity. Immunol Rev. 2009; 227(1):95– 105. https://doi.org/10.1111/j.1600-065X.2008.00730.x PMID: 19120479 77. Vance RE, Isberg RR, Portnoy DA. Patterns of pathogenesis: discrimination of pathogenic and non- pathogenic microbes by the innate immune system. Cell Host Microbe. 2009; 6(1):10–21. https://doi. org/10.1016/j.chom.2009.06.007 PMID: 19616762 78. Man SM, Tourlomousis P, Hopkins L, Monie TP, Fitzgerald KA, Bryant CE. Salmonella infection induces recruitment of Caspase-8 to the inflammasome to modulate IL-1beta production. J Immunol. 2013; 191(10):5239–46. 79. 80. Zhang Z, Zhang Y, Xia S, Kong Q, Li S, Liu X, et al. Gasdermin E suppresses tumour growth by activat- ing anti-tumour immunity. Nature. 2020; 579(7799):415–20. https://doi.org/10.1038/s41586-020-2071- 9 PMID: 32188940 Tan G, Huang C, Chen J, Chen B, Zhi F. Gasdermin-E-mediated pyroptosis participates in the patho- genesis of Crohn’s disease by promoting intestinal inflammation. Cell Rep. 2021; 35(11):109265. https://doi.org/10.1016/j.celrep.2021.109265 PMID: 34133932 81. Rogers C, Erkes DA, Nardone A, Aplin AE, Fernandes-Alnemri T, Alnemri ES. Gasdermin pores per- meabilize mitochondria to augment caspase-3 activation during apoptosis and inflammasome activa- tion. Nat Commun. 2019; 10(1):1689. https://doi.org/10.1038/s41467-019-09397-2 PMID: 30976076 82. Chen KW, Demarco B, Ramos S, Heilig R, Goris M, Grayczyk JP, et al. RIPK1 activates distinct gasder- mins in macrophages and neutrophils upon pathogen blockade of innate immune signaling. Proc Natl Acad Sci U S A. 2021; 118(28). https://doi.org/10.1073/pnas.2101189118 PMID: 34260403 83. Roca FJ, Whitworth LJ, Redmond S, Jones AA, Ramakrishnan L. TNF Induces Pathogenic Pro- grammed Macrophage Necrosis in Tuberculosis through a Mitochondrial-Lysosomal-Endoplasmic Reticulum Circuit. Cell. 2019; 178(6):1344–61 e11. https://doi.org/10.1016/j.cell.2019.08.004 PMID: 31474371 84. Kawamoto Y, Morinaga Y, Kimura Y, Kaku N, Kosai K, Uno N, et al. TNF-alpha inhibits the growth of Legionella pneumophila in airway epithelial cells by inducing apoptosis. J Infect Chemother. 2017; 23 (1):51–5. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 25 / 26 PLOS PATHOGENSTNF licenses macrophages to die and restrict Legionella replication 85. Santic M, Asare R, Doric M, Abu Kwaik Y. Host-dependent trigger of caspases and apoptosis by Legio- nella pneumophila. Infect Immun. 2007; 75(6):2903–13. https://doi.org/10.1128/IAI.00147-07 PMID: 17420236 86. Simpson DS, Pang J, Weir A, Kong IY, Fritsch M, Rashidi M, et al. Interferon-gamma primes macro- phages for pathogen ligand-induced killing via a caspase-8 and mitochondrial cell death pathway. Immunity. 2022; 55(3):423–41 e9. 87. Berger KH, Isberg RR. Two distinct defects in intracellular growth complemented by a single genetic locus in Legionella pneumophila. Mol Microbiol. 1993; 7(1):7–19. https://doi.org/10.1111/j.1365-2958. 1993.tb01092.x PMID: 8382332 88. Sadosky AB, Wiater LA, Shuman HA. Identification of Legionella pneumophila genes required for growth within and killing of human macrophages. Infect Immun. 1993; 61(12):5361–73. https://doi.org/ 10.1128/iai.61.12.5361-5373.1993 PMID: 8225610 89. Rauch I, Deets KA, Ji DX, von Moltke J, Tenthorey JL, Lee AY, et al. NAIP-NLRC4 Inflammasomes Coordinate Intestinal Epithelial Cell Expulsion with Eicosanoid and IL-18 Release via Activation of Cas- pase-1 and -8. Immunity. 2017; 46(4):649–59. https://doi.org/10.1016/j.immuni.2017.03.016 PMID: 28410991 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010767 June 6, 2023 26 / 26 PLOS PATHOGENS
10.1371_journal.ppat.1010981	RESEARCH ARTICLE Enhanced stability of the SARS CoV-2 spike glycoprotein following modification of an alanine cavity in the protein core 1,2, Christine Langer1, Irene Boo1, Tasnim Zakir1, Rob Pantelis PoumbouriosID J. Center1,3, Anouschka Akerman4, Vanessa Milogiannakis4, Anupriya Aggarwal4, Bronte A. Johnstone5,6, Jungmin Ha5,6, Fasse´ li Coulibaly5,6, Stuart G. Turville4, Heidi E. Drummer1,2,3 1 Burnet Institute, Melbourne, Australia, 2 Department of Microbiology, Monash University, Clayton, Australia, 3 Department of Microbiology at The Peter Doherty Institute for Infection and Immunity, The University of Melbourne, Parkville, Australia, 4 Kirby Institute, University of New South Wales, Kensington, Australia, 5 Infection Program, Biomedicine Discovery Institute, Monash University, Clayton, Australia, 6 Department of Biochemistry and Molecular Biology, Monash University, Clayton, Australia andy.poumbourios@burnet.edu.au Abstract The spike (S) glycoprotein of SARS CoV-2 is the target of neutralizing antibodies (NAbs) that are crucial for vaccine effectiveness. The S1 subunit binds ACE2 while the S2 subunit mediates virus-cell membrane fusion. S2 is a class I fusion glycoprotein subunit and con- tains a central coiled coil that acts as a scaffold for the conformational changes associated with fusion function. The coiled coil of S2 is unusual in that the 3–4 repeat of inward-facing positions are mostly occupied by polar residues that mediate few inter-helical contacts in the prefusion trimer. We examined how insertion of bulkier hydrophobic residues (Val, Leu, Ile, Phe) to fill a cavity next to Ala1016 and Ala1020 in the 3–4 repeat affects the stability and anti- genicity of S trimers. Substitution of Ala1016 with bulkier hydrophobic residues in the context of a prefusion-stabilized S trimer, S2P-FHA, was associated with increased thermal stability. S glycoprotein membrane fusion function was retained with Ala1016/Ala1020 cavity-filling mutations associated with improved recombinant S2P-FHA thermostability, however 2 mutants, A1016L and A1016V/A1020I, lacked ability to mediate entry of S-HIV-1 pseudo- particles into 293-ACE2 cells. When assessed as immunogens, two thermostable S2P-FHA mutants derived from the ancestral isolate, A1016L (16L) and A1016V/A1020I (VI) elicited neutralizing antibody with 50%-inhibitory dilutions (ID50s) in the range 2,700–5,110 for ancestral and Delta-derived viruses, and 210–1,744 for Omicron BA.1. The antigens elicited antibody specificities directed to the receptor-binding domain (RBD), N-terminal domain (NTD), fusion peptide and stem region of S2. The VI mutation enabled the production of intrinsically stable Omicron BA.1 and Omicron BA.4/5 S2P-FHA-like ectodomain oligomers in the absence of an external trimerization motif (T4 foldon), thus representing an alternative approach for stabilizing oligomeric S glycoprotein vaccines. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Poumbourios P, Langer C, Boo I, Zakir T, Center RJ, Akerman A, et al. (2023) Enhanced stability of the SARS CoV-2 spike glycoprotein following modification of an alanine cavity in the protein core. PLoS Pathog 19(5): e1010981. https://doi.org/10.1371/journal.ppat.1010981 Editor: Tongqing Zhou, National Institutes of Health, UNITED STATES Received: November 7, 2022 Accepted: April 24, 2023 Published: May 18, 2023 Copyright: © 2023 Poumbourios et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. Funding: This work was funded by the Perpetual Impact Philanthropy Application Program (IPAP2021/0251) (PP and HED), the Burnet Vaccine Initiative (PP and HED), mRNA Victoria Activation Program (mAP) (PP and HED), the Victorian Operational Infrastructure Fund (HED), the Australian Medical Foundation Research Grants (MRF2005760, MRF2001684) and New South PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 1 / 33 PLOS PATHOGENSWales Health COVID-19 Research Grants Round 2 (SGT). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: I have read the journal’s policy and the authors of this manuscript have the following competing interests: The work described in the manuscript forms part of a patent application belonging to the Burnet Institute (PP, CL, IB, TZ, HED). All other authors have declared that no competing interests exist. Stabilization of the SARS CoV-2 spike glycoprotein Author summary First-generation SARS CoV-2 vaccines that generate immune responses to ancestral Spike glycoprotein sequences have averted at least 14.4 million deaths, but their effectiveness against the recently emerged Omicron lineages is reduced. The updating of booster vac- cines with variant Spike sequences are therefore likely required to maintain population immunity as the pandemic continues to evolve. The Spike is a trimeric integral membrane protein with a membrane spanning sequence at its C-terminus. The Spike protein-based vaccine that is currently licensed for human use is produced by a complex process that reconstitutes the Spike in an adjuvanted nanoparticle. Alternatively, production of the Spike trimer as a soluble protein generally requires replacement of the membrane span- ning sequence with a foreign often highly immunogenic trimerization motif that can com- plicate clinical advancement. We used systematic structure-directed mutagenesis coupled with functional studies to identify an alternative stabilization approach that negates the requirement for an external trimerization motif or membrane-spanning sequence. The replacement of 2 alanine residues that are associated with a cavity in the core of the Spike trimer with bulkier hydrophobic residues resulted in increased Spike thermal stability. Thermostable Spike mutants retained major conserved neutralizing antibody epitopes and the ability to elicit broad neutralizing antibody responses. One such mutation, referred to as VI, enabled the production of intrinsically stable Omicron variant Spike ectodomain oligomers in the absence of an external trimerization motif. The VI mutation potentially enables a simplified method for producing a stable oligomeric S ectodomain glycoprotein vaccine. Introduction The SARS CoV-2 betacoronavirus has led to the death of more than 6.4 million people with a strong age dependent fatality rate. First-generation vaccines that deliver ancestral SARS CoV- 2-derived viral Spike glycoprotein (S) sequences for its in vivo expression and neutralizing antibody (NAb) induction have been rolled out across the globe and have proven highly effec- tive at preventing symptomatic and severe COVID-19. The viral S glycoprotein mediates receptor attachment and virus-cell membrane fusion and is the target of NAbs, which play a critical role in protection against SARS CoV-2 transmission and disease progression [1–4]. The mature spike comprises 2 functional subunits, S1 and S2, that are derived from a polypro- tein precursor, S, by furin cleavage of an oligobasic motif as it transits the Golgi. ACE2 recep- tor attachment is mediated by the RBD within the large subunit, S1, while membrane fusion is mediated by the small subunit, S2, which contains the fusion peptide. S1 and S2 form a hetero- dimer via non-covalent interactions. S1-S2 is assembled into a higher-order trimeric structure with a coiled coil-forming α-helix of S2 (amino acids 986–1033; referred to as CH) forming the core of the trimer [5] (Fig 1A). A membrane-spanning sequence at the C-terminus of S2 stabilizes the trimer and anchors it to the viral or cell membrane [6]. The ACE2 RBD sits atop the S1 glycoprotein trimer and presents in “up” ACE2-binding-ready and “down” inert orien- tations [7]. Following receptor attachment, S2 is cleaved by the TMPRSS2 protease at the cell surface or by cathepsin L following endocytosis to liberate the fusion peptide of S2 and enable full fusion activation (For review: [8]. The S glycoprotein mediates membrane fusion via a class I mechanism whereby an activation trigger (ACE2-binding by S1, TMPRSS2 cleavage of S2) causes the S2 subunit of the metastable pre-fusion trimer to refold into a stable trimer of PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 2 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein Fig 1. A, Three-dimensional structure of the SARS CoV-2 S ectodomain drawn with the coordinates PDB ID 6VSB [4]. S1 is shown in green and blue, RBM, ACE2 receptor binding motif with receptor-interacting amino acids in red; RBD, receptor binding domain in blue; NTD, N-terminal domain in teal. S2 is shown in yellow and pink: FP, fusion peptide in red; central coiled coil in pink with Ala cavity highlighted in purple. B, heptad repeat motifs within the coiled coil sequences of 3 betacoronaviruses. C, a comparison of the central coiled coil in the prefusion (PDB ID 6VSB [4]) and postfusion (PDB 6XRA [9]) conformations. Although incompletely shown here, HR1 helices (C-terminal part shown in yellow) extend the coiled coil upward toward the cellular membrane in the postfusion conformation [9]. D, Illustration of how the Ala cavity of prefusion S might be filled following substitution with hydrophobic amino acids. 16L and VI are codes for A1016L and A1016V/ A1020I mutations. https://doi.org/10.1371/journal.ppat.1010981.g001 hairpins, bringing the N-terminal fusion peptide and C-terminal membrane spanning sequences together such that their associated membranes fuse [9]. The sites of vulnerability to NAbs within S have been revealed with the isolation of mono- clonal NAbs (mNAbs) from COVID-19 patients and vaccinees. The RBD is an PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 3 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein immunodominant antibody target in natural infection [10] and highly potent NAbs directed to the RBD can block infection by binding to the ACE2 receptor-binding motif (RBM) and directly blocking ACE2 binding, via steric blockade of ACE2 binding, by locking RBDs in the down orientation to preclude ACE2 binding, or by triggering premature refolding of S1-S2 into the post-fusion state with shedding of S1 [11–18]. The N-terminal domain (NTD) of S1 has also been identified as a supersite of vulnerability and comprises multiple antigenic sites [12,15,19–21]. This region exhibits a high degree of plasticity, acquiring point mutations, dele- tions, insertions and glycan additions enabling antibody evasion [22]. Conserved neutraliza- tion epitopes external to the RBD and NTD have also been identified including within the fusion peptide [23] and stem helix of S2 [12,24,25]. Immunity induced by natural infection or vaccination is believed to be driving the emergence of variants of concern (VOCs) which pri- marily contain mutations in the RBD and NTD [26–29]. Variants of concern are associated with successive waves of infections across the globe with Alpha, Delta and now Omicron line- ages, respectively, dominating the pandemic. Key mutations observed in the RBD of VOCs include K417T/N, N439K, N440K, L452R, T478K, E484K/Q/A, F486V and N501Y, while in the NTD, deletion of amino acids 24–26, 69–70, 142–144, 156–157 and 242–245 have been observed. COVID-19 vaccines based on ancestral SARS Cov-2 sequences proved to be highly effective against Alpha and Delta variants, however this was reduced with the emergence of the Omicron lineage of VOCs, which contains at least 30 mutations in S, including at least 15 mutations within the RBD [30–32]. Reduced vaccine efficacy against Omicron lineage subvar- iants appears to correlate with reduced in vitro neutralization potency of vaccinee sera against these VOCs, especially when antibody levels wane over time [13,33–36]. Boosting with VOC- matched vaccines may help to maintain immunity against emerging viral variants as the COVID-19 pandemic evolves [35,37–39]. Class I viral fusion glycoproteins such as Env of retroviruses, HA of orthomyxoviruses and S of coronaviruses contain a central coiled coil, which acts as a scaffold for the conformational changes associated with the membrane fusion process [4,5,9,40–47] (Fig 1A). In the former 2 viral families, the inward-facing positions of the coiled coil are occupied by hydrophobic resi- dues in a 3–4 repeat that stabilize the coiled coil via knob into-holes packing interactions. In the case of SARS CoV-2, these positions within the central coiled coil of S2 (formed by CH helices) are mostly occupied by polar residues that mediate few inter-helical contacts in the prefusion trimer (Fig 1B and 1C). In the postfusion S trimer, the N-terminal 2/3 of the coiled coil is brought together by the packing of HR1 helices that extends the coiled coil in an N-ter- minal direction by 110Å. In this conformation the inward facing residues are close enough for hydrogen bonds to form (Fig 1C). Within the prefusion S coiled coil, Ile1013 is a point of con- tact between the 3 CH helices and forms a small hydrophobic core through inter-helical con- tacts with Ile1013 and with Leu1012. These interactions form a hydrophobic ceiling above a cavity associated with Ala1016 and Ala1020 that occupy central positions of the coiled coil (Fig 1A–1D). Classical studies on protein folding and stability revealed that cavities in a protein’s core are destabilising and filling the cavity with bulkier hydrophobic residues can improve thermal stability and function [48]. In this study, we examined how filling the Ala1016 and Ala1020 cavity with bulkier hydrophobic amino acids influences protein expression, stability, antigenicity and immunogenicity in guinea pigs. Our study identified thermostable S mutants that elicit NAbs in small animals that maintain potency against Delta and Omicron VOCs. The cavity filling A1016V/A1020I mutation stabilized Omicron BA.1 and BA.4/5 S ectodo- mains with S2P-foldon-like quaternary and antigenic structures and negated the requirement for an external trimerization motif. The A1016V/A1020I mutation represents a method for producing an intrinsically stable S ectodomain glycoprotein vaccine in the absence of a foreign trimerization tag. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 4 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein Results Characterization of S2P-FHA A CMV promoter driven expression vector was used to produce a soluble form of the S glyco- protein known as S2P [5], which comprises residues 16–1208 of the ancestral Hu-1 S glycopro- tein, a furin cleavage site mutation, R682RAR-> G682SAS, and a di-Pro substitution at positions 986 and 987. T4 foldon, octa-His and avitag sequences were added to the C-terminus to give S2P-FHA. Wrapp and others have shown that S2P with a C-terminal foldon clamp is a trimer [5]. Following partial purification by divalent cation affinity chromatography, size exclusion chromatography (SEC) of the S2P-FHA protein revealed a major peak coeluting with thyroglobulin (669 kDa) that was collected as a homogenous protein as indicated by SEC and SDS-PAGE (S1A and S1B Fig). The elution profile of S2P-FHA is consistent with that of the S2P-foldon trimer described by Wrapp et al [5] and is thus likely to be a trimer. A biotiny- lated form of purified S2P-FHA exhibited binding activity with hACE2-Fc, (human ACE2 resi- dues 19–615 linked to the Fc domain of human IgG1) and various human mNAbs in ELISA (S1C Fig). Differential scanning fluorimetry (DSF) indicated that the S2P protein possessed relatively low thermal stability, exhibiting a melting temperature (Tm) of 43.6˚C (S1D Fig). Effects of Ala cavity substitutions on thermostability We asked how replacement of Ala1016 and Ala1020 with bulkier hydrophobic residues (Val, Leu, Ile, Phe) affects the stability of the SARS CoV-2 S trimer. (2 examples are shown in Fig 1D). Alanine1016 and Ala1020 were singly and doubly substituted with Val, Ile Leu and Phe in S2P-FHA and the 293F-expressed glycoproteins partially purified by divalent cation affinity chromatography. SDS-PAGE indicated a range of yields: S2P, A1016V, A1020V, A1020I > A1016V/A1020V, A1016I, A1016L, A1020L, A1016V/A1020I > A1016I/A1020I, A1016L/A1020L, A1016V/A1020L, A1016V/A1020F, A1016I/A1020F > A1016F/A1020F (Fig 2A). Differential scanning fluorimetry (DSF) was used to examine the thermostability of S2P-FHA and the mutants. S2P-FHA comprised a major species with a Tm of 43.6˚C and a more stable minor species with a Tm of 58˚C (Fig 2B). The substitution of hydrophobic resi- dues into position 1016 increased the proportion of the 58˚C species with A1016L being the most stabilizing mutation. By contrast, the 43.6˚C species remained the predominant form with substitutions of A1020. Double substitutions were all associated with the stable form. Little or no S2P-FHA was produced with A1016F/A1020F indicating that the large sidechain of Phe was not accommodated at these positions. A subset of mutants with favourable thermostability/yield characteristics were purified to homogeneity and re-analysed by DSF (Fig 3A–3C). The DSF data obtained with partially pure proteins in Fig 2B were largely recapitulated with the purified S2P-FHA proteins. Increasing proportions of the 58˚C form relative to the 43.6˚C form was observed for the mutants as fol- lows: A1016V/A1020I (VI) = A1016I/A1020I (II) > A1016L (16L) > A1016V (16V) > S2P-FHA (Fig 3D). Interestingly, this hierarchy of stability was inversely correlated to putative trimer yield (Fig 3A and 3C). Functional properties of Ala cavity mutants We next examined the effects of mutating Ala1016 and Ala1020 on the membrane fusion func- tion of the ancestral SARS CoV-2 S glycoprotein. A1016V, A1016L, A1020V, A1020L and A1016V/A1020I mutations were introduced to the WH-Human1_EPI_402119 expression plasmid bearing codon-optimized full-length S. A western blot confirmed that the WT and mutated S glycoproteins were expressed and cleaved to S1 following transfection of 293T cells PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 5 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein Fig 2. Expression and thermostability screen of Ala cavity mutants. A, Coomassie blue stained S2P-FHA mutants purified from culture supernatants with TALON resin following SDS-PAGE under reducing conditions. The Tms of S2P-FHA mutants obtained in B are indicated below each lane. Bold type: major species; small font: minor species; nd: not determined. B, Differential scanning fluorimetry of S2P-FHA proteins shown in A using SYPRO Orange. The rate of change of fluorescence over time [–d(RFU)/dt] as a function of temperature is shown. Graphs are representative of at least two independent experiments. https://doi.org/10.1371/journal.ppat.1010981.g002 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 6 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein Fig 3. Purification and characterization of selected Ala cavity mutants. A, Superose 6 SEC of S2P-FHA proteins following elution from TALON resin. The calibration standards are thyroglobulin (669 kDa), ferritin, (440 kDa) and IgG (150 kDa). B, Superose 6 SEC of purified trimers following a freeze (-80˚C)-thaw cycle. C, Differential scanning fluorimetry of purified S2P-FHA trimers. D, SDS-PAGE and Coomassie blue staining of purified proteins under nonreducing (top) and reducing (bottom) conditions. https://doi.org/10.1371/journal.ppat.1010981.g003 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 7 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein with the WH-Human1_EPI_402119 plasmids (Fig 4A). A cell-cell fusion assay was established to measure the membrane fusion function of the S glycoproteins. 293T effector cells were cotransfected with S expression vectors, a bacteriophage T7 RNA polymerase expression plas- mid (pCAG-T7) [49], a furin expression plasmid (pcDNA3.1-Furin) [50] and a T7 promotor- driven Gaussia princeps luciferase reporter plasmid (pTM-Gaussia) to monitor T7 polymerase expression in the effector cells. 293-ACE2 target cells [51] were transfected with a T7 pro- moter-driven firefly luciferase reporter plasmid (pTMluc) [52] and a TMPRSS2 expression plasmid [53]. At 36.5 h post transfection, the supernatants of effector cell cultures were sam- pled for Gaussia luciferase activity after which effector and target cells were cocultured for 3 h. Firefly luciferase activity was measured and normalized against relative Gaussia luciferase activity (S2A Fig) to account for differences in T7 polymerase expression in effector cells. The Fig 4. Effects of Ala cavity mutations on fusion function. A, Reducing SDS-PAGE and western blotting of SARS CoV-2 S glycoproteins expressed in 293T cells with rabbit anti-S1 polyclonal antibody. B, Cell-cell fusion activity of S glycoproteins determined in a luciferase reporter assay. Firefly luciferase relative light units obtained for each coculture was normalized according to Gaussia luciferase activity produced by the corresponding well of effector cells used for each coculture. Relative Gaussia activity was first obtained by dividing Gaussia RLU of each transfection (see S2A Fig) by Gaussia RLU produced by the WT transfection. Firefly luciferase relative light units were then divided by relative Gaussia activity produced by the corresponding effector cells. The mean ± SEM from at least 3 independent experiments shown. ns, not significant, , P < 0.05, * P < 0.01, versus WT, Kruskal-Wallis test. C, Representative microscopy fields at 10x magnification of cocultures prior to lysis and firefly luciferase assay. D, Cell-cell fusion activity of S glycoproteins determined in a fluorescence assay. The nuclei of S-expressing 293T effector cells stained with Hoechst 33342 (blue) were cocultured for 24 h with ACE-293 target cells expressing EGFP (green). Images were acquired on the ImageXpress Pico Automated Cell Imaging System (Molecular Devices) using a 10x objective. Images were analysed using ImageJ Version 2.9.0/1.53t. Control vectors: S2P-1273 contains proline at positions 986 and 987 and lacks a furin cleavage site; HIV-1 Env: a HIV-1 glycoprotein expression vector, pcDNA3.1AD8-WT. https://doi.org/10.1371/journal.ppat.1010981.g004 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 8 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein firefly luciferase assay revealed that all S mutants possessed fusion activity that was not signifi- cantly different to that of WT (Fig 4B). Two control plasmids, S2P-1273, which expresses the full-length S glycoprotein containing Pro at positions 986 and 987 and a mutated cleavage site, and an HIV-1 Env expression plasmid, pcDNA3.1AD8-WT, lacked fusion activity. No lucifer- ase activity was detected by the firefly luciferase substrate when WT S effector cells were cocul- tured with 293-ACE2 cells in which the pTMluc plasmid was replaced with pcDNA3 empty vector (Fig 4B, WT/noFFluc). Consistent with the production of luciferase due to S-mediated fusion between effector and target cells, large multinucleated syncytia were observed for all S constructs in cocultures prior to lysis and luciferase assay but not for the controls S2P-1273 and HIV-1 Env (Fig 4C) A fluorescence-based assay was established to aid in the visualization of cell-cell fusion. 293T effector cells were cotransfected with WT or mutant WH-Human1_EPI_402119 S expression plasmids and pcDNA3.1-Furin, whereas 293-ACE2 target cells were cotransfected with TMPRSS2 plasmid and pEGFP-N1. At 36.5 h posttransfection, the nuclei of effector cells were stained with Hoechst 33342, washed, detached and cocultured with EGFP-expressing tar- get cells in black-walled 96-well culture plates for 24 h prior to visualization in an ImageXpress Pico cell imaging system. Syncytia with diffuse EGFP cytoplasmic staining and deep teal nuclear staining were observed for all S constructs, except S2P-1273 (Figs 4D and S2B and S2C). In S2P-1273 and HIV-1 Env control wells, blue and green fluorescence was intense and associated with distinct single cells. The data indicate that Ala cavity filling mutations do not adversely affect membrane fusion function. S-pseudotyped HIV-1 luciferase reporter viruses [54] were prepared to examine the ability of S mutants to mediate entry of viral pseudoparticles into 293-ACE2 cells. The entry activities of A1016V, A1020V and A1020L pseudotypes were not significantly different to that of WT particles whereas A1016L and A1016V/A1020I particles lacked entry ability (S3A Fig). West- ern blotting of pelleted pseudoparticles from the same stock as used in entry assays indicated the similar levels of HIV-1 p24/CA in all cases (S3B Fig). To detect the presence of S, pseudo- particles were pelleted through a sucrose cushion followed by western blotting with S1-specific polyclonal antibody. Bands consistent with S1 were observed for WT, and all mutants but not for empty particles produced in the absence of an S expression vector (S3C Fig). Whereas the stabilizing A1016L and A1016V/A1020I mutations do not adversely affect the membrane fusion function of S, these mutations block the ability of S to mediate entry of HIV-1 pseudo- particles into 293-ACE2 cells. Immunogenicity of Ala cavity mutants Guinea pigs were used to examine whether the Ala cavity mutations A1016L (16L) and A1016V/A1020I (VI) can affect the magnitude and specificity of antibody responses to S2P-FHA putative trimers. Outbred guinea pigs were immunized with 30 μg of S2P-FHA, S2P.16L-FHA and S2P.VI-FHA in Addavax adjuvant at weeks 0, 4 and 14 and bleeds per- formed at weeks 6 and 16 (Fig 5A). The neutralizing activity in vaccinal sera was determined using S-pseudotyped HIV luciferase reporter viruses and 293-ACE2 target cells as described previously [54]. A comparison of week-6 and week-16 sera (bleeds taken 2 weeks after the 1st and 2nd boosts, respectively) using pseudotypes containing ancestral (Hu-1) S glycoprotein indicated neutralizing activity in S2P-FHA-, S2P.16L-FHA- and S2P.VI-FHA-immune sera with mean ID50s ranging from 1,700–1,900 for week-6 sera and 6,000–9,100 for week-16 sera. These data equate to ~3–5.4-fold increases in mean neutralization ID50 following the 2nd boost, although statistical significance was not reached for S2P-FHA- and S2P.VI-FHA- immune sera (Fig 5B). A capture ELISA, employing plate-bound avidin to capture PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 9 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein Fig 5. Immunogenicity of Ala cavity mutants. A, Immunization protocol. B, Pseudovirus neutralization ID50s of vaccinal sera obtained at weeks 6 and 16. The geometric mean pseudovirus neutralization ID50 of the control group of 8 animals receiving 3 doses of 50% Addavax-PBS was <1/120 at both time-points. C and D, ELISA binding titers of week-16 vaccinal sera to streptavidin-captured biotinylated RBDs or S2P-FHA trimers, respectively, derived from ancestral Hu-1, Delta and Omicron BA.1 (indicated below the graphs). The endpoint was determined as 5-times background optical density obtained in the absence of primary antibody. The horizontal bars are the geometric means. The horizontal dotted line is the geometric mean binding titer of the control group receiving 3 doses of 50% Addavax-PBS. A Kruskal-Wallis test was used to determine whether the differences in ID50s or binding titers are significantly different. ns, not significant; , P < 0.05; , P < 0.01. https://doi.org/10.1371/journal.ppat.1010981.g005 biotinylated ancestral, Delta and Omicron BA.1 RBDs, was used to determine the RBD bind- ing titers of vaccinal sera. Fig 5C shows no significant differences in binding profile between immunogen groups with geometric mean reciprocal binding titers in the range 3.5x104– 1.36x105. A small (~3-fold) but significant reduction in reciprocal binding titer was observed for the Omicron BA.1 RBD for all 3 immunogen groups. Serum binding titers to ancestral, Delta and Omicron BA.1 S2P-FHA spike proteins bound directly to ELISA plates were next determined. Fig 5D indicates no significant differences in reciprocal binding titers to the 3 S2P-FHA variants for the 3 immunogen groups, which were in the range 3.0x105-1.1x106. These results suggest that the 3 immunogens generated similar titers of antibody able to bind the S2P-FHA spike. We next assessed neutralizing activity against pseudotypes containing ancestral, Delta or Omicron BA.1 S glycoprotein and 293-ACE2 cells [54]. Potent neutralizing activity against ancestral and Delta pseudoviruses was observed with S2P-FHA-, S2P.16L-FHA- and S2P. VI-FHA-immune sera with mean ID50s ranging from 3,900–5,100 (Fig 6A and S1 Table). Serum neutralizing activities against pseudoviruses containing the Omicron BA.1 Spike were not significantly different relative to Hu-1 and Delta, although the IC50s against Omicron BA.1 trended downwards with < 3-fold reductions in mean titer observed for the 3 groups. The neutralizing activity of sera against authentic ancestral, Delta, and Omicron BA.1 SARS-CoV-2 using HAT-24 cells in the R-20 microneutralization assay developed by Aggar- wal et al. [33] was examined. Whereas ancestral clade A2.2 and Delta viruses were potently neutralized by sera from the 3 immunogen groups, Omicron BA.1 virus neutralization was reduced by ~1log10 (Fig 6B). In Fig 6C, log10 reciprocal RBD binding titers were plotted against log10 neutralization ID50 for individual sera and color-coded according to SARS CoV- 2 variant. Strong and highly significant correlations between RBD binding and authentic PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 10 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein Fig 6. Neutralization activity of immune sera. A) Pseudovirus neutralization ID50s of week-16 vaccinal sera. The S variant genotypes used in S-HIV pseudotype assays are shown below the x axis. Neutralization ID50 obtained with sera from individual animals are indicated with various symbols. The bars are geometric mean ID50s. The geometric mean pseudovirus neutralization ID50 of the control group receiving 3 doses of 50% Addavax-PBS was <1/200. B) Neutralisation assays performed in high-throughput format with authentic SARS-CoV-2 variants. The SARS-CoV-2 variants are shown below the x axis. Horizontal bars are the geometric mean ID50s for each immunogen group. The geometric mean neutralization ID50 of the control group receiving 3 doses of 50% Addavax-PBS was <1/320 for Ancestral and Delta and <1/40 for Omicron BA.1. A Kruskal-Wallis test was used to determine whether the differences in ID50s observed between variants was significant: ns, not significant; , P < 0.05; *, P < 0.01. C) Correlations PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 11 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein between RBD binding titer (obtained in Fig 5C) and pseudovirus neutralization ID50 (top) and authentic virus ID50 (bottom). The data points are color-coded according to variant. Spearman ρ and P values were determined using GraphPad Prism v9.3.0. https://doi.org/10.1371/journal.ppat.1010981.g006 SARS CoV-2 neutralization (top panel) and to a lesser extent pseudoviral neutralization (bot- tom panel) were observed in keeping with the known important role that RBD-directed anti- bodies play in SARS CoV-2 neutralization. This analysis suggests that the 15 mutations in the RBD of Omicron BA.1 contribute to the decreased neutralization potency of ancestral S2P-FHA-elicited sera for Omicron BA.1 virus. Specificity of antibody responses An ELISA employing S2P-FHA and S subdomains that contain known neutralization epitopes was employed to examine the specificity of antibodies elicited by the 3 spike vaccines. The sub- domains included S1 (amino acids 16–681), the NTD (amino acids 16–305), the RBD (amino acids 332–532), a synthetic fusion peptide (amino acids 808–832 corresponding to the epitope of mNAb COV44-79 [23]) and a synthetic peptide derived from the stem region of S2 (amino acids 1142–1165 corresponding to the epitope of CV3-25 [25]). Mean reciprocal binding titers of greater than 105 were observed in sera from the 3 immunogen groups against S2P-FHA, S1 and RBD antigens, whereas reciprocal binding titers approaching 103 and in the range 104−4.3 were observed with NTD and stem peptide, respectively (S4A Fig). Low-level specific binding to the synthetic fusion peptide was observed only for S2P.VI-FHA immune sera. The reactivity of S2P-FHA and its subdomains with human mNAbs is shown in S4B Fig. S2P-FHA was bound by mNAbs directed to the RBD (COVOX222), NTD (C1520), fusion peptide (COV44-79) and stem CV3-25 but not by the HCV E2-specific mNAb HCV1 [55]. S1 was specifically bound by COVOX222 and C1520, NTD by C1520, the fusion peptide by COV44-79 and the stem peptide by CV3-25. A serum-mNAb cross-competition ELISA was employed as a complementary approach to gain an understanding of the antigenic sites within S that are targeted by vaccinal antibodies. This experiment was included to account for epi- topes that may depend on S quaternary structure. Biotinylated ancestral Hu-1 S2P-FHA cap- tured onto streptavidin coated ELISA plates was incubated with a mixture comprising a sub- saturating amount of human mNAbs or ACE2-Fc and a dilution series of sera. The assay was developed with anti-human F(ab’)2-HRP for the mNAbs or anti-human IgG-HRP for ACE2-Fc. The data (S4C Fig) show that S2P-FHA-, S2P.16L-FHA- and S2P.VI-FHA-elicited sera contained antibody specificities with largely equivalent abilities to block binding by ACE2-Fc and mNAbs directed to the ACE2-binding site (COVOX222), the stem region of S2 (CV3-25) and an undefined epitope in S (COVA1-25) with mean reciprocal blocking titers being �103. Weaker inhibitory activities were observed with C1520 to the NTD and COV44- 79 to the fusion peptide, consistent with relatively low binding titers to the corresponding S fragments shown in S4A Fig. The binding curves of sera from individual animals to streptavi- din-captured S2P-FHA in the absence of competitor indicate that there are no major differ- ences in S2P trimer binding ability between the 3 immunogen groups (S4D Fig). The data indicate that broad specificity antibody responses were elicited by the 3 spike antigens. The VI mutation confers stability to the Omicron BA.1 S2P-FHA oligomer Thirty amino acid changes occur in the Omicron BA.1 Spike compared to the ancestral strain. Eight are in the NTD and 15 in the RBD, consistent with decreased neutralization of Omicron BA.1 by human vaccinee sera [33–36,56] and low vaccine efficacy against Omicron infection [30]. One approach being taken to improve the efficacy of COVID-19 vaccines against PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 12 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein Fig 7. VI stabilizes S2P-FHA trimers derived from Omicron BA. 1. A, Superose 6 SEC profiles of S2P. OmiBA1-FHA and, S2P.OmiBA1.VI-FHA eluted from TALON or HiTRAP columns, respectively. B, Superose 6 SEC profiles of presumed trimers obtained in A. C, Differential scanning fluorimetry of purified S2P.OmiBA1-FHA and, S2P.OmiBA1.VI-FHA trimers. D, SDS-PAGE/Coomassie blue staining of purified trimers under non-reducing and reducing (1% v/v betamercaptoethanol) conditions. The samples were not boiled prior to electrophoresis. m, markers. E, SDS-PAGE of trimers under non-reducing and reducing (1% v/v betamercaptoethanol) conditions after boiling for 5 min. The position of the S2P.omicron.VI-FHA major band under the various conditions is indicated with an arrow. m, markers. F, SDS-PAGE of S2P-FHA, S2P.OmiBA1-FHA, S2P.Omi.VI-FHA trimers under non-reducing (left) and reducing (1% v/v betamercaptoethanol) (right) conditions. Sample buffer containing SDS was added to the samples PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 13 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein with and without 1% betamercaptoethanol and the samples were either left at room temperature (22˚C) or boiled (100˚C) for 3 min prior to electrophoresis. Thyroglobulin that had been chemically crosslinked with 1 mM bis (sulfosuccinimidyl)suberate was included to mark the expected position (669 kDa) of the S2P.OmiBA1.VI-FHA oligomer (S2PO). https://doi.org/10.1371/journal.ppat.1010981.g007 Omicron subvariants is to include Omicron-derived sequences in second-generation vaccines [37,38]. This prompted an assessment of whether the Omicron BA.1 Spike could be stabilized by mutations in the alanine cavity. Omicron BA.1 versions of S2P-FHA and S2P.VI-FHA con- structs were prepared and are referred to as S2P.OmiBA1-FHA and S2P.OmiBA1.VI-FHA, respectively. The His681ArgAlaArgArg furin site was changed to Pro681GlySerAlaSer in both constructs. The glycoproteins were expressed in Expi293F cells, extracted by divalent cation affinity chromatography and further purified by Superose 6 SEC. S2P.OmiBA1-FHA eluted as a major peak close to the position of thyroglobulin (669 kDa) (Fig 7A, top panel) and had an almost identical profile to that of S2P-FHA derived from the Hu-1 isolate (see Fig 3). The putative tri- mer was collected as a homogenous protein as indicated by analytical SEC (Fig 7B, top panel). S2P.OmiBA1.VI-FHA eluted as 4 species including a prominent presumed trimer peak (dashed box, Fig 7A, bottom panel). Fractions corresponding to S2P-foldon trimer were pooled, concentrated and re-chromatographed in a Superose 6 column following a freeze (-80˚C)-thaw cycle, revealing a largely homogeneous species consistent with trimeric structure (Fig 7B). DSF indicated that the purified S2P.OmiBA1-FHA possessed higher thermal stability relative to its ancestral Hu-1 counterpart with a Tm of 61˚C versus 43.6˚C for the latter (Fig 7C, top panel). The introduction of VI increased the Tm to 65˚C for S2P.OmiBA1.VI-FHA (Fig 7C, lower panel). The purified S2P.OmiBA1-FHA and S2P.OmiBA1.VI-FHA oligomers were further ana- lysed by SDS-PAGE under non-reducing and reducing (1% betamercaptoethanol) conditions in the absence of sample boiling (Fig 7D). S2P.OmiBA1-FHA was largely resolved to its expected monomer molecular weight of ~ 160–180 kDa in both the presence and absence of reducing agent. By contrast, the major S2P.OmiBA1.VI-FHA species (indicated by an arrow) was retained as a high molecular weight species migrating close to the top of the gel; a minor species was also observed at the monomer position. Fig 7E shows a repeat experiment in which the samples were boiled for 5’ prior to electrophoresis. Under non-reducing conditions, the results seen in Fig 7D were largely recapitulated with the high molecular weight major S2P.OmiBA1.VI-FHA species again observed. However, boiling in the presence of reducing agent resolved this species to its monomer molecular weight. The data suggest that the S2P. OmiBA1.VI-FHA oligomer resists disruption by 0.8% w/v sodium dodecyl sulfate denaturant with and without 1% betamercaptoethanol at 22˚C. To obtain an SDS-PAGE marker with a theoretical molecular weight of 669 kDa, which is close to that of the S2P-foldon trimer in SEC [5], thyroglobulin subunits were covalently crosslinked with bis(sulfosuccinimidyl) suberate. Fig 7F shows that crosslinked thyroglobulin co-migrated with the SDS/betamercaptoethanol- resistant high molecular weight form of S2P.OmiBA1.VI-FHA following treatment with 0.8% SDS at 22˚C or 100˚C for 3 min, or 0.8% SDS + 1% betamercaptoethanol at 22˚C for 3 min prior to electrophoresis. Again, S2P.OmiBA1.VI-FHA resolved as a monomer following boil- ing in 0.8% SDS + 1% betamercaptoethanol. S2P.OmiBA1-FHA and Hu-1 S2P-FHA proteins migrated as monomers following all treatments except for some residual putative trimeric S2P-FHA after treatment with 1% SDS at 22˚C. The data illustrate the stabilizing effect of the VI mutation on quaternary structure. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 14 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein Epitope profiles of Ala cavity mutants Next, biolayer interferometry was used to compare the effects of VI on the exposure of epitopes recognized by human mNAbs and ACE2-Fc in S2P-FHA putative trimers derived from ances- tral Hu-1 and Omicron BA.1. The human mNAbs examined were C1520 directed to the NTD [57], Omi-18 and Omi-42 to the RBM [35], S2H97 [18] and SP1-77 [58] to conserved epitopes within the RBD excluding the RBM, COV44-79 to the fusion peptide [23], and CV3-25 to the stem of S2 [24,25]. ACE2-Fc and the human mNAbs were attached to anti-human IgG Fc cap- ture biosensors while the S2P glycoproteins were in the analyte phase. The sensograms shown in S5 Fig correlate with the 1:1 bimolecular interaction model with R2 values being >0.93 and χ2 values being <1 for all but one case (Table 1). In all cases, affinity constants (KD) of less- than 4x10-9M were observed with most interactions achieving less-than 10-12M. A comparison of dissociation rates (kdis) showed that the introduction of VI to the ancestral S2P-FHA stabi- lized the interaction with mNAbs directed to the NTD (C1520), RBD (S2H97), fusion peptide (COV44-79) and to the S2 stem (CV3-25) which translated to an ~ 2log10 improvement in affinity constant in most cases (Table 1). This effect of the VI mutation was not observed in the context of S2P.OmiBA1-FHA. The data suggest that the improved thermostability of the ancestral S2P.VI-FHA trimer is associated with stabilization of its interaction with key mNAbs, however, this is not observed with its Omicron BA.1-derived counterpart. The VI mutation stabilizes the Omicron BA.1 and BA.4/5 S2P ectodomain in oligomeric form in the absence of an external foldon trimerization domain We next determined whether the VI mutation could stabilize the ancestral and Omicron BA.1 S ectodomains (residues 16–1208) in a form consistent with S2P-foldon trimers [5] in the absence of the C-terminal foldon motif. The last residue of the ectodomain, Q1208, was appended with GlySerGlySer-His8 to generate S2P-1208.H8, S2P.VI-1208.H8 (derived from ancestral Hu-1) and S2P.OmiBA1-1208.H8 and S2P.OmiBA1.VI-1208.H8 (derived from Omi- cron BA.1). Superose 6 SEC revealed that the ancestral S2P-1208.H8 co-eluted with ferritin (440 kDa) suggesting that it was secreted from transfected 293F cells as a lower-order species whereas S2P.VI-1208.H8 was purified in a form that was consistent with a trimer [5], as indi- cated by its coelution with thyroglobulin (669 kDa) (Fig 8A). The purified ancestral S2P.VI- 1208.H8 putative trimer was reanalyzed by Superose 6 SEC following a freeze (-80˚C)-thaw cycle revealing that ~ 50% of the S2P.VI-1208.H8 protein had dissociated to lower order spe- cies (Fig 8B). DSF indicated that S2P.VI-1208.H8 comprised 2 species with melting tempera- tures of 43˚C and 58˚C, respectively (Fig 8C). The data indicate that whereas VI enables purification of the ancestral S2P ectodomain as a putative trimer in the absence of an external trimerization tag, a large proportion is unstable, dissociating into lower-order species after a freeze-thaw cycle. Both S2P.OmiBA1-1208.H8 and S2P.OmiBA1.VI-1208.H8 were obtained in a form consis- tent with a trimer [5] in the absence of the external foldon trimerization tag (Fig 8A). Superose 6 SEC of the putative trimers following a freeze (-80˚C)-thaw cycle revealed that ~ 15% of the S2P.OmiBA1-1208.H8 protein had dissociated to a lower-order form. By contrast, the putative trimeric structure of S2P.Omi1.VI-1208.H8 was retained (Fig 8B). A thermofluor assay indi- cated that the majority of purified S2P.OmiBA1-1208.H8 had a melting temperature of 41˚C, whereas the melting temperature of S2P.OmiBA1.VI-1208.H8 was much higher at 63˚C (Fig 8C). The data indicate that the S2P.OmiBA1 ectodomain is intrinsically unstable requiring the foldon domain for thermal stability. Addition of VI to the Omicron BA.1 ectodomain enables PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 15 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein Table 1. Binding kinetics of Spike trimers to ACE2-Fc and human monoclonal NAbs. Ligand S2P glycoprotein a Req (nm) R2b χ2c KD (M) C1520 S2P-FHA 0.74 0.988 0.815 5.89x10- 11 kon (1/ Msec) 5.4x105 kdis (1/ sec) S2P glycoprotein 3.2x10-5 OmiBA1-FHA a Req (nm) 0.82 R2b χ2c KD (M) 0.999 0.135 <10−12 kon (1/ Msec) 3.3x105 kdis (1/ sec) <10−7 S2P.VI-FHA 0.7 0.997 0.113 <10−12 3.8x105 <107 OmiBA1. VI-FHA 0.84 0.996 0.495 <10−12 4.5x105 <10−7 ACE2-Fc S2P-FHA 0.26 0.994 0.056 1.65x10-9 4.3x105 7.0x10-4 OmiBA1-FHA 0.35 0.994 0.104 S2P.VI-FHA 0.2 0.995 0.017 4.22x10 9 1.02x105 Omi-18 S2P-FHA S2P.VI-FHA Omi-42 S2P-FHA S2P.VI-FHA 0.84 0.65 0.74 0.61 0.991 0.812 <10−12 0.999 0.051 <10−12 0.995 0.431 <10−12 0.999 0.047 <10−12 4.4x105 2.1x105 5.1x105 2.3x105 4.3x10-4 OmiBA1. VI-FHA <10−7 OmiBA1-FHA <10−7 OmiBA1. VI-FHA <10−7 OmiBA1-FHA <10−7 OmiBA1. VI-FHA 0.3 0.998 0.028 6.41x10- 10 1.87x10- 10 4.2x105 2.7x10-4 2.7x105 5.1x10-5 0.63 0.79 0.49 0.56 0.998 0.112 <10−12 0.996 0.184 <10−12 3.0x105 4.0x105 <10−7 <10−7 0.999 0.014 <10−12 4.82x10- 0.998 0.081 11 8.5x102 1.8x103 <10−7 1.1x10-5 S2H97 S2P-FHA 0.9 0.981 1.12 2.03x10- 10 6.1x105 1.2x10-4 OmiBA1-FHA 0.82 0.998 0.19 <10−12 2.4x105 <10−7 S2P.VI-FHA 0.45 0.987 0.39 <10−12 2.1x105 SP1-77 S2P-FHA S2P.VI-FHA 0.8 0.68 0.986 0.953 <10−12 0.995 0.167 <10−12 9.6x105 4.0x105 <10−7 OmiBA1. VI-FHA <10−7 OmiBA1-FHA <10−7 OmiBA1. VI-FHA 0.78 0.999 0.159 <10−12 2.6x105 <10−7 0.76 0.74 0.998 0.179 <10−12 0.32 <10−12 0.996 2.6x105 3.4x105 <10−7 <10−7 COV44- 79 S2P-FHA 0.12 0.988 0.019 5.55x10-9 1.9x105 1.1x10-3 OmiBA1-FHA 0.23 0.934 0.02 <10−12 2.0x104 <10−7 S2P.VI-FHA 0.09 0.99 0.025 <10−12 8.8x104 <10−7 OmiBA1. VI-FHA 0.31 0.992 0.02 <10−12 5.3x104 <10−7 CV3-25 S2P-FHA 0.55 0.986 0.359 2.48x10- 10 4.2x105 1.1x10-4 OmiBA1-FHA 0.48 0.999 0.036 <10−12 1.4x105 <10−7 S2P.VI-FHA 0.29 0.99 0.073 <10−12 1.3x105 Ligand C1520 S2P glycoprotein OmiBA1.VI- 1208 Req (nm) R2 χ2 KD (M) kon (1/Ms) 1.04 0.993 0.938 1.13x10- 10 6.9x105 ACE2-Fc OmiBA1.VI- 0.2 0.997 0.02 2.31x10-9 3.2x105 Omi-18 Omi-42 S2H97 SP1-77 COV44- 79 CV3-25 1208 OmiBA1.VI- 1208 OmiBA1.VI- 1208 OmiBA1.VI- 1208 OmiBA1.VI- 1208 OmiBA1.VI- 1208 OmiBA1.VI- 1208 0.83 0.99 0.858 3.01x10- 12 7.1x105 0.63 0.996 0.245 <10−12 3.5x105 0.62 0.999 0.02 1.66x10- 10 1.6x105 0.88 0.996 0.493 <10−12 5.3x105 0.1 0.986 0.008 4.41x10- 11 1.2x105 0.27 0.97 0.251 <10−12 2.6x105 <10−7 OmiBA1. VI-FHA kdis (1/ sec) S2P glycoprotein 7.8x10-5 OmiBA4/5.VI- 1208 7.3x10-4 OmiBA4/5.VI- 1208 2.2x10-6 OmiBA4/5.VI- 1208 <10−7 OmiBA4/5.VI- 1208 2.7x10-5 OmiBA4/5.VI- 1208 <10−7 OmiBA4/5.VI- 1208 5.4x10-6 OmiBA4/5.VI- 1208 <10−7 OmiBA4/5.VI- 1208 0.51 0.999 0.015 2.63x10- 10 1.3x105 3.5x10-5 Req (nm) R2 χ2 KD (M) 1 0.991 1.19 <10−12 kon (1/ Ms) 6.8x105 kdis (1/ sec) <10−7 0.15 0.995 0.016 3.72x10-9 2.3x105 8.5x10-4 0.69 0.997 0.17 1.06x10-9 4.8x105 5.1x10-4 0.65 0.997 0.19 0.64 0.999 0.03 9.33x10- 11 5.27x10- 10 4.8x105 4.5x10-5 3.2x105 1.7x10-4 0.91 0.996 0.527 <10−12 5.5x105 <10−7 0.08 0.986 0.008 4.41x10- 11 1.2x105 5.4x10-6 0.34 0.999 0.018 3.09x10-9 1.8x105 5.5x10-4 aReq: Response units at binding equilibrium bR2: Square of the coefficient of correlation between sensogram data and 1:1 bimolecular binding model cχ2: Sum of the squared deviations: sensogram data versus 1:1 bimolecular binding model https://doi.org/10.1371/journal.ppat.1010981.t001 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 16 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein Fig 8. The VI mutation stabilizes the trimerization of the omicron BA.1 and BA.4/5 S ectodomain in the absence of the T4 foldon trimerization motif. A) Superose 6 size exclusion chromatography of S2P-1208.H8 proteins purified from 293F (ancestral) or Expi293F (Omicron BA.1 and BA.4/5) cells by hiTRAP affinity chromatography. B, Superose 6 size exclusion chromatography of S2P-1208.H8 glycoproteins purified in A following a freeze (-80˚C)-thaw cycle. C, Differential scanning fluorimetry of purified S2P-1208.H8 proteins following a freeze (-80˚C)-thaw cycle performed using SYPRO Orange. D, SDS-PAGE and Coomassie blue staining of purified proteins under reducing conditions. WT: Ala at amino positions 1016 and 1020; VI: Val and Ile at amino positions 1016 and 1020, respectively. m = markers. https://doi.org/10.1371/journal.ppat.1010981.g008 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 17 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein highly stable soluble oligomers to be obtained thereby obviating the requirement for an exoge- nous trimerization domain to maintain quaternary structure and relative thermal stability. The Omicron BA.4 and BA.5 subvariants recently evolved from the Omicron BA.2 VOC lineage and were briefly the dominant variant due to an apparent transmission advantage and immune evasion properties [59]. BA.4 and BA.5 S amino acid sequences are identical. In com- parison to BA.1, 5/6 mutations within the NTD and 5/17 in the RBD are unique to BA.4/5; L452R and F486V within the BA.4/5 RBD are believed to be responsible for its ability to evade immunity due to vaccination and/or previous infection by Omicron BA.1 [60]. Current biva- lent mRNA booster vaccines therefore comprise the ancestral and BA.4/5 spike sequences [39]. The S2 subunits of the BA.4/5 spike trimer exhibit relatively tight packing, which may impose steric restrictions to NAb binding [61]. We therefore prepared S2P.OmiBA.4/5-1208. H8 and S2P.OmiBA.4/5.VI-1208.H8 constructs to determine if they could also form stable oligomers consistent with trimeric structure. The proteins were obtained largely as oligomers coeluting with thyroglobulin consistent with the trimeric structure of S2P-foldon [5] following HiTRAP divalent affinity chromatography of transfected Expi293F culture supernatants (Fig 8A). The putative trimeric structure of S2P.OmiBA.4/5.VI-1208.H8 was retained following a freeze-thaw cycle, whereas a significant proportion of S2P.OmiBA.4/5-1208.H8 had dissoci- ated to smaller molecular-weight species after the treatment (Fig 8B). These data were reflected in the thermofluor assay which revealed 2 species for S2P.OmiBA.4/5-1208.H8 with Tms of 43˚C and 59˚C, whereas S2P.OmiBA.4/5.VI-1208.H8 presented as a single species with a relatively high Tm of 61˚C (Fig 8C). Reducing SDS-PAGE and Coomassie blue staining indi- cated a single ~160-180kDa band for all constructs (Fig 8D). We asked whether the VI muta- tion can overcome the requirement for the 2P mutation for efficient S2P.OmiBA.4/5-1208.H8 putative trimer expression and thermostability. Pro986Pro987 were mutated back to the native amino acids Lys and Val, respectively in SnoP.OmiBA4/5.VI-1208.H8. The P986K/P987V reversion led to an ~ 10-fold reduction in yield of spike oligomer relative to when S2P was present (S6A Fig), consistent with previous observations with MERS and SARS CoV Spike gly- coproteins [62]. DSF performed on SnoP.OmiBA4/5.VI-1208.H8 trimer indicated a melting temperature of 61˚C, which is equivalent to that of S2P.OmiBA.4/5.VI-1208.H8 (S6B Fig). The S2P mutation is therefore essential for efficient expression of the omicron BA45 S ectodo- main containing VI but not for its thermostability. The data indicate that VI provides a means for producing soluble S2P glycoprotein putative trimers derived from Omicron subvariants in the absence of foreign, potentially immunogenic trimerization sequences. S2P.OmiBA.1.VI-FHA, S2P.OmiBA1.VI-1208.H8 and S2P.OmiBA4/5.VI-1208.H8 were subjected to negative stain electron microscopy (EM). The proteins are relatively homoge- neous in sizes and overall shapes with no sign of aggregation. The 2D class averages shown in S7 Fig indicate that the majority of the glycoproteins are in a conformation compatible with a pre-fusion trimeric form with only a minor fraction of S2P.OmiBA.1.VI-FHA (8%) and S2P. OmiBA1.VI-1208 (8.8%) observed in putative post-fusion forms (rod-shaped molecules indi- cated by red boxes) [4,5,9,63–65]. Rod-shaped molecules that may represent post-fusion forms were not observed for S2P.OmiBA4/5.VI-1208.H8. ACE2-Fc and broadly neutralizing monoclonal antibodies were used to reveal the epitope profile of S2P.OmiBA1.VI-1208.H8 and S2P.OmiBA4/5.VI-1208.H8 proteins in biolayer inter- ferometry. The sensograms shown in S5 Fig all correlate with a 1:1 bimolecular binding model with R2 values being �0.97 and χ2 values being �1 for all cases (Table 1). In all cases, KD val- ues of � 3.74x10-9M were achieved, the most stable interactions being S2P.OmiBA1.VI-1208 with Omi-42, SP1-77 and CV3-25 and between S2P.OmiBA4/5.VI-1208 and C1250, and SP1- 77. The data indicate that key conserved neutralization epitopes within the NTD, RBD, fusion peptide and stem are exposed in S2P.OmiBA1.VI-1208.H8 and S2P.OmiBA.45.VI-1208.H8 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 18 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein oligomers. The foldon trimerization clamp is thus not required for the formation and presen- tation of key conserved neutralization epitopes in VI-stabilized S2P putative trimers. Discussion Class I fusion glycoproteins from a number of viral families comprise a central trimeric coiled coil that acts as a scaffold for the conformational changes required for the membrane fusion process [4,5,9,40–42,45–47,66]. In orthomyxoviruses and retroviruses, a 3–4 repeat of largely hydrophobic residues mediates knobs-into-holes interhelical contacts to help stabilize the tri- mer. By contrast, the coiled coil at the center of the SARS-CoV-2 prefusion S glycoprotein tri- mer is formed by 3 bow-shaped helices that expand away from each other from a point of contact mediated by the inward-facing Ile1013 and Leu1012. The remainder of the 3–4 repeat is largely comprised of polar residues that mediate few inter-helical contacts. This unusual structure may contribute to the relatively low thermostability of the ancestral prefusion S tri- mer, as evidenced by the Tm of 43.6˚C observed here, even in the presence of the S2P muta- tion, shown to stabilize SARS CoV-2 S trimers in the pre-fusion conformation [5], and the C- terminal T4 foldon trimerization domain. In this study, we found that the ancestral S2P trimer could be further stabilized by the creation of an artificial hydrophobic core in the center of the coiled coil of S2 CH helices by replacing Ala1016 and Ala1020, that form part of the 3–4 repeat, with bulkier hydrophobic residues to fill the cavity associated with these Ala residues. Mutagenesis of Ala1016 shifted the Tm of S2P-FHA putative trimers from 43.6˚C to 58˚C with A1016L (‘16L’) giving the highest proportion of the 58˚C species. By contrast, the 43.6˚C species was the predominant form with substitutions of Ala1020, including A1020L. The more pronounced stabilizing effect of the 1016 versus 1020 hydrophobic substitutions may be due to the former’s proximity to Ile1013 and substitutions at this position will enlarge the Leu1012-Ile1013 hydrophobic network that forms in the core of the coiled coil trimer. Double substitutions were all associated with the stable form, except for 1016/20VV. Interestingly, increased thermostability was generally associated with decreased soluble S2P-FHA expression suggesting that the bulk and/or geometry of the sidechain chosen to fill the cavity can impact the folding of the S2P trimer. Membrane fusion is triggered by ACE2-S1 binding and cleavage of S2 by cellular TMPRSS2, leading to viral entry and replication [4]. The stabilizing effects of substitution of either Ala1016 or Ala1020 individually with Val or Leu. or in combination with Val and Ile, respectively, did not lead to changes in membrane fusion function. However, 2 sta- bilizing mutations, A1016L and A1016V/A1020I blocked the ability of S to mediate entry when pseudotyped into HIV-1 particles. These effects could be due to structural constraints being imposed on the S ectodomain due to interactions between the S cytoplasmic tail and HIV-1 matrix protein that underlies the viral envelope. Indeed, the structure and function of the HIV-1 gp41 fusion glycoprotein ectodomain can be modulated by interactions between matrix and the cytoplasmic tail of gp41 [67,68]. Overall, the data indicate that creation of an artificial hydrophobic core in the center of the CH1 coiled coil of S2 does not alter the intrinsic membrane fusion function of S. Even though the S2P.16L-FHA and S2P.VI-FHA putative trimers exhibited higher thermo- stability than unmutated S2P-FHA, the 3 antigens exhibited very similar abilities to elicit gly- coprotein-binding and neutralizing antibodies against the ancestral variant (that they were derived from) as well as the Delta and Omicron BA.1 VOCs. The 3 antigens elicited very high RBD-binding titers (1/95,000-1/137,000), pseudovirus neutralizing ID50s (3,950–5,110) and authentic virus neutralizing ID50s (2700–3150) against ancestral and Delta-derived viruses and proteins, whereas 3-fold reductions in binding titer and neutralization ID50 were observed with RBD and pseudovirus derived from Omicron BA.1 and a more pronounced ~log10 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 19 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein reduction in neutralization ID50 against Omicron BA.1 authentic virus was observed (see S1 Table). Strong and significant correlations were observed between RBD-binding titer and authentic virus neutralizing ID50 suggesting that a large component of neutralizing antibody was directed to the RBD and this activity was likely compromised by the 15 mutations arising in the Omicron BA.1 RBD. The results of the S fragment binding ELISA and serum-NAb cross-competition assay indicated that the 3 antigens elicited very similar antibody specificities that could bind to the RBD, NTD and S2 stem as well as compete with ACE2-Fc for binding to the RBM and mNAbs directed to key neutralization epitopes within the RBD, NTD fusion peptide and S2 stem region of the ancestral S2P-FHA trimer. S2P.OmiBA1-FHA exhibited greater thermostability than its ancestral counterpart (Tm of 58˚C versus 43.6˚C, respectively) and introduction of VI increased the Tm further to 65˚C. The stabilizing effects of the Ala cavity-filling VI mutation are therefore transferrable to this highly divergent variant. An examination of the binding of ACE2-Fc and mNAbs to key neu- tralization epitopes including the NTD, RBM, RBD, fusion peptide and stem indicated highly stable and high affinity interactions with VI and non-VI forms of ancestral and omicron BA.1 S2P-FHA indicating that the key broad neutralization epitopes are present on these proteins. The presence of the epitopes in the ancestral S2P-FHA and S2P.VI-FHA immunogens may in part account for the broad range of antibody specificities observed in the immune sera. Nota- bly, the VI mutation in ancestral S2P-FHA stabilized the interaction with C1520 to the NTD, S2H97 to the RBD, COV44-79 to the fusion peptide and CV3-25 to the stem-directed mNAbs whereas such an effect was not observed with the Omicron BA.1 version. Stabilization of the ancestral S2P-FHA by VI may cause structural changes that subtly affect access to neutraliza- tion epitopes in these regions or enable the optimization of epitope-paratope interactions such that highly stable interactions form. In theory, this may be beneficial for higher affinity interac- tions with B cell receptors if this construct were to be used as a vaccine immunogen. Despite these subtle changes in antigenicity, the S2P.VI-FHA trimer was as immunogenic as S2P-FHA in guinea pigs, whose B cell repertoire may or may not include the human mNAbs used in this study. Omicron BA.1 S trimers have been shown to have relatively compact domain organiza- tions with increased intersubunit buried surface and relatively high stability [61,63,64]. A more rigid Omicron BA.1 S architecture may inhibit structural changes caused by the stabiliz- ing VI mutation in the core of the trimer such that full exposure of neutralization epitopes is maintained. The VI mutation enabled the production of intrinsically stable Omicron BA.1 and Omicron BA.4/5 S ectodomain oligomers with elution properties consistent with that of trimeric S2P- foldon [5] in the absence of an external trimerization motif (S2P.OmiBA1.VI-1208.H8 and S2P.OmiBA.4/5.VI-1208.H8, respectively). The ‘foldonless’ ancestral S2P ectodomain (S2P- 1208.H8) had the propensity to form lower molecular weight forms, requiring VI for the puta- tive trimer to form. This result suggests that the sidechains of Val and Ile at positions 1016 and 1020, respectively, prefer a trimeric structure for their accommodation in the coiled coil. Despite the ancestral S2P.VI-1208.H8 ectodomain forming an oligomer consistent with tri- meric structure, a large proportion dissociated to lower molecular weight forms following a freeze-thaw cycle indicating that the ancestral ‘foldonless’ S2P.VI-1208 oligomer is unstable. By contrast, the Omicron BA.1 and Omicron BA.4/5 ‘foldonless’ ectodomains were able to form putative trimers in both the presence and absence of VI. However, whereas the VI mutants exhibited melting curves pointing to a single species with melting temperatures of >60˚C that were stable against a freeze-thaw cycle, the non-VI versions comprised 2 species with lower melting temperatures that partially dissociated to lower-order species following freeze-thawing. The BLI data presented in Table 1 and S5 Fig indicated stable and high-affinity mNAb binding to key conserved neutralization epitopes within the NTD, RBM, RBD, fusion PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 20 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein peptide and stem of S2P.OmiBA1.VI-1208 and S2P.OmiBA.45.VI-1208 as observed with S2P. VI-FHA. The foldon trimerization clamp is thus not required for the formation and presenta- tion of key conserved neutralization epitopes in these VI-stabilized S2P glycoproteins suggest- ing that they too may elicit broad neutralizing responses as did S2P.VI-FHA, but perhaps skewed towards omicron variants rather than ancestral or delta variants. The VI mutation can therefore be used to enhance the biophysical properties of candidate purified trimeric Spike glycoprotein vaccines such that foreign, highly immunogenic trimerization domains, such as T4 foldon or the HIV-1 gp41 6-helix bundle [69,70], are no longer required. Such off-target antibody responses have halted the progression of an otherwise immunogenic Spike candidate through human clinical trials [69]. The unequal global distribution of approved COVID-19 vaccines, the nondurable nature of vaccine-elicited protective NAb responses and the emergence of the SARS CoV-2 omicron lin- eages has led to continuing waves of SARS CoV-2 transmission. mRNA- and chimpanzee ade- novirus-based COVID-19 vaccines exhibit substantial initial effectiveness against omicron BA.1, however, this wanes over time [30]. NAb titers can be increased by second and third mRNA booster doses but these increases are transient [71,72], raising the prospect of periodic boosting with circulating strain-matched vaccines to maintain protective immunity in the human population [38,39]. Our use of hydrophobic residues to fill the Ala cavity in the core of the S trimer enabled the production of intrinsically stable omicron lineage S trimers thereby providing an avenue for developing a simple trimeric S glycoprotein vaccine that could have utility as a heterologous booster. Materials and methods Ethics statement All animal experiments were performed in accordance with the eighth edition of the Austra- lian Code for the Care and Use of Animals for Scientific Purposes and were approved by the South Australian Health and Medical Research Institute (SAHMRI) Animal Ethics Commit- tee, project number SAM-20-030. Recombinant proteins and synthetic peptides S2P-FHA. A synthetic gene encoding the SARS CoV-2 (ancestral Hu-1 isolate; Genbank accession number YP_009724390.1) S ectodomain, corresponding to the S2P protein described by Wrapp et al. [5], was obtained from GeneART-ThermoFisher Scientific. The gene encodes S residues 16–1208, the furin cleavage site mutation, R682RAR-> G682SAS, and a di-Pro substitution at positions 986 and 987. The C-terminus of S2P was appended with foldon (YIPEAPRDGQAYVRKDGEWVLLSTFL), octa-His and avitag (GLNDIFEAQKIEWHE) sequences, each separated by GSGS linkers to give S2P-FHA. The synthetic S2P-FHA gene was ligated downstream of a DNA sequence encoding the tissue plasminogen activator leader via NheI, within pcDNA3 (Invitrogen). Mutations were introduced into S2P-FHA expression vec- tors using synthetic genes encoding mutated S2P subfragments produced by GeneART-Ther- moFisher Scientific. An S2P-FHA expression vector encoding the Omicron BA.1 S ectodomain was also produced by the same method (S2P.OmiBA1-FHA). S1. A synthetic gene encoding the S1 domain (amino acids 16–681) of the ancestral Hu-1 isolate was obtained from GeneART-ThermoFisher Scientific and ligated to the tissue plasminogen activator leader via NheI in pcDNA3. The protein encodes a C-terminal hexa-His tag and Avitag sequence. RBD. Synthetic genes encoding the receptor binding domain (RBD; amino acids 332–532) of ances- tral Hu-1, Delta, Omicron BA.1 and Omicron BA.4/5 isolates were obtained from GeneART-- ThermoFisher Scientific and ligated to the tissue plasminogen activator leader via NheI in PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 21 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein pcDNA3. Both proteins encode a C-terminal hexa-His tag and Avitag sequence. NTD. A syn- thetic gene encoding the N-terminal domain (NTD; amino acids 16–305) of the ancestral Hu- 1 isolates was obtained from GeneART-ThermoFisher Scientific and ligated to the tissue plas- minogen activator leader via NheI in pcDNA3. The protein encodes a C-terminal hexa-His tag and Avitag sequence. S2P-1208.H8 expression vectors containing ancestral Hu-1, Omicron BA.1 and Omicron BA.4/5 S ectodomain sequences were derived from S2P-FHA plasmids by exchanging the foldon-His8-Avitag sequence with GlySerGlySer-His8. S2P-1208.H8 represents the S ectodomain (residues 16–1208) appended with a C-terminal His8 tag. In one variant, the 2P mutation (Pro986Pro987) was reverted to the WT sequence (Lys986Val987) in SnoP. OmiBA4/5.VI-1208.H8 by exchanging the S gene fragment containing 2P with a synthetic gene fragment containing the reversion. hACE2-Fc is a recombinant fusion protein comprising amino acids 19–615 of the human ACE2 ectodomain linked to the Fc domain of human IgG1 via a GS linker. A synthetic gene encoding hACE2-Fc was obtained from GeneART-Thermo- Fisher Scientific and ligated downstream of the tissue plasminogen activator leader via NheI in pcDNA3. The DNA sequences of S and hACE2 clones were verified by fluorescent Sanger sequencing (BigDye, ABI). Synthetic peptides. Synthetic peptides corresponding to the fusion peptide (amino acids 808–832, DPSKPSKRSFIEDLLFNKVTLADAG), and stem (amino acids 1142–1165, QPELDSFKEELDKYFKNHTSPDVD), were synthesized by Genscript. The pep- tides contain an N-terminal biotin moiety and C-terminal amide. Expression and purification of recombinant proteins S2P-FHA and S2P-1208.H8 expression vectors were transfected into 293Freestyle or Expi293F cells using 293fectin or Expifectamine, respectively, as recommended by the manufacturer (ThermoFisher Scientific). To produce biotinylated S2P-FHA and recombinant RBD proteins, the appropriate expression vectors were transfected into Expi293F-BirA cells [73] using Expi- fectamine. The cells were cultured for 4–5 days at 34˚C after which the transfection superna- tants were clarified by centrifugation and filtration through 0.45 μm nitrocellulose filters. The SARS CoV-2 glycoproteins were then purified by divalent cation affinity chromatography using TALON resin (Merck) or HiTrap IMAC FF followed by size exclusion chromatography using a Superose 6 Increase 10/300 column linked to an AKTApure instrument (Cytiva). hACE2-Fc was produced in Expi293F cells and purified from the clarified culture supernatant using Protein G-Agarose (Genscript) followed by SEC on a Superdex 200 16/600 column linked to an AKTApure instrument (Cytiva). All proteins were concentrated using Amicon centrifugal filter units. The protein solutions were filter-sterilized using 0.45 μm nitrocellulose filters and protein aliquots stored at -80˚C. Protein purity was assessed by SDS-PAGE and SEC. Recombinant mNAbs pCDNA3-based human IgG1 heavy and kappa and lambda light chain expression vectors [74] containing the variable regions of SARS CoV-2 directed mNAbs COVOX222 [13], S2E12 and S2H97 [18], Omi-18 and Omi-42 [35], SP1-77 [58], C1520 [57], COV44-79 [23], CV3-25 [24] COVA1-25 [12], and HCV1 [55] were produced in-house using synthetic gene fragments encoding the mNAb heavy and light chain variable regions produced by GeneART-Thermo- Fisher Scientific. The mNAbs were produced by transfection of Expi293 cells with equal amounts of matched heavy and light chain vectors using Expifectamine according to the man- ufacturer’s instructions (ThermoFisher Scientific). After 5 days of culture at 37˚C, the transfec- tion supernatants were clarified by centrifugation and filtration through 0.45 μm nitrocellulose filters. The IgG was purified by affinity chromatography using Protein G-agarose (Genscript) PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 22 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein and exchanged into PBS. The antibodies were concentrated using Amicon centrifugal filter units. IgG solutions were filter-sterilized using 0.45 μm nitrocellulose filters and aliquots stored at -80˚C. Differential scanning fluorimetry Differential scanning fluorimetry was used to assess protein thermostability [75]. 10 μg of pro- tein was diluted into 25μL with 5x concentration SYPRO Orange Protein Gel Stain (Sigma Aldrich) in duplicate. The samples were then heated in an QuantStudio 7 qPCR System in 0.5˚C increments from 25˚C to 95˚C for 1 minute per increment. 3 measurements of fluores- cence were taken at the end of each increment. Excitation was at 492nm, and emission at 610nm. The Tm was determined to be the minimum of the negative first derivative of the melt- ing curve. Cell-cell fusion assay Luciferase reporter assay. Mutations were introduced to the WH-Human1_EPI_402119 expression plasmid bearing codon-optimized full-length S by overlap extension PCR using Phusion DNA polymerase (ThermoFisher). The sequences of mutants were confirmed by fluo- rescent Sanger sequencing (BigDye Terminator v3.1, ABI). 293T effector cells were plated at 250,000 cells per well of 12-well tissue culture plates (Nunc) in Dulbecco’s modified minimal essential medium containing 10% v/v fetal bovine serum and transfected with WH-Huma- n1_EPI_402119 plasmid (1 μg), a bacteriophage T7 RNA polymerase expression plasmid, (1 μg, pCAG-T7) [49], a T7 promoter-driven Gaussia luciferase reporter plasmid (0.5 μg, pTM.Gaussia) and a furin expression plasmid (0.25 μg, pcDNA3.1-Furin) [50]. pTM.Gaussia comprises the Gaussia princeps luciferase open reading frame (Geneart-Thermo Fisher) ligated into the NcoI-SalI sites of pTM.1 [76]. 293T-ACE2 target cells [51] in the same medium were transfected with a luciferase reporter plasmid (1 μg, pTMluc) [52] and a TMPRSS2 expression plasmid (0.25 μg) [53]. The transfection reagent was FuGENE HD (Promega). At 36.5 h post transfection, the culture supernatants of effector cells were assayed for Gaussia luciferase activ- ity. Effector and target cells were then resuspended in fresh medium and cocultured for 3 h in round-bottomed 96-well tissue culture plates after which luciferase activity was measured using Steady-Glo luciferase reagent (Promega) in a Clariostar plate reader (BMG Labtech). Fluorescent assay. 293T effector cells were plated at 250,000 cells per well of 12-well tissue culture plates (Nunc) in Dulbecco’s modified minimal essential medium containing 10% v/v fetal bovine serum and transfected with WH-Human1_EPI_402119 plasmid (1 μg) plus pcDNA3.1-Furin (0.25 μg). 293T-ACE2 target cells [51] in the same medium were transfected with an EGFP expression vector (1 μg, pEGFP-N1) and a TMPRSS2 expression plasmid (0.25 μg) [53]. The transfection reagent was FuGENE HD (Promega). At 36.5 h post transfec- tion, the effector cells were stained with Hoechst 33342 nuclear stain (Invitrogen) for 1 h, washed with PBS, detached with versene solution, and resuspended in culture medium. 293T-ACE2 cells were detached and resuspended by the same process. Effector and target cells were mixed and added to poly-L-lysine coated black walled 96-well culture dishes (Corning). The cells were cocultured for 24 h and whole well images were acquired on the ImageXpress Pico Automated Cell Imaging System (Molecular Devices) using a 10x objective. Images were analysed using ImageJ Version 2.9.0/1.53t. Western blot Transfected 293T cells were lysed for 10 min on ice in phosphate-buffered saline containing 1% Triton X-100, 0.02% sodium azide, 1 mM EDTA. Lysates were clarified by centrifugation PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 23 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein at 10,000xg at 4˚C prior to polyacrylamide gel electrophoresis in the presence of SDS in 10% polyacrylamide gels under reducing conditions. Proteins were transferred to nitrocellulose prior to Western blotting with anti-S1 polyclonal rabbit antibody (Sino biological) and Goat anti-rabbit IR-Dye800CW (Odyssey). The blots were scanned in a LI-COR Mol2800 (Millen- ium Science) and visualized using Image Studio v1.0 (LI-COR). Pseudotyped virus production and analysis S-pseudotyped HIV luciferase reporter viruses were prepared according to the method of Jack- son et al. [54]. Plasmids used for the production of S-HIV pseudoparticles were a kind gift of Professor Doria-Rose, NIH Vaccine Research Center, and include the WH-Huma- n1_EPI_402119 expression plasmid bearing codon-optimized full-length S (Genbank # MN908947.3), the packaging plasmid pCMVΔR8.2 and luciferase reporter plasmid pHR’ CMV Luc [77], and a TMPRSS2 plasmid [53]. The 4 plasmids were co-transfected into HEK293T cells and after 18 h of incubation, the medium was replaced with fresh Dulbecco’s modification of minimal essential medium containing 10% fetal bovine serum (DMF10) and cultured for a further 2 days. Supernatants containing pseudoviruses were filtered through 0.45μm membrane filters prior to use. WH-Human1_EPI_402119 vectors encoding synthetic Delta and Omicron BA.1 S genes (GeneArt) were also produced. To visualize the p24 content of pseudoviruses, filtered culture supernatants were centrifuged at 10,000 xg for 2 h. To visual- ize the S content of pseudoviruses, filtered culture supernatants were ultracentrifuged over a sucrose cushion (1.5 ml, 25% w/v sucrose in PBS) using a Beckman SW41 Ti rotor (25,000 rpm, 2.5 h, 4˚C). The pelleted pseudovirions were subjected to reducing SDS-PAGE and west- ern blotting with monoclonal antibody 183-H12-5C (NIH HIV Reagent Program, Division of AIDS, NIAID, NIH, contributed by Dr. Bruce Chesebro and Dr. Hardy Chen [78]) to detect p24 and anti-S1 polyclonal rabbit antibody (Sino biological) to detect S. The blots were devel- oped with goat anti-mouse immunoglobulins-AlexaFluor688 (Invitrogen) or goat anti-rabbit immunoglobulins-IR-Dye800CW (Odyssey). The blots were scanned in a LI-COR Mol2800 (Millenium Science) and visualized using Image Studio v1.0 (LI-COR). The infectivity of fil- tered pseudoparticle-containing supernatants was determined 3 days after inoculation of 293-ACE2 cells plated on poly-L-lysine coated 96-well culture plates (10,000 cells per well). Luciferase activity was measured using Promega’s luciferase assay system in a Clariostar plate reader (BMG Labtech). Biolayer interferometry BLI-based measurements were determined using an OctetRED96 System (ForteBio, Fremont CA). Antibodies were diluted in kinetic buffer to 10 μg/ml and immobilized onto anti-human IgG Fc capture biosensors (AHC, ForteBio). Kinetics assays were carried out at 30˚C using standard kinetics acquisition rate settings (5.0 Hz, averaging by 20) at a sample plate shake speed of 1,000 rpm. The kinetic experiments included five steps: (a) baseline (180 s); (b) anti- body loading (300 s); (c) second baseline (180 s); (d) association of antigen (300 s), and (e) dis- sociation of antigen (300 s). Fitting curves were constructed using ForteBio Data Analysis 10.0 software using a 1:1 binding model, and double reference subtraction was used for correction. Immunizations Groups of 8 guinea pigs (outbred tricolor) that were matched for gender, weight, and age were immunized subcutaneously with 30 μg of S2P proteins in PBS in a 1:1 (v/v) mix with AddaVax adjuvant (InvivoGen, San Diego, CA) at weeks 0, 4 and 14. A negative control group was immunized as above with a 1:1 (v/v) mix of PBS and adjuvant. Blood was collected at 2 weeks PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 24 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein after the 2nd dose via the saphenous vein, and at 2 weeks after the 3rd dose by terminal cardiac puncture and allowed to clot for serum preparation. Sera were stored at -80˚C, with heat inac- tivation at 56˚C for 30 min prior to use in immunological assays. Animals were housed and all procedures were performed at the Preclinical, Imaging, and Research Laboratories, South Aus- tralian Health and Medical Research Institute (Gilles Plains, Australia). ELISA Streptavidin capture format. Nunc Maxisorp 96 well plates were coated with streptavidin (5 μg/ml, 50 mM carbonate buffer) overnight at 4˚C, washed with PBS and blocked with BSA (10 mg/ml, PBS) at room temperature for 1 h. The plates were again washed and then incu- bated with biotinylated S glycoproteins or synthetic peptides (5 μg/ml, PBS) overnight at 4˚C. After further washing, the plates were incubated with serially diluted serum samples or mNAbs for 2 h at room temperature and antibody binding detected using horseradish peroxi- dase–labelled rabbit anti-guinea pig antibody (Dako, Glostrup, Denmark) or horseradish per- oxidase–labelled anti-human IgA, IgG, IgM (Dako, Glostrup, Denmark) and 3,3’5,5’ tetramethylbenzidine. Direct binding format. Nunc Maxisorp 96 well plates were coated with S2P-FHA, S1, RBD or NTD protein solutions (2 μg/ml, PBS) at 4˚C overnight. The plates were washed with PBS and blocked with BSA (10 μg/ml, PBS) at room temperature for 1 h. The plates were again washed and then incubated with serially diluted serum samples for 2 h at room temperature. Antibody binding was detected using horseradish peroxidase–labelled rabbit anti-guinea pig antibody (Dako, Glostrup, Denmark) or horseradish peroxidase–labelled anti-human IgA, IgG, IgM (Dako, Glostrup, Denmark) and 3,3’5,5’ tetramethylbenzidine. Color reactions were measured with a Clariostar plate reader (BMG LabTech). Optical density was plotted against the reciprocal dilution of plasma in Prism v9.3.0 and curves fitted using the Sigmoidal, 4PL, X is concentration model. The binding titer was defined as the reciprocal dilution of serum giv- ing an optical density ten-times that of background, as defined by binding to BSA. S-HIV pseudovirus neutralizing assay Neutralization assays were conducted according to the method of Jackson et al. [54]. Heat inactivated sera (56˚C for 30 minutes) were serially diluted in DMF10 and each dilution mixed with an equal volume of S-pseudotyped HIV luciferase reporter viruses and incubated for 1h at 37˚C in triplicate. Virus-serum mixtures were added to 293T-ACE2 cell [51] monolayers attached to poly-L-lysine (0.01% w/v) coated 96 well plates the day prior at 10,000 cells/well and incubated for 2h at 37˚C before addition of an equal volume of DMF10. After 3 days, tis- sue culture fluid was removed, monolayers were washed once with PBS and lysed with cell cul- ture lysis reagent (Promega) and luciferase measured using luciferase substrate (Promega) in a CLARIOstar plate reader (BMG LabTech). The mean percentage entry was calculated as (RLU plasma+virus)/(RLU medium+virus)100. The percentage entry was plotted against the recip- rocal dilution of plasma in Prism v9.3.0 and curves fitted using the Sigmoidal, 4PL, X is con- centration model. The reciprocal dilution of plasma required to prevent 50% virus entry was calculated from the non-linear regression line (ID50). The lowest amount of neutralizing anti- body detectable is a titer of 200. All samples that did not reach 50% neutralization were assigned an arbitrary value of 100. Authentic virus neutralization assay The neutralizing activity of sera against authentic ancestral hCoV-19/Australia/NSW2715/ 2020 (Spike sequence identical to Hu-1), Delta, and Omicron BA.1 SARS-CoV-2 was PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 25 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein determined with the rapid high-content SARS-CoV-2 microneutralization assay described by Aggarwal et al. [33]. Briefly, Hoechst-33342-stained HAT-24 cells were seeded in 384-well plates (Corning, CLS3985). Serially diluted heat-inactivated vaccinal sera were coincubated with an equal volume of SARS-CoV-2 virus solution at twice the median lethal dose for 1 h at 37˚C. 40 μl of serum-virus mixtures were added to an equal volume of pre-plated cells, incu- bated for 20 h and then directly imaged on an InCell Analyzer HS2500 high-content fluores- cence microscopy system (Cytiva). Cellular nuclei counts were obtained with IN Carta automated image analysis software (Cytiva), and the percentage of virus neutralization was cal- culated as described in [33]. The neutralization ID50 was the last consecutive dilution reaching �50% neutralization. Serum-mNAb cross-competition ELISA Nunc Maxisorp 96-well plates were coated with streptavidin (Sigma) (5 μg/ml in 50 mM car- bonate buffer) at 4˚C, overnight after which they were blocked with BSA (10 mg/ml in PBS) at room temperature for 1h. After 2 washes, the plates were incubated with biotinylated S2P-FHA trimers (2 μg/ml in 5 mg/ml BSA/PBS containing 0.05% Tween 20) at room temper- ature for 1 h. Serially diluted vaccinal sera were mixed with sub-saturating amounts of hACE2-Fc and anti-S mNAbs and incubated with the streptavidin-biotinylated S2P-FHA coated plates for a further 2 h at room temperature. mNAb binding was detected using horse- radish peroxidase–labelled goat anti-human IgG F(ab’)2 (Thermofisher-Scientific), or horse- radish peroxidase–labelled anti-human IgA, IgG, IgM (Dako, Glostrup, Denmark) for hACE2-Fc. The substrate was 3,3’5,5’ tetramethylbenzidine. Color reactions were measured with a Multiskan Ascent plate reader (Thermo Electron, Waltham, MA). Antibody binding to different antigens was compared by fitting curves using the Sigmoidal, 4PL, X is concentration model using Prism version 9 software, and ID50s obtained by interpolation. Negative stain EM Purified S2P.OmiBA1.VI-FHA, S2P.OmiBA1.VI-1208 or S2P.OmiBA4/5.VI-1208 were diluted to 10–17.5 μg/mL and 5 μl applied to glow-discharged continuous carbon grids and incubated for 2 min before staining with 2% uranyl acetate. Grids were imaged using a Tecnai Spirit G2 TEM (FEI) operating at 120 kV and equipped with an Eagle 4K camera (FEI). Images were collected at a nominal magnification of 67,000 (corresponding to pixel size of 1.64 Å/ pixel), defocus of 0.4–1.3 μm and a total exposure of 30 e-/Å2 following low-dose procedures. Image processing was conducted using RELION 3.1 [79]. Estimation of the contrast transfer function (CTF) parameters was performed using GCTF [80]. Particles were initially picked using manual picking to generate 2D classes subsequently used for template picking of the entire dataset. Extracted particles were subject to reference-free 2D classification to yield the final 2D classes. Statistical methods Data were statistically compared using the non-parametric Kruskal-Wallis test with Dunn’s multiple comparisons in Prism 9.3.0. Correlations between RBD titer and neutralization ID50 were tested using the nonparametric Spearman test. A P value of < 0.05 was considered significant. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 26 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein Supporting information S1 Fig. Characterization of purified S2P-FHA. A, Superose 6 SEC of purified S2P-FHA. Standards: thyroglobulin, 669 kDa, aldolase, 158 kDa. B, SDS-PAGE under reducing condi- tions and Coomassie blue staining of purified S2P-FHA. C, Binding of ACE2-Fc and human mNAbs to avidin-captured biotinylated S2P-FHA in ELISA. D, Differential scanning fluorim- etry of purified S2P-FHA using SYPRO Orange. The rate of change of fluorescence over time [–d(RFU)/dt] as a function of temperature is shown. (JPG) S2 Fig. A. Gaussia luciferase activity produced by effector cells used in the luciferase reporter assay of cell-cell fusion shown in Fig 2B. RLU from individual transfections shown. B. Approach used to obtain magnified fields shown in Fig 2D from the whole field image (left panel). Corresponding bright field, Hoechst 33342 and EGFP images are stacked at right. C, Bright field, Hoechst 33342, EGFP, merged images and enlargements of the merged images (left to right) corresponding to those shown in Fig 2D. (PPTX) S3 Fig. A, Infectivity of S-HIV-1 luciferase reporter pseudoviruses. S-HIV-1 pseudoviruses were produced by transfected 293T cells for 72 h after which transfection supernatants were filtered, diluted 1/10 and used to inoculate 293-ACE2 cells expressing TMPRSS2. The cells were lysed and assayed for firefly luciferase activity 72 h days later. The means ± SEM shown (n = 8). ns, not significant, *, P < 0.001, **, P < 0.0001 versus WT, Kruskal-Wallis test. B, p24/CA content of pelleted pseudoviruses from transfections in A. The p24/CA band was revealed by SDS-PAGE and western blotting with anti-CA antibody 183- and AlexaFluor688-- labeled goat-anti-mouse immunoglobulin. C, S content of S-HIV pseudoviruses pelleted by ultracentrifugation through a sucrose cushion. The S band was revealed by SDS-PAGE and western blotting with polyclonal anti-S1 antibody and goat anti-rabbit immunoglobulin IRDye800CW. m, molecular weight markers. (JPG) S4 Fig. Specificity of elicited antibodies. A, Reactivity of immune sera with S fragments in ELISA. Immune sera were titrated on S2P-FHA and subdomains thereof derived from the Hu- 1 ancestral strain. Binding titers were defined as the reciprocal dilution of serum giving an optical density ten-times that of background, as defined by binding to BSA. B. Reactivity of mNAbs and S-derived glycoproteins and subdomains used in panels A and C in ELISA. C, Epitope specificity of immune sera assessed by competition ELISA. Competition ID50s of vac- cinal sera versus ACE2-Fc or mNAbs for binding to streptavidin-captured ancestral Hu-1 bio- tin-S2P-FHA. Serially diluted vaccinal sera were mixed with constant amounts of ACE2-Fc and human monoclonal anti-S IgGs prior to incubation with streptavidin captured biotin- S2P-FHA. The immunogen groups are indicated below the graphs. A Kruskal-Wallis test was used to determine that the differences in ID50s observed between groups was not significant (ns). The horizontal dotted lines indicates that the ID50s of control sera were >1/20. D, Bind- ing of immune sera to biotin-S2P-FHA captured on streptavidin-coated plates in ELISA. (JPG) S5 Fig. BLI measurements of ACE2-Fc and mNAbs binding to Ala cavity mutants. Binding of S2P-derived analytes to S ligands immobilized on anti-human IgG capture biosensors. Asso- ciation was for 300 sec followed by dissociation for 300 sec. The binding kinetics are shown in Table 1. (JPG) PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 27 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein S6 Fig. Reversion of the 2P mutation to Lys986-Val987 in SnoP.OmiBA4/5.VI-1208 reduces trimer yield but maintains thermostability. A, Superose 6 SEC of S2P.OmiBA4/5.VI-1208 and SnoP.OmiBA4/5.VI-1208 trimers. B, SYPRO orange thermofluor assay of S2P.OmiBA4/5.VI- 1208 and SnoP.OmiBA4/5.VI-1208 trimers. (JPG) S7 Fig. Negative stain electron microscopy analysis of S2P.VI trimers. Representative raw EM micrographs of negatively-stained S2P.OmiBA1-FHA (left), S2P.OmiBA1.VI-1208 (mid- dle) and S2P.OmiBA4/5.VI-1208 (right) and corresponding representative 2D class averages. The class averages, derived from ~10,000 particles per sample, revealed the S2P.VI protein samples to be predominantly in forms consistent with the pre-fusion trimer (classes shown ranked by abundance). A small percentage of the rod-shaped putative post-fusion form (highlighted within red boxes) was observed in the S2P.OmiBA1-FHA (8.0%) and S2P. OmiBA1.VI-1208 (8.8%) samples. (JPG) S1 Table. Summary of neutralization ID50s, RBD and S2P-FHA binding titers of sera elic- ited by S2P-FHA antigens. (PDF) Acknowledgments 293-ACE2 cells were obtained from Professor Jesse Bloom, Fred Hutch Cancer Center, Seattle, WA. The WH-Human1_EPI_402119, pCMVΔR8.2, pHR’ CMV Luc, and TMPRSS2 expres- sion plasmids were obtained from Professor Nicole Doria-Rose, Vaccine Research Center, NIH, MD, USA. Expi293F-BirA cells were obtained from Dr. Bruce Wines, Burnet Institute, VIC, Australia. Author Contributions Conceptualization: Pantelis Poumbourios, Heidi E. Drummer. Data curation: Pantelis Poumbourios, Christine Langer, Irene Boo, Tasnim Zakir, Anouschka Akerman, Anupriya Aggarwal, Stuart G. Turville. Formal analysis: Pantelis Poumbourios, Christine Langer, Irene Boo, Tasnim Zakir, Anouschka Akerman, Vanessa Milogiannakis, Anupriya Aggarwal, Bronte A. Johnstone, Jungmin Ha, Fasse´li Coulibaly, Stuart G. Turville, Heidi E. Drummer. Funding acquisition: Pantelis Poumbourios, Stuart G. Turville, Heidi E. Drummer. Investigation: Pantelis Poumbourios, Christine Langer, Irene Boo, Tasnim Zakir, Rob J. Center, Anouschka Akerman, Vanessa Milogiannakis, Anupriya Aggarwal, Bronte A. Johnstone, Jungmin Ha, Fasse´li Coulibaly, Stuart G. Turville. Methodology: Pantelis Poumbourios, Christine Langer, Irene Boo, Tasnim Zakir, Rob J. Center, Anouschka Akerman, Vanessa Milogiannakis, Anupriya Aggarwal, Bronte A. Johnstone, Jungmin Ha, Fasse´li Coulibaly, Stuart G. Turville, Heidi E. Drummer. Project administration: Pantelis Poumbourios. Resources: Rob J. Center, Fasse´li Coulibaly, Stuart G. Turville. Supervision: Pantelis Poumbourios, Fasse´li Coulibaly, Stuart G. Turville, Heidi E. Drummer. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 28 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein Validation: Christine Langer, Irene Boo, Tasnim Zakir, Rob J. Center, Anouschka Akerman, Vanessa Milogiannakis, Anupriya Aggarwal, Jungmin Ha, Stuart G. Turville, Heidi E. Drummer. Visualization: Bronte A. Johnstone, Jungmin Ha. Writing – original draft: Pantelis Poumbourios. Writing – review & editing: Pantelis Poumbourios, Christine Langer, Irene Boo, Tasnim Zakir, Rob J. Center, Anouschka Akerman, Vanessa Milogiannakis, Anupriya Aggarwal, Bronte A. Johnstone, Jungmin Ha, Fasse´li Coulibaly, Stuart G. Turville, Heidi E. Drummer. References 1. Duan L, Zheng Q, Zhang H, Niu Y, Lou Y, Wang H. The SARS-CoV-2 Spike Glycoprotein Biosynthesis, Structure, Function, and Antigenicity: Implications for the Design of Spike-Based Vaccine Immunogens. Front Immunol. 2020; 11:576622. https://doi.org/10.3389/fimmu.2020.576622 PMID: 33117378 2. Finkelstein MT, Mermelstein AG, Parker Miller E, Seth PC, Stancofski ED, Fera D. Structural Analysis of Neutralizing Epitopes of the SARS-CoV-2 Spike to Guide Therapy and Vaccine Design Strategies. Viruses. 2021; 13(1):134. https://doi.org/10.3390/v13010134 PMID: 33477902 3. Hoffmann M, Kleine-Weber H, Pohlmann S. A Multibasic Cleavage Site in the Spike Protein of SARS- CoV-2 Is Essential for Infection of Human Lung Cells. Mol Cell. 2020; 78(4):779–84 e5. https://doi.org/ 10.1016/j.molcel.2020.04.022 PMID: 32362314 4. Walls AC, Park YJ, Tortorici MA, Wall A, McGuire AT, Veesler D. Structure, Function, and Antigenicity of the SARS-CoV-2 Spike Glycoprotein. Cell. 2020; 183(6):1735. https://doi.org/10.1016/j.cell.2020.11. 032 PMID: 33306958 5. Wrapp D, Wang N, Corbett KS, Goldsmith JA, Hsieh CL, Abiona O, et al. Cryo-EM structure of the 2019-nCoV spike in the prefusion conformation. Science. 2020; 367(6483):1260–3. https://doi.org/10. 1126/science.abb2507 PMID: 32075877 6. Fu Q, Chou JJ. A Trimeric Hydrophobic Zipper Mediates the Intramembrane Assembly of SARS-CoV-2 Spike. J Am Chem Soc. 2021; 143(23):8543–6. https://doi.org/10.1021/jacs.1c02394 PMID: 34086443 7. Ke Z, Oton J, Qu K, Cortese M, Zila V, McKeane L, et al. Structures and distributions of SARS-CoV-2 spike proteins on intact virions. Nature. 2020; 588(7838):498–502. https://doi.org/10.1038/s41586-020- 2665-2 PMID: 32805734 8. Jackson CB, Farzan M, Chen B, Choe H. Mechanisms of SARS-CoV-2 entry into cells. Nat Rev Mol Cell Biol. 2022; 23(1):3–20. https://doi.org/10.1038/s41580-021-00418-x PMID: 34611326 9. Cai Y, Zhang J, Xiao T, Peng H, Sterling SM, Walsh RM, Jr., et al. Distinct conformational states of SARS-CoV-2 spike protein. Science. 2020; 369(6511):1586–92. https://doi.org/10.1126/science. abd4251 PMID: 32694201 10. Piccoli L, Park YJ, Tortorici MA, Czudnochowski N, Walls AC, Beltramello M, et al. Mapping Neutralizing and Immunodominant Sites on the SARS-CoV-2 Spike Receptor-Binding Domain by Structure-Guided High-Resolution Serology. Cell. 2020; 183(4):1024–42 e21. https://doi.org/10.1016/j.cell.2020.09.037 PMID: 32991844 11. Barnes CO, West AP, Jr., Huey-Tubman KE, Hoffmann MAG, Sharaf NG, Hoffman PR, et al. Structures of Human Antibodies Bound to SARS-CoV-2 Spike Reveal Common Epitopes and Recurrent Features of Antibodies. Cell. 2020; 182(4):828–42 e16. https://doi.org/10.1016/j.cell.2020.06.025 PMID: 32645326 12. Brouwer PJM, Caniels TG, van der Straten K, Snitselaar JL, Aldon Y, Bangaru S, et al. Potent neutraliz- ing antibodies from COVID-19 patients define multiple targets of vulnerability. Science. 2020; 369 (6504):643–50. https://doi.org/10.1126/science.abc5902 PMID: 32540902 13. Dejnirattisai W, Zhou D, Supasa P, Liu C, Mentzer AJ, Ginn HM, et al. Antibody evasion by the P.1 strain of SARS-CoV-2. Cell. 2021; 184(11):2939–54 e9. https://doi.org/10.1016/j.cell.2021.03.055 PMID: 33852911 14. Gruell H, Vanshylla K, Weber T, Barnes CO, Kreer C, Klein F. Antibody-mediated neutralization of SARS-CoV-2. Immunity. 2022; 55(6):925–44. https://doi.org/10.1016/j.immuni.2022.05.005 PMID: 35623355 15. Liu L, Wang P, Nair MS, Yu J, Rapp M, Wang Q, et al. Potent neutralizing antibodies against multiple epitopes on SARS-CoV-2 spike. Nature. 2020; 584(7821):450–6. https://doi.org/10.1038/s41586-020- 2571-7 PMID: 32698192 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 29 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein 16. Qi H, Liu B, Wang X, Zhang L. The humoral response and antibodies against SARS-CoV-2 infection. Nat Immunol. 2022; 23(7):1008–20. https://doi.org/10.1038/s41590-022-01248-5 PMID: 35761083 17. Shi R, Shan C, Duan X, Chen Z, Liu P, Song J, et al. A human neutralizing antibody targets the recep- tor-binding site of SARS-CoV-2. Nature. 2020; 584(7819):120–4. https://doi.org/10.1038/s41586-020- 2381-y PMID: 32454512 18. Starr TN, Czudnochowski N, Liu Z, Zatta F, Park YJ, Addetia A, et al. SARS-CoV-2 RBD antibodies that maximize breadth and resistance to escape. Nature. 2021; 597(7874):97–102. https://doi.org/10.1038/ s41586-021-03807-6 PMID: 34261126 19. Andreano E, Piccini G, Licastro D, Casalino L, Johnson NV, Paciello I, et al. SARS-CoV-2 escape from a highly neutralizing COVID-19 convalescent plasma. Proc Natl Acad Sci U S A. 2021; 118(36) e2103154118. https://doi.org/10.1073/pnas.2103154118 PMID: 34417349 20. Cerutti G, Guo Y, Zhou T, Gorman J, Lee M, Rapp M, et al. Potent SARS-CoV-2 neutralizing antibodies directed against spike N-terminal domain target a single supersite. Cell Host Microbe. 2021; 29(5):819– 33 e7. https://doi.org/10.1016/j.chom.2021.03.005 PMID: 33789084 21. McCallum M, De Marco A, Lempp FA, Tortorici MA, Pinto D, Walls AC, et al. N-terminal domain anti- genic mapping reveals a site of vulnerability for SARS-CoV-2. Cell. 2021; 184(9):2332–47 e16. https:// doi.org/10.1016/j.cell.2021.03.028 PMID: 33761326 22. McCarthy KR, Rennick LJ, Nambulli S, Robinson-McCarthy LR, Bain WG, Haidar G, et al. Recurrent deletions in the SARS-CoV-2 spike glycoprotein drive antibody escape. Science. 2021; 371 (6534):1139–42. https://doi.org/10.1126/science.abf6950 PMID: 33536258 23. Dacon C, Tucker C, Peng L, Lee CD, Lin TH, Yuan M, et al. Broadly neutralizing antibodies target the coronavirus fusion peptide. Science. 2022; 377(6607):728–35. https://doi.org/10.1126/science. abq3773 PMID: 35857439 24. 25. Jennewein MF, MacCamy AJ, Akins NR, Feng J, Homad LJ, Hurlburt NK, et al. Isolation and characteri- zation of cross-neutralizing coronavirus antibodies from COVID-19+ subjects. Cell Rep. 2021; 36 (2):109353. https://doi.org/10.1016/j.celrep.2021.109353 PMID: 34237283 Li W, Chen Y, Prevost J, Ullah I, Lu M, Gong SY, et al. Structural basis and mode of action for two broadly neutralizing antibodies against SARS-CoV-2 emerging variants of concern. Cell Rep. 2022; 38 (2):110210. https://doi.org/10.1016/j.celrep.2021.110210 PMID: 34971573 26. Chen KK, Tsung-Ning Huang D, Huang LM. SARS-CoV-2 variants—Evolution, spike protein, and vac- cines. Biomed J. 2022; 45(4):573–9. https://doi.org/10.1016/j.bj.2022.04.006 PMID: 35526825 27. Ghimire D, Han Y, Lu M. Structural Plasticity and Immune Evasion of SARS-CoV-2 Spike Variants. Viruses. 2022; 14(6). https://doi.org/10.3390/v14061255 PMID: 35746726 28. 29. Jacobs JL, Haidar G, Mellors JW. COVID-19: Challenges of Viral Variants. Annu Rev Med. 2022; 74:31–53. https://doi.org/10.1146/annurev-med-042921-020956 PMID: 35850493 Li J, Jia H, Tian M, Wu N, Yang X, Qi J, et al. SARS-CoV-2 and Emerging Variants: Unmasking Struc- ture, Function, Infection, and Immune Escape Mechanisms. Front Cell Infect Microbiol. 2022; 12:869832. https://doi.org/10.3389/fcimb.2022.869832 PMID: 35646741 30. Andrews N, Stowe J, Kirsebom F, Toffa S, Rickeard T, Gallagher E, et al. Covid-19 Vaccine Effective- ness against the Omicron (B.1.1.529) Variant. N Engl J Med. 2022; 386(16):1532–46. https://doi.org/ 10.1056/NEJMoa2119451 PMID: 35249272 31. Lopez Bernal J, Andrews N, Gower C, Gallagher E, Simmons R, Thelwall S, et al. Effectiveness of Covid-19 Vaccines against the B.1.617.2 (Delta) Variant. N Engl J Med. 2021; 385(7):585–94. https:// doi.org/10.1056/NEJMoa2108891 PMID: 34289274 32. Supasa P, Zhou D, Dejnirattisai W, Liu C, Mentzer AJ, Ginn HM, et al. Reduced neutralization of SARS- CoV-2 B.1.1.7 variant by convalescent and vaccine sera. Cell. 2021; 184(8):2201–11 e7. https://doi. org/10.1016/j.cell.2021.02.033 PMID: 33743891 33. Aggarwal A, Stella AO, Walker G, Akerman A, Esneau C, Milogiannakis V, et al. Platform for isolation and characterization of SARS-CoV-2 variants enables rapid characterization of Omicron in Australia. Nat Microbiol. 2022; 7(6):896–908. https://doi.org/10.1038/s41564-022-01135-7 PMID: 35637329 34. Dejnirattisai W, Huo J, Zhou D, Zahradnik J, Supasa P, Liu C, et al. SARS-CoV-2 Omicron-B.1.1.529 leads to widespread escape from neutralizing antibody responses. Cell. 2022; 185(3):467–84 e15. https://doi.org/10.1016/j.cell.2021.12.046 PMID: 35081335 35. Nutalai R, Zhou D, Tuekprakhon A, Ginn HM, Supasa P, Liu C, et al. Potent cross-reactive antibodies following Omicron breakthrough in vaccinees. Cell. 2022; 185(12):2116–31 e18. https://doi.org/10. 1016/j.cell.2022.05.014 PMID: 35662412 36. Hachmann NP, Miller J, Collier AY, Ventura JD, Yu J, Rowe M, et al. Neutralization Escape by SARS- CoV-2 Omicron Subvariants BA.2.12.1, BA.4, and BA.5. N Engl J Med. 2022; 387(1):86–8. https://doi. org/10.1056/NEJMc2206576 PMID: 35731894 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 30 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein 37. Branche AR, Rouphael NG, Diemert DJ, Falsey AR, Losada C, Baden LR, et al. SARS-CoV-2 Variant Vaccine Boosters Trial: Preliminary Analyses. medRxiv. 2022. https://doi.org/10.1101/2022.07.12. 22277336 PMID: 35898343 38. Chalkias S, Harper C, Vrbicky K, Walsh SR, Essink B, Brosz A, et al. A Bivalent Omicron-Containing Booster Vaccine against Covid-19. N Engl J Med. 2022; 387(14):1279–91. https://doi.org/10.1056/ NEJMoa2208343 PMID: 36112399 39. Tenforde MW, Weber ZA, Natarajan K, Klein NP, Kharbanda AB, Stenehjem E, et al. Early Estimates of Bivalent mRNA Vaccine Effectiveness in Preventing COVID-19-Associated Emergency Department or Urgent Care Encounters and Hospitalizations Among Immunocompetent Adults—VISION Network, Nine States, September-November 2022. MMWR Morb Mortal Wkly Rep. 2023; 71(53):1637–46. https://doi.org/10.15585/mmwr.mm7153a1 PMID: 36921274 40. Bullough PA, Hughson FM, Skehel JJ, Wiley DC. Structure of influenza haemagglutinin at the pH of membrane fusion. Nature. 1994; 371(6492):37–43. https://doi.org/10.1038/371037a0 PMID: 8072525 41. Chan DC, Fass D, Berger JM, Kim PS. Core structure of gp41 from the HIV envelope glycoprotein. Cell. 1997; 89(2):263–73. https://doi.org/10.1016/s0092-8674(00)80205-6 PMID: 9108481 42. Julien JP, Cupo A, Sok D, Stanfield RL, Lyumkis D, Deller MC, et al. Crystal structure of a soluble cleaved HIV-1 envelope trimer. Science. 2013; 342(6165):1477–83. https://doi.org/10.1126/science. 1245625 PMID: 24179159 43. McLellan JS, Yang Y, Graham BS, Kwong PD. Structure of respiratory syncytial virus fusion glycopro- tein in the postfusion conformation reveals preservation of neutralizing epitopes. J Virol. 2011; 85 (15):7788–96. https://doi.org/10.1128/JVI.00555-11 PMID: 21613394 44. McLellan JS, Chen M, Joyce MG, Sastry M, Stewart-Jones GB, Yang Y, et al. Structure-based design of a fusion glycoprotein vaccine for respiratory syncytial virus. Science. 2013; 342(6158):592–8. https:// doi.org/10.1126/science.1243283 PMID: 24179220 45. Walls AC, Tortorici MA, Snijder J, Xiong X, Bosch BJ, Rey FA, et al. Tectonic conformational changes of a coronavirus spike glycoprotein promote membrane fusion. Proc Natl Acad Sci U S A. 2017; 114 (42):11157–62. https://doi.org/10.1073/pnas.1708727114 PMID: 29073020 46. Weissenhorn W, Dessen A, Harrison SC, Skehel JJ, Wiley DC. Atomic structure of the ectodomain from HIV-1 gp41. Nature. 1997; 387(6631):426–30. https://doi.org/10.1038/387426a0 PMID: 9163431 47. Wilson IA, Skehel JJ, Wiley DC. Structure of the haemagglutinin membrane glycoprotein of influenza virus at 3 A resolution. Nature. 1981; 289(5796):366–73. https://doi.org/10.1038/289366a0 PMID: 7464906 48. Baase WA, Liu L, Tronrud DE, Matthews BW. Lessons from the lysozyme of phage T4. Protein Sci. 2010; 19(4):631–41. https://doi.org/10.1002/pro.344 PMID: 20095051 49. Takikawa S, Ishii K, Aizaki H, Suzuki T, Asakura H, Matsuura Y, et al. Cell fusion activity of hepatitis C virus envelope proteins. J Virol. 2000; 74(11):5066–74. https://doi.org/10.1128/jvi.74.11.5066-5074. 2000 PMID: 10799580 50. Binley JM, Sanders RW, Clas B, Schuelke N, Master A, Guo Y, et al. A recombinant human immunode- ficiency virus type 1 envelope glycoprotein complex stabilized by an intermolecular disulfide bond between the gp120 and gp41 subunits is an antigenic mimic of the trimeric virion-associated structure. J Virol. 2000; 74(2):627–43. https://doi.org/10.1128/jvi.74.2.627-643.2000 PMID: 10623724 51. Crawford KHD, Eguia R, Dingens AS, Loes AN, Malone KD, Wolf CR, et al. Protocol and Reagents for Pseudotyping Lentiviral Particles with SARS-CoV-2 Spike Protein for Neutralization Assays. Viruses. 2020; 12(5):513. https://doi.org/10.3390/v12050513 PMID: 32384820 52. Maerz AL, Center RJ, Kemp BE, Kobe B, Poumbourios P. Functional implications of the human T-lym- photropic virus type 1 transmembrane glycoprotein helical hairpin structure. J Virol. 2000; 74(14):6614– 21. https://doi.org/10.1128/jvi.74.14.6614-6621.2000 PMID: 10864675 53. Bottcher E, Matrosovich T, Beyerle M, Klenk HD, Garten W, Matrosovich M. Proteolytic activation of influenza viruses by serine proteases TMPRSS2 and HAT from human airway epithelium. J Virol. 2006; 80(19):9896–8. https://doi.org/10.1128/JVI.01118-06 PMID: 16973594 54. Jackson LA, Anderson EJ, Rouphael NG, Roberts PC, Makhene M, Coler RN, et al. An mRNA Vaccine against SARS-CoV-2—Preliminary Report. N Engl J Med. 2020; 383(20):1920–31. https://doi.org/10. 1056/NEJMoa2022483 PMID: 32663912 55. Morin TJ, Broering TJ, Leav BA, Blair BM, Rowley KJ, Boucher EN, et al. Human monoclonal antibody HCV1 effectively prevents and treats HCV infection in chimpanzees. PLoS Pathog. 2012; 8(8): e1002895. https://doi.org/10.1371/journal.ppat.1002895 PMID: 22952447 56. Tada T, Zhou H, Dcosta BM, Samanovic MI, Chivukula V, Herati RS, et al. Increased resistance of SARS-CoV-2 Omicron variant to neutralization by vaccine-elicited and therapeutic antibodies. EBioMe- dicine. 2022; 78:103944. https://doi.org/10.1016/j.ebiom.2022.103944 PMID: 35465948 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 31 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein 57. Wang Z, Muecksch F, Cho A, Gaebler C, Hoffmann HH, Ramos V, et al. Analysis of memory B cells identifies conserved neutralizing epitopes on the N-terminal domain of variant SARS-Cov-2 spike pro- teins. Immunity. 2022; 55(6):998–1012 e8. https://doi.org/10.1016/j.immuni.2022.04.003 PMID: 35447092 58. Luo S, Zhang J, Kreutzberger AJB, Eaton A, Edwards RJ, Jing C, et al. An antibody from single human V(H)-rearranging mouse neutralizes all SARS-CoV-2 variants through BA.5 by inhibiting membrane fusion. Sci Immunol. 2022; 7(76):eadd5446. https://doi.org/10.1126/sciimmunol.add5446 PMID: 35951767 59. Aggarwal A, Akerman A, Milogiannakis V, Silva MR, Walker G, Stella AO, et al. SARS-CoV-2 Omicron BA.5: Evolving tropism and evasion of potent humoral responses and resistance to clinical immunother- apeutics relative to viral variants of concern. EBioMedicine. 2022; 84:104270. https://doi.org/10.1016/j. ebiom.2022.104270 PMID: 36130476 60. Tuekprakhon A, Nutalai R, Dijokaite-Guraliuc A, Zhou D, Ginn HM, Selvaraj M, et al. Antibody escape of SARS-CoV-2 Omicron BA.4 and BA.5 from vaccine and BA.1 serum. Cell. 2022; 185(14):2422–33 e13. https://doi.org/10.1016/j.cell.2022.06.005 PMID: 35772405 61. Cao Y, Yisimayi A, Jian F, Song W, Xiao T, Wang L, et al. BA.2.12.1, BA.4 and BA.5 escape antibodies elicited by Omicron infection. Nature. 2022; 608(7923):593–602. https://doi.org/10.1038/s41586-022- 04980-y PMID: 35714668 62. Pallesen J, Wang N, Corbett KS, Wrapp D, Kirchdoerfer RN, Turner HL, et al. Immunogenicity and structures of a rationally designed prefusion MERS-CoV spike antigen. Proc Natl Acad Sci U S A. 2017; 114(35):E7348–E57. https://doi.org/10.1073/pnas.1707304114 PMID: 28807998 63. Cerutti G, Guo Y, Liu L, Liu L, Zhang Z, Luo Y, et al. Cryo-EM structure of the SARS-CoV-2 Omicron spike. Cell Rep. 2022; 38(9):110428. https://doi.org/10.1016/j.celrep.2022.110428 PMID: 35172173 64. Cui Z, Liu P, Wang N, Wang L, Fan K, Zhu Q, et al. Structural and functional characterizations of infec- tivity and immune evasion of SARS-CoV-2 Omicron. Cell. 2022; 185(5):860–71 e13. https://doi.org/10. 1016/j.cell.2022.01.019 PMID: 35120603 65. Stalls V, Lindenberger J, Gobeil SM, Henderson R, Parks R, Barr M, et al. Cryo-EM structures of SARS-CoV-2 Omicron BA.2 spike. Cell Rep. 2022; 39(13):111009. https://doi.org/10.1016/j.celrep. 2022.111009 PMID: 35732171 66. Benton DJ, Gamblin SJ, Rosenthal PB, Skehel JJ. Structural transitions in influenza haemagglutinin at membrane fusion pH. Nature. 2020; 583(7814):150–3. 67. Edwards TG, Wyss S, Reeves JD, Zolla-Pazner S, Hoxie JA, Doms RW, et al. Truncation of the cyto- plasmic domain induces exposure of conserved regions in the ectodomain of human immunodeficiency virus type 1 envelope protein. J Virol. 2002; 76(6):2683–91. https://doi.org/10.1128/jvi.76.6.2683-2691. 2002 PMID: 11861835 68. Narasimhulu VGS, Bellamy-McIntyre AK, Laumaea AE, Lay CS, Harrison DN, King HAD, et al. Distinct functions for the membrane-proximal ectodomain region (MPER) of HIV-1 gp41 in cell-free and cell-cell viral transmission and cell-cell fusion. J Biol Chem. 2018; 293(16):6099–120. https://doi.org/10.1074/ jbc.RA117.000537 PMID: 29496992 69. Chappell KJ, Mordant FL, Li Z, Wijesundara DK, Ellenberg P, Lackenby JA, et al. Safety and immuno- genicity of an MF59-adjuvanted spike glycoprotein-clamp vaccine for SARS-CoV-2: a randomised, dou- ble-blind, placebo-controlled, phase 1 trial. Lancet Infect Dis. 2021; 21(10):1383–94. https://doi.org/10. 1016/S1473-3099(21)00200-0 PMID: 33887208 70. Sliepen K, van Montfort T, Melchers M, Isik G, Sanders RW. Immunosilencing a highly immunogenic protein trimerization domain. J Biol Chem. 2015; 290(12):7436–42. https://doi.org/10.1074/jbc.M114. 620534 PMID: 25635058 71. Bar-On YM, Goldberg Y, Mandel M, Bodenheimer O, Amir O, Freedman L, et al. Protection by a Fourth Dose of BNT162b2 against Omicron in Israel. N Engl J Med. 2022; 386(18):1712–20. https://doi.org/10. 1056/NEJMoa2201570 PMID: 35381126 72. Pajon R, Doria-Rose NA, Shen X, Schmidt SD, O’Dell S, McDanal C, et al. SARS-CoV-2 Omicron Vari- ant Neutralization after mRNA-1273 Booster Vaccination. N Engl J Med. 2022; 386(11):1088–91. https://doi.org/10.1056/NEJMc2119912 PMID: 35081298 73. Wines BD, Vanderven HA, Esparon SE, Kristensen AB, Kent SJ, Hogarth PM. Dimeric FcgammaR Ectodomains as Probes of the Fc Receptor Function of Anti-Influenza Virus IgG. J Immunol. 2016; 197 (4):1507–16. 74. Center RJ, Boo I, Phu L, McGregor J, Poumbourios P, Drummer HE. Enhancing the antigenicity and immunogenicity of monomeric forms of hepatitis C virus E2 for use as a preventive vaccine. J Biol Chem. 2020; 295(21):7179–92. https://doi.org/10.1074/jbc.RA120.013015 PMID: 32299914 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 32 / 33 PLOS PATHOGENSStabilization of the SARS CoV-2 spike glycoprotein 75. Niesen FH, Berglund H, Vedadi M. The use of differential scanning fluorimetry to detect ligand interac- tions that promote protein stability. Nat Protoc. 2007; 2(9):2212–21. https://doi.org/10.1038/nprot.2007. 321 PMID: 17853878 76. Moss B, Elroy-Stein O, Mizukami T, Alexander WA, Fuerst TR. Product review. New mammalian expression vectors. Nature. 1990; 348(6296):91–2. https://doi.org/10.1038/348091a0 PMID: 2234068 77. Naldini L, Blomer U, Gage FH, Trono D, Verma IM. Efficient transfer, integration, and sustained long- term expression of the transgene in adult rat brains injected with a lentiviral vector. Proc Natl Acad Sci U S A. 1996; 93(21):11382–8. https://doi.org/10.1073/pnas.93.21.11382 PMID: 8876144 78. Chesebro B, Wehrly K, Nishio J, Perryman S. Macrophage-tropic human immunodeficiency virus iso- lates from different patients exhibit unusual V3 envelope sequence homogeneity in comparison with T- cell-tropic isolates: definition of critical amino acids involved in cell tropism. J Virol. 1992; 66(11):6547– 54. https://doi.org/10.1128/JVI.66.11.6547-6554.1992 PMID: 1404602 79. 80. Zivanov J, Nakane T, Forsberg BO, Kimanius D, Hagen WJ, Lindahl E, et al. New tools for automated high-resolution cryo-EM structure determination in RELION-3. Elife. 2018; 7. https://doi.org/10.7554/ eLife.42166 PMID: 30412051 Zhang K. Gctf: Real-time CTF determination and correction. J Struct Biol. 2016; 193(1):1–12. https:// doi.org/10.1016/j.jsb.2015.11.003 PMID: 26592709 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1010981 May 18, 2023 33 / 33 PLOS PATHOGENS
10.1371_journal.ppat.1009980	RESEARCH ARTICLE A novel class of Candida glabrata cell wall proteins with β-helix fold mediates adhesion in clinical isolates Viktoria ReithoferID GrootID 2, Lars-Oliver EssenID 1 1☯, Jordan Ferna´ ndez-PereiraID 2☯, Marı´a AlvaradoID 2, Piet de a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Reithofer V, Ferna´ndez-Pereira J, Alvarado M, de Groot P, Essen L-O (2021) A novel class of Candida glabrata cell wall proteins with β- helix fold mediates adhesion in clinical isolates. PLoS Pathog 17(12): e1009980. https://doi.org/ 10.1371/journal.ppat.1009980 Editor: George Deepe, jr., University of Cincinnati College of Medicine, UNITED STATES Received: September 25, 2021 Accepted: November 30, 2021 Published: December 28, 2021 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.ppat.1009980 Copyright: © 2021 Reithofer et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All crystal structures are available from the RCSB protein structure database (accession numbers 7O9O, 7O9P, 7O9Q). 1 Department of Biochemistry, Philipps-Universita¨t, Marburg, Germany, 2 Regional Center for Biomedical Research, Castilla-La Mancha Science & Technology Park, University of Castilla–La Mancha, Albacete, Spain ☯ These authors contributed equally to this work. * Piet.deGroot@uclm.es (PdG); essen@chemie.uni-marburg.de (LOE) Abstract Candida glabrata is an opportunistic pathogenic yeast frequently causing infections in humans. Though it lacks typical virulence factors such as hyphal development, C. glabrata contains a remarkably large and diverse set of putative wall adhesins that is crucial for its success as pathogen. Here, we present an analysis of putative adhesins from the homology clusters V and VI. First, sequence similarity network analysis revealed relationships between cluster V and VI adhesins and S. cerevisiae haze protective factors (Hpf). Crystal structures of A-regions from cluster VI adhesins Awp1 and Awp3b reveal a parallel right- handed β-helix domain that is linked to a C-terminal β-sandwich. Structure solution of the A- region of Awp3b via single wavelength anomalous diffraction phasing revealed the largest known lanthanide cluster with 21 Gd3+ ions. Awp1-A and Awp3b-A show structural similarity to pectate lyases but binding to neither carbohydrates nor Ca2+ was observed. Phenotypic analysis of awp1Δ, awp3Δ, and awp1,3Δ double mutants did also not confirm their role as adhesins. In contrast, deletion mutants of the cluster V adhesin Awp2 in the hyperadhesive clinical isolate PEU382 demonstrated its importance for adhesion to polystyrene or glass, biofilm formation, cell aggregation and other cell surface-related phenotypes. Together with cluster III and VII adhesins our study shows that C. glabrata CBS138 can rely on a set of 42 Awp1-related adhesins with β-helix/α-crystallin domain architecture for modifying the sur- face characteristics of its cell wall. Author summary Adhesion to host cells and abiotic, often hydrophobic surfaces, e.g. that of medical equip- ment like catheters, is an indispensable virulence factor for many pathogenic fungi. Among the latter, the yeast Candida glabrata excels by encoding in its genome large sets of surface-exposed cell wall proteins. Here, we show that in the clinical isolate PEU382 of C. glabrata, hyper-adhesiveness to plastics and the tendency to biofilm formation is PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 1 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins Funding: LOE and VR received funding from Deutsche Forschungsgemeinschaft [SFB 987]. PdG, JFP and MA received funding from the Spanish Ministry of Economy and Competitiveness (MINECO) [SAF2013-47570-P and SAF2017- 86188-P], and the Regional government of Castilla- La Mancha [SBPLY/19/180501/000114 and SBPLY/19/180501/000356], the Spanish grants co-financed by the EU (FEDER). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. conferred by a single adhesin, Awp2. An integrative bioinformatic and structural analysis of this and the related Awp1 and Awp3 adhesins unifies four, so far separately assigned Awp clusters—III, V, VI and VII–into one consisting of 42 Awp1-related adhesins. These adhesins commonly present an N-terminal module consisting of a right-handed β-helix and an α-crystallin domain on the yeast surface and use a calcium-independent mode for adhesion. Their sheer number contrasts to the 20 members of the well characterized Epa and 7 members of the Pwp family of surface proteins. Given these findings we suggest that C. glabrata utilizes just two structurally distinct motifs for colonizing different host niches by adhesion: the β-helix/α-crystallin module of Awp1-related adhesins and the C-type lec- tin PA14-domain for Epa and Pwp proteins. Introduction The yeast Candida glabrata is an opportunistic human pathogen that can cause mucosal, bloodstream, and medical device-related infections [1,2]. Despite a closer relationship to Sac- charomyces than to the pathogenic CTG-clade Candida spp. [2], C. glabrata is regarded the second most common causative agent of candidiasis worldwide after Candida albicans. C. glabrata has the ability to adhere to a wide variety of biotic and abiotic surfaces [1,3], mediated by a remarkably large number of putative adhesins [4,5]. As adhesion to host tissues or to med- ical devices is an important first step in the establishment of fungal infections, it is regarded as an important pathogenicity factor. C. glabrata wall adhesins are large, modular glycosylpho- sphatidylinositol (GPI)-modified proteins. The N-terminal region of mature GPI cell wall pro- teins (GPI-CWPs)–also referred to as A-region–is believed to define the ligand-binding function. This region is followed by a low complexity region, the B-region, which is usually rich in serine and threonine residues, thus presenting abundant acceptor sites for O-glycosyla- tion, and usually contains a variable number of large tandem repeats. By being linked with its processed C-terminus to cell wall β-1,6-glucans via the glycan remnant of the GPI anchor, the heavily glycosylated B-region can interact with other cell wall glycans and act as spacer mole- cule to present the A-region along the cell surface [1,6]. Screening the C. glabrata genome for genes encoding GPI-CWPs with the typical architec- ture of adhesins identified about 70 putative adhesins in various studies that–based on A- region sequence similarity–were categorized into seven phylogenetic clusters [5,7,8]. Well described adhesins are proteins of the Epa (epithelial adhesion) protein family (cluster I), which specify lectins that mediate adhesion to host cells by binding ligands containing termi- nal galactose residues [8–11]. Most proteins in the other clusters remain totally uncharacter- ized and their biological function or role in C. glabrata pathogenesis remains unknown. Unlike Epa adhesins, no orthologs of the uncharacterized Awp proteins have been described, a finding, which raises doubt about their relevance in this particular organism. However, a sub- set of the putative adhesins from different clusters was identified by mass spectrometric analy- sis of isolated cell walls in different C. glabrata strains and under various growth conditions. Among the identified GPI-CWPs were Epa1, 3, 6, and 7, but also a number of uncharacterized proteins from other clusters that were named Awp1–13 [4,5,12]. Their presence at the cell sur- face implies that these unknown putative adhesins are relevant for C. glabrata and support its colonization by promoting adhesion to surfaces. This study focuses on putative C. glabrata adhesins from clusters V and VI. Proteomic stud- ies identified the cluster VI protein Awp3 in walls from exponentially growing cells of refer- ence strain C. glabrata CBS138/ATCC 2001, whereas Awp1 (also in cluster VI) was identified PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 2 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins in stationary phase cells from low-adherent strain ATCC90876 [5]. Quantitative PCR expres- sion analysis of AWP1 and AWP3 in CBS138 showed higher expression of both genes in bio- film conditions compared to planktonic growth [4], however, the proteins were not identified in cell walls from CBS138 biofilm cultures, nor in hyperadhesive clinical isolates [4]. From cluster V, two GPI-CWPs that were commonly identified in isolated walls from CBS138 and hyperadhesive strains in biofilms as well as under planktonic conditions are Awp2 and Awp4. Interestingly, Awp8–11 from this same cluster have been identified only in biofilms of hyper- adhesive isolates grown on polystyrene, suggesting a role for cluster V in adhesion and biofilm formation on abiotic surfaces. Recently, an update of the C. glabrata CBS138 genome has been published [7]. This new genome assembly resolved several sequence errors, omissions and misassemblies in the 2004 version of the reference genome, particularly affecting (sub)telomeric regions where most of the putative adhesins are located. This also altered cluster V and VI adhesins. For instance, in the new assembly, a 5 kb insertion in the locus of AWP3 results in two tandem genes, named Awp3a and Awp3b. Also corrected were the misassembled sequences of the highly homolo- gous genes AWP4 and AWP11 (= AWP2D), which shifted their positions on chromosomes J and M. Sequence analysis revealed weak similarities leading to predictions of structural similarities between the A-regions of clusters V and VI putative adhesins. The A-region of cluster V pro- tein Awp2 also shows some similarity to the GPI-CWP Hyr1/Iff family of putative adhesins in C. albicans. Functional charactization studies documented that deletion of the hyphally-regu- lated gene HYR1 of this family led to attenuated virulence in the mouse oral biofilm model of infection [13] and, in a different study, IFF4 null mutants showed reduced adhesion to plastic substrate and attenuated virulence in a murine model of disseminated candidiasis [14]. In this study, we aim to shed light on the biological role of uncharacterized cluster V and VI wall adhesins in C. glabrata. We present an integrated study including sequence similarity network (SSN) analysis, descriptions of high-resolution three-dimensional structures of the A-regions of Awp1 and Awp3b, as well as phenotypic characterization of AWP1-3 deletion and overexpression mutants. We show that Awp1 and Awp3b A-regions have a two domain architecture, where the N-terminal domain is structurally related to pectate lyases and bacte- rial β-helix domains, while the C-terminal domain adopts an α-crystallin fold. We further demonstrate that the related Awp2 GPI-CWP regulates adhesiveness to polystyrene and also controls other cell surface-related phenotypes of the hyperadhesive strain PEU382. Our study opens the path to better understand the role of these adhesin-like wall proteins in pri- mary processes such as adhesion and biofilm formation that underlie the establishment of C. glabrata infections. Results Sequence similarity network of Awp1-related GPI-CWPs Sequence similarity network analysis is a valuable tool for identifying isofunctional subfamilies within a set of related sequences. Sequences within the network are represented as “nodes”, and their pairwise relationships as deduced from E-values by all-by-all BLAST analyses is depicted by lines connecting those, referred to as “edges”. This procedure leads to the forma- tion of clusters of nodes that represent protein subfamilies [15]. We generated a SSN using the A-regions of the previously identified C. glabrata GPI-CWPs Awp1-Awp4 from cluster VI as seeds for identifying related orthologs by extensive PSI-BLAST searches. Interestingly, the resulting SSN of the Awp1/Awp3 orthologs (Fig 1) is dominated by bacterial species (93.0%; 9569/10290) whereas only 6.9% belong to fungi and here exclusively to ascomycetes (711/ PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 3 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins Fig 1. SSN analysis of the class of fungal cell wall proteins of the Awp1/Hyr1/Hpf1-type. Upper left, the SSN of orthologs of Awp1-Awp4 is shown (E-value cut-off 10−20), which have been accumulated by ten repeated rounds of PSI-BLAST for each before merging the results and removing redundancy prior to SSN-analysis. The Interpro family of ‘N-terminal, hyphally-regulated cell wall proteins’ (IPR021031) is a subset of this SSN. Each of the 4507 nodes (left, E-value cutoff 10−20) covers sequences with pairwise sequence identities of at least 80% (total: 11737 sequences), 445 nodes (711 sequences) are of fungal (red), the remainder (grey) of bacterial origin. Right, the majority of fungal orthologs are found in Saccharomycetes (E-value cut-off 10−25). The structurally characterized A-regions of Awp1 and Awp3b fall in two separate C. glabrata subclusters besides the Iff4 subcluster. In C. albicans, at least twelve paralogs of the same domain type can be identified (orange labels), in S. cerevisiae strains Awa1, Hpf1 and Css1 form a common subcluster. The nodes are colored by fungal clade, with red being C. glabrata (Nakaseomyces); light orange designated the CTG clade, including Candida albicans. https://doi.org/10.1371/journal.ppat.1009980.g001 10290). The latter contain as a subset the INTERPRO protein family IPR021031, which covers conserved N-terminal domains of fungal proteins like Hpf1 from Saccharomyces cerevisiae and Hyr1 from Candida albicans. Hyr1 is apparently involved in hyphal formation and promotion of biofilm formation. In contrast, the former protein, Hpf1, is a heavily glycosylated manno- protein with still unknown biological function for S. cerevisiae. Its assignment as a haze- protective factor was solely derived from the oenologically useful trait during white wine pro- duction [16]. Another difference between them is the presence of at least 12 Hyr1 paralogs in C. albicans (Fig 1), named as Iff3-Iff6, Iff8-Iff9, Iff11, Iff14, Flo9 besides Hyr3 and Hyr4, whereas S. cerevisiae S288c harbors only two, Hpf1 and Css1. In Candida glabrata CBS138 we found 22 paralogs of Awp1 (S1 Fig), which exceeds even the number of Epa1-like adhesins with 20 paralogs [7]. Accordingly, we simplify the nomen- clature for Awp1-related GPI-CWPs in C. glabrata as follows: eight paralogs (Awp1, Awp1a- Awp1e, Awp3a, and Awp3b) belong to the previously annotated cluster VI. Thirteen form cluster V with ten of them (Awp2a-Awp2i, Awp4) being closely related to Awp2 (pairwise sequence identities 54% - 95%). Consequently, former Awp8-Awp11 were reassigned to Awp2a, Awp2b, Awp2c, Awp2d, respectively. The remaining two cluster V members (Gen- bank IDs: QHS67215.1, QHS65613.1) share more similarities with C. albicans proteins (Fig 1). One paralog (QHS68879.1) has not been assigned before to cluster V or VI and lacks any pre- dicted disulfide bridges for its A-region (see below). Awp2 but not Awp1 or Awp3 governs adhesion to polystyrene The importance of Awp1-related GPI-CWPs for fungal adhesion and other surface character- istics was studied by generating deletion mutants for AWP1, AWP2 and AWP3 using sequence information from genome assembly 2 in the Candida Genome Database. So far, Awp1 and PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 4 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins Fig 2. Awp2 mediates adhesion to plastic. (A) Adhesion to polystyrene after 24 h measured by Crystal violet (CV) staining. (B) Adhesion to polystyrene after 4 h measured by FC. (C) Adhesion to glass being measured after 4 h by flow cytometry (FC). Data in (A) are normalized to CBS138. AWP2 was deleted in three genetic backgrounds: reference strain CBS138, and hyperadhesive strains PEU382 and PEU427. AWP1 and AWP3 were deleted only in CBS138 as the corresponding proteins have not been identified in the hyperadhesive strains. (D) Adhesion to polystyrene after 24 h of gain-of-function mutants in S. cerevisiae, normalized to parental strain BY4741. Asterisks indicate significant differences of mutants compared to their parental strains (A) and (B) or the empty plasmid control (D). https://doi.org/10.1371/journal.ppat.1009980.g002 Awp3 are the only two cluster VI GPI-CWPs identified in cell wall preparations. A recent update of the C. glabrata genome [7] indicated that our awp3Δ mutants in fact lack the whole gene tandem AWP3a - AWP3b (S2 Fig). As neither the awp1Δ nor the awp3Δ mutants showed any phenotypes in preliminary assays of fitness, adhesion, aggregation, and drug sensitivities, we also created a double mutant where the tandem AWP3a/AWP3b was deleted in the awp1Δ background and used this mutant for the studies described below. Of cluster V, two proteins that are often encountered in cell wall preparations are Awp2 and Awp4, which are part of the same SSN cluster (Fig 1). As the AWP4 sequence in genome assembly 2 contains various errors leading to frameshifts, we focused on AWP2 and deleted this gene in three genetic back- grounds: in reference strain CBS138 as well as in the hyperadherent clinical strains PEU382 and PEU427. Mutants of Awp1-related GPI-CWPs were first analyzed for their capacity to adhere to polystyrene (Fig 2A). Upon 24 h of incubation, adhesion to polystyrene was clearly diminished by deleting AWP2 in PEU382, a strain with strong hyperadherence to plastic, resulting in a low level of adherence resembling that of reference strain CBS138. In CBS138 or PEU427, however, no lowering in adherence to polystyrene was observed upon deleting this gene. One could hypothesize that the effect of loss of Awp2 might be obscured by functional redundancy with other cluster V adhesins, and this may be particularly relevant in PEU427, where the related cluster V proteins Awp2a, Awp2b, and Awp2c (previous annotations: Awp8, 9 and 10) have been identified in cell walls of 24 h biofilms [12] (S2 Table). Therefore, we tested PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 5 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins adherence to polystyrene after shorter incubation for 4 h in PBS to avoid extensive biofilm for- mation (Fig 2B). In this case, in both hyperadherent strains PEU382 and PEU427, deletion of Awp2 clearly diminished adherence to a level similar to the reference strain. This Awp2-me- diated adhesion of C. glabrata to polystyrene is independent of calcium and other bivalent cat- ions, as addition of EDTA to the assays caused no loss of strain-specific adhesion (S3 Fig). A dominant role of Awp2 for causing the hydrophobic surface character of the PEU382 strain can also be inferred from the twofold diminished adhesion to hydrophilic glass surfaces by PEU382 when compared to CBS138 and PEU427 (Fig 2C). Deletion of AWP2 in PEU382 restores the adhesion to glass surfaces to even higher levels than exercised by CBS138 and PEU427, indicating that at least some Awp1-related A-regions have anti-adhesive functions as well. We then tested the mutants´ adhesiveness to human Hela and HaCaT cells but found no difference with their respective parental strains (Table 1). This implies no role of Awp2 in binding of these C. glabrata strains to human cells. For the cluster VI awp1,3Δ mutants, no phenotypes in adhesion were detected that could confirm their role as adhesins. Interestingly, the hyperadhesive PEU427 strain is the only strain that mediates adhesion to keratinocyte-like HaCaT cells at a similar level as found for the adhesion of all examined C. glabrata strains against cervical cancer-derived HeLa cells. Accordingly, the PEU427 strain may have a unique set of cell surface adhesins, that has been evolved towards the specific interaction with the epidermis. To further demonstrate the involvement of Awp2 in adhesion to polystyrene, hybrid gain- of-function constructs, containing the A-regions of Awp2 and Awp3b in front of the Epa1 Ser/ Thr-rich region [8], were expressed at the cell surface of a non-adhesive S. cerevisiae strain (Fig 2D). Consistent with the results obtained with the deletion mutants, expression of Awp2 but not of Awp3 paralogs in S. cerevisiae led to a clear, almost two-fold, increase of adhesiveness to polystyrene during a 24 h incubation. We then checked the AWP deletion mutants for a range of other phenotypes including fit- ness, sensitivity to cell wall-perturbing agents like calcofluor white (CFW) and SDS, antifun- gals like amphotericin B, flucanozole, isavucanozole, caspofungin, and micafungin, sedimentation, aggregation, surface hydrophobicity, and contents of cell wall components like chitin and protein (Figs 3 and S4). Compared to reference strain CBS138, the hyperadhesive strain PEU382 has been documented to have lower growth rates, increased resistance to the β- 1,3-glucan hydrolyzing enzyme zymolyase and to the cell wall perturbant CFW, fast sedimen- tation, and hyperaggregation, among other phenotypes [12]. All these PEU382 phenotypes were completely eliminated by deleting awp2Δ in this background (Fig 3A–3D and 3F). Inter- estingly, the hyperaggregation and sedimentation of parental PEU382 cells grown to stationary Table 1. Adhesion of C. glabrata awp mutants and parental strains to human HeLa and HaCaT cells. C. glabrata strain CBS138 PEU382 PEU427 Parental awp1,3Δ awp2Δ Parental awp2Δ Parental awp2Δ HeLa (%) 100 ± 4.2a 103.7 ± 17.4 100.1 ± 9.5 103.7 ± 5.4 122.8 ± 17.0 152.7 ± 11.2 158.8 ± 14.3 HaCaT (%) 18.2 ± 1.4 19.9 ± 2.4 18.9 ± 1.7 20.5 ± 2.5 21.9 ± 2.0 92.2 ± 11.6 94.7 ± 6.6 a All data in the table is normalized to CBS138 adhesion to Hela cells showing an absolute adhesion level of 31% https://doi.org/10.1371/journal.ppat.1009980.t001 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 6 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins Fig 3. Deletion of AWP2 overrides cell surface-related phenotypes of hyperadhesive strain PEU382. (A) Growth rate, (B) zymolyase sensitivity, (C) drug sensitivities using microbroth dilution methods, (D) Calcofluor white drop assay, (E) sedimentation and (F) aggregation of awp mutants. For reasons of clarity (C-F) are limited to CBS138, PEU382, and awp2Δ in PEU382. https://doi.org/10.1371/journal.ppat.1009980.g003 phase in YPD was abolished when the cells were washed with PBS (Fig 3E). This also occurred by adding PBS to stationary phase cultures in YPD prior to the sedimentation test (S5A Fig) but not when NaOH was added to reach the same pH of 6.5. Interestingly, the same effect could be achieved by adding fresh YPD media, but not by EDTA or YPD media conditioned before by C. glabrata growth therein (S5B and S5C Fig). These results suggest that there may other factors than phosphate that may inhibit hyperaggregation of PEU382. Nevertheless, col- orimetric assay measurements of protein and chitin content of cell walls isolated from plank- tonic (exponential phase) and 24 h biofilm cells (S4 Fig) did not reveal drastic changes between deletion mutants and parental strains. The increased protein (planktonic) and chitin (biofilms) contents observed in PEU382 compared to CBS138 were diminished by deletion of AWP2 in PEU382. In drug sensitivity assays (antifungals, SDS, S3 Table) no differences between strains PEU382 and CBS138 were observed and deletion of AWP2 in PEU382 did not result in any phenotype. Altogether, these data underline the importance of Awp2 in PEU382, not only for hyperadherence but also for other altered cell surface characteristics that make this clinical isolate so atypical. Contrasting with the many surface phenotypes of awp2Δ in PEU382, awp2Δ mutants in PEU427 and CBS138 as well as awp1,3Δ double mutants in CBS138 lacked obvious cell PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 7 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins surface-related phenotypes as only minor alterations were observed. In detail, the increased zymolyase resistance of PEU427 compared to CBS138 was not abolished by deletion of AWP2 as happened in PEU382 background but instead became slightly more pronounced. In CBS138 background, awp2Δ mutants showed slight alterations in surface hydrophobicity in stationary phase cells and cell wall protein content in both planktonic as well as biofilm cells. For the awp1,3Δ mutants, the only observed alterations were lower protein and higher chitin contents in biofilm cells. Structures of Awp1-related A-regions For structural studies, we overproduced the A-regions of Awp1 and Awp3b in E. coli and read- ily obtained crystals for both of them. In contrast, the recombinant Awp2 A-region was found to form only insoluble inclusion bodies. Crystals of the Awp3b A-region (Awp3b-A) belong to rhombohedral space group R32 with one molecule per asymmetric unit. As no protein struc- ture with sufficient sequence identity for molecular replacement (MR) was available in the Protein Databank (PDB), we determined phases by single wavelength anomalous diffraction (SAD). Awp3b A-region crystals soaked with Gd(OAc)3 diffracted to a resolution of 2 Å, there- from derived SAD phases were used to solve a native dataset of 1.55 Å resolution. Only few regions with uninterpretable electron density are contained in the Awp3b-A•Gd3+-complex, namely the loops consisting of residues S75–D82 and S320–E322. However, these regions are clearly defined in the Awp3b-A structure. Data collection and refinement statistics are summa- rized in Table 2. The conformations of some loops, especially T1 (see below), differ in the Awp3b A-region structures and hence indicate some flexibility for these regions. The Awp1-A crystals belong to space group P43212 with two molecules per asymmetric unit and diffracted to a resolution of 1.85 Å. The Awp1-A structure was solved by MR using the Awp3b-A structure as template, followed by 20 cycles of model building in ARP/wARP. Struc- tural similarity between the A-regions of Awp1 and Awp3b is evident from a calculated root mean square deviation (RMSD) of 1.36 Å for 223 Cα positions. This is consistent with the notion that both proteins belong to the same cluster of putative C. glabrata adhesins as they share a sequence identity of 25.1% for their A-regions (Awp1: S18-A324; Awp3b: D20-E344). The A-regions of Awp1 and Awp3b (Fig 4A) consist of 34 and 33 β-strands, respectively, which form a two-domain architecture with a parallel right-handed β-helix at its N-terminus (Awp1: S18-V240 with strands β0-β26, Awp3b: D20-I254 with β1-β26) that is followed by a β- sandwich domain with α-crystallin fold (Awp1: V241-Q314, Awp3b: V255-E334; both with strands β27- β33). The C-termini of the A-regions are covalently attached to the α-crystallin domain via a disulfide bridge to a cysteine residue in the loop linking β-strands 31 and 32 (Figs 4A and 5B; Awp1: C284-C322, Awp3b: C304-C341). Disulfide-bridging between A-domain cores and their C-termini has been observed before in fungal adhesins with PA14 [6,9,17] and fibronectin-type domains [18]. Due to the lack of an experimental structure, we subjected the sequence of the Awp2 A- region to template-independent structure prediction [19] by transformer-restrained Rosetta (trRosetta). The structural model of Awp2 (Fig 4B) is based on 195 homologous sequences and shows the same fold as the Awp1 and Awp3b A-regions with r.m.s.d. values of 2.60 and 2.68 Å for 223/239 Cα positions, respectively. These modest deviations are comparable to differences between the experimental structures of Awp1 and Awp3b and their corresponding trRosetta models (3.39 Å for 216 and 1.52 Å for 198 Cα positions, Fig 5A). Like other cluster V adhesins, Awp2 is predicted to form other disulfide bridges (Fig 5C) than the cluster VI adhesins by link- ing the C-terminal cysteine residues of the CxxC motif subsequent to its A-region to cysteines in β-helix turns 7 and 8 (Awp2: C327-C201, C330-C184; S1 and S6 Figs). As a consequence the PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 8 / 27 PLOS PATHOGENSTable 2. Data collection and refinement statistics for crystal structures of Awp1- and Awp3b-A regions. Deposition numbers for the RCSB protein data bank are given in parentheses. Data collection statistics is generated by SCALA [65]. Structural and functional analysis of β-helical Candida glabrata adhesins Awp1-A (7O9Q) ESRF, ID29 0.97717 P 43 21 2 a = b = 83.28, c = 274.24 45.81–1.85 (1.92–1.85) 165589 (16103) 83156 (8101) 2.97 (34.94) 12.64 (1.62) 99.16 (97.48) 2.0 (2.0) 0.998 (0.95) Awp3b-A (7O9O) ESRF, ID23-1 0.97625 R 3 2 a = b = 147.97, c = 117.77 53.51–1.55 (1.61–1.55) 134731 (13321) 69278 (6889) 3.627 (42.99) 10.68 (1.84) 96.96 (97.30) 1.9 (1.9) 0.999 (0.431) Awp3b-A•Gd3+ (7O9P) ESRF, ID29 1.71237 R 3 2 a = b = 144.4, c = 113.95 27.41–1.99 (2.06–1.99) 62641 (6200) 31321 (3100) 3.672 (12.86) 18.42 (4.90) 99.92 (100.00) 2.0 (2.0) 0.997 (0.924) 18.79/20.83 15.93/18.79 19.03/22.78 5262 51.43 0.004 0.67 0.99 96.89 2.62 0.49 1.31 3117 28.78 0.008 0.99 2.02 96.93 3.07 0.00 0.35 2658 37.13 0.014 2.02 3.59 96.68 2.99 0.33 3.09 Data collection X-ray source Wavelength Space group Unit cell parameters (Å) Resolution range (Å)a Total No. of reflectionsa No. of unique relfectionsa Rmerge (%)a I/σ(I)a Completeness (%)a Multiplicitya a CC1/2 Refinement Rwork/Rfree (%) No. of atoms Average B factor (Å2) R. m.s. deviations Bond length (Å2) Bond angles (˚) Clash scoreb Ramachandran plot (%)b Favouredb Allowedb Outliersb Rotamer outliers (%)b a Values for the outer resolution shell are given in parentheses. b Quality values as provided by MolProbity. https://doi.org/10.1371/journal.ppat.1009980.t002 C-termini of the rod-like A-regions of cluster V adhesins are placed on the central part of the A-regions rather than at their tip as found for Awp1 and Awp3b. Given that these A-regions are followed by long, heavily O-glycosylated repeats of the B-region, cluster-specific differences may cause different orientations of the A-regions along the surface of the C. glabrata cell wall. According to the nomenclature for β-helices by Yoder & Jurnak [20], the three β-strands forming a single turn in β-helices are referred to as PB1, PB2, and PB3 (Fig 4C); loops between them are labeled T1 (connecting PB1 and PB2), T2 (PB2 and PB3), and T3 (PB3 and PB1 of next turn). Additionally, Awp1-A has a truncated turn comprising only PB3-T3 at the N-ter- minus (β0: S18-P24). As right-handed β-helices, the T2 and T3 loops of Awp1-A and Awp3b- A are very short, whereas T1 loops are more extended. Interestingly, the T1-loops of β-helix turns 4–6 from Awp3b-A are cross-linked by two disulfide bridges, C115-C145 and C144-C178 (Fig 4A). These features, which were found to stabilize analogous loop conforma- tions in Epa-like adhesion domains of C. glabrata [9], are lacking in Awp1 and other Awp1-re- lated GPI-CWPs (S1 Fig). PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 9 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins Fig 4. Structures of the Awp1, Awp2 and Awp3b A-regions. (A) Left, crystal structure of the Awp1 A-region. The nine β-helix turns are colored according to S1 Fig. The C-terminal α-crystallin domain is depicted in orange. Right, crystal structure of the Awp3b A-region. (B) Model of the Awp2 A-region based on transformer-restrained modelling by trRosetta. (C) Architecture and nomenclature of β-helix turns. https://doi.org/10.1371/journal.ppat.1009980.g004 The two-domain architecture of Awp1-related A-regions A feature often found in β-helices is the repetitive stacking of distinct residue types along the outer and inner sides of the β-helix [21]. These stacks are regarded to provide additional stabil- ity and rigidity to the structures of β-helix proteins. On the inner sides, Awp1-A and Awp3b-A exhibit stacks of hydrophobic amino acids placed on all three sheets composing the β-helix, mainly Val, Leu, Ile, and Phe, and occasionally Tyr (Fig 6A and 6B). The twisted conformation of the β-helix results in a slight offset between following consecutive residues of the stack, thereby preventing unfavourable alignment of aromatic side chains due to overlap of Fig 5. Similarity and disulfide bridges in Awp1, Awp2 and Awp3b A-regions. (A) Superpositions of the Awp1 and Awp3b A-region crystal structures with the trRosetta model of Awp2A. Despite the limited sequence identity of 24% to the Awp3b A-region, the Awp3b-A structure closely resembles the Awp1 A-region with an r.m.s.d. of 1.4 Å for 223 Cα-atoms. (B) Disulfide bridges between the C-terminal cysteine residue and the α-crystallin domains of Awp1 and Awp3b A-regions. (C) Predicted disulfide bridges between the C-terminus and β-helix turns of the trRosetta model of the Awp2 A-region. https://doi.org/10.1371/journal.ppat.1009980.g005 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 10 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins Fig 6. Stacking patterns and O-glycosylation sites within the Awp1 and Awp3b A-regions. Top, view along the β- helix domains showing the repetitive bands of hydrophobic (gray) and serine, threonine and asparagine residues (colored) along the inner and outer β-helix areas of Awp1 (A) and Awp3b (B) A-regions. Bottom, O-glycosylation sites as predicted by NetOGlyc 4.0 for the A-regions of Awp1 (C) and Awp3b (D). The hydroxyl groups to be covalently modified by a fungal oligomannan are highlighted as spheres. Only eight of nine predicted O-glycosylation sites in Awp1 (Awp3b: 16 of 17) are depicted because S254 (Awp3b: T268) is buried within the α-crystallin domain (gray) and hence unaccessible for O-glycosylation. Notably, no or only a single O-glycosylation site was predicted for both A- regions in the β-helix domain. https://doi.org/10.1371/journal.ppat.1009980.g006 π-electron systems. β-helices of Awp1-A and Awp3b-A are characterized by highly conserved internal Asn ladders (Awp1: N89, N117, N145, N179; Awp3b: N99, N129, N158, N193; Awp2 model: N103, N129, N152, N179), whose asparagines adopt trans/gauche− rotamers (χ1 ~-179˚, χ2 ~44˚) and are found as last residues of PB3 strands starting from β-helix turn 4. These internal Asn ladders are capped by hydrogen-bonds to a tilted asparagine or glutamine residue (Awp1: N221, Awp2: Q220, Awp3b: Q234) that follows the PB3 strand of helix turn 8. Another feature of striking regularity can be observed along the PB1 sheet of the Awp1-A β- helix. Here, a Ser/Thr ladder stretches over 8 β-turns along the exterior face at the beginning of each β-strand (Fig 6A; S42, S65, T92, T120, T148, T182, S203, and T222). In a +2 position, a second outer stack consisting of four Thr and four Asn, interrupted by an isoleucine residue, stretches along the whole PB1 sheet. The four asparagine side chains (χ1 ~-63˚, χ2 ~-53˚) resemble the interior Asn stack by forming a hydrogen bonding network between the side chains’ carboxamide groups. Stacking on the exterior side of β-helix sheets has been described in several proteins, but seems to play a role in providing a hydrophobic environment for PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 11 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins packing of other structural features (e. g. an α-helix) towards the β-helix in many cases [21], which is not the case for Awp1-A. Serine and threonine residues are known to be potential target sites for O-linked glycosyla- tion, whereas asparagines within the sequence motif N-X-(S/T) may be subjected to N-linked glycosylation. Prediction of O-glycosylation sites in Awp1-related adhesins was done using NetOGlyc 4.0 [22] and identified numerous potential glycosylation sites, especially in the B-regions, where one third, on average, of the residues are predicted to carry a covalently attached oligomannan modification. (S4 Table). This coincides with glycosylation predictions performed on B-regions of many other fungal cell wall proteins [23]. Interestingly, the modifi- cation pattern is diverse for the A-region of the Awp1-related adhesins. The A-regions of the Awp1 and Awp3 subclusters (Fig 1) are almost all predicted to be heavily O-glycosylated at the α-crystallin domain and the following cysteine-comprising stretch (Awp1: 8 sites; Awp3b: 16 Sites), but not along the β-helix domain (Fig 6C and 6D). Accordingly, none of the potentially O-glycosylated residues is involved in formation of the Ser/Thr ladder at the PB1 sheet of the Awp1-A β-helix domain. The roles of the Ser/Thr and the Asn stacks on the surfaces Awp1-re- lated A-domain β-helices apart from structural stabilization of the β-helical turns [24] hence remain elusive. In contrast to the Awp1 and Awp3 subclusters, members of the Awp2 subclus- ter, i.e. Awp2, Awp2a-Awp2i and Awp4, apparently lack O-glycosylation sites on their α-crys- talline domains (S4 Table). Here, the only exceptions are Awp2d with five and Awp2i with two predicted sites. Obviously, one cannot decide whether this different modification pattern is caused either by the alternative attachment mode of Awp2 A-regions to the B-region via the CxxC motif that follows the A-region or the sheer size of their B-regions, on average ~2000 res- idues (Awp1 subcluster: ~1000; Awp3 subcluster: ~850). N-linked glycosylation sites as predicted by NetNGlyc 1.0 [25] are only scarcely found in Awp1-related adhesins (S4 Table). In Awp1, N224, the most C-terminal residue of the Thr/ Asn-stack on the PB1 sheet, may be such a potential target site for N-linked glycosylation, but this site is not predicted for other A-regions of Awp1-related adhesins. Interestingly, in the α- crystallin domains of Awp2-subcluster members a conserved site for N-linked glycosylation can be found. In Awp2, the N291 site resides on the loop linking β-strands 31 and 32; in Awp1 and Awp3 subcluster adhesins, this loop is used for linking the C-terminal end of the A-region to the α-crystallin domain by a disulfide bridge (Fig 5B). Other surface cues that may play a role in the adhesive/anti-adhesive properties of Awp1-re- lated adhesins are surface electrostatics and hydrophobicity. In terms of electrostatics, the Awp3b A-region is clearly distinct from Awp1 and Awp2 due the former’s negatively charged surface (Fig 7A). This charge distribution of the Awp3b A-region is apparently the reason for its capability to bind in crystallo 42 Gd3+ ions on its surface, with 21 of them forming the so far largest structurally defined lanthanide cluster (Fig 7B). Only five Gd3+ ions are found to coordi- nate to the C-terminal α-crystallin domain. This propensity for coordinating multivalent cat- ions may modulate any adhesive/anti-adhesive effects caused by Awp3b. Hydrophobic surface bands made up of aromatic residues have been described for the Flo11 adhesins of Saccharomy- ces cerevisiae [18,26], which mediate, similar to Awp2 of C. glabrata, the adhesiveness to plastic surfaces. Interestingly, groups of surface-exposed tyrosine side chains can be delineated only in the model of the Awp2 A-region (S7 Fig) with overall 12 residues, whereas the crystal structures of Awp1 and Awp3b show only 4 and 6 tyrosines on the protein surfaces, respectively. Structural relationship to other β-helix proteins Structural similarity of Awp1-A and Awp3b-A to proteins deposited in the PDB was analyzed by pairwise 3D alignments with PDBeFold v2.59 with the default cut-off of 70% for lowest PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 12 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins Fig 7. Electrostatics and lanthanide binding by Awp1, Awp2, and Awp3b A-regions. (A) Electrostatic molecular surface potentials of Awp1, Awp2, and Awp3b. The Awp3b surface structure is shown in superposition to the Gd3+ binding sites of the Awp3b•Gd complex. (B) The 42 Gd3+ binding sites of the Awp3b•Gd complex. Three Gd3+ clusters are found on the surface of the β-helix domain, comprising either 21, 4 or two lanthanide ions. Cluster 1 with its 21 Gd3+ ions is coordinated by only two residues, D40 and E59. It contains four tetrahedral subclusters (A, B, C, and D); subclusters A, B, and C with distances of 3.3–4.1 Å between their Gd3+ ions are connected to each other by triangular planar clusters composed of three Gd3+, with which they form a basket-like shape with three-fold symmetry. Distances of ions participating in composing those triangles range from 3.5 to 4.4 Å. Subcluster D is associated to the basket-like shape via a single Gd3+ ion. Atoms in this subcluster are a bit further apart from each other when compared to other tetrahedral clusters, namely 3.7–4.4 Å. Cluster 2 with four Gd3+ ions forms a regular tetrahedron that is connected to the protein via Q102, E132 and E134. Distances between the Gd3+ ions range from 2.8 to 3.8 Å, and the Gd3+ ions are 2.4 Å away from the carboxyl group O of the coordinating residues. Tetrahedral geometries of lanthanide clusters have been reported previously for complexes of the S. cerevisiae Flo5 A-domain with Gd3+[38]. Such clusters are of high interest for the development of novel contrast agents in magnetic resonance imaging [72]. https://doi.org/10.1371/journal.ppat.1009980.g007 acceptable similarity [27]. A number of proteins was identified to be similar to the β-helices of these putative adhesins, including the heme-haemopexin binding HxuA from Haemophilus influenza [28], a variety of polysaccharide lyases from different organisms (e. g. the pectate lyase Bsp165PelA from Bacillus sp. N16-5 [29], pectate lyase A from Erwinia chrysanthemi [30], alginate lyase from Paenibacillus sp. str. FPU-7 [31], as well as other polysaccharide-bind- ing proteins (e. g. the chitin-binding polysaccharide lyase-like protein Cthe_2159 from C. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 13 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins thermocellum [32], the Vi-antigen lyase VexL from Achromobacter denitrificans [33] or the ser- ine-rich repeat protein (SRRP100-23) from Lactobacillus reuteri [34]. These proteins all contain a three-faced right-handed β-helix. In general, sequence conservation was observed to be low, with sequence identities between Awp3b-A and search results ranging from 4.3% to 14.8% and RMSD values ranging from 2.63 to 6.04 Å, which indicates structural similarity. Similar results have also been observed for other β-helix proteins [28,32]. Interestingly, characteristic features of β-helix proteins such as the vWiDH sequence motif are not observed in the Awp1-A and Awp3b-A β-helices. Furthermore, many of the other β-helix proteins contain extensively long T3 loop regions, which often form additional secondary structure elements packed against the β-helix [21]. Such extrahelical domains are not seen in Awp1-related A-regions where the T1 loops represent the longest loops and do not form any regular structures. Parallel β-helices have been identified in the polysaccharide lyase families PL1, PL3, PL6, and PL9 [35]. In those enzymes, as well as in the polysaccharide lyase-like Cthe_2159 encoun- tered in the PDBeFold search, Ca2+ is required for ligand recognition [32,35]. Ca2+ depen- dency has also been observed for ligand binding in the Epa family of C. glabrata adhesins [36]. The use of lanthanides as probes for Ca2+ binding sites has been described on several occasions including the C-type lectin adhesion domain of the yeast adhesin Flo5 [37,38]. Consistent with this, potential Ca2+ coordination sites in Awp3b-A were delineated by numerous Ca2+ mim- icking Gd3+ ions in the structure of Awp3b-A-Gd. Most residues liganding to Gd3+ ions like D40 and E59, which coordinate the Gd21-cluster 1, or Q102, E132 and E134, which bind to the tetranuclear Gd3+ cluster 2 (Fig 7B) are not conserved among Awp1-related A-regions. An analysis of the Awp3b-A•Gd3+ complex by the CMM server [39] shows that only the binding site for Gd3+ cluster 3, which involves the side chains of E141, D169 and the carbonyl groups of K109 and R110, fulfills the geometric criteria expected for a mononuclear Ca2+-binding site. Accordingly, a structural alignment with the pectate lyase C from Dickeya chrysanthemi (PDB: 2EWE) as a representative of the search results from the PDBeFold search indicates that none of these putative Ca2+ binding sites in Awp3b-A is located at positions equivalent to Ca2+ bind- ing sites in the active site of pectate lyases and pectate lyase-related enzymes. Furthermore, there is no Ca2+ binding site predicted for Awp1-A because none of the lanthanides from the crystallization condition was observed in the Awp1 A-region structure. Despite their structural similarity to the aforementioned glycan-processing enzymes, our recombinant Awp1 and Awp3b A-regions failed to bind to any glycans as covered by the mammalian glycan array, ver- sion 5.2, from the Consortium of Functional Glycomics (deposit id: cfg_rRequest_3531). Discussion C. glabrata possesses a large number of putative GPI-modified wall adhesins proposed to be important for the pathogenicity of this fungus. Based on homology in their effector domains these proteins were classified into seven different groups/clusters [1,5,7]. Members of the Epa cluster have been well-characterized, both functionally as well as structurally [8–11]. However, till now, the majority of the putative adhesins in other clusters lack characterization. Proteomic studies revealed that, besides members of the Epa cluster, cluster V and VI proteins, i.e. Awp1 —Awp4, can be commonly found in cell wall preparations of C. glabrata strains [4,5,12]. We showed by bioinformatic analysis that the N-terminal A-regions of cluster V and VI cell wall proteins belong to the same protein family and are highly related to other fungal proteins like the haze-protetective factor (Hpf1) from Saccharomyces cerevisiae [40] or hypha-specific GPI-CWPs from Candida albicans like Hyr1 or Iff4, members of the Iff protein family [1]. Interestingly, our SSN analysis found that these and other, related, fungal A-regions belong to a much larger cluster involving many bacterial orthologs (Fig 1). Structural analysis of the PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 14 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins Awp1 and Awp3b A-regions proved this relationship, as these A-regions correspond to a fusion between a right-handed parallel β-helix fold and an α-crystallin domain; Awp1-A and Awp3b-A are hence structurally very similar with a pairwise r.m.s.d of only 1.4 Å. Our Awp1- like A-domains are distinct from so far available structures of A-domains of C. glabrata adhe- sins, namely Epa1, Epa6, and Epa9, all belonging to adhesin cluster I. These Ca2+-dependent adhesins contain a PA-14 domain that mediates glycan binding [8,9,41]. The Pwp family of C. glabrata adhesins, i.e. cluster II, is predicted to contain a PA-14 domain as well; however, no information for members of this family in terms of structures and cognate ligands is yet avail- able [36]. The parallel β-helix of Awp1-related adhesins structurally resembles pectate lyases and pec- tate methyl transferases of fungal and bacterial origin, respectively, as all have eight complete β-turns [42,43]. However, there is no predictable, conserved active site found along the β-helix surfaces of Awp1-related adhesins. Conserved Ca2+-binding sites, a prerequisite of enzymatic activity in polysaccharide lyases, are hence missing in structures of Awp1 and Awp3b A- regions. Furthermore, we failed to show by glycan arrays any interactions between these A- regions and a cognate oligosaccharide ligand. Besides other, glycan-dependent β-helix enzymes and tailspike proteins from bacteriophages, the Awp1-3 β-helices are structurally closely related to several bacterial secretion domains from autotransporter (AT) families (Fig 8A), some of them can form rather large β-helices. In pathogenic Gram-negative bacteria, these virulence factors are known to act as adhesins or haemolysins during host colonization, biofilm formation or adhesion to abiotic surfaces [44]. For example, a 500 nm long β-helix region that is located at the N-terminus of the filamentous hemagglutinin (FHA) from Borde- tella pertussis sticks out of the outer membrane and apparently mediates a multitude of inter- actions, e.g. to host glycolipids, host proteins and abiotic surfaces [45]. Another cell-exposed protein, HxuA from Haemophilus influenza (Fig 8A), mediates binding to the host’s serum Fig 8. Structural relationships and predicted arrangement on glycan cell wall of Awp1-related adhesins. (A) Structural superposition of the Awp1 A-region (black ribbon) and three bacterial right-handed β-helix proteins colored from the N-terminus (blue) to the C-terminus (red): The periplasmic alginate epimerase AlgG from Pseudomonas syringae, the tailspike protein from the E. coli bacteriophage HK620 and the hemopexin binding protein HxuA from Haemophilus influenzae. The r.m.s.d. values after superposition were calculated for the respective β-helix domains by PDBeFold [27]. Right, Alphafold2 models of the Ecm33 domain from S. cerevisiae (Uniprot entry P38248, residues S28-S362) and Awp12 from C. glabrata CBS138 (Q6FSI9, D20-C314). (B) Predicted association modes of the A-regions of cluster V and VI adhesins to the cell wall of C. glabrata. The repetitive, highly O-glycosylated repeats of the B-region are schematically depicted as spheres (ochre). The C-region (red) of these C. glabrata GPI-CWPs links the cell wall proteins via a GPI-remnant (hexagon, orange) to the outer β-1,6-glucan layer (grey). https://doi.org/10.1371/journal.ppat.1009980.g008 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 15 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins protein haemopexin for triggering bacterial iron uptake [46]. Given the topological similari- ties, we suggest that at least some of the Awp1-related adhesins may mediate specific interac- tions to host proteins, glycans or lipids as well. In addition, Awp1-related adhesins as analyzed in this study are apparently not the only fungal cell-wall anchored proteins with right-handed β-helices. For example, Alphafold2’s template-free prediction [47] of Ecm33, a cell wall protein widely distributed in Ascomycetes, shows a β-helix with 13 turns (Fig 8A). This shows that β- helices apparently form a large group of ascomycetal cell wall proteins besides the well charac- terized adhesion domains of the PA-14 type [6,8,10,11,17,41]. In the context of C. glabrata CBS138, we found by Alphafold2 that adhesins from cluster III with its 13 members and clus- ter VII with six members [1] also adopt the A-region architecture of Awp1-related adhesins [48](Fig 8A). Surprisingly, this prompts the notion that the β-helix/α-crystallin family with its 42 assignable GPI-CWP (S5 Table) may play a similarly important role for the biology of C. glabrata as the PA-14 adhesion domain harboring GPI-CWP as the latter harbor only 20 members from the Epa [8–10] and 7 members from the Pwp-family [1,7]. The α-crystallin domain of Awp1-related adhesins apparently serves as a C-terminal cap for the β-helix fold, whose ninth turn is already incomplete with only two β-strands. N- and O- glycosylation sites are predicted to be rare in the β-helix fold, but numerous in the α-crystallin domain. We suggest that the glycosylated α-crystallin domain, alike the majority of the B- region, is embedded into the β-glucan portion of the Candida cell wall, where these Awp1-re- lated adhesins are covalently linked via their GPI remnant at their C-terminal end. Cluster V adhesins like Awp2 but not cluster VI adhesins like Awp1 and Awp3b are predicted to cross- link the C-terminal CXXC motif with the central region of the β-helix region. The difference between cluster V and VI adhesins is then the presentation mode of the β-helix region, as for cluster VI members it is predicted to stick straight out, whereas cluster V adhesins place this elongated domain flat-on along the cell wall surface (Fig 8B). Our phenotypic characterization of awp1,3 and awp2 deletion mutants demonstrated clearly a role of Awp2 but not Awp1 or the Awp3 paralogs in mediating hyperadhesiveness to polystyrene. For this, we deleted AWP2 in three distinct genetic backgrounds of C. glabrata, i.e. reference strain CBS138 and the hyperadhesive isolates PEU382 and PEU427. Surprisingly, adherence to polystyrene was diminished only for the latter two but not the lowly adherent ref- erence strain. Involvement of Awp2, but not Awp3, in adherence to polystyrene was further supported by studies with gain-of-function mutants expressing the A-domains of these puta- tive adhesins on the yeast cell surface. Interestingly, consequences of deleting AWP2 hugely depended on the genetic background. The awp2Δ mutants in PEU382 showed lowered adher- ence both after short (4 h) and long (24 h) incubation times. In contrast, strain PEU427, whose adherence to polystyrene is in-between PEU382 and reference strain CBS138, showed this behavior only for short incubation times (4 h). Furthermore, unlike the awp2Δ mutants in CBS138 and PEU427, the awp2Δ mutants in PEU382 also show a variety of other cell surface- related phenotypes, all of them converting the atypical surface features of PEU382 into a refer- ence strain-like characteristics. As CBS138, PEU382, and PEU427 all incorporate Awp2 in their cell walls the question is why the observed phenotypes are so distinct in these different genetic backgrounds. Based on proteo- mic peptide counts from biofilm cell wall preparations [12] (S2 Table), Awp2 is not more abun- dant in PEU382 than in the other two strains, therefore not providing an obvious explanation for its dominant role in PEU382. Another possiblity could be mutations in the AWP2 gene in the different strains, but sequence analysis of the AWP2 gene showed for both PEU strains, that orig- inate from the same hospital, only a single A2162G mutation relative to CBS138. This mutation causes a N721S conversion in the C-terminal spacer domain of Awp2 and hence fails to provide a plausible explanation for phenotypical differences. Nevertheless, when analyzing the total peptide PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 16 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins counts of putative adhesins in proteomic data from biofilms of the three parental strains remark- able differences become apparent [12] (S2 Table). Compared to CBS138, both hyperadhesive strains show increased numbers of adhesin peptides. In PEU427, the dominant adhesin family appears to be cluster I, i.e. Epa-like adhesins, whereas in PEU382 cluster V, to which Awp2 belongs, is enriched. Thirteen genes of this cluster have been identified in the genome of CBS138. Therefore, we cannot exclude that specific regulation and functional redundancy com- plicate studies that aim to decipher the contribution or importance of an individual gene of such a large family. Apart of Awp2, all three strains harbor other cluster V adhesins in their cell walls. More specifically, Awp4 is present in all three strains like Awp2; Awp2a was found in both PEU strains, Awp2b and Awp2c in PEU427, and Awp2d is only present and abundant in PEU382. Future studies with mutants in other genes of the family, including deletions of multiple genes are therefore desirable to further functional characterization of these adhesins. Our observation that Awp2 mediates C. glabrata adhesion to abiotic surfaces in some clini- cal isolates reminds of the function of other fungal cell wall proteins. For example, in the case of C. albicans, a phenotypical screen of 25 cell wall proteins for the ability to confer plastic adhesion identified Iff4, which additionally affects the infection burden in a mouse model for vaginal candidiasis [49]. Clearly, the N-terminal region of Iff4 is structurally related to Awp2 as it belongs to the family of Awp1-related adhesins as well (Fig 1). In S. cerevisiae, adhesion to hydrophobic surfaces like polystyrene as well as exceedingly stable cell-cell interactions are realized by Flo11 [26,50]. The adhesiveness of the Flo11 A-domain, at least to plastics, is medi- ated by two hydrophobic surface-exposed clusters of aromatic residues [18,26,51]. Interest- ingly, our model of the Awp2 A-region, but not the structures of Awp1 and Awp3b, shows a clustering of surface-exposed aromatic residues in the α-crystallin domain as well (S7 Fig). Together with the flat-on presentation mode of the A-region of Awp2 (Fig 8B), this suggests a distinct function for the α-crystallin domain, where the N-terminal β-helix domain may still fulfill other specificities in terms of adhesion or modulation of fungal cell wall characteristics. In this context one may even wonder whether some Awp1-related adhesins might confer an anti-adhesive function similar to Ywp1 from C. albicans [52]. Slightly reduced adhesiveness of Awp3b-overexpressing mutants was indeed observable (Fig 2D). The impressive multitude of paralogous genes encoding for Awp1-related adhesins in C. glabrata or C. albicans, implies a certain degree of functional redundancy, which currently impedes clear-cut phenotypical assignments in Candida yeasts. Analogous to Epa-type adhe- sins, the batteries of GPI-CWPs with β-helix/α-crystallin modules most likely allow pathogenic yeasts to exploit different ecological niches in host organisms by modulating either their adhe- sive specificity or overall surface characteristics. An intriguing possibility is that some of the adhesins are involved in cell-cell interactions, e.g. for promoting binding to hyphae of C. albi- cans (Tati et al. 2016). This might allow C. glabrata, which strictly grows as a budding yeast, to enter host tissues, and is consistent with the notion that C. glabrata is often found in mixed infections with C. albicans causing oral candidiasis (Coco et al., 2008; Miranda-Cadena et al. 2018). Based on our results on Awp1- and Epa-like adhesins, future studies will enable us to understand how C. glabrata modifies its surface characteristics in response to different envi- ronmental surface cues. Materials & methods Sequence similarity network analysis of the A-domains of Awp1-related cell wall proteins The A-regions of Awp1 (L19-R244, QHS67468.1) and Awp3a (G24-Q301, QHS67883.1) were used for iterative PSI-BLAST searches against the NCBI database of non-redundant protein PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 17 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins sequences. The cut-off criterion for the E-value was set to 10−4, the sequence coverage to at least 70%. After ten rounds, the aligned sequences were combined, redundant hits removed before subjecting the 11737 remaining sequences to a sequence similarity network (SSN) anal- ysis with the Enzyme Similarity Tool of the Enzyme Function Initiative (EFI-EST)[53]. For ini- tial SSN analysis, we used a BLAST-derived E-value stringency of 10−5. For edge removal, the alignment score E-value cut-off was decreased stepwise to finally 10−20 and 10−25, respectively. The resulting network with a pair-wise sequence identity greater than 80% for each node resulted in 4507 nodes. 10290 sequences were associated with full taxonomic description. The data were analyzed with Cytoscape 3.8.2, Clustal Omega and WEBLOGO [54–56]. Generation of deletion mutants Based on previous proteomic studies [12], AWP2 deletion mutants were generated in reference strain CBS138 as well as in two hyperadherent strains (PEU427 and PEU382) [12], whereas mutants for cluster VI genes AWP1 and AWP3 were generated in CBS138 only. An awp1Δ/ awp3Δ double mutant was also generated. The new version of the CBS138 genome [7] was published in the course of this work. The described 5 kb genomic insertion in the AWP3 locus converts this gene in two tandem genes, AWP3a and AWP3b. Our awp3Δ and awp1Δ/awp3Δ mutants are lacking both AWP3a and AWP3b. Deletion mutants were generated using the SAT1-flipper system [57] in combination with CRISPR-Cas9 to achieve more efficient integration into the correct genomic loci. The SAT1- flipper system, originally designed for C. albicans, harbours a SAT1 gene for transformant selection and employs an inducible FLP1 gene for excision of the integrated cassette from the genome to obtain clean mutants. To improve functionality of the flippase encoded in the deletion cassette in C. glabrata, the C. albicans adapted version (CaFLP1) of this gene in pSFS1a was replaced by the original FLP1 gene from S. cerevisiae. For preparation of the AWP deletion cassettes, flanking regions of about 0.5 kb of the adhesin genes were amplified using proofreading KAPA polymerase and cloned into the KpnI and XhoI (upstream frag- ment) and NotI and SacI (downstream fragments) sites of the modified pSFS1a vector. Cor- rectness of the deletion cassettes was verified by DNA sequencing. The RNA-protein (RNP) complex-based CRISPR-Cas9 system of IDT (Leuven, Belgium), whose application in Can- dida was described in detail by Grahl et al. (2017) [58], was employed for achieving correct integration into the genome of C. glabrata. Selection of transformants was performed on YPD (1% yeast extract, 2% peptone, 2% dextrose) agar containing 200 μg/mL nourseothricin (NT). Deletion mutants among transformants were identified by PCR. Correct integration in the genome was verified for both ends of the deletion cassettes using a combination of prim- ers that covered the junctions, and deletion of the AWP genes was also verified with internal AWP gene primers. Deletion cassettes were removed from the genome by inducing the flip- pase gene, which is under control of the CaSAP2 promoter, by growth in YCB-BSA-YE (1.17% yeast carbon base, 0.4% bovine serum albumin, 0.2% yeast extract, pH 4,0) and select- ing slow-growing colonies on plates containing a low concentration of 10 μg/mL NT. Loss of the deletion cassettes was verified by PCR (Fig 2). For each AWP gene, at least two deletion mutants from independent transformation experiments were obtained and in all phenotypic assays data of mutants therefore represent the average of two mutants. Notably, the AWP2 gene of the hyperadherent strains PEU382 and PEU427 harbors the point mutation N721S for the encoded adhesin as shown by sequencing. Oligonucleotides and gene-specific sgRNA guides used for this work are listed in S1 Table. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 18 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins Adhesion to polystyrene Adhesion to polystyrene was measured using two different assays depending on the time allowed to adhere (4 or 24 h). For the 24 h experiment, overnight cultures in YPD at 37˚C were adjusted to a cell density of OD600 = 0.5 in fresh YPD, and 200 μL of the cell suspension was pipetted into a 96-well plate and incubated at 37˚C for 24 h in a humid environment. Unattached cells were removed by gentle washing with mQ water, and the remaining adhered cells were stained with 0.1% crystal violet (CV) solution for 30 min followed by washing with mQ water. Finally, CV was solubilized in 33% glacial acetic acid and quantified by measuring the OD595 using a microplate reader (Molecular Devices). Data for each strain are the average of six technical and at least two biological replicates. Polystyrene adhesion of S. cerevisiae strains that overexpress hybrid C. glabrata constructs containing the A-domains of Awp2 and Awp3b (from BG2 background) at the cell surface, detailed in Zupancic et al., 2008 [8] and Tati et al., 2016 [59], was tested using the same assay but incubating in synthetic complete medium (2% glucose, 1.1% acid casein peptone, 0.6% ammonium sulfate, 0.2% yeast nitrogen base (YNB, without ammonium sulfate and without amino acids) at 30˚C. Non-adhesive parental strain BY4741 and C. glabrata reference strain CBS138 were added as controls. For the 4 h adhesion experiment, overnight cultures were diluted to OD600 = 0.05 in PBS, and 0.5 mL was pipetted in 12-wells plates or glass tubes and incubated for 4 h at 37˚C. Unbound cells were removed by two washes with PBS after which adhered cells were treated for 10 min with a 2.5% trypsin (from porcine pancreas, Sigma) in PBS solution. Finally, PBS was added to a final volume of 0.5 mL, cells were resuspended and measured using a MACS- Quant (Miltenyi Biotec) flow cytometer (FC). Data are the average of two biological replicates measured in triplicate. Adhesion to HeLa and HaCaT cells Confluent layers of HeLa and HaCaT cells in 12-well plates were prepared as described [60] but without antimycotics in the medium. Overnight cultures (YPD, 37˚C) of C. glabrata were diluted in pre-warmed RPMI medium to obtain a cell density of about 6.0 × 103 cells/mL. Forty μl of the cell suspensions were added to the HeLa cells and incubated for 2 h at 37˚C with 5% CO2. Non-adhered cells were pipetted off. Adhered cells were scraped off with a cell scraper and collected in 2 mL PBS. Cells in both fractions were concentrated by centrifuging and plated on YPD plates containing 2 μg/mL chloramphenicol for colony-forming unit (CFU) counting after growth. For each sample, a control without HeLa cells was performed to discard adhesion to plastic. Data shown are averages of at least six biological replicates per sample. Drug sensitivity assays Susceptibility to antifungals and cell wall perturbants was tested in 96-well plates following EUCAST guidelines. Twofold serial dilutions of compounds were prepared in YPD and mixed 1:1 with cells from overnight cultures diluted to an OD600 of 0.01. Plates were incubated for 24 h at 37˚C. Minimal inhibitory concentrations (MIC) were determined after reading the OD600 in a microplate reader. Compounds tested were amphotericin B, fluconazole, isavuconazole, caspofungin, micafungin, Calcofluor white and SDS. For CFW drop assays were done as an alternative assay for sensitivity. Tenfold serial dilutions were prepared from overnight cultures, and 4 μl were spotted on YPD plates containing 200 μg/mL CFW. Growth was monitored after 24 and 48 h of growth at 37˚C. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 19 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins Growth rate and sensitivity to zymolyase For determination of growth rates, cells from overnight cultures were inoculated 1/100 in fresh YPD and incubated in 200 μl volumes in 96-well plates at 37˚C with agitation in a SpectraMax 340PC microplate reader. The increase in OD600 was followed in time. For deter- mination of zymolyase sensitivity, cells were grown in YPD until log phase, collected, and resuspended in 10 mM Tris-HCl, pH 7.5 at an OD600 of 2.0 to which 0.25% of β-mercaptoetha- nol was added. After 1 h of incubation at room temperature (RT), 180 μl of cell suspensions and 20 μl of 10 U/mL zymolyase were mixed in a 96-well plate and placed in the microplate reader at 37˚C. Decrease in OD600 was measured each minute after a short mixing pulse. Curves represent averages of three technical and two biological replicates. Cell surface hydrophobicity Exponential phase and overnight cultures in YPD at 37˚C were washed two times with PBS and resuspended at an OD600 of 0.7. Cell suspensions were mixed with hexadecane in glass tubes at a 15:1 volume ratio. Upon 1 min of gentle vortexing, the phases were allowed to settle for 10 min after which the OD600 of the aquous phase was measured. Each strain was assayed twice with two technical replicates each. Sedimentation and aggregation Overnight cultures in YPD at 37˚C were transferred to glass tubes. Every 10 min 50 μl of sam- ple was carefully taken 2 cm below the surface of the cell suspension for OD600 measurements. After 1 h the cells were washed with PBS and the sedimentation experiment was repeated. The sedimentation test with PEU382 was also performed after adjusting the pH of the cell culture to 6.5 with PBS or NaOH. Cell aggregation of overnight cultures in YPD was observed with a Leica DM1000 micro- scope mounted with a MC170 HD digital camera. Cell wall protein and chitin content Cells grown to logarithmic phase in YPD at 37˚C were harvested or seeded in Petri dishes for biofilm development as detailed in Go´mez-Molero et al. [12]. Preparation of cell walls was per- formed using a Fastprep 24 instrument (MP Biomedicals) as described in Go´mez-Molero et al. [12] and references therein. Protein and chitin content were determined using colorimetric assays as described by Kapteyn et al. [61]. Statistical analysis Statistical significance of phenotypic data was analyzed by Student’s t-tests or one-way ANOVA followed by post hoc Delayed Matching to Sample (DMS) tests. P values <0.05 were considered statistically significant. Cloning of AWP1 and AWP3b in expression plasmids N-terminal effector domains of Awp1 and Awp3b were amplified using KAPA polymerase (Kapa Biosystems) and cloned into the BamHI and HindIII restriction sites of expression plas- mid pET28a (Novagen). Correctness of the clones was verified by DNA sequencing (STAB- Vida). Expression was performed in E. coli SHuffle T7 Express cells (New England Biolabs). Cells were grown at 37˚C in Luria broth (LB) medium to an OD600 of 0.6. Overexpression was induced by addition of 0.1 mM isopropyl-β-D-thiogalactopyranoside and 72 h further incuba- tion at 12˚C and 140 rpm in baffled flasks. Cells were harvested by centrifugation at 3200 g, PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 20 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins 4˚C, 20 min, washed with 50 mM NaH2PO4, 300 mM NaCl, pH 8.0 buffer, and stored at -80˚C until use. Protein purification Cell pellets were resuspended in Ni-NTA buffer and lysed using a microfluidizer (Microflui- dics). Cell lysate was cleared by centrifugation at 18000 rpm, 4˚C, for 40 min. Supernatant was filtered using a 0.45 μm syringe filter (Millipore) and applied on a 5 mL Ni-NTA column (Macherey Nagel), equilibrated with Ni-NTA buffer. Protein was eluted stepwise using increas- ing imidazole concentrations; eluted fractions were analyzed by SDS-PAGE. Fractions con- taining the target protein were pooled and concentrated using an Amicon Ultra concentrator with 30 kDa cut-off (Millipore). Concentrated protein samples were further purified by size exclusion chromatography, on a Superdex 200 pg column (GE Healthcare), equilibrated with SEC buffer (20 mM Tris-HCl, 300 mM NaCl, pH 8.0). All purification steps were performed at a temperature of 4˚C. Finally, protein purity was assessed by SDS-PAGE; protein solutions were stored at 4˚C. Protein crystallization The Awp1 A-region was crystallized in a hanging-drop vapor diffusion setup in a 1.2 μL drop (0.6 μL protein solution + 0.6 μL reservoir), equilibrated against 1 mL 0.1 M MOPSO/bis-tris pH 6.5, 10% (w/v) PEG 8000, 20% 1,5-pentanediol, 0.5 mM erbium (III) chloride hexahydrate, 0.5 mM terbium (III) chloride hexahydrate, and 0.5 mM ytterbium (III) chloride hexahydrate at a temperature of 18˚C. A protein concentration of 48 mg/mL was used, crystals were grown for 3–4 weeks. The Awp3b A-region was crystallized at a protein concentration of 24 mg/mL in a hanging-drop vapor diffusion setup in a 0.6 μL drop (0.3 μL protein solution + 0.3 μL res- ervoir), equilibrated against 1 mL of 0.2 M magnesium chloride, 0.1 M Tris pH 7.0, 3.0 M sodium chloride at an incubation temperature of 18˚C. Crystals were allowed to grow for 2–3 weeks. Crystals used for collection of the native dataset of Awp3b A-region were grown in a sit- ting drop vapor diffusion setup. A 0.6 μL drop (0.3 μL protein solution + 0.3 μL reservoir) was equilibrated against 50 μL 0.1 M sodium phosphate, 0.1 M potassium phosphate, 0.1 M MES pH 6.5, 2.0 M sodium chloride. Crystals were harvested after several months. Data collection and processing For collection of native datasets, crystals were harvested and directly flash-frozen in liquid nitrogen without additional cryoprotection. To enable phase determination of the Awp3b A- domain by single wavelength anomalous diffraction (SAD), crystals were transferred to a drop of mother liquid, containing 50 mM gadolinium (III) acetate. Crystals were allowed to sit in the drop for 90 min, and then flash-frozen in liquid nitrogen without additional cryoprotec- tion. Datasets were collected at 100 K at the European Synchrotron Radiation Facility (ESRF), Grenoble. Diffraction images were integrated in iMOSFLM [62] or XDS [63], data reduction was done in AIMLESS [64], run in the CCP4i2 [65] software suite. Structure solution and refinement Crystallographic phases of the Awp3b-Gd3+ complex were determined using CRANK2 [66], followed by refinement in REFMAC5 [67] and model building in ARP/wARP [68]. The Gd3+-derivative was used to solve a native dataset of the Awp3b A A-region with higher resolu- tion. The phases of the Awp1 A-region were determined via molecular replacement in Phaser [69], using the structure of Awp3b A-region as a starting model, followed by 20 cycles of PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 21 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins structure building using the ARP/wARP Web Service [68]. All structures were refined running iterative cycles model building in Coot [70] and phenix.refine [71] or REFMAC5 [67]. A sum- mary of the data collection and refinement statistics is given in Table 2. Supporting information S1 Table. Oligonucleotides and sgRNA guides used in this study. (XLSX) S2 Table. Semi-quantitative analysis of proteomics data of biofilm wall samples from Gomez-Molero et al. 2015 [12] based on peptide counting. (XLSX) S3 Table. Drug sensitivity of AWP mutants. (XLSX) S4 Table. O- and N-glycosylation sites of cluster V/VI Awp1-related adhesins of C. glabrata as predicted by NetOGlyc4.0. For comparison the prediction is also shown for the haze-pro- tective factor, Hpf1, of Saccharomyces cerevisiae. Notably, the Awp1-related region of Hpf1 (S337-S629) is not located at the N-terminus, but flanked by highly O-glycosylated repeat regions. (XLSX) S5 Table. Other Awp1-related adhesins in C.glabrata with right-handed β-helix domain as predicted by Alphafold 2. Besides cluster III and VII adhesins, three additional have been found by our SSN analysis (Fig 1). The open reading frame CAGL0J05159g of C. glabrata CBS138 (XP_002999571.1) has not been assigned to a gene entry in the assembly of Xu et al. 2020, due to a frameshift that is not found in other strains. However the N-terminal region also adopts the fold of an Awp1-related adhesin. (XLSX) S1 Fig. Structure-based multiple sequence alignment (MSA) of Awp1-like GPI-CWPs of C. glabrata. The MSA was generated for A-regions of C. glabrata GPI-CWPs with β-helix domains by 3D-coffee using the structural information of the Awp1-A and Awp3b-A domains. The nine β-helix turns are highlighted by blue-to-yellow coloured boxes; the α-crystalline domain by a cyan box. Cysteines are highlighted in yellow. Notably, pairwise sequence identi- ties of Awp2, Awp2a-Awp2i and Awp4 A-regions vary from 53.6% to 94.8% (cluster V), those of Awp1,Awp1a-Awp1e, Awp3a and Awp3b (cluster VI) from 18.1% to 64.1%. For clarity, dis- tantly related Awp1-like adhesins of cluster V (QHS65613.1, QHS67215.1, QHS68879.1) are omitted from the MSA. Anyway, pairwise sequence identities between clusters V and VI A- regions are low, i.e. in the range of 14.4% to 22.9%. (TIF) S2 Fig. AWP gene deletion strategy and PCR confirmation of deletion mutants. (A) Sche- matic structure of SAT1-flipping deletion constructs and AWP gene deletion procedure. CRISPR-Cas9 RNP complexes (not indicated) are added during transformation to aid integra- tion into the correct locus. Mutants are selected by PCR analysis of 5’ and 3’ integration junc- tions, after which the cassette is excised by inducing FLP1 expression. (B-D). PCR verification of AWP deletion mutants after excision of the cassette using external (Ext) and internal (Int) primers as indicated in (A). (B) Verification of awp1Δ mutants. (C) Verification of awp3abΔ mutants and awp1Δ /awp3abΔ mutants. (D) Verification of awp2Δ mutants in three different PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 22 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins genetic backgrounds. (TIF) S3 Fig. Awp2-mediated adhesion to plastics is metal-independent. Adhesion to polystyrene after 24 h of incubation in YPD in the presence of the indicated EDTA concentrations. Nota- bly, the MIC50 for all four tested strains is 0.8 mM. (TIF) S4 Fig. Protein and chitin content. Protein and chitin amount in the cell wall was measured using colorimetric assays as described in Kapteyn et al. (2001). Each strain was assayed twice with two technical replicates each. Statistically significant differences (Student’s t-tests or ANOVA, p<0.05) of mutants compared to their parental strain under the same conditions are indicated by asterisks. (TIF) S5 Fig. Sedimentation and aggregation of C. glabrata strain PEU382. (A) Sedimentation of a 37˚C overnight culture of hyperadhesive aggregating strain PEU382 in YPD with and with- out pH adjustment by adding PBS (final conc. 1×) or NaOH was followed in time. (B) Sedi- mentation of PEU382 cells grown overnight at 37˚C in YPD without or with supplementations as indicated. (C) Microscopical images of cells from (B) at t = 0. (TIF) S6 Fig. A priori predicted contact maps of Awp1-related A-regions. Contact maps were pre- dicted by trRosetta [19] using therewith derived multiple sequence alignments of 532, 195 and 140 homologous sequences for Awp1, Awp2, and Awp3b, respectively. The TM-scores of the obtained models were very high with values of 0.738–0.839. Predicted contacts between the C- terminal end of the Awp2 A-region and β-helix turns 7 and 8 are highlighted by arrows. These contacts are the base for the disulfide bridges C327-C201 and C330-C184 in the Awp2 model. (TIF) S7 Fig. Aromatic residues exposed on the surfaces of Awp1-related A-regions. Only tyro- sines have been found to be surface-exposed on Awp1-related A-regions. Awp2 harbors six residues each in the β-helix (coloured) and α-crystallin (gray) domain. (TIF) Acknowledgments We thank Diana Kruhl (Philipps-Universita¨t Marburg), Ana Moreno and Albert de Boer (UCLM, Molecular Mycology), Diego Ferna´ndez (UCLM, Molecular Oncology), and the staff of Beamline 14.1 at the BESSYII synchrotron in Berlin (Germany) and Beamline ID23-1 at the European Synchrotron Radiation Facility (ESRF) in Grenoble (France) for technical support. David F. Smith and Jamie Heimburg-Molinaro of the Consortium for Functional Glycomics (CFG) are thanked for performing in vitro screening of glycan specificity. Brendan Cormack (Baltimore, MD) is thanked for kindly providing S. cerevisiae AWP overexpression plasmids, and Oliver Bader (Go¨ttingen, Germany) for providing PEU strains. Author Contributions Conceptualization: Piet de Groot, Lars-Oliver Essen. Data curation: Viktoria Reithofer. Formal analysis: Lars-Oliver Essen. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 23 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins Funding acquisition: Piet de Groot, Lars-Oliver Essen. Investigation: Viktoria Reithofer, Jordan Ferna´ndez-Pereira, Marı´a Alvarado. Methodology: Piet de Groot, Lars-Oliver Essen. Project administration: Lars-Oliver Essen. Resources: Piet de Groot. Supervision: Piet de Groot, Lars-Oliver Essen. Validation: Lars-Oliver Essen. Visualization: Viktoria Reithofer, Marı´a Alvarado, Piet de Groot. Writing – original draft: Viktoria Reithofer, Jordan Ferna´ndez-Pereira, Piet de Groot, Lars- Oliver Essen. Writing – review & editing: Piet de Groot, Lars-Oliver Essen. References 1. de Groot PWJ, Bader O, de Boer AD, Weig M, Chauhan N. Adhesins in human fungal pathogens: Glue with plenty of stick. Eukaryotic Cell. 2013. pp. 470–481. https://doi.org/10.1128/EC.00364-12 PMID: 23397570 2. Kaur R, Domergue R, Zupancic ML, Cormack BP. A yeast by any other name: Candida glabrata and its interaction with the host. Curr Opin Microbiol. 2005; 8: 378–384. https://doi.org/10.1016/j.mib.2005.06. 012 PMID: 15996895 3. Valotteau C, Prystopiuk V, Cormack BP, Dufrêne YF. Atomic Force Microscopy Demonstrates that Candida glabrata Uses Three Epa Proteins To Mediate Adhesion to Abiotic Surfaces. mSphere. 2019; 4. https://doi.org/10.1128/mSphere.00277-19 PMID: 31043520 4. Kraneveld EA, de Soet JJ, Deng DM, Dekker HL, de Koster CG, Klis FM, et al. Identification and Differ- ential Gene Expression of Adhesin-Like Wall Proteins in Candida glabrata Biofilms. Mycopathologia. 2011; 172: 415–427. https://doi.org/10.1007/s11046-011-9446-2 PMID: 21769633 5. De Groot PWJ, Kraneveld EA, Qing YY, Dekker HL, Groß U, Crielaard W, et al. The cell wall of the human pathogen Candida glabrata: Differential incorporation of novel adhesin-like wall proteins. Eukar- yot Cell. 2008; 7: 1951–1964. https://doi.org/10.1128/EC.00284-08 PMID: 18806209 6. Veelders M, Bru¨ckner S, Ott D, Unverzagt C, Mo¨sch H-UU, Essen L-OO. Structural basis of flocculin- mediated social behavior in yeast. Proc Natl Acad Sci U S A. 2010; 107: 22511–22516. https://doi.org/ 10.1073/pnas.1013210108 PMID: 21149680 7. Xu Z, Green B, Benoit N, Schatz M, Wheelan S, Cormack B. De novo genome assembly of Candida glabrata reveals cell wall protein complement and structure of dispersed tandem repeat arrays. Mol Microbiol. 2020; 113: 1209–1224. https://doi.org/10.1111/mmi.14488 PMID: 32068314 8. Zupancic ML, Frieman M, Smith D, Alvarez RA, Cummings RD, Cormack BP. Glycan microarray analy- sis of Candida glabrata adhesin ligand specificity. Mol Microbiol. 2008; 68: 547–559. https://doi.org/10. 1111/j.1365-2958.2008.06184.x PMID: 18394144 9. Maestre-Reyna M, Diderrich R, Veelders MS, Eulenburg G, Kalugin V, Bruckner S, et al. Structural basis for promiscuity and specificity during Candida glabrata invasion of host epithelia. Proc Natl Acad Sci. 2012; 109: 16864–16869. https://doi.org/10.1073/pnas.1207653109 PMID: 23035251 10. Diderrich R, Kock M, Maestre-Reyna M, Keller P, Steuber H, Rupp S, et al. Structural Hot Spots Deter- mine Functional Diversity of the Candida glabrata Epithelial Adhesin Family. J Biol Chem. 2015; 290: 19597–19613. https://doi.org/10.1074/jbc.M115.655654 PMID: 26105055 11. Hoffmann D, Diderrich R, Reithofer V, Friederichs S, Kock M, Essen LO, et al. Functional reprogram- ming of Candida glabrata epithelial adhesins: The role of conserved and variable structural motifs in ligand binding. J Biol Chem. 2020; 295: 12512–12524. https://doi.org/10.1074/jbc.RA120.013968 PMID: 32669365 12. Go´mez-Molero E, de Boer AD, Dekker HL, Moreno-Martı´nez A, Kraneveld EA, Ichsan, et al. Proteomic analysis of hyperadhesive Candida glabrata clinical isolates reveals a core wall proteome and differen- tial incorporation of adhesins. FEMS Yeast Res. 2015; 15: 1–10. https://doi.org/10.1093/femsyr/fov098 PMID: 26546455 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 24 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins 13. Dwivedi P, Thompson A, Xie Z, Kashleva H, Ganguly S, Mitchell AP, et al. Role of Bcr1-activated genes Hwp1 and Hyr1 in Candida albicans oral mucosal biofilms and neutrophil evasion. PLoS One. 2011; 6: e16218. https://doi.org/10.1371/journal.pone.0016218 PMID: 21283544 14. Kempf M, Cottin J, Licznar P, Lefranc¸ois C, Robert R, Apaire-Marchais V. Disruption of the GPI protein- encoding gene IFF4 of Candida albicans results in decreased adherence and virulence. Mycopatholo- gia. 2009; 168: 73–77. https://doi.org/10.1007/s11046-009-9201-0 PMID: 19347602 15. Zallot R, Oberg N, Gerlt JA. The EFI Web Resource for Genomic Enzymology Tools: Leveraging Protein, Genome, and Metagenome Databases to Discover Novel Enzymes and Metabolic Path- ways. Biochemistry. 2019; 58: 4169–4182. https://doi.org/10.1021/acs.biochem.9b00735 PMID: 31553576 16. Brown SL, Stockdale VJ, Pettolino F, Pocock KF, De Barros Lopes M, Williams PJ, et al. Reducing hazi- ness in white wine by overexpression of Saccharomyces cerevisiae genes YOL155c and YDR055w. Appl Microbiol Biotechnol. 2007; 73: 1363–1376. https://doi.org/10.1007/s00253-006-0606-0 PMID: 17024473 17. Kock M, Bru¨ ckner S, Wozniak N, Maestre-Reyna M, Veelders M, Schlereth J, et al. Structural and Func- tional Characterization of PA14/Flo5-Like Adhesins From Komagataella pastoris. Front Microbiol. 2018; 9: 1–17. 18. Kraushaar T, Bru¨ ckner S, Veelders M, Rhinow D, Schreiner F, Birke R, et al. Interactions by the fungal Flo11 adhesin depend on a fibronectin type III-like adhesin domain girdled by aromatic bands. Struc- ture. 2015; 23: 1005–1017. https://doi.org/10.1016/j.str.2015.03.021 PMID: 25960408 19. Yang J, Anishchenko I, Park H, Peng Z, Ovchinnikov S, Baker D. Improved protein structure prediction using predicted interresidue orientations. Proc Natl Acad Sci U S A. 2020; 117: 1496–1503. https://doi. org/10.1073/pnas.1914677117 PMID: 31896580 20. Yoder MD, Jurnak F. The refined three-dimensional structure of pectate lyase C from Erwinia chry- santhemi at 2.2 Angstrom resolution. Implications for an enzymatic mechanism. Plant Physiol. 1995; 107: 349–364. https://doi.org/10.1104/pp.107.2.349 PMID: 12228363 21. Jenkins J, Mayans O, Pickersgill R. Structure and evolution of parallel β-helix proteins. J Struct Biol. 1998; 122: 236–246. https://doi.org/10.1006/jsbi.1998.3985 PMID: 9724625 22. Hansen JE, Lund O, Tolstrup N, Gooley AA, Williams KL, Brunak S. NetOglyc: Prediction of mucin type O-glycosylation sites based on sequence context and surface accessibility. Glycoconj J. 1998; 15: 115– 130. https://doi.org/10.1023/a:1006960004440 PMID: 9557871 23. Boisrame´ A, Cornu A, da Costa G, Richard ML. Unexpected Role for a Serine/Threonine-Rich Domain in the Candida albicans Iff Protein Family r. Eukaryot Cell. 2011; 10: 1317–1330. https://doi.org/10. 1128/EC.05044-11 PMID: 21841123 24. Richardson JS. The Anatomy & Taxonomy of Protein Structure. Adv Protein Chem. 1981; 34: 167–339. https://doi.org/10.1016/s0065-3233(08)60520-3 PMID: 7020376 25. Gupta R, Jung E, Brunak S. NetNGlyc: Prediction of N-glycosylation sites in human proteins. Prep. 2004. http://www.cbs.dtu.dk/services/NetNGlyc/ 26. Bru¨ckner S, Schubert R, Kraushaar T, Hartmann R, Hoffmann D, Jelli E, et al. Kin discrimination in social yeast is mediated by cell surface receptors of the Flo11 adhesin family. Elife. 2020; 9: 1–26. https://doi.org/10.7554/eLife.55587 PMID: 32286952 27. Krissinel E, Henrick K. Secondary-structure matching (SSM), a new tool for fast protein structure align- ment in three dimensions. Acta Crystallogr Sect D Biol Crystallogr. 2004; 60: 2256–2268. https://doi. org/10.1107/S0907444904026460 PMID: 15572779 28. Baelen S, Dewitte F, Clantin B, Villeret V. Structure of the secretion domain of HxuA from Haemophilus influenzae. Acta Crystallogr Sect F Struct Biol Cryst Commun. 2013; 69: 1322–1327. https://doi.org/10. 1107/S174430911302962X PMID: 24316822 29. 30. 31. Zheng Y, Huang CH, Liu W, Ko TP, Xue Y, Zhou C, et al. Crystal structure and substrate-binding mode of a novel pectate lyase from alkaliphilic Bacillus sp. N16-5. Biochem Biophys Res Commun. 2012; 420: 269–274. https://doi.org/10.1016/j.bbrc.2012.02.148 PMID: 22414692 Thomas LM, Doan CN, Oliver RL, Yoder MD. Structure of pectate lyase a: Comparison to other iso- forms. Acta Crystallogr Sect D Biol Crystallogr. 2002; 58: 1008–1015. https://doi.org/10.1107/ s0907444902005851 PMID: 12037303 Itoh T, Nakagawa E, Yoda M, Nakaichi A, Hibi T, Kimoto H. Structural and biochemical characterisation of a novel alginate lyase from Paenibacillus sp. str. FPU-7. Sci Rep. 2019; 9: 14870. https://doi.org/10. 1038/s41598-019-51006-1 PMID: 31619701 32. Close DW, D’angelo S, Bradbury ARM. A new family of β-helix proteins with similarities to the polysac- charide lyases. Acta Crystallogr Sect D Biol Crystallogr. 2014; 70: 2583–2592. https://doi.org/10.1107/ S1399004714015934 PMID: 25286843 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 25 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins 33. Liston SD, McMahon SA, Le Bas A, Suits MDL, Naismith JH, Whitfield C. Periplasmic depolymerase provides insight into ABC transporter-dependent secretion of bacterial capsular polysaccharides. Proc Natl Acad Sci U S A. 2018; 115: E4870–E4879. https://doi.org/10.1073/pnas.1801336115 PMID: 29735649 34. Sequeira S, Kavanaugh D, MacKenzie DA, Sˇ uligoj T, Walpole S, Leclaire C, et al. Structural basis for the role of serine-rich repeat proteins from Lactobacillus reuteri in gut microbe–host interactions. Proc Natl Acad Sci. 2018; 115: E2706–E2715. https://doi.org/10.1073/pnas.1715016115 PMID: 29507249 35. 36. Lombard V, Bernard T, Rancurel C, Brumer H, Coutinho PM, Henrissat B. A hierarchical classification of polysaccharide lyases for glycogenomics. Biochem J. 2010; 432: 437–444. https://doi.org/10.1042/ BJ20101185 PMID: 20925655 Timmermans B, D Las Peñas A, Castaño I. Adhesins in Candida glabrata. Fungi J. 2018; 1–16. https:// doi.org/10.3390/jof4020060 PMID: 29783771 37. Martin RB, Richardson FS. Lanthanides as probes for calcium in biological systems. Q Rev Biophys. 1979; 12: 181–209. https://doi.org/10.1017/s0033583500002754 PMID: 386412 38. Veelders M, Essen L-O. Complex Gadolinium-Oxo Clusters Formed along Concave Protein Surfaces. ChemBioChem. 2012; 13: 2187–2190. https://doi.org/10.1002/cbic.201200441 PMID: 22936626 39. Zheng H, Cooper DR, Porebski PJ, Shabalin IG, Handing KB, Minor W. CheckMyMetal: A macromolec- ular metal-binding validation tool. Acta Crystallogr Sect D Struct Biol. 2017; 73: 223–233. https://doi. org/10.1107/S2059798317001061 PMID: 28291757 40. Waters EJ, Pellerin P, Brillouet JM. A Saccharomyces mannoprotein that protects wine from protein haze. Carbohydr Polym. 1994; 23: 185–191. https://doi.org/10.1016/0144-8617(94)90101-5 41. Ielasi FS, Decanniere K, Willaert RG. The epithelial adhesin 1 (Epa1p) from the human-pathogenic yeast Candida glabrata: Structural and functional study of the carbohydrate-binding domain. Acta Crys- tallogr Sect D Biol Crystallogr. 2012; 68: 210–217. https://doi.org/10.1107/S0907444911054898 PMID: 22349222 42. Pelloux J, Ruste´ rucci C, Mellerowicz EJ. New insights into pectin methylesterase structure and function. Trends in Plant Science. 2007. pp. 267–277. https://doi.org/10.1016/j.tplants.2007.04.001 PMID: 17499007 43. Yoder MD, Lietzke SE, Jurnak F. Unusual structural features in the parallel β-helix in pectate lyases. Structure. 1993; 1: 241–251. https://doi.org/10.1016/0969-2126(93)90013-7 PMID: 8081738 44. Meuskens I, Saragliadis A, Leo JC, Linke D. Type V secretion systems: An overview of passenger domain functions. Front Microbiol. 2019; 10: 1–19. 45. Serra DO, Conover MS, Arnal L, Sloan GP, Rodriguez ME, Yantorno OM, et al. FHA-mediated cell-sub- strate and cell-cell adhesions are critical for Bordetella pertussis biofilm formation on abiotic surfaces and in the mouse nose and the trachea. PLoS One. 2011; 6: e28811. https://doi.org/10.1371/journal. pone.0028811 PMID: 22216115 46. 47. 48. 49. Zambolin S, Clantin B, Chami M, Hoos S, Haouz A, Villeret V, et al. Structural basis for haem piracy from host haemopexin by Haemophilus influenzae. Nat Commun. 2016; 7: 1–12. https://doi.org/10. 1038/ncomms11590 PMID: 27188378 Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly accurate protein struc- ture prediction with AlphaFold. Nature. 2021. https://doi.org/10.1038/s41586-021-03819-2 PMID: 34265844 Fernandez-Pereira J, Alvarado M, Go´ mez-Molero E, Dekker HL, Teresa M, Eraso E, et al. Characteri- zation of Awp14, A Novel Cluster III Adhesin Identified in a High Biofilm-Forming Candida glabrata Iso- late. Front Cell Infect Microbiol. 2021; 11: 1123. https://doi.org/10.3389/fcimb.2021.790465 PMID: 34869084 Fu Y, Luo G, Spellberg BJ, Edwards JE, Ibrahim AS. Gene overexpression/suppression analysis of candidate virulence factors of Candida albicans. Eukaryot Cell. 2008; 7: 483–492. https://doi.org/10. 1128/EC.00445-07 PMID: 18178776 50. Bru¨ckner S, Mo¨ sch HU. Choosing the right lifestyle: Adhesion and development in Saccharomyces cer- evisiae. FEMS Microbiol Rev. 2012; 36: 25–58. https://doi.org/10.1111/j.1574-6976.2011.00275.x PMID: 21521246 51. Bouyx C, Schiavone M, Teste MA, Dague E, Sieczkowski N, Julien A, et al. The dual role of amyloid-β- sheet sequences in the cell surface properties of flo11-encoded flocculins in saccharomyces cerevisiae. Elife. 2021; 10: 1–25. https://doi.org/10.7554/eLife.68592 PMID: 34467855 52. Granger BL. Insight into the antiadhesive effect of yeast wall protein 1 of Candida albicans. Eukaryot Cell. 2012; 11: 795–805. https://doi.org/10.1128/EC.00026-12 PMID: 22505336 53. Gerlt JA, Bouvier JT, Davidson DB, Imker HJ, Sadkhin B, Slater DR, et al. Enzyme Function Initiative- Enzyme Similarity Tool (EFI-EST): A web tool for generating protein sequence similarity networks. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 26 / 27 PLOS PATHOGENSStructural and functional analysis of β-helical Candida glabrata adhesins Biochim Biophys Acta—Proteins Proteomics. 2015; 1854: 1019–1037. https://doi.org/10.1016/j. bbapap.2015.04.015 PMID: 25900361 54. Sievers F, Wilm A, Dineen D, Gibson TJ, Karplus K, Li W, et al. Fast, scalable generation of high-quality protein multiple sequence alignments using Clustal Omega. Mol Syst Biol. 2011; 7: 539. https://doi.org/ 10.1038/msb.2011.75 PMID: 21988835 55. Cline MS, Smoot M, Cerami E, Kuchinsky A, Landys N, Workman C, et al. Integration of biological net- works and gene expression data using cytoscape. Nat Protoc. 2007; 2: 2366–2382. https://doi.org/10. 1038/nprot.2007.324 PMID: 17947979 56. Crooks GE, Hon G, Chandonia JM, Brenner SE. WebLogo: A sequence logo generator. Genome Res. 2004; 14: 1188–1190. https://doi.org/10.1101/gr.849004 PMID: 15173120 57. Reuß O, Vik Å, Kolter R, Morschha¨ user J. The SAT1 flipper, an optimized tool for gene disruption in Candida albicans. Gene. 2004; 341: 119–127. https://doi.org/10.1016/j.gene.2004.06.021 PMID: 15474295 58. Grahl N, Demers EG, Crocker AW, Hogan DA. Use of RNA-Protein Complexes for Genome Editing in Non- albicans Candida Species. mSphere. 2017; 2: e00218–17. https://doi.org/10.1128/mSphere. 00218-17 PMID: 28657070 59. 60. Tati S, Davidow P, Mccall A, Hwang-Wong E, Rojas IG, Cormack B, et al. Candida glabrata Binding to Candida albicans Hyphae Enables Its Development in Oropharyngeal Candidiasis. PLOS Pathog. 2016; 12: e1005522. https://doi.org/10.1371/journal.ppat.1005522 PMID: 27029023 Ferna´ ndez-Aroca DM, Roche O, Sabater S, Pascual-Serra R, Ortega-Muelas M, Sa´ nchez Pe´rez I, et al. P53 pathway is a major determinant in the radiosensitizing effect of Palbociclib: Implication in can- cer therapy. Cancer Lett. 2019; 451: 23–33. https://doi.org/10.1016/j.canlet.2019.02.049 PMID: 30872077 61. Kapteyn JC, Ter Riet B, Vink E, Blad S, De Nobel H, Van Den Ende H, et al. Low external ph induces HOG1-dependent changes in the organization of the Saccharomyces cerevisiae cell wall. Mol Microbiol. 2001; 39: 469–480. https://doi.org/10.1046/j.1365-2958.2001.02242.x PMID: 11136466 62. Battye TGG, Kontogiannis L, Johnson O, Powell HR, Leslie AGW. iMOSFLM: A new graphical interface for diffraction-image processing with MOSFLM. Acta Crystallogr Sect D Biol Crystallogr. 2011; 67: 271– 281. https://doi.org/10.1107/S0907444910048675 PMID: 21460445 63. Kabsch W. XDS. Acta Crystallogr D. 2010; 66: 125–132. https://doi.org/10.1107/S0907444909047337 PMID: 20124692 64. Evans PR, Murshudov GN. How good are my data and what is the resolution? Acta Crystallogr Sect D Biol Crystallogr. 2013; 69: 1204–1214. https://doi.org/10.1107/S0907444913000061 PMID: 23793146 65. Winn MD, Ballard CC, Cowtan KD, Dodson EJ, Emsley P, Evans PR, et al. Overview of the CCP4 suite and current developments. Acta Crystallographica Section D: Biological Crystallography. 2011. pp. 235–242. https://doi.org/10.1107/S0907444910045749 PMID: 21460441 66. Skuba´ k P, Pannu NS. Automatic protein structure solution from weak X-ray data. Nat Commun. 2013; 4: 2777. https://doi.org/10.1038/ncomms3777 PMID: 24231803 67. Vagin AA, Steiner RA, Lebedev AA, Potterton L, McNicholas S, Long F, et al. REFMAC5 dictionary: Organization of prior chemical knowledge and guidelines for its use. Acta Crystallogr Sect D Biol Crys- tallogr. 2004; 60: 2184–2195. https://doi.org/10.1107/S0907444904023510 PMID: 15572771 68. Langer G, Cohen SX, Lamzin VS, Perrakis A. Automated macromolecular model building for X-ray crys- tallography using ARP/wARP version 7. Nat Protoc. 2008; 3: 1171–1179. https://doi.org/10.1038/nprot. 2008.91 PMID: 18600222 69. McCoy AJ, Grosse-Kunstleve RW, Adams PD, Winn MD, Storoni LC, Read RJ. Phaser crystallographic software. J Appl Crystallogr. 2007; 40: 658–674. https://doi.org/10.1107/S0021889807021206 PMID: 19461840 70. Emsley P, Lohkamp B, Scott WG, Cowtan K. Features and development of Coot. Acta Crystallogr Sect D Biol Crystallogr. 2010; 66: 486–501. https://doi.org/10.1107/S0907444910007493 PMID: 20383002 71. Adams PD, Afonine P V., Bunko´ czi G, Chen VB, Davis IW, Echols N, et al. PHENIX: A comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr Sect D Biol Crystallogr. 2010; 66: 213–221. https://doi.org/10.1107/S0907444909052925 PMID: 20124702 72. Caravan P. Strategies for increasing the sensitivity of gadolinium based MRI contrast agents. Chem Soc Rev. 2006; 35: 512–523. https://doi.org/10.1039/b510982p PMID: 16729145 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009980 December 28, 2021 27 / 27 PLOS PATHOGENS
10.1371_journal.ppat.1008708	RESEARCH ARTICLE Rho factor mediates flagellum and toxin phase variation and impacts virulence in Clostridioides difficile Dominika TrzilovaID, Brandon R. Anjuwon-Foster, Dariana Torres Rivera, Rita TamayoID* Department of Microbiology and Immunology, University of North Carolina at Chapel Hill School of Medicine, Chapel Hill, North Carolina, United States of America * rita_tamayo@med.unc.edu Abstract The intestinal pathogen Clostridioides difficile exhibits heterogeneity in motility and toxin production. This phenotypic heterogeneity is achieved through phase variation by site-spe- cific recombination via the DNA recombinase RecV, which reversibly inverts the “flagellar switch” upstream of the flgB operon. A recV mutation prevents flagellar switch inversion and results in phenotypically locked strains. The orientation of the flagellar switch influences expression of the flgB operon post-transcription initiation, but the specific molecular mecha- nism is unknown. Here, we report the isolation and characterization of spontaneous sup- pressor mutants in the non-motile, non-toxigenic recV flg OFF background that regained motility and toxin production. The restored phenotypes corresponded with increased expression of flagellum and toxin genes. The motile suppressor mutants contained single- nucleotide polymorphisms (SNPs) in rho, which encodes the bacterial transcription termina- tor Rho factor. Analyses using transcriptional reporters indicate that Rho contributes to het- erogeneity in flagellar gene expression by preferentially terminating transcription of flg OFF mRNA within the 5’ leader sequence. Additionally, Rho is important for initial colonization of the intestine in a mouse model of infection, which may in part be due to the sporulation and growth defects observed in the rho mutants. Together these data implicate Rho factor as a regulator of gene expression affecting phase variation of important virulence factors of C. difficile. Author summary Phenotypic heterogeneity maintained by phase variation allows bacterial subpopulations to overcome potentially detrimental stresses in the environment, contributing to bacterial survival. Phase variation of flagella and toxins in C. difficile suggests that maintaining het- erogeneity of their production may be important for survival and virulence. In this study, we identified Rho as a trans-acting factor that mediates the differential gene expression that imparts heterogeneity in flagellum and toxin production. These results reveal a new role for Rho-mediated transcription termination in regulation of gene expression. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Trzilova D, Anjuwon-Foster BR, Torres Rivera D, Tamayo R (2020) Rho factor mediates flagellum and toxin phase variation and impacts virulence in Clostridioides difficile. PLoS Pathog 16(8): e1008708. https://doi.org/10.1371/journal. ppat.1008708 Editor: Bruce A. McClane, University of Pittsburgh School of Medicine, UNITED STATES Received: March 3, 2020 Accepted: June 16, 2020 Published: August 12, 2020 Copyright: © 2020 Trzilova et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The sequencing data are posted in the NCBI Sequence Read Archive Database, accession number PRJNA630461. Funding: This work was supported by National Institutes of Health awards R01-AI107029 and R01-AI143638 to R.T. D.T.R. was supported by a fellowship from the University of North Carolina Chapel Hill Postbaccalaureate Research Education Program (NIH R25-GM089569). B.R.A-F. was supported by a National Research Service Award Individual Predoctoral Fellowship (F31-AI120613), PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 1 / 28 PLOS PATHOGENSa UNC-CH Dissertation Completion Fellowship, and a GlaxoSmithKline Science Achievement Award from the United Negro College Fund. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Rho factor mediated phase variation Introduction Clostridioides difficile is a gram-positive, spore-forming anaerobe and the leading cause of anti- biotic-associated diarrheal disease [1]. C. difficile infections (CDI) are currently among the most common hospital-acquired infections and exhibit a high rate of recurrence [2]. Disease is largely mediated by two large glucosylating exotoxins, TcdA and TcdB [3–5]. These toxins inactivate members of the Rho family of GTPases, causing a loss of interaction between the GTPase and its downstream effectors including those controlling the integrity of the actin cytoskeleton [6, 7]. In cell culture, toxin activity results in disruption of tight junctions, cell rounding, and eventually cell death [8, 9]. During infection of a host intestine, the toxins cause disruption of the epithelial barrier, leading to development of diarrheal symptoms, immune cell recruitment, and inflammation [10]. These toxins are also necessary for development of diarrheal disease in animal models of infection [4, 11, 12]. C. difficile encodes multiple surface appendages that contribute to cell adherence and/or colonization of the intestine, including fibronectin-binding protein Fbp68, type IV pili, and flagella [13–17]. Flagella are required for swimming motility and are also important for adher- ence to intestinal epithelial cells, colonization, and virulence in animal models of infection [18–20]. Similar to Bacillus subtilis and numerous gram-negative bacteria, the flagellar genes are expressed in a hierarchical manner to ensure the coordinated assembly of the complex structure [21–23]. C. difficile contains flagellar genes in at least four operons [24]. The early stage (flgB) flagellar operon includes the gene encoding SigD, a sigma factor essential for expression of late stage flagellar genes [23, 25, 26]. SigD also positively regulates the expression of toxin genes by activating transcription of tcdR, which encodes a positive regulator of tcdA and tcdB expression [23, 27]. Accordingly, factors that regulate expression of the flgB operon, such as cyclic diguanylate, impact toxin production via SigD [22, 25]. Other regulators, includ- ing Agr, SinR, and RstA, also affect both flagellum and toxin gene expression [28–32]. For example, SinR activates both flagellar and toxin gene expression [29, 30], while RstA acts as a negative regulator of both flagellum and toxin genes [31, 32]. We recently demonstrated the phase variation of flagella and toxins in multiple C. difficile ribotypes [33, 34]. Phase variation is a process that generates phenotypic heterogeneity in a bacterial population to help ensure the survival of the population as a whole in response to environmental selective pressures [35]. Typically, phase variation modulates the production of surface structures that directly interface with the bacterium’s environment. Several mucosal pathogens use phase variation to promote immune evasion and persistence in the host [36]. Phase variation occurs stochastically and is also fully reversible to maintain heterogeneity [35]. Multiple genetic and epigenetic mechanisms can lead to phase variation. One mechanisms is conservative site-specific DNA recombination, which involves recombinase-mediated inver- sion of a specific DNA sequence [37]. The invertible sequence typically contains regulatory information, such as a promoter element, that affects downstream gene expression [38]. DNA inversion requires the action of one or multiple serine or tyrosine recombinases that recognize inverted repeats flanking the invertible genetic sequence and catalyze DNA strand exchange that results in inversion of the intervening sequence [37, 39]. Recent work suggests that C. difficile employs recombination-mediated phase variation to generate an extensive degree of phenotypic heterogeneity. The first biologically characterized example of phase variation involved the cell wall protein V (CwpV) [40, 41]. Two additional potential DNA inversions were identified upstream of genes involved in cyclic diguanylate metabolism [42]. Most recently, high-throughput sequencing was used to identify a total of seven invertible sequences flanked by inverted repeats in the C. difficile NAP1/B1/027 ribotype strain R20291, and an eighth in several 017 ribotype strains [43, 44]. The identified sequences PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 2 / 28 PLOS PATHOGENSRho factor mediated phase variation were shown to undergo inversion, and the sequence upstream of the cmrRST operon was con- firmed to regulate expression of the downstream genes in a manner consistent with phase vari- ation [44, 45]. We subsequently showed that site-specific recombination also mediates phase variation of flagella and toxins by inversion of a genetic sequence called the “flagellar switch” [33]. The 154 bp flagellar switch is flanked by imperfect inverted repeats and is located upstream of the flgB operon. In one orientation, the flagellar switch is permissive for expres- sion of the flgB operon, resulting in “flg ON” bacteria exhibiting flagellum biosynthesis, swim- ming motility, and toxin production. Conversely, the inverse orientation reduces flgB operon transcription, yielding “flg OFF” bacteria that are aflagellate, non-motile, and attenuated for toxin production. Inversion of the flagellar switch requires the site-specific tyrosine recombi- nase RecV. Mutation of recV leads to genotypically and phenotypically “phase locked” strains that no longer undergo phase variation [33]. In contrast to these phase locked mutants, wild- type C. difficile R20291 in broth culture consists of a heterogeneous population of flg ON and OFF bacteria. Notably, RecV is also required for inversion of the cwpV switch, as well as two of the other identified invertible sequences including one shown to impact multiple phenotypes including virulence [40, 44, 45]. The canonical regulatory mechanism of phase variation by site-specific recombination involves an invertible promoter element that, when correctly oriented, promotes gene tran- scription in cis. For example, production of fimbriae by Escherichia coli and related species is regulated by the invertible fimbrial switch fimS, which contains a promoter for the adjacent fimA gene that encodes the fimbrial subunit [46, 47]. The flagellar switch in C. difficile lies in the 5’ untranslated region of the mRNA, between the previously identified σA-dependent pro- moter and the flgB coding sequence. Previous work showed that the flagellar switch does not contain a functional promoter [33]. Instead, regulation occurs post-transcription initiation and involves an unidentified trans-acting posttranscriptional regulator specific to C. difficile. In this study, we sought to identify the mechanism by which the flagellar switch controls expression of flagellum and toxin genes. Specifically, how does the orientation of the flagellar switch mediate the phase variable expression of the flgB operon? To answer this question, we used a non-motile recV phase-locked OFF strain in a suppressor analysis to identify factor(s) involved in inhibiting flgB operon transcription, swimming motility, and virulence character- istics. We recovered suppressor mutants that retained the flagellar switch in the OFF orienta- tion but gained extragenic mutations that restored swimming motility, toxin production, and expression of flagellum and toxin genes. The extragenic mutations conferring the suppressor phenotypes mapped to rho, which encodes the transcription termination factor Rho. Using a series of reporter fusions in C. difficile with or without an intact rho allele, we determined that Rho inhibits transcription from the 5’ leader sequence of flgB mRNA with the flagellar switch in the OFF orientation but not the ON orientation. These results suggest that Rho can discrim- inate between flg ON and flg OFF mRNA to selectively inhibit transcription in flagellar phase OFF variants, and reveal a role for Rho-mediated transcription termination phase variation of C. difficile flagella and toxins. Further phenotypic characterization of rho mutants additionally linked the loss of Rho to altered sporulation and ability to colonize the intestine in a murine model of infection, indicating a broader role for Rho-mediated transcription termination in C. difficile physiology. Results Motile suppressors arise from a non-motile recV flg OFF mutant In C. difficile strain R20291 (ribotype 027), the DNA recombinase RecV is required for flagellar switch inversion, and loss of recV leads to phenotypically-locked bacteria [33]. These recV PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 3 / 28 PLOS PATHOGENSRho factor mediated phase variation locked ON (recV flg ON, RT1702) and locked OFF (recV flg OFF, RT1693) strains are pheno- typically distinct. As previously shown, recV flg ON bacteria are motile in soft agar, while recV flg OFF bacteria typically remain non-motile even after 72 hours (Fig 1A); in a subset of experi- ments, we observed spreading of the recV flg OFF mutant after prolonged incubation in soft agar (�72 hours) (Fig 1A, right panel) [33]. In contrast to the uniform motility exhibited by WT and recV flg ON bacteria, these ‘motile flares’ were irregular and asymmetrical. The same phenomenon was observed for an independent recV flg OFF mutant, RT1694, which differs from RT1693 in the orientation of the invertible sequence upstream of cwpV, another phase variable locus (S1 Fig) [33, 48]. To recover the motile bacteria, which we term motile suppres- sors (MS) hereafter, we collected growth from the outer edge of the motile flares and subcul- tured on BHIS agar to obtain isolated colonies. When inoculated into soft agar, these MS showed motility similar to that of WT bacteria; an example is shown in Fig 1B. A total of 14 MS isolated from RT1693 and RT1694 were assayed for swimming motility in soft agar. All 14 MS regained swimming motility to the level of the recV flg ON strain, while the recV flg OFF and non-motile sigD-null control remained non-motile (S2 Fig); the results for 6 representa- tive MS are shown in Fig 1C. Consistent with these phenotypes, production of the flagellin FliC was restored in these 6 representative MS, with FliC levels equivalent to or higher than recV flg ON (Fig 1D). FliC was undetectable in the recV flg OFF parental strain. This observation suggested that the recV flg OFF mutant is capable of recovering motility. One possible explanation is that a recombinase other than recV catalyzed inversion of the fla- gellar switch. To test this possibility, we used orientation-specific polymerase chain reaction (OS-PCR), which employs primer pairs that specifically amplify one orientation of an invert- ible sequence or the other (S1 Table) [33]. Wild-type C. difficile R20291 grown in BHIS broth consists of a heterogeneous population. Accordingly, both OFF and ON flagellar switch orien- tations were detected by OS-PCR (Fig 1E). As determined previously, the recV flg OFF and recV flg ON strains yielded 280bp and 375bp product sizes, respectively, which correspond to the known flagellar switch orientations in these mutants [33]. The 14 MS contained the flagel- lar switch exclusively in the OFF orientation, with a representative 6 MS shown in Fig 1D. These results indicate that flagellar switch inversion did not occur, and the MS retain the flg OFF genotype despite their motile phenotype (Fig 1B). We also sequenced the promoter region and 5’ UTR of the flgB operon and did not find any mutations in the MS compared to the parental strains. Together, these results suggest that additional mechanisms are involved in inhibiting motility in C. difficile when the flagellar switch is in the OFF orientation. Identification of Rho as a negative regulator of flagellar motility Soft agar swimming motility assays provide a strong selective pressure for bacterial self-propul- sion to enable access to nutrients as they become depleted locally. The results above suggest that extragenic mutations arose in the MS that alleviated the negative effect of the flg OFF genotype on swimming motility in this assay. We postulated that a suppressor mutation occurred in a gene(s) involved in inhibiting flagellar gene expression in recV flg OFF bacteria. We thus performed whole genome sequencing of seven motile suppressors (MS 1–7) to iden- tify single nucleotide polymorphisms (SNPs) compared to the R20291 wild type reference genome (detailed in S2 Table, S1 Data). Compared to the recV flg OFF strain, five of seven MS carried SNPs upstream of CDR20291_1465, which encodes a putative Mn2+-containing cata- lase. The same five MS (MS 2–6) also carried SNPs in the region between CDR20291_1414 and 1415, which are convergently transcribed and encode the putative acetolactate synthase subunit IlvB and a phage-associated integrase, respectively. Additional SNPs appeared within the inverted repeats of switches upstream of cwpV and the flgB operon, in accordance with PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 4 / 28 PLOS PATHOGENSRho factor mediated phase variation Fig 1. Motile suppressor mutants arise from a non-motile recV flg OFF strain. (A) Swimming motility in soft agar medium of C. difficile R20291 (WT), recV flg ON (RT1702), recV flg OFF (RT1693), and sigD-null non-motile control. Images were taken of representative plates after 24, 48 and 72 h. The arrow on the 72 h image points to a motile flare arising from recV flg OFF. (B) Swimming motility in soft agar medium of C. difficile R20291 (WT), recV flg OFF (RT1694), MS5, and sigD-null non-motile control. Image of a representative plate at 48 h. (C) Quantification of swimming motility of 6 motile suppressors (MS) selected for further characterization. Swim diameter was measured after 48 h. The means and standard deviation from 4 biological replicates are shown. ��p<0.0001 by one-way ANOVA and Dunnett’s post-test compared to recV flg OFF. (D) Immunoblot detection of FliC in recV flg ON and OFF and select MS. Shown is a representative image of three independent experiments. Numbers represent quantification of band intensity expressed as fold-change relative to recV flg ON. ND–non-detectable. (E) Orientation-specific PCR for flagellar switch in six motile suppressors, WT, and recV flg ON and OFF controls. Band sizes– 280bp (OFF) or 375bp (ON). Shown is a representative image of three independent experiments. https://doi.org/10.1371/journal.ppat.1008708.g001 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 5 / 28 PLOS PATHOGENSRho factor mediated phase variation Table 1. Summary of single nucleotide polymorphisms in recV flg OFF motile suppressor mutants. recV flg OFF Suppressor Position on Chromosomea Ref Alt MS1 MS2 MS3 MS4 MS5 MS6 MS7 MS8 MS9 MS10 MS11 MS12 MS13 MS14 3955138 3955300 3955885 3955391 3956539 3955955 3956119 1433 1226 337 1598 980 973 754 T C C C T C C C T G A G G G C A T A / A A T C T G T A A Nucleotide Freqb 99.28 Coverb 978 99.13 97.66 99.05 89.83 99.22 99.02 N/A N/A N/A N/A N/A N/A N/A 2295 2055 2114 2056 2055 2447 N/A N/A N/A N/A N/A N/A N/A Locus Tag Gene Amino Acid Change 3324 3324 3324 3324 3324 3324 3324 3324 3324 3324 3324 3324 3324 3324 rho rho rho rho rho rho rho rho rho rho rho rho rho rho N533S G479V G284E E449-Stop N66-FS E261-Stop R206I S478F L409P E113-Stop N533S R327I D325N E252-Stop a Note that rho is encoded on the complementary strand, and reference and alternate (Ref/Alt) nucleotides are designated based on chromosome position for MS#1–7 and gene position, i.e. sense orientation, for MS#8–14. b Frequency of the alternate nucleotide and coverage of reads at each of the indicated positions. https://doi.org/10.1371/journal.ppat.1008708.t001 these strains containing inversions in the cwpV and flg switches relative to the R20291 refer- ence genome. Similarly, SNPs were also present in the inverted repeats flanking the invertible sequence upstream of CDR20291_0685, which was previously shown to be heterogeneous compared to the R20291 reference [44, 45]. A single locus, CDR20291_3324, contained SNPs in MS 1–7 compared to the allele in R20291 (Table 1, S1 Data); SNPs in this gene did not appear in either recV flg OFF parent. These SNPs showed a high frequency (89% or greater) in the sequencing reads. In lieu of whole genome sequencing, we Sanger sequenced CDR20291_3324 in MS 8–14 and identified additional SNPs in all of them. CDR20291_3324 encodes Rho factor, a transcriptional terminator known to act at the 3’ ends of genes or operons as an alternative mechanism to the use of an intrinsic terminator [49]. Rho binds nascent transcripts at Rho utilization (rut) sites, cytosine-rich sequences with poor con- sensus [49]. Hexameric Rho then uses ATPase activity to translocate 5’ to 3’ along the RNA. Although the mechanism of Rho-dependent transcription termination is not fully understood, it is thought to occur when Rho reaches the RNA polymerase, e.g. at an RNAP pause site, and forces its movement on template DNA without nucleotide addition, leading to destabilization of the transcription complex and mRNA release [50]. The SNPs identified in rho mapped to different domains of the Rho protein, including an N-terminal insertion domain (NID), primary binding site (PBS), and the C-terminal ATPase domain (Fig 2A). Mutations included substitutions (e.g. R206I, G284E) as well as nonsense mutations resulting either from a SNP creating a stop codon (e.g. E113-Stop) or a frameshift (e.g. N66-FS, resulting in a stop codon at residue 73). Rho is essential for viability in gram-negative bacteria but dispensable in gram-positive bac- teria studied to date, such as Bacillus subtilis and Staphylococcus aureus [51, 52]. Based on a transposon mutagenesis screen, Rho is not essential for viability in C. difficile [53]. However, loss of Rho can cause growth defects potentially due to inappropriate, pervasive read-through transcription [54–56]. Therefore, we assessed the growth of the 14 motile suppressors in BHIS broth. All motile suppressors reached a lower final optical density (OD600) compared to recV flg OFF (S3 Fig). While doubling times during exponential growth for the recV flg ON and PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 6 / 28 PLOS PATHOGENSRho factor mediated phase variation Fig 2. Rho is a negative regulator of flagellar motility. (A) Diagram of Rho factor domains and single nucleotide polymorphisms (SNPs) identified in the motile suppressors (MS). NHB–N-terminal helix bundle, NID–N-terminal insertion domain, PBS–primary binding site, FS–frameshift, St–stop codon. N66-FS results in a stop codon at residue 73. (B) Swimming motility in soft agar medium of motile suppressor 5 (MS5; N66-FS) and 10 (MS10; E113-St) expressing wild-type rho (pRho) or bearing vector at 48 hours. Controls included recV flg ON, recV flg OFF and non-motile sigD mutant carrying vector. (C) Quantification of swimming motility after 48 h of strains in (B). The means and standard deviation from 3 independent experiments are shown. ��p<0.0001, �p<0.05 by one-way ANOVA and Tukey’s post-test. (D) Immunoblot detection of FliC by strains in (B, C). Shown is a representative image of three independent experiments. Numbers represent quantification of band intensity expressed as the fold change compared to recV flg ON for the image shown. ND–non-detectable. (E) Relative transcript abundance of flgB and fliC measured by qRT-PCR. The means and standard deviation from 4 biological replicates are shown. ��p<0.01, �p<0.05 by one-way ANOVA and Dunnett’s post-test compared to recV flg OFF. https://doi.org/10.1371/journal.ppat.1008708.g002 OFF strains were 56.5 ± 6.0 and 63.6 ± 3.5, respectively, growth rates for the MS ranged from 71.8 to 129.9 minutes, and 10 of the MS showed significant attenuation of doubling time (Table 2). PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 7 / 28 PLOS PATHOGENSTable 2. Doubling times in rich medium. Rho factor mediated phase variation Strain recV flg ON recV flg OFF MS1 MS2 MS3 MS4 MS5 MS6 MS7 MS8 MS9 MS10 MS11 MS12 MS13 MS14 Doubling Time (min)a, b 56.5 ± 6.0 63.6 ± 3.5 75.7 ± 6.6 129.9 ± 12.2 98.0 ± 10.5 122.2 ± 10.5 123.1 ± 27.8 126.1 ± 27.8 75.9 ± 4.6 71.8 ± 5.5 98.2 ± 3.5 91.0 ± 3.3 76.0 ± 4.5 103.5 ± 3.4 85.1 ± 1.5 97.0 ± 4.7 a Means and standard deviations for data from two experiments each with 3 biological replicates (n = 6). b Bold indicates p < 0.05 by one-way ANOVA and Dunnett’s post-test compared to recV flg OFF. https://doi.org/10.1371/journal.ppat.1008708.t002 Because the rho mutations restored motility to the recV flg OFF mutants, we next tested the effect of expressing the wild-type rho allele in rho-null strains. We were unsuccessful in generat- ing an in-frame deletion of rho, likely because of the associated growth defects. In lieu of a tar- geted mutant, we utilized MS5 and MS10, which contain stop codons early in the rho coding sequence: N66-FS (stop codon at position 73) and E113-St, respectively (Fig 2A). A plasmid car- rying wild-type rho under control of an inducible promoter, pRho, was introduced into these MS, and the resulting strains were assayed for swimming motility. The vector control strains showed the expected swimming motility behaviors after 48 hours (Fig 2B and 2C). Expression of rho in MS5 and MS10 significantly inhibited swimming motility compared to the respective vec- tor controls, effectively complementing the effects of the SNPs in rho. Consistent with these phe- notypes, expression of rho led to 3- to 4-fold decreases in FliC levels in MS5 and MS10 compared to vector controls (Fig 2D). To examine whether changes in protein production and swimming were due to changes in flagellar gene transcription, we used quantitative reverse transcriptase PCR (qRT-PCR) to measure the transcript abundance of flgB, the first gene of the operon con- trolled by the flagellar switch, and fliC, which encodes flagellin and is regulated by SigD [22, 25]. The flgB transcript abundance was higher in MS5 and MS10 than recV flg OFF, though the differ- ences were not significant (Fig 2E, left). Both MS5 and MS10 had significantly increased fliC tran- script levels compared to recV flg OFF (Fig 2E, right). Providing rho in trans reduced flgB and fliC transcript abundance compared to the vector control. Notably, expression of rho also corrected the growth defect of MS5 and MS10 to the level of recV flg OFF parent with vector (S4 Fig). Together these results indicate that Rho regulates flagellum production and swimming motility in C. difficile by directly or indirectly inhibiting transcription of flagellar genes. Rho inhibits heterogeneous flagellar gene expression Analysis of flagellar gene expression by qRT-PCR reflects the transcript abundance averaged across the bacterial population. Yet, phase variation generates a heterogeneous population of PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 8 / 28 PLOS PATHOGENSRho factor mediated phase variation bacteria, some of which express flagellar genes, and some of which do not. To analyze the effects of Rho on flagellar gene expression at the single cell level, we used fluorescence micros- copy with a red fluorescence protein (mCherryOpt) reporter gene under the control of the flgM promoter (PflgM), which is a SigD-dependent promoter in the late stage flagellar operon [22, 23]. Expression of mCherryOpt from the flgM promoter is thus an indirect indication of the flagellar switch orientation. As previously observed, a population of wild-type C. difficile R20291 exhibited heterogeneity in fluorescence, with the majority of cells expressing mCher- ryOpt (Fig 3) [33]. In contrast, virtually no red cells were detectable for recV flg OFF C. difficile. Heterogeneous fluorescence was restored in MS5 and MS10 carrying PflgM::mCherryOpt (Fig 3). Thus, Rho is necessary for suppression of flagellar gene expression in recV flg OFF bacteria. Rho negatively impacts toxin production The alternative sigma factor SigD, encoded in the flgB operon, positively regulates the expres- sion of tcdR, tcdA, and tcdB [23, 27]. We therefore predicted that inhibition of flgB operon Fig 3. Rho inhibits heterogeneous flagellar gene expression in a recV flg OFF background. Micrographs of C. difficile R20291 (WT), MS5 and its recV flg OFF cwpV ON parent (RT1694), and MS10 and its recV flg OFF cwpV OFF parent (RT1693) transformed with the pPflgM::mCherryOpt reporter plasmid. Representative images for three independent experiments. Channels used are indicated for each column. White bars = 10 microns. https://doi.org/10.1371/journal.ppat.1008708.g003 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 9 / 28 PLOS PATHOGENSRho factor mediated phase variation Fig 4. Mutations in rho impact toxin production. (A) Immunoblot detection of TcdA by MS5 and MS10 bearing pRho for expression of wild-type rho or bearing vector. Controls included recV flg ON, recV flg OFF and sigD mutant carrying vector. Shown is a representative image for three independent experiments. Numbers represent quantification of band intensity expressed as the fold change compared to recV flg ON for the image shown. (B) Relative transcript abundance of tcdA, tcdB, and tcdR measured by qRT-PCR. The means and standard deviation from 3 to 5 biological replicates per strain are shown. ��p<0.0001, ��p<0.001, ��p<0.01, �p<0.05 by one-way ANOVA and Dunnett’s post-test compared to recV flg OFF. https://doi.org/10.1371/journal.ppat.1008708.g004 transcription by Rho would concomitantly inhibit toxin gene expression. We evaluated TcdA production in MS5 and MS10 carrying vector or pRho by immunoblot. MS5 and MS10 showed a 3- to 4-fold increase in TcdA production, respectively, compared to the recV flg OFF parent (Fig 4A). Expression of wild-type rho in MS5 and MS10 decreased TcdA levels, resulting in TcdA levels comparable to recV flg OFF bacteria. To determine whether changes in protein lev- els correlate with changes in transcript abundance, we examined expression of tcdR, tcdA, and tcdB by qRT-PCR. As observed previously, tcdA and tcdB transcript abundance was signifi- cantly higher in recV flg ON than recV flg OFF bacteria; tcdR was similarly altered, though the difference was not statistically significant (Fig 4B) [33]. Consistent with the negative impact of Rho on flagellar gene expression, tcdA, tcdB, and tcdR transcript levels were higher in MS5 and MS10 than in the recV flg OFF parent (Fig 4B). Providing rho in trans decreased tcdA, tcdB, and tcdR transcript abundances of MS5 and MS10 to the parental recV flg OFF levels. Therefore, in addition to inhibiting motility and growth, Rho negatively affects toxin gene expression and production. This effect is likely mediated through SigD encoded in the flgB operon [23, 27]. Mutant rho alleles confer dominant negative motility phenotypes Rho factor functions as a homohexamer [57]. We hypothesized that overexpression of mutant rho alleles could result in incorporation of aberrant subunits into the hexamer, interfering PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 10 / 28 PLOS PATHOGENSRho factor mediated phase variation with Rho function. To test for this dominant negative effect, we introduced mutant rho alleles from six different MS into the recV flg OFF strain, with the expectation that incorporation of non-functional Rho monomers would prevent inhibition of swimming motility by the wild- type, chromosomally-encoded Rho. The wild-type rho allele was also introduced as a control. To achieve overexpression of the cloned alleles, the genes were placed under the control of an anhydrotetracycline (ATc)-inducible promoter in the multi-copy plasmid pRPF185. Strains bearing these expression plasmids were assayed for swimming motility in soft agar medium. Expression of five of the six of the mutant rho alleles restored swimming motility to varying extents, while the wild-type allele did not alter motility (Fig 5A and 5B). These results were obtained regardless of ATc induction, suggesting leaky expression from the Ptet promoter as previously reported [58, 59]. The only mutant allele that did not lead to a dominant negative motility phenotype was N66-FS derived from MS5 (Fig 2A). Because this rho allele contains a nonsense mutation resulting in an early stop codon at residue 73, the truncated gene product may be unstable or unable to incorporate into the Rho hexamer, further justifying the use of the MS5 strain as a rho-null mutant. Rho preferentially inhibits transcription of flg OFF mRNA Our prior work suggests that transcription termination of flg OFF bacteria is mediated by a trans-acting factor specific to C. difficile [33]. This role could be fulfilled by Rho. While Rho typically terminates transcription of genes and operons 3’ of coding sequences, recent studies have shown that Rho can also terminate transcription in some 5’ leader regions in gram-nega- tive bacteria [60–63]. This mechanism would have a regulatory effect on the downstream gene (s). We therefore hypothesized that Rho is this trans-acting factor affecting transcription of the flgB operon, preferentially inhibiting transcription in flgB transcripts containing the flagellar switch in the OFF orientation. To test this hypothesis, we transcriptionally fused a phoZ reporter gene to the flgB coding sequence and the upstream regulatory region of the flgB operon. This 1045 bp region includes the σA-dependent promoter and the 498 bp 5’ untrans- lated region containing with the flagellar switch in either the ON or OFF orientation: PflgB- UTRON::phoZ and PflgB-UTROFF::phoZ respectively. Promoterless (::phoZ) and promoter-only Fig 5. Mutant rho alleles confer dominant negative motility phenotypes. (A) Swimming motility in soft agar medium after 72 hours for the recV flg OFF (RT1693) expressing rho alleles encoding the indicated Rho proteins. Labels correspond to SNPs in rho (Fig 2A), and WT corresponds to the wild-type rho allele. Dominant negative effects are present regardless of anhydrotetracycline (ATc) induction suggesting leaky expression from the Ptet promoter. (B) Quantification of swim diameter from soft agar plates containing 15ng/mL ATc after 72 hours. The means and standard deviation of three biological replicates are shown. ��p<0.0001, ��p<0.001, ��p<0.01 by one-way ANOVA and Dunnett’s post-test compared to recV flg OFF. https://doi.org/10.1371/journal.ppat.1008708.g005 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 11 / 28 PLOS PATHOGENSRho factor mediated phase variation Fig 6. C. difficile Rho directly inhibits flagellar gene expression in flg OFF orientation. (A) Alkaline phosphatase (phoZ) reporter fusions to the flgB operon regulatory region, with the flagellar switch in either the ON or OFF orientation, were introduced into the recV flg ON (RT1702), recV flg OFF (RT1693), MS5 and MS10 strains. Alkaline phosphatase activity was normalized to the recV flg OFF values. The means and standard deviation of 5 biological replicates are shown. ��p<0.0001 by two-way ANOVA and Tukey’s post-test. (B) The C. difficile rho gene was introduced into Bacillus subtilis bearing the phoZ fusions to the flgB regulatory region with the flagellar switch in either the ON or OFF orientation. C. difficile rho (Cd-rho) was introduced, and alkaline phosphatase activity was assayed. Control–B. subtilis reporter without Cd-rho; BCM–bicyclomycin, 50 μg/mL. The means and standard deviation of three biological replicates are shown. ��p<0.0001, �p<0.05 by two-way ANOVA with Sidak’s post-test comparing strains with the same reporter construct. https://doi.org/10.1371/journal.ppat.1008708.g006 constructs (PflgB::phoZ) were included as controls. These plasmid-borne reporters were intro- duced into recV flg ON and OFF strains, which encode wild-type Rho, and MS5 and MS10, which contain mutant rho alleles, and alkaline phosphatase activity was assayed [64]. As antici- pated, the no-promoter control lacked activity in all strains, and no differences were observed for activity of the promoter only reporter suggesting that Rho does not regulate transcription at the level of promoter (S5 Fig). For the PflgB-UTRON::phoZ reporter, activity was modestly (~3-fold) higher in MS5 and MS10 compared to the recV flg OFF parent (Fig 6A). In compari- son, activity was ~15-fold higher in MS5 and MS10 compared to recV flg OFF for the PflgB- UTROFF::phoZ reporter. Therefore, mutation of rho had a greater effect on flagellar gene tran- scription for bacteria with the flagellar switch in the OFF orientation, suggesting that Rho pref- erentially inhibits transcription of the flg OFF transcript. The results of alkaline phosphatase assays in C. difficile imply that Rho negatively regulates flgB operon transcription, but do not distinguish between a direct and indirect mechanism of regulation. Rho could directly act on the flgB UTR to terminate transcription from flgB OFF mRNA, or Rho could impact the production of another protein involved in flgB regulation. We previously showed that, whereas the PflgB-UTRON::phoZ reporter resulted in significantly higher activity than the PflgB-UTROFF::phoZ reporter in C. difficile, the difference was lost when these reporters were assayed in B. subtilis [33]. These results indicate that B. subtilis does not encode the factor that mediates regulation. We postulated that if Rho directly terminates tran- scription within the flgB 5’ UTR, introducing C. difficile rho (Cd-rho) into B. subtilis strains carrying the PflgB-UTRON::phoZ and PflgB-UTROFF::phoZ reporters would restore the regula- tion seen in C. difficile. To test this idea, the wild-type Cd-rho allele was introduced into the previously constructed B. subtilis reporter strains [33], and alkaline phosphatase activity was assayed. As seen previously, reporter activity was the same in B. subtilis with PflgB-UTRON:: PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 12 / 28 PLOS PATHOGENSRho factor mediated phase variation phoZ and PflgB-UTROFF::phoZ (Fig 6B). Expression of Cd-rho resulted in decreased activity only in B. subtilis with the PflgB-UTROFF::phoZ reporter. Finally, the addition of bicyclomycin, a specific inhibitor of Rho ATPase activity [65], abrogated this effect. These data indicate that Rho, and not another C. difficile protein, is the trans-acting factor which directly inhibits fla- gellar gene expression, and that Rho selectively prevents transcription in flg OFF bacteria. Rho is important for early colonization in a mouse model of infection and efficient sporulation In vitro, Rho affects several phenotypes including growth, motility, and toxin production. Because these characteristics are important during CDI, we analyzed the effect of a rho muta- tion in a mouse model of infection. MS5 and MS10 were derived from two recV mutant strains that differ in cwpV status–MS5 was derived from RT1694 (cwpV ON) while MS10 was derived from RT1693 (cwpV OFF) [48]. The role of cwpV in vivo has not been previously reported, so both recV flg OFF parental strains were tested. RecV is required for site-specific recombination of multiple invertible sequences, not only the flagellar and cwpV switches [44, 45, 48]. To ensure appropriate attribution of phenotypes, we confirmed that the parental strains and motile suppressors are isogenic for the other sequences and only differ in the cwpV switch (S6A Fig). Male and female C57BL/6 mice were treated with antibiotics to render them suscep- tible to C. difficile and then inoculated by oral gavage with 105 spores of wild-type R20291, MS5, MS10, and their respective parent strains. As observed previously, wild-type R20291 colonized the mice within 1 day post-inoculation (reaching more than 106 CFU/g feces), maintained this level of colonization for 1–3 days, then was gradually cleared typically between days 3 and 7 post-inoculation (S6B Fig). A similar, but not identical pattern of colonization was seen for both recV flg OFF strains, suggesting that cwpV expression does not consistently impact colonization in this model. For MS5 and MS10, the bacterial burden in feces was significantly lower on day 1 post-inoculation compared to the respective recV flg OFF parents and wild-type R20291 (Fig 7A). About 50% of the animals inoculated with MS5 or MS10 had undetectable levels of C. difficile in their feces; most of the remaining animals showed intermediate or parental levels of colonization. Interestingly, colo- nization of both MS5 and MS10 recovered to parental levels starting at day 2 post infection and were cleared within a similar time frame (S6B Fig), suggesting that Rho is important for initial colonization in a mouse model of infection. Notably, although the recV flg ON, MS5, and MS10 strains differ in toxin production compared to the recV flg OFF strain in vitro, we did not observe significant differences in weight loss or diarrheal symptoms between the groups of infected animals. These results may be attributable to the subclinical colitis caused by R20291 in this animal model, as previously reported [66–68]. It is also possible that other regulators of toxin gene expression, such as CodY and CcpA, unlink co-expression of the toxin and flagellar genes [69–72]. Sporulation and germination are important for colonization of the mouse model [73]. Because both MS5 and MS10 were attenuated for colonization on day 1 of infection, we con- sidered that this difference is attributable to a germination and/or sporulation defect. We assessed sporulation and spore viability by enumerating ethanol resistant spores as a percent- age of total cells (spore plus vegetative) [74]. While the sporulation efficiency for the wild type and both recV flg OFF strains was between 10 and 15%, sporulation efficiency was <1% for both motile suppressors (Fig 7B). Our data implicate Rho as an important factor that positively regulates sporulation, which may contribute to the colonization defect observed in the mouse model. It is possible that additional SNPs in the motile suppressors also contribute to the spor- ulation defect. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 13 / 28 PLOS PATHOGENSRho factor mediated phase variation Fig 7. Rho is important for early colonization in a mouse model of infection and efficient sporulation. (A) Antibiotic-treated male and female C57BL6 mice were inoculated with 105 spores of the indicated C. difficile strain. CFU in feces collected at 24 hours post inoculation were enumerated. Data are combined from two independent infection studies that each included 3 male and 3 female mice for 12 total mice per strain. Symbols indicate CFU from individual animals, and bars indicate the medians. ��p<0.0001, �p<0.05 by Kruskal-Wallis test with Dunn’s post-test comparing motile suppressors to their parent strains: RT1694 (recV flg OFF cwpV ON) and MS5, RT1693 (recV flg OFF cwpV OFF) and MS10. Dotted line represents a limit of detection. (B) Sporulation efficiency was evaluated by ethanol resistance and calculated as the total number of spores divided by the total number of viable cells (spores plus vegetative). A sporulation-deficient spo0A mutant was included as a control. The means and standard deviation of four biological replicates are shown. ��p<0.0001 by one-way ANOVA with Tukey’s post-test comparing all strains. n.d.–non-detectable. https://doi.org/10.1371/journal.ppat.1008708.g007 To examine germination, purified spores of wild-type R20291, both recV flg OFF strains, MS5, and MS10 were assayed in buffer supplemented with the spore germinant 10 mM tauro- cholic acid (TA) as previously described [75]. In the absence of TA, no germination was detected. In the presence of TA, all strains germinated to the same level, indicating that Rho does not influence germination rate (S7 Fig). Discussion In this study, we identified Rho as a trans-acting factor that controls phase variation of flagella and toxins in vitro. The regulation exerted by Rho contributes to the ability of a population of C. difficile to continually maintain motile, toxin-producing flg ON cells as well as non-motile, atoxigenic flg OFF cells. Comparative transcriptional analyses using C. difficile as well as B. sub- tilis as a heterologous system support that Rho inhibits flagellar gene transcription selectively in flg OFF bacteria. These results implicate Rho as an important regulatory component medi- ating phase variation of flagella, and by extension toxins, and reveal a new role for Rho-medi- ated transcription termination in regulation of gene expression. All of the 14 motile suppressors (MS) contained nucleotide polymorphisms in Rho confer- ring a missense or nonsense mutation that presumably abrogated Rho function. Five of six rho alleles, which correlated with a range of growth defects in the respective MS, led to a dominant negative effect and relieved inhibition of motility in flg OFF bacteria when over expressed. In E. coli, a dominant negative effect resulted from less efficient binding of mRNA to the Rho sec- ondary binding site and decreased translocation of Rho along the mRNA towards RNA poly- merase [76]. In C. difficile, incorporation of mutant subunits into the Rho homohexamer could negatively affect mRNA binding, ATP processing and/or resulting helicase activity. These mechanisms are not mutually exclusive–mutations in different domains of Rho may affect Rho activity by different mechanisms while imparting the same effect on motility. We PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 14 / 28 PLOS PATHOGENSRho factor mediated phase variation note that we were unable to generate an independent mutation in rho in a wild-type R20291 background, however the MS5 and MS10 strains with truncated Rho represent useful mutants for characterization of rho-null C. difficile. Further studies will determine how the mutant Rho proteins are altered in function and the mechanism by which mutant rho alleles interfere with function of wild-type Rho. Rho suppressor mutants exhibit restored motility compared to the flg OFF bacteria from which they were derived. The increased motility of motile suppressors corresponds with increased expression of flagellar genes flgB and fliC and higher levels of the major flagellin FliC. Furthermore, Rho contributes to heterogeneity of flagellar expression at the single cell level as evidenced by differences in mCherry signal driven by the SigD-dependent flgM pro- moter. Unlike the recV flg OFF strain that lacks fluorescence, motile suppressors derived from flg OFF bacteria are mCherry positive and appear similar to a wild-type population. That loss of Rho in the MS resulted in heterogenous fluorescence intensity among individual cells indi- cates that another factor influences expression. A c-di-GMP riboswitch is encoded between the flgB transcriptional start site and the flagellar switch, and we speculate that fluorescence intensity reflects varying levels of c-di-GMP [77, 78]. Experiments with transcriptional phoZ fusions in C. difficile and B. subtilis indicate that Rho strongly inhibits transcription when the flagellar switch is in the OFF orientation. We pro- pose two alternative models for direct control of phase variable expression of the flgB operon by Rho through the selective, premature termination of flg OFF transcripts (Fig 8). In model 1, Rho distinguishes between flg ON and OFF mRNA by preferentially binding to flg OFF mRNA due to the presence of rut sequences that are absent in the flg ON. In model 2, Rho binds flg ON and OFF mRNAs equally, 5’ of the flagellar switch. Transcription termination may then be differentially influenced by the presence of an additional sequence required for termination. For example, an RNA polymerase pause site may appear only in flg OFF mRNA [61]. In either model, Rho would selectively terminate flgB operon transcription and inhibit linked phenotypes in bacteria with the flagellar switch in the OFF orientation. More work is needed to distinguish between these two models. Using the RhoTermPredict algorithm [79], we were unable to identify any predicted rut sites within the flgB leader sequence of either flg ON or OFF sequences. RhoTermPredict is based on E. coli, B. subtilis, and S. enterica databases and searches for rut sites with regularly spaced C residues and C>G content followed by a Fig 8. Proposed models of direct Rho binding and inhibition of transcription readthrough of flg OFF mRNA. Model 1: Rho distinguishes between flg ON and OFF mRNA by preferentially binding to flg OFF mRNA. This may be due to the presence of rut sequences that are absent in the flg ON. Model 2: Rho binds flg ON and OFF mRNAs equally, 5’ of the flagellar switch. Transcription termination may then be differentially influenced by the presence of an additional sequence required for termination (e.g. a RNA polymerase pause site) only in flg OFF mRNA. https://doi.org/10.1371/journal.ppat.1008708.g008 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 15 / 28 PLOS PATHOGENSRho factor mediated phase variation putative RNA polymerase pause site [79]. However, the C. difficile genome has low G+C con- tent (<30%) [80], and the flgB leader sequence has 23% G+C content. The rut site characteris- tics in C. difficile therefore may be different from those previously described in other bacteria, so the RhoTermPredict algorithm may be unsuitable for predicting rut sites in C. difficile. There are several examples of Rho exerting regulation on 5’ leader sequences in gram-nega- tive bacteria, where gene regulation by Rho is achieved by multiple mechanisms involving other proteins, small RNAs, and potentially yet unidentified factors. In E. coli, Rho seems to preferentially regulate expression of >250 genes with long 5’ UTRs [81]. In addition, the RNA-binding protein CsrA binds to the 5’ UTR of pgaA to prevent the formation of an RNA secondary structure that otherwise sequesters the rut site [82]. In S. enterica serovar Typhimur- ium, Rho binds within the leader sequences of three genes encoding Mg2+ transporters to con- trol expression [60, 62, 83], and the small RNA ChiX inhibits expression of the chiPQ operon by inducing premature Rho-dependent termination [84]. Finally, most known E. coli ribos- witches modulate gene expression by either translational regulation or Rho-dependent termi- nation [60, 85]. However, riboswitches in C. difficile, including the c-di-GMP riboswitch upstream of the flgB operon, appear to act through Rho-independent mechanisms [77, 78]. To our knowledge, this is the first example of Rho-mediated transcription termination within a 5’ UTR that results in modulation of downstream gene expression in a gram-positive species. In E. coli, Rho requires cofactors NusA and NusG to terminate transcription at many sites [86]. NusG and NusA are essential for growth of C. difficile R20291 [53], but the cofactor requirements for C. difficile Rho are currently unknown. Introduction of C. difficile rho into a heterologous host B. subtilis did not alter the ability of C. difficile Rho to terminate transcrip- tion of the flg OFF construct. These data suggest that either Rho is able to terminate flagellar transcription without additional cofactors, or it is able to use homologs of NusA, NusG, or other potential cofactors present in B. subtilis. Interestingly, while both C. difficile and B. subti- lis encode Rho, only the C. difficile factor terminates flg OFF transcription. This difference could be caused by a difference in structure of these two proteins. Although many of the fea- tures of Rho are conserved across bacteria, in ~35% of species, including C. difficile, Rho con- tains an N-terminal insertion domain (NID) whose length and composition are not conserved among species [87]. In other bacterial species with an NID-containing Rho, the NID imparts diverse functions [88–90]. It is therefore possible that the insertion domain of C. difficile Rho confers the ability to terminate flg OFF transcription. Mutations in rho negatively affect initial colonization in a mouse model of infection, result- ing in a delay in colonization. We ruled out contributions from other phase-variable loci by ensuring that MS5 and MS10 are isogenic with the parental strains at these sites. The rho muta- tions likely have pleiotropic effects that impact colonization [91, 92], however the delayed colo- nization may be due in part to the defects in growth and sporulation of the motile suppressor mutants. Interestingly, a high-throughput screen in C. difficile R20291 did not identify rho as a gene required for sporulation [53]. How Rho affects growth and sporulation in C. difficile is unknown, but may arise from pervasive transcription, particularly loss of suppression of anti- sense transcription, or other potential consequences of loss of Rho [86, 93–95]. Further studies are needed to elucidate the effects of Rho on global transcription in C. difficile to determine the cause of the observed growth defects as well as other phenotypes affected by Rho. Materials and methods Growth and maintenance of bacterial strains Strains and plasmids used in this study are listed in S3 Table. C. difficile was maintained in an anaerobic chamber (Coy Laboratories) in an atmosphere of 85% N2, 5% CO2, and 10% H2. C. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 16 / 28 PLOS PATHOGENSRho factor mediated phase variation difficile and B. subtilis were routinely cultivated in Brain Heart Infusion medium (Becton Dick- inson) supplemented with 5% yeast extract (Becton Dickinson) (BHIS) at 37˚C. Where indi- cated, bacteria were cultured in Tryptone Yeast (TY) broth. All C. difficile broth cultures were grown statically, with 10 μg/mL thiamphenicol (Tm) for plasmid maintenance as needed. E. coli DH5α and HB101(pRK24) were cultured under aerobic conditions at 37˚C in Luria-Ber- tani (LB) broth. For selection of plasmids in E. coli, 100 μg/mL ampicillin (Amp) and/or 10 μg/ mL chloramphenicol (Cm) was used, as indicated. Kanamycin (Kan) 100 μg/mL was used to select against E. coli after conjugations with C. difficile. Spectinomycin (Spec) 100 μg/mL was used to select for B. subtilis transformants containing Cd-rho. Soft agar swimming motility assay Flagellum-dependent swimming motility was assayed in 0.5X BHIS-0.3% agar as previously described [25]. When appropriate, Tm was added for plasmid maintenance, and 10 ng/mL anhydrotetracycline (ATc) was added to induce gene expression. The diameter of motile growth was measured after 24, 48, and 72 hours. Three independent experiments were per- formed, each with six technical replicates. Images were taken using the G:BOX Chemi imaging system with the Upper White Light illuminator. Isolation and sequencing of motile suppressor mutants The recV flg OFF mutants RT1693 and RT1694 (which contain the cwpV switch in the OFF or ON orientation, respectively) were grown in BHIS broth until OD600 of 1, then 1.5 μL were inoculated into 0.5X BHIS-0.3% agar motility medium and incubated at 37˚C for 48–96 hours. Each plate included a non-motile sigD negative control (RT1566) and recV flg ON (RT1702) and recV flg OFF controls. Plates were examined for expansion of the recV flg OFF colonies, which appeared in a subset of plates. Bacteria were collected from the outer edge of motile growth and subcultured on BHIS agar. Genomic DNA was extracted from seven motile isolates (RT1705 to RT1711 (MS 1–7), S3 Table) and the parental recV flg OFF strains (RT1693, RT1694) as previously described [96]. Genomic DNA of MS 1–7 was prepared using the KAPA HyperPrep Kit (Roche) and sequenced using an Illumina HiSeq 2500 Rapid Run platform with paired ends and 100X cov- erage by the UNC-CH High Throughput Genomic Sequencing Facility. The sequencing data is available on the National Center for Biotechnology Information (NCBI) Sequence Read Archive Database, accession number PRJNA630461. Sequencing reads were mapped to the reference C. difficile R20291 genome (Accession No. FN545816.1) using CLC Genomics Workbench v. 9 software (Qiagen), and nucleotide polymorphisms were identified using the fixed ploidy variant detector function with default parameters. Whole genome sequencing was not performed for MS 8–14. Instead, for MS 8–14 (RT1939-1945), the rho gene (CDR20291_3324) was amplified by PCR with primers R2307 and R2308, and the products were Sanger sequenced using primers R2307, R2308, R2366, and R2367. Primer sequences are provided in S1 Table. Nucleotide polymorphisms were identified by alignment with the wild- type sequence from R20291 using ClustalOmega [97]. Determination of invertible switch orientation by orientation-specific PCR C. difficile was cultured from glycerol stocks on BHIS agar for 24 hours at 37˚C. A single col- ony was suspended in 20 μL of dH2O and heated at 100˚C for 10 minutes. These lysates served as templates for PCR using primers that discriminate between each flagellar switch sequence orientation in R20291 (S1 Table). Primers R1614 and R857 were used to amplify the ON orien- tation of the flagellar switch, which corresponds to the published sequence of R20291. Primers PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 17 / 28 PLOS PATHOGENSRho factor mediated phase variation R1615 and R857 were used to amplify the OFF orientation of the flagellar switch. Similarly, orientation-specific PCR was used to determine the orientations of the other invertible sequences using primers listed in S1 Table, which follow the naming pattern LOCUS_pubF and LOCUS_R for detection of the orientation in R20291 reference genome, and LOCUS_invF and LOCUS_R for the inverse orientations. Three independent experiments were done. Detection of FliC and TcdA by immunoblot Western blots for TcdA and FliC production were performed as previously described [33, 34]. Cultures for TcdA immunoblotting were grown in TY broth overnight (~16 hours), diluted 1:50 in fresh TY broth, and grown until late stationary phase (OD600 of 1.8 to 2.0). Cultures for immunoblotting FliC were grown overnight (~16 hours) in BHIS broth. For complementation experiments, Tm was included in all growth media, and 10 ng/mL ATc was added to induce gene expression. For both FliC and TcdA detection, samples were normalized to an OD600 1.0, and then cells were collected by centrifugation at 16,000 x g for 5 minutes (TcdA) or 2,000 x g for 10 minutes (FliC). Bacterial pellets were suspended in 1x SDS-PAGE sample buffer. The lysates were separated on a 12% SDS-polyacrylamide gel for FliC detection or on an 8% SDS- polyacrylamide gel for TcdA detection, then transferred to a nitrocellulose membrane (Bio- Rad). Membranes were stained with Ponceau S (Sigma) to assess equal loading and imaged using the G:Box Chemi imaging system. FliC was detected using α-FliC hamster sera (gener- ous gift from Dr. Ghose-Paul) [34, 98] followed by goat anti-hamster IgG (H+L) secondary antibody conjugated to DyLight 800 (Novus Biologicals). TcdA was detected using mouse α- TcdA antibody (Novus Biologicals) followed by goat anti-mouse IgG secondary antibody con- jugated to DyLight 800 4x PEG (Invitrogen). Blots were imaged using the Odyssey imaging system (LI-COR), and quantification was performed with Image Studio Software. All strains were assayed in at least three independent experiments. Growth curves Overnight cultures were diluted 1:50 into BHIS medium including 10 μg/mL Tm and 10 ng/ mL ATc as needed. Optical density (OD600) was measured every 30 minutes for 8 hours. Dou- bling times were calculated based on the change in optical density during exponential growth. Six biological replicates were assayed in two independent experiments. Quantitative reverse transcriptase-PCR Overnight cultures were diluted in BHIS medium containing thiamphenicol as needed. Cells were grown to mid-exponential phase (OD600 0.8–1) or stationary phase (OD600~1.5) for anal- ysis of flagellum (flgB, fliC) and toxin (tcdA, tcdB, tcdR) gene expression, respectively. RNA was isolated as described previously [33, 96]. Briefly, cells were collected by centrifugation and stored in ethanol:acetone (1:1) at -80˚C overnight. Cells were lysed by bead beating in cold Tri- SURE (Bioline). Nucleic acids were extracted with chloroform, precipitated from the aqueous phase with isopropanol, washed with ethanol, and suspended in RNase-free water. RNA was treated with TURBO DNase (Thermo Fisher) according to the manufacturer’s protocol. Syn- thesis of cDNA was done using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems) and random hexamers according to the manufacturer’s instructions. No-reverse transcriptase controls were included in all experiments. Real-time PCR was performed using 10 ng of cDNA, a final primer concentration of 1 μM, and SYBR Green Real-Time qPCR reagents (Bioline). Relative transcript abundance was calculated using the ΔΔCt method, with rpoC as the control gene and the indicated reference condition/strain. Primers used are listed PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 18 / 28 PLOS PATHOGENSRho factor mediated phase variation in S1 Table, with forward and reverse primers named according to the pattern gene-qF and– qR, respectively. Visualizing heterogeneity using fluorescent reporters To visually examine population heterogeneity, we used a previously described protocol [99, 100]. Briefly, overnight cultures of strains containing the PflgM::mCherryOpt reporter were diluted 1:100 into BHIS-Tm. Bacteria were grown anaerobically at 37˚C until OD600 ~0.5, 1 mL of culture was collected by centrifugation, and the remaining steps were performed aerobi- cally. Cell pellets were washed with PBS, then suspended in 500 μl PBS and 120 μl 5x fixative (20 μl NaPO4, pH 7.4; 100 μl 16% paraformaldehyde) [99]. The solution was incubated in the dark at room temperature for 30 minutes followed by 30 minutes at 4˚C. After the fixative was removed, cells were washed three times with PBS before suspension in 500 μl PBS and incuba- tion overnight in the dark at 4˚C to allow for fluorophore maturation. Slides for microscopy were prepared by placing 10 μl of concentrated culture onto a thin layer of 1% agarose applied directly to the surface of the slide. Microscopy was performed using a 60x oil immersion Nikon Plan Apo objective on a Keyence BZ-X810 equipped with Chroma 49005-UF1 for RFP detection. Generation of strains To generate rho expression plasmids, wild-type and mutant rho alleles were amplified from genomic DNA of recV flg OFF (RT1693) bacteria and 6 selected motile suppressors by PCR using primers R2308 and R2307 [96]. PCR products were cloned via the EcoRV and BamHI sites in pRT1611, a derivative of pRPF185 in which the gusA reporter gene was removed [33, 58]. After transformation into E. coli DH5α, Cm-resistant clones were recovered at 30˚C to hinder additional mutations in rho. The presence of the rho insert and its sequence integrity were confirmed using primers R2308, R2307, R2366, and R2367. The expression plasmids and the pRT1611 control were introduced into C. difficile strains RT1693 (recV flg OFF), RT1702 (recV flg ON), RT1709 (MS5), and RT1941 (MS10) via conjugation with E. coli HB101 (pRK24). The presence of the expected plasmid was confirmed by PCR with vector-specific primers R1832 and R1833. Plasmids containing transcriptional fusions of the phoZ alkaline phosphatase gene to flgB and iterations of the upstream regulatory region were created previously [33]. These plasmids were introduced into heat-shocked RT1693, RT1702, MS5, and MS10 [101]. For fluorescence microscopy, pRT1676, a pDSW1728 derivative carrying PflgM::mCherryOpt [33], was intro- duced into RT1693, RT1694, MS5, and MS10 by conjugation with E. coli HB101(pRK24) and confirmed by PCR. To introduce C. difficile rho (Cd-rho) into B. subtilis BS49, Cd-rho including its native ribo- somal binding site (RBS) was amplified from R20291 genomic DNA by PCR using R2656 and R2657, digested with HindIII and SphI, and ligated into similarly digested pDR111, which allows for integration at the amyE site [102]. The resulting plasmid was transformed into B. subtilis BS49 strains bearing previously described transcriptional fusions of phoZ to flgB and its upstream regulatory region, and transformants were selected on LB-Spec agar. Alkaline phosphatase assays Overnight (~16 h) cultures of B. subtilis BS49 and C. difficile phoZ reporter strains were diluted 1:50 (C. difficile) or 1:100 (B. subtilis) into BHIS medium. Thiamphenicol was added to C. diffi- cile growth media for plasmid maintenance. To induce expression of Cd-rho in B. subtilis, 0.5 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) was added to the growth medium when PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 19 / 28 PLOS PATHOGENSRho factor mediated phase variation cultures reached OD600 ~0.3. Controls without induction were processed in parallel. Where indicated, bicyclomycin (Cayman Chemical) was used at 50 μg/mL. Mid-exponential phase cells (OD600 0.8–1.3, 1 mL) were collected by centrifugation, the supernatant was discarded, and pellets were stored at -20˚C overnight. Frozen pellets were thawed on ice, and the alkaline phosphatase (AP) assay was performed as previously described [64]. Spore purification Overnight cultures (100 μL) were plated on ten 70:30 agar plates [103]. After 72 hours of growth at 37˚C, bacterial growth was scraped, suspended in 10 mL DPBS (Gibco) and kept at room temperature overnight. Spores were purified by collection of the growth in DPBS, then washing of the suspension four times with DPBS before purification using a sucrose gradient as described [104]. After discarding supernatant containing cell debris, the spore pellet was washed five more times with DPBS + 1% BSA. Spores were stored at room temperature until use. Germination assay Spore germination was analyzed at room temperature (27˚C) by measuring the change in OD600 [75]. Germination was carried out in clear 96-well flat bottom plates (Corning) in a final volume of 100 μl and final concentration of 30 mM glycine, 50 mM Tris, 100 mM NaCl, pH 7.5. Spores were heat-activated at 65˚C for 30 minutes, cooled on ice and suspended to a final OD600 of 0.7. At the initiation of the experiment, 10 mM sodium taurocholate (Sigma Aldrich) (TA) was added to induce germination; no-taurocholate controls were done in paral- lel. Optical density at 600 nm was measured every 2 minutes for 1 hour using a BioTek Synergy plate reader. Sporulation assay Sporulation assays were performed as described previously [74]. Briefly, C. difficile cultures were grown overnight in BHIS broth supplemented with 0.1% TA and 0.2% fructose to pre- vent spore accumulation. Cultures were diluted 1:30 in BHIS-0.1% TA-0.2% fructose, grown to an OD600 of 0.5 and 250 μl of culture applied to 70:30 agar as a lawn [103]. A control ethanol resistance sporulation assay was performed at this point to ensure no spores were present in exponential phase cultures. After 24 hours of incubation at 37˚C, cells were suspended in BHIS to an OD600 of 1.0, and an ethanol resistance sporulation assay was performed. A 0.5 ml aliquot was mixed with 0.5 ml of 57% ethanol to achieve a final concentration of 28.5% etha- nol, vortexed, and incubated for 15 minutes to eliminate all vegetative cells. Serial dilutions were made in PBS-0.1% TA and plated on BHIS-0.1% TA agar for spore enumeration. Vegeta- tive cells were enumerated by plating serial dilutions of the BHIS cell suspension on BHIS agar. Sporulation efficiency was calculated as the total number of spores divided by the total number of viable cells (spores plus vegetative). Ethics statement All animal studies were done in compliance with protocols approved by the UNC-CH Institu- tional Animal Care and Use Committee. Animal experiments Groups of eight- to ten-week-old female and male C57BL/6 mice (Charles River) were given a cocktail of antibiotics in their drinking water provided ab libitum for 3 days [105]. The PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 20 / 28 PLOS PATHOGENSRho factor mediated phase variation antibiotic cocktail consisted of kanamycin (400 mg/L), gentamicin (35 mg/L), colistin (850,000 units/L), vancomycin (45 mg/L), and metronidazole (215 mg/L) [105]. Four days prior to inoculation, the mice were switched to regular water for the remainder of the experi- ment. Clindamycin (10 μg/g body weight) was administered by intraperitoneal injection 48 hours prior to infection [14]. Mice were inoculated with 105 spores by oral gavage; control mice received PBS only. Inoculums were quantified by plating serial dilution on BHIS-0.1% TA agar and enumerating CFU. The animals were subsequently monitored for weight loss and diarrheal disease. Fecal samples were collected in pre-weighed tubes every 24 hours for 9 days. Fecal pellets were suspended in 1 mL DPBS and heated at 55˚C for 30 minutes. Serial dilutions were plated on TCCFA to enumerate CFU per gram feces [14]. Two independent experiments were done, each with 3 male and 3 female mice per C. difficile strain tested, for a total of 12 mice inoculated with each strain. Supporting information S1 Data. Single nucleotide polymorphisms identified in the MS1-7 and recV flg OFF strains compared to the R20291 reference genome. (XLS) S1 Table. Oligonucleotides used in this study. (DOCX) S2 Table. Complete analysis of single nucleotide polymorphisms (SNPs) in MS1-7 and recV flg OFF parent. (DOCX) S3 Table. Strains and plasmids used in this study. (DOCX) S1 Fig. recV flg OFF strains differ in status of cwpV switch. Orientation-specific PCR for the cwpV switch in recV flg OFF strains RT1693 and RT1694. Band sizes– 469bp (OFF) or 322bp (ON). (TIF) S2 Fig. All 14 motile suppressors have restored motility. Quantification of swimming motil- ity assays for the 14 MS, RT1693 (recV flg OFF), and RT1702 (recV flg ON). A non-motile sigD mutant was included as a control. The means and standard deviation of four biological repli- cates are shown. (TIF) S3 Fig. Motile suppressors have altered growth. Growth curves of MS1-14, recV flg ON (RT1702), the recV flg OFF parents (values combined for RT1694 (parent of MS1-7) and RT1693 (parent of MS8-14)). Shown are the means and standard deviation for 3 biological rep- licates. (TIF) S4 Fig. Expression of wild-type rho in MS5 and MS10 restores growth. Growth curves of MS5 and MS10 expressing wild-type rho (pRho) or bearing vector. The recV flg OFF and recV flg ON strains carrying vector were included. Expression of rho was induced with 10 ng/mL anhydrotetracycline (ATc). The means and standard deviation of three biological replicates are shown. (TIF) PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 21 / 28 PLOS PATHOGENSRho factor mediated phase variation S5 Fig. Rho does not affect flagellar gene expression at the level of promoter. An alkaline phosphatase (phoZ) reporter fusion to the flgB promoter (PflgB) and a promoterless:: phoZ con- struct were introduced into recV flg ON (RT1702), recV flg OFF (RT1693), MS5, and MS10. The means and standard deviation of 5 biological replicates are shown. (TIF) S6 Fig. Supporting data for animal studies. (A) Orientation-specific PCR for the 6 additional invertible sequences found in R20291, in recV flg OFF (RT1693), MS5, and MS10. WT R20291 was included as a control. Orientation is labelled as ON/OFF for the three invertible sequences whose regulation has been studied (cwpV, flg, cmrRST) or as published (pub) or inverse (inv) based on the R20291 reference genome for the Cdi2, Cdi3, and Cdi5 sequences whose effects on gene expression are not known. (B) Antibiotic-treated male and female C57BL6 mice were inoculated with 105 spores of the indicated C. difficile strain. CFU in feces collected every 24 hours post inoculation were enumerated as an indication of intestinal burden of C. difficile. Shown are the full courses of infection for two independent experiments that each included 3 male and 3 female mice. The data are separated by motile suppressor and its respective parent strain for clarity with the same data for wildtype R20291 in both upper and lower panels, with means and standard deviation shown. Dotted line represents a limit of detection. (TIF) S7 Fig. Rho is dispensable for C. difficile spore germination. Purified spores of indicated strains were germinated in the presence of taurocholate (+) or in buffer without germinant as a control (-), and optical density (OD600) was measured. Germination was plotted as the ratio of optical density (OD600) at a given time point (tx) versus initial OD600 (t0). A representative germination plot of four independent experiments is shown. (TIF) Acknowledgments We thank Kathleen Furtado for analysis using the RhoTermPredict algorithm, Elizabeth Shank for sharing pDR111, and Chandrabali Ghose-Paul and David Ho for gifting the hamster anti-FliC serum. Author Contributions Conceptualization: Dominika Trzilova, Brandon R. Anjuwon-Foster, Rita Tamayo. Data curation: Dominika Trzilova, Brandon R. Anjuwon-Foster, Rita Tamayo. Formal analysis: Dominika Trzilova, Brandon R. Anjuwon-Foster, Rita Tamayo. Funding acquisition: Brandon R. Anjuwon-Foster, Rita Tamayo. Investigation: Dominika Trzilova, Brandon R. Anjuwon-Foster, Dariana Torres Rivera, Rita Tamayo. Methodology: Dominika Trzilova, Brandon R. Anjuwon-Foster, Dariana Torres Rivera. Project administration: Rita Tamayo. Supervision: Rita Tamayo. Validation: Dominika Trzilova, Brandon R. Anjuwon-Foster, Dariana Torres Rivera. Writing – original draft: Dominika Trzilova, Brandon R. Anjuwon-Foster. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 22 / 28 PLOS PATHOGENSWriting – review & editing: Dominika Trzilova, Brandon R. Anjuwon-Foster, Dariana Torres Rho factor mediated phase variation Rivera, Rita Tamayo. References 1. Cartman ST, Heap JT, Kuehne SA, Cockayne A, Minton NP. The emergence of ’hypervirulence’ in Clostridium difficile. Int J Med Microbiol. 2010; 300:387–95. https://doi.org/10.1016/j.ijmm.2010.04. 008 PMID: 20547099 2. McFarland LV, Elmer GW, Surawicz CM. Breaking the cycle: treatment strategies for 163 cases of recurrent Clostridium difficile disease. Am J Gastroenterol. 2002; 97:1769–75. https://doi.org/10. 1111/j.1572-0241.2002.05839.x PMID: 12135033 3. Carter GP, Awad MM, Kelly ML, Rood JI, Lyras D. TcdB or not TcdB: a tale of two Clostridium difficile toxins. Future Microbiol. 2011; 6:121–3. https://doi.org/10.2217/fmb.10.169 PMID: 21366410 4. Kuehne SA, Cartman ST, Heap JT, Kelly ML, Cockayne A, Minton NP. The role of toxin A and toxin B in Clostridium difficile infection. Nature. 2010; 467:711–3. https://doi.org/10.1038/nature09397 PMID: 20844489 5. Chandrasekaran R, Lacy DB. The role of toxins in Clostridium difficile infection. FEMS Microbiol Rev. 2017; 41:723–50. https://doi.org/10.1093/femsre/fux048 PMID: 29048477 6. 7. Just I, Selzer J, Wilm M, von Eichel-Streiber C, Mann M, Aktories K. Glucosylation of Rho proteins by Clostridium difficile toxin B. Nature. 1995; 375:500–3. https://doi.org/10.1038/375500a0 PMID: 7777059 Just I, Wilm M, Selzer J, Rex G, von Eichel-Streiber C, Mann M, et al. The enterotoxin from Clostridium difficile (ToxA) monoglucosylates the Rho proteins. J Biol Chem. 1995; 270:13932–6. https://doi.org/ 10.1074/jbc.270.23.13932 PMID: 7775453 8. Nusrat A, von Eichel-Streiber C, Turner JR, Verkade P, Madara JL, Parkos CA. Clostridium difficile toxins disrupt epithelial barrier function by altering membrane microdomain localization of tight junction proteins. Infect Immun. 2001; 69:1329–36. https://doi.org/10.1128/IAI.69.3.1329-1336.2001 PMID: 11179295 9. Kasendra M, Barrile R, Leuzzi R, Soriani M. Clostridium difficile toxins facilitate bacterial colonization by modulating the fence and gate function of colonic epithelium. J Infect Dis. 2014; 209:1095–104. https://doi.org/10.1093/infdis/jit617 PMID: 24273043 10. Shen A. Clostridium difficile toxins: mediators of inflammation. J Innate Immun. 2012; 4:149–58. https://doi.org/10.1159/000332946 PMID: 22237401 11. Lyras D, O’Connor JR, Howarth PM, Sambol SP, Carter GP, Phumoonna T, et al. Toxin B is essential for virulence of Clostridium difficile. Nature. 2009; 458:1176–9. https://doi.org/10.1038/nature07822 PMID: 19252482 12. Carter GP, Chakravorty A, Pham Nguyen TA, Mileto S, Schreiber F, Li L, et al. Defining the Roles of TcdA and TcdB in Localized Gastrointestinal Disease, Systemic Organ Damage, and the Host Response during Clostridium difficile Infections. mBio. 2015; 6:e00551–15. https://doi.org/10.1128/ mBio.00551-15 PMID: 26037121 13. Hennequin C, Janoir C, Barc MC, Collignon A, Karjalainen T. Identification and characterization of a fibronectin-binding protein from Clostridium difficile. Microbiology. 2003; 149:2779–87. https://doi.org/ 10.1099/mic.0.26145-0 PMID: 14523111 14. McKee RW, Aleksanyan N, Garrett EM, Tamayo R. Type IV pili promote Clostridium difficile adher- ence and persistence in a mouse model of infection. Infect Immun. 2018; 86: e00943–17. https://doi. org/10.1128/IAI.00943-17 PMID: 29483294 15. Tasteyre A, Barc MC, Collignon A, Boureau H, Karjalainen T. Role of FliC and FliD flagellar proteins of Clostridium difficile in adherence and gut colonization. Infect Immun. 2001; 69:7937–40. https://doi. org/10.1128/IAI.69.12.7937-7940.2001 PMID: 11705981 16. Purcell EB, McKee RW, Bordeleau E, Burrus V, Tamayo R. Regulation of Type IV Pili Contributes to Surface Behaviors of Historical and Epidemic Strains of Clostridium difficile. J Bacteriol. 2015; 198:565–77. https://doi.org/10.1128/JB.00816-15 PMID: 26598364 17. Bordeleau E, Purcell EB, Lafontaine DA, Fortier LC, Tamayo R, Burrus V. Cyclic di-GMP riboswitch- regulated type IV pili contribute to aggregation of Clostridium difficile. J Bacteriol. 2015; 197:819–32. https://doi.org/10.1128/JB.02340-14 PMID: 25512308 18. Aubry A, Hussack G, Chen W, Kuolee R, Twine SM, Fulton KM, et al. Modulation of toxin production by the flagellar regulon in Clostridium difficile. Infect Immun. 2012; 80:3521–32. https://doi.org/10. 1128/IAI.00224-12 PMID: 22851750 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 23 / 28 PLOS PATHOGENSRho factor mediated phase variation 19. Faulds-Pain A, Twine SM, Vinogradov E, Strong PC, Dell A, Buckley AM, et al. The post-translational modification of the Clostridium difficile flagellin affects motility, cell surface properties and virulence. Mol Microbiol. 2014; 94:272–89. https://doi.org/10.1111/mmi.12755 PMID: 25135277 20. Baban ST, Kuehne SA, Barketi-Klai A, Cartman ST, Kelly ML, Hardie KR, et al. The role of flagella in Clostridium difficile pathogenesis: comparison between a non-epidemic and an epidemic strain. PLoS One. 2013; 8:e73026. https://doi.org/10.1371/journal.pone.0073026 PMID: 24086268 21. Mukherjee S, Kearns DB. The structure and regulation of flagella in Bacillus subtilis. Annu Rev Genet. 2014; 48:319–40. https://doi.org/10.1146/annurev-genet-120213-092406 PMID: 25251856 22. McKee RW, Mangalea MR, Purcell EB, Borchardt EK, Tamayo R. The second messenger cyclic Di- GMP regulates Clostridium difficile toxin production by controlling expression of sigD. J Bacteriol. 2013; 195:5174–85. https://doi.org/10.1128/JB.00501-13 PMID: 24039264 23. El Meouche I, Peltier J, Monot M, Soutourina O, Pestel-Caron M, Dupuy B, et al. Characterization of the SigD regulon of C. difficile and its positive control of toxin production through the regulation of tcdR. PLoS One. 2013; 8:e83748. https://doi.org/10.1371/journal.pone.0083748 PMID: 24358307 24. Stabler RA, He M, Dawson L, Martin M, Valiente E, Corton C, et al. Comparative genome and pheno- typic analysis of Clostridium difficile 027 strains provides insight into the evolution of a hypervirulent bacterium. Genome Biol. 2009; 10:R102. https://doi.org/10.1186/gb-2009-10-9-r102 PMID: 19781061 25. Purcell EB, McKee RW, McBride SM, Waters CM, Tamayo R. Cyclic diguanylate inversely regulates motility and aggregation in Clostridium difficile. J Bacteriol. 2012; 194:3307–16. https://doi.org/10. 1128/JB.00100-12 PMID: 22522894 26. Marquez LM, Helmann JD, Ferrari E, Parker HM, Ordal GW, Chamberlin MJ. Studies of sigma D- dependent functions in Bacillus subtilis. J Bacteriol. 1990; 172:3435–43. https://doi.org/10.1128/jb. 172.6.3435-3443.1990 PMID: 2111808 27. Mani N, Dupuy B. Regulation of toxin synthesis in Clostridium difficile by an alternative RNA polymer- ase sigma factor. Proc Natl Acad Sci U S A. 2001; 98:5844–9. https://doi.org/10.1073/pnas. 101126598 PMID: 11320220 28. Martin MJ, Clare S, Goulding D, Faulds-Pain A, Barquist L, Browne HP, et al. The agr Locus Regulates Virulence and Colonization Genes in Clostridium difficile 027. J Bacteriol. 2013; 195:3672–81. https:// doi.org/10.1128/JB.00473-13 PMID: 23772065 29. Ciftci Y, Girinathan BP, Dhungel BA, Hasan MK, Govind R. Clostridioides difficile SinR’ regulates toxin, sporulation and motility through protein-protein interaction with SinR. Anaerobe. 2019; 59:1–7. https://doi.org/10.1016/j.anaerobe.2019.05.002 PMID: 31077800 30. Girinathan BP, Ou J, Dupuy B, Govind R. Pleiotropic roles of Clostridium difficile sin locus. PLoS Pathog. 2018; 14:e1006940. https://doi.org/10.1371/journal.ppat.1006940 PMID: 29529083 31. Edwards AN, Anjuwon-Foster BR, McBride SM. RstA Is a Major Regulator of Clostridioides difficile Toxin Production and Motility. mBio. 2019; 10:954–71. 32. Edwards AN, Tamayo R, McBride SM. A novel regulator controls Clostridium difficile sporulation, motility and toxin production. Mol Microbiol. 2016; 100:954–71. https://doi.org/10.1111/mmi.13361 PMID: 26915493 33. Anjuwon-Foster BR, Tamayo R. A genetic switch controls the production of flagella and toxins in Clos- tridium difficile. PLoS Genet. 2017; 13:e1006701. https://doi.org/10.1371/journal.pgen.1006701 PMID: 28346491 34. Anjuwon-Foster BR, Maldonado-Vazquez N, Tamayo R. Characterization of flagellar and toxin phase variation in Clostridiodes difficile ribotype 012 isolates. J Bacteriol. 2018; 200:e00056–18. https://doi. org/10.1128/JB.00056-18 PMID: 29735765 35. 36. van der Woude MW. Phase variation: how to create and coordinate population diversity. Curr Opin Microbiol. 2011; 14:205–11. https://doi.org/10.1016/j.mib.2011.01.002 PMID: 21292543 van der Woude MW, Baumler AJ. Phase and antigenic variation in bacteria. Clin Microbiol Rev. 2004; 17:581–611. https://doi.org/10.1128/CMR.17.3.581-611.2004 PMID: 15258095 37. Grindley ND, Whiteson KL, Rice PA. Mechanisms of site-specific recombination. Annu Rev Biochem. 2006; 75:567–605. https://doi.org/10.1146/annurev.biochem.73.011303.073908 PMID: 16756503 38. Bayliss CD. Determinants of phase variation rate and the fitness implications of differing rates for bac- terial pathogens and commensals. FEMS Microbiol Rev. 2009; 33:504–20. https://doi.org/10.1111/j. 1574-6976.2009.00162.x PMID: 19222587 39. Johnson RC. Site-specific DNA Inversion by Serine Recombinases. Microbiolog Spectr. 2015; 3: MDNA3-0047-2014. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 24 / 28 PLOS PATHOGENSRho factor mediated phase variation 40. Emerson JE, Reynolds CB, Fagan RP, Shaw HA, Goulding D, Fairweather NF. A novel genetic switch controls phase variable expression of CwpV, a Clostridium difficile cell wall protein. Mol Microbiol. 2009; 74:541–56. https://doi.org/10.1111/j.1365-2958.2009.06812.x PMID: 19656296 41. Reynolds CB, Emerson JE, de la Riva L, Fagan RP, Fairweather NF. The Clostridium difficile cell wall protein CwpV is antigenically variable between strains, but exhibits conserved aggregation-promoting function. PLoS Pathog. 2011; 7:e1002024. https://doi.org/10.1371/journal.ppat.1002024 PMID: 21533071 42. Stabler RA, Valiente E, Dawson LF, He M, Parkhill J, Wren BW. In-depth genetic analysis of Clostrid- ium difficile PCR-ribotype 027 strains reveals high genome fluidity including point mutations and inver- sions. Gut Microbes. 2010; 1:269–76. https://doi.org/10.4161/gmic.1.4.11870 PMID: 21327033 43. Sekulovic O, Bourgeois J, Shen A, Camilli A. Expanding the repertoire of conservative site-specific recombination in Clostridioides difficile. Anaerobe. 2019; 60:102073. https://doi.org/10.1016/j. anaerobe.2019.102073 PMID: 31323290 44. Sekulovic O, Mathias Garrett E, Bourgeois J, Tamayo R, Shen A, Camilli A. Genome-wide detection of conservative site-specific recombination in bacteria. PLoS Genet. 2018; 14:e1007332. https://doi. org/10.1371/journal.pgen.1007332 PMID: 29621238 45. Garrett EM, Sekulovic O, Wetzel D, Jones JB, Edwards AN, Vargas-Cuebas G, et al. Phase variation of a signal transduction system controls Clostridioides difficile colony morphology, motility, and viru- lence. PLoS Biol. 2019; 17:e3000379. https://doi.org/10.1371/journal.pbio.3000379 PMID: 31658249 46. Abraham JM, Freitag CS, Clements JR, Eisenstein BI. An invertible element of DNA controls phase variation of type 1 fimbriae of Escherichia coli. Proc Natl Acad Sci U S A. 1985; 82:5724–7. https://doi. org/10.1073/pnas.82.17.5724 PMID: 2863818 47. Clegg S, Hancox LS, Yeh KS. Salmonella typhimurium fimbrial phase variation and FimA expression. J Bacteriol. 1996; 178:542–5. https://doi.org/10.1128/jb.178.2.542-545.1996 PMID: 8550478 48. Sekulovic O, Ospina Bedoya M, Fivian-Hughes AS, Fairweather NF, Fortier LC. The Clostridium diffi- cile cell wall protein CwpV confers phase-variable phage resistance. Mol Microbiol. 2015; 98:329–42. https://doi.org/10.1111/mmi.13121 PMID: 26179020 49. Ray-Soni A, Bellecourt MJ, Landick R. Mechanisms of Bacterial Transcription Termination: All Good Things Must End. Annu Rev Biochem. 2016; 85:319–47. https://doi.org/10.1146/annurev-biochem- 060815-014844 PMID: 27023849 50. Park JS, Roberts JW. Role of DNA bubble rewinding in enzymatic transcription termination. Proc Natl Acad Sci U S A. 2006; 103:4870–5. https://doi.org/10.1073/pnas.0600145103 PMID: 16551743 51. Washburn RS, Marra A, Bryant AP, Rosenberg M, Gentry DR. rho is not essential for viability or viru- lence in Staphylococcus aureus. Antimicrob Agents Chemother. 2001; 45:1099–103. https://doi.org/ 10.1128/AAC.45.4.1099-1103.2001 PMID: 11257021; PubMed Central PMCID: PMC90430. 52. Quirk PG, Dunkley EA Jr., Lee P, Krulwich TA. Identification of a putative Bacillus subtilis rho gene. J Bacteriol. 1993; 175:8053. https://doi.org/10.1128/jb.175.24.8053.1993 PMID: 8253694 53. Dembek M, Barquist L, Boinett CJ, Cain AK, Mayho M, Lawley TD, et al. High-throughput analysis of gene essentiality and sporulation in Clostridium difficile. mBio. 2015; 6:e02383–14. https://doi.org/10. 1128/mBio.02383-14 PMID: 25714712 54. Nicolas P, Mader U, Dervyn E, Rochat T, Leduc A, Pigeonneau N, et al. Condition-dependent tran- scriptome reveals high-level regulatory architecture in Bacillus subtilis. Science. 2012; 335:1103–6. https://doi.org/10.1126/science.1206848 PMID: 22383849 55. Botella L, Vaubourgeix J, Livny J, Schnappinger D. Depleting Mycobacterium tuberculosis of the tran- scription termination factor Rho causes pervasive transcription and rapid death. Nat Commun. 2017; 8:14731. https://doi.org/10.1038/ncomms14731 PMID: 28348398 56. Mader U, Nicolas P, Depke M, Pane-Farre J, Debarbouille M, van der Kooi-Pol MM, et al. Staphylococ- cus aureus Transcriptome Architecture: From Laboratory to Infection-Mimicking Conditions. PLoS Genet. 2016; 12:e1005962. https://doi.org/10.1371/journal.pgen.1005962 PMID: 27035918 57. Mitra P, Ghosh G, Hafeezunnisa M, Sen R. Rho Protein: Roles and Mechanisms. Annu Rev Microbiol. 2017; 71:687–709. https://doi.org/10.1146/annurev-micro-030117-020432 PMID: 28731845 58. Fagan RP, Fairweather NF. Clostridium difficile has two parallel and essential Sec secretion systems. J Biol Chem. 2011; 286:27483–93. https://doi.org/10.1074/jbc.M111.263889 PMID: 21659510 59. Oliveira Paiva AM, Friggen AH, Hossein-Javaheri S, Smits WK. The Signal Sequence of the Abundant Extracellular Metalloprotease PPEP-1 Can Be Used to Secrete Synthetic Reporter Proteins in Clos- tridium difficile. ACS Synth Biol. 2016; 5:1376–82. https://doi.org/10.1021/acssynbio.6b00104 PMID: 27333161 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 25 / 28 PLOS PATHOGENSRho factor mediated phase variation 60. Hollands K, Proshkin S, Sklyarova S, Epshtein V, Mironov A, Nudler E, et al. Riboswitch control of Rho-dependent transcription termination. Proc Natl Acad Sci U S A. 2012; 109:5376–81. https://doi. org/10.1073/pnas.1112211109 PMID: 22431636 61. Hollands K, Sevostiyanova A, Groisman EA. Unusually long-lived pause required for regulation of a Rho-dependent transcription terminator. Proc Natl Acad Sci U S A. 2014; 111:E1999–2007. https:// doi.org/10.1073/pnas.1319193111 PMID: 24778260 62. Kriner MA, Groisman EA. The Bacterial Transcription Termination Factor Rho Coordinates Mg(2+) Homeostasis with Translational Signals. J Mol Biol. 2015; 427:3834–49. https://doi.org/10.1016/j.jmb. 2015.10.020 PMID: 26523680 63. Kriner MA, Groisman EA. RNA secondary structures regulate three steps of Rho-dependent transcrip- tion termination within a bacterial mRNA leader. Nucleic Acids Res. 2017; 45:631–42. https://doi.org/ 10.1093/nar/gkw889 PMID: 28123036 64. Edwards AN, Pascual RA, Childress KO, Nawrocki KL, Woods EC, McBride SM. An alkaline phospha- tase reporter for use in Clostridium difficile. Anaerobe. 2015; 32:98–104. https://doi.org/10.1016/j. anaerobe.2015.01.002 PMID: 25576237 65. Kohn H, Widger W. The molecular basis for the mode of action of bicyclomycin. Curr Drug Targets Infect Disord. 2005; 5:273–95. https://doi.org/10.2174/1568005054880136 PMID: 16181146 66. Theriot CM, Koumpouras CC, Carlson PE, Bergin II, Aronoff DM, Young VB. Cefoperazone-treated mice as an experimental platform to assess differential virulence of Clostridium difficile strains. Gut Microbes. 2011; 2:326–34. https://doi.org/10.4161/gmic.19142 PMID: 22198617 67. Winston JA, Thanissery R, Montgomery SA, Theriot CM. Cefoperazone-treated Mouse Model of Clini- cally-relevant Clostridium difficile Strain R20291. J Vis Exp. 2016;(118). 68. Castro-Cordova P, Diaz-Yanez F, Munoz-Miralles J, Gil F, Paredes-Sabja D. Effect of antibiotic to induce Clostridioides difficile-susceptibility and infectious strain in a mouse model of Clostridioides dif- ficile infection and recurrence. Anaerobe. 2020; 62:102149. https://doi.org/10.1016/j.anaerobe.2020. 102149 PMID: 31940467 69. Antunes A, Martin-Verstraete I, Dupuy B. CcpA-mediated repression of Clostridium difficile toxin gene expression. Mol Microbiol. 2011; 79:882–99. https://doi.org/10.1111/j.1365-2958.2010.07495.x PMID: 21299645 70. Antunes A, Camiade E, Monot M, Courtois E, Barbut F, Sernova NV, et al. Global transcriptional con- trol by glucose and carbon regulator CcpA in Clostridium difficile. Nucleic Acids Res. 2012; 40:10701– 18. https://doi.org/10.1093/nar/gks864 PMID: 22989714 71. Dineen SS, Villapakkam AC, Nordman JT, Sonenshein AL. Repression of Clostridium difficile toxin gene expression by CodY. Mol Microbiol. 2007; 66:206–19. https://doi.org/10.1111/j.1365-2958. 2007.05906.x PMID: 17725558 72. Dineen SS, McBride SM, Sonenshein AL. Integration of metabolism and virulence by Clostridium diffi- cile CodY. J Bacteriol. 2010; 192:5350–62. https://doi.org/10.1128/JB.00341-10 PMID: 20709897 73. Deakin LJ, Clare S, Fagan RP, Dawson LF, Pickard DJ, West MR, et al. The Clostridium difficile spo0A Gene Is a Persistence and Transmission Factor. Infect Immun. 2012; 80:2704–11. https://doi. org/10.1128/IAI.00147-12 PMID: 22615253 74. Edwards AN, McBride SM. Determination of the in vitro Sporulation Frequency of Clostridium difficile. Bio Protoc. 2017; 7:e2125. 75. Shrestha R, Sorg JA. Hierarchical recognition of amino acid co-germinants during Clostridioides diffi- cile spore germination. Anaerobe. 2018; 49:41–7. https://doi.org/10.1016/j.anaerobe.2017.12.001 PMID: 29221987 76. Xu Y, Kohn H, Widger WR. Mutations in the rho transcription termination factor that affect RNA track- ing. J Biol Chem. 2002; 277:30023–30. https://doi.org/10.1074/jbc.M111009200 PMID: 12034708 77. McKee RW, Harvest CK, Tamayo R. Cyclic Diguanylate Regulates Virulence Factor Genes via Multi- ple Riboswitches in Clostridium difficile. mSphere. 2018; 3:e00423–18. https://doi.org/10.1128/ mSphere.00423-18 PMID: 30355665 78. Tamayo R. Cyclic diguanylate riboswitches control bacterial pathogenesis mechanisms. PLoS Pathog. 2019; 15:e1007529. https://doi.org/10.1371/journal.ppat.1007529 PMID: 30730989 79. Di Salvo M, Puccio S, Peano C, Lacour S, Alifano P. RhoTermPredict: an algorithm for predicting Rho- dependent transcription terminators based on Escherichia coli, Bacillus subtilis and Salmonella enter- ica databases. BMC Bioinformatics. 2019; 20:117. https://doi.org/10.1186/s12859-019-2704-x PMID: 30845912 80. Sebaihia M, Wren BW, Mullany P, Fairweather NF, Minton N, Stabler R, et al. The multidrug-resistant human pathogen Clostridium difficile has a highly mobile, mosaic genome. Nature Genet. 2006; 38:779–86. https://doi.org/10.1038/ng1830 PMID: 16804543 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 26 / 28 PLOS PATHOGENSRho factor mediated phase variation 81. Sedlyarova N, Shamovsky I, Bharati BK, Epshtein V, Chen J, Gottesman S, et al. sRNA-Mediated Control of Transcription Termination in E. coli. Cell. 2016; 167:111–21. https://doi.org/10.1016/j.cell. 2016.09.004 PMID: 27662085 82. Figueroa-Bossi N, Schwartz A, Guillemardet B, D’Heygere F, Bossi L, Boudvillain M. RNA remodeling by bacterial global regulator CsrA promotes Rho-dependent transcription termination. Genes Dev. 2014; 28:1239–51. https://doi.org/10.1101/gad.240192.114 PMID: 24888591 83. Sevostyanova A, Groisman EA. An RNA motif advances transcription by preventing Rho-dependent termination. Proc Natl Acad Sci U S A. 2015; 112:E6835–43. https://doi.org/10.1073/pnas. 1515383112 PMID: 26630006 84. Bossi L, Schwartz A, Guillemardet B, Boudvillain M, Figueroa-Bossi N. A role for Rho-dependent polarity in gene regulation by a noncoding small RNA. Genes Dev. 2012; 26:1864–73. https://doi.org/ 10.1101/gad.195412.112 PMID: 22895254 85. Bastet L, Chauvier A, Singh N, Lussier A, Lamontagne AM, Prevost K, et al. Translational control and Rho-dependent transcription termination are intimately linked in riboswitch regulation. Nucleic Acids Res. 2017; 45:7474–86. https://doi.org/10.1093/nar/gkx434 PMID: 28520932 86. Leela JK, Syeda AH, Anupama K, Gowrishankar J. Rho-dependent transcription termination is essen- tial to prevent excessive genome-wide R-loops in Escherichia coli. Proc Natl Acad Sci U S A. 2013; 110:258–63. https://doi.org/10.1073/pnas.1213123110 PMID: 23251031 87. D’Heygere F, Rabhi M, Boudvillain M. Phyletic distribution and conservation of the bacterial transcrip- tion termination factor Rho. Microbiology. 2013; 159:1423–36. https://doi.org/10.1099/mic.0.067462- 0 PMID: 23704790 88. Nowatzke WL, Richardson JP. Characterization of an unusual Rho factor from the high G + C gram- positive bacterium Micrococcus luteus. J Biol Chem. 1996; 271:742–7. https://doi.org/10.1074/jbc. 271.2.742 PMID: 8557681 89. Kalarickal NC, Ranjan A, Kalyani BS, Wal M, Sen R. A bacterial transcription terminator with inefficient molecular motor action but with a robust transcription termination function. J Mol Biol. 2010; 858 395:966–82. https://doi.org/10.1016/j.jmb.2009.12.022 PMID: 20026069 90. Yuan AH, Hochschild A. A bacterial global regulator forms a prion. Science. 2017; 355:198–201. https://doi.org/10.1126/science.aai7776 PMID: 28082594 91. Bidnenko V, Nicolas P, Grylak-Mielnicka A, Delumeau O, Auger S, Aucouturier A, et al. Termination factor Rho: From the control of pervasive transcription to cell fate determination in Bacillus subtilis. PLoS Genet. 2017; 13:e1006909. https://doi.org/10.1371/journal.pgen.1006909 PMID: 28723971 92. Nagel A, Michalik S, Debarbouille M, Hertlein T, Gesell Salazar M, Rath H, et al. Inhibition of Rho Activ- ity Increases Expression of SaeRS-Dependent Virulence Factor Genes in Staphylococcus aureus, Showing a Link between Transcription Termination, Antibiotic Action, and Virulence. mBio. 2018; 9: e01332–18. https://doi.org/10.1128/mBio.01332-18 PMID: 30228237 93. Dutta D, Shatalin K, Epshtein V, Gottesman ME, Nudler E. Linking RNA polymerase backtracking to genome instability in E. coli. Cell. 2011; 146:533–43. https://doi.org/10.1016/j.cell.2011.07.034 PMID: 21854980 94. Washburn RS, Gottesman ME. Transcription termination maintains chromosome integrity. Proc 870 Natl Acad Sci U S A. 2011; 108:792–7. 95. Peters JM, Mooney RA, Grass JA, Jessen ED, Tran F, Landick R. Rho and NusG suppress pervasive antisense transcription in Escherichia coli. Genes Dev. 2012; 26:2621–33. https://doi.org/10.1101/ gad.196741.112 PMID: 23207917 96. Bouillaut L, McBride SM, Sorg JA. Genetic Manipulation of Clostridium difficile. Curr Protoc Microbiol Unit 9A. 2:1–17. 97. Sievers F, Higgins DG. Clustal Omega, accurate alignment of very large numbers of sequences. Meth- ods Mol Biol. 2014; 1079:105–16. https://doi.org/10.1007/978-1-62703-646-7_6 PMID: 24170397 98. Ghose C, Eugenis I, Sun X, Edwards AN, McBride SM, Pride DT, et al. Immunogenicity and protective efficacy of recombinant Clostridium difficile flagellar protein FliC. Emerg Microbes Infect. 2016; 5:e8. https://doi.org/10.1038/emi.2016.8 PMID: 26839147 99. Ransom EM, Weiss DS, Ellermeier CD. Use of mCherryOpt Fluorescent Protein in Clostridium difficile. Methods Mol Biol. 2016; 1476:53–67. https://doi.org/10.1007/978-1-4939-6361-4_5 PMID: 27507333 100. Ransom EM, Ellermeier CD, Weiss DS. Use of mCherry Red fluorescent protein for studies of protein localization and gene expression in Clostridium difficile. Appl Environ Microbiol. 2015; 81(5):1652–60. https://doi.org/10.1128/AEM.03446-14 PMID: 25527559 101. Kirk JA, Fagan RP. Heat shock increases conjugation efficiency in Clostridium difficile. Anaerobe. 2016; 42:1–5. https://doi.org/10.1016/j.anaerobe.2016.06.009 PMID: 27377776 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 27 / 28 PLOS PATHOGENSRho factor mediated phase variation 102. Ben-Yehuda S, Rudner DZ, Losick R. RacA, a bacterial protein that anchors chromosomes to the cell poles. Science. 2003; 299:532–6. https://doi.org/10.1126/science.1079914 PMID: 12493822 103. Putnam EE, Nock AM, Lawley TD, Shen A. SpoIVA and SipL are Clostridium difficile spore morphoge- netic proteins. J Bacteriol. 2013; 195:1214–25. https://doi.org/10.1128/JB.02181-12 PMID: 23292781 104. Edwards AN, McBride SM. Isolating and Purifying Clostridium difficile Spores. Methods Mol Biol. 2016; 1476:117–28. https://doi.org/10.1007/978-1-4939-6361-4_9 PMID: 27507337 105. Chen X, Katchar K, Goldsmith JD, Nanthakumar N, Cheknis A, Gerding DN, et al. A mouse model of Clostridium difficile-associated disease. Gastroenterology. 2008; 135:1984–92. https://doi.org/10. 1053/j.gastro.2008.09.002 PMID: 18848941 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1008708 August 12, 2020 28 / 28 PLOS PATHOGENS
10.1371_journal.pone.0285980	RESEARCH ARTICLE A qualitative study exploring the lived experiences of patients living with mild, moderate and severe frailty, following hip fracture surgery and hospitalisation Vanisha PatelID 1☯, Antje Lindenmeyer2☯, Fang Gao3☯, Joyce YeungID 4☯ a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Patel V, Lindenmeyer A, Gao F, Yeung J (2023) A qualitative study exploring the lived experiences of patients living with mild, moderate and severe frailty, following hip fracture surgery and hospitalisation. PLoS ONE 18(5): e0285980. https://doi.org/10.1371/journal.pone.0285980 Editor: Rafael Van den Bergh, World Health Organization, BELGIUM Received: December 22, 2022 Accepted: May 7, 2023 Published: May 18, 2023 Copyright: © 2023 Patel et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting information files. Funding: Yes JY is funded by the National Institute for Health Research Post-Doctoral Fellowship (PDF-2014-07-061). FG is a National Institute for Health Research Senior Investigator. The views expressed are those of the authors and not necessarily those of the National Institute for Health Research or the Department of Health and Social Care. The funders had no role in study 1 Department of Anaesthetics, University Hospital Birmingham NHS Foundation Trust, Birmingham, United Kingdom, 2 Institute of Clinical Sciences, University of Birmingham, Birmingham, United Kingdom, 3 Birmingham Acute Care Research Centre, Institute of Inflammation and Ageing, University of Birmingham Research Laboratories, Birmingham, United Kingdom, 4 Warwick Medical School, Warwick Clinical Trials Unit, University of Warwick, Warwick, United Kingdom ☯ These authors contributed equally to this work. vanisha.patel@nhs.net Abstract It is well recognised that hip fracture surgery is associated with a negative impact on short and long-term post-operative physical health and emotional well-being for patients. Further- more, these patients are known to be frail with multiple co-morbidities. This study explores how frailty shapes the lived experiences of rehabilitation and recovery for patients who have undergone hip fracture surgery. Semi-structured interviews were conducted with sixteen participants, recently discharged from hospital following hip fracture surgery. Interpretative phenomenological analysis was applied to explore the lived experiences of frail patients and ascertain important themes. Patient experiences were captured in seven overarching themes: 1) the hospital as a place of “safety”, 2) placing trust in others, 3) the slow recovery journey impeded by attitude and support, 4) maintaining autonomy and dignity whilst feeling vulnerable, 5) seeking a new normal, 6) loneliness and social isolation and 7) the ageing body. Based on our study findings, we have been able to suggest a number of opportunities to improve support for frailer patients in finding a new routine to their everyday lives, these include on-going physical and psychological support, information and education and a robust pathway for transition of care into the community. A conceptual thematic diagram is presented which helps to understand the experience and the complex needs of frail older people undergoing hip fracture surgery. Introduction The ageing population presents serious challenges to our health and social care system. There are now 11.4 million people aged 65 or over in the UK and for the general surgical population prevalence estimates range between 10% and 37% for frailty [1, 2]. In the UK, approximately PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 1 / 15 PLOS ONEdesign, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: NO authors have competing interests. The lived experiences of patients living with frailty following hip fracture surgery 70 000 to 75 000 hip fractures occur each year with the majority of these occurring in the older person [3]. Clinical outcomes such as mortality and length of stay have long been a measure- ment of success of patient care for clinicians [4]. Clinical outcomes and patient experience are invariably linked; positive patient experiences have been found to be associated with better clinical outcomes [5, 6]. The NHS Outcomes Framework has reported a role for positive patient experience, the need to measure care as perceived by patients, and the need for the healthcare system to respond and act on such feedback. Thus, they are advocating information gathering about the lived healthcare experiences of patients [7]. Older patients have complex needs, and age alone does not simply define important patient centred care needs. Involving patients themselves in the identification of their care needs allows patient experience to become an important component in the evaluation of quality of healthcare, ensuring patients focused care and overall improving their healthcare journey, by making patients feel more supported and cared for. Frailty has been defined as a state of increased risk, a distinctive health state related to the ageing process in which multiple body systems gradually lose their in-built reserves resulting in patients having more health deficits [8, 9]. It is a condition characterised by loss of biological reserve, failure of physiological mechanisms and vulnerability to a range of adverse outcomes including increased risk of morbidity, mortality and loss of independence in the perioperative period [10]. Two main theoretical constructs exist in the assessment of frailty, Fried et al’s phe- notypic model and the deficit model described by Rockwood and Mitnitski [11, 12]. The for- mer defines five components, exhaustion, weight loss, weak grip strength, slow walking speed and low physical activity. Multi-system dysregulation is considered if three or more of these components are observed. which in the presence of three suggest multi-system dysregulation. The latter quantifies accumulated deficits; it assesses the number of health problems that a patient has accumulated over their lifetime. Frailty assessment brings together information about health deficits and their impact on the patients’ ability to think and do as they please; look after themselves; interact with other people; and move about without falling [8]. Hip fracture patients are well known to be frail with multiple co-morbidities; high pre-oper- ative frailty scores are associated with increased length of hospital stay, 30- and 90-day mortal- ity and likelihood of institutionalisation [13–15]. Moreover, frailty has also been recognised as a predictor of post-operative complications and poor functional outcomes, as well as a risk fac- tor for prolonged hospital stay, institutionalisation and worsening disability, resulting in a vicious cycle [13, 16–18]. In practice, recognising frailty can indicate the need for a holistic approach to treating older patients with complex needs across health and social care [19]. Qualitative studies evaluating the experiences of frail patients have focused on assessing those with chronic medical conditions or undergoing elective surgery [20, 21]. Patients dis- cussed fears included being ignored or feeling imprisoned due to loss of physical capability, cognitive decline leading to dementia and nursing home admission [22]. In addition, patients recognise the physical symptoms of lack of strength, weight loss and risk of falls associated with this chronic disease state but also add that there is a psychological component to being frail in addition to having physical symptoms [23]. Social interaction plays a greater role in maintaining quality of life for frail patients compared to the non-frail [24]. For many frail patients, the greatest priority is to maintain independence, with their well-being centred on their ability to complete everyday tasks [25]. Frail patients undergoing hip fracture surgery are a heterogenous group with respect to co- morbidities, functional status and social support and therefore it is likely that their rehabilita- tion and recovery experience will vary [26]. Previous studies exploring patient experiences fol- lowing a hip fracture have outlined the multifaceted experience of recovery. Most studies have focused on the period immediately after injury; including how patients make sense of the PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 2 / 15 PLOS ONEThe lived experiences of patients living with frailty following hip fracture surgery acute injury and hospital experience [27–29], communication and information provision [30, 31], interactions with others on the ward and management of acute pain [32, 33]. Other studies have explored more contextual factors such as patients’ experience of the discharge process and transition of care [30, 34], challenges patients experience to rehabilitation [28, 30, 32, 35], as well as the importance of support from family and friends [27, 30, 34, 35]. Long term reha- bilitation and recovery are of particular concern to patients as they return home with the aim to minimise loss of independence [36, 37]. To date, qualitative studies have not focused on frailty as a factor that can impact on the lived experiences of patients following hip fracture surgery. Therefore, the aim of our study was to explore the lived experiences of patients living with mild, moderate and severe frailty following hip fracture surgery, with a focus specifically on frail patients’ perceptions of their rehabilitation and recovery experience. Methods Ethics The study received approval from the West Midlands Research Ethics Committee. (REC refer- ence number: 16/WM0165) We conducted and reported this study in compliance with the consolidated criteria for reporting qualitative research (COREQ) [38]. Design In order to capture patients’ experiences of living with varying degrees of frailty, we conducted a qualitative study based on an interpretative phenomenological approach. Interpretative phe- nomenology was the chosen methodology as it explores in detail how people make sense of their personal and social worlds [39]. Interpretative phenomenology analysis (IPA) has gained prominence in health and social sciences as a way to understand and interpret topics which are complex and potentially emotional, such as illness experiences [40]. By applying this approach, we aimed to identify, explore and describe the lived experiences of patients follow- ing hip fracture surgery, with an emphasis on an individual’s personal perception. IPA focuses on small and homogeneous samples, with participants purposively selected because they have experiences of the phenomena studied. Each participant gives an in-depth reflective narrative of their own experiences from their own perspective. The participants interpret their own experiences and the researcher uses their own interpretation to come to an understanding of the participants experience, thus a two- stage interpretation process was involved known as ‘double hermeneutics’. Frailty assessment We chose to use a deficit model for frailty. The Clinical Frailty Scale (CFS) is a tool combining clinical assessment with objective evaluation of specific domains including comorbidity, func- tion, and cognition to generate a frailty score ranging from 1 (very fit) to 9 (terminally ill). For this study, patients were categorised into four groups 1) Fit (CFS score of 1–3), 2) Mild frailty (CFS score: 4–5), 2) Moderately frail (CFS score: 6) and 3) Severe frailty (CFS score: 7–9) [8, 41]. Frailty status was measured, at interview, 8–12 weeks post hip fracture surgery by the lead researcher (VP). Recruitment Purposive sampling [42] was used to recruit potential participants between December 2016 and July 2017 at Birmingham Heartlands Hospital, University Hospital Birmingham NHS Foundation Trust. The sample aimed to obtain a range of patients of varying age, gender, co- PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 3 / 15 PLOS ONEThe lived experiences of patients living with frailty following hip fracture surgery morbidities and baseline frailty status. Participants were eligible to take part in the study if they were aged 65 years and over and underwent hip fracture repair surgery. Patients were excluded if they were unable to consent or unable to partake in the interview due to a speech or language impairment. The researcher (VP), a female clinical doctor, screened the clinical admission database at Heartlands Hospital to identify potential participants. Patients were recruited, face to face, within 72 hours of admissions; the researcher (VP) introduced the study and provided a written information leaflet. Written consent was obtained. Out of thirty eligible participants, seventeen provided informed consent however one patient declined to be inter- viewed thereafter. Based on IPA methodology, a sample size of between 5 and 10 in-depth interviews is enough to discover the nuances and complexities of people’s lived experiences [43]. Data collection The primary researcher (VP) interviewed each participant. An interview guide was planned and created with input from stakeholders including a patient representative (S1 File). This ensured questions were relevant, patient focused and worded in an easy to understand man- ner. Initial questions were open-ended exploring how they broke their hip, their hospital expe- rience with prompts related to involvement and support for family/carers, relationship with healthcare professionals and dignity/respect. Further questioning explored their experiences and feelings around discharge, sources of support, any changes to relationships with partners, families and health care providers and their rehabilitation and recovery goals. Interviews were conducted by one researcher (VP) to ensure consistency of interview questions throughout the study. At the time of interview, frailty was evaluated using the CFS. Semi-structured interviews were conducted eight to twelve weeks following discharge at a time and place convenient to patients. In the majority of cases, patients chose to be interviewed following their hospital follow-up appointment. Participants were given a choice of being interviewed alone or accompanied by a relative/friend. Interviews were recorded and tran- scribed per verbatim for analysis. Field notes were also made during the interviews by VP. Data analysis Analysis was carried out using the steps outlined by Smith et al. for IPA [39]. VP carried out the initial analysis, listening to the transcripts as well as reading and re-reading. Descriptive codes were noted beside the text, leading to identification of the preliminary themes. With the aim of finding connections between these, VP and JY reflected and interpreted these prelimi- nary themes. Similar thematic concepts were grouped together into a cluster leading to devel- opment of a hierarchical system with the initial theme being the one most significant to the participant. Once all transcripts had been analysed, variations between accounts of patients who were living with fit, mild, moderately and severely frail (based on the CFS) were investi- gated using cross-case analysis resulting in the creation of a thematic matrix by two researchers (VP & JY). Development of themes was a circular process, with a focus on suspension of the researcher’s own pre-conceived ideas and judgements surrounding the text by employing indi- vidual knowledge and sensitivity surrounding the subject, to obtain a ‘clear’ view of the phe- nomena, a process otherwise known as ‘bracketing’ [44]. Re-reading of each transcript, review of assimilated themes for each group and further discussion by the research team (VP, JY, AL, FG) generated seven superordinate themes that captured experiences of the participants. In particular, different attitudes and motivators to rehabilitation emerged as an important con- cern and therefore they were explored further in our analysis, highlighting the similar and PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 4 / 15 PLOS ONEThe lived experiences of patients living with frailty following hip fracture surgery contrasting views between patients with differing frailty status. Themes were developed induc- tively and were not considered to be pre-existing entities lying within the data. The cumulative research team had extensive knowledge about peri-operative care sur- rounding hip fracture surgery together with qualitative research experience, ensuring rigour and reflexivity. Initial analysis involved noting beside the text of each transcript, codes were applied and thereafter categorised into themes using NVivo software version 11 (QSR, Burlington, Massa- chusetts, USA), this enabled teamworking, audit trail and manageability of data. In the last phase of analysis, a conceptual framework was developed indicating a number of resources and interventions useful to aid with resuming ADLs and improving their recovery journey. Results Sixteen participants between the ages of sixty-five and eighty-eight were interviewed; 11/16 participants were female (Table 1). Participants had a range of co-morbidities including hyper- tension, chronic obstructive pulmonary disease, atrial fibrillation, diabetes mellitus and osteo- arthritis. Eight interviews were carried out with the participant alone, the rest of the participants were interviewed with a family member or partner. Ten interviews were carried out at Birmingham Heartlands Hospital after a clinic follow-up appointment, the rest at the participant’s home. The mean interview time was thirty-two minutes (range 11 to 83 minutes). As outlined above, we analysed variations between patients according to their clinical assess- ment by the Clinical Frailty Scale. Three patients were evaluated to be fit (1–3), five patients were living with mild frailty (4–5), five patients with moderate frailty (6) and three patients with severe frailty (7+) (Table 1). Table 1. Patient demographics. Gender Age (years) Patient 1 Patient 2 Patient 3 Patient 4 Patient 5 Patient 6 Patient 7 Patient 8 Patient 9 Patient 10 Patient 11 Patient 12 Patient 13 Patient 14 Patient 15 Patient 16 F M M F F F F F F F F F M F M M 77 78 88 76 68 88 71 65 80 88 72 67 68 87 84 69 Clinical Frailty Scale post-surgerya 6 Discharge destination (Living alone)b Home 3 6 5 4 7 6 5 4 8 6 6 5 5 5 7 Partner’s house Home Home (alone) Home (alone) Intermediate carec Intermediate carec Home Home Home Home (alone) Home Friends’ house Home (alone) Home (alone) Home (alone) a Clinical Frailty Scale–Fit = 1–3, Living with mild frailty = 4–5 living with moderate frailty = 6, living with severe frailty = 7+. b For patients discharged home and living alone if applicable. c Intermediate Care is a form of respite care that supports someone to remain in their own home while they recover from an illness, accident or hospital stay https://doi.org/10.1371/journal.pone.0285980.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 5 / 15 PLOS ONEThe lived experiences of patients living with frailty following hip fracture surgery The following themes were identified as being important to frail patients when evaluating their lived experiences in relation to hospital care, rehabilitation and recovery, following hip fracture surgery. A description of each theme is attached with representative patient quotes. The hospital as a place of “safety” The patients living with mild frailty felt the hospital was a ‘safe’ environment which allowed for “recovery to begin”. However, they became frustrated with the limitations of the ward envi- ronment for example unfamiliarity with the ward layout and the noise and lights contributing to a poor sleep environment. The daily routine of waiting for assistance in washing and visiting the bathroom, receiving pain relief, problems and disturbances from other patients around them limited their recovery. They felt ready to go home and fully recover in their own environ- ment as described by Patient 11 “Well I’ve never been in hospital in my life, and it’s a bit old, 80 and you’ve never been in your life. It was just strange, I couldn’t wait to get out, that’s why I told them to do it.” However, frailer patients viewed the hospital as a ‘safe’ environment due to close access to bathrooms, support of healthcare professionals during the day and night, phys- iotherapy support for on-going rehabilitation and presence of other patients providing social support. For them, discharge from hospital was associated with limited support and worry about how they would cope with washing, dressing, cooking and shopping. Patient 12 described her hospital experience: “I don’t know I think I felt quite safe in the hospital, I felt really safe you know well I can’t fall over and I can’t do this, I can’t do that and you know I’m okay here but I mean you can’t stay there forever just because you feel safe can you. I felt like I’d just been thrown out and I felt like that this time.” Moderately frail patients who lived alone sought practical as well as emotional support to overcome their low confidence and fear of fall- ing again when they got home compared to those that lived with a partner or family members. Patients with ongoing chronic disease required help in the community; transition of care was viewed to be poor especially by frailer patients who felt that continuity of care was frag- mented and transfer of information, medication and support systems were unsatisfactory. There was limited information on the type of surgery and post-operative medical complica- tions and treatment that they had received, medication changes, and no link between the hos- pital and their GP. Patient and family involvement in discharge planning was ad-hoc and once in the community, identifying sources of help could be challenging. Some patients felt aban- doned (“left to my own devices” (Patient 2) with no help in organising follow up or review appointments with a medical professional. Discharge planning for Patient 14 involved “two ladies who deal with this came and they asked these questions and filled in their clipboard thing and, erm, oh you’ll be all right, you’ll be all right. You’ll have the half an hour in the morning, half an hour at night. I said yes and twenty-three hours all on my own.” Patient 15 said that he “would have liked to have had somebody come along and assess. . . I don’t know whether I’m pro- gressing well or not. . .. . . the physios came just a couple of times to assess me, erm, to see that I was doing it properly, you know. But I have no constant physio care at all.” These accounts show how participants’ level of frailty influenced how they perceived the hospital environment and their confidence in returning home. Placing trust in others Healthcare professionals and family/friends were seen as a source of comfort and support for all patients. Patient 5 described the healthcare professionals on the ward, “They just made you feel at ease really, I suppose”. They were able to provide knowledge and information, this pro- vided reassurance and therefore patients felt they were able to trust them during this time of uncertainty in an unfamiliar environment. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 6 / 15 PLOS ONEThe lived experiences of patients living with frailty following hip fracture surgery Healthcare professionals were encouraging with rehabilitation, “when I left hospital I was quite confident. I mean, we’d had a good ward. We’d had good staff and everything but when I went home, once the girls (her daughters) had gone back to work. . .. . .I’d lost all that and slowly your confidence goes because you’re not getting the contact.” Physiotherapists provided motiva- tion and built confidence resulting in a positive attitude to rehabilitation, “well, the physios are very encouraging and they say oh but you’re doing very well, you’re doing very well.” (Patient 14) Patient 11 described her attitude to rehabilitation having had some physiotherapy sessions in hospital, “I think it’s down to me. I think it’s down to me really, having the confidence to do it. And not be frightened.” The information given by healthcare professionals was often not questioned by frailer patients as they felt that they did not want to cause any trouble and saw them as “experts” who “knew what they were doing”. However, due to varying physiotherapy information and input, experiences of rehabilitation differed between patients regardless of frailty status. The slow journey of recovery Fitter patients focused on the motivators to recovery which allowed them to engage in rehabili- tation. Patient 2 stated in relation to her recovery, “I do sometimes think to myself, “It’s time we went somewhere, so we’ll go on the bus. I suppose that’s about it really.” Knowledge, support from family/friends and previous experiences shaped their views on their ability to return home. Looking towards the future, they focused on being able to return to their previous phys- ical state. A positive attitude to recovery with a pro-active approach was a prominent theme for success in their rehabilitation goals and recovery. For example, Patient 8 wanted to “Just get on with it.” The recovery journey was viewed as being slower than anticipated for patients now living with moderate frailty and they were frustrated with this. Prior to their injury they had felt independent, however sustaining a hip fracture and undergoing surgery had resulted in restriction to mobility and activities of daily living. Frustrations were directed at being unable to drive, pick up grandchildren from school, go to the shops and return to work. Patient 1, stated “I used to do the school run for the little ones, but nobody’s shown me how to get in and out of a car”. Patients felt that limited access to physiotherapy within the community restricted their progress with rehabilitation as expressed by Patient 7 who wanted to start using a stick, “No because I had physio in Ann Marie’s [rehabilitation centre] and they said that they was going to come here, but I’m just walking around with this [zimmer frame].” As outlined above, physiotherapists were seen to provide practical advice on exercises and improving mobility as well as reassurance. The majority of these patients were self-motivated and under- stood that they would have to persevere with rehabilitation to improve their own recovery. Maintaining autonomy and dignity whilst feeling vulnerable For patients living with mild frailty, sustaining a hip fracture and having to undergo surgery generated a new state of vulnerability, especially for those that were previously fitter. Patients who were now living with moderate or severe frailty acknowledged some, previous, limited baseline physical health state and independence. For them, restricted mobility after surgery resulted in requiring help with self-care and personal hygiene; a new and challenging experi- ence which made them feel vulnerable. Patients in this situation still wanted to have choices and be able to decide what was right for themselves. Using the bedpan or needing help getting to the bathroom on the ward was embarrassing. Patient 14 described her experience, “I said oh I want to go to the toilet. Argh and they said, well we’re very sorry but we can’t do anything [laughter]. So you had to hold it? No, I just had to do it. . .. Oh it was awful. I couldn’t believe it”. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 7 / 15 PLOS ONEThe lived experiences of patients living with frailty following hip fracture surgery These accounts show that patients living with mild and moderate frailty continued to want to maintain dignity and remain autonomous. Patient 10 had experienced restriction to her activities of daily living prior to sustaining a hip fracture, she explained, “X had done it all by the time I’d got home, so I had no trouble of that. But I didn’t need any advice, really, because I’d done it all before.” In comparison to those living with mild or moderate frailty, those living with severe frailty were familiar with the need to receive help and trust others. Seeking a new normal Most patients living with mild and moderate frailty said that they had been relatively indepen- dent and had experienced good health prior to their injury; chronic conditions were well con- trolled, they had limited physical frailty and were independent in caring for themselves. The impact of the injury had threatened this and therefore many found they wanted to return to their prior state of “normality”. Patient 13 stated, “Oh yes even though I’m sixty-eight I’m still fairly mobile, I haven’t been able to go out as much.” Patient 11 described her fears of falling “that’s the word, frightened. I still am. I don’t want to go back into hospital. Simple as that. I’ve tried hard. Some days I don’t want to do it [physiotherapy exercises] but I still do it.” This was also echoed by Patient 1, “I’d lost all that and slowly your confidence goes because you’re not get- ting the contact. I found that devastating because I couldn’t do what I wanted to do.” Motivation to continue rehabilitation once discharged from hospital was driven by the patients wanting to “return to normality”, which included resuming their daily routine, restor- ing their previous physical and emotional well-being. Patient 8 lost confidence after her fall, “It took a knock, I didn’t want to go out at first, then I thought ‘I’ve got to get out’”. Patient 1 described previous daily activities which gave them enjoyment “I like my gardening. I do need retail therapy. I do need to get out. There’s a limit with what I can do in the garden at the moment anyway”. Patient 11 was dissatisfied with what she was able to physically do “Now, that’s difficult. Because yes, I’m pleased I’m home, but no, I’m not satisfied because I can’t do what I want to do”. The support of a partner/spouse was paramount in being able to return to normal as they provided support, encouragement and motivation. However, the ‘new normal’ which encompassed walking aids, need for care-givers, assistance in washing and preparing meals was recognised and patients were concerned that they might not be able to return to pre- vious physical health, daily routine and independence. Loneliness and social isolation Physical and psychological impairment of sustaining a hip fracture influenced the patients’ ability to socialise and meet family and friends. Patient 5 explained “I’d like to meet some other older people, I suppose, like coffee morning and things like that. When you’re on your own all day every day. . . I mean, I do go out obviously because I get bored, but then I can’t walk far so it’s a viscous circle. For patients now living with moderate and severe frailty, the ageing body, low mood and loss of support from family and friends contributed to their feeling of loneliness. Patient 7 stated “I’m on my own and things keep passing through my head.” Negative emotions such also impacted on their desire to socialise with others or allow people to visit. For frailer patients, experiences of loneliness and social isolation also changed their attitude to recovery, with some adopting a “why bother?” outlook on rehabilitation. Patient 10 described feeling “Helpless sometimes, yes, not being able to do what I want to do. Yeah. . ..Oh, definitely. Yeah. I used to be out every day.” For example, Patient 12 was teary during her interview and stated, “Yes, yes sometimes I can just sit and have a good old weep for nothing really, well I think it’s nothing but I don’t know.” Patient 7 explained that “my brother and my sister in law died with PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 8 / 15 PLOS ONEThe lived experiences of patients living with frailty following hip fracture surgery cancer, so I’ve got nobody (to help).” Patients described seeking comfort and reassurance from loved ones and healthcare professionals in hospital as well as in the community. The ageing body Sustaining a hip fracture was seen as a sign of ageing and failure of their own body; the limita- tions in mobility, physical co-morbidities as well as psychological co-morbidities, contributed to views of “an ageing body”. Patients now living with frailty found that their physical frailty, worsened by limited mobility, impacted on their ability to engage with activities of daily living and the social world. Patient 7 described worsening leg ulcers as impacting her recovery and contributing to her limited mobility “They’re all broken, they’re absolutely soaked. (in relation to chronic leg ulcers) I told them not to put these socks on because I had them on before and they leaked, and I said ‘don’t put them on’, because I couldn’t put my slippers on, I can’t get used to wearing them.” The impact of existing chronic disease on rehabilitation and recovery was sig- nificant for patients now living with moderate and severe frailty. They described how sustain- ing a hip fracture worsened their disease or led to additional difficulty in managing it. For example, exacerbations of cardiac and respiratory diseases led to prolonged hospital stays and readmission to hospital in some cases. Patients with previous lower limb problems such as knee arthritis or peripheral vascular disease felt that these conditions had become worse, which made engaging with rehabilitation difficult. Worsening of chronic disease for Patient 3 was seen as a result of ageing, a natural process that was inevitable. He described “after all this upheaval of your body, I think it’s silly to take it on. . .. I suppose so. I mean, I’m 88 so I can’t go on forever. . .. No, got to be sensible about it”. The short and long-term changes to frail partici- pants’ daily activities put a strain on relationships; caregivers had to become involved in help- ing to manage their health problems when previously they had little input. In some cases, this resulted in the need for long-term adaptations to lifestyle e.g. no longer being able to visit friends and family. Delirium, affected one’s perception of their own health resulting in a nega- tive psychological impact on one’s body. Patient 3 commented: “Yes, because I’d wake up in the night and I didn’t really know where I was. . . well, I knew where I was but when I looked around everything was quiet and still. I looked up at the railing where the curtain goes and I could see lit- tle horses galloping. . .. A bit silly because I didn’t tell anybody about that.” Psychological factors impacted Patient 7, “I’m on my own and things keep passing through my head.” The psychologi- cal factors impeded engagement with rehabilitation and set them back on their recovery journey. Our conceptual framework (Fig 1) represents an overview of the lived experiences of the frail older person following hip fracture surgery. Discussion and implications This study has highlighted the life-changing impact that hip fracture surgery has on the physi- cal and psychological well-being of frail patients. The findings reported in this study are consistent with other studies that illustrate patient recovery and rehabilitation post-hip fracture [29, 45–54]. What our study adds is that levels of frailty, influence patients’ lived experiences of recovery from surgery. Furthermore, based on our study findings, we have been able to suggest a number of opportunities to improve support for frailer patients in finding a new routine to their everyday lives (Fig 1). For frailer patients, placing trust in others was crucial in providing psychological support and comfort which was invaluable in engaging in rehabilitation during a time of uncertainty. Our study found that information and effective communication are central to providing good support and relieving anxiety. It allows patients and their care-givers to participate in shared PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 9 / 15 PLOS ONEThe lived experiences of patients living with frailty following hip fracture surgery Fig 1. Conceptual thematic diagram of the lived experiences of the frail older person, focusing on hospital care, rehabilitation and recovery, for patients undergoing hip fracture surgery. (ADLs—activities of daily living). https://doi.org/10.1371/journal.pone.0285980.g001 decision-making and understand the hip fracture ‘pathway’; importantly this makes them feel empowered as well as helping them to understand what to expect in the future and therefore have more realistic expectations and goals for recovery. The current evidence base supports good information provision about their diagnosis, treatment and on-going management, mak- ing patients feel comfortable in trusting healthcare professionals and improving overall well- being [27, 30]. Following a hip fracture, the restrictions in activities of daily living experienced by patients lead to the disruption of ‘normal’ life. Similar to findings described by St-Cyr Tribble et al, we have highlighted that, patients ‘seek a new normal’ during their recovery journey [51]. Our participants primarily described factors such as encouragement and support and familiarity of their own surroundings e.g. hospital layout, returning home, in helping to boost motivation and influence their own recovery whilst finding their ‘new normal’. Maintaining normality in everyday life requires psychological, social and physical resources, giving patients the ability to look after themselves, as supported by Claassens et al. [47]. The more resources that are avail- able, the more a positive outlook is adopted by the patient [55]. In frailer patients, acceptability of interdependence should be promoted as well as positive ways of living with frailty. For the frailer patient, fears of interdependence can be reduced through anticipatory care planning and support from communities will promote a sense of belonging, allay social alienation and improve self-worth [56]. In addition, we found that ensuring a smooth transition pathway between the ‘safe’ hospital environment and less controlled environment such as the patient’s home or a rehabilitation unit/nursing home can make patients feel less isolated while ameliorating feelings of vulnera- bility. Other studies have found that a well-organised transition with adequate support and fol- low-up can improve quality of life [57]. Some participants felt that discharge plans were PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 10 / 15 PLOS ONEThe lived experiences of patients living with frailty following hip fracture surgery inadequate; therefore, early discharge planning, written information about in-patient diagno- sis, management and changes to medication, dosette boxes for prescriptions and information regarding community services and follow-up can help in supporting an integrated and co- ordinated care plan to address frailer patients’ individual needs [58]. Furthermore, our study suggests that for frail patients support also includes receiving ade- quate information and advice for example regarding mobility. There is good evidence that continuity of care can help prevent readmissions, medication errors and improve patient safety resulting in an overall positive patient experience for frailer patients [53]. Sustaining a hip fracture impacted on participants emotional well-being, therefore patients identified as vulnerable may benefit from review and follow-up with a psychologist, this has been shown to help reinforce being aware of ones’ own strengths and limitations and increas- ing their self-esteem with a reduction in negative feelings [51]. In this study, patients living with moderate and severe frailty, echoed the findings by Taube et al, with barriers to overcoming loneliness including their ageing body, fear of falling and loss of relatives. Additionally, we found that the psychological impact of sustaining a hip frac- ture together with being frail decreased their desire to socialise with people, in turn having a deleterious effect on mood and further likelihood of social isolation, potentially worsening frailty [59]. As a result, there is a need for social services input for patients and their carers and access to community social networks to reduce loneliness and isolation [60, 61]. This study has provided in-depth personal accounts on the lived experience of hip fracture surgery. Alongside patient safety and clinical effectiveness, patient experience completes the three pillars of quality of care within the NHS [62]. Therefore it is necessary to integrate patient experience into quality improvement and health policy development in order to improve qual- ity of care. Strengths and limitations Our findings provide valuable insight into the needs of frailer patients following hip fracture surgery which is important in guiding healthcare policies. The use of IPA enabled an in-depth thematic exploration of the lived experiences of our participants. Furthermore, semi-struc- tured interviews allowed for in-depth discussion and collection of ‘rich data’. Purposive sam- pling allowed for frail and non-frail patients to be included and their experiences to be compared. This is a single-centre study. The hospital that the participants were recruited from serves a diverse multi-cultural population, however only one patient was from an ethnic minority background and the experiences of this group of patients may be different. Whilst we are aware that up to one third of patients experience on-going cognitive dysfunction following hip fracture surgery [63], they were not included in our sample due to the difficulties this would present in conducting in-depth interviews. We acknowledge that the lived experience for these patients and their care-givers would be different, and capturing their views would be a valuable addition to the findings of this study. Interview length was led by the patients, one interview lasted 11 minutes, however the patient felt that this was adequate time for them. Frailty status, was measured post-operatively and therefore some patients may have had on-going ‘deficits’ contributing to their frailty status. However, with limited evidence that frailty is modifiable in this group of patients, measurement of frailty two months post-operatively is likely to reflect a continuing frail state. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 11 / 15 PLOS ONEThe lived experiences of patients living with frailty following hip fracture surgery Conclusion Our findings highlight the impact of different levels of frailty on patients’ experiences during their healthcare journey and therefore it is important to recognise a ‘one-size fits all’ patient pathway will not suffice. Opportunities to improve support for frailer patients highlighted by this study include the need for on-going support for rehabilitation, information and education to shape realistic expectations, a robust pathway for transition of care into the community with access to medical and allied health professionals as well as the recognition and importance of emotional well-being with patients being able to access psychological assessment. Supporting information S1 File. Interview question guide. (DOCX) Acknowledgments We would firstly like to thank all the participants of this study for their time and time and will- ingness to share their views and experiences at this difficult time. The authors would also like to thank Teresa Melody, Anne Deverill and members of the Clinical Ambassador Research Group. Author Contributions Conceptualization: Vanisha Patel, Antje Lindenmeyer, Fang Gao, Joyce Yeung. Data curation: Vanisha Patel, Antje Lindenmeyer, Fang Gao, Joyce Yeung. Formal analysis: Vanisha Patel, Antje Lindenmeyer, Fang Gao, Joyce Yeung. Investigation: Vanisha Patel. Methodology: Vanisha Patel, Antje Lindenmeyer, Fang Gao, Joyce Yeung. Project administration: Vanisha Patel. Software: Vanisha Patel. Supervision: Antje Lindenmeyer, Fang Gao, Joyce Yeung. Writing – original draft: Vanisha Patel, Antje Lindenmeyer, Fang Gao, Joyce Yeung. Writing – review & editing: Vanisha Patel, Antje Lindenmeyer, Fang Gao, Joyce Yeung. References 1. Office for National Statistics. National Population Projections: 2014-based Statistical Bulletin. 2015. http://www.ons.gov.uk/peoplepopulationandcommunity/populationandmigration/populationprojections/ bulletins/nationalpopulationprojections/2015-10-29 2. Hewitt J, Moug SJ, Middleton M, Chakrabarti M, Stechman MJ, Mccarthy K. Prevalence of frailty and its association with mortality in general surgery. The American Journal of Surgery. 2015; 209: 254–259. https://doi.org/10.1016/j.amjsurg.2014.05.022 PMID: 25173599 3. Royal College of Physicians. National Hip Fracture Database annual report 2019. London: RCP; 2019. 4. Miller TE, Mythen M. Successful recovery after major surgery: moving beyond length of stay. Periopera- tive medicine. 2014; 3. https://doi.org/10.1186/2047-0525-3-4 PMID: 25018877 5. Black N, Varaganum M, Hutchings A. Relationship between patient reported experience (PREMs) and patient reported outcomes (PROMs) in elective surgery. BMJ quality & safety. 2014; 1–9. https://doi. org/10.1136/bmjqs-2013-002707 PMID: 24508681 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 12 / 15 PLOS ONEThe lived experiences of patients living with frailty following hip fracture surgery 6. Carter J, Ward C, Wexler D, Donelan K. The association between patient experience factors and likeli- hood of 30-day readmission: a prospective cohort study. BMJ Qual Saf. 2018; 27: 683–690. https://doi. org/10.1136/bmjqs-2017-007184 PMID: 29146680 7. Department of Health. The NHS Outcomes Framework 2020/21. London huss; 2021. 8. Rockwood K, Theou O. Using the Clinical Frailty Scale in Allocating Scarce Health Care Resources. Can Geriatr J. 2020; 23: 210–215. https://doi.org/10.5770/cgj.23.463 PMID: 32904824 9. Turner G, Clegg A. Best practice guidelines for the management of frailty: a British Geriatrics Society, Age UK and Royal College of General Practitioners report. Age and Ageing. 2014; 43: 744–747. https:// doi.org/10.1093/ageing/afu138 PMID: 25336440 10. Centre for Perioperative Care & British Geriatric Society. Guideline for perioperative care for people liv- ing with frailty underging elective and emergency surgery. London: CPOC; 2021. https://cpoc.org.uk/ guidelines-resources-guidelines/perioperative-care-people-living-frailty 11. Fried LP, Tangen CM, Walston J, Newman AB, Hirsch C, Gottdiener J, et al. Frailty in Older Adults: Evi- dence for a Phenotype. Journal of Gerontology: MEDICAL SCIENCES America. 2001; 56A: M146– M157. https://doi.org/10.1093/gerona/56.3.m146 PMID: 11253156 12. Rockwood K, Mitnitski A. Frailty Defined by Deficit Accumulation and Geriatric Medicine Defined by Frailty. Clinics in Geriatric Medicine. 2011; 27: 17–26. https://doi.org/10.1016/j.cger.2010.08.008 PMID: 21093719 13. Krishnan M, Beck S, Havelock W, Eeles E, Hubbard RE, Johansen A. Predicting outcome after hip frac- ture: Using a frailty index to integrate comprehensive geriatric assessment results. Age and Ageing. 2014; 43: 122–126. https://doi.org/10.1093/ageing/aft084 PMID: 23832264 14. Chen C-L, Chen C-M, Wang C-Y, Ko P-W, Chen C-H, Hsieh C-P, et al. Frailty is Associated with an Increased Risk of Major Adverse Outcomes in Elderly Patients Following Surgical Treatment of Hip Fracture. Sci Rep. 2019; 9: 19135. https://doi.org/10.1038/s41598-019-55459-2 PMID: 31836751 15. Jorissen RN, Lang C, Visvanathan R, Crotty M, Inacio MC. The effect of frailty on outcomes of surgically treated hip fractures in older people. Bone. 2020; 136: 115327. https://doi.org/10.1016/j.bone.2020. 115327 PMID: 32209422 16. Robinson TN, Wallace JI, Wu DS, Wiktor A, Pointer L, Pfister SM, et al. Accumulated frailty characteris- tics predict postoperative discharge institutionalization in the geriatric patient. Journal of the American College of Surgeons. 2011; 213: 37–42. https://doi.org/10.1016/j.jamcollsurg.2011.01.056 PMID: 21435921 17. Kistler EA, Nicholas JA, Kates SL, Friedman SM. Frailty and Short-Term Outcomes in Patients With Hip Fracture. Geriatr Orthop Surg Rehabil. 2015; 6: 209–214. https://doi.org/10.1177/2151458515591170 PMID: 26328238 18. Xu BY, Yan S, Low LL, Vasanwala FF, Low SG. Predictors of poor functional outcomes and mortality in patients with hip fracture: a systematic review. BMC Musculoskeletal Disorders. 2019; 20: 568. https:// doi.org/10.1186/s12891-019-2950-0 PMID: 31775693 19. Manthorpe J, Iliffe S. Frailty–from bedside to buzzword? Journal of Integrated Care. 2015; 23: 120–128. https://doi.org/10.1108/JICA-01-2015-0007 20. 21. Lin H-S, McBride RL, Hubbard RE. Frailty and anesthesia—risks during and post-surgery. Local Reg Anesth. 2018; 11: 61–73. https://doi.org/10.2147/LRA.S142996 PMID: 30323657 van de Ree CLP, Landers MJF, Kruithof N, de Munter L, Slaets JPJ, Gosens T, et al. Effect of frailty on quality of life in elderly patients after hip fracture: a longitudinal study. BMJ Open. 2019; 9: e025941. https://doi.org/10.1136/bmjopen-2018-025941 PMID: 31324679 22. Kendall M, Carduff E, Lloyd A, Kimbell B, Cavers D, Buckingham S, et al. Different Experiences and Goals in Different Advanced Diseases: Comparing Serial Interviews With Patients With Cancer, Organ Failure, or Frailty and Their Family and Professional Carers. J Pain Symptom Manage. 2015; 50: 216– 224. https://doi.org/10.1016/j.jpainsymman.2015.02.017 PMID: 25828558 23. Schoenborn NL, Van Pilsum Rasmussen SE, Xue Q-L, Walston JD, McAdams-Demarco MA, Segev DL, et al. Older adults’ perceptions and informational needs regarding frailty. BMC Geriatr. 2018; 18: 46. https://doi.org/10.1186/s12877-018-0741-3 PMID: 29433426 24. Puts MTE, Shekary N, Widdershoven G, Heldens J, Lips P, Deeg DJH. What does quality of life mean to older frail and non-frail community-dwelling adults in the Netherlands? Qual Life Res. 2007; 16: 263– 277. https://doi.org/10.1007/s11136-006-9121-0 PMID: 17033894 25. Thinks B. Frailty: Language and Perceptions A report prepared by BritainThinks on behalf of Age UK and the British Geriatrics Society. 2015. http://www.ageuk.org.uk/Documents/EN-GB/For- professionals/Policy/health-and-wellbeing/report_bgs_frailty_language_and_perceptions.pdf?dtrk= true PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 13 / 15 PLOS ONEThe lived experiences of patients living with frailty following hip fracture surgery 26. ASAP. National Hip Fracture Database Anaesthesia Sprint Audit of Practice (ASAP) Falls and Fragility Fracture Audit Programme. 2014. https://www.aagbi.org/sites/default/files/NHFD anaesthestic report. pdf 27. Tutton E, Seers K, Langstaff D. Professional nursing culture on a trauma unit: experiences of patients and staff. J Adv Nurs. 2008; 61: 145–153. https://doi.org/10.1111/j.1365-2648.2007.04471.x PMID: 18186906 28. Chou P-L, Chien C-S, Lee B-O, Bai Y-L, Hung C-C. Early recovery experience of patients with injury in Taiwan. J Nurs Scholarsh. 2014; 46: 245–252. https://doi.org/10.1111/jnu.12084 PMID: 24754699 29. Griffiths F, Mason V, Boardman F, Dennick K, Haywood K, Achten J, et al. Evaluating recovery following hip fracture: a qualitative interview study of what is important to patients. BMJ open. 2015; 5: e005406– e005406. https://doi.org/10.1136/bmjopen-2014-005406 PMID: 25564138 30. Atwal A, Tattersall K, Murphy S, Davenport N, Craik C, Caldwell K, et al. Older adults experiences of rehabilitation in acute health care. Scandinavian Journal of Caring Sciences. 2007; 21: 371–378. https://doi.org/10.1111/j.1471-6712.2007.00469.x PMID: 17727550 31. Clukey L, Hayes J, Merrill A, Curtis D. “Helping them understand”: nurses’ caring behaviors as per- ceived by family members of trauma patients. J Trauma Nurs. 2009; 16: 73–81. https://doi.org/10.1097/ JTN.0b013e3181ac91ce PMID: 19543015 32. Zalon ML. Pain in frail, elderly women after surgery. Image J Nurs Sch. 1997; 29: 21–26. https://doi.org/ 10.1111/j.1547-5069.1997.tb01135.x PMID: 9127536 33. Eriksson K, Wikstro¨m L, Fridlund B, Årestedt K, Brostro¨ m A. Patients’ experiences and actions when describing pain after surgery—a critical incident technique analysis. Int J Nurs Stud. 2016; 56: 27–36. https://doi.org/10.1016/j.ijnurstu.2015.12.008 PMID: 26772655 34. Giosa JL, Stolee P, Dupuis SL, Mock SE, Santi SM. An examination of family caregiver experiences during care transitions of older adults. Can J Aging. 2014; 33: 137–153. https://doi.org/10.1017/ S0714980814000026 PMID: 24754978 35. Colleen W, Pryor J, Jeeawody B. The concerns of older women during inpatient rehabilitation after frac- tured neck of femur. International Journal of Therapy and Rehabilitation. 2009; 16: 261–270. https://doi. org/10.12968/ijtr.2009.16.5.42103 36. Aberg AC, Sidenvall B, Hepworth M, O’Reilly K, Lithell H. Continuity of the self in later life: perceptions of informal caregivers. Qual Health Res. 2004; 14: 792–815. https://doi.org/10.1177/ 1049732304265854 PMID: 15200801 37. 38. Toscan J, Systems H, Manderson B, Systems H, Santi SM, Systems H, et al. “Just another fish in the pond”: the transitional care experience of a hip fracture patient. 2013; 13. Tong A, Sainsbury P, Craig J. Consolidated criteria for reporting qualitative research (COREQ): a 32- item checklist for interviews and focus groups. Int J Qual Health Care. 2007; 19: 349–357. https://doi. org/10.1093/intqhc/mzm042 PMID: 17872937 39. Smith JA, Flower P, Larkin M. Interpretative phenomenological analysis: Theory, method, Research. London: Sage Publications, Ltd.; 2009. 40. Tuffour I. A Critical Overview of Interpretative Phenomenological Analysis: A Contemporary Qualitative Research Approach. J Healthc Commun. 2017; 2. https://doi.org/10.4172/2472-1654.100093 41. Rockwood K, Song X, Macknight C, Bergman H, Hogan DB, Mcdowell I, et al. A global clinical measure of fitness and frailty in elderly people. CMAJ. 2005; 173: 489–495. https://doi.org/10.1503/cmaj.050051 PMID: 16129869 42. Moriarty J. Qualitative Methods Overview. London: School for Social Care Research, National Institue for Health Research; 2011. 43. Hefferon K, Gil-Rodriguez E. Interpretative phenomenological analysis. The Psychologist. 2011; 24: 756–759. 44. Husserl E. The idea of phenomenology. Dordrecht: Kluwer Academic Publishers; 1999. 45. Archibald G. Patients’ experiences of hip fracture. Journal of Advanced Nursing. 2003; 44: 385–392. https://doi.org/10.1046/j.0309-2402.2003.02817.x PMID: 14651710 46. Gorman E, Chudyk AM, Hoppmann CA, Hanson HM, Guy P, Sims-Gould J, et al. Exploring older adults’ patterns and perceptions of exercise after hip fracture. Physiother Can. 2013; 65: 86–93. https://doi.org/ 10.3138/ptc.2012-01BH PMID: 24381388 47. Claassens L, Widdershoven GA, Van Rhijn SC, Van Nes F, Broese van Groenou MI, Deeg DJH, et al. Perceived control in health care: a conceptual model based on experiences of frail older adults. J Aging Stud. 2014; 31: 159–170. https://doi.org/10.1016/j.jaging.2014.09.008 PMID: 25456633 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 14 / 15 PLOS ONEThe lived experiences of patients living with frailty following hip fracture surgery 48. Ziden L, Wenestam C-G, Hansson-Scherman M. A life-breaking event: early experiences of the conse- quences of a hip fracture for elderly people. Clinical Rehabilitation. 2008; 22: 801–811. https://doi.org/ 10.1177/0269215508090204 PMID: 18728133 49. Bruun-Olsen V, Bergland A, Heiberg KE. “I struggle to count my blessings”: recovery after hip fracture from the patients’ perspective. BMC Geriatr. 2018; 18: 18. https://doi.org/10.1186/s12877-018-0716-4 PMID: 29351770 50. Pol M, Peek S, van Nes F, van Hartingsveldt M, Buurman B, Kro¨ se B. Everyday life after a hip fracture: what community-living older adults perceive as most beneficial for their recovery. Age and Ageing. 2019; 48: 440–447. https://doi.org/10.1093/ageing/afz012 PMID: 30806451 51. St-Cyr Tribble D, Gallagher F, Bell L, Caron C, Godbout P, Leblanc J, et al. Empowerment interven- tions, knowledge translation and exchange: perspectives of home care professionals, clients and care- givers. BMC Health Serv Res. 2008; 8: 177–177. https://doi.org/10.1186/1472-6963-8-177 PMID: 18713474 52. Olsson L, AEM N, Karlsson J, Ekman I. Admitted with a hip fracture: patient perceptions of rehabilita- tion. Journal of Clinical Nursing. 2007; 16: 853–859 7p. https://doi.org/10.1111/j.1365-2702.2006. 01635.x PMID: 17462035 53. 54. Laugaland K, Aase K, Barach P. Interventions to improve patient safety in transitional care—a review of the evidence. Work. 2012; 41 Suppl 1: 2915–2924. https://doi.org/10.3233/WOR-2012-0544-2915 PMID: 22317162 Taube E, Jakobsson U, Midlo¨ v P, Kristensson J. Being in a Bubble: the experience of loneliness among frail older people. J Adv Nurs. 2016; 72: 631–640. https://doi.org/10.1111/jan.12853 PMID: 26568280 55. Andreasen J, Lund H, Aadahl M, Sørensen EE. The experience of daily life of acutely admitted frail elderly patients one week after discharge from the hospital. Int J Qual Stud Health Well-being. 2015; 10: 27370. https://doi.org/10.3402/qhw.v10.27370 PMID: 26037333 56. Lloyd A, Kendall M, Starr JM, Murray SA. Physical, social, psychological and existential trajectories of loss and adaptation towards the end of life for older people living with frailty: a serial interview study. BMC Geriatr. 2016; 16: 176. https://doi.org/10.1186/s12877-016-0350-y PMID: 27765011 57. Olsson L-E, Karlsson J, Ekman I. The integrated care pathway reduced the number of hospital days by half: a prospective comparative study of patients with acute hip fracture. J Orthop Surg Res. 2006; 1: 3. https://doi.org/10.1186/1749-799X-1-3 PMID: 17150123 58. Dhesi J. Setting up a proactive service to make surgery safer for older people. 2013. 59. Davies K, Maharani A, Chandola T, Todd C, Pendleton N. The longitudinal relationship between loneli- ness, social isolation, and frailty in older adults in England: a prospective analysis. The Lancet Healthy Longevity. 2021; 2: e70–e77. https://doi.org/10.1016/S2666-7568(20)30038-6 PMID: 36098160 60. Fillit H, Butler RN. The frailty identity crisis. J Am Geriatr Soc. 2009; 57: 348–352. https://doi.org/10. 1111/j.1532-5415.2008.02104.x PMID: 19207150 61. Hoogendijk EO, Suanet B, Dent E, Deeg DJH, Aartsen MJ. Adverse effects of frailty on social function- ing in older adults: Results from the Longitudinal Aging Study Amsterdam. Maturitas. 2016; 83: 45–50. https://doi.org/10.1016/j.maturitas.2015.09.002 PMID: 26428078 62. Department of Health. High Quality Care For All. NHS Next Stage Review Final Report. London: The Stationey Office; 2008. 63. Oughli HA, Chen G, Philip Miller J, Nicol G, Butters MA, Avidan M, et al. Cognitive Improvement in Older Adults in the Year After Hip Fracture: Implications for Brain Resilience in Advanced Aging. Am J Geriatr Psychiatry. 2018; 26: 1119–1127. https://doi.org/10.1016/j.jagp.2018.07.001 PMID: 30454790 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285980 May 18, 2023 15 / 15 PLOS ONE
10.3201_eid2605.191638	Possible Transmission Mechanisms of Mixed Mycobacterium tuberculosis Infection in High HIV Prevalence Country, Botswana Yeonsoo Baik,1 Chawangwa Modongo, Patrick K. Moonan, Eleanor S. Click, James L. Tobias, Rosanna Boyd, Alyssa Finlay, John E. Oeltmann, Sanghyuk S. Shin,2 Nicola M. Zetola2 Tuberculosis caused by concurrent infection with multiple Mycobacterium tuberculosis strains (i.e., mixed infection) challenges clinical and epidemiologic paradigms. We ex- plored possible transmission mechanisms of mixed infec- tion in a population-based, molecular epidemiology study in Botswana during 2012–2016. We defined mixed infec- tion as multiple repeats of alleles at >2 loci within a discrete mycobacterial interspersed repetitive unit–variable-num- ber tandem-repeat (MIRU-VNTR) result. We compared mixed infection MIRU-VNTR results with all study MIRU- VNTR results by considering all permutations at each multiple allele locus; matched MIRU-VNTR results were considered evidence of recently acquired strains and non- matched to any other results were considered evidence of remotely acquired strains. Among 2,051 patients, 34 (1.7%) had mixed infection, of which 23 (68%) had recent- ly and remotely acquired strains. This finding might sup- port the mixed infection mechanism of recent transmission and simultaneous remote reactivation. Further exploration is needed to determine proportions of transmission mech- anisms in settings where mixed infections are prevalent. Tuberculosis (TB) caused by concurrent infection with multiple strains of Mycobacterium tuberculo- sis during 1 episode is commonly referred to as mixed infection. In 1972, Canetti et al. suggested the concept of mixed infection of exogenous reinfection of nonpri- mary TB among elderly patients in France (1). Their Author affiliations: University of California, Los Angeles, Los Angeles, California, USA (Y. Baik); Botswana–Upenn Partnership, Gaborone, Botswana (C. Modongo, N.M. Zetola); US Centers for Disease Control and Prevention, Atlanta, Georgia, USA (P.K. Moonan, E.S. Click, J.L. Tobias, R. Boyd, A. Finlay, J.E. Oeltmann); US Centers for Disease Control and Prevention, Gaborone (R. Boyd, A. Finlay); University of California, Irvine, Irvine, California, USA (S.S Shin) DOI: https://doi.org/10.3201/eid2605.191638 observation was followed by phage typing of cultured isolates from patients with concurrent disease in mul- tiple organ sites observed during clinical practice in North America, mixed cultures among Eskimo pa- tients during the mid-1970s (2,3), and cultures collected during outbreak investigations in the 1980s and 1990s (4,5). However, more recent applications of advanced molecular tools suggest mixed infection might occur more frequently than initially expected (6,7). This pos- sibility led to many research studies of mixed infec- tion, which found that mixed infection is associated with poor treatment outcomes (6,8), including acqui- sition of multidrug-resistant TB (7,8). Mixed infection research contributed to the discovery that exogenous reinfection was responsible for a substantial portion of incident TB, implying incomplete protection from a primary infection in subsequent infections (9,10). Despite the clinical importance of mixed infection, its potential leading mechanisms of transmission have not been examined using empirical data. Infections caused by multiple M. tuberculosis strains can occur af- ter simultaneous transmission of multiple strains dur- ing a single transmission episode (i.e., the index patient transmits multiple strains) or by sequential infections of >2 strains acquired at different times, resulting in superinfection (10). So far, transmission mechanisms of mixed infection and its population-level effect have been explored only hypothetically (9,11). Research on the transmission mechanisms for mixed infection with empirical data might improve understanding of M. tuberculosis dynamics and designing effective TB con- trol interventions (12). Our objective was to explore possible transmission mechanisms leading to mixed 1Current affiliation: Johns Hopkins University, Baltimore, Maryland, USA. 2These senior authors contributed equally to this article. Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 5, May 2020 953 RESEARCH M. tuberculosis infections by comparing genotypes and spatial proximity of all detected M. tuberculosis strains. Methods Study Setting This analysis was part of a population-based, molec- ular epidemiology study in Botswana (the Kopanyo Study). The study design and methods were previ- ously described (13). In brief, the study recruited and enrolled patients with newly diagnosed TB at 30 TB and HIV clinics during 2012–2016. Behavioral, clini- cal, and demographic information (including resi- dential address at enrolment) were collected during medical record abstraction and standardized patient interview. Sputum collected from participants under- went smear-microscopy, culture, drug-susceptibility testing, and 24-locus mycobacterial interspersed re- petitive unit–variable-number tandem-repeat (MI- RU-VNTR) genotyping using a standard internation- al protocol (14), when applicable. Definition of Mixed Infection MIRU-VNTR genotyping counts the numbers of tan- dem repeats at the selected loci, which are unique in different strains of M. tuberculosis. We defined mixed infection as multiple allele repeat numbers (e.g., dou- ble allele) at >2 loci within a discrete MIRU-VNTR re- sult (10). We defined possible mixed infection as mul- tiple allele repeat numbers at 1 locus within a discrete MIRU-VNTR result and single infection as a discrete MIRU-VNTR result with single alleles at all 24 loci (Figure 1; Appendix Tables 1, 2, https://wwwnc.cdc. gov/EID/article/26/5/19-1638-App1.xlsx). Definition of Genotype Cluster We defined TB genotype clusters as >2 patient iso- lates with exact match 24-loci results, suggesting re- cently acquired strains (12,15). We considered geno- type results that matched no other patient isolate results in the dataset nonclustered, suggesting re- motely acquired strains (12,15). To identify putative mixed infection M. tuberculosis genotype clusters, we compared MIRU-VNTR results for each mixed infec- tion patient to MIRU-VNTR results of all other M. tuberculosis strains, considering permutations of each repeat number at multiple allele loci. We also con- sidered 24-loci results to be nonclustered if no per- mutation of the mixed MIRU-VNTR result matched any other study strain; if >1 permutation of the mixed Figure 1. Mixed-strain infection MIRU-VNTR permutations and genotype cluster/noncluster examples of Mycobacterium tuberculosis (the Kopanyo Study), Botswana, 2012–2016. On the basis of mixed-strain MIRU-VNTR patterns, all possible permutations at each of multiple allele loci were considered. The MIRU-VNTR result of each strain in a possible permutation set was compared with that of all strains identified in the study. Assuming numbers of tandem repeats at other 19 loci are identical, 4 genomes (strains A–D) in the genotype cluster example (bottom left) have matched tandem repeats at the presented 5 loci of tandem repeats in the mixed- strain infection. Strains E and F in the genotype noncluster example have nonmatched tandem repeats at the second and third locus, respectively. MIRU-VNTR, mycobacterial interspersed repetitive unit–variable-number tandem-repeat. 954 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 5, May 2020 Mixed M. tuberculosis Infection, Botswana MIRU-VNTR result matched any other study strain, we considered it to be clustered. When we considered all permutations at each double allele locus, if >1 per- mutation accounting for each repeat number at each locus matched another study strain but no permuta- tion accounting for the alternate repeat number at each locus matched another study strain, we consid- ered it to be evidence of simultaneously clustered and nonclustered strains. For example, if the patient iso- late results had repeat numbers 4 and 5 at the third lo- cus, the matched M. tuberculosis strain’s MIRU-VNTR results should include repeat numbers 4, 5, or both at the same locus (Figure 1). We excluded patients with isolates that had missing or incomplete MIRU-VNTR results. We reviewed all laboratory procedures (i.e., sputum collection and processing, culture isolation and storage, DNA abstraction and storage, and MI- RU-VNTR batching processes) to identify potential points of cross-contamination or mishandling. We reviewed all laboratory registries and electronic data- bases to record processing and reporting dates for all patient isolates. Classification of Mixed Infection Mechanisms On the basis of the genotype cluster analysis, we classified patients with mixed infection into 1 of 3 categories: 1) simultaneous reactivation of >2 re- motely acquired strains if no mixed infection MIRU- VNTR permutations accounting for multiple differ- ent repeat numbers at each locus matched any other study strain; 2) infection from a recently acquired strain and simultaneous reactivation of a remotely acquired strain if >1 permutation accounting for 1 repeat number at each locus matched another study strain but no permutation accounting for the other repeat number at each locus matched any other study strain; and 3) rapid progression of >2 recently ac- quired strains if >1 permutation accounting for each repeat number at each double allele locus matched another study strain. Statistical Analyses For each mixed infection MIRU-VNTR result, we wrote a loop function using SAS (SAS Institute Inc., https://www.sas.com) to compare tandem numbers from the first locus to 24th locus with all other MIRU- VNTR results locus by locus. When discrepancies ex- isted between tandem numbers at a locus, the locus was flagged. We counted the number of flags after 24 loci were compared. If all numbers matched, the number of flags was 0; if no loci matched, the number was 24. We used the number of flags to classify the degree to which the MIRU-VNTR pattern matched that of the mixed infection MIRU-VNTR result. At the end of the loop function, we created a subset dataset with all MIRU-VNTR results by descending order of the number of exactly matched loci (from 0 for exactly matched at all 24 loci), for each mixed infection MI- RU-VNTR result. We calculated simple frequencies and propor- tions for the main outcomes (mixed infection, pos- sible mixed infection, and single infection) stratified by patient sex, HIV status, and residential address. Primary residential address of each patient was geocoded and mapped using ArcGIS (ESRI, https://www.esri.com). We showed the distribution of M. tuberculosis genotype clusters if found within 1 km of one another to add epidemiologic plausibil- ity. We excluded patients with missing residential geocoding from the spatial analysis. Sensitivity Analysis To assess potential variation within genotype related- ness, we explored an alternative clustering definition to include 1 locus difference. For this sensitivity anal- ysis, potential near matches (i.e., matched on all other loci results but with a nonmatched tandem number at the locus of interest) were considered genotype clusters. We excluded patients with isolates with missing or incomplete MIRU-VNTR results from the sensitivity analysis. Ethics Approval This study was approved by the Institutional Review Boards of the US Centers for Disease Control and Pre- vention (#6291; Atlanta, GA, USA); Health Research and Development Committee, Botswana Ministry of Health and Wellness (Gaborone, Botswana); Univer- sity of Pennsylvania (Philadelphia, PA, USA); and University of California, Irvine (Irvine, CA, USA). Participants provided written informed consent. Results A total of 2,137 patients were enrolled, of whom 1,130 (53%) were HIV positive (Table 1). After excluding patients with missing or incomplete MIRU-VNTR results (including 3 patients with mixed infection), we included 2,051 patients in the analyses (Figure 2). A total of 862 discrete genotyping MIRU-VNTR re- sults were obtained (a more detailed strain analysis is available elsewhere [15]). We detected no evidence of laboratory cross-contamination events within spu- tum processing, culturing, DNA abstraction, or geno- typing processing. All mixed infection patient isolates were processed on different days from isolates from other purported patients in the cluster. Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 5, May 2020 955 RESEARCH Table 1. Characteristics of persons in a study to assess mixed-strain transmission of Mycobacterium tuberculosis (the Kopanyo Study), Botswana, 2012–2016 HIV status Negative 948 (44) 32.5 (16) Unknown 59 (3) 33.5 (14) 558 (49) 572 (51) 584 (62) 364 (38) Positive 1,130 (53) 36.9 (10) Characteristic Total, no. (%), N = 2,137 Age, y, mean ( SD) Sex, no. (%) M F Primary residential site, no. (%) Gaborone Ghanzi District Other Botswana, not in study region Missing residential address Previous tuberculosis history, no. (%) Yes No Infection status, no. (%)* Mixed Possible mixed Single Different strains, no. (%)† *After excluding 86 patients with mycobacterial interspersed repetitive unit–variable-number tandem-repeat results with missing alleles. †Total number of different mycobacterial interspersed repetitive unit–variable-number tandem-repeat results = 862. 17 (2) 50 (5) 1,008 (93) 570 (66) 15 (2) 35 (4) 869 (94) 453 (53) 565 (60) 215 (23) 108 (11) 60 (6) 828 (73) 109 (10) 116 (10) 77 (7) 150 (16) 798 (84) 227 (20) 903 (80) 43 (73) 16 (27) 43 (73) 8 (14) 0 8 (14) 11 (19) 48 (81) 2 (4) 3 (5) 52 (91) 50 (6) Thirty-four (2%) patients had mixed infection, and 88 (4%) patients had possible mixed infection. Overall, we classified mixed infection in 23 (68%) patients as infection from a recently acquired strain and simultaneous reactivation of a remotely ac- quired strain, 7 (21%) as simultaneous reactivation of >2 remotely acquired strains, and 4 (12%) as >2 re- cently acquired strains (Table 2). Mixed infection in 27 (79%) patients involved recently acquired strains (Appendix Tables 1–3). The MIRU-VNTR results of 34 patients with mixed infection had a median of 7.5 loci (interquartile range 3–11) of multiple tandem re- peats. The most prevalent MIRU-VNTR result in the population, MIRU identification no. [ID] 644 (n = 147 isolates), was not included in any genotype clusters with potential mixed infection transmission events. The second most prevalent strain, MIRU ID 382 (n = 81), matched with 2 genotype clusters involving Figure 2. Flowchart of population-based, molecular epidemiology study (the Kopanyo Study) of mixed Mycobacterium tuberculosis strains, Botswana, 2012–2016. 956 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 5, May 2020 Mixed M. tuberculosis Infection, Botswana patients with mixed infection (MIRU ID 838 and MIRU ID 970) (Appendix Table 1). After excluding additional 137 patients with no residential address (including 3 patients with mixed infection and 6 with possible mixed infection), we explored spatiotemporal transmission among 1,914 patients (Figure 2). We found 4 genotype clusters of mixed infection within 1 km of the location of patient with mixed infection as the center: 3 in Gaborone (Figure 3) and 1 in Ghanzi (Figure 4). In sensitivity analysis, we allowed MIRU-VNTR patterns to differ by 1 locus, which changed the trans- mission category for 7 patients with mixed infection. Our main finding that the highest proportion (19 [51%]) of mixed infection occurred through a combi- nation of genotype clustered and nonclustered strains did not change. The second highest proportion (10 [27%]) of mixed infection was a combination of mul- tiple genotype clustered strains. No. (%) Table 2. Characteristics of 34 patients with mixed-strain Mycobacterium tuberculosis infection (the Kopanyo Study), Botswana, 2012–2016 Characteristic Primary residential site Gaborone Ghanzi District Other Botswana Missing residential address HIV infection status Positive Negative Unknown Transmission mechanism Recently acquired + recently acquired Recently acquired + remotely acquired Remotely acquired + remotely acquired HIV infection & transmission mechanism Recently acquired + recently acquired 22 (65) 7 (21) 2 (6) 3 (9) 18 (53) 14 (41) 2 (6) 4 (12) 23 (68) 7 (21) Recently acquired + remotely acquired Remotely acquired + remotely acquired Positive: 1 (6); negative: 2 (14) Positive: 12 (66); negative: 10 (72) Positive: 5 (28); negative: 2 (14) Discussion We describe genotype patterns consistent with hy- pothesized mixed infection transmission mechanisms, using a multiyear, population-based TB cohort. In our study, most patients with mixed infection (68%) had both recently and remotely acquired strains, suggest- ing recent transmission and simultaneous remote re- activation. Recent infection that progresses to disease might further compromise the immune system, lead- ing to reactivation. A previous case study described a patient with mixed infection with an apparent trig- gering of a remote multidrug-resistant M. tuberculosis strain after recent exposure to a drug-sensitive strain (16). A similar phenomenon has been described for relapse of Plasmodium vivax malaria triggered by in- fection with P. falciparum (17). Figure 3. Potential spatial relationships (residence within 1 km of another patient) between patients with mixed-strain infection and with other genotype-clustered strains, Gaborone, Botswana, 2012–2016. Shown are location of patients with mixed Mycobacterium tuberculosis infection and other genotype-clustered cases in Gaborone. Each color represents each genotype cluster. The 1-km radius blue-shaded area from each mixed infection patient shows the neighborhood boundary. Three patients with mixed infection had potential spatial relationships with 3–6 other patients within the neighborhood. Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 5, May 2020 957 RESEARCH Figure 4. Potential spatial relationships (residence within 1 km of another patient) between mixed infection and other genotype-clustered cases, Ghanzi, Botswana, 2012–2016. Shown are locations of patients with mixed Mycobacterium tuberculosis infection and other genotype-clustered cases. Each color represents each genotype cluster. The 1-km radius blue-shaded area from each mixed infection patient shows the neighborhood boundary. Two patients with mixed infection were genotype- clustered and had a potential spatial relationship. (Their mycobacterial interspersed repetitive unit–variable-number tandem-repeat results were not exactly matched.) Similarly, our findings suggest that most mixed infection transmission events included reactivation of remotely acquired strains triggered by recently acquired strains, implying that mixed infection may be affected by the force of infection in communities (10,11). We estimated the prevalence of each discrete MIRU-VNTR result as a proxy measure of force of in- fection in our study population. Contrary to our ex- pectation, the 2 most prevalent strains (MIRU IDs 644 and 382) appeared in only 1 mixed infection transmis- sion event. The dominate strain in the mixed infection was MIRU ID 838, which appeared 3 times. Further studies can show whether less transmissible strains outcompete other strains within the host to establish long-term persistence (11). Our results add to the complexity of TB trans- mission dynamics in high TB prevalence settings (7). Current TB prevention strategies primarily focus on interrupting recent TB transmission through early detection and treatment of sputum smear–positive patients (18). Although interventions to interrupt transmission can reduce opportunities of exoge- nous re-infection and hence reduce the prevalence of mixed infection (10), our findings also imply the importance of treating latent TB infection to reduce the risk for mixed infection (19). No statistical as- sociation between HIV status—a proxy for reduced latency—and mixed infection (data not shown [odds ratio 1.15 (95% CI 0.79–1.68)]) also further supports the influence of remotely acquired strains in poly- clonal transmission events. Our study alone might not be sufficient to generalize the results and em- phasize reactivation. However, we envision further exploration of our suggested 3 transmission mecha- nisms in a setting where the transmission intensity is expected to be higher (e.g., high population density or dense slum area) and the role of reactivation is accounted for accordingly. We added the spatial information to provide epidemiologic evidence of possible M. tuberculosis transmission. If patients whose isolates are in the same genotype cluster are spatially close to each 958 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 5, May 2020 Mixed M. tuberculosis Infection, Botswana other (i.e., within 1 km), they might be more like- ly to be in a transmission network than otherwise. This interpretation may be limited as we accounted only for the patients’ residential address as the spa- tial information, the close proximity set as 1 km was arbitrary, and the few TB clusters (and number of patients therein) may be missed if mixed infection is not included in transmission network reconstruc- tions. However, our finding reconfirmed that TB transmission was ongoing in the community. A com- prehensive molecular characterization of within- host M. tuberculosis diversity, as well as an attempt to temporally identify the primary source or index of transmission by comparing diagnosis times and the times of symptom onset (20), might be needed to fully capture TB transmission chains and accurately infer TB transmission (21,22). Our results should be interpreted with caution. The prevalence of mixed infection was lower than in other studies (9,23) because of the method of mo- lecular analyses. Although 24-loci MIRU-VNTR is a standardized molecular characterization tool and of- fers simple results that can be readily used to identi- fy mixed infection (7,24), it has limited resolution to distinguish mixed infection from clonal heterogene- ity or within-host bacterial microevolution (9,25,26). Different tools, such as whole-genome sequencing and 2 lineage-specific PCRs, might identify M. tu- berculosis strains more sensitively and lead to a dif- ferent dominating transmission mechanism if more patients with mixed infection were detected (23,27). In the meantime, we defined and analyzed possible mixed infection and mixed infection separately in an attempt to more conservatively differentiate mixed infection from within-host heterogeneity. Another limitation involves misclassification bias from detec- tion sensitivity (28); that is, all potential genotyping matches depend on the sensitivity of the character- ization method. The 24-loci MIRU-VNTR method is relatively sensitive and has high discriminatory power; however, it characterizes only part of the M. tuberculosis genome (7). Hence, we might have missed genetic heterogeneity present in loci not cov- ered by this method (12). The prevalence of each M. tuberculosis strain also depended on the degree to which we captured all M. tuberculosis strains present in the community. Although our study was multi- year and covered a broad geographic area, some im- portant patients in the transmission network could have been missed (e.g., their TB was diagnosed be- fore the study period, they resided in areas not cov- ered by the study, or they refused enrolment), lead- ing to clustering misclassification. We recruited TB patients through both passive and active case find- ing (13) to increase coverage, but not every patient produced sputum, and not all sputum samples led to M. tuberculosis isolation or valid genotype results (15). This limitation might lead to missed transmis- sion links (18,21). Given generally low bacillary load among children, the transmission mechanism would have been affected in a way that the role of reacti- vated strains was reduced if missing sputum sam- ples had been successfully identified. On the other hand, by enabling multiple permutations of possible MIRU-VNTR results for mixed infection and possi- ble mixed infection cases, MIRU-VNTR results with multiple alleles had more possible combinations and higher chance of matching with other genotypes. This finding may imply an imbalanced chance of be- ing a member of a genotype cluster. Future studies to investigate molecular profiles of M. tuberculosis with serial sputum collection, includ- ing nonrespiratory samples, and use of more sensitive and specific genome sequencing technologies, will be of interest to thoroughly assess possible transmission events leading to mixed infection. Despite the lower prevalence of mixed infection in the population in this study, the proposed mixed infection transmission mechanisms can be useful to characterize how simi- lar or different mixed-infection transmission mecha- nisms would be across different settings with differ- ent burden of mixed infection. Acknowledgments We are grateful to the patients with TB and their families for participating in this study. We thank the district health team and recruitment and retention officers who helped us coordinate with and contact patients. We are indebted to the Botswana National Tuberculosis Program and Botswana Ministry of Health for their partnership in this effort. We also thank the following persons for their thoughtful reviews of this manuscript: Benjamin Silk, Thomas Navin, James Posey, Carla Winston, Susan Cookson, Katina Pappas-DeLuca, and William Levine. This study was funded by US National Institutes of Health (grant nos. R01AI097045, K01AI118559) and by the President’s Emergency Plan for AIDS Relief (PEPFAR) through the US Centers for Disease Control and Prevention. About the Author Dr. Baik is a postdoctoral fellow at Johns Hopkins University, Baltimore, Maryland, USA. Her primary research interests include applying epidemiologic methods in TB transmission and implementing public health interventions. Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 5, May 2020 959 RESEARCH References 1. Canetti G, Sutherland I, Svandova E. Endogenous reactivation and exogenous reinfection: their relative importance with regard to the development of non-primary tuberculosis. Bull Int Union Tuberc. 1972;47:116–34. 2. Mankiewicz E, Liivak M. Phage types of Mycobacterium tuberculosis in cultures isolated from Eskimo patients. Am Rev Respir Dis. 1975;111:307–12. 3. Bates JH, Stead WW, Rado TA. Phage type of tubercle bacilli isolated from patients with two or more sites of organ involvement. Am Rev Respir Dis. 1976;114:353–8. 4. Nardell E, McInnis B, Thomas B, Weidhaas S. Exogenous reinfection with tuberculosis in a shelter for the homeless. N Engl J Med. 1986;315:1570–5. https://doi.org/10.1056/ NEJM198612183152502 5. Braden CR, Morlock GP, Woodley CL, Johnson KR, Colombel AC, Cave MD, et al. Simultaneous infection with multiple strains of Mycobacterium tuberculosis. Clin Infect Dis. 2001;33:e42–7. https://doi.org/10.1086/322635 6. van Rie A, Victor TC, Richardson M, Johnson R, van der Spuy GD, Murray EJ, et al. Reinfection and mixed infection cause changing Mycobacterium tuberculosis drug- resistance patterns. Am J Respir Crit Care Med. 2005;172:636– 42. https://doi.org/10.1164/rccm.200503-449OC 7. Tarashi S, Fateh A, Mirsaeidi M, Siadat SD, Vaziri F. Mixed infections in tuberculosis: the missing part in a puzzle. Tuberculosis (Edinb). 2017;107:168–74. https://doi.org/10.1016/j.tube.2017.09.004 8. Shin SS, Modongo C, Baik Y, Allender C, Lemmer D, Colman RE, et al. Mixed Mycobacterium tuberculosis–strain infections are associated with poor treatment outcomes among patients with newly diagnosed tuberculosis, independent of pretreatment heteroresistance. J Infect Dis. 2018;218:1974–82. 9. Cohen T, van Helden PD, Wilson D, Colijn C, McLaughlin MM, Abubakar I, et al. Mixed-strain Mycobacterium tuberculosis infections and the implications for tuberculosis treatment and control. Clin Microbiol Rev. 2012;25:708–19. https://doi.org/10.1128/CMR.00021-12 10. McIvor A, Koornhof H, Kana BD. Relapse, re-infection and mixed infections in tuberculosis disease. Pathog Dis. 2017;75. https://doi.org/10.1093/femspd/ftx020 11. Sergeev R, Colijn C, Cohen T. Models to understand the population-level impact of mixed strain M. tuberculosis infections. J Theor Biol. 2011;280:88–100. https://doi.org/ 10.1016/j.jtbi.2011.04.011 12. Mathema B, Kurepina NE, Bifani PJ, Kreiswirth BN. Molecular epidemiology of tuberculosis: current insights. Clin Microbiol Rev. 2006;19:658–85. https://doi.org/10.1128/ CMR.00061-05 13. Zetola NM, Modongo C, Moonan PK, Click E, Oeltmann JE, Shepherd J, et al. Protocol for a population-based molecular epidemiology study of tuberculosis transmission in a high HIV-burden setting: the Botswana Kopanyo study. BMJ Open. 2016;6:e010046. https://doi.org/10.1136/ bmjopen-2015-010046 14. Supply P, Allix C, Lesjean S, Cardoso-Oelemann M, Rüsch-Gerdes S, Willery E, et al. Proposal for standardization of optimized mycobacterial interspersed repetitive unit– variable-number tandem repeat typing of Mycobacterium tuberculosis. J Clin Microbiol. 2006;44:4498–510. https://doi.org/10.1128/JCM.01392-06 15. Click ES, Finlay A, Oeltmann JE, Basotli J, Modongo C, Boyd R, et al. Phylogenetic diversity of Mycobacterium tuberculosis in two geographically distinct locations in Botswana—the Kopanyo Study. Infect Genet Evol. 2020:104232. 16. Pérez-Lago L, Lirola MM, Navarro Y, Herranz M, Ruiz-Serrano MJ, Bouza E, et al. Co-infection with drug-susceptible and reactivated latent multidrug-resistant Mycobacterium tuberculosis. Emerg Infect Dis. 2015;21:2098– 100. https://doi.org/10.3201/eid2111.150683 17. Ashley EA, Phyo AP, Carrara VI, Tun KM, Nosten F, Smithuis F, et al. Plasmodium vivax relapse rates following Plasmodium falciparum malaria reflect previous transmission intensity. J Infect Dis. 2019;220:100–4. https://doi.org/10.1093/infdis/jiz052 18. Mathema B, Andrews JR, Cohen T, Borgdorff MW, Behr M, Glynn JR, et al. Drivers of tuberculosis transmission. J Infect Dis. 2017;216(suppl_6):S644–53. 19. Behr MA, Edelstein PH, Ramakrishnan L. Revisiting the timetable of tuberculosis. BMJ. 2018;362:k2738. https://doi.org/10.1136/bmj.k2738 20. Xu Y, Cancino-Muñoz I, Torres-Puente M, Villamayor LM, Borrás R, Borrás-Máñez M, et al. High-resolution mapping of tuberculosis transmission: whole genome sequencing and phylogenetic modelling of a cohort from Valencia Region, Spain. PLoS Med. 2019;16:e1002961. https://doi.org/10.1371/journal.pmed.1002961 21. Worby CJ, Lipsitch M, Hanage WP. Within-host bacterial diversity hinders accurate reconstruction of transmission networks from genomic distance data. PLOS Comput Biol. 2014;10:e1003549. https://doi.org/10.1371/ journal.pcbi.1003549 22. Martin MA, Lee RS, Cowley LA, Gardy JL, Hanage WP. Within-host Mycobacterium tuberculosis diversity and its utility for inferences of transmission. Microb Genom. 2018;4(10). 23. Sobkowiak B, Glynn JR, Houben RMGJ, Mallard K, Phelan JE, Guerra-Assunção JA, et al. Identifying mixed Mycobacterium tuberculosis infections from whole genome sequence data. BMC Genomics. 2018;19:613. https://doi.org/10.1186/s12864-018-4988-z 24. Sadegh H, Kargarpour Kamakoli M, Farmanfarmaei G, Masoumi M, Abdolrahimi F, Fateh A, et al. Pros and cons of direct genotyping on tuberculosis clinical samples. Microb Pathog. 2017;103:135–8. https://doi.org/10.1016/ j.micpath.2016.12.025 25. Cohen T, Wilson D, Wallengren K, Samuel EY, Murray M. Mixed-strain Mycobacterium tuberculosis infections among patients dying in a hospital in KwaZulu-Natal, South Africa. J Clin Microbiol. 2011;49:385–8. https://doi.org/10.1128/ JCM.01378-10 26. Streit E, Millet J, Rastogi N. Mycobacterium tuberculosis polyclonal infections and microevolution identified by MIRU-VNTRs in an epidemiological study. Int J Mycobacteriol. 2015;4:222–7. https://doi.org/10.1016/ j.ijmyco.2015.05.005 27. Mallard K, McNerney R, Crampin AC, Houben R, Ndlovu R, Munthali L, et al. Molecular detection of mixed infections of Mycobacterium tuberculosis strains in sputum samples from patients in Karonga District, Malawi. J Clin Microbiol. 2010;48:4512–8. https://doi.org/10.1128/ JCM.01683-10 28. Plazzotta G, Cohen T, Colijn C. Magnitude and sources of bias in the detection of mixed strain M. tuberculosis infection. J Theor Biol. 2015;368:67–73. https://doi.org/10.1016/ j.jtbi.2014.12.009 Address for correspondence: Sanghyuk Shin, University of California, Irvine, 106F Berk Hall, Irvine, CA 92697, USA; email: ssshin2@uci.edu 960 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 5, May 2020
10.3201_eid2507.181794	RESEARCH Asymptomatic Dengue Virus Infections, Cambodia, 2012–2013 Sowath Ly,1 Camille Fortas,1 Veasna Duong, Tarik Benmarhnia, Anavaj Sakuntabhai, Richard Paul, Rekol Huy, Sopheak Sorn, Kunthy Nguon, Siam Chan, Souv Kimsan, Sivuth Ong, Kim Srorn Kim, Sowathy Buoy, Lim Voeung, Philippe Dussart, Philippe Buchy,1,2 Arnaud Tarantola1 We investigated dengue virus (DENV) and asymptomatic DENV infections in rural villages of Kampong Cham Prov- ince, Cambodia, during 2012 and 2013. We conducted perifocal investigations in and around households for 149 DENV index cases identified through hospital and village surveillance. We tested participants 0.5–30 years of age by using nonstructural 1 rapid tests and confirmed DENV infec- tions using quantitative reverse transcription PCR or non- structural 1–capture ELISA. We used multivariable Poisson regressions to explore links between participants’ DENV in- fection status and household characteristics. Of 7,960 study participants, 346 (4.4%) were infected with DENV, among whom 302 (87.3%) were <15 years of age and 225 (65.0%) were <9 years of age. We identified 26 (7.5%) participants with strictly asymptomatic DENV infection at diagnosis and during follow-up. We linked symptomatic DENV infection status to familial relationships with index cases. During the 2-year study, we saw fewer asymptomatic DENV infections than expected based on the literature. testing of dengue-like cases in referral pediatric hospitals in Cambodia likely underestimate the true disease burden (4). By definition, syndromic surveillance does not detect asymptomatic DENV infections, which increase vector transmission potential (5). Mammen et al. used both den- gue-positive and dengue-negative index cases of febrile children to initiate perifocal investigations and found no cases in proximity to dengue-negative index cases (6). To maximize the number of recruited cases, we investigated homes around preidentified, dengue-positive index cases, as per a previous study (7). Our objectives were to docu- ment the proportion of strictly asymptomatic infections in this region of Cambodia; characterize human, sociode- mographic, household-level, and mosquito control–related factors associated with DENV infection; and identify fac- tors associated with asymptomatic DENV infection. Methods Annually, ≈390 million people in >100 countries are infected with dengue virus (DENV); 70% of cases occur in countries in Asia (1). DENV is a flavivirus trans- mitted by Aedes aegypti and Ae. albopictus anthropo- philic female mosquitoes. DENV has 4 distinct serotypes, DENV-1–4 (2); DENV infections can range from asymp- tomatic to life-threatening. In Cambodia, the national dengue surveillance system reported 60,000 cases and 135 deaths attributed to DENV in 2012 and 2013 (3). Syndromic surveillance and random Author affiliations: Institut Pasteur du Cambodge, Phnom Penh, Cambodia (S. Ly, C. Fortas, V. Duong, S. Sorn, K. Nguon, S. Chan, S. Kimsan, S. Ong, P. Dussart, P. Buchy, A. Tarantola); University of California, San Diego, California, USA (T. Benmarhnia); Institut Pasteur, Paris, France (A. Sakuntabhai, R. Paul); Malaria National Center, Phnom Penh (R. Huy); Kampong Cham Provincial Hospital, Kampong Cham, Cambodia (K.S. Kim); Prey Chhor District Referral Hospital, Kampong Cham (S. Buoy); Tboung Khmum District Referral Hospital, Thoung Khmum, Cambodia (L. Voeung) DOI: https://doi.org/10.3201/eid2507.181794 Ethics Considerations The study protocol was approved by the Cambodian National Ethics Committee on Health Research. We ob- tained informed consent from participants or their guard- ians documented during hospital or village surveillance or perifocal investigations. Study Site We conducted a study in rural villages of Kampong Cham Province, 120 km northeast of Cambodia’s capital, Phnom Penh. The study area included 368 villages with ≈60,000 households and 3 hospitals within a 30-km radius. Dengue is endemic in the region and mainly affects children <15 years of age during the annual rainy season (June–October). Identification of Dengue Index Cases in Hospitals and Villages During June 1–October 31, 2012 and 2013, we identified DENV index cases in 3 referral hospitals and 26 villages 1These authors contributed equally to this article. 2Current affiliation: GlaxoSmithKline Vaccines Research and Design, Singapore. 1354 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 25, No. 7, July 2019 under active surveillance for febrile illness (5). We targeted persons 0.5–30 years of age. In the 3 hospitals, blood sam- ples were drawn at admission and discharge for all patients suspected of having DENV infection on the basis of clinical assessment and platelet count. In the 26 villages, volunteers monitored eligible residents weekly, measuring axillary body temperature using a digital thermometer to identify persons with temperatures >38°C. Blood samples were drawn 1–2 days after fever onset, as described elsewhere (4,8,9). All sam- ples were screened for DENV infection by using a nonstruc- tural (NS) 1 IgM/IgG combination rapid test. We confirmed DENV by using quantitative reverse transcription PCR (qRT- PCR) or NS1-capture ELISA and included case-patients with confirmed DENV infection as index cases in the study. Perifocal Investigations Within 1–2 days of identifying an index case, whether from village or hospital surveillance, we began a perifocal inves- tigation of the index case-patient’s village of origin (7). For each perifocal investigation, we used a rapid dengue test kit to screen eligible residents in the index case-patient’s household for DENV and completed a baseline question- naire on individual symptoms, socioeconomic status, and household characteristics. We did the same in 20 house- holds in a 100-meter radius of the index case-patient’s household. We included persons 0.5–30 years of age who consented or whose guarantor consented. We tested adults >20 years of age during the first year of the study but found no DENV-positive cases and did not test this age group during the second year. All consecutive cases were eligible for inclusion. To avoid bias through overlapping investi- gations of a potentially common source of infection, we did not conduct a perifocal investigation within 1 week of a previous investigation for >2 index cases consecutively detected from the same village. DENV Testing and Case Definitions To screen for DENV infection during surveillance and perifocal investigations, investigators tested all blood samples on-site using SD BIOLINE Dengue Duo kit (Stan- dard Diagnostics, https://www.alere.com), according to the manufacturer's instructions. Investigators interpreted results within 15–20 minutes and ruled out possible cas- es if the control band was negative. Blood samples from DENV-positive participants were sent to Institut Pasteur du Cambodge (Phnom Penh, Cambodia) for qRT-PCR testing, as described previously (10,11), or confirmation using an NS1-capture ELISA (11,12) with positive controls diluted to the limit of detection, negative, and nontemplate controls used during extraction and PCR steps to reduce inaccura- cies (10). We considered cases confirmed when a blood sample tested positive by NS1 rapid test and was confirmed by NS1-capture ELISA or qRT-PCR. During the first year, Asymptomatic Dengue Virus Infections, Cambodia we also tested participants for Japanese encephalitis virus (JEV) and chikungunya virus (CHIKV) IgM antibodies by ELISA and confirmed IgM-positive results using spe- cific RT-PCR to ensure that symptoms were not related to CHIKV, JEV, or co-infections (11–13). Symptomatic DENV-confirmed case-patients had fe- ver, muscle or joint pain, rash, bleeding, prolonged head- aches, or digestive signs. We asked participants whether they had taken antipyretics in the previous 24 hours. We termed afebrile all symptomatic DENV-positive partici- pants without a fever and no antipyretic use. We consid- ered participants asymptomatic when they had confirmed DENV infection, no antipyretic use, and no signs or symp- toms, including fever. Participants who were symptomatic at initial diagnosis on day 0 received follow-up monitoring on days 2 and 7. We monitored asymptomatic participants daily on days 0–7 and again on day 10 using a question- naire to document signs and symptoms of DENV. In our analyses, we recategorized participants who were asymp- tomatic at baseline to symptomatic if they reported any symptoms during the follow-up period. Statistical Analysis We described DENV infection attack rates for perifocal investigations and the proportion of asymptomatic cases among all DENV infections and circulating serotypes. To explore participant- and household-level factors associated with DENV infection, we conducted a multivariable Pois- son regression estimating attack rate ratios (ARRs) (14), excluding index cases. We built explanatory models around each participant-level and household-level factor, with and without adjusting for covariates. Participant-level factors included age, sex, occupation or schooling, and relation- ship to an index case-patient. Because we found collinear- ity between age and occupation, we adjusted only for age. We placed participants 0.5–1 year of age into a specific cat- egory to account for differences in immunity and exposure to vectors due to reduced mobility. Household-level fac- tors included the main source of income, source of water, measures against mosquitoes, and environmental factors favorable to mosquito development. We further divided the source of water into 2 categories: piped water (from indoor or outdoor taps with a tube well and pump) or nonpiped water (from a pond, river, lake, or a well without pump). Considering the limited flight range of a female Aedes mosquito, we assumed that the probability of DENV trans- mission would be higher within a household than across households. To account for this factor and measure poten- tial clustering, we developed a random-effects multilevel model. We computed the intraclass correlation coefficient as the proportion of the variability in the probability of infection attributable to differences between households versus differences within households (15). We excluded Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 25, No. 7, July 2019 1355 RESEARCH 19 participants with missing covariates or predictors from the regression analyses. We explored associations between asymptomatic DENV infection and DENV serotype, par- ticipant-level factors, and the main source of income as so- cioeconomic indicators. We used the Fisher exact test for comparing proportions, the Student t test for means, and an empty multilevel model to search for a cluster effect. We conducted analyses using Stata version 13 (StataCorp, https://www.stata.com). Results Dengue Surveillance for Index Case Identification We identified 1,294 suspected DENV-infected persons, 834 (64.5%) among hospital inpatients and 460 (35.5%) through febrile illness surveillance in villages. Our testing confirmed 555 (66.5%) DENV-positive cases among hos- pital patients and 36 (7.8%) DENV-positive cases through febrile illness surveillance in villages. Perifocal Investigations From the 591 DENV-positive patients, we selected 149 (25.2%) consecutive cases for which we conducted perifocal investigations: 131 from hospital patients, termed PI-H, and 18 from village febrile surveillance, termed PI-V. Perifocal investigations took place in 104 villages over the 2 rainy sea- sons and documented 7,960 participants, 6,811 (86%) male and 1,149 (14%) female, in 2,988 households (Figure). We found 346 (4.3%) persons who were positive for DENV infection, 225 (65.0%) of whom were <9 years of age. We determined attack rates of 14.7/1,000 partici- pants (14/952) in PI-V and 47.4/1,000 (332/7,008) in PI-H (p<0.05). The attack rate over the 2 outbreak seasons in- creased marginally from 37/1,000 persons 0.5–30 years of age during the 2012 season to 46/1,000 among those 0.5– 20 years of age during 2013 (p = 0.056). Only 26 (7.5%) of 346 DENV-positive participants remained strictly asymptomatic during the 10-day follow-up, an asymptom- atic DENV-infection attack rate of 3.3/1,000 over the 2 years of our study. The proportion of asymptomatic infec- tions was 21.4% (3/14) in PI-V and 6.9% (23/332) in PI-H. Besides headache and fever, symptomatic case- patients mainly experienced muscle, retro-orbital, and joint pain. Although fever is considered a typical symptom of DENV infection, careful interview, rigorous clinical assessment, and follow-up interviews showed that partici- pants remained afebrile in 110 (31.8%) of the 320 symp- tomatic DENV infections, even without antipyretics. Only 6 (1.7%) of the DENV-positive case-patients required hos- pitalization, 2 with bleeding. The 2 annual outbreaks were dominated by DENV-1. However, DENV-2 and DENV-4 emerged in 2013, and we detected DENV-3 sporadically (Table 1). During the first year of the study, samples from all symptomatic and asymptomatic DENV cases were negative for CHIKV by MAC-ELISA. Because we diagnosed no CHIKV in year 1, and our national surveillance system also did not detect any CHIKV cases (data not shown), we did not perform CHIKV testing during year 2. Of 26 asymptomatic cases confirmed by qRT-PCR or NS1-capture ELISA, 6 had positive JEV serology and also were positive for DENV IgM. We could not conclude whether JEV-positive results were indicative of a recent or acute JEV co-infection or the result of cross-reaction among flaviviruses. Among hospi- talized patients, 2 had positive JEV results without detect- able DENV IgM, even though qRT-PCR or NS1-capture ELISA was positive. These results could suggest a recent or acute JEV co-infection. During perifocal investigations, 42 participants tested positive for JEV by MAC-ELISA Figure. Participant screening and data flowchart for perifocal investigations for asymptomatic DENV infection, Cambodia, 2012–2013. Initial DENV screening of febrile cases was conducted using nonstructural (NS) 1 IgM/IgG combo rapid test. Perifocal investigations took place in villages of index cases; we screened all persons in 20 households within a 100-m radius of an index case household. We excluded persons <0.5 and >30 years of age. Laboratory confirmation of DENV was conducted through quantitative reverse transcription PCR or NS1-capture ELISA. DENV, dengue virus; PI-H, perifocal investigations conducted for index cases identified through hospital surveillance; PI-V, perifocal investigations conducted for index cases identified through village febrile surveillance. 1356 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 25, No. 7, July 2019 Asymptomatic Dengue Virus Infections, Cambodia with negative DENV results, NS1, and qRT-PCR, support- ing evidence of JEV co-circulation in the country (16). Screened participants had a mean age (+ SD) of 11.7 (+ 7.9; median 10; interquartile range 6–16); 6,207 (77.9%) were schoolchildren, university students, or nonschooled children. The main sources of household income were planting crops (61.0%), working in a factory (14.3%), and keeping a shop (13.4%). Participants reported low use of protective measures against mosquitoes, including mos- quito coils in 787 (26.3%) households, insecticide sprays in 557 (18.6%) households, and larvicidal temephos in 374 (12.5%) households. Our investigation found uncov- ered water jars in 1,867 (62.7%) households and mosquito larvae in water containers of 1,663 (55.7%) households (Table 2). Among DENV-positive cases, boys and girls were equally affected at a mean (+ SD) age of 8.5 (± 5.7) years. Compared with persons 15–30 years of age, we found that children 1–10 years of age had a higher ARR of DENV infection (ARR 4.04 [95% CI 2.72–5.98] for those 1–5 years of age and ARR 3.83 [95% CI 2.59–5.67] for those 6–10 years of age). Siblings and cousins of index case- patients were more prone to DENV infection than neigh- bors were; siblings were 2.24 (95% CI 1.42–3.53) times and cousins 1.40 (95% CI 1.02–1.90) times more at risk for infection than neighbors. Participants who used piped wa- ter had a higher risk for DENV infection (ARR 1.35 [95% CI 1.06–1.71]) than did those who used nonpiped water. Households in which the main source of income was fish- ing, farming, or animal husbandry also had higher risks for infection (ARR 2.02 [95% CI 1.18–3.45]). Households re- porting mosquito control–related parameters did not have a lower risk for DENV infection (Table 2). The main source of income was similarly distributed between households with ≥1 case and households with no cases (p = 0.272). Our multilevel model showed a notable clustering effect at the household level after adjustment (in- traclass correlation coefficient 40.8%). We found 26 (7.5%) case-patients, 17 (65.4%) male and 9 (34.6%) female, who were positive for DENV in- fection but remained asymptomatic. We found serotypes DENV-1, DENV-2, and DENV-4 in our study group (Ta- ble 3). We used a multilevel approach to explore the role of specific serotypes and participant-level factors, such as age, gender, and relationship to the index case-patient, a proxy for common genetic background, with being DENV- positive and asymptomatic. We found that only family relationship to the index case-patient was associated with asymptomatic infection. We did not identify a cluster effect or associated factors. Discussion We screened 7,960 participants in communities in Cambo- dia during 2012 and 2013 and found 346 (4.3%) participants Table 1. Surveillance data from perifocal investigations for asymptomatic dengue virus, Cambodia, 2012–2013* Surveillance data No. participants No. villages investigated No. perifocal investigations conducted Mean no. participants per perifocal investigation Confirmed infections, no. (%) Strictly asymptomatic Afebrile Symptomatic Attack rate/1,000 participants, % Asymptomatic infections Symptomatic infections Afebrile infections 2012 2,391 35 47 51 88 5 (5.7) 33 (37.5) 83 (94.3) 36.8 2.1 34.7 13.8 83 55 (66.2) 52 (62.7) 16 (19.3) 17 (20.5) 17 (20.5) 15 (18.1) 13 (15.7) 3 (3.5) 88 2013 5,569 77 102 55 258 21 (8.1) 77 (29.8) 237 (91.9) 46.3 3.8 42.6 13.8 237 180 (75.9) 169 (71.3) 73 (30.8) 73 (30.8) 68 (28.7) 53 (22.4) 50 (21.1) 8 (3.3) 258 82 (98.8) 1 (1.2) 0 0 0 5 188 (72.9) 36 (13.9) 2 (0.8) 31 (12.0) 1 (0.4) 0 Total 7,960 104 149 53 346 26 (7.5) 110 (31.2) 320 (92.5) 43.5 3.3 40.2 13.8 320 236 (73.8) 221 (69.1) 89 (27.8) 90 (28.1) 85 (26.5) 68 (21.3) 63 (19.7) 11 (3.3) 346 270 (78.0) 37 (10.7) 2 (0.6) 31 (9.0) 1 (0.3) 5 Symptoms at diagnosis or follow-up, no. (%) Fever Headache Muscle pain Retro-orbital pain Joint pain Rash Any bleeding Hospitalizations, no. (%) DENV infections Serotype, no. (%) DENV-1 DENV-2 DENV-3 DENV-4 DENV-1 and DENV-2 Missing Participants 0.5–30 years of age in 2012 and 0.5–20 years of age in 2013. DENV, dengue virus. Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 25, No. 7, July 2019 1357 RESEARCH infected by DENV; 26 (7.5%) remained asymptomatic be- fore, during, and after our study. We found comparable at- tack rates, 37/1,000 persons in 2012 and 46/1,000 persons in 2013, to other community investigations conducted in Cam- bodia. For instance, another study reported DENV attack rates of 13.4–57.8 cases/1,000 persons during 2006–2008 (4). Previous studies only included participants ≤20 years of age, but we included persons 0.5–30 years of age with confirmed DENV infection, even symptomatic but afebrile case-patients, who were 31.8% of the DENV infections in Table 2. Participant and household characteristics with unadjusted and adjusted attack rate ratios for factors potentially associated with dengue virus infection, Cambodia, 2012–2013 Characteristics Participants Sex M F Age, y† 0.5–<1 1–<5 5–<10 10–<15 15–30 Mean (+ SD, median) Occupation‡ Student, school or university Preschool or unschooled Planting crops Other Missing Relationship to index case-patient§ Neighbor Cousin Sibling Other Missing Households Water source# Nonpiped Piped Primary source of income Planting crops Working in a factory Shopkeeping Fishing, farming, animal husbandry Working in government Other Mosquito control measures†† Temephos Larvivorous fish Treated mosquito netting Treated jar cover Coils Insecticide spray Environmental factors Vegetable garden Water collection around house Uncovered water jars Larvae in water containers Distance from house to nearest water jar, m (+ SD) Missing for all items Infected 346 Uninfected 7,614 171 175 4,103 3,511 Total 7,960 4,272 3,686 9 (2.6) 108 (31.2) 126 (36.4) 71 (20.5) 32 (9.3) 150 (2.0) 1,701 (22.3) 2,083 (27.4) 1,675 (22.0) 2,005 (26.3) 8.5 (+ 5.7, 7) 11.9 (+ 8.0, 10) 11.7 (+ 7.9, 10) 159 (2.0) 1,809 (22.7) 2,209 (27.8) 1,746 (21.9) 2,037 (25.6) 171 (49.8) 149 (43.2) 20 (5.8) 5 (1.5) 1 (0.2) 260 (75.4) 58 (16.8) 23 (6.7) 5 (1.2) 1 (0.2) 292¶ 3,588 (47.2) 2,299 (30.2) 910 (12.0) 809 (10.6) 8 (0.1) 6,309 (83.0) 991 (13.0) 251 (3.3) 55 (0.7) 8 (0.1) 2,706 3,759 (47.3) 2,448 (30.8) 930 (11.7) 814 (10.2) 9 (1.1) 6,569 (82.6) 1,049 (13.2) 274 (3.5) 59 (0.7) 9 (1.1) 2,988 Unadjusted ARR (95% CI) Adjusted ARR (95% CI) Referent Referent 1.14 (0.92–1.40) 1.01 (0.82–1.24) 3.47 (1.65–7.32) 3.53 (1.67–7.46) 3.98 (2.69–5.90) 4.04 (2.72–5.98) 3.79 (2.56–5.60) 3.83 (2.59–5.67) 2.59 (1.70–3.94) 2.55 (1.67–3.88) Referent – Referent – 2.14 (1.35–3.41) 2.14 (1.34–3.41) 2.84 (1.79–4.54) 2.84 (1.78–4.54) Referent Referent 0.28 (0.10–7.76) 0.28 (0.10–7.76) Referent Referent 1.38 (1.01–1.89) 1.40 (1.02–1.90) 2.11 (1.33–3.34) 2.24 (1.42–3.53) 1.66 (0.59–4.65) 1.76 (0.34–4.90) 108 (36.3) 184 (63.7) 1,186 (43.7) 1,520 (56.3) 1,284 (43.0) 1,704 (57.0) Referent Referent 1.32 (1.03–1.69) 1.35 (1.06–1.71) 176 (60.9) 42 (14.5) 37 (12.8) 14 (4.8) 5 (1.7) 15 (5.2) 26 (9.0) 26 (9.0) 27 (9.3) 3 (1.0) 77 (26.6) 44 (15.2) 57 (9.7) 126 (43.6) 178 (61.6) 168 (58.1) 1.5 (+ 2.2) 1,648 (61.0) 384 (14.2) 362 (13.4) 55 (2.0) 57 (2.1) 193 (7.2) 348 (12.9) 214 (7.9) 311 (11.5) 47 (1.7) 710 (26.3) 513 (19.0) 1,824 (61.0) 426 (14.3) 399 (13.4) 69 (2.3) 62 (2.1) 208 (7.0) Referent Referent 1.16 (0.84–1.62) 1.20 (0.87–1.66) 0.97 (0.67–1.40) 1.03 (0.72–1.48) 1.98 (1.15–3.43) 2.02 (1.18–3.45) 0.94 (0.38–2.30) 0.99 (0.41–2.37) 0.76 (0.43–1.32) 0.85 (0.50–1.46) 374 (12.5) 240 (8.3) 338 (11.3) 50 (1.7) 787 (26.3) 557 (18.6) 0.70 (0.47–1.06) 0.73 (0.48–1.10) 1.14 (0.75–1.74) 1.18 (0.78–1.79) 0.78 (0.52–1.17) 0.82 (0.55–1.21) 0.73 (0.24–2.24) 0.77 (0.26–2.27) 1.08 (0.82–1.41) 1.16 (0.89–1.51) 0.79 (0.57–1.10) 0.88 (0.63–1.22) 546 (20.2) 1,180 (43.7) 1,689 (62.6) 1,495 (55.4) 1.3 (+ 2.0) 603 (20.2) 1,306 (44.7) 1,867 (62.5) 1,663 (55.7) 1.3 (+ 2.0) 0.89 (0.66–1.21) 0.89 (0.66–1.20) 0.91 (0.71–1.15) 0.88 (0.70–1.12) 0.96 (0.75–1.22) 0.97 (0.76–1.23) 1.09 (0.86–1.39) 1.07 (0.85–1.37) 1.00 (0.98–1.02) 1.00 (0.98–1.02) 3 7 10 Values are no. or no. (%) except as indicated. ARR, attack rate ratio; DENV, dengue virus. †Participants 0.5–30 years of age in 2012 and 0.5–20 years of age in 2013. ‡Adjusted for sex. §Adjusted for age. ¶No. housesholds with >1 DENV case. #Nonpiped water comes from a river, pond, lake, or a well that does not have a pump; piped water comes from indoor or outdoor taps with a tube well and pump. Adjusted for sex and occupation. ††Adjusted for age, relationship to index case-patient, occupation, and primary source of income. 1358 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 25, No. 7, July 2019 Table 3. Univariate tests for associations between sociodemographic factors and infecting serotypes with asymptomatic dengue virus infections, Cambodia, 2012–2013 Asymptomatic, n = 26 Symptomatic, n = 320 p value† 17 9 154 166 Factor Sex M F Age, y 0.5 to <1 1–5 6–10 11–14 15–30 Mean (+ SD, median) 0 9 (34.6) 9 (34.6) 5 (19.2) 3 (11.5) 9.2 (+ 7.2, 8) Relationship to index case-patient 17 (65.4) Neighbor 5 (19.2) Cousin 1 (3.9) Sibling Other 3 (11.5) Source of household income Planting crops Working in a 14 (53.8) 3 (11.5) 0.09 0.976 0.004 0.812 9 (2.8) 99 (30.9) 117 (36.6) 66 (20.6) 29 (9.1) 11.0 (+ 7.1, 10) 243 (76.0) 53 (16.6) 22 (6.9) 1 (0.3) 192 (60.0) 53 (16.6) factory 0 6 (1.9) 3 (11.5) 5 (19.2) 1 (3.9) 36 (11.3) 17 (5.3) Shopkeeping Fishing, farming, animal husbandry Working in government Other or missing DENV serotype‡ 21 (80.8) DENV-1 2 (7.7) DENV-2 0 DENV-3 DENV-4 3 (11.5) *Values are no. (%) patients except as indicated. DENV, dengue virus. †By Fisher test or 2 test. ‡Data for 5 symptomatic patients were missing, and another patient was excluded from analysis because of co-infection with DENV-1 and DENV-2. 249 (79.1) 35 (11.1) 2 (0.6) 28 (8.9) 16 (5.0) 0.892 this study. We noted that attack rates were lower in PI-V, 14.7/1,000 participants (14/952), than in PI-H, 47.4/1,000 participants (332/7,008). Circulation of DENV around fe- brile index case-patients identified through PI-V was less intense, but with more asymptomatic cases, than around in- dex case-patients identified through PI-H. Aside from pos- sible detection biases (17), multiple factors could explain this observation and deserve further research. Our study documented cases of DENV infection in trans- mission clusters located around index case-patients. We found that 26.6% of DENV-confirmed case-patients reported clini- cal symptoms, including headache and muscle pain, but no fever even in the absence of antipyretics, comparable to data from Thailand, where 40.4% of the DENV cases remained afebrile (17). The appearance of afebrile DENV-infected pa- tients raises potential concerns for case definitions for detec- tion, especially of imported cases in at-risk countries. An additional 7.5% of DENV-confirmed case-patients had no symptoms during the 10-day course of clinical monitoring, a considerably lower rate than estimates from other prospective studies (5,18–21). Published sources refer Asymptomatic Dengue Virus Infections, Cambodia to inapparent infections, often defined as afebrile clinical complaints with biologic evidence of DENV infection, rang- ing from 20% to 80% of cases (19,22,23). Previous studies used different definitions of asymptomatic infection than ours, but the major difference lies in follow-up monitoring. Other retrospective studies used school or work absentee- ism as a basis for follow-up (19). Strictly asymptomatic patients, such as those we describe, escape detection by sur- veillance or control measures, infect mosquitoes, and might disproportionately contribute to DENV transmission (5). The DENV burden documented through hospital- based surveillance of febrile case-patients in Thailand and Vietnam showed a shift to older age groups (24,25). Our active, systematic case-finding system to identify DENV in villages in Cambodia found the attack rate was highest in children <10 years of age, which is what we expected in a dengue-endemic country with frequent outbreaks demon- strated in other careful studies (26). This finding raises con- cerns because recommendations for the only licensed den- gue vaccine are for use in persons 9–45 years of age with demonstrated evidence of past DENV infection (27). Our study demonstrates that children in rural Cambodia might have undergone >1 DENV infection before 9 years of age, reducing the potential cost-effectiveness of vaccination. Few studies have explored the role of socioeconom- ic status, which might be a proxy for peridomestic envi- ronmental management, on DENV infection in Southeast Asia. Often, the direction of the association is unclear and socioeconomic status has entirely different associations de- pending on the setting (28). Our study shows that the ad- justed risk for DENV infection was highest in households in which the main source of income was from fishing, farm- ing, or animal husbandry, activities associated with lower average household income in Cambodia. We found temephos provided no additional protection against DENV infection after adjusting for other factors. Although temephos is effective in reducing Aedes spp. lar- val populations in water storage jars, its use did not cor- relate with lower DENV transmission in Cambodia or else- where (8), due to incorrect distribution coverage, dosage, and placement (29) or multiple vector breeding sites. In ad- dition, increases in temephos-resistant A. aegypti mosquito larvae have been documented in Cambodia (30). Unexpectedly, we found a higher risk for DENV with piped water as a main water source after adjusting for other factors, contrary to a study in Thailand (6). However, piped water with suboptimal sanitation in Cambodia might contrib- ute to collection of water in or around households that could become breeding sites for DENV-transmitting mosquitoes. We found that 40.8% of the variability in probabil- ity of being DENV infected was explained by differences between households. Those living in the same household as an index case-patient were 2.11 times more likely to be Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 25, No. 7, July 2019 1359 RESEARCH infected, consistent with other published sources. A study in Mexico found that the risk for infection for those liv- ing with an index case-patient was twice that of someone living in a 50-meter radius of an index case-patient (31). This relationship was further described in a cluster study in Thailand that showed decreased risk for infection with increasing distance from the index case household (31). This clustering effect around an index case, however, seems to occur only on a short temporal scale, at least in urban settings (32). Rates and severity of illness after infection with the different DENV serotypes differ widely (33,34). The only notable epidemiologic factor associated with asymptomatic DENV infections in our study was being family-related to the index case-patient. Another study showed that adaptive immune responses against DENV differ between persons with symptomatic and asymptomatic DENV infection (35), which might explain our observations. We found no other associated epidemiologic factor, including age or cluster effects. Although the ratio of male to female participants was twice as high among asymptomatic than symptomatic participants, this finding was not statistically significant, likely due to sample size. Although the size and duration of our study confer strength to our data, it might suffer from biases and limita- tions, especially due to the small number of strictly asymp- tomatic DENV-positive participants after stratifying by DENV serotype. We found dengue incidence rates high- est in young children. These data might be biased because we focused on investigating clusters around an index case, perhaps overestimating the incidence in the general popula- tion. DENV circulation, however, is intense in children in Cambodia, and these figures remain comparable to those found in dengue studies that use different methods, ranging from 20 to 80 per 1,000 person-seasons (1). Furthermore, we did not capture cases referred to the private sector, low- ering our estimates somewhat. Healthy male workers often were away at the time of the investigations, possibly lead- ing to an overestimation of DENV incidence. These work- ers, however, are >18 years of age, but DENV infections occur mainly in persons <15 years of age in Cambodia (4). Documentation bias might also have pulled our risk factor estimates toward the null. We did not document solid waste disposal in our study, but comparatively high Bre- teau index values have been reported in Cambodia (29). In addition, we could have missed details or misrepresented implementation of mosquito-control measures. Despite the potential misclassification, mosquito-control measures re- main nondifferential and likely had no major effect on our risk estimates. Further, 7.5% of our DENV-infected participants re- mained strictly asymptomatic. Aside from case definition issues we discuss, our method of screening for DENV around symptomatic cases might have underestimated the number of asymptomatic DENV infections. In addition, we did not enroll persons who tested negative for DENV IgM, NS1-capture ELISA, and qRT-PCR. Some of these persons might have been infected but not yet mounted an IgM response, so that NS1 and viral RNA titers had al- ready receded to undetectable levels when we tested them. This strict case definition might have underestimated the incidence of asymptomatic cases, but a precise retrospec- tive documentation of such cases would be extremely difficult. Similarly, we retrospectively conducted MAC- ELISA on samples collected during perifocal investiga- tions and identified 11 cases of IgM seroconversion in the absence of PCR- or NS1-positive tests. Even in the context of JEV cocirculation, some of these cases could have been true DENV infections, but including them would not have changed the overall estimated attack rate. Previous stud- ies suggested virus serotype might affect severity and types of symptoms and observed that DENV-1 infections more frequently were associated with clinically apparent illness (36,37). Virus molecular analysis studies are ongoing to determine whether specific strains cause more asymptom- atic infection than others. Furthermore, DENV infection in Cambodia occurs mainly in children who might be more likely to answer positively to daily-repeated questions on dengue symptoms, somewhat underestimating asymptom- atic cases. Having implemented careful and thorough 10- day clinical assessment of objective symptoms in each as- ymptomatic DENV-positive participant, we believe these figures reflect the true proportion of strictly asymptomatic DENV infection in our setting. However, we collected our findings mainly in children with DENV-1 infection in Cambodia. Whether these findings are directly applicable to other epidemiologic settings, populations, or virus sero- types or genotypes remains to be determined (33). Finally, vaccination against JEV might have led to cross-protection against symptomatic dengue. Data on JEV vaccination were not collected during perifocal investiga- tions. According to local health centers, however, JEV vac- cine has been provided only recently and only for children 9–24 months of age. In our study, only 3 children were DENV-positive in that age category. Our study demonstrates that systematically relying on fever for DENV case definition can underestimate cases and hinder control efforts in areas with potential vectors and at risk for DENV introduction. We found 7.5% of DENV-infected participants remained strictly asymptom- atic, which has wide-ranging epidemiologic consequences. Undetected sources can increase transmission (5), a factor that must be taken into account in future vaccine coverage and vaccine effectiveness studies. The attack rate differ- ences observed around febrile index case-patients detected in village surveillance and index case-patients detected in 1360 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 25, No. 7, July 2019 hospital surveillance deserve further study. In-depth virus (36) and human genetic studies could contribute useful in- sights (33,35). Our strict definition of asymptomatic DENV infections should be considered when designing studies that aim to elucidate the pathophysiological mechanisms of dengue disease. Acknowledgments The authors gratefully acknowledge participating hospitals, villages, and study participants, as well as Tineke Cantaert for her editorial comments and suggestions. Dengue Framework for Resisting Epidemics in Europe studies were funded by a grant (no. 282378) from the European Union 7th FP. P.B. is a former head of virology at Institut Pasteur du Cambodge and is an employee of GSK Vaccines, Singapore. About the Authors Dr. Ly is a medical doctor and epidemiologist at the Epidemiology & Public Health Department at Institut Pasteur du Cambodge; his primary research interests are epidemiology of endemic and epidemic arboviruses and zoonoses in the Mekong Region. Ms. Fortas is an epidemiologist whose research interests are tropical infectious diseases in low- and middle- income countries. References 1. Bhatt S, Gething PW, Brady OJ, Messina JP, Farlow AW, Moyes CL, et al. The global distribution and burden of dengue. Nature. 2013;496:504–7. http://dx.doi.org/10.1038/nature12060 2. Simmons CP, Farrar JJ, van Vinh Chau N, Wills B. Dengue. N Engl J Med. 2012;366:1423–32. http://dx.doi.org/10.1056/ NEJMra1110265 3. Western Pacific Regional Office of the World Health Organization. Dengue situation updates [cited 2017 Jul 12]. http://www.wpro. who.int/entity/emergencies/wp_dengue_biweekly/en 4. Vong S, Goyet S, Ly S, Ngan C, Huy R, Duong V, et al. Under-recognition and reporting of dengue in Cambodia: a capture- recapture analysis of the National Dengue Surveillance System. Epidemiol Infect. 2012;140:491–9. http://dx.doi.org/10.1017/ S0950268811001191 5. Duong V, Lambrechts L, Paul RE, Ly S, Lay RS, Long KC, et al. Asymptomatic humans transmit dengue virus to mosquitoes. Proc Natl Acad Sci U S A. 2015;112:14688–93. http://dx.doi.org/ 10.1073/pnas.1508114112 6. Mammen MP, Pimgate C, Koenraadt CJM, Rothman AL, Aldstadt J, Nisalak A, et al. Spatial and temporal clustering of dengue virus transmission in Thai villages. PLoS Med. 2008;5:e205. http://dx.doi.org/10.1371/journal.pmed.0050205 7. Dussart P, Baril L, Petit L, Beniguel L, Quang LC, Ly S, et al. Clinical and virological study of dengue cases and the members of their households: the multinational DENFRAME Project. PLoS Negl Trop Dis. 2012;6:e1482. http://dx.doi.org/10.1371/ journal.pntd.0001482 8. Huy R, Buchy P, Conan A, Ngan C, Ong S, Ali R, et al. National dengue surveillance in Cambodia 1980–2008: epidemiologi- cal and virological trends and the impact of vector control. Bull Asymptomatic Dengue Virus Infections, Cambodia World Health Organ. 2010;88:650–7. http://dx.doi.org/10.2471/ BLT.09.073908 9. Vong S, Khieu V, Glass O, Ly S, Duong V, Huy R, et al. Dengue incidence in urban and rural Cambodia: results from population- based active fever surveillance, 2006–2008. PLoS Negl Trop Dis. 2010;4:e903. http://dx.doi.org/10.1371/journal.pntd.0000903 10. Hue KDT, Tuan TV, Thi HTN, Bich CTN, Anh HHL, Wills BA, et al. Validation of an internally controlled one-step real-time multiplex RT-PCR assay for the detection and quantitation of dengue virus RNA in plasma. J Virol Methods. 2011;177:168–73. http://dx.doi.org/10.1016/j.jviromet.2011.08.002 11. Andries A-C, Duong V, Ngan C, Ong S, Huy R, Sroin KK, et al. Field evaluation and impact on clinical management of a rapid diagnostic kit that detects dengue NS1, IgM and IgG. PLoS Negl Trop Dis. 2012;6:e1993. http://dx.doi.org/10.1371/ journal.pntd.0001993 12. Andries A-C, Duong V, Ong S, Ros S, Sakuntabhai A, Horwood P, et al. Evaluation of the performances of six commercial kits designed for dengue NS1 and anti-dengue IgM, IgG and IgA detection in urine and saliva clinical specimens. BMC Infect Dis. 2016;16:201. http://dx.doi.org/10.1186/s12879-016-1551-x 13. Duong V, Andries A-C, Ngan C, Sok T, Richner B, Asgari-Jirhandeh N, et al. Reemergence of chikungunya virus in Cambodia. Emerg Infect Dis. 2012;18:2066–9. http://dx.doi.org/ 10.3201/eid1812.120471 14. Zou G. A modified Poisson regression approach to prospective studies with binary data. Am J Epidemiol. 2004;159:702–6. http://dx.doi.org/10.1093/aje/kwh090 15. Wu S, Crespi CM, Wong WK. Comparison of methods for estimating the intraclass correlation coefficient for binary responses in cancer prevention cluster randomized trials. Contemp Clin Trials. 2012;33:869–80. http://dx.doi.org/10.1016/j.cct.2012.05.004 16. Tarantola A, Goutard F, Newton P, de Lamballerie X, Lortholary O, Cappelle J, et al. Estimating the burden of Japanese encephalitis virus and other encephalitides in countries of the Mekong region. PLoS Negl Trop Dis. 2014;8:e0002533. https://doi.org/10.1371/ journal.pntd.0002533 17. Yoon I-K, Srikiatkhachorn A, Hermann L, Buddhari D, Scott TW, Jarman RG, et al. Characteristics of mild dengue virus infection in Thai children. Am J Trop Med Hyg. 2013;89:1081–7. http://dx.doi.org/10.4269/ajtmh.13-0424 18. Porter KR, Beckett CG, Kosasih H, Tan RI, Alisjahbana B, Rudiman PIF, et al. Epidemiology of dengue and dengue hemorrhagic fever in a cohort of adults living in Bandung, West Java, Indonesia. Am J Trop Med Hyg. 2005;72:60–6. http://dx.doi.org/10.4269/ajtmh.2005.72.60 19. Endy TP, Chunsuttiwat S, Nisalak A, Libraty DH, Green S, Rothman AL, et al. Epidemiology of inapparent and symptomatic acute dengue virus infection: a prospective study of primary school children in Kamphaeng Phet, Thailand. Am J Epidemiol. 2002;156:40–51. http://dx.doi.org/10.1093/aje/kwf005 20. Kosasih H, Alisjahbana B, Nurhayati, de Mast Q, Rudiman IF, Widjaja S, et al. The epidemiology, virology and clinical findings of dengue virus infections in a cohort of Indonesian adults in western Java. PLoS Negl Trop Dis. 2016;10:e0004390. http://dx.doi.org/10.1371/journal.pntd.0004390 21. Yoon I-K, Rothman AL, Tannitisupawong D, Srikiatkhachorn A, Jarman RG, Aldstadt J, et al. Underrecognized mildly symptomatic viremic dengue virus infections in rural Thai schools and villages. J Infect Dis. 2012;206:389–98. http://dx.doi.org/10.1093/infdis/jis357 22. Balmaseda A, Standish K, Mercado JC, Matute JC, Tellez Y, Saborío S, et al. Trends in patterns of dengue transmission over 4 years in a pediatric cohort study in Nicaragua. J Infect Dis. 2010;201:5–14. http://dx.doi.org/10.1086/648592 23. Endy TP, Anderson KB, Nisalak A, Yoon I-K, Green S, Rothman AL, et al. Determinants of inapparent and symptomatic Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 25, No. 7, July 2019 1361 RESEARCH dengue infection in a prospective study of primary school children in Kamphaeng Phet, Thailand. PLoS Negl Trop Dis. 2011;5:e975. http://dx.doi.org/10.1371/journal.pntd.0000975 24. Anders KL, Nga H, Thuy NTV, Ngoc TV, Tam CT, Tai LTH, et al. Households as foci for dengue transmission in highly urban Vietnam. PLoS Negl Trop Dis. 2015;9:e0003528. http://dx.doi.org/10.1371/journal.pntd.0003528 25. Limkittikul K, Brett J, L’Azou M. Epidemiological trends of dengue disease in Thailand (2000–2011): a systematic literature review. PLoS Negl Trop Dis. 2014;8:e3241. http://dx.doi.org/ 10.1371/journal.pntd.0003241 26. Hadinegoro SR, Arredondo-García JL, Capeding MR, Deseda C, Chotpitayasunondh T, Dietze R, et al.; CYD-TDV Dengue Vaccine Working Group. Efficacy and long-term safety of a dengue vaccine in regions of endemic disease. N Engl J Med. 2015;373:1195–206. http://dx.doi.org/10.1056/NEJMoa1506223 27. Dengue vaccine: WHO position paper, September 2018— recommendations. Vaccine. 2018;S0264-410X(18)31339-2. 28. Mulligan K, Dixon J, Joanna Sinn C-L, Elliott SJ. Is dengue a disease of poverty? A systematic review. Pathog Glob Health. 2015; 109:10–8. http://dx.doi.org/10.1179/2047773214Y.0000000168 29. Khun S, Manderson LH. Abate distribution and dengue control in rural Cambodia. Acta Trop. 2007;101:139–46. http://dx.doi.org/ 10.1016/j.actatropica.2007.01.002 30. Polson KA, Curtis C, Seng CM, Olson JG, Chantha N, Rawlins SC. Susceptibility of two Cambodian populations of Aedes aegypti mosquito larvae to temephos during 2001. Dengue Bull. 2001; 25:79–83. 31. Martínez-Vega RA, Danis-Lozano R, Díaz-Quijano FA, Velasco-Hernández J, Santos-Luna R, Román-Pérez S, et al. Peridomestic infection as a determining factor of dengue transmission. PLoS Negl Trop Dis. 2015;9:e0004296. http://dx.doi.org/10.1371/journal.pntd.0004296 32. Yoon I-K, Getis A, Aldstadt J, Rothman AL, Tannitisupawong D, Koenraadt CJM, et al. Fine scale spatiotemporal clustering of dengue virus transmission in children and Aedes aegypti in rural Thai villages. PLoS Negl Trop Dis. 2012;6:e1730. http://dx.doi.org/ 10.1371/journal.pntd.0001730 33. Grange L, Simon-Loriere E, Sakuntabhai A, Gresh L, Paul R, Harris E. Epidemiological risk factors associated with high global frequency of inapparent dengue virus infections. Front Immunol. 2014;5:280. http://dx.doi.org/10.3389/fimmu.2014.00280 34. Clapham HE, Cummings DAT, Johansson MA. Immune status alters the probability of apparent illness due to dengue virus infection: evidence from a pooled analysis across multiple cohort and cluster studies. PLoS Negl Trop Dis. 2017;11:e0005926. http://dx.doi.org/10.1371/journal.pntd.0005926 35. Simon-Lorière E, Duong V, Tawfik A, Ung S, Ly S, Casadémont I, et al. Increased adaptive immune responses and proper feedback regulation protect against clinical dengue. Sci Transl Med. 2017;9:eaal5088. http://dx.doi.org/10.1126/scitranslmed. aal5088 36. Li D, Lott WB, Lowry K, Jones A, Thu HM, Aaskov J. Defective interfering viral particles in acute dengue infections. PLoS One. 2011;6:e0019447. https://doi.org/10.1371/journal. pone.0019447 37. Yung C-F, Lee K-S, Thein T-L, Tan L-K, Gan VC, Wong JGX, et al. Dengue serotype-specific differences in clinical manifestation, laboratory parameters and risk of severe disease in adults, Singapore. Am J Trop Med Hyg. 2015;92:999–1005. http://dx.doi.org/10.4269/ajtmh.14-0628 Address for correspondence: Arnaud Tarantola, Institut Pasteur du Cambodge, Epidemiology and Public Health Unit, PO Box 983, Phnom Penh, Cambodia; email: arnaud.tarantola@pasteur.fr EID Podcast: Antimicrobial Drug Resistance and Gonorrhea Neisseria gonorrhoeae, the causative pathogen of gon- orrhea, has been designated an urgent antimicrobial drug resistance threat by the Centers for Disease Control and Prevention. Since the introduction of antimicrobial drugs in the first half of the 20th century, N. gonorrhoeae has successively developed resistance to each antimicrobial agent recommended for gonorrhea treatment. In the Unit- ed States, the prevalence of resistance in N. gonorrhoeae often varies by sex of partner and by geographic region. Prevalence is often greater in isolates from gay, bisexual, and other men who have sex with men than those from men who have sex only with women, and prevalence is often highest in the West and lowest in the South. Resis- tant strains, in particular penicillinase-producing N. gon- orrhoeae, fluoroquinolone-resistant N. gonorrhoeae, and gonococcal strains with re- duced cephalosporin suscep- tibility, seemed to emerge ini- tially in the West (Hawaii and the West Coast) before spread- ing eastward across the coun- try. These geographic patterns seem to support the idea that importation of resistant strains from other regions of the world, such as eastern Asia, is a primary factor of the emer- gence of resistant gonococci in the United States. Whereas antimicrobial drug prescribing patterns have been clearly associated with the emergence of resistance in other bac- terial pathogens, the degree to which domestic antimicro- bial use and subsequent selection pressure contributes to the emergence of gonococcal antimicrobial resistance in the United States is unclear. Using an ecologic approach, we sought to investigate the potential geographic and tem- poral association between antimicrobial drug susceptibil- ity among US N. gonorrhoeae isolates and domestic out- patient antimicrobial drug prescribing rates in the United States during 2005–2013. Visit our website to listen: https://www2c.cdc.gov/podcasts/ player.asp?f=8647449 ® 1362 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 25, No. 7, July 2019
10.1371_journal.pstr.0000049	RESEARCH ARTICLE A transition support system to build decarbonization scenarios in the academic community 1, 1, Je´ re´ mie Klein2, Marceau Challet2, Olivier Dangles3, Serge Janicot4, 14, Alexandre Caron15, Guillaume Piton16, Aure´ lie Verney-CarronID Nicolas GratiotID Miriam Candelas5, Ge´ raldine Sarret6, Ge´ remy Panthou1, Benoıˆt HingrayID Nicolas Champollion1, Julien Montillaud7, Pascal Bellemain8, Odin Marc9, Ce´ dric- Ste´ phane Bationo10, Loïs Monnier9, Laure Laffont9, Marie-Alice Foujols11, Ve´ ronique Riffault12, Liselotte Tinel12, Emmanuel Mignot13, Nathalie Philippon1, 17, Alain DezetterID Anne Delaballe18, Nelly Bardet19, Florence Nozay-Maurice20, Anne-Sophie Loison21, Franck Delbart22, Sandrine AnquetinID Fabien MalbetID Elodie Petitdidier28, Olivier Aumont4, Florence Michau29, Nicolas Bijon30, Jean- Philippe VidalID Louise Mimeau16, Anne Biarnès34, Charlotte Re´ capet35, Morgane Costes-Thire´ 36, Mariline Poupaud15, Maialen Barret37, Marie Bonnin38, Virginie MournetasID Bernard Tourancheau29, Bertrand GoldmanID Soret42 25, Ce´ line Berni26, Laurence Delattre27, Vincent EchevinID 26, Se´ bastien Pinel31, Oce´ ane Biabiany32, Cathy Grevesse33, 23, Christophe Baehr24, 1, Franc¸ oise ImmelID 40, Marie Paule Bonnet41, Isabelle Michaud 39, 4, 1 Institute of Environmental Geosciences (IGE), Grenoble Alps University (UGA), French National Centre for Scientific Research (CNRS), Grenoble Institute of Technology (Grenoble-INP), French Research Institute for Development (IRD), Saint-Martin d’Hères, France, 2 School of engineering in Physics, Applied Physics, Electronics & Materials Science, Grenoble Alps University (UGA), Grenoble Institute of Technology (Grenoble-INP), Grenoble, France, 3 Center in Ecology and Evolutionary Ecology (CEFE), University of Montpellier (UM), French National Centre for Scientific Research (CNRS), Practical School of Advanced Studies (EPHE), French Research Institute for Development (IRD), Montpellier, France, 4 Laboratory of Oceanography and Climate: Experiments and Numerical Approaches (LOCEAN), Sorbonne University, French National Centre for Scientific Research (CNRS), French National Museum of Natural History, French Research Institute for Development (IRD), Paris, France, 5 Institute of Physiology, Czech Academy of Sciences (ASCR), Praha, Czechia, 6 Earth Sciences Institute (ISTerre), Grenoble Alps University (UGA), University Savoie Mont Blanc (USMB), French National Centre for Scientific Research (CNRS), French Research Institute for Development (IRD), French Institute of Science and Technology for Transport, Development and Networks (IFFSTAR), Grenoble, France, 7 Research unit "Universe, Time-frequency, Interfaces, Nanostructures, Atmosphere and environment, Molecules" (UTINAM), French National Centre for Scientific Research (CNRS), University Burgundy Franche-Comte´ , Besanc¸on, France, 8 Grenoble Images Parole Signal Automatique (GIPSA-lab), Grenoble Alps University (UGA), French National Centre for Scientific Research (CNRS), Grenoble Institute of Technology (Grenoble-INP), Saint-Martin d’Hères, France, 9 Geosciences Environnement Toulouse (GET), French National Centre for Scientific Research (CNRS), French Research Institute for Development (IRD), French National Centre for Space Studies (CNES), Paul Sabatier University (UPS), Midi-Pyre´ ne´ es Observatory (OMP), Toulouse, France, 10 Economic and Social Sciences of Health and Medical Information Processing (SESSTIM), institute of public health sciences (ISSPAM), Aix-Marseille University (AMU), French National Institute of Health and Medical Research (INSERM), French Research Institute for Development (IRD), Marseille, France, 11 Institute Pierre-Simon Laplace (IPSL), Sorbonne University, French National Centre for Scientific Research (CNRS), Paris, France, 12 Centre for Education, Research and Innovation in Energy Environment (CERI EE), IMT Nord Europe, Institut Mines-Te´ le´ com (IMT), University of Lille (UDL), Lille, France, 13 Fluid Mechanics and Acoustics Laboratory (LMFA), University of Lyon (UDL), National Institute for Applied Sciences of Lyon (INSA Lyon), Central School of Lyon (ECL), UCBL, French National Centre for Scientific Research (CNRS), Villeurbanne, France, 14 Research Unit “HydroSciences Montpellier” (HSM), University of Montpellier (UM), French Research Institute for Development (IRD), French National Centre for Scientific Research (CNRS), Montpellier, France, 15 Research Unit “Animals, Health, Territories, Risks and Ecosystems” (ASTRE), University of Montpellier (UM), Agricultural Research Centre for International Development (CIRAD), French National Research Institute for Agriculture, Food and Environment (INRAE), Montpellier, France, 16 Research Unit “Torrential Erosion, Snow & Avalanches” (ETNA), Grenoble Alps University (UGA), French a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Gratiot N, Klein J, Challet M, Dangles O, Janicot S, Candelas M, et al. (2023) A transition support system to build decarbonization scenarios in the academic community. PLOS Sustain Transform 2(4): e0000049. https://doi.org/ 10.1371/journal.pstr.0000049 Editor: Atiq Zaman, Curtin University, AUSTRALIA Received: July 1, 2022 Accepted: February 12, 2023 Published: April 3, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pstr.0000049 Copyright: © 2023 Gratiot et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All underlying data is available from https://cloud.univ-grenoble-alpes.fr/ s/yDojHCrPHBdKY8D. Funding: NG, MB, AD, VE, OD, SJ, OM, NB, FN-M, VE, EP, OA, M-PB, AB were financed by the Institut PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 1 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONde Recherche pour le De´veloppement (IRD- France). OM, MAF, PBE, GS, BH, NC, PB, NP, SA, FI, FM, SA, FD, LM, IMS were financed by the Centre National de la Recherche Scientifique (CNRS-France). BG was financed by the International Space University (ISU). BT, GP, AD, JK were financed by University Grenoble Alps (UGA-France). JM was financed by the University of Franche Comte´. CG, MP, NB, AC were financed by the French Agricultural Research Centre for International Development (Centre de coope´ration internationale en recherche agronomique pour le de´veloppement, CIRAD). MC carried out this study during his internship financed by the Institut des Ge´osciences de l’Environnement (IGE UMR 5001 UGA/CNRS/IRD/Grenoble INP). MC were financed by Institute of Physiology ASCR (Czech Republic). GP, CB, JPV, OB, LM were financed by the French National Research Institute for Agriculture, Food and Environment (INRAE). LL, LM were financed by the Paul Sabatier University, Toulouse. VR and LT were financed by IMT Nord Europe. CB was financed by Me´te´o-France. LD was financed by the University of Lille. FM is retired, she carried out this study on her leisure time. CR was financed by Universite´ Pau Pays de l’Adour (UPPA). MCT, MB were financed by Institut National Polytechnique de Toulouse. AVC was financed by Univ. Paris Est Cre´teil. VM carried out this study on her personal time. SP was financed by Univ. Perpignan UPVD. A-S Loison was financed by Univ. Catholique de Lille. JK carried out this study during his intership financed by the University Grenoble Alps (UGA- France). MC carried out this study during his intership financed by the Institut des Ge´osciences de l’Environnement. EM (Emmanuel Mignot) was financed by INSA Lyon. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Decarbonization scenarios in the academic community National Research Institute for Agriculture, Food and Environment (INRAE), Grenoble, France, 17 Interuniversity Laboratory of Atmospheric Systems (LISA), Paris-East Cre´ teil University (UPEC), Cre´ teil, France, 18 UGA Design Factory, Grenoble Alps University (UGA), Saint-Martin d’Hères, France, 19 Department of distance education, French Research Institute for Development (IRD), Marseille, France, 20 French Research Institute for Development (IRD), Marseille, France, 21 HEMiSF4iRE Design School, Catholic University of Lille, Lille, France, 22 French Polar Institute (IPEV), University of Western Brittany (UBO), Brest, France, 23 Research unit “Chrono-environnement”, French National Centre for Scientific Research (CNRS), University Burgundy Franche-Comte´, Besanc¸on, France, 24 National Centre for Meteorological Research (CNRM), French National Centre for Scientific Research (CNRS), Toulouse, France, 25 Planetology and Astrophysics Institute of Grenoble (IPAG), Grenoble Alps University (UGA), French National Centre for Scientific Research (CNRS), Grenoble, France, 26 Research Unit “RiverLy “, French National Research Institute for Agriculture, Food and Environment (INRAE), Villeurbanne, France, 27 Research unit “Lille Economics Management” (LEM), University of Lille (UDL), French National Centre for Scientific Research (CNRS), IESEG School of Management, Lille, France, 28 Research unit INTERTRYP, University of Montpellier (UM), French Research Institute for Development (IRD), Agricultural Research Centre for International Development (CIRAD), Montpellier, France, 29 Grenoble Informatics Laboratory (LIG), Grenoble Alps University (UGA), Saint-Martin d’Hères, France, 30 Research Unit “Recycling and Risk”, Agricultural Research Centre for International Development (CIRAD), Montpellier, France, 31 Centre of Education and Research on Mediterranean Environments (CEFREM), University of Perpignan Via Domitia (UPVD), Perpignan, France, 32 Agroecology, genetics, and tropical livestock systems (ASSET), French National Research Institute for Agriculture, Food and Environment (INRAE), Petit-Bourg, France, 33 Office of the Director General in charge of Research and Strategy, Agricultural Research Centre for International Development (CIRAD), Montpellier, France, 34 Laboratory for the study of Soil-Agrosystem-Hydrosystem interactions (LISAH), University of Montpellier (UM), French National Research Institute for Agriculture, Food and Environment (INRAE), French Research Institute for Development (IRD), Institute Agro, Montpellier, France, 35 Behavioural Ecology and Fish Population Biology (ECOBIOP), University of Pau and the Adour Region (UPPA), Energy Environment Solutions (E2S), French National Research Institute for Agriculture, Food and Environment (INRAE), Saint-Pe´ e-sur-Nivelle, France, 36 Genetics, Physiology and Livestock Systems (GenPhySE), French National Research Institute for Agriculture, Food and Environment (INRAE), University of Toulouse, ENVT, Toulouse INP, Castanet Tolosan, France, 37 Laboratory of functional ecology and environment (ECOLAB), French National Centre for Scientific Research (CNRS), Paul Sabatier University (UPS), Toulouse Institute of Technology (Toulouse-INP), Toulouse, France, 38 Laboratory of Marine Environmental Sciences (LEMAR), French Research Institute for Development (IRD), French National Centre for Scientific Research (CNRS), French Research Institute for Exploitation of the Sea (IFREMER), University of Western Brittany (UBO), 29280 Plouzane´, France, 39 ADLIN Science, Evry, France, 40 Max- Planck-Institut fu¨r Astronomie, Heidelberg, Germany, 41 Research unit Espace-DEV, University of Montpellier (UM), Institute of Research for Development, Guyane University, University of Reunion, Montpellier, France, 42 Chemistry and Biology of Metals laboratory (LCBM), Interdisciplinary Research Institute of Grenoble (IRIG), University Grenoble Alps (UGA), French National Centre for Scientific Research (CNRS), Alternative and Atomic Energies Agency (CEA), Grenoble, France nicolas.gratiot@ird.fr Abstract A growing portion of scientists realises the need to not only alert about climate change, but also change their professional practices. A range of tools have emerged to promote more sustainable activities, yet many scientists struggle to go beyond simple awareness-raising to create concrete transition actions. Here we propose a game-based transition support sys- tem MaTerre180’, which has been designed to build scenarios of greenhouse gas (GHG) emission reductions in the academic community. After providing a common scientific back- ground about the context (global warming issue, its causes and consequences) and setting up a challenge (50% reduction of carbon budget by 2030), the participants belonging to the academic community and its governance bodies immerse themselves into fictional charac- ters, to simulate the behaviour of real research groups. The game has been deployed during the year 2021, with six hundred participants from nine countries and 50 cities. Results PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 2 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community explore clear pathways for GHG reductions between 25 and 60%, and a median reduction of 46%. The alternatives allowing the greatest reduction are video communication tools (36%), followed by mutualization of professional activities and voluntary cancellation or reduction, that represent 22 and 14% of reduction, respectively. The remaining 28% of reduction consists of transport alternative, relocation of professional activities, extended duration of some travels, etc. In addition, the analyses pointed out the importance of the guided negotiation phase to bring out some alternatives such as relocation, local partners and computing optimization. An added value of this transition support system is that the information it collects (anonymously) will be used to answer pressing research questions in climate change science and environmental psychology regarding the use of serious games for promoting changes in attitudes and behaviours towards sustainability, and including broader questions on how network structures influence “climate behaviour”, knowledge and the governance of the commons. Modestly, MaTerre180’ offers an innovative game-based transition support system to build scenarios of greenhouse gas (GHG) emission reductions in the academic community. It is not simply a question of moving tokens on a virtual gameboard and a playful adjustment of practices, but rather a question of brainstorming about possible and desirable ways of remodelling research and teaching communities and embracing a new paradigm. After tens of workshops, our results show clear pathways for reaching up to 50% GHG reductions and stress the importance of guided negotiations to bring out alternatives to carbonized activi- ties. This first attempt reinforces our belief that scientific engagement is at the heart of the international development agenda and a key approach to tear down the institutional barriers that inhibit the transformation needed to achieve a more sustainable society. 1. Author summary For the last centuries, humans have upscaled their socio-economic structures and global- ized their interactions; and these unprecedented developments have been largely driven by our capacity to extract energy and resources from the Earth. In developed countries, people live in a carbonized world, where almost unlimited access to fossil resources and derived goods has become the norm. Generations after generations, homo sapiens switched and installed themselves in the ideology of a no limit planet. For some decades now, scientists have warned about the inadequacy between this commonly shared belief and the physical and biogeochemical limits. In simple words, the “carbonized sapiens” now know the threats but miss guidelines to reinvent themselves. 2. Introduction Since 1972, Meadows et al. (1972[1]) were probably the first to point out some major problems faced by humankind and how the exponential growth of population, food production, indus- trialization, pollution, and consumption of non-renewable natural resources, would ultimately lead to an overshoot of the Earth’s capacity. New concepts emerged later such as planetary boundaries (Rockstrom et al., 2009[2]), which were defined as nine parameters whose trans- gressions could lead to catastrophic consequences for humankind, due to the existence of PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 3 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community thresholds triggering abrupt changes. Carbon footprint–as addressed in this role-playing game–is only one measure of our impact on the planet, among others; but it translates our use of fossil fuels in various sectors and its decrease would certainly affect many of the current neg- ative feedbacks of anthropogenic activities, such as loss of biodiversity and shortage of critical resources. Since the Paris agreement on climate change in 2015, and the Intergovernmental Panel on Climate Change Special Report on Global Warming of 1.5˚C (IPCC, 2018[3]), 191 states have committed to set ever more stringent policies of greenhouse gas (GHG) reduction (UNFCCC report, 2021[4]). In this context, the European Union has set the target of achieving, at least, a 55% reduction in GHG by 2030, compared to 1990. On July 8 2021, the European Central Bank took a historic step by announcing, for the first time, the integration of climate change into its monetary policy. Earlier in 2021, the International Energy Agency called on govern- ments to ensure that their economic recovery plans focus on clean energy investments in order to create the conditions for a sustainable recovery and long-term structural decline in carbon emissions (IEA report, 2021[5]). At the global scale, a systemic change through moderate to low GHG emissions can only be reached if both individuals and communities endorse a dual responsibility to inform policy makers and citizens about the threatening situation for humans and life on Earth. It requires action to promote a form of frugality (Vaden et al., 2020[6]) and embody a socio-ecological transition toward low carbon societies (IPCC, 2018[3]; Otto et al., 2020[7]). In France, this dual responsibility is unavoidable since individual actions, such as commitments and financial investments, can at best reach a 45% reduction of GHG emission (Dugast et al., 2019[8]). Changing the individual and collective behaviors of society is quite challenging but is key in implementing efficient public policies. Behavioral science can help in designing tools that pro- mote sustainable behaviors (OCDE, 2017[9]). Besides, serious games (or learning games) have now been developed for decades in various fields, with a common feature which is to not be targeting mere entertainment. These games allow players to develop problem-solving skills in a real-world context, as well as engagement and responsibility (Cheng et al., 2020[10]). In a game-based approach, participants are required to adopt roles with possibly competing inter- ests and various perspectives depending on the consequences of the issue on their character(s), which foster multiple opinions and collaborative argument to reach a common goal (Doerr- Stevens et al., 2011[11]; Guigon et al., 2021[12]). Role-playing games (RPG) have already been used and their efficacy assessed for various environmental issues. Salvini et al. (2016[13]) explored their use to promote sustainable land-use agricultural practices with Brazilian farm- ers. Besides technical knowledge, they observed socio-institutional learning and engagement in collective action for one of the three groups of farmers involved. Meinzen-Dick et al. (2018 [14]) report of games used in a pilot study to improve groundwater resource management in Andhra Pradesh, India. Communities where the games were played were more prone to adopt water registers and rules to govern groundwater, compared to other communities which did not follow the games. In the SECOLOZ game, the impact on ecosystem services as a function of three different farming practices in Lozère, France, has been facilitated between local stake- holders (Moreau et al., 2019[15]). Agusdinata and Lukosh (2019[16]) designed the HomeRUN RPG to decrease the amount of GHG emissions arising from the consumption of food, energy, and water resources at the household level. GHG emissions of academic activities can no longer be ignored. As highlighted by the IPCC (2018[1]), limiting global warming to 1.5˚C or even 2˚C requires a drastic and rapid reduction of GHG emissions that must concern all sectors of activity, particularly in developed countries (Mahlstein et al., 2011[17]). In this respect, the academic world is not an exception (Attari et al., 2016[18]). Besides, cognitive dissonance is high in all spheres and perhaps even PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 4 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community more within the academic world, which can no longer afford to only raise awareness and alarm about the upcoming crisis, but must act as pioneers and embody changes (Schrems and Upham, 2020[19]; Whitmarsh et al., 2020[20]). Defining a robust strategy of emissions reduction implies, firstly, to accurately monitor GHG emissions. In the academic sector, a group of French researchers, named Labos1point5 (https://labos1point5.org/), developed an open-source tool called ‘GES1point5’ to help research units calculate their carbon footprint (Mariette et al., 2022[21]). Monitoring is a first step but it is insufficient to lead to in-depth changes of our professional behaviour (Hulme, 2020[22]). Yet, a growing portion of the scientific community realises the need to not only alert but also change their professional practices. Moreover, according to Attari et al. (2016 [18]), the credibility of scientists and of their warnings is increased when they behave in a non- dissonant manner. According to a study carried out among 6000 people (Labos 1point5, 2020 [23]), 88% of French researchers "completely agree" or "somewhat agree" that the climate emergency requires profound changes in their practices; however, the structural and func- tional framework of the academic sector and the evaluation of academic performances do not favour the emergence of sustainable trajectories. On the contrary, it largely promotes research- ers’ behaviours that lead to high carbon pathways (e.g. international travel, promotion of inter- national network, use of high-technology and unique scientific instruments). Nowadays, whether for conferences, field surveys, highly specific instrument experiments, thesis defences or project meetings, the emissions linked to researchers’ mobilities are an important (and sometimes predominant) contribution of a laboratory GHG footprint (Whit- marsh et al., 2020[20]). In addition, travel practices are inequitably distributed among individ- uals, reaching per instance 10.8 tCO2e (i.e. where all GHG have been converted to an equivalent CO2 greenhouse forcing) per capita on a yearly average for a professor at the Uni- versity of Montreal (Arsenault et al., 2019 [24]) and 7.5 tCO2e at the University of British Columbia (Wynes et al., 2019 [25]). For both locations in Canada, this sole activity corre- sponds to a vast proportion of what the average person from that country emits during a year (19.4 t CO2e/capita in 2019) (Canada, 2021[26]), which is far away from Canada’s commit- ment to reach net-zero emissions by 2050. The use of aircraft is a predominant source of GHG emissions and according to some authors (Wynes et al., 2019[25]), it would not necessarily bring a clear benefit in terms of career development and enhancement of professional relations. A range of tools, of varying degrees of entertainment and constraint, are gradually emerg- ing, but many of them struggle to go beyond simple awareness-raising to create concrete tran- sition actions (Galeote, et al., 2021[27]). In France, as in many other countries, a growing number of researchers organize themselves to change their work habits and embrace more sus- tainable practices; a trend that was accelerated due to the COVID pandemic crisis and the increase of video communications. Some alternatives should be in place to enlarge the scien- tific community involved, but also to provide an overall vision of possible pathways of GHG emission reductions. Current approaches include incitative measures (carbon tax, ecological money), regulatory measures (carbon quotas, green charter, carbon offsetting) and several gamification approaches, such as for France: (i) The Climate Fresk, a 3-hour collaborative workshop to understand the scientific bases of climate change and start taking action (https:// climatefresk.org/). (ii) ClimaTicTac, a French collaborative strategy board game for players above 9 years old developed by French CNRS and CEA scientists (https://climatictac.ipsl.fr/). (iii) 2 Tonnes, a 3-hour workshop to find solutions to reach the desirable footprint of 2 tons equivalent CO2 per year per person, an objective to be reached by 2050 to respect the commit- ments of the Paris Agreement, that is to say to keep the increase in global temperature at a level below 2 degrees (https://en.2tonnes.org/atelier), and (iv) Carbon Lean, a card game for PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 5 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community players above 8 years old, to discover their own carbon footprint and try to minimize it (https://www.carbon-lean.com/). The latter can take the form of serious games, which simulate multi-actor systems for tackling the complexity of environmental issues and their interplay with many other domains (Oliver, 2016[28]). In the context of climate change, digital serious games have been used for almost forty years (Robinson and Ausubel, 1983[29]). In their literature review, comprising tens of gamified approaches, Galeote et al. (2021[27]) showed that serious games stimulate cognitive engage- ment, affect the perception of climate change-related topics and behavioural engagement with others, by combining learning and entertainment. Serious games create a sphere of thinking around a complex topic while maintaining a playful atmosphere. As players, participants then embody positions or roles that are not necessarily their own, and relate more easily with issues that do not concern them directly or by which they did not think they were concerned. More- over, serious games generate dynamics of opposition or cooperation involving the players’ emotions to immerse them further in their character and promote the players’ empathy towards roles different from their real-life conditions (Wiemeyer et al, 2016[30]). They favour moments that create links and encourage sincere exchanges. According to Gee (2008) [31], serious games need to be moderately funny or "pleasantly frustrating" to be serious enough. This characteristic makes the adaptation of serious games on the theme of climate change or socio-ecological concerns perfectly appropriate. Indeed, these topics are surely some major issues of our time, and at the same time the most postponed ones. In this context, there are more and more serious games being set up to raise awareness on these issues among the vari- ous social, political and economic stakeholders (Onencan et al., 2016[32]; Terti et al., 2019[33]; Undorf et al., 2020[34]). In this perspective, we developed MaTerre180’ (i.e. MyEarth180’), a transition support sys- tem including a game-based participatory tool, that aims at raising awareness regarding the carbon footprint of the academic world, and identifying ways of reductions through social interactions. MaTerre180’ particularly focuses on the predominant proportion of air travel in the academic carbon footprint, but also includes other means of transportation (train, car or boat for oceanographic surveys) as well as additional sources of emissions such as numerical simulations and the access to highly technologic and unique scientific instruments (e.g., parti- cle collider). MaTerre180’ goes beyond the mere framework of learning by first identifying solutions, then embracing action and bringing to light concrete solutions to reduce academic GHG emissions. After a general description of the timeline, materials and methods, results focus on the anal- ysis of the eighty-five game-based phases played to date. These games have been analysed in order to discuss the applicability of the suggested solutions for GHG emission reduction within the academic world. In particular, it has been possible to assess the robustness of the proposed alternatives through indicators of their spontaneity and popularity. Finally, we ques- tioned the indicators used to measure academic performance and their consistency with the GHG emission reduction objectives in order to open discussions on the possible and most effective ways to implement the proposed strategies. 3. Material and methods a. Ethics statement All aspects of the experimental procedures were reviewed and approved by the “scientific board” of the French National Research Institute for Sustainable Development (IRD-France, approval n˚ D2S-2022-002). All participants gave consent to the facilitators prior to their par- ticipation: once the online session was opened, the facilitator of each table asked each PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 6 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community participant of the workshop for the right to record the videos as a source of raw data for further non-profit research. When the agreement was not obtained for all individual participants, the session was not recorded and the corresponding table was not considered for further analysis. When the agreement was obtained the session was recorded and the facilitator notified it by signing a letter agreement. We do remind that each participant role-play two fictive characters; no personal information on individuals were collected, only on the actions of their fictive char- acters during the game. b. MaTerre180’, a game-based participatory tool MaTerre180’ is a Transition Support System organised in four distinct phases, through which an academic institute/group will seek to change the organisation of its academic work to reach a target GHG emission reduction. Fig 1 summarises the timeline. The deployment of MaTerre180’ lasts 180 minutes (+ 30 minutes debrief time). It runs over two half-days, to help the participants gain sufficient introspection and encourage their cognitive engagement. As an adaptation to the COVID pandemic, MaTerre180’ has been designed to be deployed online, which proved to be particularly useful for the massification and the digitization of this game- based approach. In this paper, the analysis focuses on the role-playing phase of the MaTerre180’ workshop (phase 3 in Fig 1). Each MaTerre180’ individual workshop aims to gather a facilitator, six participants, one of them playing the role of team leader, and an advisor. Phase 1: The awareness-raising phase This first phase intends to build a common background on the topic among participants, and to offer them the opportunity to know each other, a key prerequisite before the further discus- sions and negotiations. Phase 1 is based on a set of documents containing general ecological statements: the crossing of four of the nine global limits (Rockstrom et al. 2009[2]; Steffen et al. 2015[35]) and the theory of the doughnut economy (Raworth, 2012[36]). Then follows a more specific section on climate change, with an overview on global temperatures (https:// showyourstripes.info/) and their possible evolution in France (Bador et al, 2017[37]). The rest of the awareness-raising documents deal more specifically with the academic world, presenting the carbon footprint of some French research groups (IGE, ISTerre and LOCEAN), the impact of some research activities at the individual scale (Berthoud et al., 2019[38]) and the results of Fig 1. Timeline of the MaTerre180’ Transition Support System. Each workshop is composed of four phases to raise awareness (phase 1), make some introspection (phase 2), participate in a role-playing serious game (phase 3) and debrief about results and postures (phase 4). https://doi.org/10.1371/journal.pstr.0000049.g001 PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 7 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community the survey on academic practices and awareness "Les personnels de la recherche face au change- ment climatique" conducted by Labos 1point5 (Labos 1point5, 2020[23]). Emerging initiatives in some French research groups are then presented. The awareness-raising phase ends with a debrief time for sharing feelings, reactions, personal experiences and opinions through discus- sions. The next phases of MaTerre180’, including the role-playing phase, are also introduced during this first 90-minute session. Phase 2: The intersession phase Participants are invited, in the few days between the two sessions, to compute their personal carbon footprint with an open access simulator (https://avenirclimatique.org/micmac/ simulationCarbone.php). They also familiarise themselves with the two characters they will play during the role-playing (i.e. game-based) phase, each related to a technician, researcher or professor profiles (see below). Phase 3: The role-playing phase During the role-playing phase, five out of the six participants play the roles of two different characters resulting from a fictitious research group. The sixth participant takes on the role of team leader, which will be detailed hereafter. At the time of the study, 12 virtual research teams, each composed of ten characters, were available to simulate groups working on various topics with distinct approaches (laboratory experiments, numerical simulations, field surveys. . .). Each of them has its own characteristics and has been inspired from a real research group. In particular, the starting emissions of each virtual research team was inspired from real research group emissions computed with the GES1point5 online tool, designed specifically for research groups (Mariette et al., 2022[21]). At that time, both in GES1point5 computation and in the MaTerre180’ virtual teams, the GHG emissions associated with purchases of research groups (whether for services, scientific equipment or consumables) were not considered yet. With this limit, the diversity of emissions of the virtual teams is representative of real research groups, representing different topics, institute policy, or approaches. Given the wide diversity of the games, most participants could choose a familiar research environment, which they generally did. Table 1 lists the different virtual teams considered here, the team’s initial GHG footprint and some keywords related to the scientific topics addressed. Their full description is available at https://materre.osug.fr/-Les-jeux-. Each participant chose two cards describing his/her fictive characters and their respective activities and emissions. The set of 10 characters per virtual team includes senior and junior permanent researchers, PhD and postdoc students, engineers, technical and administrative staff. The description includes their links with the other team members, their academic reputa- tion and lastly, their "ecological awareness profile". There are five types of "ecological aware- ness profile", ranging from a person fully concerned about climate change and already involved in collective actions (profile "Time for actions"), to someone considering that his/her career and duties justify a high carbon footprint (profile "I make the difference"). A game facil- itator is in charge of animating the game, and an advisor (ideally chosen outside of the aca- demic community) brings his/her external vision on the discussions and comments on the final results of the negotiations. In total, eight people are involved during the role-playing phase: the game facilitator, five participants that embody the 10 characters, one participant act- ing as team leader and one advisor, which ensures rich and open-minded social interactions. In case of registered participants not showing-up during the role-playing phase (or unable to attend), the game can be played with down to four participants (instead of six), with some PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 8 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community Table 1. List of the 12 virtual teams with their characteristics. Name of the virtual team Climatology Geophysics Earth Dynamics Environment International Joint Laboratory Society and environment Ocean & Climate Computer science (Informatics) Water Resources Development & Environment Air quality Technology & transition Initial GHG footprint (sum in tCO2e/year for ten characters) 42.0 62.0 43.5 48.0 78.0 68.0 70.0 58.0 63.0 53.0 61.0 63.0 Topics and keywords Climate change, local field studies, glaciers, snow science Earthquakes and volcanoes, near and far field studies, databases, modelling Near and far field studies, geochemistry, partnerships with southern countries Environmental sciences, geochemistry, mineralogy, unique instrument, near and far field studies International laboratory, partnerships with southern countries (e.g. in South-eastern Asia), oceanography campaigns, numerical modelling Sociology, anthropology, ecology, near and far field studies, collaborations with Southern partners Oceanography, high sea missions, high performance computing Parallel programming, artificial intelligence, image processing Hydrology, critical zone, field studies (e.g. in Patagonia), with strong partnership with European partners (e.g. France and Germany) Near and far field studies Geochemistry, near and far field studies, biological and chemical analysis Automation, signal processing, control https://doi.org/10.1371/journal.pstr.0000049.t001 participants playing up to three characters and the team leader. For each virtual team, a virtual board is initially set with tokens representing the individual emitting activities of the 10 char- acters (see Fig 2A and Fig 2B). The surface area of the tokens is proportional to the GHG emis- sion (Table 2) and labelled with a specific icon that symbolises the corresponding activity. The activities considered in the different virtual teams are listed in Fig 2B. They will be further referred to as “emission motives”. After 25 minutes of introduction and presentation of the board, characters and tokens, the strictly speaking role-playing phase takes place in three sub-phases: a free negotiation phase (20 min), a phase of publication of results of research funding applications (about 10min), and a guided negotiation phase (25 min). The remaining 10 minutes of phase 3 are dedicated to short debriefs by the advisor that will be extended in phase 4. The objective for the team during the negotiation phases is to perform their research while reducing the carbon footprint of their virtual team to a given target of fifty percent (50%). During the “free negotiation” sub-phase, the virtual characters played by the participants discuss how to reduce by half the GHG footprint of their virtual research team. Each decision leads to an action: the game facilitator moves tokens on the virtual play mat, in or out of the game board and writes down the suggested alternatives through the digital interface (Fig 2A). Tokens can be substituted by others of smaller sizes, for instance if an intra continental (or domestic) travel by plane is substituted by a train journey. All proposed alternatives are eligible as long as they are accepted by the game facilitator, and co-opted by the participants and the advisor. The free negotiation phase ends by a short debriefing (5–10 minutes) during which the mid-term GHG footprint is presented by the advisor. The advisor also comments on the PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 9 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community Fig 2. A. Digital interface used during the role-play phase. Example for the geophysics research team. The upper left-hand panel is the bank of tokens, the lower left-hand panel is the project’s related tokens, the right hand panel is the area for low carbon alternatives. All research teams’ interfaces are freely available from http://51.178.55.78/MT180/ mt180.htm (the digital interface is coded in javascript). B. Emission motives considered in the 12 virtual teams. https://doi.org/10.1371/journal.pstr.0000049.g002 negotiations, shares his/her feelings and motivates the team to go beyond the efforts already undertaken. The funding application sub-phase then begins. Before the free negotiations sub-phase, the characters were given the possibility to apply for French (French National Research Agency, ANR) or European (European Research Council, ERC) research funds. Each application has a Table 2. Token sizes, related CO2e emissions and corresponding characteristics of emission sources considered so far (Mariette et al., 2022[21]). Details on tokens can be found in appendix A. Token Size Small Medium Large X-Large CO2e emissions (in kg) 20 100 500 3000 Characteristics of emission sources 500 km journey by train 500 km journey by car 2500 km journey by train Short and medium-haul journey by plane 300,000 hours of CPU calculation 1 day of coastal ship mission Long-haul journey by plane 1,800,000 hours of CPU calculation 3 days of high-sea ship mission https://doi.org/10.1371/journal.pstr.0000049.t002 PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 10 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community ⅙ probability of being awarded, close to the current real-life situation in France. Handling such projects implies additional travels that were estimated at 4.0 and 8.0 tCO2e per year for French and European projects, respectively. During the research funding application sub- phase, the results of the applications are published and presented by the facilitator. The success (or failure) of project application is determined by simply rolling a digital dice. Additional tokens are then granted to the successful characters for each awarded project and displayed on the playmat, so that the GHG footprint of the team is increased. Thirdly, the “guided negotiation” sub-phase led by the team leader takes place. He/she man- ages the negotiation phase as a research group leader and is free to choose his/her management strategy (authoritarian, consensual, persuasive. . .). This guided negotiation phase is also timed and lasts 25 minutes. At the end of the three sub-phases, the final GHG footprint is presented and a debriefing period starts. In MaTerre180’ transition support system, the role-playing phase allows participants to put their own research activities and professional constraints into perspective. Working in groups stimulates context-specific abstraction and active experimentation (Morris, 2020[39]). Phase 4: The debriefing phase This last 30-minutes phase closes the workshop. During the debriefing phase, the advisor gives his/her opinion on the suggested alternatives, on the way the characters were played and on the highlights of the role-playing phase. The team, the facilitator and the advisor come back to the highlights, share their opinions on the game-based phase and discuss the relevance and robustness of the proposals made to reduce the research team GHG emissions. c. Database management The role-playing can take place in a classical–i.e., physical–way around a table with all the material previously prepared (game board, character cards, tokens). The role-playing can also be performed online on an open access digital interface (Fig 2A and http://51.178.55.78/ MT180/mt180.htm). In the digital interface, game information is recorded automatically. Each action (e.g. removing a token) is associated with the name of the character to whom the token belongs, the motive for the removal of the token and its value in kg CO2e. Some additional information concerns the phase of negotiation (free or guided) during which the action was played, and whether the token was attributed as a success to a research project application (French or European projects), the name of the alternative to which the token was moved, the reduction in kg CO2e induced by this alternative and the time in seconds at which the token was last moved. Each record is then concatenated in a database to group together all the games that have been played. Four meta information are thus added to identify individual games. Lastly the cat- egory of alternatives (see section on “alternative categorization” below) is specified for each of them. The database obtained is then cross-referenced with another one containing informa- tion specific to each virtual team as described in Table 1 (initial CO2 balance, characters, psy- chological profiles, etc.) for further analysis. This makes it possible, for example, to analyse the results by table, by character, by sessions of the workshop, or by alternatives, in order to pay attention to specific points and decision processes. d. Alternative categorization As mentioned above, the suggested alternatives that emerged were expressed freely by each individual participant. They cover a rich and varied lexical field that had to be categorised in PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 11 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community order to analyse them. These alternatives (translated in the appendix B from French to English) were classified in nine categories that were neither too general nor too specific in order to obtain a fair balance in the information provided. This categorization stems from reading the recorded games by some experts, which consequently involves a degree of subjec- tivity. Categories are described in Table 3. e. Studied parameters i. Trajectories of the different games, in terms of GHG footprint. For each workshop session, we look at the evolution of its GHG footprint according to the modifications (increase or reduction) of the absolute quantity of emissions Qi j in tCO2e, where subscripts refer to each specific sub-phase j and superscripts to the individual workshop session number i. Here, the potential emissions added or removed during the game, linked for instance to new funded projects or to behavioural changes, are taken into account in Q (e.g. using train instead of aircraft for a domestic journey both introduces several tokens of 20 kgCO2e for the train, the number depending on the distance, and removes the 500 kgCO2e token for the plane). • Initial time (j = 0): the initial carbon footprint of the virtual team is equal to the initial GHG emission assigned to each game (see Table 2): CFi 0 ¼ Qi 0 • After the free negotiation phase (j = FN): the new carbon footprint CFi FN is obtained by sub- FN that were proposed during the free negotiation phase tracting the emission reductions Qi CFi FN ¼ Qi 0 (cid:0) Qi FN Table 3. Description of the alternatives’ categories identified so far. Alternative category Description Video communication Mutualization Reduction/cancellation All telecommunication activities between people, whether or not there is interaction. This includes video conferencing/communication, teleworking, e- learning such as Massive Open Online Courses (MOOCs), webinars, etc. Pooling of a large diversity of activities. It includes the use of the terms: mutualization, merging, combination, pooling, association, grouping, etc. Covers voluntary reduction of activity. It includes the words: cancellation, deletion, reduction, halving, etc. Train / public transportation Contains all plane or car trips replaced by train, long-distance buses and all types of public transportation. Relocation Duration extension IT (Information Technology) optimization Other Local Partners Brings the location of an activity at a closer distance, for example by preferring regional conferences or local field areas. This can be associated with the use of public transportation. The words used by participants can be: relocation, bringing closer, regional, local, etc. Includes extension of the time spent on-site after travelling to avoid returning to the same place several times, or combination of several missions. Can sometimes be related to mutualization. This includes the terms: extension, expatriation, prolongation, long, duration, etc. Any solution that aims at reducing the energy consumption of intensive calculations, for example by making the codes less complex and/or better optimised. It covers the words: calculation, optimization, computing, data, etc. Includes some hardly classified alternatives and some original but infrequent ones. For example, the use of sailing boats for missions at sea, volunteer work or carbon offsetting inspired by Miyawaki forest restoration methods, etc. Explicitly cite some local partners from foreign countries to mutualize some activities https://doi.org/10.1371/journal.pstr.0000049.t003 PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 12 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community • After results of ANR/ERC project calls (j = ANR/ERC): depending whether research projects ANR=ERC can be added to the carbon footprint before are granted or not, an emission surplus Qi the guided negotiation phase: CFi ANR=ERC ¼ Qi 0 (cid:0) Qi FN þ Qi ANR=ERC • After the guided negotiation phase (j = GN = f): the final (index f) carbon footprint is calcu- lated by subtracting the additional emission reductions Qi GN suggested CFi GN ¼ CFi f ¼ Qi 0 (cid:0) Qi FN þ Qi ANR=ERC (cid:0) Qi GN • These absolute CF can be converted into a cumulative relative reduction R, for the corre- sponding phase j, using: Ri j ¼ CFi 0 j (cid:0) CFi CFi 0 ii. Alternatives and motives: Frequency, spontaneity and intensity of reductions. We also consider the amount of CO2e avoided from the emission motive m to the alternative a. This allows us to describe in more detail pathways of GHG reductions for each emission motive and thus to deduce the total amount of GHG avoided by each alternative. It will also help to describe whether the emission motives are removed to alternatives or retained in the final GHG footprint of the team. We define the frequency of a given alternative (see Table 3) as the ratio between the number of games that have used this alternative and the total number of games. For motives (Fig 3) a weighted calculation of the frequency of appearance is applied, since games present various initial types and numbers of activities. Then, the spontaneity of the alternative (respectively motive) preferentially chosen (respec- tively removed) is defined as the minimum time before it first appears (respectively, is removed) in the game. This minimum time is then averaged over all games for each variable to deduce its average spontaneity. Finally, we are interested in the GHG reduction intensity caused by an alternative or motive, i.e. the ratio between the total absolute reduction and the number of tokens moved. This allows us to estimate the ability of an alternative or the reduction motive to decrease the team’s GHG footprint more or less efficiently. Thus, the more this ratio tends towards 3000 kg CO2e per token (activity of maximum CO2 emission for X-Large token, as presented in Table 2), the more efficient the variable considered is, in terms of reduction intensity. 4. Results From November 6th 2020 to June 18th 2021, 85 workshop sessions brought together more than 600 participants (mostly academic professionals) from nine countries and more than 50 cities. a. Alternative categorization Fig 3 summarizes the categorization of alternatives in the form of a histogram, showing the fre- quency of each category in the proposed alternatives, and a pie chart for the relative contribu- tions of lexical items to a given category. In total, 407 different alternatives were expressed; some of them being considered by many participants, so that the total individual number of actions (move of tokens) performed to PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 13 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community Fig 3. Classified alternatives by categories, and pie chart of the relative contributions of lexical items to each category. All lexical items are reported in the appendix. https://doi.org/10.1371/journal.pstr.0000049.g003 reduce carbon footprint was 2241. Videocommunication receives wide approval (35%) by most participants. This alternative is seen as an easy way to continue to engage in meetings, confer- ences and even PhD defense committees, or even replace in person training by remote or virtual sessions, while reducing their carbon footprint. It is worth noting that this alternative can also offer additional benefits such as reduced costs and personal life constraints while improving diversity, equity and inclusion (Skiles et al., 2022[40]). Replacing air flights by other public transportation options is the second most frequent suggestion (21%), although this option appeared sometimes limited for activities requiring the transport of instruments for instance. The third alternative–Mutualization (18%)–also presents the largest lexical field (~25% of the lexical items) since it requires a degree of interaction between two, or more, characters, and thus covers a great lexicological plurality. More than one hundred (101) different wordings of this alternative were voiced by participants. It gathers a variety of options such as combined trips for project meetings and/or conferences and/or field campaigns for a single individual, selecting one representative to attend a conference for a research group, training one person on many instruments for a field campaign or choosing similar mission fields. Reduction/cancella- tion (10%) has been chosen mostly for conference attendance (once every two years; limitation of either the number of people attending from the same research group or the number of inter- national conferences per individual for instance), and voluntary decrease of research activities (waiving field trips, numerical computations, grant applications). Overall, these four categories account for more than 84% of all the alternatives proposed. PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 14 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community b. Trajectories of the different game tables The GHG emissions trajectories are first presented through the absolute reduction of GHG of each game table (Fig 4A, CF); then, the relative reduction is shown (Fig 4B, R) to facilitate intercompari- son given that not all the game tables/teams start with the same initial emission level (Table 1). Fig 4. Virtual GHG footprint trajectories. (a) Absolute and (b) relative GHG trajectories for 85 game tables coloured by virtual teams. The horizontal solid black line represents the 50% reduction goal. https://doi.org/10.1371/journal.pstr.0000049.g004 PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 15 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community The x-axis reports the four successive sub-phases of the role-playing game, as previously detailed in section 3b, phase 3, namely the initial footprint of the different virtual teams, the GHG footprint decrease after the free negotiation phase, French and European project grants, and the final reduction after the guided negotiation phase. Beyond the general decreasing tra- jectory of all broken lines observed in Fig 4A and 4B, we can emphasise a strong variety of ini- tial budget (ranging from 42 tCO2e to 78 tCO2e per virtual teams), and of games trajectories. Overall, in all workshop sessions, all virtual research groups managed to reduce their car- bon footprint after the free negotiation phase. The variability of the final emissions at the end of the games overpasses the variability of initial GHG footprint, which clearly highlights the importance of the interactions between players during the game. To compare the trajectories of the different tables, we displayed the relative reduction in GHG footprint (Fig 4B). Here, all tables start from 0% and reach between 5% and 45% reduc- tion at the end of the free negotiation phase. As previously pointed out in Fig 4A, the successful application to French or European funding programs increases some of the footprints, some- times wiping out the efforts that have been made during the free negotiation (e.g. one game of the Environment virtual team in brown). Finally, the range of reduction after guided negotia- tion is narrowed down to a final average reduction of 44% and a median of 46%. The variability between games is high, the less efficient groups of participants reducing by 25–30% their emissions, while the most efficient ones reach reductions close to 60%. Despite the variety of situations, the virtual reductions obtained during all games are promising and show that substantial opportunities for GHG emissions reduction exist within the academic world. The high variability between games suggests that the reduction does not depend on the intrinsic characteristics of the twelve virtual teams (initial carbon footprint, distribution of motives, psychological profiles, etc.), but rather on the way participants of a game interplay through the ten characters they embody. To go further in the analysis, it is interesting to show the density distribution of the final relative GHG reductions, which is represented in Fig 5. On this figure, no colour clusters are observable, suggesting that the final GHG footprint of virtual teams are approximately evenly distributed. For example, among the twenty games of the “Society and Environment” virtual team (blue squares), there is one at each extreme (-27.5% and -62.5%): the final result therefore depends more on social interactions that have been created during the game between participants, than on the characteristics of virtual teams played. However, in addition to this observation, there is a threshold effect related to the target of -50% proposed to win the game: before this target, the distribution increases gently and gradually, whereas after -50%, it suddenly drops. The target seems to affect the result obtained so that, as long as the target is not reached, the participants imagine solutions to reduce by 50% their emissions, but as soon as the target is reached, there is no reason to do more than necessary. The distribution peak, observed for a value of 50%, seems to indicate that the moti- vation of the participants is highly driven by the objective to be reached. Another interesting aspect concerns the impact of additional fundings on the final GHG footprint. In Fig 5, games that did not receive additional fundings (i.e. additional GHG emis- sions) have an average reduction of -46.5%, logically beyond the ones that were overloaded by additional emissions. For games receiving additional fundings, the corresponding additional GHG emission averaged 12.8%. If participants were not influenced by these “penalties” the reduction of GHG emission should be around -33.9%, which is actually not the case. After the guided negotiation phase, the average GHG emission reduction was established at -42.8%. It means that corresponding participants made a substantial effort (+8.9%) to reduce their foot- print and tentatively reach the targeted -50% of reduction. It is worth noting that none of the games with additional funding overpasses the target, while 14 of the 64 games without addi- tional fundings overpass the target. PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 16 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community Fig 5. Density distribution of the final GHG reduction. It synthesised data presented in Fig 4B, for the 85 game tables colored by virtual teams. Black arrows indicates the -50% target and the mean reduction of games that did not obtained projects fundings (46.5%), The blue arrows indicate the mean reduction of games with funding (adding ~12.8% to the initial emissions) which could have been expected to achieve 33.9% but actually reached 42.8% reduction. https://doi.org/10.1371/journal.pstr.0000049.g005 c. Alternatives chosen and motives The previous section indicates that the interaction between the participants and the resulting synergies predominate in the achievement of the reduction objective. However, are the alter- natives chosen by the participants of the different games the same or, on the contrary, are they very diverse and dependent on the synergies specific to each game table? To answer the question, the games were also analysed and compiled to emphasise the alter- natives selected by participants, in the nine categories detailed previously (Table 3) and catego- rised in Fig 3. Results are reported in Fig 6. The predominant alternative (36.1%) is the use of video communication tools. It is followed by the mutualization of some professional activities (22.3%) and by voluntary cancellation or reduction of research activities (14.4%). Train (6.9%), relocation (4.9%) and duration exten- sion of journeys (4.7%) contribute a smaller part to the total virtual reduction. Finally, local partners (3.0%), IT optimization of numerical calculations (2.0%) and others (2.2%) account for a small share of the virtual emission reduction. Overall, almost 80% of the reduction is achieved through four categories of alternatives. Reduction of the GHG footprint through the implication of “local partners” category is believed to be underestimated, probably as a result of mixing with the mutualization category. The relatively low effect of IT optimization is attrib- uted to the small fraction of emissions from computer simulations present in the 12 virtual teams considered. At a global scale, IT optimization is probably much more important. Fig 7 shows which alternatives were chosen for each of major research activities, their corre- sponding alternative proportion, and how much GHG emissions were reduced. PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 17 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community Fig 6. Repartition of the total GHG reduction by categories. The GHG reduction is, by average, 44% of the GHG initial footprint. The alternative categories are the ones expressed by participants and synthesised in Fig 3. https://doi.org/10.1371/journal.pstr.0000049.g006 Video communication (blue bars) is an efficient factor to reduce GHG footprint for six emission motives, by replacing physical meetings for conferences, projects, juries (PhD, staff recruitment, etc.) as well as training, institutional and expertise meetings by some distant video interactions. Field trips (on the continent or at sea), which are highly contributing to GHG footprint, are most often mutualized. In general, the alternatives are dependent on the motives. A diversity of alternatives is required to maximize the reduction, which emphasizes the complexity and richness of interac- tion between participants. Fig 8 shows in more detail the distribution of GHG emissions and pathways for reductions. The grey vertical bars and colored bands are proportional to the global GHG emissions for the 85 games considered. This Sankey diagram complements the information given in Fig 7. It becomes clearer why the total emissions from conferences are predominant: it is also the larg- est share of the initial distribution. Some motives appear to be difficult to substitute, for instance intensive computing and sea cruises, while others seem easier to reduce, juries in particular. PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 18 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community Fig 7. Absolute GHG reduction distribution. The reduction is subdivided by alternative categories depending on the emission motives: air travel to reach a conference, to meet for a project, for field trip, jury, for training, oceanographic campaigns, air travel for institutional meeting, cost of numerical computing, air travel to make an expertise, to access to a large unique instrument. https://doi.org/10.1371/journal.pstr.0000049.g007 Fig 8. Distribution of the GHG emissions from the role-play initial balance to the selected alternatives and the final balance. On this Sankey diagram, the initial distribution of emissions can be seen, to which the emissions generated by the funded French and European projects (resp. ANR or ERC) during the game can be added. The initial distribution according to the motives can be seen in the centre of the diagram. On the right-hand side are the selected alternatives and the remaining emissions. The flow bands indicate the distribution between motives and selected alternatives. https://doi.org/10.1371/journal.pstr.0000049.g008 PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 19 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community d. Frequency, spontaneity and emission intensity As the role-playing phase takes place in two sub-phases of 20 and 25 minutes each, it is inter- esting to look at the influence of the time when the tokens are replaced for a given alternative. Three characteristics are particularly meaningful: first, the spontaneity of an action, i.e. the minimum time of appearance of the variable (motive or alternative); secondly, its frequency of appearance on all the games and finally its reduction intensity in kg CO2e per token. Fig 9 depicts the frequency of appearance of each alternative as a function of its spontaneity. The size of the bubbles is proportional to the reduction effectiveness of the alternative in kgCO2e per token. Overall, four clusters of bubbles can be observed. First is the “video com- munication” alternative, which is very spontaneous (less than 10 minutes for its first appear- ance), very frequent (proposed by 95% of games) and rather effective. Cluster two includes three alternatives, namely “mutualization”, “cancellation” and “train”, which also come fairly early during games and remain fairly frequent but are unequally effective in reducing GHG emission, especially “train” which is rather low as it cannot substitute long-distance air travels. The following cluster is composed of the “duration extension” and “local partners” alterna- tives, which are proposed later and are less popular (around 25% of occurrence) but rather effective in terms of intensity of reduction. The last cluster includes “relocation”, “IT optimisa- tion” and “others”. It arrives very late in the games, on average during the guided negotiation phase (after 30 minutes on average), is infrequent and unequally effective: “relocation” is the most effective alternative, while “IT optimisation” appears to be poorly effective. Fig 10 represents the frequency of each motive removal as a function of its spontaneity. The participation in international conferences is globally the only motive to be withdrawn fre- quently (more than 95% of games played) and getting a high spontaneity (<10min). In con- trast and logically, flight to access to “unique instruments” are the least frequently removed (just over 50% of game tables initially having them), which is understandable as it is the core of Fig 9. Spontaneity of the different alternatives sized by reduction intensity. https://doi.org/10.1371/journal.pstr.0000049.g009 PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 20 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community Fig 10. Spontaneity of the different emission motives removal. It is sized by reduction intensity. The size of the bubbles is proportional to the reduction effectiveness of the motive removal in kg CO2 equivalent per token. https://doi.org/10.1371/journal.pstr.0000049.g010 some research activities and cannot be substituted. Finally, IT optimization is less spontane- ously mentioned (beyond 30 minutes of play). The effectiveness of reduction, represented by the size of circles, is also rather variable, ranging from more than 1500 kg CO2e per token for projects and conferences meetings, to less than 500 kg CO2e equivalent per token for computing. 5. Discussion a. Synergy during the role-play sessions and influence of the target According to Pohlmann et al. (2021)[41], the normalisation of climate-friendly behaviours in a given social group will not occur through the sum of individuals. Gamification thus often pro- vides interactive spaces where reality can be experienced and transformed, which is a rich basis for knowledge creation (Kolb, 2014[42]). Our study shows that most of the variability of the results can broadly be explained by two independent factors: the synergy that was created between the participants during each game and the target that is given to win the game (in our case -50% of GHG footprint). As far as syn- ergy between participants is concerned, an in-depth anthropological and sociological work would be needed to assess the brakes and leverages to GHG footprint reduction (Whitmarsh et al., 2020[20]). An in-depth analysis of this hypothesis in this study goes beyond our scope but is a key perspective for further analysis of the data collected during the games. Focusing on a more quantitative analysis, some interesting elements can be deduced from final GHG footprints (Fig 5). In this figure, the density distribution shows an asymmetry, which corresponds to a threshold effect: below 50% of reduction, the game tables are distrib- uted rather gradually, but once the objective is reached, the density distribution suddenly drops. Thus, as long as the objective is not reached, the participants make all the efforts they PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 21 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community can and as soon as the objective is reached, the participants stop making efforts. The question then arises whether setting a target of 75% would also result in this threshold effect with an average reduction slightly below the target. We may hypothesise that a reduction of 50% finally remains acceptable and reachable, but a target at 75% would probably discourage participants and require more profound and systemic changes of the academic sector practices. It is worth noting that the median of final reduction was about 45% which is believed to be a positive sig- nal for reaching significant reduction of GHG emission in real life. b. Frequency, spontaneity and effectiveness of alternatives Here, our interest was to identify how to articulate the emission motives and the alternatives, as expressed in Fig 8, in order to build realistic scenarios for reducing the carbon footprint of the academic world. Virtual teams in MaTerre180’ are as much realistic as possible to prevent too wide a gap between the game and the real world. Besides, once awareness is raised in the first phase, some individual choices can be made—when possible—without the approval of the employer or much impact on one’s career (such as meal choices, mode of commuting to work), yet have some benefits on the academic footprint in real life although they are not explored in the game itself. In order to analyse the reduction choices made by the participants, it was decided to focus the study’s attention on specific characteristics. To do this, it is impor- tant to understand which emission motives are favoured for reduction and towards which alternatives by looking at the frequency, spontaneity, effectiveness and efficiency of these choices (Figs 9 and 10). However, passing from the virtual space of a role-playing game to the real world of research, may introduce unexpected difficulties due to the current functioning of research, which promotes individual performance and competition (van Dalen, 2021[43]) instead of building bridges toward global sustainability (Irwin et al., 2018[44]). Our results showed that 80% of the GHG reduction was possible thanks to four alternatives, namely video communication, mutualization of means or activities, cancellation of activities and lower carbon emission transportation (train). The use of video communication is the most spontaneous and frequent proposal, which enables the greatest reduction (16.2%), because it can be adapted to a large number of activities, with the notable exception of field/sea campaigns. The spontaneity and efficiency for video communication have probably been pro- pelled by the COVID-19 pandemic crisis that has recently imposed such means of communi- cation due to lockdowns and remote working (Nguyen et al. 2020[45]). Video communication practice had however already been raised within the scientific community as an alternative to conferences (Jordan and Palmer, 2020[46]). Nevertheless, the advantages and disadvantages of virtual conferences are debated. Another suggested option is to attend conferences in person, but to be more selective (see below, cancellation). The second option is the mutualization of activities or means, which also leads to a strong overall reduction of GHG footprint (10.0%) by combining several field trips of different purposes or by delegating specific tasks to limit the number of participants during field/sea campaigns. Yet, experts of oceanographic campaigns consider that a reliable mutualization of onboard activities is an uneasy task. In real life, one can anticipate non-negligible organisational obstacles and an expected resistance of research- ers and their stakeholders (community, hierarchy, partners) for such suggestions. While grouping several activities on a personal basis is not excessively complex, mutualization between colleagues requires a high degree of communication, preparation and trust. At pres- ent, mutualization is not sufficiently recognized by academic institutions to become popular, in view of the time required and the risks involved for careers, in case of failure of uneasily rescheduled campaigns. According to Shove and Walker (2014[47]), individual actions are embedded in institutional, social and infrastructural frameworks, which ensure that climate- PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 22 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community damaging behaviours remain the norm. The academics need to be proactive to shift these norms through more mutualized and frugal research. The third alternative concerns cancella- tion or rationalisation of research activities. It is by nature very simple to be done technically, but seems to be over-represented in our results. The main limitation is the psychological acceptance by participants, in link with social habits and pressures (Gifford, 2011[48]). The lack of institutional recognition of the efforts made and risk-taking by researchers in the case of a cancellation or drastic reduction of field/oceanographic surveys seems also to be a limita- tion. It is the same in the case of limitation to in-person meeting participation. As long as a car- bon quota or any other indicator, based on the sustainability of activities, is not put in place by academic institutions, reducing one’s activity brings at best a saving of time and an improved work-life balance, at worst, a devaluation of research performance and researcher’s recogni- tion. An in-depth analysis of costs and benefits for the society should be considered. The fourth alternative is train travel, which is often mentioned in the literature as a solution for decarbonizing research. However, train travel quickly reaches its limits in the sense that it is neither easily accepted to take the train if several train changes are required or heavy/cumber- some equipment needs to be transported. Trains cannot substitute long-distance air travel. For most regional activities however, train is even very efficient (Ciers, et al., 2019[49]). The train must thus be promoted both as an efficient practice on a regional scale, and as a marker of change in our practices. The remaining 20% of the reduction is made up of solutions that occurred less frequently and were less spontaneous, but which can compensate for the limitations of the first four. Relo- cation, coupled with the use of trains, is thus very efficient as it directly addresses long-distance air travel, particularly for conferences. The extension of the mission duration is similarly very interesting but is proposed more specifically for field trips or sea cruises which allow for more expatriation. Local partners and expatriation are specific to some research groups and topics. Reducing the corresponding GHG footprint will require first to understand people’s beliefs, values and norms, second to engage in-depth discussions between all actors and policy makers to break psychological and other limits (Gifford, 2011[48]). Regarding the emission motives, they are globally withdrawn from the playmat in propor- tion to their initial distribution within the eighty-five tables. Conferences are naturally removed the fastest and most often, but this should not overshadow the other motives for the teams’ emissions, as is often the case in scientific works that consider conferences for the most part. However, this raises the question of the acceptability of replacing a conference with a vid- eoconference or cancelling it, and the valuation of conferences in the research indicators. There are also many motives that can be played on. For example, thesis juries are especially reduced, as they can essentially be carried out by videocommunication, with an associated gain in personal life quality. Conversely, certain motives are under-represented, like oceano- graphic surveys, intensive computing or travel for the use of unique instruments, as they are specific to the activity of the research labs and so more difficult to reduce, which may explain the lower spontaneity and frequency for the latter two. c. Steps and timetable for achieving the -50% target by 2030 The key point now is to consider how to transform the virtual pathways of GHG, expressed during the role play phase, into real measures. In the virtual format, participants detach them- selves from their emotions but have the difficult task of projecting themselves into the skin of a fictional character. Some participants may find it difficult to make this change of posture and to become imbued with the personal motivations, posture and convictions of the embodied characters. The difficulty is even greater when each participant plays two characters, and when PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 23 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community these characters’ behaviour and profile are different from their own (for instance when a PhD student must play a senior researcher). The complexity therefore lies in knowing to what extent the proposals emanating from fictional discussions can be directly transposed into the everyday life of an actor in the academic world. Nevertheless, no justification could discredit an alternative a priori. It is still necessary to encourage their implementation in order to judge their acceptance in the framework of a functional research group. Two main directions for their implementation can be distinguished: First, promoting and recognizing the efforts made by individuals to reduce one’s GHG foot- print would be a preliminary step. One point that came up several times in the discussions dur- ing the debriefing phases was the importance of indicators of academic performance. Indeed, the current indicators encourage productivity and do not take into account the social and eco- logical impact of research and education activities, in particular in terms of GHG footprint. It seems inappropriate to keep the same evaluation criteria for academia in the context of the socio-ecological transition. We know that conferences play a major role in the dissemination of work and the construction of a professional network. They are more important for young researchers compared to senior ones who have already obtained permanent positions and built up their network. Nevertheless, it is the latter who travel the most to participate in inter- national conferences (Wynes, 2019[25]). The evolution of indicators and evaluation criteria therefore appears to be a relevant option for taking better account of criteria compatible with global limits. The second option is for the functional teams to take control of the results. The digital inter- face used during the role-playing phase of MaTerre180’ constitutes a powerful tool for develop- ing new techniques of communication and negotiation between peers. We can imagine that some research groups could take advantage of this transition support system to experiment with various strategies of research projects and define the ones that best balance benefits for society and sustainable GHG footprint. In their exhaustive review, Flood et al. (2018[50]) reported various climate related games or role playing focusing on water management, long term farming or risk disasters; but none of them was dedicated to the academic world and its non-negligible GHG footprint. Knowing the peculiar role of scientists in society, we may hope that the use of a tool such as MaTerre180’ could accelerate a shift in the scientific community and provide a persuasive argument for a broader shift in other sectors. Transition support system could certainly facilitate the transition, but this will depend on our capacity to follow at least two recommendations (Galeote et al., 2021[27]): first, it is impor- tant to promote interventions in emerging and developing countries and to extend the target to young students and more social, political, and economic actors. Secondly, gamification and transition support system techniques should be massive and lead to large data series in order to get statistically robust and unbiased scenarios of reduction. Some collaboration with research institutions with a broad national and international presence, could favorably help for reaching these recommendations. d. Limits and ongoing improvements of MaTerre180’ Several limits of the current set up of the workshop are already identified and will lead to future, improved versions of the workshop. First, the rationale of the workshop was designed before the COVID crisis which has imposed most researchers to drastically reduce their trav- els. However, it is clear that the possibilities and incentives for long distance travel for various reasons (jury, conferences, fieldwork) are currently resuming, and therefore the need for research communities to reflect on how to perform sustainable research is very timely. PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 24 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community Additionally, the COVID crisis has made concrete many of the alternatives discussed during the role-played negotiations (e.g., relocation, online meetings, etc.), which could facilitate dis- cussions and implementations of these options in both virtual and actual research units. Second, the workshops were devised based on CO2e budgets which considered air travel as the dominant source of emission. However, recent GHG budgets from research institutions emphasised that together with air-travel, research-driven purchases are often a dominant source of emissions (Martin et al., 2022 [51]). They are currently being included in the initial CO2e budget and will require different alternatives than the ones envisioned for air travel reduction. Obviously, the complex and international activities of research institutions mean some substantial part of their CO2 budget may still not be captured by the game (e.g., large infrastructures and satellites, Kno¨dleser et al., 2022 [52]). Nevertheless, the goal of the game is to accelerate the emergence and implementation of alternatives allowing to decarbonize research activities, even if all sources and alternatives are not adequately quantified and decar- bonisation itself cannot be the only measure of sustainability. The game can be currently played for research activities located in France (where electricity is mostly decarbonized), since the emission factors associated with activities depend also on the energy mix of each country. Last, a limit of the game may be its tendency to underestimate resistance to alternative ways of performing research. Indeed, the proportion of researchers with low environmental com- mitments may be larger than in virtual teams, and actual research unit heads may not always be very pro-active in negotiating overall reduction of CO2e emissions, as it is assumed in the second phase of the role-play game. Varying the proportion of virtual characters resisting changes and analysing separately games outcome as a function of the personality of the virtual team leaders may be a way forward to assess this potential bias. 6. Conclusion The authors of this study are convinced that the state of scientific knowledge on the current and coming social and ecological crises, caused or enhanced by global warming, is not enough to bring about a systemic and rapid change that is commensurate with the issues at stake (Hulme, 2020[22]). In this context, the academic world is not an exception and must act and embody changes (Attari et al., 2016[18]; Whitmarsh et al., 2020[20]). Mathematically-based methods, such as simply assessing the GHG balance of research activities (e.g. Mariette et al., 2022[21]), are essential but lack the ability to engage deeply all those involved in academic research, from the management to the technical staff, from PhD students to senior researchers. For that purpose, a game-based transition support system, MaTerre180’ (https://materre.osug. fr/), was created to build scenarios of GHG emission reductions in the academic community. MaTerre 180’ reproduces–even at a small scale–a laboratory group where people perform vari- ous duties, are at all levels of their career and can pursue different professional objectives, not necessarily compatible with a path for emission reduction. This tool has been deployed during the year 2021 with around 600 participants. The analysis of all the games played is encouraging and expresses clear pathways for reductions: given a target of 50% reduction, the range of GHG reduction at the end of the game-played phase is between 25 to 60% with a median reduction of 46%, independently of the virtual research team played. Although the game is time-limited, its potential to recreate similar group dynamics as in real life interactions was appreciated by the participants. This result highlights that, virtually, the objective of 50% of GHG emission reduction in 2030 is reachable for the academic world. More in-depth analyses were conducted to understand the dynamics of reduction, the remaining obstacles to endorse a reduction strategy, and to spark all ideas about possible alter- natives. The alternatives allowing the greatest reduction are the video communication tools PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 25 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community (35%), followed by the mutualization of the professional activities and the voluntary cancella- tion or reduction that represents 22 and 14% of reduction, respectively. The remaining 28% of reduction is composed by the use of trains as a transport alternative, the relocation of profes- sional activities, the duration extension of some missions, the optimization of the information technology and other marginal ideas. Our results also confirm the necessity of alternatives adapted to specific research activities: the most effective tool to reduce the GHG emissions from conferences, projects and juries is, as expected, the video communication tool whereas mutualization and duration extension are the most important alternatives for field trips. The initial footprint of the research activities explains the dominance of some activities to the total emission that remains even after the game phase (like conferences). It also shows the small part of cancellation in the GHG emission reduction from the different categories, except for conferences, and thus shows the relatively easy way for academics to reduce their emissions without tremendously affecting their research activities. Finally, the analyses of all the game dynamics, i.e. when, which and how often the alternatives are proposed, show some obstacles to use some types of alternatives and the necessity to have a person that guides the discussion (second part of the game phase): relocation, local partners and computing optimization need more guided discussions than individual choices of video communication, and free discussion for mutualization. Overall, most solutions proposed by the participants are known or have already been experienced by them, but calculating in real time the potential of each of them to reduce effectively the GHG emissions of the team makes them more palatable. Following the informative documents provided ahead of the game phase, the authors expect the game to trigger some behavioral changes at the individual (personal) level (such as the use of decarbonized transportation outside of commuting/business trips, or the decrease in meat consumption). Moreover, the game can contribute to professional structural changes by rais- ing a collective momentum on this issue that warrants collective endeavors on the part of the academic community. Diverse game reviews from the last decade show that the tendency of gamification has only grown in recent decades (Reckien and Eisenack, 2013[53]; Flood et al., 2018[49]; Galeote, 2021 [27]). However, to the best of our knowledge, this is the first time that such a role-playing game is deployed and used to determine the possible scenarios to reduce GHG emissions in the academic world. Gamification is relevant because it allows participants to fail with low con- sequence (Plass et al., 2015[54]). Some further sessions of MaTerre180’ need to be performed in order to consolidate the results and explore the participants sociological synergies during the workshops: changing the 50% target of GHG emission reduction, using virtual teams exploring other field of research, adding other kinds of virtual characters, incorporating the purchases (services, consumables, materials and equipments) into the initial carbon budget, etc.. Additionally, deploying MaTerre180’ at different scales and within varied academic con- texts (universities vs. national research institutes, students vs. university staff) will help to tackle possible biases. Last but not least remains the transition between virtual and real world, i.e. to find the method to adapt the scenarios imagined with the virtual game-based tool into the real world of academic research. This necessarily requires the participation and involve- ment of the institutional governance of research organisations. Supporting information S1 File. Details on tokens and characters of the “geophysics” research team. (DOCX) S2 File. Details on the 2241 expressed alternatives. (DOCX) PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 26 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community S3 File. Gender distribution by sessions: female and male for both facilitators and partici- pants are reported in orange and green horizontal bars, respectively. (DOCX) Acknowledgments The authors would like to thank Ignacio Palomo for reading the draft version and advice. The authors would like to thank beta-testers: Yann Echinard (Sciences Po Grenoble), Isabella Zin (Grenoble-INP), Thierry Lebel (IRD), Geraldine Sarret (CNRS), Florence Michau (Grenoble- INP) and Sigrid Thomas (CEA). Ludovic Eugenot is acknowledged for improving the ergon- omy of the serious game phase and the guideline for facilitators, and Caroline Play for financial support and institutional collaboration with the French National Research Institute for Sus- tainable Development (IRD). A special thanks to Martine Ahrweiller, Lydie Civilleti and their team. We would also like to acknowledge the advisors of sessions and all people who contrib- uted to the deployment. Author Contributions Conceptualization: Nicolas Gratiot, Ge´raldine Sarret, Ge´remy Panthou, Benoıˆt Hingray, Nicolas Champollion, Olivier Aumont, Florence Michau, Cathy Grevesse. Data curation: Nicolas Gratiot, Olivier Dangles, Serge Janicot, Miriam Candelas, Ge´raldine Sarret, Ge´remy Panthou, Benoıˆt Hingray, Nicolas Champollion, Julien Montillaud, Pascal Bellemain, Odin Marc, Ce´dric-Ste´phane Bationo, Loïs Monnier, Laure Laffont, Marie-Alice Foujols, Ve´ronique Riffault, Liselotte Tinel, Emmanuel Mignot, Nathalie Philippon, Alain Dezetter, Alexandre Caron, Guillaume Piton, Aure´lie Verney-Carron, Nelly Bardet, Florence Nozay-Maurice, Anne-Sophie Loison, Franck Delbart, Sandrine Anquetin, Franc¸oise Immel, Christophe Baehr, Fabien Malbet, Ce´line Berni, Laurence Delattre, Vincent Echevin, Elodie Petitdidier, Olivier Aumont, Florence Michau, Nicolas Bijon, Jean-Philippe Vidal, Se´bastien Pinel, Oce´ane Biabiany, Cathy Grevesse, Louise Mimeau, Anne Biarnès, Charlotte Re´capet, Morgane Costes-Thire´, Mariline Poupaud, Maialen Barret, Marie Bonnin, Virginie Mournetas, Bernard Tourancheau, Bertrand Goldman, Marie Paule Bonnet, Isabelle Michaud Soret. Formal analysis: Nicolas Gratiot, Je´re´mie Klein, Marceau Challet, Serge Janicot, Ge´remy Panthou, Benoıˆt Hingray, Nicolas Champollion, Odin Marc, Marie-Alice Foujols, Florence Michau, Se´bastien Pinel. Funding acquisition: Nicolas Gratiot, Olivier Dangles, Julien Montillaud, Anne Delaballe. Investigation: Nicolas Gratiot, Marceau Challet, Julien Montillaud. Methodology: Nicolas Gratiot, Ge´raldine Sarret, Ge´remy Panthou, Benoıˆt Hingray, Nicolas Champollion, Emmanuel Mignot, Nathalie Philippon, Guillaume Piton, Franc¸oise Immel, Olivier Aumont, Florence Michau, Oce´ane Biabiany, Cathy Grevesse. Project administration: Nicolas Gratiot, Odin Marc, Nelly Bardet, Florence Michau, Jean-Philippe Vidal, Se´bastien Pinel. Resources: Nicolas Gratiot, Olivier Dangles, Miriam Candelas, Ge´raldine Sarret, Ge´remy Panthou, Benoıˆt Hingray, Nicolas Champollion, Julien Montillaud, Odin Marc, Ce´dric-Ste´phane Bationo, Loïs Monnier, Laure Laffont, Marie-Alice Foujols, Ve´ronique Riffault, Liselotte Tinel, Alain Dezetter, Alexandre Caron, Guillaume Piton, Anne Delaballe, PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 27 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community Florence Nozay-Maurice, Sandrine Anquetin, Franc¸oise Immel, Fabien Malbet, Ce´line Berni, Nicolas Bijon, Jean-Philippe Vidal, Oce´ane Biabiany, Cathy Grevesse, Louise Mimeau, Anne Biarnès, Charlotte Re´capet, Morgane Costes-Thire´, Mariline Poupaud, Maialen Barret, Marie Bonnin, Virginie Mournetas, Bernard Tourancheau, Bertrand Goldman, Marie Paule Bonnet, Isabelle Michaud Soret. Software: Pascal Bellemain. Supervision: Nicolas Gratiot, Anne Delaballe. Validation: Nicolas Gratiot, Serge Janicot, Miriam Candelas, Ge´raldine Sarret, Ge´remy Panthou, Benoıˆt Hingray, Julien Montillaud, Florence Michau, Jean-Philippe Vidal, Se´bastien Pinel. Visualization: Je´re´mie Klein, Marceau Challet, Pascal Bellemain. Writing – original draft: Nicolas Gratiot, Je´re´mie Klein, Marceau Challet. Writing – review & editing: Nicolas Gratiot, Olivier Dangles, Serge Janicot, Miriam Candelas, Ge´raldine Sarret, Ge´remy Panthou, Benoıˆt Hingray, Nicolas Champollion, Julien Montillaud, Pascal Bellemain, Odin Marc, Ce´dric-Ste´phane Bationo, Loïs Monnier, Laure Laffont, Marie-Alice Foujols, Ve´ronique Riffault, Liselotte Tinel, Emmanuel Mignot, Nathalie Philippon, Alain Dezetter, Alexandre Caron, Guillaume Piton, Aure´lie Verney-Carron, Anne Delaballe, Nelly Bardet, Florence Nozay-Maurice, Anne-Sophie Loison, Franck Delbart, Sandrine Anquetin, Franc¸oise Immel, Christophe Baehr, Fabien Malbet, Ce´line Berni, Laurence Delattre, Vincent Echevin, Elodie Petitdidier, Olivier Aumont, Nicolas Bijon, Jean-Philippe Vidal, Se´bastien Pinel, Oce´ane Biabiany, Cathy Grevesse, Louise Mimeau, Anne Biarnès, Charlotte Re´capet, Morgane Costes-Thire´, Mariline Poupaud, Maialen Barret, Marie Bonnin, Virginie Mournetas, Bernard Tourancheau, Bertrand Goldman, Marie Paule Bonnet, Isabelle Michaud Soret. References 1. Meadows DH, Meadows DL, Randers J, Behrens WW III. The limits to growth-club of rome. 1972. 2. Rockstro¨m J, Steffen W, Noone K, Persson A, Chapin FS, Lambin EF, et al. A safe operating space for humanity. Nature. 2009; 461: 472–475. https://doi.org/10.1038/461472a PMID: 19779433 3. IPCC. Global Warming of 1.5˚C. An IPCC Special Report on the impacts of global warming of 1.5˚C above pre-industrial levels and related global greenhouse gas emission pathways, in the context of strengthening the global response to the threat of climate change, sustainable development, and efforts to eradicate poverty [Masson-Delmotte V, Zhai P, Po¨ rtner HO, Roberts D, Skea J, Shukla PR, Pirani A, Moufouma-Okia W, Pe´an C, Pidcock R, Connors S, Matthews JBR, Chen Y, Zhou X, Gomis MI, Lonnoy E, Maycock T, Tignor M, Waterfield T, editors]. Cambridge, UK and New York, NY, USA: Cambridge University Press; 2018. https://doi.org/10.1017/9781009157940 4. UNFCC. Nationally determined contributions under the Paris agreement. Synthesis report by the secre- tariat; 2021 Sep. United Nations Framework Convention on Climate Change. Report FCCC/PA/CMA/ 2021/8. Available from: https://unfccc.int/documents/306848 5. IEA. Net Zero by 2050: A Roadmap for the Global Energy Sector. Paris: International Energy Agency; 2021. Available from: https://www.iea.org/reports/net-zero-by-2050 6. Vade´ n T, La¨hde V, Majava A, Ja¨ rvensivu P, Toivanen T., Hakala E, et al. Decoupling for ecological sus- tainability: a categorisation and review of research literature. Environ Sci Policy. 2020; 112: 236–244. https://doi.org/10.1016/j.envsci.2020.06.016 PMID: 32834777 7. Otto IM, Donges JF, Cremades R, Bhowmik A, Hewitt RJ, Lucht W, et al. Social tipping dynamics for stabilizing Earth’s climate by 2050. Proc Natl Acad Sci U S A. 2020; 117(5): 2354–2365. https://doi.org/ 10.1073/pnas.1900577117 PMID: 31964839 8. Dugast C, Soyeux A, Castelli B, Cassagnaud C, Ledoux R, Jancovici JM, et al. Faire sa part? Pouvoir et responsabilite´ des individus, des entreprises et de l’e´ tat face au re´ chauffement climatique. Carbone 4 report. 2019. Available from: https://www.carbone4.com/wp-content/uploads/2019/06/Publication- Carbone-4-Faire-sa-part-pouvoir-responsabilite-climat.pdf PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 28 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community 9. OCDE (2017), Behavioural Insights and Public Policy: Lessons from Around the World, E´ ditions OCDE, Paris, https://doi.org/10.1787/9789264270480-en 10. Cheng PH, Yeh TK, Chao YK, Lin J, Chang CY. Design Ideas for an Issue-Situation-Based Board Game Involving Multirole Scenarios. Sustainability. 2020; 12(5), 2139. https://doi.org/10.3390/ su12052139 11. Doerr-Stevens C, Candance R, Boeser E. Using Online Role-Play to Promote Collaborative Argument and Collective Action. The English Journal. 2011; 100, 5, pp. 33–39. JSTOR, http://www.jstor.org/ stable/23047800. 12. Guigon G, Vermeulen M, Muratet M, Carron T. Towards an Integration of the Multi-role Dimension in the Design of Learning Games: a Review of the Literature. In: de Rosa F., Marfisi Schottman I., Baals- rud Hauge J., Bellotti F., Dondio P., Romero M. (eds) Games and Learning Alliance. GALA. 2021; 13134. https://doi.org/10.1007/978-3-030-92182-8_26 13. Salvini G, van Paassen A, Ligtenberg A, Carrero GC, Bregt AK. A role-playing game as a tool to facili- tate social learning and collective action towards Climate Smart Agriculture: Lessons learned from Apuı´, Brazil. Environmental Science & Policy. 2016; 63 113–121, Available from: https://doi.org/10. 1016/j.envsci.2016.05.016. 14. Meinzen-Dick R, Janssen MA, Kandikuppa S, Chaturvedi R, Rao K, Theis S. Playing games to save water: Collective action games for groundwater management in Andhra Pradesh, India, World Develop- ment. 2018; 107 40–53. Available from: https://doi.org/10.1016/j.worlddev.2018.02.006 15. Moreau C, Barnaud C, Mathevet R. Conciliate Agriculture with Landscape and Biodiversity Conserva- tion: A Role-Playing Game to Explore Trade-Offs among Ecosystem Services through Social Learning. Sustainability. 2019; 11 310. Available from: https://doi.org/10.3390/su11020310 16. Agusdinata DB, Lukosch H. Supporting Interventions to Reduce Household Greenhouse Gas Emis- sions: A Transdisciplinary Role-Playing Game Development. Simulation & Gaming. 2019; 50 359–376. Available from: https://doi.org/10.1177/1046878119848135 17. Mahlstein I, Knutti R, Solomon S. Portmann RW. Early onset of significant local warming in low latitude countries. Environ Res Lett. 2011; 6: 034009. https://doi.org/10.1088/1748-9326/6/3/034009 18. Attari SZ, Krantz DH, Weber EU. Statements about climate researchers’ carbon footprints affect their credibility and the impact of their advice. Clim Change. 2016; 138: 325–338. https://doi.org/10.1007/ s10584-016-1713-2 19. Schrems I, Upham P. Cognitive dissonance in sustainability scientists regarding air travel for academic purposes: a qualitative study. Sustainability. 2020; 12(5): 1837. https://doi.org/10.3390/su12051837 20. Whitmarsh L, Capstick S, Moore I, Ko¨hler J, Le Que´ re´ C. Use of aviation by climate change research- ers: Structural influences, personal attitudes, and information provision. Glob Environ Change. 2020; 65: 102184. https://doi.org/10.1016/j.gloenvcha.2020.102184 21. Mariette J, Blanchard O, Berne´ O, Ben-Ari T. An open-source tool to assess the carbon footprint of research. Environ. Res.: Infrastruct. Sustain. 2022; 2 035008. https://doi.org/10.1088/2634-4505/ ac84a4 22. Hulme M. One Earth, many futures, no destination. One Earth. 2020; 2(4): 309–311. https://doi.org/10. 1016/j.oneear.2020.03.005 23. Labos 1point5. Enquête #1: « Les personnels de la recherche face au changement climatique » 2020. Available from: https://labos1point5.org/les-enquetes 24. Arsenault J, Talbot J, Boustani L, Gonzalès R, Manaugh K. The environmental footprint of academic and student mobility in a large research-oriented university. Environ Res Lett. 2019; 14: 095001. https://doi.org/10.1088/1748-9326/ab33e6 25. Wynes S, Donner SD, Tannason S, Nabors N. Academic air travel has a limited influence on profes- sional success. J Clean Prod. 2019; 226: 959–967. https://doi.org/10.1016/j.jclepro.2019.04.109 26. Canada (2021). National Inventory Report 1990–2019: Greenhouse gas sources and sinks in Canada —Part 1. https://unfccc.int/documents/271493 27. Galeote DF, Rajanen M, Rajanen D, Legaki NZ, Langley DJ, Hamari J. Gamification for climate change engagement: review of corpus and future agenda. Environ Res Lett. 2021; 16: 063004. https://doi.org/ 10.1088/1748-9326/abec05 28. Oliver S. Integrating role-play with case study and carbon footprint monitoring: a transformative approach to enhancing learners’ social behavior for a more sustainable environment. Int J Environ Sci Educ. 2016; 11(6): 1323–1335. https://doi.org/10.12973/ijese.2016.346a 29. Robinson J, Ausubel JH. A game framework for scenario generation for the Co2 issue. Simul Gaming. 1983; 14: 317–344. https://doi.org/10.1177/104687818301400306 PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 29 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community 30. Wiemeyer J, Nacke L, Moser C, Mueller FF. Player Experience. In: Do¨ rner R, Go¨ bel S, Effelsberg W, Wiemeyer J, editors. Serious Games. Foundations, Concepts and Practice. Cham: Springer. 2016. pp. 243–271. https://doi.org/10.1007/978-3-319-40612-1_9 31. Gee JP. Good Video Games and Good Learning. Collected Essays on Video Games, Learning and Lit- eracy. 2nd ed. York New: Peter Lang Editions; Series New Literacies and Digital Epistemologies. 2008; 67 167. 32. Onencan A, Van de Walle B, Enserink B, Chelang’a J, Kulei F. WeShareIt. Game: Strategic Foresight for Climate-change Induced Disaster Risk Reduction, Procedia Eng. 2016; 159: 307–315. https://doi. org/10.1016/j.proeng.2016.08.185 33. Terti G, Ruin I, Kalas M, La´ng I, Cangròs I Alonso A, et al. ANYCaRE: a role-playing game to investigate crisis decision-making and communication challenges in weather related hazards. Nat Hazards Earth Syst Sci. 2019; 19 (3), 507–533. https://doi.org/10.5194/nhess-19-507-2019 34. Undorf S, Tett SF, Hagg J, Metzger MJ, Wilson C, Edmond G, et al. Understanding interdependent cli- mate change risks using a serious game, Bull Am Meteorol Soc. 2020; 101(8): E1279–E1300. https:// doi.org/10.1175/BAMS-D-19-0177.1 35. Steffen W, Richardson K, Rockstro¨ m J, Cornell SE, Fetzer I, Bennett EM, et al. Planetary boundaries: Guiding human development on a changing planet. Science. 2015; 347(6223): 1259855. https://doi. org/10.1126/science.1259855 PMID: 25592418 36. Raworth K. A safe and just space for humanity: can we live within the doughnut? Oxfam Discussion papers. 2012. Available from: https://www-cdn.oxfam.org/s3fs-public/file_attachments/dp-a-safe-and- just-space-for-humanity-130212-en_5.pdf 37. Bador M, Terray L, Boe´ J, Somot S, Alias A, Gibelin AL, et al. Future summer mega-heatwave and record-breaking temperatures in a warmer France climate. Environ Res Lett. 2017; 12(7): 074025. https://doi.org/10.1088/1748-9326/aa751c 38. Berthoud F, Guitton P, Lefèvre L, Quinton S, Rousseau A, Sainte-Marie J, et al. Sciences, Environne- ments et Socie´te´s: Rapport long du groupe de travail MakeSEnS d’Inria. Inria report. 2019. Available from: https://hal.inria.fr/hal-02340948 39. Morris TH. Experiential learning—a systematic review and revision of Kolb’s model. Interact Learn Envi- ron. 2020; 28(8): 1064–1077. https://doi.org/10.1080/10494820.2019.1570279 40. Skiles M, Yang E, Reshef O, et al. Conference demographics and footprint changed by virtual platforms. Nat Sustain. 2022; 5, 149–156. Available from: https://doi.org/10.1038/s41893-021-00823-2 41. Pohlmann A, Walz K, Engels A, Aykut SC, Alstaedt S, Colell A, et al. It’s not enough to be right! The cli- mate crisis, power, and the climate movement. GAIA—Ecological Perspectives for Science and Soci- ety. 2021; 30(4): 231–236. https://doi.org/10.14512/gaia.30.4.5 42. Kolb DA. Experiential Learning: Experience as the Source of Learning and Development. 2nd edition. Upper Saddle River, New Jersey: Pearson Education; 2014. 43. Van Dalen HP. How the publish-or-perish principle divides a science: the case of economists. Sciento- metrics. 2021; 126: 1675–1694. https://doi.org/10.1007/s11192-020-03786-x 44. Irwin EG, Culligan PJ, Fischer-Kowalski M, Law KL, Murtugudde R, Pfirman S. Bridging barriers to advance global sustainability. Nat Sustain. 2018; 1(7): 324–326. https://doi.org/10.1038/s41893-018- 0085-1 45. Nguyen MH, Gruber J, Fuchs J, Marler W, Hunsaker A, Hargittai E. Changes in digital communication during the COVID-19 global pandemic: implications for digital inequality and future research. Soc Media Soc. 2020; 6(3). https://doi.org/10.1177/2056305120948255 PMID: 34192039 46. Jordan C, Palmer A. Virtual meetings: A critical step to address climate change. Sci Adv. 2020; 6 (38). https://doi.org/10.1126/sciadv.abe5810 PMID: 32938670 47. Shove E, Walker G. What is energy for? Social practice and energy demand. Theory Cult Soc. 2014; 31(5): 41–58. https://doi.org/10.1177/0263276414536746 48. Gifford R. The Dragons of inaction: Psychological barriers that limit climate change mitigation and adap- tation. Am Psychol. 2011; 66(4): 290–302. https://doi.org/10.1037/a0023566 PMID: 21553954 49. Ciers J, Mandic A, Toth LD, Op ‘t Veld G. Carbon footprint of academic air travel: a case study in Swit- zerland. Sustainability. 2019; 11(1): 80. https://doi.org/10.3390/su11010080 50. Flood S, Cradock-Henry NA, Blackett P, Edwards P. Adaptive and interactive climate futures: system- atic review of ‘serious games’ for engagement and decision-making. Environ Res Lett. 2018; 13(6): 063005. https://doi.org/10.1088/1748-9326/aac1c6 51. Martin P, Brau-Nogue´ S, Coriat M, Garnier P, Hughes A, Kno¨dlseder J, et al. The carbon footprint of IRAP. arXiv preprint. 2022. Available from: https://doi.org/10.48550/arXiv.2204.12362 PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 30 / 31 PLOS SUSTAINABILITY AND TRANSFORMATIONDecarbonization scenarios in the academic community 52. Kno¨ dlseder J, Brau-Nogue´ S, Coriat M, Garnier P, Hughes A, Martin P, Tibaldo L. Estimate of the car- bon footprint of astronomical research infrastructures. Nature Astronomy. 2022; 6(4), 503–513. Avail- able from: https://doi.org/10.1038/s41550-022-01612-3 53. Reckien D, Eisenack K. Climate change gaming on board and screen: a review. Simul Gaming. 2013; 44(2–3): 253–271. https://doi.org/10.1177/1046878113480867 54. Plass JL, Homer BD, Kinzer CK. Foundations of game-based learning. Educ Psychol. 2015; 50(4) 258–283. https://doi.org/10.1080/00461520.2015.1122533 PLOS Sustainability and Transformation \| https://doi.org/10.1371/journal.pstr.0000049 April 3, 2023 31 / 31 PLOS SUSTAINABILITY AND TRANSFORMATION
10.3201_eid2602.190434	Chronic Human Pegivirus 2 without Hepatitis C Virus Co-infection Kelly E. Coller, Veronica Bruce, Michael Cassidy, Jeffrey Gersch, Matthew B. Frankel, Ana Vallari, Gavin Cloherty, John Hackett, Jr., Jennifer L. Evans, Kimberly Page, George J. Dawson Most human pegivirus 2 (HPgV-2) infections are associ- ated with past or current hepatitis C virus (HCV) infection. HPgV-2 is thought to be a bloodborne virus: higher prev- alence of active infection has been found in populations with a history of parenteral exposure to viruses. We eval- uated longitudinally collected blood samples obtained from injection drug users (IDUs) for active and resolved HPgV-2 infections using a combination of HPgV-2– specific molecular and serologic tests. We found evi- dence of HPgV-2 infection in 11.2% (22/197) of past or current HCV-infected IDUs, compared with 1.9% (4/205) of an HCV-negative IDU population. Testing of available longitudinal blood samples from HPgV-2–positive partici- pants identified 5 with chronic infection (>6 months vire- mia in >3 timepoints); 2 were identified among the HCV- positive IDUs and 3 among the HCV-negative IDUs. Our findings indicate that HPgV-2 can establish chronic infec- tion and replicate in the absence of HCV. The recently identified second human pegivirus (HPgV-2 or HHpgV-1) is a bloodborne flavivirus: little is known about the potential clinical significance of infection (1,2). Active or resolved HPgV-2 infec- tion has been detected worldwide in cohorts associ- ated with risk for parenteral exposure to bloodborne pathogens (1,3–5). In a study in which hepatitis C vi- rus (HCV) status was determined (3), 1.2% of HCV positive were actively infected with HPgV-2; none of the 1,306 HCV-negative participants (volunteer blood donors, HBV infected, HIV infected) were actively in- fected. In another study, participants with concurrent HIV/HCV infection or injection drug users (IDUs) had a higher prevalence (3.0%–5.7%) of active HPgV-2 in- fection (4–6). Although active HPgV-2 has been found in other populations (e.g., hemophiliacs or others, Author affiliations: Abbott Laboratories, Abbott Park, Illinois, USA (K.E. Coller, M. Cassidy, J. Gersch, M.B. Frankel, A. Vallari, G. Cloherty, J. Hackett, Jr., G.J. Dawson); University of New Mexico, Albuquerque, New Mexico, USA (V. Bruce, K. Page); University of California San Francisco, San Francisco, California, USA (J.L. Evans) DOI: https://doi.org/10.3201/eid2602.190434 with risk for parenteral exposure), their HCV anti- body status was not determined (6,7). Previous studies indicate that HPgV-2 can estab- lish a chronic infection characterized by detectable viremia for >6 months (2,6). Most chronic HPgV-2 cases are associated with active HCV infection (2,6). In chronic HPgV-2 cases in which HCV RNA has not been detected, the presence of HCV antibodies (indi- cating a resolved infection) was not determined; thus, it is unclear whether previous HCV infection played a role in the initial HPgV-2 infection. Despite observa- tions of HCV and HPgV-2 co-infection, no evidence has been reported that HPgV-2 infection exacerbates HCV infection (5,6,8) or that co-infection influences the replication of either virus. We examined a cohort of IDUs for whom longi- tudinal samples were available. We monitored the cohort for HCV status by HCV antibodies, RNA, or both, with the intent of capturing nascent HCV in- fections (9). We performed initial testing for HPgV-2 (RNA and antibodies) on baseline and last collected samples. We further characterized longitudinal sam- ples from available participants that showed active or resolved HPgV-2 infection upon testing initial or last timepoints. We hypothesized that the IDUs would have similar prevalence of HPgV-2 as shown in a pre- vious study of HCV-infected persons with unknown IDU status, and that by studying IDUs without HCV infection we would uncover HPgV-2 infection in the absence of HCV. Last, we hypothesized that longi- tudinal samples from IDUs would reveal whether HPgV-2 can establish a persistent infection in the ab- sence of HCV co-infection. Materials and Methods Samples We obtained samples from the U-Find-Out (UFO) Study, an ongoing prospective observational study of young adult active injectors, <30 years of age at en- rollment, that was initiated in 2003 in the San Francis- co Bay area (California, USA). Details of enrollment Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 2, February 2020 265 RESEARCH methods and follow-up have been described previ- ously (9,10). In brief, young adult IDUs were recruit- ed from neighborhoods where IDUs were known to congregate and invited to participate in a field study for eligibility screening. Eligible persons were those who reported injection drug use in the prior 30 days, were <30 years of age, spoke English, had no plans to travel outside of the San Francisco Bay area for >3 months, and had negative or unknown HCV status. HCV antibody–positive persons were admitted into the study if their HCV RNA status was negative or unknown; those identified as HCV infected (RNA positive) at baseline were not enrolled into follow-up. Eligible consenting participants were asked to complete a baseline interviewer-administered struc- tured questionnaire that queried sociodemographics, parenteral and sexual risk behaviors and exposures, injecting exposures (e.g., frequency of injecting, num- ber of persons injected with, types of drugs injected), alcohol use, and prevention and health service use. They were also asked to provide blood samples for HCV testing, including both HCV antibodies and HCV RNA, and for storage. Before 2012, all partici- pants provided samples for HCV antibodies (using standard laboratory-based testing) and for a qualita- tive HCV RNA status determination using a nucleic acid amplification test (Procliex HIV-1/HCV assay; Gen-Probe Inc., https://www.novartis.com). Begin- ning in May 2012, HCV antibody testing was pri- marily conducted using a rapid test (OraSure Tech- nologies, https://www.orasure.com) by fingerstick capillary blood collection; however, venipuncture was still used to collect specimens for RNA testing and sample storage. Baseline samples on 402 partici- pants were selected as the initial sample set (Figure). A total of 205 (51.0%) samples were negative and 197 (49.0%) positive for HCV antibodies and HCV RNA at baseline. HCV-positive in this current study is defined as any evidence of HCV infection (RNA or HCV antibodies), past or present; HCV-negative is defined as no evidence (RNA or HCV antibodies), past or present. HPgV-2 Prevalence Study Design We used previously described HPgV-2 molecular (11) and serologic (3) assays to test all samples for deter- mining HPgV-2 prevalence. We divided them into 3 testing groups: sample set 1, initial blood samples (n = 402); sample set 2, all last available follow-up samples (n = 200); and sample set 3, any longitudi- nal samples available for samples that were HPgV- 2 (RNA or antibody) positive at initial or last draw timepoint (n = 70) (Figure). Because the initial study collection targeted incident HCV infection, a limited number of participants who were HCV positive at ini- tial collection had follow-up samples available; only 8 HCV-positive and 192 HCV-negative participants from sample set 1 had follow-up (last) draw available for testing, constituting sample set 2. HPgV-2 Molecular Assay We used a modified version of the HPgV-2 reverse transcription PCR (RT-PCR) to determine HPgV-2 vi- remia (11). The RT-PCR targets 2 conserved regions of the HPgV-2 genome within the 5′ untranslated region (UTR) and the nonstructural (NS) 2–3 coding region (11). We modified the assay to replace detection of HPgV RNA with an internal control. The internal con- trol was derived from the hydroxypyruvate reductase gene from the pumpkin plant, Cucurbita pepo, and is delivered in an Armored RNA (Ambion, Inc., https:// www.thermofisher.com) particle that has been dilut- ed in negative human plasma (nonreactive for HBsAg, HIV RNA, HCV RNA, HBV DNA, HIV-1/-2 antibod- ies, and HCV antibodies). We introduced the internal control into each specimen at the beginning of sample preparation as a control for extraction and ampli- fication. We extracted samples from plasma using Figure. Design of study of chronic human pegivirus-2 and hepatitis C virus co-infection in injection drug users in the San Francisco Bay area, California, USA. Samples were tested using HPgV-2 molecular and serologic assays in 3 sample sets. HCV, hepatitis C virus; HPgV-2, human pegivirus 2. 266 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 2, February 2020 Human Pegivirus 2 without Hepatitis C Co-infection the Abbott m2000sp instrument (Abbott Molecular, https://www.molecular.abbott) (open mode proto- col m2000-RNA-Plasma-LL-500–110-v71408, version 1.0). We used eluted nucleic acids immediately for subsequent PCR analysis or stored them in the deep well plate at –80°C. HPgV-2 Antibody Testing We screened research use–only assays to detect IgG re- sponse to HPgV-2 proteins for HPgV-2 seroconversion (3). In brief, we built 2 separate indirect IgG assays for use on the ARCHITECT instrument (Abbott Laborato- ries). The capture antigen for the E2 assay was mam- malian expressed glycoprotein E2, and for the NS4AB assay a portion of the NS4AB region. We generated signal to cutoff values for each assay by determining a provisional cutoff from testing a population of low- risk volunteer donors and calculating the median + 10 SD of relative light units (RLU) generated using the individual assays (3). Both E2 and NS4AB assays de- tected active and resolved HPgV-2 infection (3). Statistical Analyses We used the Fisher exact test to examine differenc- es in prevalence of HPgV-2 between subgroups (for example, by HCV status). We performed unpaired Student t-tests to determine if there was a significant difference (p<0.05) between the average HPgV-2 log copies/mL (NS2/3 or 5’UTR) of the HCV positive and negative groups. We used GraphPad Prism ver- sion 6.04 for Windows (GraphPad Software, https:// www.graphpad.com). Results Baseline Sample Testing The overall prevalence of HPgV-2 (presence of RNA or antibodies) among baseline samples in the IDU cohort was 6.5% (Table 1; Figure). We determined a higher HPgV-2 prevalence in the HCV-positive group (11.2%) compared with the HCV-negative group (1.9%) (p = 0.0002 by Fisher exact test). We observed HPgV-2 infection (HPgV-2 RNA) more frequently in the HCV-positive group (6.1%; 12/197 samples) than in the HCV-negative group (1.0%; 2/197 samples). Last Sample Testing Follow-up specimens were available for some study participants (sample set 2), primarily those who were HCV negative, due to the prospective design of the study that did not require follow-up samples from HCV-positive participants (Figure). However, the study also included persons with newly detected HCV infection whom we followed to examine natural history and resolution of incident HCV infection (9). A total of 200 participants from baseline collection had a final follow-up sample available for evaluation in the HPgV-2 RNA and antibody assays; this group in- cluded 8 HCV-positive and 192 HCV-negative partic- ipants (Figure). Although 26 participants were HPgV- 2 positive (by RNA or antibodies) at baseline (Table 1), only 4 participants were available for follow-up; they provided 3 HPgV-2 RNA-positive samples and 1 HPgV-2 RNA-negative seropositive sample. We saw evidence of HPgV-2 infection (antibodies or RNA) in 9 samples in set 2, 6 HCV-negative sam- ples and 3 HCV-positive samples (Table 2). Among the HCV-negative participants, 3 (QM0003, VH0052, VP0295) showed active HPgV-2 infection during ei- ther the baseline or last draw timepoint. Participant QM0003 showed chronic (>6 mos viremia) HPgV-2 infection during the study, with detection of HPgV-2 RNA at time points spanning 832 days (Table 2). Par- ticipant VH0052 was actively infected with HPgV-2 at baseline and resolved infection by the last draw date (201 days elapsed), whereas participant VP0295 acquired HPgV-2 infection during the study and was RNA positive at the last draw date (on study for 553 days). Three participants (GG0012, RM0095, RM0337) were HPgV-2 RNA and antibody negative at enroll- ment and had seroconverted to have HPgV-2 antibod- ies by the last draw. One participant, VH0044, was HPgV-2 seropositive at baseline but was seronegative (seroreverted) by the final draw (250 days elapsed). In the HCV-positive group, 1 participant (VH0085) was HPgV-2 RNA positive both on the first and last draw dates (2,805 days elapsed). The other 2 HCV-positive participants (GG0038 and VT0031) were HPgV-2 RNA negative on the first draw dates but showed active HPgV-2 infection at the last draw date. A single participant (RM0337) acquired both Table 1. Prevalence of HPgV-2 antibodies or RNA, initial blood draw samples from study of chronic HPgV-2 infection and HCV co-infection in injection drug users in the San Francisco Bay area, California, USA* HCV status HCV positive HCV negative Totals Ab, antibody; HCV, hepatitis C virus; HPgV-2, human pegivirus 2; +, positive; −, negative. No. HPgV-2 Ab+ 18 3 21 No. tested 197 205 402 No. HPgV-2 Ab+/RNA+ 8 1 9 No. HPgV-2 Ab−/RNA+ 4 1 5 Total RNA+ 12 2 14 Total HPgV-2 RNA or Ab+ (%) 22 (11.2) 4 (1.9) 26 (6.5) Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 2, February 2020 267 RESEARCH Table 2. Evidence of HPgV-2 infection in most recent samples from study of chronic HPgV-2 infection and HCV co-infection in injection drug users in the San Francisco Bay area, California, USA HCV RNA status at initial draw Negative Positive HPgV-2 status at initial blood draw HPgV-2 status at last blood draw Sample ID QM0003 VH0052 VP0295 GG0012 RM0095 RM0337§ VH0044 VH0085 GG0038 VT0031 RNA Pos Pos Neg Neg Neg Neg Neg Pos Neg Neg Antibody Neg Pos Neg Neg Neg Neg Pos Pos Neg Neg RNA Pos Neg Pos Neg Neg Neg Neg Pos Pos Pos Antibody Pos Pos Pos Pos Pos Pos Neg Pos Neg Pos HCV RNA status Neg Neg Neg Neg Neg Pos Neg Neg Neg Pos No. days† 832 201 553 689 201 1,680 250 2,805 461 818 Comments‡ Chronic Resolved Active Resolved Resolved Resolved Resolved Chronic Active Active HCV, hepatitis C virus; HPgV-2, human pegivirus 2; neg, negative; pos, positive. †Days elapsed between initial and last blood draw. ‡Comments of HPgV-2 status based on initial and last draw testing. Chronic indicates >6 months with detectable active viremia. Active indicates viremia at last draw. Resolved indicates no viremia detected at last draw, but antibodies were present. §Participant RM0337 acquired HCV during the study. HPgV-2 and HCV during the course of the study, with HPgV-2 infection preceding HCV infection by 280 days (Table 3). Within the last sample set, 6 par- ticipants demonstrated HPgV-2 infection after initial collection. Three of the participants showed active HPgV-2 viremia and 3 showed resolved HPgV-2 in- fection as indicated by detection of antibodies only. Longitudinal Sample Testing We tested longitudinal samples (N = 70, sample set 3) from 9/10 participants (Table 2) for HPgV-2 anti- bodies and RNA (Tables 3, 4; Figure). We reported data available for age, sex, years of injection drug use, HIV status, HCV RNA, HCV antibodies, alanine aminotransferase (HCV-positive only), and HCV drug treatment therapy. We did not follow up with additional HPgV-2 testing on participant VH0044, who was negative for HPgV-2 RNA and antibod- ies (seroreverted) by last sample date (Table 2). We found chronic HPgV-2 (active viremia >6 months) in 5 different IDU participants, 3 male and 2 female; 3 (QM0003, VP0295, RM0095) were HCV negative and 2 (VH0085, GG0038) HCV positive. None of the participants demonstrating chronic HPgV-2 infec- tion or seroconversion had evidence of HIV infec- tion. Among the participants who demonstrated chronic HPgV-2 infection and were HCV negative, participant QM0003 had a longer (>5 y) history of exposure to injection drug use than the other partici- pants, RM0095 and VP0295 (<5 y). Both participants with active HPgV-2 and HCV co-infection (GG0038 and VH0085) had long histories of injecting expo- sure (>5 y) and maintained HPgV-2 after HCV infec- tion was resolved. All chronic HPgV-2 infections demonstrated active viremia despite the presence of HPgV-2 an- tibodies, with most participants generating an IgG response to the glycoprotein E2 (Tables 3, 4). E2 an- tibodies developed in all participants with chron- ic HPgV-2 samples and observed seroconversion (GG0038, QM0003, VP0295, and RM0095) before the other marker, NS4AB antibody (Tables 3, 4). One chronically infected sample, VH0085, contained anti- bodies to both E2 and NS4AB, but we did not observe the initial seroconversion. Compared with the other samples in this study, VH0085 had the highest signal to cutoff values for both E2 and NS4AB antibody as- says, and the HPgV-2 RNA log copies/mL were high- er than in most other samples (Table 3). Some par- ticipants (VP0295, RM0095, GG0038) demonstrated seroconversion after several months of initial HPgV-2 RNA detection; these participants were chronically infected with HPgV-2. The median using the NS2/3 assay was 3.26 HPgV-2 log10 copies/mL for the HCV-negative group and 3.21 HPgV-2 log10 copies/mL for the HCV- positive group; using the 5′ UTR, results were 1.71 log10 copies/mL for the HCV-negative group and 1.60 log10 copies/mL for the HCV-positive group (Table 5). HCV co-infection did not appear to influ- ence HPgV-2 viral load; the average value showed no significant difference between the HCV-positive and HCV-negative groups (NS2/3, p = 0.11; 5′ UTR, p = 0.36). One HCV-positive participant, VH0085, was positive for HPgV-2 and HCV RNA at baseline, received HCV treatment (8 weeks ledipasvir/sofos- buvir), and cleared HCV infection. After clearance of HCV, the participant remained HPgV-2 viremic and went on to establish a chronic HPgV-2 infection that lasted 8 years (2,805 days). Participant GG0038 acquired HPgV-2 infection after spontaneous resolu- tion of HCV infection (RNA negative and HCV anti- body positive) and maintained active HPgV-2 infec- tion for >232 days. 268 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 2, February 2020 Human Pegivirus 2 without Hepatitis C Co-infection Conclusions The recently identified human pegivirus HPgV-2 has yet to be linked with any disease in humans. Several groups have shown HPgV-2 infection associated with HCV co-infection (1,3–6). HPgV-2 is a bloodborne virus, and a higher HPgV-2 prevalence is observed among HCV-positive IDUs (4,6). We decoupled the behavior of injection drug use from HCV status by monitoring HCV-negative or HCV-positive IDUs for HPgV-2 infection (RNA and antibodies). We also ob- served the enrichment of HPgV-2 infection in HCV- positive IDUs (Table 1), as was reported previously (4,6). We found, through longitudinal surveillance of both HCV-negative participants, that HPgV-2 can establish infection and maintain chronic infection in the absence of HCV. We defined chronic infection as detectable HPgV-2 viremia for >6 months in >3 time- points that was not associated with particular symp- toms. Of 9 participants with evidence of HPgV-2 infection (by RNA or serology), 2 HCV-infected par- ticipants demonstrated chronic HPgV-2 infection (Table 3), and 3 participants demonstrated chronic HPgV-2 without evidence of past or present HCV in- fection (Table 4). Several limitations can contribute to the under- estimation of chronic HPgV-2 infection. We identi- fied HPgV-2 in baseline samples from 12 HCV-pos- itive participants; but because of the study design most HCV-positive participants were not followed through subsequent timepoints. Three of these par- ticipants did provide longitudinal samples; 2 par- ticipants demonstrated chronic HPgV-2 infection. A second limitation is that participants testing negative for HPgV-2 during the timepoints evaluated may be- come positive following the last timepoint sampled, if they continue the risk behavior of intravenous drug use. Alternatively, false-positive chronic infections could result from long lapses in sampling, in which Table 3. Information about participants with HPgV-2 infection with HCV co-infection in study of injection drug users in the San Francisco Bay area, California, USA No. years Participant drug use† Collection date age, y/sex 25.6/F Sample ID RM0337 11.1 VH0085§ 22.2/F 9.9 GG0038§ 26.5/M 8.8 VT0031 19.8/M 1.8 2013 Jul 10 2013 Oct 9 2014 Jan 29 2016 Jul 27 2016 Oct 26 2017 Jan 18 2017 Apr 26 2017 Jul 19 2017 Oct 18 2018 Jan 31 2018 Feb 14 2010 May 14 2017 Feb 15 2017 May 17 2017 Aug 9 2017 Nov 1 2018 Jan 17 2016 Jul 14 2017 Mar 1 2017 May 3 2017 May 23 2017 Oct 18 2005 Feb 2 2005 May 24 2005 Jun 14 2005 Aug 9 2005 Nov 8 2006 Jan 31 2006 May 9 2006 Aug 8 2007 Feb 6 2007 May 1 HCV RNA Neg Neg Neg Neg Neg Neg Neg Neg Neg Pos Pos Pos Neg Neg Neg Neg Neg Pos Pos Pos Neg Neg Pos Pos Pos Pos Pos Pos Pos Pos Pos Pos HCV antibody Neg Neg Neg Neg Neg Neg Neg Neg Neg Neg Neg Pos Not tested Not tested Not tested Not tested Not tested Pos Not tested Not tested Not tested Not tested Pos Pos Pos Pos Pos Pos Pos Pos Pos Pos NS2/3 log10 copies/mL Neg Neg Neg Neg Neg Neg 0.96 Neg Neg Neg Neg 3.22 3.14 1.87 4.30 3.40 3.86 Neg 3.04 2.14 3.01 4.15 Neg Neg Neg Neg Neg Neg Neg Neg 2.30 0.48 5 UTR log10 copies/mL Neg Neg Neg Neg Neg Neg 0.42 Neg Neg Neg Neg 2.3 2.07 0.69 2.25 1.68 2.26 Neg 1.02 0.64 1.16 1.91 Neg Neg Neg Neg Neg Neg Neg Neg 0.53 0.41 NS4AB S/CO‡ 0.16 0.14 0.13 0.12 0.12 0.13 0.13 0.09 0.13 0.14 0.25 1.38 2.96 3.33 6.13 5.13 3.75 0.16 0.17 0.20 0.18 0.16 0.09 0.11 0.10 0.12 0.08 0.14 0.09 0.21 0.14 0.14 E2 S/CO‡ 0.14 0.13 0.12 0.11 0.13 0.13 1.46 1.34 0.88 1.02 1.26 9.37 12.92 12.35 18.71 19.45 16.16 0.12 ¶ 0.37 0.36 2.96 0.22 0.15 0.17 0.14 0.13 0.13 0.19 1.25 1.19 1.82 Participants were determined positive for HCV by RNA or antibody test. HCV, hepatitis C virus; HPgV-2, human pegivirus 2; ID, identification; neg, negative; NS, nonstructural protein; pos, positive; S/CO, signal to cutoff value; UTR, untranslated region. †Number of years participant had injected drugs as of the time of the initial blood draw. ‡S/CO >1.0 is considered positive. §Participants VH0085 and GG0038 demonstrate resolved HCV infection. VH0085 was administered direct active antiviral drugs (8 weeks ledipasvir/sofosbuvir, LDV-SOF), and GG0038 spontaneously cleared HCV infection. ¶Volume not available for testing. Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 2, February 2020 269 RESEARCH the participant could become infected, clear the infec- tion, then become reinfected with HPgV-2. The mo- lecular and serologic assays cannot distinguish rein- fection from chronic infection. We observed HPgV-2 seroconversion in the longi- tudinal surveillance of 7 participants; the detectable IgG response occurred several months after initial HPgV-2 RNA detection in several of the chronically infected par- ticipants (Tables 3, 4). Similar to HCV infection, which demonstrates a seronegative viremic window period of 50–60 days (12–14), the detection of HPgV-2–specific an- tibodies lagged behind detectable HPgV-2 RNA (Tables 3, 4). In this study, all participants demonstrating chron- ic HPgV-2 viremia did so in the presence of antibodies to the glycoprotein E2, which suggests that E2 antibod- ies are not neutralizing. In contrast, active viremia and E2 IgG are rarely co-detected in persons with the closest human virus, human pegivirus-1 (HPgV, GBV-C); the presence of E2 antibodies in HPgV infection often indi- cates resolution of infection (3,15–17). Our data indicate that chronic HPgV-2 infection among IDUs does not require active HCV to establish infection or maintain chronic infection. It is possible that participants with no detectable HCV antibodies were seroreverters from previous cleared HCV infec- tions; however, this is unlikely in our study because the cohort of active IDUs had ongoing exposure to HCV, and because seroreversion in immunocompe- tent persons has been shown to occur after a long time (>7 years) (18), exceeding the observation period of this study. Furthermore, 2 participants showed spontane- ous (GG0038) or therapeutic (VH0085) resolution of Table 4. Information about participants with HPgV-2 infection without HCV co-infection in study of injection drug users in the San Francisco Bay area, California, USA Sample ID QM0003 Age, y/sex 28.4/F No. years drug use† 14.4 VP0295 24.6/M 3.1 GG0012 23.4/F 3.4 VH0052 RM0095 23.7/M 27.1/M 3.5 1.1 Collection date 2015 Oct 10 2015 Nov 4 2016 Jan 27 2016 Apr 18 2016 Jul 13 2016 Oct 5 2017 Jan 4 2017 Apr 10 2017 Jul 5 2018 Feb 7 2016 Jan 6 2016 Apr 6 2016 Aug 24 2017 Jan 4 2017 Apr 12 2017 Jul 12 2016 Apr 29 2016 Aug 17 2016 Nov 29 2017 Mar 1 2017 May 31 2017 Sep 20 2017 Dec 13 2018 Mar 19 2006 Dec 14 2007 Jul 3 2011 Jan 19 2011 Jul 27 2011 Oct 19 2012 Jan 11 2012 Apr 11 2012 Jul 3 2012 Oct 23 2013 Jan 15 2013 Apr 10 2013 Jul 2 2013 Sep 18 2013 Dec 11 NS2/3 log10 copies/mL 3.76 0.93 2.66 3.56 3.66 3.22 4.22 3.85 4.08 3.41 Neg Neg Neg 4.44 3.19 1.42 Neg Neg Neg Neg Neg Neg Neg Neg 2.13 Neg Neg 0.02 2.33 0.20 0.02 Neg 0.03 0.45 1.37 Neg 0.46 Neg 5 UTR log10 copies/mL 2.84 0.55 1.10 1.62 1.61 1.74 2.30 2.10 2.39 1.34 Neg Neg Neg 2.82 1.06 0.82 Neg Neg Neg Neg Neg Neg Neg Neg 1.23 Neg Neg 0.14 1.54 0.50 0.16 Neg 0.27 0.61 1.08 Neg 0.55 Neg NS4AB S/CO‡ 0.12 0.08 0.16 0.07 § 0.13 0.25 0.17 0.28 0.37 0.10 0.09 0.12 0.11 0.09 0.15 0.23 1.77 3.18 4.90 2.60 4.05 6.03 3.77 2.00 0.81 0.06 0.07 0.06 0.09 0.24 0.14 0.12 0.15 0.13 0.17 0.14 0.14 E2 S/CO‡ 0.08 0.08 0.72 0.85 § 2.76 4.67 3.58 3.73 2.66 0.12 0.11 0.11 0.14 0.89 5.51 0.52 3.28 5.76 8.90 4.51 10.94 11.04 6.67 1.91 1.63 0.06 0.09 0.29 0.59 0.50 0.59 1.09 2.45 2.37 3.53 2.64 3.07 Participants were determined negative for HCV by RNA or antibody test. HCV, hepatitis C virus; HPgV-2, human pegivirus 2; ID, identification; neg, negative; NS, nonstructural protein; pos, positive; S/CO, signal to cutoff value; UTR, untranslated region. †Number of years participant had injected drugs as of the time of the initial blood draw. ‡S/CO >1.0 is considered positive. §No blood sample available for testing. 270 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 2, February 2020 Human Pegivirus 2 without Hepatitis C Co-infection Table 5. Characteristics of findings in study of chronic HPgV-2 infection for participants with and without HCV co-infection in injection drug users in the San Francisco Bay area, California, USA Finding Average HPgV-2, log10 copies/mL NS2/3 5′ UTR Average years injection drug use† Average age at detection of HPgV-2 RNA, y† *HCV, hepatitis C virus; HPgV-2, human pegivirus 2; NS, nonstructural protein. †No p values available because of low number of samples. 3.26 1.71 5.5 25.6 3.21 1.60 7.9 26.1 0.11 0.36 NA NA HCV negative, n = 3 HCV positive, n = 2 p value HCV infection but HPgV-2 chronic infection remained (Table 3); thus, there appeared to be no reliance on HCV to establish or maintain HPgV-2 infection. We observed no difference in HPgV-2 viral loads whether HCV was present or absent (Table 5). The relative ratio of resolved to active infections between the HCV-pos- itive and HCV-negative cohorts was similar (Table 1). As noted in this and other studies, high-risk pop- ulations that are exposed to parenterally transmitted viruses experience an increase in HPgV-2 prevalence (4,6,7). We observed similar higher incidence of both active and resolved HPgV-2 infection in the HCV- positive IDU cohort (Table 1). The HCV-positive and HCV-negative infected persons within the IDU cohort share many common behaviors with no dis- cernable characteristics, except total number of years of injection drug use (average 7.6 y for HCV-positive users, 4.7 y for HCV-negative users). Similarly, the HCV-positive HPgV-2 carriers identified in this study demonstrated injection drug use behavior longer (av- erage 7.9 y) than the HCV-negative group (average 4.4 y). The greater number of potential exposures to parentally transmitted bloodborne viruses is prob- ably a major contributing factor for increased preva- lence of HPgV-2 in the HCV-positive IDU cohort. The pathogenic potential of HPgV-2 in humans remains unknown; no clinical symptoms have been associated with HPgV-2 infection. We gathered no additional clinical information from HCV-negative study participants. Most HPgV-2 studies have shown the virus is associated with HCV co-infection, which can mask any pathogenicity associated with HPgV- 2 infection. Identifying populations that show higher prevalence of HPgV-2 monoinfection and monitor- ing these persons over time may help identify clini- cal symptoms associated with HPgV-2 infection, thus enabling researchers to categorize HPgV-2 as human pathogen or benign infection. Acknowledgments We acknowledge the ongoing support and efforts of the Blood Systems Research Institute, especially Michael Busch, Mars Stone, and Honey Dave, for their expert management of the UFO Study specimen bank. About the Author Dr. Coller is a research scientist at Abbott Laboratories, Abbott Diagnostics Division. Her research interests include developing serologic assays for the detection of emerging infectious diseases. References 1. Berg MG, Lee D, Coller K, Frankel M, Aronsohn A, Cheng K, et al. Discovery of a novel human pegivirus in blood associated with hepatitis C virus co-infection. PLoS Pathog. 2015;11:e1005325. https://doi.org/10.1371/ journal.ppat.1005325 2. Kapoor A, Kumar A, Simmonds P, Bhuva N, Singh Chauhan L, Lee B, et al. Virome analysis of transfusion recipients reveals a novel human virus that shares genomic features with hepaciviruses and pegiviruses. MBio. 2015;6:e01466–15. https://doi.org/10.1128/mBio.01466-15 3. Coller KE, Berg MG, Frankel M, Forberg K, Surani R, Chiu CY, et al. Antibodies to the novel human pegivirus 2 are associated with active and resolved infections. J Clin Microbiol. 2016;54:2023–30. https://doi.org/10.1128/ JCM.00515-16 4. Wang H, Wan Z, Sun Q, Zhu N, Li T, Ren X, et al. Second human pegivirus in hepatitis C virus-infected and hepatitis C virus/HIV-1-co-infected persons who inject drugs, China. Emerg Infect Dis. 2018;24:908–11. https://doi.org/10.3201/ eid2405.161162 5. Wang H, Wan Z, Xu R, Guan Y, Zhu N, Li J, et al. A novel human pegivirus, HPgV-2 (HHpgV-1), is tightly associated with hepatitis C virus (HCV) infection and HCV/human Immunodeficiency virus type 1 coinfection. Clin Infect Dis. 2018;66:29–35. https://doi.org/10.1093/cid/cix748 6. Kandathil AJBF, Thomas DL, Balagopal A, Mehta SH, Timp W, Salzberg SL, et al. Presence of human hepegivirus-1 in a cohort of people who inject drugs. Ann Intern Med. 2018;168:158–9. https://doi.org/10.7326/L17-0527 7. Bonsall D, Gregory WF, Ip CLC, Donfield S, Iles J, Ansari MA, et al. Evaluation of viremia frequencies of a novel human pegivirus by using bioinformatic screening and PCR. Emerg Infect Dis. 2016;22:671–8. https://doi.org/ 10.3201/eid2204.151812 8. Bijvand Y, Aghasadeghi MR, Sakhaee F, Pakzad P, Vaziri F, Saraji AA, et al. First detection of human hepegivirus-1 (HHpgV-1) in Iranian patients with hemophilia. Sci Rep. 2018;8:5036. https://doi.org/10.1038/s41598-018-23490-4 9. Page K, Hahn JA, Evans J, Shiboski S, Lum P, Delwart E, et al. Acute hepatitis C virus infection in young adult injection drug users: a prospective study of incident infection, resolution, and reinfection. J Infect Dis. 2009;200:1216–26. https://doi.org/10.1086/605947 10. Hahn JA, Page-Shafer K, Lum PJ, Bourgois P, Stein E, Evans JL, et al. Hepatitis C virus seroconversion among Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 2, February 2020 271 RESEARCH young injection drug users: relationships and risks. J Infect Dis. 2002;186:1558–64. https://doi.org/10.1086/345554 antibody. J Med Virol. 1999;57:370–5. https://doi.org/10.1002/ (SICI)1096-9071(199904)57:4<370::AID-JMV8>3.0.CO;2-K 11. Frankel M, Forberg K, Coller KE, Berg MG, Hackett J Jr, Cloherty G, et al. Development of a high-throughput multiplexed real time RT-PCR assay for detection of human pegivirus 1 and 2. J Virol Methods. 2017;241:34–40. https://doi.org/10.1016/j.jviromet.2016.12.013 12. Cox AL, Netski DM, Mosbruger T, Sherman SG, Strathdee S, Ompad D, et al. Prospective evaluation of community- acquired acute-phase hepatitis C virus infection. Clin Infect Dis. 2005;40:951–8. https://doi.org/10.1086/428578 13. Maheshwari A, Ray S, Thuluvath PJ. Acute hepatitis C. Lancet. 2008;372:321–32. https://doi.org/10.1016/ S0140-6736(08)61116-2 14. Glynn SA, Wright DJ, Kleinman SH, Hirschkorn D, Tu Y, Heldebrant C, et al. Dynamics of viremia in early hepatitis C virus infection. Transfusion. 2005;45:994–1002. https://doi.org/10.1111/j.1537-2995.2005.04390.x 15. Tanaka T, Hess G, Schlueter V, Zdunek D, Tanaka S, Kohara M. Correlation of interferon treatment response with GBV-C/HGV genomic RNA and anti-envelope 2 protein 16. Gutierrez RA, Dawson GJ, Knigge MF, Melvin SL, Heynen CA, Kyrk CR, et al. Seroprevalence of GB virus C and persistence of RNA and antibody. J Med Virol. 1997;53:167–73. https://doi.org/10.1002/(SICI)1096- 9071(199710)53:2<167::AID-JMV10>3.0.CO;2-G 17. Dille BJ, Surowy TK, Gutierrez RA, Coleman PF, Knigge MF, Carrick RJ, et al. An ELISA for detection of antibodies to the E2 protein of GB virus C. J Infect Dis. 1997;175:458–61. https://doi.org/10.1093/infdis/175.2.458 18. Lefrère JJ, Girot R, Lefrère F, Guillaume N, Lerable J, Marrec NL, et al. Complete or partial seroreversion in immunocompetent individuals after self-limited HCV infection: consequences for transfusion. Transfusion. 2004; 44:343–8. https://doi.org/10.1111/j.1537-2995.2004.00656.x Address for correspondence: Kelly E. Coller, Abbott Laboratories, Abbott Diagnostics Division, 100 Abbott Park Rd, Abbott Park, IL 60064-6400, USA; email: kelly.coller@abbott.com etymologia Pegivirus [pegʺi-viʹrǝs] Ronnie Henry In 1967, researchers studying non-A, non-B hepatitis identified a transmissible agent in the serum of a surgeon (initials G.B.) with acute hepatitis and named it the GB agent. In the 1990s, researchers from Abbott Laboratories identified 3 GB viruses (A, B, and C) at the same time as a group at Genelabs isolated RNA from patients with non-A, non-B hepatitis and named it hepa- titis G virus. Later research showed that GB virus C and hepatitis G virus were the same species. Subsequent phylogenetic analysis showed that GB viruses A and C (and GB virus D, lat- er identified in bats) should be classified un- der a new genus, Pegivirus (because they cause persistent infection and because of the historic association with hepatitis G), and GB virus B should be classified as a second species (with hepatitis C virus) in the genus Hepacivirus. As of 2016, 11 species of Pegivirus had been identified (Pegivirus A–K). Sources 1. Linnen J, Wages J Jr, Zhang-Keck ZY, Fry KE, Krawczynski KZ, Alter H, et al. Molecular cloning and disease association of hepatitis G virus: a transfusion-transmissible agent. Science. 1996;271: 505–8. https://doi.org/10.1126/science.271.5248.505 2. Simons JN, Leary TP, Dawson GJ, Pilot-Matias TJ, Muerhoff AS, Schlauder GG, et al. Isolation of novel virus-like sequences associated with human hepatitis. Nat Med. 1995;1:564–9. https://doi.org/ 10.1038/nm0695-564 3. Smith DB, Becher P, Bukh J, Gould EA, Meyers G, Monath T, et al. Proposed update to the taxonomy of the genera Hepacivirus and Pegivirus within the Flaviviridae family. J Gen Virol. 2016;97:2894–907. https://doi.org/10.1099/jgv.0.000612 4. Stapleton JT, Foung S, Muerhoff AS, Bukh J, Simmonds P. The GB viruses: a review and proposed classification of GBV-A, GBV-C (HGV), and GBV-D in genus Pegivirus within the family Flaviviridae. J Gen Virol. 2011;92:233–46. https://doi.org/10.1099/vir.0.027490-0 Address for correspondence: Ronnie Henry, Centers for Disease Control and Prevention, 1600 Clifton Rd NE, Mailstop E28, Atlanta, GA 30333, USA; email: boq3@cdc.gov DOI: https://doi.org/10.3201/eid2602.ET2602 272 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 26, No. 2, February 2020
10.1371_journal.pone.0287610	RESEARCH ARTICLE How does the integration of cultural and tourism industries impact the value added to tourism value chain: Evidences from Jiangsu Province of China Meiling ZengID ¤☯, Suyan Shen¤☯, Jie Gu¤ Department of Business Administration, College of Economics and Management, Nanjing Forestry University, Nanjing, Jiangsu Province, People’s Republic of China ☯ These authors contributed equally to this work. ¤ Current address: Nanjing Forestry University, Nanjing, People’s Republic of China shensuyan@njfu.edu.cn Abstract China has been fully implementing the policy of the cultural and tourism industrial integration since 2018. However, the value-added benefits of this policy are not prominent, and the rela- tionship between industrial integration and the value added to the tourism value chain was seldom addressed by researchers. In the context of China’s high-quality development, it is necessary to conduct the impact of the integration of cultural and tourism industries on the value added to tourism value chain. This paper proposed four theoretical hypotheses and the corresponding econometric models based on the panel data from 2013 to 2020 in Chi- na’s Jiangsu Province. According to empirical results, the integration of cultural and tourism industries is spatially unbalanced, with notable imbalances between the south and the north. This paper identified a new connection between cultural and tourism integration and the tourism value chain. It is found that the integration of cultural and tourism industries can enhance the value added to tourism value chain either directly or indirectly through the infor- mation technology, with the direct effect being positively moderated by tourism agglomera- tion. Moreover, this paper may overturn how people generally think about the integration between cultural and tourism industries. It reveals a single-threshold effect that only when the integration of cultural and tourism industries reached a high level will it exert a positive effect. To be more specific, not all Chinese cities are suitable for implementing cultural and tourism integration, because the integration is likely to be ineffective in regions where the cultural industry is substantially less developed than the tourism industry. Introduction Cultural tourism refers to tourism activities that involve heritage tourism [1] and creative tour- ism [2]. The value added of cultural tourism depends on the consumption activities occurred within the cultural scenic spots, which have generally a low value. The value added by cultural a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Zeng M, Shen S, Gu J (2023) How does the integration of cultural and tourism industries impact the value added to tourism value chain: Evidences from Jiangsu Province of China. PLoS ONE 18(6): e0287610. https://doi.org/10.1371/ journal.pone.0287610 Editor: Han Lin, Nanjing Audit University, CHINA Received: March 30, 2023 Accepted: June 8, 2023 Published: June 29, 2023 Copyright: © 2023 Zeng et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The data underlying the results presented in the study are available from Chinese public database of Statistics Yearbook, i.e. Jiangsu Statistics Yearbook (http://tj. jiangsu.gov.cn/col/col86293/index.html), Jiangsu Culture and Tourism Statistics Yearbook (https:// navi.cnki.net/knavi/yearbooks/YJSWH/detail? uniplatform=NZKPT&language=chs) China City Statistics Yearbook (https://navi.cnki.net/knavi/ yearbooks/YZGCA/detail?uniplatform= NZKPT&language=chs), and the Statistics Yearbooks of the 13 cities (https://navi.cnki.net/ PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 1 / 20 PLOS ONEknavi/yearbooks/YNJTJ/detail?uniplatform= NZKPT&language=chs). Funding: Major Project of Philosophy and Social Science Research in Colleges and Universities of Jiangsu Province (Grant No. 2021SJZDA052). National Natural Science Foundation of China (Grant No. 32171856). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. How does the integration of cultural and tourism industries impact the value added to tourism value chain and tourism integration has been recognized in four parts [3]. Following the occurrence of industrial integration, actors in the tourism industrial chain will engage in new value-adding activities, resulting in the phenomenon of value reconstruction [4]. However, the existing stud- ies of cultural and tourism integration rarely addressed this phenomenon, but mainly focused on the current situation and its action mechanism. Tourism value chain analysis has been regarded as a useful tool for assessing the impact of the tourism industry on developing coun- tries [5]. It can help them identify the connections between tourism industry and the other industries [6]. Value chain governance and poverty reduction are two main topics addressed by the studies of tourism value chain, but there is a research gap on the relationship between cultural and tourism integration and the tourism value chain. Nowadays, tourists tend to pursue more authentic experiences from tourism activities, so the boundaries between the cultural and tourism industries are gradually blurred and the scope of the tourism value chain is being expanded continuously [7]. The integration between cultural and tourism industries may promote other industries’ consumption due to the tour- ism multiplier effect [8], and realize industrial transformation and upgrading during the inte- gration process. In some countries, cultural and tourism integration has been incorporated into the overall national economic development strategy [3]. China, for example, has not only released the policy of cultural and tourism integration but also implemented the correspond- ing institutional restructuring. Since then, China has been undergoing a boom in cultural and tourism integration all over the country. However, some problems have arisen during the inte- gration process, such as the imbalance of industrial structure [9]. The use of value chain analy- sis can provide theoretical support for solving the problems caused by the integration between the two industries. Taking Jiangsu Province of China as a case study, this paper aimed to explore the impact of the integration between cultural and tourism industries on the value added to the tourism value chain. Specifically, it addressed the following issues: (a) what is the integration level of cultural and tourism industries. (b) whether this integration level impacts tourism value chain and whether there is any heterogeneity; and (c) how does the industrial integration impact tourism value chain adding. From the perspective of tourism value chain, this paper provided creative responses to the question as how to effectively implement cultural and tourism inte- gration. First, on the basis of the value chain theory, a new connection between cultural and tourism integration and the tourism value chain was identified, which was in line with China’s national strategies of high-quality development. Second, the general impression that cultural and tourism integration can bring over significant benefits was overturned. It was revealed that only high-level integration between the cultural and tourism industries might lead to value addition. The structure of this paper is as follows: Section 2 provides literature reviews on the related topics; Section 3 proposes four theoretical hypotheses; Section 4, Section 5 and Section 6 pres- ent the empirical process and results; and Section 7 is the conclusions and implications. Literature review Integration of cultural and tourism industries The integration between cultural and tourism industries is a common market phenomenon, but not much research has been conducted on this topic yet. The existing literature mainly focuses on two aspects: the evaluation of the current situation and the action mechanism in different aspects. On the one hand, researchers have evaluated the current situation of the regional integra- tion of cultural and tourism industries by measuring the spatial and temporal evolution. Grey PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 2 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain relational analysis, data mining [10], and the coupling coordination degree model [9, 11, 12] are common models that can reflect the differences of regional development of cultural and tourism industries. For example, it was reported that the cultural industry in Guangxi province [12] was developing slightly faster than the tourism industry, but Shaanxi [9] lagged behind the tourism industry. On the other hand, driving factors and influence paths have emerged as hotspots in the research of the related field. The integration between cultural and tourism industries has apparent advantages in promoting industrial and economic development [13]. With the devel- opment of digital economy, information technology has rebuilt the tourism value chain [14] and greatly improved the tourism efficiency [15]. According to the available literature, researchers have not comprehensively discussed the necessity of the integration between cul- tural and tourism industries, nor have they elaborated how to do it from the perspective of high-quality tourism development. Tourism value chain Value chain analysis is widely used in the research of manufacturing industry based on Porter’s value chain theory. However, the uniqueness of the tourism industry makes it necessary to spe- cially define tourism value chain rather than directly adopting the value chain theory of the manufacturing industry. According to Hjalager [16], there are supply-oriented and demand- oriented definitions for tourism value chain. From the supply-oriented perspective, tourism value chain can be considered as the supply chain of tourism products based on Porter’s value chain model. It involves four types of stakeholders [17]. From the demand-oriented perspec- tive, tourism value chain has the meaning of Value Shop [18] and is a continuum of the related economic activities associated with visitors [19]. Tourism value chain involves multiple actors and the distribution of value among multiple industries. Through value chain analysis, the dynamic flow of economic and organizational activities among actors of different industries can be easily uncovered by focusing on the inter- linkage. Most of the studies on tourism value chain are qualitative analyses of the governance model [5, 17], the rural tourism value chain reconstruction [20], and the value co-creation model [21]. In addition, there are also a few scholars who quantitatively analyzed the impact of tourism value chain on tourism poverty alleviation [22], regional economic leakage [23], and tourism destination management [18]. However, how to improve the regional tourism value chain from the perspective of industrial integration was rarely addressed. In fact, this is the perspective that can best reflect the concept of value chain. Theoretical hypotheses The relationship between industrial integration and tourism value chain The integration of cultural and tourism industries refers to the process of mutual penetration, continuous reorganization, and optimization of culture and tourism elements [3], which can promote tourism value chain from three aspects, i.e., integration of markets, integration of resources, and integration of supply chains. Firstly, the market of cultural industry has been highly valued by the suppliers due to its excellent quality, high consumption potential, and generally-steady visitor flow [15]. Markets integration can promote the transfer of tourism value chain to high-consumer groups and improve consumption level. The integration of cultural and tourism industries can also opti- mize the entire value chain by encouraging the participation of tourists from design, manufacturing, management, marketing, and other multi-links [24]. Secondly, the advantage of cultural and tourism integration lies in the combination of the aesthetic value of cultural PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 3 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain resources with the experience value of tourism resources. Such integration can facilitate the flow of resources and value restructuring, which means that materials, knowledge, and human resources will be allocated to higher-value parties for the purpose of generating higher returns [25]. Finally, the primary tourism-related industries, covering the six elements of tourism (i.e., food, accommodation, transportation, sightseeing, shopping, and entertainment), used to be the essential participants and the main value-added body in the tourism value chain. But now, they have declined to the low end of the value chain because of low barriers to entry, limited resources renewal, and insufficient knowledge. The integration of supply chains between cul- tural and tourism industries implies that resources and values are redistributed among differ- ent actors. It can help the primary tourism-related industries evolve to a higher value-added stage in the value chain and, at the same time, add value to the whole tourism value chain. Hypothesis 1a. The integration of cultural and tourism industries can promote the value added to tourism value chain. Tourism has evolved into an important stage featuring the transformation from high-speed to high-quality development [26]. However, not every area’s tourism has advanced to the point of high-quality development. One key element determining the high-quality growth of tourism is the degree of industrial integration. High-level integration of cultural and tourism industries is a comprehensive concept covering cultural tourism resources, facilities and envi- ronments [27]. By supplying effectively, it can meet market demands and guide consumptions. Contrarily, low-level integration of cultural and tourism industries describes the early stages of integration that are low-coupling, uncoordinated, and characterized by disparities between supply and demand [28]. Over-commercialization and homogeneity might become issues as a result. The resource dependency theory suggests that enterprises with superior resources have the power to determine the flow of resources in competing relationships. Therefore, the unco- ordinated development of cultural industry and tourism industry is not conducive to the effi- cient and reasonable flow of information, resources and other factors. Hypothesis 1b. The value-added effect of high-level integration between cultural and tourism industries is superior to that of low-level integration. The moderating effect of tourism agglomeration Tourism agglomeration refers to the spatial proximity of tourism businesses on the basis of the relationship in the supply chain, which emphasizes the importance of value chain and spatial agglomeration [29]. In view that the quantity and variety of products in a tourism destination have an obvious effect on tourists’ choices [30], stronger tourism agglomeration means a more variety of goods to serve tourists and a higher value of experience gained by tourists. In addi- tion, tourism agglomeration can be utilized as a collaboration platform among local tourism companies [31]. It can accelerate the flow of various factors in the agglomeration space to form the scaling effect through the mechanisms of information sharing, resource allocation, talent exchange and policy support across enterprises [12]. The higher the degree of tourism agglom- eration, the greater the value added of tourism value chain is. Therefore, tourism agglomera- tion adds value not only to the various products clustered by enterprises, but also to the tourists’ experience. Hypothesis 2. Tourism agglomeration can intensify the promoting effect of integration between cultural and tourism industries on the value added to tourism value chain and play a positive moderating role. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 4 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain The mediating effect of information technology Information technology aids industrial integration, which will tear down barriers across indus- tries. The integration of cultural and tourism industries has been substantially improved by information technology, especially during the epidemic when tourist activities are restricted [13]. It can play positive effect by achieving product innovation and precise marketing. Firstly, information technology plays a significant role in knowledge creation [32]. Technologies such as VR and AR are carriers of invisible cultural products, which are helpful for the innovation of tourism products by breaking through the temporal and spatial restrictions. Thus, tourists can experience the dialogue with history and activate the traditional culture. Secondly, tourism is a very information-intensive activity [33]. Information technology not only affects tourists’ information acquisition and consumption decision making, but also plays a positive role in the marketing of suppliers [34]. Due to the intangibility and unpredictability of tourism products, tourists need to collect adequate information when choosing their desired products. At the same time, tourists’ demands are also diverse and volatile, so the supplies of tourism products and services are required to respond to market changes in a timely manner. The market orien- tation of actors guiding the production process at multiple stages of the chain based on market information is the prerequisite for the creation of value [35]. Hypothesis 3. The integration of cultural and tourism industries achieves value added to tour- ism value chain through information technology. It plays a mediating role in this process. Methods The coupling coordination degree evaluation model Coupling refers to the motion of a system where the subsystems interact with each other, and coor- dination means the relationship between subsystems that work together in a harmonious way [36]. The coupling coordination degree (CCD) has been widely used to measure the level of industry integration. Although the conventional concept of CCD has been studied by plenty of scholars, it involves two obvious problems [37]. Firstly, the coupling model didn’t satisfy the assumption of normal distribution but was explained by it. Secondly, the contribution coefficients of the coordi- nation model were defined in an artificial and subjective way, which simplified the model but devi- ated from the true level. Correspondingly, Wang [37] modified the conventional coupling model and solved the first problem, while Shen [38] proposed an improved coordination model to address the second problem. Therefore, the improved CCD models were used to measure the inte- gration between cultural and tourism industries. The detailed calculation steps are as follows. First, the information entropy method is applied to calculate the weights of indicators. It assigns a weight to each indicator mainly on the basis of the information contained in this indicator rather than the data linearity [39], which can avoid bias by subjective influence. Standardize each positive indicator to eliminate the differences in units. Mij ¼ Xij (cid:0) minjðXijÞ maxjðXijÞ (cid:0) minjðXijÞ þ 0:001 Calculate the proportion of the indicator j in the sample i. Pij ¼ Mij i¼1Mij Sm PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 ð1Þ ð2Þ 5 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain Calculate the information entropy of the indicator j. ej ¼ (cid:0) 1 lnðmÞ Xm 1 PijlnðPijÞ Calculate the weight of the indicator j. wj ¼ 1 (cid:0) ej i¼11 (cid:0) ej Sn ð3Þ ð4Þ Where Mij and Xij denote the standardized value and the original value of indicator j in the sample i, respectively; maxj(Xij) and minj(Xij) refer to the maximum and minimum value of indicator j among all the samples, respectively; n refers to the number of indicators in each sys- tem; m refers to the total sample covering all the cities over the entire observation period. Second, the technique for order preference by similarity to an ideal solution (TOPSIS) can reflect the relative importance of each indicator with the time sequence. It serves an effective tool to evaluate the development degree of a subsystem [36]. Calculate the distance from a sample to the positive ideal solution and negative ideal solu- tion. 8 >>>>>< >>>>>: dþ j ¼ d(cid:0) j ¼ s s ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Xn j Þ2 wjðMij (cid:0) Mþ j¼1 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Xn j Þ2 wjðMij (cid:0) M(cid:0) j¼1 ð5Þ Where Mþ j and M(cid:0) samples, respectively. j refer to the maximum and minimum value of indicator j among all the Calculate the relative closeness of a sample to the ideal solution. ci ¼ d(cid:0) i dþ i þ d(cid:0) i Finally, the improved CCD is used to reflect the integration of the two industries. Calculate the coupling degree following Wang’s [37] modification. Assuming max Ui is Uc, then s ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Ut Uc 1 (cid:0) ðUc (cid:0) UtÞ � C ¼ ð6Þ ð7Þ Where Uc and Ut refer to the comprehensive development level of the cultural industry and tourism industry measured by the relative closeness ci, respectively. Calculate the coordination degree following Shen’s [38] modification. 8 >>>>>< >>>>>: T ¼ a � Uc þ b � Ut a ¼ b ¼ Ut UC þ Ut UC UC þ Ut ð8Þ Where α and β are contribution coefficients of the two systems, meeting the condition of α +β = 1. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 6 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain Calculate the improved CCD as follow. p ffiffiffiffiffiffiffiffiffiffiffiffiffi C � T D ¼ ð9Þ Index selection The evaluation index system was used to calculate the CCD of the cultural and tourism indus- tries in this paper. According to the industrial integration mechanism, the integration of cul- tural and tourism industries was evaluated in terms of markets integration, supply chains integration and resources integration. Considering the consistency and availability, 14 indica- tors were selected from 5 aspects: total market revenue, reception volume, core industry income, number of core industries, and number of core resources (see Table 1). The results are shown in S1 Table. Research design Study area Located in the eastern part of China, Jiangsu is a major economic province nourished by the Yangtze River and the Grand Canal. It is home to several UNESCO world heritage sites, including China’s Grand Canal, Suzhou classical gardens, and Ming Xiaoling Mausoleum, as well as intangible cultural heritages like Kunqu Opera, woodblock and Yunjin embroidery. With a long history, profound culture and picturesque natural sceneries, each city in Jiangsu has its distinctive characters, making Jiangsu a huge tourism market in China. In 2019 (before Covid-19), Jiangsu received 3.9 million overseas tourists and earned $4.74 billion in foreign exchange earnings from tourism. During the May Day holiday in 2023 (the first major public holiday in China after the pandemic), Jiangsu received 39.8 million tourists and brought in tourism revenue of 9.96 billion yuan, ranking the first over the country. Besides, the “Charm of Jiangsu” brand has spread out of China as one of the top three most influential international tourism brands. As one of the most popular tourism destinations, Jiangsu has made outstand- ing achievements in cultural tourism festivals, performing arts programs and utilization of intangible cultural heritages. Table 1. Evaluation index system for the cultural industry and tourism industry. Cultural industry Markets integration Total revenue of cultural market Visitors of cultural activities Supply chains integration Operating income in cultural-related industries Employees in cultural-related industries Resources integration Number of libraries Number of museums Number of cultural centers Tourism industry Markets integration Total tourism revenue Number of domestic and foreign tourists Supply chains integration Sum of operating income in travel agencies, star hotels and A-grade tourist attractions Sum of employees in travel agencies, star hotels and A-grade tourist attractions Resource integration Number of travel agencies Number of star hotels Number of A-grade tourist attractions https://doi.org/10.1371/journal.pone.0287610.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 7 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain Model specification The data used in this paper covers 13 cities of Jiangsu Province over a period of 8 years. A fixed effect regression model with a controlling year effect was employed for analysis. Because heteroskedasticity, autocorrelation and cross-section relation might cause biased estimates, the regression with Driscoll-Kraay standard errors was used in the fixed effect model to obtain valid and consistent unbiased estimates as far as possible. First, the main regression model established for the impact of integration between cultural and tourism industries on the value added to tourism value chain is as follows. VALUEit ¼ a0 þ b1CCDit þ b2TAit þ lkControlsit þ ui þ ti þ εi ð10Þ Where i denotes the city and t denotes the year; a0 is a constant, and β1, β2, λk are regression parameters to be estimated; ui refers to the unobserved individual effect; ti refers to the year effect; εi is the random error term. Second, the panel threshold model is used to analyze the heterogeneity impact of the inte- gration of cultural and tourism industries on tourism value chain. According to the statistical effect, this model may categorize industrial integration into low and high phases by describing the leaping character or structural break in the link between different variables [40]. The panel threshold regression model is established as follows. VALUEit ¼ a0 þ g1CCDit � IðCCDit � dÞ þ g2CCDit � IðCCDit � dÞ þ b2TAit þ lkControlsit þ ui þ ti þ εi ð11Þ Third, to test the moderating effect of tourism agglomeration on the value added of indus- trial integration to tourism value chain, an interaction term (CCDTA) is introduced. The moderation analysis model is established as follows. VALUEit ¼ a0 þ b1CCDit þ b2TAit þ b3CCD∗TAit þ lkControlsit þ ui þ ti þ εi ð12Þ The mediating effect is an intermediate path underlying the effect of X to Y. Specifically, X influences the mediator variable M (path a is described by Eq (13)), which in turn influences Y (path b is described by Eq (14)). If the coefficients of a1 and b2 are statistically significant, the mediating effect can be established. 8 < : ITit ¼ i1 þ a1CCDit þ a2TAit þ akControlsit þ ui þ ti þ εi VALUEit ¼ i0 þ b1CCDit þ b2ITit þ b3TAit þ lkControlsit þ ui þ ti þ εi ð13Þ ð14Þ Finally, Instrumental Variable Estimates are widely used in solving almost all types of endogenous problems. In this paper, the two-stage least-squares (2SLS) estimates were employed for robust test. Variable measurement The integration level of cultural and tourism industries was taken as the main explanatory variable in this paper, which is calculated through the improved CCD as above. The CCD can reflect the process of benign interaction and synergistic development of the two industries. The value added to tourism value chain was selected as an explained variable. Hjalager [16] proposed to measure the value chain as the difference between turnover and costs, i.e., profit. Combining the concept of tourism value chain and the data from statistical yearbooks, the total tourism revenue can be considered as the total value paid by tourists, i.e., the turnover of PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 8 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain the tourism industry. The operating costs of the tourism industry can be reflected by the GDP contributed by the six elements of tourism (i.e., food, accommodation, transportation, sight- seeing, shopping, and entertainment) based on the expenditure method. The subtraction of the two derives the final value addition to tourism value chain. Therefore, the value added to tourism value chain can be calculated as follows. TVC ¼ TTR (cid:0) GDPi ð15Þ Where TVC refers to the value added to tourism value chain. TTR refers to the total tourism revenue. GDPi refers to the GDP of industry i which belongs to one of the six ele- ments of tourism. The Chinese System of National Account divides industries and measures GDP based on the production at the supply side, while the six elements of tourism are defined based on the tourists’ demand. Such a mismatch shows that tourism value chain is actually concealed in different economic departments [31]. For example, the transportation industry is involved for providing tourist transportation services; the accommodation and catering industry is involved for providing tourist catering and accommodation services; the retail industry is involved for providing tourist shopping services; and the entertain- ment and recreation industry is involved for providing tourist sightseeing and entertain- ment services. Therefore, Eq (15) can effectively reflect the value added to tourism value chain. As a moderated variable, tourism agglomeration reflects the scaling effect of tourism indus- try and the intensity of tourism activities. Thus, the total tourism revenue divided by regional GDP was used in this paper to measure tourism agglomeration [41, 42] Information technology is a mediating variable. From the industrial level, the postal busi- ness volume [29] has difficulty in reflecting the development level of 5G Internet. The Internet penetration rate [14] ignores the overall digital economy. Therefore, this paper used the ratio of the GDP of the information transmission, computer services and software industry to the total regional GDP as the proxy variable for information technology. A larger ratio refers to a higher development level of regional informatization and greater opportunities to apply infor- mation technology in other industries. The traffic passenger volume, the upgrading of industrial structure, and the government consumption were selected as control variables. The traffic passenger volume reflects the tour- ism traffic accessibility [43], which needs to be guaranteed by transportation infrastructure. The upgrading of industrial structure was measured by the industrial structure supererogation. An upgraded industrial structure can play a positive role in improving the efficiency of the tourism industry [44]. Government consumption reflects the government support for eco- nomic development, which is measured by the ratio of fiscal expenditure to regional GDP as a proxy variable [14]. Data source and descriptive statistics The data used in this paper is the panel data collected from 13 cities of Jiangsu province from 2013 to 2020 (Jiangsu began to conduct statistics on culture-related industries since 2013). As macroscopic statistical data is characterized by authenticity, objectivity and comparability, it is suitable for horizontal and vertical analyses. Thus, all the data used in this paper was directly collected or calculated from Jiangsu Statistics Yearbook, Jiangsu Culture and Tourism Statistics Yearbook, China Statistics Yearbook, and the Statistics Yearbooks of the 13 cities. The raw data used for analysis of this paper is as shown in S2 Table. The definitions and descriptive statistics on the variables used in this paper are shown in Table 2. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 9 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain Table 2. Definitions and descriptive statistics of the variables. Variable Measurement Explained Variable VALUE Total tourism revenue subtracts the GDP of the six elements of tourism Explanatory Variable CCD The CCD of cultural industry and tourism industry Moderating Variable TA Total tourism revenue divided by regional land area Mediating Variable IT GDP of the information technology industry divided by regional GDP Control Variable TRANS Logarithm of traffic volume GOV ISS Fiscal expenditure divided by regional GDP GDP of the tertiary industry divided by the GDP of the secondary industry https://doi.org/10.1371/journal.pone.0287610.t002 Obs Mean Std.Dev. Min Max 104 104 104 104 104 104 104 171.4 358.5 -445.5 1297 0.363 0.158 0.176 0.856 0.299 0.256 0.0341 1.248 0.022 0.016 0.011 0.093 9.043 0.124 1.037 0.653 0.0293 0.190 7.640 0.0851 0.763 10.810 0.200 1.785 Empirical findings and discussions The CCD of cultural and tourism industries Table 3 presents the CCD of cultural and tourism industries in 13 cities of Jiangsu province. According to Geng [36], the CCD can be classified into 8 categories. By observing the average value of the 13 cities (see S1 Fig), it was found that the coupling level of Jiangsu was basically balanced, but the coordination level was imbalanced. Only 3 cities achieved a balanced devel- opment between the two industries. Consequently, the integration between the cultural and tourism industries was still of a low quality in Jiangsu province. In particular, the development of the cultural industry generally lagged behind the tourism industry, which seriously restricted the progression towards high-level integration [9]. The low-level integration between the two industries is not conducive to the value added to the whole tourism value chain. Fan and Xue [9] found that the integration of cultural and tourism industries in Shaanxi had high and low-value clusters. Similarly, the integration of cultural and tourism industries in Jiangsu also showed a clear spatial difference (see S2 Fig), which is mainly attributed to the Table 3. The status of the integration between culture and tourism industries in Jiangsu province. City Nanjing Wuxi Suzhou Changzhou Xuzhou Nantong Yancheng Huaian Taizhou Suqian Yangzhou Zhenjiang Lianyungang Total CCD 0.738 0.533 0.521 0.435 0.370 0.367 0.303 0.293 0.283 0.249 0.211 0.207 0.205 0.363 Coupling Coordination Cultural industry Tourism industry Integration level 0.834 0.777 0.596 0.824 0.741 0.777 0.886 0.890 0.922 0.878 0.462 0.477 0.610 0.744 0.656 0.368 0.456 0.231 0.185 0.174 0.104 0.097 0.087 0.072 0.097 0.090 0.070 0.207 0.640 0.312 0.348 0.201 0.153 0.147 0.094 0.089 0.086 0.083 0.062 0.057 0.049 0.179 0.691 0.451 0.666 0.274 0.245 0.218 0.117 0.108 0.090 0.066 0.236 0.216 0.123 0.269 Intermediately coordinated Reluctantly coordinated Reluctantly coordinated Approaching imbalanced Slightly imbalanced Slightly imbalanced Slightly imbalanced Moderately imbalanced Moderately imbalanced Moderately imbalanced Moderately imbalanced Moderately imbalanced Moderately imbalanced Slightly imbalanced https://doi.org/10.1371/journal.pone.0287610.t003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 10 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain factors of resource endowment and industrial foundation [45]. The cities in Jiangsu can be divided into three tiers in terms of integration level. The first tier includes Nanjing, Wuxi and Suzhou, whose cultural and tourism industries have developed in good coordination due to their strong economy and profound historical culture. As a result, the integration between cul- tural and tourism industries has effectively promoted their industrial value addition and gener- ated positive spillover effects [12]. The second tier includes Changzhou, Xuzhou, Nantong and Yancheng, whose industrial integration is slightly imbalanced. In general, the development of their cultural industry lags slightly behind that of the tourism industry, but the potential of industrial integration is enormous. The third tier includes Huaian, Taizhou, Suqian, Yangzhou, Zhenjiang and Lianyungang, whose industrial integration is moderately imbalanced. In these cities, both the cultural and tourism industries are at a low development level, lacking of driving force from advantageous industries. Yangzhou and Zhenjiang are two exceptions in the third- tier cities, as the development of their tourism industry is at a much higher level than that of their cultural industry. However, the significant development gap between the two industries greatly weakens the value added to the tourism value chain. The basic goal of high-quality devel- opment is to achieve fair and mutual benefits [46]. The cultural and tourism industries must achieve a coordinated development for a high-level integration and high value addition. The impact of industrial integration on tourism value chain According to Table 4 Model 1 illustrates the relationship between the cultural and tourism integration and the value added to tourism value chain. It can be seen that the integration between cultural and tourism industries has contributed to the value added to tourism value chain at the 1% statistic level, with the regression coefficient of 1,252.262. In terms of the eco- nomic significance, every increase by one unit of the level of CCD will improve the value added to tourism value chain by 1,252.262 units. Therefore, hypothesis 1a was supported. Under global value chain governance, there are four approaches to increase the industry value: process upgrading, product upgrading, internal chain upgrading, and inter-industry upgrading [35, 47]. The integration between cultural and tourism industries is the embodi- ment of cross-industry restructuring, which can add value to the tourism value chain in three ways. Firstly, a high-consumption cultural tourism market has been developed, which has greatly increased the income from tourism-related industry. Secondly, the integration of cul- tural and tourism resources has added value by embedding diverse experience values and increasing the efficiency of resource allocation. And thirdly, the integration of cultural and tourism supply chains has enhanced the value added via the upgrading of industrial structures. Of course, the integration between cultural and tourism industries also has negative effects on the tourism value chain. Firstly, the culture and tourism integration advocates resources sharing between hosts and guests, which may exhaust local public resources for the local resi- dents, increase their costs of living and reduce their quality of life [48]. Secondly, the integra- tion may cause a surge in market entities in each link of the tourism value chain, and the homogeneity in cultural resources can easily lead to vicious competition, such as excessive commercialization. Thirdly, while pursuing economic benefits, the protection and inheritance of local culture may be neglected [49]. Commercialization of cultural resources through tour- ism development can easily distort the local culture, leading to the disappearance of cultural authenticity [50, 51]. The heterogeneity impact of industrial integration on tourism value chain As indicated in Table 4 the integration of cultural and tourism industries may be divided into two phases depending on the selected threshold values from Model 2. It can illustrate the PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 11 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain Table 4. Regression results of the impact of cultural and tourism integration on the tourism value chain and its mechanism. MODEL VARIABLES CCD CCD(CCD<0.4657) CCD(CCD>0.4657) DUM_CCD TA CCDTA IT TRANS GOV ISS CONSTANT R-squared Year control First stage of F F (1) FE 1,252.262* (4.60) (2) TR VALUE -447.891 (-0.92) 2499.591* (5.88) 1,051.395* 1,008.746* (15.94) (7.41) -1,373.425 -9,499.273* (-0.93) 5.030 (0.13) 226.967 (0.17) 21.508 (0.18) -645.358 (-1.37) 0.814 YES (-3.55) -29.467 (-0.72) -565.890 (-0.57) -105.704 (-0.51) 231.968 (0.46) 0.857 YES (3) MOD 1,126.785 (3.12) (4) MED1 IT (5) MED2 VALUE (6) IV 1,098.118 (2.23) 847.605* (43.50) 2,768.672* (5.27) -3,788.082* (-4.85) -37.734 (-1.49) 101.489 (0.11) -58.835 (-0.80) -56.073 (-0.17) 0.866 YES 0.010*** (3.28) -0.008* (-1.76) -0.003* (-1.77) -0.063* (-1.83) 0.070* (12.28) -0.007 (-0.38) 0.787 YES 176.446* (2.71) 716.235* (7.90) 1,037.821* (6.64) 9,885.303* -975.412 (4.78) -34.431 (-0.97) 1,884.945* (2.71) 244.451 (1.33) -425.578 (-1.20) 0.841 YES (-0.39) 7.245 (0.16) 371.272 (0.32) 49.324 (0.21) 0.814 YES 118.51* 25.81* 18509.54* 33.02* 8437609.86* 28.01* 36.55* t-statistics in parentheses * p<0.01 ** p<0.05 * p<0.1 https://doi.org/10.1371/journal.pone.0287610.t004 heterogeneous influence of industrial integration on the impact of value added to tourism value chain. From Table 5 the estimator of the single-threshold model is 0.467 (P = 0.022), while that of the double-threshold model fails to reject the null hypothesis (P = 0.440). It indicates that the integration of cultural and tourism industries has a single-threshold effect. Model 2 demon- strates that high-level integration of cultural and tourism industries (higher than 0.466) has a positive effect on the value added to tourism value chain, which has passed the significance of 1%-level test. Surprisingly, the low-level integration of cultural and tourism industries (lower Table 5. The results of panel threshold regression. Single threshold Double threshold Threshold 0.466 0.567 F 28.850** 17.320 confidence interval [0.463,0.467] [0.552,0.574] https://doi.org/10.1371/journal.pone.0287610.t005 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 12 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain than 0.466) may reduce the value added to tourism value chain although failing the statistical test. The results supported hypothesis 1b. The findings above suggest that the negative effect of the integration of cultural and tourism industries tends to predominate when the CCD of cultural and tourism industries is lower than 0.466. This threshold value is very close to that of Shi et al. [27] for high-quality cultural tourism in the Yangtze River Delta (0.46). Meanwhile, the integration of cultural and tourism industries may differ from other industries’ integration. For example, the integration between the AI and energy industries only needs to be coupled rather than coordinated to achieve a 20% increase in annual growth rate for the energy industry [52]. In comparison, the integra- tion between cultural and tourism industries has to be both coupled and coordinated to achieve sustainable development [53]. Nevertheless, the negative impact does not deny the rationality of the China’s policy of culture and tourism integration but rather confirms the need for in-depth researches. Accordingly, future researches should shift from the argument on whether to implement the policy of culture and tourism integration to how to promote the cultural and tourism industries toward high-level integration. The moderating effect of tourism agglomeration Model 3 (Table 4) with an additional interaction term coefficient (CCDTA) on the basis of Model 1, estimates the moderating effect of tourism agglomeration, and centralized treatment was performed to avoid multicollinearity. The results of Model 3 indicate that the interaction term coefficient (CCDTA) is 2768.672, with the P-value suggesting statistically significant. Meanwhile, the variables of TA and CCD have both passed the significance test and show the same direction, implying that tourism agglomeration plays an assisting role rather than a substituting role. Therefore, hypothesis 2 was supported. The moderating effect of tourism agglomeration suggests that tourism agglomeration can promote the cultural and tourism industries to the high-level integration. This is consistent with the conclusion of Yan et al. [54], which argued that the scaling effect of the logistics indus- try would enhance the coupling quality of the two industries. The industrial agglomeration [55] and agricultural agglomeration [56] have posed both a positive spillover effect and nega- tive crowding effect on economic sustainability. Compared to other industries, the negative crowding impact of tourism agglomeration is not obvious. This is because the value added of the tourism value chain is contributed from the tourists. The clustering of tourism-related companies can help enrich the types of tourism supply and reduce the average transportation cost [57]. Furthermore, tourism agglomeration stems from the attractiveness of tourism desti- nations, on the basis of strong attraction to enterprises and tourists [57]. The tourism flow formed by the gathering of tourists has an external spillover effect, which can promote the inte- gration of culture and tourism industries across multiple regions. The mediating effect of information technology The causal mediation analysis proposed by Imai et al. [58] accords with causal inference and is suitable for testing the causal mediation mechanism. Thereby, this paper applied the dummy variable of CCD (1, if higher than the threshold value; otherwise, 0) for exposure to treatment, and performed sensitivity analysis as robust test. As shown in Model 4 and Model 5 from Table 4 the test of CCD to IT (path a), and the joint test of CCD and IT to VALUE (path b) are both statistically significant. As shown in Table 6, the estimated coefficient of ACME is 106.334, with the 95% confidence interval [45.059,178.377] failing to include 0, implying that the integration of cultural and tourism industries partially promotes the value added to PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 13 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain Table 6. The results of information technology based on causal mediation analysis. ACME Direct effect Total effect Mediated proportion Mean 106.334 175.652 281.985 36.611% [95% Conf. Interval] [45.059, 178.377] [72.601, 312.298] [157.811, 401.878] [0.265, 0.674] https://doi.org/10.1371/journal.pone.0287610.t006 tourism value chain via information technology at the level of 36.611%. Therefore, hypothesis 3 was supported. Besides, sensitivity analysis indicates how an estimated quantity that violates the key assumption will change for different degrees [59]. A larger value of Rho corresponds to a greater difficulty in overturning the causal mediating effect. As shown in Fig 1, the value of Rho (0.45) is large enough to confirm the credibility of the causal mediating effect. Information technology can make up for the deficiency of the low-level integration of cul- tural and tourism industries. It is consistent with the view that digital technologies can pro- mote the sustainable development of cultural and tourism industries after crisis [3]. Different from the conclusion of Li et al. [13] that information technology played a mediating role in alleviating the impact of epidemic on cultural and tourism industries, the findings of this paper suggest that there may be a two-way causal relationship between information technology and the integration of cultural and tourism industries. On the one hand, information technol- ogy can promote the high-level integration of cultural and tourism industries and on the other hand, high-level integration stimulates the cultural and tourism industries to apply informa- tion technology in a more effective way, which in turn significantly adding value to the tourism value chain. Fig 1. Visual presentation of sensitivity analysis for causal mediation analysis. https://doi.org/10.1371/journal.pone.0287610.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 14 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain Endogeneity test The coupling degree was used as the instrumental variable because CCD is calculated based on it and the evaluation indicators of coupling degree are different from regression control vari- ables. Model 6 in Table 4 shows the results of 2SLS. The first-stage F statistic is 118.510, which needs to be higher than 104.7 for a conventional t-test [60]. The second-stage conventional t- test of CCD is 2.23 (bigger than 1.96, corresponding to the significance level of 5%). Therefore, the 2SLS model has passed the weak instrument test and over-identification test, indicating that IV has a strong explanatory power. Moreover, the P-value and the direction of CCD to VALUE are consistent with the fixed effect, implying that the empirical results are robust. Conclusions and implications The policy of cultural and tourism integration, as an important strategy in China’s 14th Five- year Plan, contributes significantly to the tourism high-quality development. The key to high- quality tourism development lies in value addition. Based on these, the paper explored how the integration of cultural and tourism industries affected the value added to tourism value chain. By building an industrial integration evaluation system based on resources integration, mar- kets integration and supply chains integration, this paper evaluated the integration between the cultural and tourism industries of 13 cities in Jiangsu, and applied four econometric mod- els to analyze the impact mechanisms. The following conclusions are drawn from the findings. Firstly, the integration of cultural and tourism industries in Jiangsu Province is generally at an imbalanced stage, with significant differences between the south and the north. Only three cities, namely Nanjing, Suzhou and Wuxi, have achieved high-level integration, all located in economically developed areas. It means that that not all cities are suitable for the policy of cul- tural and tourism integration, because the integration of cultural and tourism industries is ineffective in regions where the cultural industry is substantially less developed than the tour- ism industry. Secondly, the integration between cultural and tourism industries not only improves tour- ism value chain directly but also promotes its value added indirectly through information tech- nology. Information technology plays a mediating role in industrial integration since it can break down barriers across industries. Tourism agglomeration can intensify the promoting effect of cultural and tourism integration on the value added to the tourism value chain, and plays a positive moderating role due to the scaling effect. Thereby, information technology and tourism agglomeration can be used to adjust places that lack integration prerequisites. Thirdly, high-level integration provides a stronger value-added effect than low-level inte- gration. The tourism value chain may be inhibited when the development of the cultural industry lags behind that of the tourism industry by a significant gap. It seems obvious that the two industries cannot achieve high-level integration and maximize the benefits of industrial integration without engaging in synergistic development. Theory implications Firstly, the tourism value chain involves multi-stakeholder activities [18] and has indirect chain-reaction effects on other economic activities [22]. Therefore, it is difficult to define the division of products in the tourism industry and the value-added process of intermediate links. Based on the tourism value chain system and tourism satellite accounts [22], this paper successfully quantified the value added to the tourism value chain by industrial integration, which is equal to the total output of the tourism subsectors minus the total tourist consump- tion. This finding may contribute to the quantitative research of the tourism value chain. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 15 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain Secondly, Porter’s value chain model defines the actions of a firm that create value addition, while tourism is a demand-oriented business that caters to the needs of tourists. The value added to the tourism value chain can occur in four stages [61]. This paper revealed that the key to the value added to the tourism value chain is to enhance the tourist experience value, which is consistent with the idea of value shop [18]. In response to Hjalager’s horizontal coordination method based on the destination logic [16], this paper addressed the research gap between industrial integration and the tourism value chain by clarifying the relationship between the two. Thirdly, this paper extended the application of the threshold model to different stages of industrial integration, which are usually analyzed by the average level of CCD. In general, the threshold model is used to address the link between exposure and reaction, such as the rela- tionship between financial constraints and company investment decisions [62]. This model can automatically calculate the turning points in various stages and display the effects on the periods that follow. Therefore, it is suggested that the threshold model can be used to analyze the problems related to industrial integration in different stages. Policy implications According to the impact mechanism between the cultural and tourism integration and the value added to tourism value chain, information technology and tourism agglomeration are two key driving forces for high-level integration. This theoretically supports the government to implement incentives for the integration between cultural and tourism industries. Suppose a region lacks cultural resources but has a sizable tourist base. In that case, it may benefit from the economic value generated by the integration of cultural and tourism industries through the scaling effect and external spillover effect of tourism agglomeration. For regions with abun- dant cultural resources but an immature tourism market, information technology can be help- ful for quickly positioning and responding to the market. It can assist the actors at each node of the tourism value chain in obtaining and sharing market information, so that avoiding the crowding effect of tourism agglomeration. Industrial integration is a stepwise process, and value-added benefits will not be achieved from initial industrial integration but require constant and comprehensive restructuring. It enlightens the government to establish a hierarchical management mechanism for the gover- nance of both cultural and tourism industries. For regions at a low level of integration, the local government should consider implementing exit mechanisms and exploring other forms of tourism development, such as natural science tourism. For regions at a high level of integra- tion, incentive policies can be implemented vigorously to promote the orderly and sustainable development of cultural tourism integration. Limitations It should be admitted that this paper has certain limitations in terms of data sampling, model selection and value type. First, the study area is mainly concentrated in Jiangsu province, and the sample is relatively limited. Further studies are required to confirm whether the results of this paper can be generalized. Second, the possible nonlinear relationship between the integra- tion of cultural and tourism industries and the value added to tourism value chain in different stages was ignored. In the future, the negative effect of the low-level integration of cultural and tourism industries and the measures to reduce such effect should be clarified. Lastly, this paper mainly focused on the economic value of tourism value chain, while future research can address the social or other values of tourism value chain. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 16 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain Supporting information S1 Fig. Regional differences of coupling coordination among 13 cities in Jiangsu Province. (TIF) S2 Fig. Time trends of cultural tourism integration in 13 cities in Jiangsu Province. (TIF) S1 Table. Raw data of evaluation index system for the cultural industry and tourism indus- try. (XLSX) S2 Table. Raw data of regression analysis of cultural tourism integration on tourism value chain. (XLSX) Acknowledgments We are very grateful of anonymous reviewers for their insightful comments on this manu- script. We also would like to show our sincere gratitude to academic editors and journal edi- tors for their careful guidance and suggestions. Author Contributions Conceptualization: Meiling Zeng, Suyan Shen. Formal analysis: Meiling Zeng. Funding acquisition: Suyan Shen. Methodology: Meiling Zeng, Jie Gu. Writing – original draft: Meiling Zeng. Writing – review & editing: Meiling Zeng, Suyan Shen. References 1. Loulanski T, Loulanski V. The sustainable integration of cultural heritage and tourism: a meta-study. J Sustain Tour. 2011 Sep; 19(7):837–62. 2. Richards G. Creativity and tourism: The State of the Art. Ann Tour Res. 2011 Oct; 38(4):1225–53. 3. Zhao X, Xie C, Huang L, Wang Y, Han T. How digitalization promotes the sustainable integration of cul- ture and tourism for economic recovery. Econ Anal Policy. 2023 Mar 1; 77:988–1000. 4. Clay PM, Feeney R. Analyzing agribusiness value chains: a literature review. Int Food Agribus Manag Rev. 2019 Jan 28; 22(1):31–46. 5. Adiyia B, Stoffelen A, Jennes B, Vanneste D, Ahebwa WM. Analysing governance in tourism value chains to reshape the tourist bubble in developing countries: the case of cultural tourism in Uganda. J Ecotourism. 2015 Sep 2; 14(2–3):113–29. 6. Tejada P, Santos FJ, Guzma´ n J. Applicability of global value chains analysis to tourism: issues of gover- nance and upgrading. Serv Ind J. 2011 Aug; 31(10):1627–43. 7. McKercher B. Cultural tourism market: a perspective paper. Tour Rev. 2020; 75(1):126–9. 8. Scarlett HG. Tourism recovery and the economic impact: A panel assessment. Res Glob. 2021 Dec; 3:100044. 9. Fan T, Xue D. Spatial correlation of cultural industry and tourism industry in Shaanxi Province, China: LISA analysis based on coordination model. Asia Pac J Tour Res. 2020 Sep 1; 25(9):967–80. 10. Chang W. An Integrated Development Model of Cultural and Creative Industry and Rural Tourism Industry Based on Data Mining. Wirel Commun Mob Comput. 2022 Apr 25; 2022:2351906. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 17 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain 11. Chen H, Chen T, Li L, Chen X, Huang J. Testing Convergence of Tourism Development and Exploring Its Influencing Factors: Empirical Evidence from the Greater Bay Area in China. Sustainability. 2022 May 28; 14(11):6616. 12. 13. 14. Zhou Z, Yang Q, Kim DJ. An Empirical Study on Coupling Coordination between the Cultural Industry and Tourism Industry in Ethnic Minority Areas. J Open Innov Technol Mark Complex. 2020 Aug 20; 6 (3):65. Li X, Liang X, Yu T, Ruan S, Fan R. Research on the Integration of Cultural Tourism Industry Driven by Digital Economy in the Context of COVID-19—Based on the Data of 31 Chinese Provinces. Front Public Health. 2022 Mar 9; 10:780476. https://doi.org/10.3389/fpubh.2022.780476 PMID: 35356017 Zhou C, Sotiriadis M. Exploring and Evaluating the Impact of ICTs on Culture and Tourism Industries’ Convergence: Evidence from China. Sustainability. 2021 Oct 25; 13(21):11769. 15. Su Z, Aaron JR, McDowell WC, Lu DD. Sustainable Synergies between the Cultural and Tourism Indus- tries: An Efficiency Evaluation Perspective. Sustainability. 2019 Dec; 11(23):6607. 16. Hjalager AM, Tervo-Kankare K, Tuohino A. Tourism value chains revisited and applied to rural well- being tourism. Tour Plan Dev. 2016 Oct; 13(4):379–95. 17. Song H, Liu J, Chen G. Tourism Value Chain Governance: Review and Prospects. J Travel Res. 2013 Jan; 52(1):15–28. 18. Bakucz M. Tourism Value Chain Management as a Tool for Effective Tourism Destination Development The Case of Pe´ cs ECoC 2010. Acta Univ Danub Oeconomica. 2011; 7(3):46–63. 19. Spencer JP, Safari E, Dakora EAN. An evaluation of the tourism value-chain as an alternative to socio- economic development in Rwanda, Africa. Afr J Phys Health Educ Recreat Dance. 2014; 20(2:1):569– 83. 20. Zheng L, Wang H, Li G, Guo Y. Construction Scenario for a Rural Tourism Value Chain: A Case Study from Rural China. Am J Ind Bus Manag. 2021; 11(1):1–18. 21. Pesce D, Neirotti P, Paolucci E. When culture meets digital platforms: value creation and stakeholders’ alignment in big data use. Curr Issues Tour. 2019 Sep 14; 22(15):1883–903. 22. Mitchell J. Value chain approaches to assessing the impact of tourism on low-income households in developing countries. J Sustain Tour. 2012; 20(3):457–75. 23. Rylance A, Spenceley A. Reducing economic leakages from tourism: A value chain assessment of the tourism industry in Kasane, Botswana. Dev South Afr. 2017 May 4; 34(3):295–313. 24. 25. Zhu Y. An Economic Model for Studying the Role of Cultural Industries on Social Development in Cross-border Contexts. Emerg Mark Finance Trade. 2020 May 27; 56(7):1581–600. Zhou WZ, Xu YY, Zhou ZC. Management systems of flower-themed tourism in China: a value chain analysis. Axelson LE, Fernqvist F, editors. Xviii Int Symp Hortic Econ Manag. 2016; 1132:113–20. 26. Wang G, Ye L. Spatial-Temporal Pattern of Mismatch Degree of High-Quality Tourism Development and Its Formation Mechanism in Taihu Lake Basin, China. Sustainability. 2022 Apr; 14(8):4812. 27. Shi Z, Xu D, Xu L. Spatiotemporal characteristics and impact mechanism of high-quality development of cultural tourism in the Yangtze River Delta urban agglomeration. Plos One. 2021 Jun 22; 16(6): e0252842. https://doi.org/10.1371/journal.pone.0252842 PMID: 34157034 28. Hu Y. The Integration and Development path of Culture and Tourism Industry based on Digital Technol- ogy Enablement. J Commer Econ. 2022;(1):182–4. 29. Li Z, Liu H. How tourism industry agglomeration improves tourism economic efficiency? Tour Econ. 2021 May 10;135481662110091. 30. Adachi Y. Applicability of agglomeration to tourism economics. Jpn World Econ. 2018 Sep; 47:58–67. 31. Yang Y. Agglomeration density and tourism development in China: An empirical research based on dynamic panel data model. Tour Manag. 2012 Dec; 33(6):1347–59. 32. Xiang Z. From digitization to the age of acceleration: On information technology and tourism. Tour Manag Perspect. 2018 Jan; 25:147–50. 33. 34. Tajzadeh-Namin A. A review on value creation in tourism industry. Manag Sci Lett. 2012 Jan 1; 2 (1):203–12. Law R, Buhalis D, Cobanoglu C. Progress on information and communication technologies in hospitality and tourism. Int J Contemp Hosp Manag. 2014; 26(5):727–50. 35. Pietrzak M, Chlebicka A, Kraciński P, Malak-Rawlikowska A. Information Asymmetry as a Barrier in Upgrading the Position of Local Producers in the Global Value Chain—Evidence from the Apple Sector in Poland. Sustainability. 2020 Sep 23; 12(19):7857. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 18 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain 36. Geng Y, Wei Z, Zhang H, Maimaituerxun M. Analysis and Prediction of the Coupling Coordination Rela- tionship between Tourism and Air Environment: Yangtze River Economic Zone in China as Example. Discrete Dyn Nat Soc. 2020 Jan 31; 2020:1–15. 37. Wang S jia, Kong W, Ren L, Zhi D dan, Dai B ting. Research on misuses and modification of coupling coordination degree model in China. J Nat Resour. 2021; 36(3):793. 38. Shen L, Huang Y, Huang Z, Lou Y, Ye G, Wong SW. Improved coupling analysis on the coordination between socio-economy and carbon emission. Ecol Indic. 2018 Nov; 94:357–66. 39. Li J, Xia J, Zuo Y, Cui J, Qiu Q, Liu X, et al. Spatiotemporal Evolution Patterns and Driving Factors of Synergistic Development of Culture, Sports, and Tourism Industries: The Case Study of China. Math Probl Eng. 2021 Nov 26; 2021:1–13. 40. Wang Q. Fixed-effect panel threshold model using Stata. Stata J Promot Commun Stat Stata. 2015; 15 (1):121–34. 41. Wu Q, Zhang C, Wang H, Hao J. Study on the relationship between agglomeration of service industry and economic growth in Yangtze River Delta based on spatial econometric models. J Phys Conf Ser. 2019 Oct 1; 1324(1):012089. 42. Wang J, Ma J. Has Tourism Industry Agglomeration Improved the Total Factor Productivity of Chinese Urban Agglomerations?—The Moderating Effect of Public Epidemic. Front Public Health. 2022 Mar 17; 10:854681. https://doi.org/10.3389/fpubh.2022.854681 PMID: 35372188 43. 44. 45. Tang R, Xiu P. The modern service industry agglomeration and tourism efficiency in China: regional dif- ference and influencing mechanism. J Asia Pac Econ. 2021 Jul 27;1–20. Li Y, Li R, Ruan W, Liu CH. Research of the Effect of Tourism Economic Contact on the Efficiency of the Tourism Industry. Sustainability. 2020 Jul 14; 12(14):5652. Feng J, Wang C, Zhao Y. Measurement and causes of cultural industry agglomeration in Jiangsu Prov- ince. AIP Conf Proc. 2023 May 5; 2685(1):040021. 46. Shen W, Huang Z, Yin S, Hsu WL. Temporal and Spatial Coupling Characteristics of Tourism and Urbanization with Mechanism of High-Quality Development in the Yangtze River Delta Urban Agglom- eration, China. Appl Sci-Basel. 2022 Apr; 12(7):3403. 47. Kadarusman Y, Nadvi K. Competitiveness and Technological Upgrading in Global Value Chains: Evi- dence from the Indonesian Electronics and Garment Sectors. Eur Plan Stud. 2013 Jul; 21(7):1007–28. 48. Wu Z, Lai IKW, Tang H. Evaluating the Sustainability Issues in Tourism Development: An Adverse- Impact and Serious-Level Analysis. Sage Open. 2021 Oct; 11(4):21582440211050384. 49. Qu C, Timothy DJ, Zhang C. Does tourism erode or prosper culture? Evidence from the Tibetan ethnic area of Sichuan Province, China. J Tour Cult Change. 2019 Jul 4; 17(4):526–43. 50. 51. Intason M, Lee C, Coetzee W. Examining the interplay between a hallmark cultural event, tourism, and commercial activities: A case study of the Songkran Festival. J Hosp Tour Manag. 2021 Dec; 49:508– 18. Zhang SN, Ruan WQ, Li YQ, Huang H. Local cultural distortion risk at tourist destinations: connotation deconstruction and theoretical construction. Curr Issues Tour [Internet]. 2023 [cited 2023 Apr 30]; Avail- able from: https://www.webofscience.com/wos/woscc/full-record/WOS:000939508000001 52. Dong F, Zhang S, Zhu J, Sun J. The Impact of the Integrated Development of AI and Energy Industry on Regional Energy Industry: A Case of China. Int J Environ Res Public Health. 2021 Aug 25; 18(17):8946. https://doi.org/10.3390/ijerph18178946 PMID: 34501536 53. Środa-Murawska S, Grzelak-Kostulska E, Biegańska J, Dąbrowski LS. Culture and Sustainable Tour- ism: Does the Pair Pay in Medium-Sized Cities? Sustainability. 2021 Aug 13; 13(16):9072. 54. Yan B, Dong Q, Li Q, Yang L, Amin FU I. A study on the interaction between logistics industry and manufacturing industry from the perspective of integration field. Plos One. 2022 Mar 3; 17(3): e0264585. https://doi.org/10.1371/journal.pone.0264585 PMID: 35239696 55. Han C, Hua D, Li J. A View of Industrial Agglomeration, Air Pollution and Economic Sustainability from Spatial Econometric Analysis of 273 Cities in China. Sustainability. 2023 Jan; 15(9):7091. 56. Zhang H, Zhang J, Song J. Analysis of the threshold effect of agricultural industrial agglomeration and industrial structure upgrading on sustainable agricultural development in China. J Clean Prod. 2022; 341:130818. 57. Majewska J. Inter-regional agglomeration effects in tourism in Poland. Tour Geogr. 2015 May 27; 17 (3):408–36. 58. Imai K, Keele L, Tingley D. A general approach to causal mediation analysis. Psychol Methods. 2010; 15(4):309–34. https://doi.org/10.1037/a0020761 PMID: 20954780 59. Hicks R, Tingley D. Causal mediation analysis. Stata J Promot Commun Stat Stata. 2011; 11(4):605– 19. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 19 / 20 PLOS ONEHow does the integration of cultural and tourism industries impact the value added to tourism value chain 60. Lal A, Lockhart MW, Xu Y, Zu Z. How Much Should We Trust Instrumental Variable Estimates in Politi- cal Science? Practical Advice based on Over 60 Replicated Studies. SSRN Electron J. 2023; (2303):11399. 61. Yilmaz Y, Bititci U. Performance measurement in the value chain: manufacturing v. tourism. Int J Prod- uct Perform Manag. 2006 Jul 1; 55(5):371–89. 62. Hansen BE. Threshold effects in non-dynamic panels: Estimation, testing, and inference. J Econom. 1999 Dec 1; 93(2):345–68. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287610 June 29, 2023 20 / 20 PLOS ONE
10.1371_journal.ppat.1009495	RESEARCH ARTICLE Fatty acid oxidation participates in resistance to nutrient-depleted environments in the insect stages of Trypanosoma cruzi 1, Fla´ via Silva DamascenoID Rodolpho Ornitz Oliveira SouzaID Marc Biran2, Gilson Murata3, Rui Curi3,4, Fre´ de´ ric BringaudID 1, Sabrina Marsiccobetre1, 1* 5, Ariel Mariano SilberID 1 University of São Paulo, Laboratory of Biochemistry of Tryps–LaBTryps, Department of Parasitology, Institute of Biomedical Sciences–São Paulo, São Paulo, Brazil, 2 Centre de Re´ sonance Magne´tique des Systèmes Biologiques (RMSB), Universite´ de Bordeaux, Bordeaux, France, 3 University of São Paulo, Department of Physiology, Institute of Biomedical Sciences–São Paulo, São Paulo, Brazil, 4 Cruzeiro do Sul University, Interdisciplinary Post-Graduate Program in Health Sciences—São Paulo, São Paulo, Brazil, 5 Laboratoire de Microbiologie Fondamentale et Pathoge´ nicite´ (MFP), Universite´ de Bordeaux, Bordeaux, France * asilber@usp.br Abstract Trypanosoma cruzi, the parasite causing Chagas disease, is a digenetic flagellated protist that infects mammals (including humans) and reduviid insect vectors. Therefore, T. cruzi must colonize different niches in order to complete its life cycle in both hosts. This fact deter- mines the need of adaptations to face challenging environmental cues. The primary environ- mental challenge, particularly in the insect stages, is poor nutrient availability. In this regard, it is well known that T. cruzi has a flexible metabolism able to rapidly switch from carbohy- drates (mainly glucose) to amino acids (mostly proline) consumption. Also established has been the capability of T. cruzi to use glucose and amino acids to support the differentiation process occurring in the insect, from replicative non-infective epimastigotes to non-replica- tive infective metacyclic trypomastigotes. However, little is known about the possibilities of using externally available and internally stored fatty acids as resources to survive in nutri- ent-poor environments, and to sustain metacyclogenesis. In this study, we revisit the meta- bolic fate of fatty acid breakdown in T. cruzi. Herein, we show that during parasite proliferation, the glucose concentration in the medium can regulate the fatty acid metabo- lism. At the stationary phase, the parasites fully oxidize fatty acids. [U-14C]-palmitate can be taken up from the medium, leading to CO2 production. Additionally, we show that electrons are fed directly to oxidative phosphorylation, and acetyl-CoA is supplied to the tricarboxylic acid (TCA) cycle, which can be used to feed anabolic pathways such as the de novo biosyn- thesis of fatty acids. Finally, we show as well that the inhibition of fatty acids mobilization into the mitochondrion diminishes the survival to severe starvation, and impairs metacyclogenesis. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Souza ROO, Damasceno FS, Marsiccobetre S, Biran M, Murata G, Curi R, et al. (2021) Fatty acid oxidation participates in resistance to nutrient-depleted environments in the insect stages of Trypanosoma cruzi. PLoS Pathog 17(4): e1009495. https://doi.org/10.1371/journal. ppat.1009495 Editor: Michael L. Ginger, University of Huddersfield, UNITED KINGDOM Received: January 15, 2021 Accepted: March 23, 2021 Published: April 5, 2021 Copyright: © 2021 Souza et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. Funding: This work was supported by Fundac¸ão de Amparo à Pesquisa do Estado de São Paulo (www. fapesp.br) grants 2016/06034-2 and 2018/14432-3 (awarded to AMS); Conselho Nacional de Desenvolvimento Cientı´fico e Tecnolo´gico (www. cnpq.br/) grants 308351/2013-4 and 404769/ 2018-7 (awarded to AMS), Research Council PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 1 / 25 Fatty acid metabolism in insect stages of Trypanosoma cruzi United Kingdom Global Challenges Research Fund under grant agreement “A Global Network for Neglected Tropical Diseases” (https://www.ukri. org/research/global-challenges-research-fund/) (grant MR/P027989/1) (awarded to AMS). In addition, FB was supported by the Universite´ de Bordeaux (https://www.u-bordeaux.fr/), the Centre National de la Recherche Scientifique (CNRS, http://www.cnrs.fr/), the Agence Nationale de la Recherche (ANR, http://www.agence-nationale- recherche.fr) through GLYCONOV and ADIPOTRYP grants of the "Ge´ne´rique » call and the Laboratoire d’Excellence (LabEx) ParaFrap ANR-11-LABX- 0024. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Author summary Trypanosoma cruzi is a protist parasite with a life cycle involving two types of hosts, a ver- tebrate one (which includes humans, causing Chagas disease) and an invertebrate one (kissing bugs, which vectorize the infection among mammals). In both hosts, the parasite faces environmental challenges such as sudden changes in the metabolic composition of the medium in which they develop, severe starvation, osmotic stress and redox imbalance, among others. Because kissing bugs feed infrequently in nature, an intriguing aspect of T. cruzi biology (it exclusively inhabits the digestive tube of these insects) is how they subsist during long periods of starvation. In this work, we show that this parasite performs a met- abolic switch from glucose consumption to lipid oxidation, and it is able to consume lipids and the lipid-derived fatty acids from both internal origins as well as externally supplied compounds. When fatty acid oxidation is chemically inhibited by etomoxir, a very well- known drug that inhibits the translocation of fatty acids into the mitochondria, the prolif- erative insect stage of the parasites has dramatically diminished survival under severe met- abolic stress and its differentiation into its infective forms is impaired. Our findings place fatty acids in the centre of the scene regarding their extraordinary resistance to nutrient- depleted environments. Introduction T. cruzi, a flagellated parasite, is the causative agent of Chagas disease, a neglected health prob- lem endemic to the Americas [1]. The parasite life cycle is complex, alternating between repli- cative and non-replicative forms in two types of hosts, mammalians and triatomine insects [2]. In mammalian hosts, two primary forms are recognized: replicative intracellular amastigotes and nondividing trypomastigotes, which are released from infected host cells into the extracel- lular medium. After being released from infected cells, trypomastigotes can spread the infec- tion by infecting new cells, or they can be ingested by a triatomine bug during its blood meal. Once inside the invertebrate host, the ingested trypomastigotes differentiate into epimasti- gotes, which initiate their proliferation and colonization of the insect digestive tract [3]. Once the epimastigotes reach the final portion of the digestive tube, they initiate differentiation into non-proliferative, infective metacyclic trypomastigotes. These forms will be expelled during a new blood meal and will be able to infect a new vertebrate host [2,4–6]. The diversity of environments through which T. cruzi passes during its life cycle (i.e., the digestive tube of the insect vector, the bloodstream and the mammalian cells cytoplasm) sub- jects it to different levels of nutrient availability [3,7]. Therefore, this organism evolved a robust, flexible and efficient metabolism [4,8]. As an example, it was recognized early on that epimastigotes are able to rapidly switch their metabolism, allowing the consumption of carbo- hydrates and different amino acids [9,10]. Several studies identified aspartate, asparagine, glu- tamate [11], proline [12–14], histidine [15], alanine [11,16] and glutamine [11,17] as oxidisable energy sources. T. cruzi glycolysis is by far the most studied metabolic pathway in trypanosomatids. It hap- pens in a specific peroxisome-related organelles named glycosomes and the cytosol [18,19]. The absence of the classical controls by ATP and O2 (Pasteur effect) is compensated by such compartmentalization [18]. It has been suggested that glycosomes do not exchange ATP and glycolytic cofactors such as NAD (H) with the cytosol [20]. Therefore, the intra-glycosomal ATP/ADP ratio prevents the deleterious effects of the uncontrolled activation of ATP-depen- dent glycolytic enzymes (hexokinase and phosphofructokinase) by the surplus ATP which is PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 2 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi produced in the cytoplasm, the denominated “turbo design” [21]. In addition, many pathways contribute to maintain the intra-glycosomal redox balance. A remarkable one consists in the conversion of phosphoenolpyruvate into succinate, a NADH reoxidizing pathway which includes a cytosolic shunt, involving a cytosolic fumarase, and finally a glycosomal fumarate reductase [22]. The last two enzymes are responsible for part of the reoxidation of NADH [22]. This pathway produces significant amounts of succinate, a main end product of T. cruzi glycol- ysis. Pyruvate can be also a substrate of transaminases as a –NH2 acceptor, resulting in forma- tion of alanine that is exported from the cell, or can be imported into the mitochondrion where the pyruvate dehydrogenase complex can convert it into Acetyl-CoA and subsequently into acetate, which can be excreted. Acetyl-CoA can be also incorporated into the TCA cycle, to produce reducing equivalents for the mitochondrial respiratory chain and carbon skeletons for producing other metabolites. Importantly, there is a mitochondrial isoform of fumarate reductase, which, as happens with its glycosomal counterpart, is able to convert mitochondrial fumarate into succinate. Much less is known about how T. cruzi uses fatty acids and how these compounds contribute to the parasite´s metabolism and survival. The NADP+-dependent β- oxidation of palmitoyl-CoA has been shown in mitochondrial and glycosomal fractions of T. cruzi epimastigotes [23], and a recent proteomic analysis of glycosomes showed the presence of all the required enzymes for this process [24]. In this study, we explore fatty acid metabolism in T. cruzi. We also address fatty acid regulation by external glucose levels and the involvement of their oxidation in the replication and differentiation of T. cruzi insect stages. Methods Parasites Epimastigotes of T. cruzi strain CL clone 14 were maintained in the exponential growth phase by sub-culturing them for 48 h in Liver Infusion Tryptose (LIT) medium at 28˚C [25]. Meta- cyclic trypomastigotes were obtained through the differentiation of epimastigotes at the sta- tionary growth phase by incubation for 2 h in TAU (Triatomine Artificial Urine) medium (190 mM NaCl, 8 mM phosphate buffer pH 6.0, 17 mM KCl, 2 mM CaCl2, 2 mM MgCl2) fol- lowed by an incubation for 6 days in TAU-3AAG (TAU supplemented with 10 mM proline, 50 mM glutamate, 2 mM aspartate and 10 mM glucose) as previously reported [17], or TAU- 3AAG supplemented with 5, 10, 25 or 50 mM etomoxir (ETO). Fatty acid oxidation assays Preparation of palmitate-BSA conjugates. Sodium palmitate at 70 mM was solubilized in water by heating it up to 70˚C. BSA free fatty acids (FFA BSA) (Sigma Aldrich) was dis- solved in PBS and warmed up to 37˚C with continuous stirring. Solubilized palmitate was added to BSA at 37˚C with continuous stirring (for a final concentration of 5 mM in 7% BSA). The conjugated palmitate-BSA was aliquoted and stored at −80˚C [26]. CO2 production from oxidisable carbon sources. To test the production of CO2 from palmitate, glucose or histidine, exponentially growing epimastigotes (5x107 mL-1) were washed twice in PBS and incubated for different times (0, 30, 60 and 120 min) in the presence of 0.1 mM of palmitate spiked with 0.2 μCi of 14C-U-substrates. To trap the produced CO2, What- man paper was embedded in 2 M KOH solution and was placed in the top of the tube. The 14CO2 trapped by this reaction was quantified by scintillation [15,16]. 1H-NMR analysis of the exometabolome. Epimastigotes (1x108 mL-1) were collected by centrifugation at 1,400 x g for 10 min, washed twice with PBS and incubated in 1 mL (single point analysis) of PBS supplemented with 2 g/L NaHCO3 (pH 7.4). The cells were maintained for 6 h at 27˚C in incubation buffer containing [U-13C]-glucose, non-enriched palmitate or no PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 3 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi carbon sources. The integrity of the cells during the incubation was checked by microscopic observation. The supernatant (1 mL) was collected and 50 μl of maleate solution in D2O (10 mM) was added as an internal reference. 1H-NMR spectra were collected at 500.19 MHz on a Bruker Avance III 500 HD spectrometer equipped with a 5 mm Prodigy cryoprobe. The mea- surements were recorded at 25˚C. The acquisition conditions were as follows: 90˚ flip angle, 5,000 Hz spectral width, 32 K memory size, and 9.3 sec total recycling time. The measurements were performed with 64 scans for a total time of close to 10 min and 30 sec. The resonances of the obtained spectra were integrated and the metabolite concentrations were calculated using the ERETIC2 NMR quantification Bruker program. Oxygen consumption. To evaluate the importance of internal fatty acid sources in O2 consumption, exponentially growing parasites were treated or not treated with 500 μM ETO (the inhibitor of carnitine palmitoyltransferase 1), washed twice in PBS and resuspended in Mitochondrial Cellular Respiration (MCR) buffer. The rates of oxygen consumption were measured using intact cells in a high-resolution oxygraph (Oxygraph-2k; Oroboros Instru- ments, Innsbruck, Austria). Oligomycin A (0.5 μg/mL) and FCCP (0.5 μM) were sequentially added to measure the optimal non-coupled respiration and the respiration leak state, respec- tively. The data were recorded and treated using DatLab 7 software [15,16,27]. Mitochondrial activity assays MTT. The parasites were washed twice and incubated in PBS supplemented with 0.1 mM palmitate in 0.35% FFA BSA, 0.35% FFA BSA alone, and 5 mM glucose, and 5 mM histidine or not supplemented media were used as controls (positives and negative, respectively). The cell via- bility was evaluated at 24 h and 48 h after incubation using the MTT assay, as described in [15,16]. Alamar blue. The parasites were washed twice and incubated in PBS or PBS supple- mented with 500 μM ETO in 96-well plates. The plates were maintained at 28˚C during all the experiments. After every 24 h, the cells were incubated with 0.125 μg.mL-1 of Alamar Blue reagent and kept at 28˚C for 2 h under protection from light. The fluorescence was accessed using the wavelengths λexc = 530 nm and λem = 590 nm in the SpectraMax i3 (Molecular Devices) plate reader [17]. Measurement of intracellular ATP content The intracellular ATP levels were assessed using a luciferase assay kit (Sigma-Aldrich), as described in [15–17]. In brief, the parasites were incubated in PBS supplemented (or not) with 0.1 mM palmitate, 0.35% FFA BSA, 5 mM glucose or 5 mM histidine for 24 h at 28˚C. The ATP concentrations were determined by using a calibration curve with ATP disodium salt (Sigma), and the luminescence at 570 nm was measured as indicated by the manufacturer. Enzymatic activities Carnitine palmitoyltransferase 1 (CPT1). The epimastigotes were washed twice in PBS (1,000 x g, 5 min at 4˚C), resuspended in buffered Tris-EDTA (100 mM, 2.5 mM and 0.1% Tri- ton X-100) containing 1 μM phenylmethyl-sulphonyl fluoride (PMSF), 0.5 mM N-alpha-p- tosyl-lysyl-chloromethyl ketone (TLCK), 0.01 mg aprotinin and 0.1 mM trans-epoxysuccinyl- L-leucyl amido (4-guanidino) butane (E-64) as a protease inhibitors (Sigma Aldrich) and lysed by sonication (5 pulses for 1 min each, 20%). The lysates were clarified by centrifugation at 10,000 x g for 30 min at 4˚C [28]. The soluble fraction was collected and the proteins were quantified by Bradford method [29] and adjusted to 0.1 mg/mL protein. The reaction mixture contained 0.5 mM L-carnitine, 0.1 mM palmitoyl-CoA and 2.5 mM DTNB in Tris-EDTA buffer (pH = 8.0). The CPT1 activity was measured spectrophotometrically at 412 nm by PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 4 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi DTNB reaction with free HS-CoA, forming the TNB- ion. To calculate the specific activity, the absorbance values were converted into molarity by using the TNB- extinction molar coefficient of 12,000 M-1.s-1 [30]. As a blank, we performed the same assay without adding the substrate. All the enzymatic assays were performed in 96-well plates at a final volume of 0.2 mL in the SpectraMax i3 (Molecular Devices). Acetyl-CoA carboxylase (ACC). The ACC activity was measured spectrophotometrically by coupling its enzymatic reaction with that of citrate synthase (CS), which uses oxaloacetate and ace- tyl-CoA to produce citrate. Measurements were performed at the end-points in two steps. First, the reaction mixture contained 100 mM potassium phosphate buffer (pH = 8.0), 15 mM KHCO3, 5 mM MnCl2, 5 mM ATP, 1 mM acetyl-CoA and 0.1 μM biotin. The reaction was initiated by adding 0.1 mg of cell extract and developed using 15 min incubations at 28˚C. The reaction was stopped by adding perchloric acid 40% (v/v) and centrifuged 10,000 x g for 15 min at 4˚C. The second reac- tion was performed by using 0.1 mL of the supernatant from the first reaction, 20 mM oxaloacetate and 0.5 mM of DTNB in 100 mM potassium phosphate buffer (pH = 8.0). The reaction was initi- ated by adding 0.5 units of CS (Sigma Aldrich). To calculate the specific activity of ACC, we con- verted the absorbance values to molarity by using the TNB- extinction molar coefficient of 12,000 M-1.s-1. For the blank reaction, we performed the same assay without acetyl-CoA [31]. Hexokinase (HK). The HK activity was measured as described in [32]. Briefly, the activity was measured by coupling the hexokinase activity with a commercial glucose-6-phosphate dehydrogenase, which oxidizes the glucose-6-phosphate (G6PD, SIGMA) resulting from the HK activity with the concomitant reduction of NADP+ to NADPH. The resulting NADPH was spectrophotometrically monitored at 340 nm. The reaction mixture contained 50 mM Triethanolamine buffer pH 7.5, 5 mM MgCl2, 100 mM KCl, 10 mM glucose, 5 mM ATP and 5 U of commercial G6PD. To calculate the specific activity, the absorbance values were con- verted to molarity using the NADP(H) extinction molar coefficient of 6,220 M-1.s-1. Serine palmitoyltransferase (SPT). The SPT activity was measured through the reduc- tion of the DTNB reaction by the free HS-CoA, forming the TNB- ion, which was measured spectrophotometrically at 412 nm as previously described [30]. In brief, the epimastigotes were washed twice in PBS, resuspended in Tris-EDTA buffer (100 mM/2.5 mM) containing Triton X-100 0.1% and lysed by sonication (20% of potency, during 2 min). The reaction mixture contained 0.1 mg of protein free-cell extract, 0.5 mM L-serine, 0.1 mM palmitoyl-CoA and 2.5 mM DTNB in Tris-EDTA buffer (100 mM/2.5 mM) pH = 8.0 [33]. To calculate the specific activity, we converted the absorbance values to molarity using the TNB- extinction molar coef- ficient of 12,000 M-1.s-1. For the blank reaction, we performed the same assay without adding palmitoyl-CoA. All the enzymatic assays were performed in 96-well plates in a final volume of 0.2 mL in the SpectraMax i3 (Molecular Devices). Glucose and triglyceride quantification Spent LIT medium from epimastigote cultures was collected by recovering the supernatants from a centrifugation (10,000 x g for 15 min at 4˚C). Each sample of spent LIT was analysed for its glucose and triglyceride contents using commercial kits (triglyceride monoreagent and glucose monoreagent by Bioclin Brazil) according to the manufacturer’s instructions. These kits are based on colorimetric enzymatic reactions, and the absorbance of each assay was mea- sured in 96-well plates at a final volume of 0.2 mL in the SpectraMax i3 (Molecular Devices). Proliferation assays Exponentially growing T. cruzi epimastigotes (5x107 mL-1) were treated with different concen- trations of ETO or not treated (negative control) in LIT medium. As a positive control for PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 5 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi growth inhibition, we used a combination of rotenone (60 μM) and antimycin (0.5 μM) [34]. The parasites (2.5x106 mL-1) were transferred to 96-well plates and then incubated at 28˚C. The cell proliferation was quantified by reading the optical density (OD) at 620 nm for eight days. The OD values were converted to cell numbers using a linear regression equation previ- ously obtained under the same conditions. Each experiment was performed in quadruplicate [35]. Flow cytometry analyses Cell death. Epimastigotes in the exponential phase of growth were maintained in LIT and treated with ETO 500 μM for 5 days. After the incubation time, the parasites were analysed as described in [35]. The cells were analysed by flow cytometry (FACScalibur BD Biosciences). Cell cycle (DNA content). Epimastigotes in the exponential phase of growth were main- tained in LIT and treated with ETO 500 μM over 5 days. After the incubation time, the para- sites were washed twice in PBS and resuspended in lysis buffer (phosphate buffer Na2HPO4 7.7 mM; KH2PO4 2.3 mM; pH = 7.4) and digitonin 64 μM. After incubating on ice for 30 min, propidium iodide 0.2 μg/mL was added. The samples were analysed by flow cytometry (Guava) adapted from [36]. Fatty acid staining using BODIPY 500/510. Exponentially growing epimastigotes were kept in LIT medium to reach three different cell densities (2.5x107 mL-1, 5x107 mL-1 and 108 mL-1) in 24-well plates at 28˚C. Twenty-four hours before the flow cytometry analysis, the par- asites were treated with 1 μM C1-BODIPY 500/510-C12. This fluorophore allows for measure- ments of the relationship between fatty acid accumulation and consumption by shifting the fluorescence filter. The samples were collected, washed twice in PBS and incubated in 4% para- formaldehyde for 15 min. After incubation, the cells were washed twice with PBS and sus- pended in the same buffer. Flow cytometry analysis was performed with FL-1 and FL-2 filters in a FACS Fortessa DB. The results were analysed using FlowJo software. Fluorescence microscopy The parasites were maintained in LIT medium as previously reported for fatty acid staining using BODIPY 500/510. After incubation, the cells were washed twice in PBS and placed on glass slides. The images were acquired with a digital DFC 365 FX camera coupled to a DMI6000B/AF6000 microscope (Leica). The images were analysed using ImageJ software. Results Palmitate supports ATP synthesis in T. cruzi We initially investigated the ability of T. cruzi epimastigotes to oxidize fatty acids (for a scheme see Fig 1A). To this end, we used palmitate as a proxy for fatty acids in general. The parasites were incubated with 0.1 mM 14C-[U]-palmitate, which allowed us to measure the production of 1.3 nmoles of CO2 derived from palmitate oxidation during the first 60 min and 1.5 extra nmoles during the following 60 min (Fig 1B). This finding indicated that beta-oxidation and the further ‘burning’ of the resulting acetyl-CoA is operative in epimastigote mitochondria. Because palmitate is taken up from extracellular medium and oxidized to CO2, it is reasonable to assume that it could contribute to resistance to severe nutritional stress. To support this idea, we tested the ability of palmitate to extend parasite survival under extreme nutritional stress. Parasites were incubated for 24 and 48 h in PBS (negative control, in this condition we expected the lower viability after the incubations), 0.1 mM palmitate in PBS supplemented with BSA (as a palmitate carrier), 5.0 mM histidine in PBS or 5.0 mM glucose in PBS (both PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 6 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi Fig 1. Palmitate oxidation promotes ATP production and viability in epimastigote forms under starvation. A) Schematic representation of 14C-U-palmitate metabolism. The metabolites corresponding to labelled palmitate metabolism are presented in green. B) 14CO2 production from epimastigotes incubated in PBS with 14C-U-palmitate 100 μM. The 14CO2 was captured at 0, 30, 60 and 120 min. C) Viability of epimastigote forms after incubation with different carbon sources and palmitate. The viability was assessed after 24 and 48 h by MTT assay. D) The intracellular ATP content was evaluated following incubation with different energy substrates or not (PBS, negative control). The ATP concentration was determined by luciferase assay and the data were adjusted by the number of cells. A statistical analysis was performed with one-way ANOVA followed by Tukey’s post-test at p < 0.05 using the GraphPad Prism 8.0.2 software program. We represent the level of statistical significance in this figure as follows: �� p value < 0.001; �� p value < 0.01; � p value < 0.05. For a p value > 0.05 we consider the differences to be not significant (ns). https://doi.org/10.1371/journal.ppat.1009495.g001 positive controls, since it is well knowing the ability of both metabolites to extend the parasites ´ viability in metabolic stress conditions, see [15]). As an additional negative control, we used PBS supplemented with BSA without added palmitate. The viability of these cells was assayed by measuring the total reductive activity by MTT assay. Additionally, we measured the total ATP levels. Cells incubated in the presence of palmitate showed higher viability than the nega- tive controls, but not as high as that of parasites incubated with glucose or histidine (Fig 1C). Consistently, parasites incubated in the presence of palmitate showed higher ATP contents than both negative controls. However, the intracellular ATP levels in the cells incubated with palmitate were diminished by half when compared to parasites incubated with histidine. Inter- estingly, the palmitate kept the ATP content at levels comparable to glucose (Fig 1D). Epimastigote forms excrete acetate as a primary end-product of palmitate oxidation Because the epimastigotes were able to oxidize 14C-U-palmitate to 14CO2, we were interested in analysing their exometabolome and comparing it with that of parasites exclusively consum- ing glucose, palmitate or without any carbon source. Thus, we subjected exponentially growing parasites to 16 h of starvation and then incubated them for 6 h in the presence of 0.3 mM pal- mitate, 10 mM 13C-U-glucose or without any carbon source. For the control, we analysed a PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 7 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi sample of non-starved parasites. After the incubations, the extracellular media were collected and analysed by 1H-NMR spectrometry. As expected, all the incubation conditions produced different flux profiles for excreted metabolites (Figs 2 and S1). Under our experimental condi- tions, the non-starved parasites primarily excreted succinate and acetate in similar quantities, Fig 2. Excreted end products of glucose and palmitate metabolism in epimastigote forms of T. cruzi. A) The extracellular medium of epimastigote forms incubated under different conditions was analysed by 1H-NMR spectrometry to detect and quantify the end-products. The resulting data were expressed in nmoles/h/108 cells. Means ± SD of three independent experiments. ICS is internal carbon sources; nd is non-detectable. B) and C) Schematic representation of the contribution of glucose and palmitate to the metabolism of epimastigote forms of T. cruzi. The glycosomal compartment and TCA cycle are indicated. The amount of end- product determined by the font size. Numbers indicates enzymatic steps. 1. Glycolysis; 2. pyruvate dehydrogenase; 3. citrate synthase; 4. aconitase; 5. isocitrate dehydrogenase; 6. α-ketoglutarate dehydrogenase; 7. succinyl-CoA synthetase; 8. Succinate dehydrogenase/complex II/fumarate reductase NADH-dependent; 9. fumarate hydratase; 10. malate dehydrogenase; 11. Malic enzyme; 12. alanine dehydrogenase/alanine aminotransferase; 13. lactate dehydrogenase; 14. acetate:succinyl-CoA transferase; 15. acetyl-CoA hydrolase; 16. succinyl-CoA synthetase; 17. Glycosomal fumarate reductase and 18. Palmitate oxidation by beta- oxidation, resulting in FADH2, NADH and acetyl-CoA; Abbreviations: Cit: Citrate, Aco: Aconitate, IsoC: Isocitrate, α-kg: α- Ketoglutarate, Suc-CoA: Succinyl-CoA, Suc: Succinate, Fum: Fumarate, Mal: Malate, and Oxa: Oxaloacetate. https://doi.org/10.1371/journal.ppat.1009495.g002 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 8 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi and alanine and lactate to a lesser extent. Parasites starved for 16 h in PBS and left to incubate in the absence of other metabolites had diminished succinate production (~7-fold) but increased acetate production three-fold compared to the non-starved parasites. It is relevant to stress that the only possible origin for these metabolites are internal carbon sources (ICS). Notably, no other excreted metabolites were detected under these conditions, indicating that under starvation, most of the ICS are transformed into acetate as an end product, which is compatible with the oxidation of internal fatty acids. These results raise the question about the metabolic fates of glucose or fatty acids in previously starved parasites. Starved epimastigotes that recovered in the presence of glucose exhibited a profuse excretion of succinate (450-fold the quantity excreted by the starved cells) and roughly equivalent quantities of acetate com- pared with the starved cells. Interestingly, lactate and alanine were also excreted at similar lev- els. As expected, the recovery with glucose produced an increase in all the secreted metabolites. However, analysing their distribution is a reconfiguration of the metabolism towards a majority production of succinate. Finally, in epimastigotes incubated with palmitate, we observed an increase in the acetate and alanine production of approximately 2.5 times to the levels in parasites that recovered in the presence of glucose. Interestingly, succinate is excreted in a smaller quantity than acetate and alanine, but still at 10-fold the rate observed in the starved non-recovered cells. Surprisingly, there was also a significant production of pyru- vate (not previously described in the literature, and not observed under any other conditions) and a small amount of lactate derived from palmitate. Glucose metabolism represses the fatty acid oxidation in epimastigotes Glucose is the primary carbon source for exponentially proliferating epimastigotes, and after its exhaustion from the culture medium, the parasites change their metabolism to use amino acids as carbon sources preferentially [10]. Therefore, we were interested in analysing if this preference for glucose is maintained in relation to the consumption of lipids. To determine if glucose metabolism interferes with the consumption of fatty acids, we created a 48 h prolifera- tion curve using parasites with an initial concentration adjusted to 2.5 x 107 mL-1 and quanti- fied them for 24 h each. Under these conditions, the parasites from the beginning of the experiment, at 0 h, are at mid-exponential phase, they are at late exponential phase at 24 h, and at 48 h they reached stationary phase at a concentration of 10 x 107 mL-1 (Fig 3A). At 0 h, 24 h and 48 h, the culture medium was collected to measure the remaining glucose and triacylgly- cerol (TAGs) concentrations (Fig 3B and 3C). Most of the glucose was consumed during the first 24 h (during proliferation), while the concentration of TAGs remained the same. After 48 h of proliferation (stationary phase), the TAG levels and lipid contents of the droplets were decreased by 1.5-fold and 2-fold, respectively, suggesting that glucose is preferentially con- sumed relative to fatty acids. These data show a decrease in the extracellular TAGs between 24 and 48 h, while the glucose was already almost entirely consumed, suggesting that glucose is negatively regulating the fatty acid catabolism. Epimastigote forms use endogenous fatty acids to support growth after glucose exhaustion From the previous results, we learned that under glucose deprivation, TAGs are taken up by the epimastigotes, and internally stored fatty acids are mobilized. However, to date, we did not provide any evidence pointing to their use as reduced carbon sources. To confirm this idea, exponentially proliferating epimastigotes were incubated in PBS supplemented with palmitate and 14C-U-glucose, or reciprocally, glucose and 14C-U-palmitate. In both cases, the production of 14C-labelled CO2 was quantified. The presence of 5 mM glucose diminished the release of PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 9 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi Fig 3. Changes in glucose and triacylglycerol contents in LIT medium. A) Growth curve of epimastigote forms. B) Glucose quantification over 48 h. C) Triacylglycerol levels over 48 h. In each experiment, we collected each medium at different times and subjected it to quantification according to the manufacturer’s instructions. All the experiments were performed in triplicates. Statistical analysis was performed with one-way ANOVA followed by Tukey’s post-test p < 0.05 using the GraphPad Prism 8.0.2 software program. We represent the levels of statistical significance in this figure as follows: �� p value < 0.001; �� p value < 0.01; and � p value < 0.05. For p value > 0.05, we consider the differences not significant (ns). https://doi.org/10.1371/journal.ppat.1009495.g003 14CO2 from 14C-U-palmitate by 90% while the presence of palmitate did not interfere with the production of 14CO2 from 14C-U-glucose (Fig 4). Taken together, our results show that glucose inhibits TAGs and fatty acid consumption, and after glucose exhaustion, a metabolic switch occurs towards the oxidation of internally stored fatty acids. To monitor the dynamics of use or accumulation of fatty acids in lipid droplets, we used as a probe a fluorescent fatty acid analogue called BODIPY 500/510 C1-C12. BODIPY shifts its fluorescence from red to green upon the uptake and catabolism of fatty acids, and from green to red when fatty acids are accumulated in the lipid droplets. Parasites collected at the mid and late exponential proliferation phases and the stationary phase were incubated with 1 μM BOD- IPY 500/510 C1-C12 for 16 h, before fluorescence determination by flow cytometry (Fig 5A, 5B and 5C). The fluorescence values increased with the harvesting time (and therefore, with the glucose depletion), indicating the increased uptake and use of fatty acids as substrates by a fatty acyl-CoA synthetase. These data were confirmed by fluorescence microscopy (Fig 5D). Interestingly, parasites in stationary phase showed an accumulation of activated fatty acids in PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 10 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi Fig 4. Glucose metabolism inhibits FAO. Parasites were incubated in the presence of 14C-U-palmitate + 5 mM glucose and 14C-U-glucose + 0.1 mM palmitate in PBS. 14CO2 production from epimastigotes incubated in PBS. The 14CO2 was captured after 120 min of incubation. The experiments were performed in triplicates. Statistical analysis was performed with one-way ANOVA followed by Tukey’s post-test p < 0.05 using the GraphPad Prism 8.0.2 software program. We represent the level of statistical significance in this figure as follows: �� p value < 0.001; �� p value < 0.01; and � p value < 0.05. For p value > 0.05, we consider the differences not significant (ns). https://doi.org/10.1371/journal.ppat.1009495.g004 spots along the cell. However, the number of lipid droplets increased upon parasite prolifera- tion (Fig 6A, 6B and 6C). This observation indicates that not only fatty acids metabolism is activated after glucose exhaustion, but also the parasite storage of fatty acids into lipid droplets. To find if the increase in fatty acid pools is accompanied by a change in the levels of enzymes related to fatty acid metabolism, we evaluated the specific activities of the enzymes hexokinase (HK), which is responsible for the initial step of glycolysis and an indicator of active glycolysis; acetyl-CoA carboxylase (ACC), which produces malonyl-CoA for fatty acid synthesis and carnitine palmitoyltransferase 1 (CPT1), the complex that plays a central role in fatty acid oxidation (FAO) by controlling the entrance of long-chain fatty acids into the mito- chondria [37]. For the control, we selected the enzyme serine palmitoyltransferase 1 (SPT1), a constitutively expressed protein in T. cruzi [38] (Fig 7). The hexokinase activity diminished up to 30% with the progression of the proliferation curve and the correlated depletion of glucose (Fig 7A). In addition, the ACC activity is no more detectable in the stationary phase cells (Fig 7B). By contrast, the CPT1 activity is increased by ~4-fold when the stationary phase is reached (Fig 7C), which confirms that fatty acid degradation occurs in the absence of glucose. It is noteworthy that the high levels of ACC activity in the presence of glucose supports the idea that under these conditions, fatty acids are probably synthesized instead of being catabolized. As expected, SPT1 did not change during the analysed time frame (Fig 7D). ETO, a CPT1 inhibitor, affects T. cruzi proliferation and mitochondrial activity To investigate the role of FAO in T. cruzi, we tested the effect of a well characterized inhibitor of CPT1, ETO, on the proliferation of epimastigotes. Among the ETO concentrations tested PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 11 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi Fig 5. Flow cytometry reveals distinct patterns in fatty acid pools during epimastigote growth. The epimastigotes were treated with 1 μM of BODIPY C1-C12 (500/510) and analysed by flow cytometry and fluorescence microscopy. A) 0 h. B) 24 h. C) 48 h. In the flow cytometry histograms, dashed peaks represent unstained parasites. Green-filled peaks represent stained parasites. D) Mean fluorescence per cell. The fluorescence for each cell was calculated using ImageJ software. All the experiments were performed in triplicates. Statistical analysis was performed with one-way ANOVA followed by Tukey’s post-test p < 0.05 using the GraphPad Prism 8.0.2 software program. We represent the level of statistical significance in this figure as follows: �� p value < 0.001; �� p value < 0.01; and � p value < 0.05. For p value > 0.05, we consider the differences not significant (ns). https://doi.org/10.1371/journal.ppat.1009495.g005 here (from 0.1 to 500 μM), only the higher concentration arrested parasite proliferation (Fig 8A). Importantly, the ETO effect was manifested when the parasites reached the late exponen- tial phase (a cell density of approximately 5x107 mL-1). This result is consistent with our previ- ous findings showing that FAO (and thus CPT1 activity) acquires an important role at this point in the proliferation curve. To confirm that CPT1 is in fact a target of ETO in T. cruzi, we assayed the drug’s effect on the enzyme activity in free cell extracts. Our results showed that 500 μM ETO diminished the CPT1 activity by almost 80% (Fig 8B). To confirm the interfer- ence of ETO with the beta-oxidation of fatty acids, parasites incubated in PBS containing 14C-U-palmitate were treated with 500 μM ETO to compare their production of 14CO2 with that of the untreated controls. Palmitate-derived CO2 production diminished by 80% in ETO- treated cells compared to untreated parasites (Fig 8C). In addition, ETO treatment did not PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 12 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi Fig 6. Epimastigote forms accumulates fatty acids into lipid droplets during growth. The epimastigotes were treated with 1 μM BODIPY C1-C12 (500/510) and analysed by flow cytometry and fluorescence microscopy. A) 0 h. B) 24 h. C) 48 h. In the flow cytometry histograms, dashed peaks represent unstained parasites. Yellow filled peaks represent positively stained parasites. The number of green/yellow spots for each cell was calculated using ImageJ software. All the experiments were performed in triplicates. https://doi.org/10.1371/journal.ppat.1009495.g006 affect the metabolism of 14C-U-glucose or 14C-U-histidine, ruling out a possible unspecific reaction of this drug with CoA-SH as described by [38,39]. Other compounds described as FAO inhibitors were also tested, but none of them inhibited epimastigote proliferation or 14CO2 production from 14C-U-palmitate (S3 Fig). In addition, the BODIPY cytometric analy- sis of cells treated with 500 μM ETO showed a strong decrease in the CoA acylation levels (acti- vation of fatty acids) with respect to the untreated controls (Fig 8D), as confirmed by fluorescence microscopy (Fig 8D). To reinforce the validation of ETO for further experiments, a set of controls are offered in S3 Fig. Our preliminary conclusion is that ETO inhibited beta- oxidation by inhibiting CPT1, confirming that the breakdown of fatty acids is important to proliferation progression in the absence of glucose. ETO treatment affects cell cycle progression The metabolic interference of ETO diminished epimastigote proliferation; however, this find- ing could be due to a decrease in the parasite proliferation rate or an increase in the death rate. Therefore, we checked if this compound could induce cell death through programmed cell death (PCD) or necrosis. PCD is characterized by biochemical and morphological events such as exposure to phosphatidylserine, DNA fragmentation, decreases (or increases) in the ATP levels, and increases in reactive oxygen species (ROS), among others [40]. The parasites were treated with 500 μM of ETO for 5 days, followed by incubation with propidium iodide (PI) for PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 13 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi Fig 7. Activities of enzymes related to lipid and glucose metabolism during T. cruzi growth curves. A) (HK) Hexokinase B) (ACC) acetyl-CoA carboxylase, C) (CPT1) carnitine-palmitoyltransferase, and D) (SPT) serine palmitoyltransferase. All these activities were measured in crude extracts from epimastigote forms at different moments of the growth curve. All the experiments were performed in triplicates. Time course activities and controls shown in S2 Fig. Statistical analysis was performed with one-way ANOVA followed by Tukey’s post-test at p < 0.05, using the GraphPad Prism 8.0.2 software program. We represent the level of statistical significance in this figure as follows: �� p value < 0.001; �� p value < 0.01; and � p value < 0.05. For p value > 0.05 we consider the differences not significant (ns). https://doi.org/10.1371/journal.ppat.1009495.g007 cell membrane integrity analysis and annexin-V FITC to evaluate the phosphatidylserine expo- sure. Parasites treated with ETO showed negative results for necrosis or programmed cell death markers (Fig 9A), indicating that the cell proliferation was arrested but cell viability was maintained. Because the multiplication rates seemed to be diminished, we performed a cell cycle analysis. Noticeably, the treated parasites were enriched in G1 (85.9%) with respect to non-treated cells (43.6%), suggesting that ETO prevented the entry of epimastigotes into the S phase of the cell cycle (Fig 9B). Last, we noticed that after washing out the ETO, the parasites recovered their proliferation at rates comparable to our untreated controls (Fig 9C). Inhibition of FAO by ETO affects energy metabolism, impairing the consumption of endogenous fatty acids The evidence obtained to date suggests that parasites resist metabolic stress by mobilizing and consuming stored fatty acids. Therefore, it is reasonable to hypothesize that ETO, which blocks the mobilization of fatty acids into the mitochondria for oxidation, probably perturbs the O2 consumption and ATP levels in late-exponential or stationary phase cells. Parasites growing for 5 days under 500 μM ETO treatment or no treatment were collected to evaluate their ability PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 14 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi Fig 8. ETO inhibits CPT1 and interferes with cell proliferation in epimastigote forms. (A) Proliferation of epimastigote forms in the presence of 0.1 to 500 μM ETO. For the positive control of dead cells, a combination of antimycin (0.5 μM) and rotenone (60 μM) was used. (B) Inhibition of CPT1 activity in crude extracts using 250 and 500 μM of ETO. C) 14CO2 capture from 14C-U-palmitate oxidation. D) Flow cytometry analysis and fluorescence microscopy of epimastigote forms treated (or not) with ETO. In the histograms, dashed peaks represent unstained parasites and green-filled peaks represent parasites stained with BODIPY C1-C12. All the experiments were performed in triplicates. Statistical analysis was performed with one-way ANOVA followed by Tukey’s post-test at p < 0.05 using the GraphPad Prism 8.0.2 software program. We represent the level of statistical significance in this figure as follows: �� p value < 0.001; �� p value < 0.01; and � p value < 0.05. For p values > 0.05, we consider the differences not significant (ns). https://doi.org/10.1371/journal.ppat.1009495.g008 to trigger O2 consumption (Fig 10A and 10B). The rates of O2 consumption corresponding to basal respiration were measured in cells resuspended in MCR respiration buffer. We then measured the leak respiration by inhibiting the ATP synthase with oligomycin A. Finally, to measure the maximum capacity of the electron transport system (ETS), we used the uncoupler FCCP [27]. Our results demonstrate that compared to no treatment, ETO treatment dimin- ishes the rate of basal O2 consumption, the leak respiration and the ETS capacity (Fig 10C). In general, respiratory rates diminished in parasites treated with ETO when compared to the untreated ones. As expected, ETO treatment led to a 75% decrease in the levels of total intracel- lular ATP compared to untreated parasites (Fig 10D). To complement this result, because all these experiments were conducted in the complete absence of an oxidizable external metabo- lite, our results show that the parasite is able to oxidize internal metabolites (Fig 10A and 10B). Taking into account that treating parasites with ETO diminished the basal respiration rates of these parasites by approximately one-half (Fig 10A and 10B), it is reasonable to conclude that a relevant part of the respiration in the absence of external oxidisable metabolites is based on the PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 15 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi Fig 9. Analysis of extracellular phosphatidylserine exposure, membrane integrity and cell cycle after ETO treatment. Parasites in the exponential growth phase were treated with 500 μM of ETO for 5 days. (A) Following the incubation period, the parasites were labelled with propidium iodide (PI) and annexin V-FITC (ANX) and analysed by flow cytometry. (B) The cell cycle was assessed using PI staining. (C) Growth curves of epimastigote forms before and after removing the treatment. All the experiments were performed in triplicates. Statistical analysis was performed with one-way ANOVA followed by Tukey’s post-test p < 0.05, using the GraphPad Prism 8.0.2 software program. We represent the level of statistical significance in this figure as follows: �� p value < 0.001; �� p value < 0.01; and � p value < 0.05. For p values > 0.05, we consider the differences not significant (ns). https://doi.org/10.1371/journal.ppat.1009495.g009 consumption of internal lipids. This is consistent with the confirmation that epimastigotes maintain their viability in the presence of non-fatty acid carbon sources in the presence of ETO (S4 Fig). In summary, these results confirm that ETO is interfering with ATP synthesis through oxidative phosphorylation in epimastigote forms. Endogenous fatty acids contribute to long-term starvation resistance in epimastigote forms As previously demonstrated, ETO interferes with the consumption of endogenous fatty acids, and this impairment causes ATP depletion and cell cycle arrest. One intriguing characteristic of the insect stages of T. cruzi is their resistance to starvation. To observe the importance of internal fatty acids in this process, we incubated epimastigotes in PBS in the presence (or absence) of 500 μM ETO. The mitochondrial activity of these cells was followed for 24 h with Alamar Blue. Our results showed that the mitochondrial activity of the parasites in the presence of ETO was reduced by 31% after 48 h of starvation, and 65% after 72 h of starvation (Fig 11) compared to the controls (untreated parasites). These data confirmed our hypothesis that the breakdown of accu- mulated fatty acids partially contributes to the resistance of the parasite under severe starvation. Inhibition of CPT1 impairs metacyclogenesis Considering that the FAO increases in the epimastigotes during the stationary phase, and that differentiation into infective metacyclic trypomastigotes (metacyclogenesis) is triggered in the PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 16 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi Fig 10. Effects of ETO on respiration and ATP production in epimastigote forms of T. cruzi. (A) Oxygen consumption of epimastigote forms after normal growth in LIT medium. (B) Oxygen consumption after ETO 500 μM treatment. Parasite growth in LIT medium with the compound until the 5th day. In black, a time-course register of the concentration (pmols) of O2 in the respiration chamber. In blue, negative of the concentration derivative (pmols) of O2 with respect to time (velocity of O2 consumption in pmoles per second). The parasites were washed twice in PBS and kept in MRC buffer at 28˚C during the assays. (C) The basal respiration (initial oxygen flux values, MRC), respiration leak after the sequential addition of 0.5 μg/mL of oligomycin A (2 μg/mL), and electron transfer system (ETS) capacity after the sequential addition of 0.5 μM FCCP (2 μM) were measured for each condition. (D) Intracellular levels of ATP after treating with 500 μM ETO. The intracellular ATP content was assessed following incubation with different energy substrates or not (PBS, negative control). The ATP concentration was determined by luciferase assay and the data were adjusted by the number of cells. All the experiments were performed in triplicates. Statistical analysis was performed with one-way ANOVA followed by Tukey’s post-test at p < 0.05 using GraphPad Prism 8.0.2 software. We represent the level of statistical significance as follows: �� p value < 0.001; �� p value < 0.01; and � p value < 0.05. For p values > 0.05, we consider the differences not significant (ns). https://doi.org/10.1371/journal.ppat.1009495.g010 stationary phase of epimastigote parasites, one might expect a possible relationship between the consumption of fatty acids and metacyclogenesis. To approach this possibility, we initially compared the CPT1 activity of stationary epimastigote forms before and after a 24 h incuba- tion in the differentiation medium TAU-3AAG. As observed, there is an increase in CPT1 activity after submitting the parasites to the metacyclogenesis in vitro (Fig 12A). Parasites were then submitted to differentiation with TAU-3AAG medium in the presence of the probe BOD- IPY. The probe was incorporated into lipid droplets, confirming that fatty acids metabolism was active during the beginning of metacyclogenesis (Fig 12B). To address the importance of FAO during differentiation, metacyclogenesis was induced in vitro on ETO-treated or untreated (control) parasites. ETO treatment interfered with differentiation, diminishing the number of metacyclic forms present in the culture (Fig 12C). In addition, this inhibition was dose-dependent, with an IC50 = ± 32.96 μM (Fig 12D). Importantly, we ruled out that the vari- ation found in the differentiation rates was due to a selective death of treated epimastigotes, since their survival during this experiment in the presence or absence of ETO (from 5 to 500 μM) was not significantly different (S5 Fig). Based on these data, we could conclude that fatty acid oxidation, at the level of the CPT1, was also participating in the regulation of metacyclogenesis. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 17 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi Fig 11. Internal fatty acid consumption contributes to parasite viability under severe nutritional starvation. Viability of epimastigote forms after incubation in PBS with or without ETO. The viability was assessed every 24 h using Alamar Blue. Statistical analysis was performed with one-way ANOVA followed by Tukey’s post-test p < 0.05 using GraphPad Prism 8.0.2 software. We represent the levels of statistical significance as follow: �� p value < 0.001, and for p values > 0.05, we consider the differences not significant (ns). https://doi.org/10.1371/journal.ppat.1009495.g011 Discussion During the journey of T. cruzi inside the insect vector, the glucose levels decrease rapidly after each blood meal [41], leaving the parasite exposed to an environment rich in amino and fatty Fig 12. ETO inhibits metacyclogenesis. A) CPT1 activity of epimastigote forms in stationary phase and 24h after incubated in TAU-3AAG medium (for triggering metacyclogenesis). B) Fluorescence microscopy of cells incubated in TAU-3AAG in the presence of BODIPY 500–510 C1-C12. C) Effects of different ETO concentrations on metacyclogenesis. The differentiation was evaluated by counting the cells in a Neubauer chamber each day for 6 days. This experiment was performed in triplicate. D) Percentage of differentiation at the 5th day of differentiation. Inset: IC50 of metacyclogenesis inhibition by ETO. The enzymatic activities were measured in duplicate. All the other experiments were performed in triplicates. https://doi.org/10.1371/journal.ppat.1009495.g012 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 18 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi acids in the digestive tube of Rhodnius prolixus [42,43]. Because the digestive tract of triato- mine insects possesses a perimicrovillar membrane, which is composed primarily of lipids and is enriched by glycoproteins [44], it has been speculated that its degradation could provide lip- ids for parasite metabolism [45]. In this study, we showed that the insect stages of T. cruzi coordinate the activation of fatty acid consumption with the metabolism of glucose. Our experiments corroborate early studies about the relatively slow use of palmitate as an energy source by proliferating epimastigotes [46,47]. In addition, our results shed light on the end product excretion by epimastigote forms during incubation under starvation conditions, and during their recovery from starvation using glucose or palmitate. First, we showed that non- starved and starved parasites recovered in the presence of glucose, excreting succinate as their primary metabolic waste, as expected [48,49]. After 16 h of nutritional starvation, the con- sumption of internal carbon sources produces acetate as the primary end-product. In the pres- ence of glucose after 16 h of starvation, we found that glucose-derived carbons contribute to the excreted pools of acetate and lactate. Interestingly, palmitate metabolism contributed to the increase in acetate production, followed by the production of alanine, pyruvate, succinate and lactate. The unexpected production of alanine, pyruvate and lactate can be explained by an increase in the TCA cycle activity, producing malate, which can be converted into pyruvate by the decarboxylative reaction of the malic enzyme (ME) [50]. Pyruvate can be converted into alanine through a transamination reaction by an alanine- [51], a tyrosine- [52] an aspartate aminotransferase [53], or a reductive amination by an alanine dehydrogenase [54]. The excre- tion of lactate could be a consequence of lactate dehydrogenase activity. However, it should be noted that this enzymatic activity has not been observed to date. In relation to the succinate production, a relevant factor favouring this process is the production of NADH by the third step of the beta-oxidation (3-hydroxyacyl-CoA dehydrogenase). This NADH can be oxidized through the activity of NADH-dependent mitochondrial fumarate reductase [55], which con- comitantly converts NADH into NAD+ and fumarate into succinate. This succinate can be excreted or re-used by the TCA cycle, and the resulting NAD+ can be used as a cofactor for other enzymes. As previously mentioned, it is well known that during the initial phase of proliferation, epi- mastigotes preferentially consume glucose, and during the stationary phase, a metabolic switch occurs towards the consumption of amino acids [8,10,56]. Our results show that this switch constitutes a broader and more systemic metabolic reprogramming, which also includes FAO. We detected this switch through changes in the enzymatic activities of key enzymes responsi- ble for the regulation of FAO, such as CPT1 and ACC, which have increased and decreased activities, respectively, in the presence of glucose. Our findings showed that the inhibition of CPT1 affects the late phase of proliferation of epimastigotes when the switch to FAO has already occurred. An interesting question about T. cruzi epimastigotes is how they survive long periods of starvation. Early data showed high respiration levels in epimastigotes incubated in the absence of external oxidisable carbon sources. This oxygen consumption was attributed to the break- down of TAGs into free fatty acids and their further oxidation [57]. Here, we confirmed this finding by inhibiting the internal fatty acid consumption, which in turn diminished the oxida- tive phosphorylation activity, internal ATP levels and the total reductive activity of parasites under severe nutritional stress. Even more notably, we showed that under these conditions, the lipids stored in lipid droplets [58,59] are consumed. Unlike what has been observed in pro- cyclic forms of T. brucei, in which the function of lipid droplets is not clear [60], our results show that in T. cruzi, they are committed to epimastigote survival under extreme metabolic stress. Of course, the contribution of other metabolic sources and processes such as autophagy in coping with nutritional stress cannot be ruled out [61]. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 19 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi Multiple metabolic factors have been involved in metacyclogenesis, such as the proline, aspartate, glutamate [62], glutamine [17] and lipids present in the triatomine digestive tract [63]. Interestingly, the occurrence of metacyclic trypomastigotes in culture leads to an increase in CO2 production from labelled palmitate [47]. The ETO treatment inhibited metacyclogen- esis in vitro, showing that the consumption of internal fatty acids is important for cell differen- tiation. Consequently, we propose that lipids are not only external signals of metacyclogenesis, as previously suggested [63], but they also have a central role in the bioenergetics of metacyclo- genesis. It was previously described that this differentiation process is triggered by a starvation followed by the addition of energy and carbon sources such as a combination of glucose, pro- line, aspartate and glutamate [64], just proline [65], or just glutamine [17] must be supplied. In our experiment, we added the carbon and energy sources alone, or in combination with ETO, which allowed us to dissect the effect of lipids mobilization into the mitochondria after the starvation. Our results indicate that, in addition to the oxidation of several metabolites (glucose and amino acids), the mitochondrial metabolism of lipids, and probably the reduced cofactors resulting from these processes are contributing to the mitochondrial ATP production neces- sary to support this differentiation step. Summarizing, this observation points to the participa- tion of lipids metabolism in T. cruzi metacyclogenesis. In conclusion, fatty acids are important carbon sources for T. cruzi epimastigotes in the absence of glucose. The completeness and function of the TCA cycle in trypanosomatids have been a matter of controverse [66–68]. However, several works have shown in T. cruzi the func- tionality of TCA cycle and more recently, most of the TCA cycle intermediates has been detected by metabolomic analysis [10]. In addition, Villafraz et al demonstrated that even in procyclics of T. brucei the TCA cycle is complete and works in the oxidative direction [69]. Therefore, we propose that palmitate can be taken up by the cells and fuel the TCA cycle by producing acetyl-CoA, the oxidation of which generates CO2. However, in the absence of external carbon sources, lipid droplets become the primary sources of fatty acids, helping the organism to survive nutritional stress. Importantly, FAO supports endogenous respiration rates and ATP production and powers metacyclogenesis. Supporting information S1 Fig. 1H-NMR analysis of excreted end products from glucose and threonine metabo- lism. The metabolic end products (succinate, acetate, alanine and lactate) excreted by the epi- mastigote cells that were incubated after 6 h in PBS (A), PBS after 16 h of starvation without (B) or with D-[U-13C]-glucose (C) or palmitate (D) were determined by 1H-NMR. Each spec- trum corresponds to one representative experiment from a set of at least 3. A part of each spec- trum ranging from 0.5 ppm to 4 ppm is shown. The resonances were assigned as indicated: A12, acetate; A13, 13C-enriched acetate; Al12, alanine; Al13, 13C-enriched alanine; G13, 13C- enriched glucose; L12, lactate; L13, 13C-enriched lactate; P12, palmitate; S12, succinate; and S13, 13C-enriched succinate. (TIF) S2 Fig. Time course activities of enzymes measured in this work. A) (ACC) acetyl-CoA car- boxylase, B) (CPT1) carnitine-palmitoyltransferase, and C) (SPT) serine palmitoyltransferase. All the activities were measured in cell-free extracts of epimastigote forms at different moments of the growth curve as indicated in the main text. All the measurements were per- formed in triplicates. (TIF) PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 20 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi S3 Fig. Other FAO inhibitors did not affect cell proliferation and FAO in the epimastigote forms. Well-known FAO inhibitors such as valproic acid [57], trimetazidine [58,59] and β- hydroxybutyrate [60], were assayed for their effect on the proliferation of T. cruzi epimasti- gotes. None of the compounds at the tested concentrations inhibit the epimastigotes prolifera- tion. We used the compounds at the higher concentration tested for epimastigotes proliferation to assay their ability to inhibit FAO by 14CO2 trapping using U-14C-palmitate as a substrate. As observed, none of these compound inhibited the 14CO2 production from palmi- tate, confirming that they are not inhibiting FAO in T. cruzi. The compounds were assayed at concentrations between 0.1 and 1000 μM. For positive controls of dead cells, a combination of antimycin (0.5 μM) and rotenone (60 μM) were used. The maximum concentration tested for these compounds does not diminish CO2 liberation from FAO. A) Valproic Acid (AV). B) Tri- metazidine (TMZ). C) β-hydroxybutyrate (βHOB). (TIF) S4 Fig. ETO affected specifically FAO and did not affect the viability of epimastigote forms in the presence of other carbon sources. To confirm that ETO is acting on FAO and to rule out that our findings are due to off-target effects that can happenas at concentrations of up to 200 μM [61,62] we assayed its effect in the presence of non-lipidic carbon sources. The parasites were incubated for 24 h in PBS (negative control), 0.1 mM palmitate supplemented with BSA, 5.0 mM histidine, 5 mM glucose, 0.1 mM carnitine and BSA without adding palmitate in the pres- ence (or not) of 500 μM ETO. As expected, ETO treatment did not affect the viability of cells incu- bated in glucose or histidine but did affect the viability of the cells incubated with palmitate or carnitine. Surprisingly, we also observed an ETO effect on parasites under metabolic stress, such as those incubated with PBS or BSA. This finding could be explained by the fact that under meta- bolic stress, the parasite mobilizes and consumes its internal lipids. The viability of these cells was inferred from by measuring the total reductive activity using MTT assays after 24 h. (TIF) S5 Fig. Viability of epimastigote forms subjected to metacyclogenesis under different ETO concentrations. We assassed the viability of parasites during metacyclogenesis in the presence or not of ETO. Stationary epimastigotes in TAU-3AAG media were treated with different con- centrations of ETO (5 to 500 μM for 24 hs, panel A, or 5 to 50 μM for 120 h, panel B). The via- bility of these cells was inferred by measuring the total reductive activity using an Alamar blue assay [63]. In addition, the parasites were maintained during metacyclogenesis in the presence of 50 μM ETO (or not, control) and differentiation was followed up by daily counts, based on the percentage of metacyclic trypomastigotes collected in culture supernatant (panel C). Con- sidering that TAU-3AAG contains glucose in its composition, we performed an in vitro meta- cyclogenesis using only proline as a metabolic inducer [64]. As observed, even in the absence of glucose, ETO treatment affects metacyclogenesis. (TIF) Acknowledgments We thank the Core Facility for Scientific Research at the University of Sao Paulo (CEFAP- USP/FLUIR) for the flow cytometry analysis and Dr. Mauro Javier Veliz Cortez (Department of Parasitology, ICB-USP) for the microscopy support. Author Contributions Conceptualization: Rodolpho Ornitz Oliveira Souza, Rui Curi, Fre´de´ric Bringaud, Ariel Mari- ano Silber. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 21 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi Data curation: Rodolpho Ornitz Oliveira Souza, Fla´via Silva Damasceno, Sabrina Marsiccobe- tre, Marc Biran, Fre´de´ric Bringaud, Ariel Mariano Silber. Formal analysis: Rodolpho Ornitz Oliveira Souza, Fla´via Silva Damasceno, Marc Biran, Gil- son Murata, Fre´de´ric Bringaud, Ariel Mariano Silber. Funding acquisition: Fre´de´ric Bringaud, Ariel Mariano Silber. Investigation: Rodolpho Ornitz Oliveira Souza, Fla´via Silva Damasceno, Sabrina Marsiccobe- tre, Gilson Murata, Ariel Mariano Silber. Methodology: Rodolpho Ornitz Oliveira Souza, Fla´via Silva Damasceno, Sabrina Marsiccobe- tre, Marc Biran, Fre´de´ric Bringaud. Project administration: Ariel Mariano Silber. Resources: Fre´de´ric Bringaud, Ariel Mariano Silber. Supervision: Rui Curi, Fre´de´ric Bringaud, Ariel Mariano Silber. Validation: Fla´via Silva Damasceno, Rui Curi, Ariel Mariano Silber. Visualization: Fre´de´ric Bringaud. Writing – original draft: Rodolpho Ornitz Oliveira Souza, Ariel Mariano Silber. Writing – review & editing: Fre´de´ric Bringaud, Ariel Mariano Silber. References 1. WHO (last visit on March, 2021) Chagas disease (American trypanosomiasis). WHO https://www.who. int/health-topics/chagas-disease. 2. Perez-Molina JA, Molina I (2018) Chagas disease. Lancet. 391: 82–94. https://doi.org/10.1016/S0140- 6736(17)31612-4 PMID: 28673423 3. Melo RdFP, Guarneri AA, Silber AM. The influence of environmental cues on the development of Trypa- nosoma cruzi in triatominae vector. 2020; Front Cell Infect Microbiol 10: 27–27. https://doi.org/10.3389/ fcimb.2020.00027 PMID: 32154185 4. Lisvane Silva P, Mantilla BS, Barison MJ, Wrenger C, Silber AM. The uniqueness of the Trypanosoma cruzi mitochondrion: opportunities to identify new drug target for the treatment of Chagas disease. Curr Pharm Des. 2011; 17: 2074–2099. https://doi.org/10.2174/138161211796904786 PMID: 21718252 5. Bern C. Chagas’ Disease. N Engl J Med. 2015; 373: 1882. 6. De Souza W. Basic cell biology of Trypanosoma cruzi. Curr Pharm Des. 2002; 8: 269–285. https://doi. org/10.2174/1381612023396276 PMID: 11860366 7. Li Y, Shah-Simpson S, Okrah K, Belew AT, Choi J, Caradonna KLet al. Transcriptome remodeling in Trypanosoma cruzi and human cells during intracellular infection. PLOS Pathog. 2016; 12: e1005511– e1005511. https://doi.org/10.1371/journal.ppat.1005511 PMID: 27046031 8. Marchese L, Nascimento J, Damasceno F, Bringaud F, Michels P, Silber AM. The uptake and metabo- lism of amino acids, and their unique role in the biology of pathogenic trypanosomatids. Pathogens. 2018; 7: 36–36. 9. Cazzulo JJ. Energy metabolism in Trypanosoma cruzi. Subcell Biochem. 1992; 18: 235–257. https:// doi.org/10.1007/978-1-4899-1651-8_7 PMID: 1485353 10. Barison MJ, Rapado LN, Merino EF, Furusho Pral EM, Mantilla BS, Marchese Let al. Metabolomic pro- filing reveals a finely tuned, starvation-induced metabolic switch in Trypanosoma cruzi epimastigotes. J Biol Chem. 2017; 292: 8964–8977. https://doi.org/10.1074/jbc.M117.778522 PMID: 28356355 11. Zeledon R. Comparative physiological studies on four species of hemoflagellates in culture. II. Effect of carbohydrates and related substances and some amino compounds on the respiration. J. Parasitol. 1960; 46: 541–541. PMID: 13788134 12. Sylvester D, Krassner SM. Proline metabolism in Trypanosoma cruzi epimastigotes. Comp Biochem Physiol B: Comp. Biochem. 1976; 55: 443–447. https://doi.org/10.1016/0305-0491(76)90318-7 PMID: 789007 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 22 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi 13. Paes LS, Suarez Mantilla B, Zimbres FM, Pral EM, Diogo de Melo P, Tahara EBet al. Proline dehydro- genase regulates redox state and respiratory metabolism in Trypanosoma cruzi. PLoS ONE. 2013; 8: e69419–e69419. https://doi.org/10.1371/journal.pone.0069419 PMID: 23894476 14. Mantilla BS, Paes LS, Pral EMF, Martil DE, Thiemann OH, Ferna´ndez-Silva P et al. Role of Δ1-pyrro- line-5-carboxylate dehydrogenase supports mitochondrial metabolism and host-cell invasion of Trypa- nosoma cruzi. J Biol Chem. 2015; 290: 7767–7790. https://doi.org/10.1074/jbc.M114.574525 PMID: 25623067 15. Bariso´ n MJ, Damasceno FS, Mantilla BS, Silber AM. The active transport of histidine and its role in ATP production in Trypanosoma cruzi. J Bioenerg Biomembr. 2016; 48: 437–449. https://doi.org/10.1007/ s10863-016-9665-9 PMID: 27222029 16. Girard RMBM, Crispim M, Alencar MB, Silber AM. Uptake of L-alanine and its distinct roles in the bioen- ergetics of Trypanosoma cruzi. 2018; mSphere 3: e00338–18. https://doi.org/10.1128/mSphereDirect. 00338-18 PMID: 30021880 17. Damasceno FS, Bariso´ n MJ, Crispim M, Souza ROO, Marchese L, Silber AM. L-Glutamine uptake is developmentally regulated and is involved in metacyclogenesis in Trypanosoma cruzi. Mol Biochem Parasitol. 2018; 224:17–25. https://doi.org/10.1016/j.molbiopara.2018.07.007 PMID: 30030130 18. Gualdro´n-Lo´pez M, Brennand A, Hannaert V, Quiñones W, Ca´ ceres AJ, Bringaud F, et al. When, how and why glycolysis became compartmentalised in the Kinetoplastea. A new look at an ancient organelle. Int J Parasitol. 2012; 42: 1–20. https://doi.org/10.1016/j.ijpara.2011.10.007 PMID: 22142562 19. Gabaldo´ n T, Ginger ML, Michels PA. Peroxisomes in parasitic protists. Mol Biochem Parasitol. 2016; 209: 35–45. https://doi.org/10.1016/j.molbiopara.2016.02.005 PMID: 26896770 20. Gualdron-Lo´pez M, Vapola MH, Miinalainen IJ, Hiltunen JK, Michels PA, Antonenkov VD. Channel- forming activities in the glycosomal fraction from the bloodstream form of Trypanosoma brucei. PLoS One. 2012; 7: e34530. https://doi.org/10.1371/journal.pone.0034530 PMID: 22506025 21. Bakker BM, Mensonides FI, Teusink B, van Hoek P, Michels PA, Westerhoff HV. Compartmentation protects trypanosomes from the dangerous design of glycolysis. Proc Natl Acad Sci U S A. 2000; 97: 2087–2092. https://doi.org/10.1073/pnas.030539197 PMID: 10681445 22. Quiñones W, Acosta H, Gonc¸alves CS, Motta MCM, Gualdro´ n-Lo´ pez M, Michels PA. Structure, proper- ties, and function of glycosomes in Trypanosoma cruzi. Front Cell Infect Microbiol. 2020; 10: 25. https:// doi.org/10.3389/fcimb.2020.00025 PMID: 32083023 23. Acosta H, Dubourdieu M, Quiñones W, Ca´ ceres A, Bringaud F, Concepcio´n JL. Pyruvate phosphate dikinase and pyrophosphate metabolism in the glycosome of Trypanosoma cruzi epimastigotes. Comp Biochem Physiology B: Biochem Mol Biol. 2004; 38: 347–356. https://doi.org/10.1016/j.cbpc.2004.04. 017 PMID: 15325334 24. Acosta H, Burchmore R, Naula C, Gualdro´n-Lo´pez M, Quintero-Troconis E, Ca´ ceres AJ, et al. Proteo- mic analysis of glycosomes from Trypanosoma cruzi epimastigotes. Mol Biochem Parasitol. 2019; 229: 62–74. https://doi.org/10.1016/j.molbiopara.2019.02.008 PMID: 30831156 25. Camargo EP. Growth and differentiation in Trypanosoma cruzi. I. Origin of metacyclic trypanosomes in liquid media. Rev Inst Med Trop Sao Paulo. 1964; 6: 93–100. PMID: 14177814 26. Huynh FK, Green MF, Koves TR, Hirschey MD. Measurement of fatty acid oxidation rates in animal tis- sues and cell lines. Meth Enzymol. 2014; 542: 391–405. https://doi.org/10.1016/B978-0-12-416618-9. 00020-0 PMID: 24862277 27. Alencar MB, Girard RBMM, Silber AM. Measurement of energy states of the trypanosomatid mitochon- drion. Meth Mol Biology (Clifton, NJ). 2020; 2116: 655–671. https://doi.org/10.1007/978-1-0716-0294- 2_39 PMID: 32221948 28. Marchese L, Olavarria K, Mantilla BS, Avila CC, Souza ROO, Damasceno FS, et al. Trypanosoma cruzi synthesizes proline via a Δ1-pyrroline-5-carboxylate reductase whose activity is fine-tuned by NADPH cytosolic pools. Biochem J. 2020; 477: 1827–1845. https://doi.org/10.1042/BCJ20200232 PMID: 32315030 29. Bradford MM. A rapid and sensitive method for the quantitation of microgram quantities of protein utiliz- ing the principle of protein-dye binding. Anal Biochem. 1976; 72: 248–254. https://doi.org/10.1006/abio. 1976.9999 PMID: 942051 30. Bieber PLL, Abraham T, Helmrath T, Kim Y, Dehlin R. A rapid spectrophotometric assay for carnitine palmitoyltransferase. Anal Biochem. 1972; 50: 509–518 https://doi.org/10.1016/0003-2697(72)90061- 9 PMID: 4630394 31. Willis LB, Saridah W, Omar W, Ravigadevi S, Sinskey AJ. Non-radioactive assay for Acetyl-CoA car- boxylase activity. J. Oil Palm Res. Special Issue on Malaysia-MIT Biotechnology Partnership Pro- gramme V. 2008; 2: 30–36 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 23 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi 32. Racagni GE, Machado de Domenech EE. Characterization of Trypanosoma cruzi hexokinase. Mol Bio- chem Parasitol. 1983; 9: 181–188. https://doi.org/10.1016/0166-6851(83)90108-1 PMID: 6366548 33. Ru¨tti MF, Richard S, Penno A, von Eckardstein A, Hornemann T. An improved method to determine ser- ine palmitoyltransferase activity. J Lip Res. 2009; 50: 1237–1244. https://doi.org/10.1194/jlr.D900001- JLR200 PMID: 19181628 34. Magdaleno A, Ahn IY, Paes LS, Silber AM. Actions of a proline analogue, L-thiazolidine-4-carboxylic acid (T4C), on Trypanosoma cruzi. PLoS One. 2009; 4: e4534–e4534. https://doi.org/10.1371/journal. pone.0004534 PMID: 19229347 35. Damasceno FS, Barison MJ, Pral EM, Paes LS, Silber AM. Memantine, an antagonist of the NMDA glu- tamate receptor, affects cell proliferation, differentiation and the intracellular cycle and induces apopto- sis in Trypanosoma cruzi. PLoS Negl Trop Dis. 2014; 8: e2717–e2717. https://doi.org/10.1371/journal. pntd.0002717 PMID: 24587468 36. 37. Figarella K, Rawer M, Uzcategui NL, Kubata BK, Lauber K, Madeo F, et al. Prostaglandin D2 induces programmed cell death in Trypanosoma brucei bloodstream form. Cell Death Differ. 2005; 12: 335– 346. https://doi.org/10.1038/sj.cdd.4401564 PMID: 15678148 Lopaschuk GD, Wall SR, Olley PM, Davies NJ. Etomoxir, a carnitine palmitoyltransferase I inhibitor, protects hearts from fatty acid-induced ischemic injury independent of changes in long chain acylcarni- tine. Circ Res. 1988; 63: 1036–1043. https://doi.org/10.1161/01.res.63.6.1036 PMID: 3197271 38. Koeller CM, Heise N. The sphingolipid biosynthetic pathway is a potential target for chemotherapy against Chagas disease. Enzyme Res. 2011; 2011: 1–13. https://doi.org/10.4061/2011/648159 PMID: 21603271 39. Divakaruni AS, Hsieh WY, Minarrieta L, Duong TN, Kim KKO, Desousa BR, et al. Etomoxir inhibits mac- rophage polarization by disrupting CoA homeostasis. Cell Metab. 2018; 28: 490–503.e497. https://doi. org/10.1016/j.cmet.2018.06.001 PMID: 30043752 40. Duszenko M, Figarella K, Macleod ET, Welburn SC. Death of a trypanosome: a selfish altruism. Trends Parasitol. 2006; 22: 536–542. https://doi.org/10.1016/j.pt.2006.08.010 PMID: 16942915 41. Mariano AC, Santos R, Gonzalez MS, Feder D, Machado EA, Piscarelli B, et al. Synthesis and mobiliza- tion of glycogen and trehalose in adult male Rhodnius prolixus. Arch Insect Biochem Physiol. 2009; 72: 1–15. https://doi.org/10.1002/arch.20319 PMID: 19514081 42. Ribeiro JMC, Genta FA, Sorgine MHF, Logullo R, Mesquita RD, Paiva-Silva GO, et al. An insight into the transcriptome of the digestive tract of the bloodsucking bug, Rhodnius prolixus. PLoS Negl Trop Dis. 2014; 8: e2594–e2594. https://doi.org/10.1371/journal.pntd.0002594 PMID: 24416461 43. Antunes L, Han J, Pan J, Moreira CJC, Azambuja P. Metabolic signatures of triatomine vectors of Try- panosoma cruzi unveiled by metabolomics. PLoS ONE. 2013; 8: 77283–77283. https://doi.org/10. 1371/journal.pone.0077283 PMID: 24204787 44. Gondim KC, Atella GC, Pontes EG, Majerowicz D. Lipid metabolism in insect disease vectors. Insect Biochem Mol Biol. 2018; 101: 108–123. https://doi.org/10.1016/j.ibmb.2018.08.005 PMID: 30171905 45. Bittencourt-Cunha PR, Silva-Cardoso L, Oliveira GAd, Silva JRd, Silveira ABd, Kluck GEG, et al. Peri- microvillar membrane assembly: the fate of phospholipids synthesised by the midgut of Rhodnius pro- lixus. Mem Inst Oswaldo Cruz. 2013; 108: 494–500. https://doi.org/10.1590/S0074- 0276108042013016 PMID: 23827998 46. Wood DE. Trypanosoma cruzi: fatty acid metabolism in vitro. Exp Parasitol. 1975; 37: 60–66. https:// doi.org/10.1016/0014-4894(75)90052-1 PMID: 1090440 47. Wood DE, Schiller EL. Trypanosoma cruzi: comparative fatty acid metabolism of the epimastigotes and trypomastigotes in vitro. Exp Parasitol. 1975; 38: 202–207. https://doi.org/10.1016/0014-4894(75) 90022-3 PMID: 1100424 48. Cazzulo JJ. Aerobic fermentation of glucose by trypanosomatids. FASEB J. 1992; 6(13): 3153–3161. https://doi.org/10.1096/fasebj.6.13.1397837 PMID: 1397837 49. 50. 51. Frydman B, Santos C, Cannata JJB, Cazzulo JJ. Carbon-13 nuclear magnetic resonance analysis of [1-13C]glucose metabolism in Trypanosoma cruzi. Evidence of the presence of two alanine pools and of two CO2 fixation reactions. Europ J Biochem. 1990; 192: 363–368. https://doi.org/10.1111/j.1432- 1033.1990.tb19235.x PMID: 2120054 Leroux AE, Maugeri DA, Opperdoes FR, Cazzulo JJ, Nowicki C. Comparative studies on the biochemi- cal properties of Microbiol Lett. 2011; 314: 25–33. https://doi.org/10.1111/j.1574-6968.2010.02142.x PMID: 21105905 the malic enzymes from Trypanosoma cruzi and Trypanosoma brucei. FEMS Zelada C, Montemartini M, Cazzulo JJ, Nowicki C. Purification and partial structural and kinetic charac- terization of an alanine aminotransferase from epimastigotes of Trypanosoma cruzi. Mol Biochem Para- sitol. 1996; 79: 225–228. https://doi.org/10.1016/0166-6851(96)02652-7 PMID: 8855559 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 24 / 25 PLOS PATHOGENSFatty acid metabolism in insect stages of Trypanosoma cruzi 52. Montemartini M, Bua´ J, Bontempi E, Zelada C, Ruiz AM, Santome´ JA, et al. A recombinant tyrosine aminotransferase from Trypanosoma cruzi has both tyrosine aminotransferase and alanine aminotrans- ferase activities. FEMS Microbiol Lett. 1995; 133: 17–20. https://doi.org/10.1111/j.1574-6968.1995. tb07854.x PMID: 8566704 53. Marciano D, Llorente C, Maugeri DA, de la Fuente C, Opperdoes F, Cazzulo JJ, et al. Biochemical char- acterization of stage-specific isoforms of aspartate aminotransferases from Trypanosoma cruzi and Trypanosoma brucei. Mol Biochem Parasitol. 2008; 161: 12–20. https://doi.org/10.1016/j.molbiopara. 2008.05.005 PMID: 18602174 54. Cazzulo JJ, Arauzo S, Franke de Cazzulo BM, Cannata JJB. On the production of glycerol and L-ala- nine during the aerobic fermentation of glucose by trypanosomatids. FEMS Microbiol Lett. 1988; 51: 187–191. 55. Boveris A, Hertig CM, Turrens JF. Fumarate reductase and other mitochondrial activities in Trypano- soma cruzi. Mol Biochem Parasitol. 1986; 19: 163–169. https://doi.org/10.1016/0166-6851(86)90121-0 PMID: 3523239 56. Cazzulo JJ. Intermediate metabolism in Trypanosoma cruzi. J Bioenerg Biomembr. 1994; 26: 157– 165. https://doi.org/10.1007/BF00763064 PMID: 8056782 57. Rogerson GW, Gutteridge WE. Catabolic metabolism in Trypanosoma cruzi. Int J Parasitol. 1980; 10: 131–135. https://doi.org/10.1016/0020-7519(80)90024-7 PMID: 6989775 58. Pereira MG, Visbal G, Salgado LT, Vidal JC, Godinho JLP, De Cicco NNT, et al. Trypanosoma cruzi epi- mastigotes are able to manage internal cholesterol levels under nutritional lipid stress conditions. PLoS ONE. 2015; 10: e0128949–e0128949. https://doi.org/10.1371/journal.pone.0128949 PMID: 26068009 59. Pereira MG, Visbal G, Costa TFR, Frases S, de Souza W, Atella G, et al. Trypanosoma cruzi epimasti- gotes store cholesteryl esters in lipid droplets after cholesterol endocytosis. Mol Biochem Parasitol. 2018; 224: 6–16. https://doi.org/10.1016/j.molbiopara.2018.07.004 PMID: 30016698 60. Allmann S, Mazet M, Ziebart N, Bouyssou G, Fouillen L, Dupuy JW, et al. Triacylglycerol storage in lipid droplets in procyclic Trypanosoma brucei. PLoS ONE. 2014; 9: e114628–e114628. https://doi.org/10. 1371/journal.pone.0114628 PMID: 25493940 61. Alvarez VE, Kosec G, Sant’Anna C, Turk V, Cazzulo JJ, Turk B. Autophagy is involved in nutritional stress response and differentiation in Trypanosoma cruzi. J Biol Chem. 2008; 283: 3454–3464. https:// doi.org/10.1074/jbc.M708474200 PMID: 18039653 62. Contreras VT, Salles JM, Thomas N, Morel CM, Goldenberg S. In vitro differentiation of Trypanosoma cruzi under chemically defined conditions. Mol Biochem Parasitol. 1985; 16: 315–327. https://doi.org/ 10.1016/0166-6851(85)90073-8 PMID: 3903496 63. Wainszelbaum MJ, Belaunzara´ n ML, Lammel EM, Florin-Christensen M, Florin-Christensen J, Isola ELD. Free fatty acids induce cell differentiation to infective forms in Trypanosoma cruzi. Biochem J. 2003; 375: 705–712. https://doi.org/10.1042/BJ20021907 PMID: 12887332 64. Contreras VT, Morel CM, Goldenberg S. Stage specific gene expression precedes morphological changes during Trypanosoma cruzi metacyclogenesis. Mol Biochem Parasitol. 1985; 14: 83–96. https://doi.org/10.1016/0166-6851(85)90108-2 PMID: 3885031 65. Homsy JJ, Granger B, Krassner SM. Some factors inducing formation of metacyclic stages of Trypano- soma cruzi. J Protozool. 1989; 36: 150–153. https://doi.org/10.1111/j.1550-7408.1989.tb01063.x PMID: 2657033 66. van Hellemond JJ, Opperdoes FR, Tielens AGM. The extraordinary mitochondrion and unusual citric acid cycle in Trypanosoma brucei. Biochem Soc Trans. 2005; 33: 967–967. https://doi.org/10.1042/ BST20050967 PMID: 16246022 67. Docampo R, de Boiso JF, Stoppani AO. Tricarboxylic acid cycle operation at the kinetoplast-mitochon- drion complex of Trypanosoma cruzi. Biochim Biophys Acta. 1978; 502: 466–476. https://doi.org/10. 1016/0005-2728(78)90079-8 PMID: 350277 68. Bringaud F, Rivière L, Coustou V. Energy metabolism of trypanosomatids: Adaptation to available car- bon sources. Mol Biochem Parasitol. 2006; 149: 1–9. https://doi.org/10.1016/j.molbiopara.2006.03.017 PMID: 16682088 69. Villafraz O, Biran M, Pineda E, Plazolles N, Cahoreau E, Souza ROO, et al. Procyclic trypanosomes recycle glucose catabolites and TCA cycle intermediates to stimulate growth in the presence of physio- logical amounts of proline. PLoS Pathog. 2021; 17: e1009204. https://doi.org/10.1371/journal.ppat. 1009204 PMID: 33647053 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009495 April 5, 2021 25 / 25 PLOS PATHOGENS
10.1371_journal.pone.0287250	RESEARCH ARTICLE Quantifying the economic effects of different fishery management regimes in two otherwise similar fisheries Christopher Liese, Scott CrossonID* NOAA Southeast Fisheries Science Center, Miami, Florida, United States of America * scott.crosson@noaa.gov Abstract In the southeast U.S., two very similar fisheries are managed by very different management regimes. In the Gulf of Mexico Reef Fish fishery, all major species are managed by individual transferable quotas (ITQs). The neighboring S. Atlantic Snapper-Grouper fishery continues to be managed by traditional regulations such as vessel trip-limits and closed seasons. Using detailed landings and revenue data from logbooks together with trip-level and annual, vessel-level economic survey data, we develop financial statements for each fishery to esti- mate cost structures, profits, and resource rent. By comparing the two fisheries from an eco- nomic perspective, we illustrate the detrimental effects of the regulatory measures on the S. Atlantic Snapper-Grouper fishery and quantify the difference in economic outcomes, includ- ing estimating the difference in resource rent. We find that the choice of fishery management regime shows up as a regime shift in the productivity and profitability of the fisheries. The ITQ fishery generates substantially more resource rents than the traditionally managed fish- ery; the difference is a large fraction of revenue (~30%). In the S. Atlantic Snapper-Grouper fishery, the potential value of the resource has almost completely dissipated via lower ex- vessel prices and hundreds of thousands of gallons of wasted fuel. Excess use of labor is a lesser issue. Introduction Fisheries management takes a variety of different forms, but the most important distinction from a resource economics perspective is the choice between controls to manage the fishery. In traditional U.S. fisheries management, total commercial harvest is indirectly managed by limiting how many vessels are able to harvest the resource by limiting the number of licenses, implementing maximum per-trip catch levels, opening and closing seasons for different spe- cies, requiring or forbidding particular types of gear, and other controls of fishing effort [1]. These approaches to management effectively raise the cost of effort in order to limit it. In con- trast, fisheries management with catch shares directly manages harvest by dividing the total harvest into separate quotas that exclusively harvested by individuals or groups with much fewer restrictions on fishing effort. As a result, fishers can focus on economic efficiency, i.e., maximizing revenues while minimizing costs. Hence, while both forms of regulation can a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Liese C, Crosson S (2023) Quantifying the economic effects of different fishery management regimes in two otherwise similar fisheries. PLoS ONE 18(6): e0287250. https://doi.org/10.1371/ journal.pone.0287250 Editor: John A. B. Claydon, Ocean Frontier Institute, CANADA Received: November 14, 2022 Accepted: June 2, 2023 Published: June 20, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0287250 Copyright: This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication. Data Availability Statement: NOAA landings data are considered confidential, including the economic information. However, properly aggregated data can be found at https://repository.library.noaa.gov/. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 1 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries Funding: The author(s) received no specific funding for this work. achieve biological goals (resource sustainability), catch shares are much better at capturing economic value for society from a renewable resource [2–4]. Competing interests: The authors have declared that no competing interests exist. Among the most economically efficient of catch shares are individual transferable quotas (ITQs) [5, 6]. Economists have demonstrated conceptually and empirically that introducing ITQs into fisheries rationalizes them, i.e., reduces excessive fishing effort and redundant investment thereby increasing productivity and hence economic profitability. In the U.S., fish- eries that have been able to increase profitability following the introduction of ITQs include Alaskan halibut [7], West Coast groundfish [8], and Gulf of Mexico red snapper [9]. In the southeast U.S., two otherwise very similar fisheries operate under two very different management regimes. Their similarity in all but management allow us to explore the divergent economic behavior and outputs that derive from them. The South Atlantic Snapper-Grouper fishery and the Gulf of Mexico Reef Fish fishery are geographically adjacent, utilize the same vessel and gear, catch the same species complex, are integrated into the same regional and national markets for inputs and outputs, and are both under U.S. federal management over- sight (and the same data collection regime). However, the two different regional fisheries man- agement councils have taken completely different approaches to management. The Snapper- Grouper fishery is intensely managed by traditional regulations while the Reef Fish fishery has been mostly transitioned to ITQs. We wish to add to the empirical literature by carefully comparing these two fisheries from an economic perspective. Using detailed landings and revenue data from logbooks together with trip-level and annual, vessel-level economic survey data, we develop financial statement for each fishery to estimate cost structures, profits, and resource rent. These detailed economic measures allow us to contrast the economic performance of these fisheries and to illustrate the detrimental effects of the regulations on the S. Atlantic Snapper-Grouper fishery; as well as to quantify the difference in economic outcomes, including estimating the difference in resource rent generated. We take a slightly different perspective from much of the empirical literature on resource rent, in that our primary focus is on the inefficiencies brought about in the tradi- tionally managed fishery rather than focusing on the efficiency gains of a fishery that transi- tioned to an ITQ [10]. Materials and methods Discussion of framing of rents / brief literature review Here, we define a commercial fishery that has achieved a state where overall fishing activity only covers business costs and opportunity costs as lacking resource rents [11]. We define any profits beyond business and opportunity costs as fishery rents. Our analysis below will use these definitions to illustrate the impact of traditional regulations and ITQs on two similar fisheries. The ability of an economy to extract resource rents from fisheries has long been a concern of resource economists. Businesses tend to focus on profits, but some profits are expected even in industries that lack significant barriers to entry (beyond the capital and skills required to enter a market). Many economists that study fisheries assume that for society to maximize the value of natural resources—a “gift of nature”—those resources should be extracted in a way that minimizes costs [12]. If a natural resource is privately owned, that may well occur, but commonly owned resources are a different story [13]. There are several significant barriers to extracting resource rents from commonly owned renewable resources (as most all wild caught fisheries are). The first is the long established sus- ceptibility of fishery stocks to overexploitation [14]. This is a standard commons-type prob- lem, with individual fishermen perhaps not intending to extract from the stock to the point of PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 2 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries diminishing returns, but the lack of incentives to manage jointly leading to that result overall. To a large extent, the point of fishery management is to prevent that overexploitation from occurring [15]. Unfortunately, a second barrier to extracting potential rents from fishery stocks may well be the fisheries management process itself. There are a number of potential ways to limit over- fishing by commercial fishermen. Limiting entry is certainly one of those, but even a limited- entry fishery will not necessarily lead to an efficient outcome because the commons problem still exists for the current participants [1]. In fact, regulated open access can lead to additional inefficiencies [16, 17]. Additional man- agement measures may include putting up obstacles to rational fishing behavior, such as limit- ing the harvest of particular species on each vessel or trip, limiting the use of the most efficient fishing gears, limiting fishing seasons, creating closed areas, and other restrictions that manag- ers develop to make the business of fishing less profitable and hence less palatable. An unregu- lated fishery will eventually drive down harvest to the point that fishermen will only pursue the species to the point of covering their business costs and opportunity costs. A regulated, limited entry fishery may do the same, although at a higher harvest level and with better protection of the underlying stock [18, 19]. In either case, the regulations restrict options and hence lead to inefficiencies. Contrasting systems have many names: rights based, market-based, rationalized, catch shares, ITQs, individual fishing quotas. They have in common that harvest is exclusively assigned to certain actors, thereby deterministically achieving a set quota (short of cheating, of course). The resource economics literature on the effects of ITQs on fisheries is vast. Research- ers have noted the effects of ITQs on sustainability [20], communities [21], safety [22], equality [23], and diversification [24]. Some of the social effects of ITQs can be negative, but our inten- tion here is not to add another case study of the effects of ITQs, but instead to demonstrate the inefficiencies of standard regulations by contrasting them with a neighboring alternative rights-based management system. The fisheries of the snapper-grouper species complexes in the Southeast U. S. The U.S. federal law decentralizes much of its fisheries management by handing management to regional fishery management councils (FMC). While the National Oceanic and Atmo- spheric Administration (NOAA) and its parent the U.S. Department of Commerce retain the ultimate responsibility for preventing overfishing, the FMCs debate and choose most of the actual regulations after consulting with their scientific advisory committees [25]. In the south- east U.S., the South Atlantic Fishery Management Council (SAFMC) and Gulf of Mexico Fish- ery Management Council (GFMC) are responsible for federal fisheries management. The SAFMC is responsible for managing fishing in federal waters from the North Carolina-Vir- ginia border to the bottom of the Florida Keys, and the GFMC oversees the federal waters from the Keys border to the Texas border with Mexico (Fig 1). The two areas are mostly at the same latitudes (although the SAFMC jurisdiction stretches further north), share similar weather (and hurricanes), are culturally similar, and are split by the Florida peninsula. The state of Florida is represented on both the SAFMC and GFMC. Most commercially valuable species found in one area are also present in the other, although the size of the stocks varies due to topography and ecology. The Gulf of Mexico area is both larger and contains a larger, shallower continental shelf. In both regions, commercial fishermen harvest the snapper-grouper species complex often associated with bottom structures; also known as reef fish. Species include over 40 different PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 3 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries Fig 1. Waters of the GFMC and SAFMC. Map created by Megan Slemons, Emory Center for Digital Scholarship. https://doi.org/10.1371/journal.pone.0287250.g001 snappers, groupers, tilefishes, and others. These fish are not the economically dominant spe- cies in either region as shrimp, lobsters and crabs, as well as other finfish, e.g., menhaden, make up the large majority of the seafood produced in lbs and dollar terms [26]. The commercial reef fish fisheries in both regions are also conducted in a similar manner. Vessels are very similar—around 35 feet in total length, built of fiberglass in the late 1980s, with about 400 horsepower engines, and using ice as refrigerant. They use the same type of gear, predominantly vertical lines, including hand lines, electric reels, and bandit gear. The next biggest gear groups in both regions are bottom longlines and diving. Also, very important from an economic perspective, the two fisheries are embedded in the same markets. The mar- kets for fish extends throughout the region and the nation, with dealers regularly shipping to NY. Also, fuel, labor, gear, and vessels are all sourced from the same southeast U.S. market, with prices matching and fluctuating together. In fact, the two fisheries are so similar, the managers and fishery scientists refer to these two fisheries by different names to help keep them apart. In the SAT, the species complex is usually referred to as snapper-grouper while in the GOM this complex is called reef fish. For the remainder of the paper, we will follow this convention and abbreviate “South Atlantic Snap- per-Grouper” as SAT SG and “Gulf of Mexico Reef Fish” as GOM RF. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 4 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries Managing these reef fish fisheries is biologically and technically complex. It is biologically complex because many aspects of their life cycles make them particularly susceptible to over- fishing. Some species live for decades and do not mature until they are several years old, and some are sequential hermaphrodites (beginning life as males and maturing into females later in life). Fisheries that disproportionately target larger fish, either because of desirability or reg- ulations (i.e., minimum size limits), will disproportionately affect the reproductive capacity of the entire stock. Another complication is that reef fish that are caught in waters deeper than approximately 100 feet will suffer barotrauma when quickly brought to the survey. This increases the discard mortality for caught fish that are released due to being the wrong size or out-of-season species (regulatory discards). This problem is compounded by the inability of fishermen to selectively target many of the individual species as they cohabitate the same eco- system. Most reef fish are also considered tasty fare and in high demand from both commercial and recreational fishermen. Overall, reef fish stocks are easy to target and easy to overfish, dif- ficult to select for, face a high discard mortality rate, and are easier to find in recent decades due to the advent of fish finders and GPS technology (storing and finding reefs or structure precisely). The GMFC and SAFMC have faced these challenges simultaneously. Beginning in the late 1980s, federal fisheries management in the U.S. began moving from an emphasis on commer- cial harvesting to biological sustainability [25]. While the GFMC Reef Fish Fishery Manage- ment Plan and the SAFMC Snapper-Grouper Fishery Management Plan both date to the mid 1980s, the Councils only began seriously regulating, i.e., restricting, the reef fish fisheries in the 1990s. This included the passage of various size limitations and then a new commercial license moratorium passed by the GFMC in 1996 and the SAFMC in 1997, limiting access to new fish- ers. The SAFMC started the first ITQ for finfish in the U.S. in 1993 for the wreckfish fishery, but has not passed another since [27]. After experiencing many of the problems with ever increasing regulations (short mini-sea- sons, derby fishing, high size limits and much discarding), the GFMC started its first ITQ for red snapper in 2007 [9], and has since started another for 13 species of groupers and tilefish [28, 29]. With the introduction of ITQ for red snapper, the GFMC removed the management by seasons and closures but also reduced the size limit, and instituted a big quota cut (which was reversed fairly quickly as the stock recovered). The majority of revenue from the GOM RF fishery now comes from species managed under ITQs, as we will describe below. In contrast, the SAT SG fishery is managed entirely with traditional regulations. Golden tilefish provides a useful practical example of the difference between the two management regimes in the two adjoining regions. In the GOM RF fishery, the commercial quotas for ITQ-managed species are annually converted into pounds and split among the quota owners proportionately to their share of the total quota, so a quota owner with 1% of the quota for tilefish, for example, is eligible to catch poundage equivalent to that portion of the overall quota or (alternatively) lease it to another vessel for its use. All commercial vessels harvesting tilefish thus need a limited-access reef fish permit and can either lease tilefish quota allocation (valid for a year) or permanently purchase tilefish shares (that annually spawn quota allocation) from others in the ITQ program (if they were not allocated them at the start of the program). The quota allocation applies to any of the tilefish species caught on the trip, and non-tilefish species co-caught on the trip (generally yellowedge grouper) will require a separate quota of its own (deepwater complex quota). There are some gear restric- tions, but most vessels will use the most efficient gear (bottom longline) as the species lives in deeper waters. The fishery is year-round, and vessels can time their trips to balance the freshness of the harvest with expected demand and the variable and opportunity costs incurred on trips of varying lengths. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 5 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries On the other side of Florida in the SAT SG fishery, the commercial tilefish quota is divided into a hook and line component (for all fishers possessing a limited entry snapper-grouper per- mit) and a longline component (for fishers possessing one of 23 vessels possessing an addi- tional golden tilefish endorsement). Regulations are designed to slow the overall harvest. The hook and line catch of golden tilefish is limited to 500 lbs per trip, and when that portion of the quota is met, the fishery is closed until the following year. In 2020, that portion of the fish- ery closed on July 23rd. This is still longer than the longline component, as their fishery was closed on February 18th as a precautionary measure; despite a 4000 lbs trip limit. After the quota was found not to have been met, their season was reopened for nine additional days on March 14th, then closed until January 2021. Golden tilefish from the Gulf is available year round, but the same species is only available until mid-year from the South Atlantic, and suf- fers from pulses and surges as fishermen race to catch as much as possible before the closures begin. Ethics statement Economic surveys of commercial fishermen were approved by the United States Office of Management and Budget (OMB) as part of NOAA’s regulatory authority. As with landings data, survey results tied to individual fishermen are considered confidential data and can be released only after aggregating to include at least three participants. Aggregated data can be found at the NOAA repositories at https://repository.library.noaa.gov/. Data For our comparative analysis, we use fisheries logbook and economic data from NOAA’s National Marine Fisheries Service’s (NMFS) Southeast Fisheries Science Center (SEFSC). Since 1993, the SEFSC has required all fishing vessels to report on their commercial fishing activity for federally managed species, including the SAT SG and GOM RF fisheries. Fishers must complete and submit a trip report (logbook) for every fishing trip to remain compliant with their federal fishing permits. The fishing logbooks are nearly a complete census of land- ings and effort in the federally managed commercial fisheries in the southeast U.S. To estimate revenue at the trip level (i.e., for each logbook), we multiply the logbook’s landings poundage by the most appropriate price available from the dealer landings data summarized in the SEFSC’s Accumulated Landings System (ALS). An algorithm matches dealer, state, month, and year between logbooks and ALS records for each species at the highest resolution possible. Since 2006, SEFSC economists have conducted two economic surveys to collect economic data at both the trip-level and annual, vessel-level to complement the logbook data. The eco- nomic surveys are designed to provide data that in turn can produce fishery-level financial statements; to measure and track the economic developments in these federally-managed fisheries. Each year, a random stratified sample of permitted vessels is selected to provide trip-level economic information. Selection eligibility is based on whether a vessel has a valid federal per- mit of interest during late November of the previous year. Approximately 30% of active vessels and 10% of inactive vessels are randomly sampled. For each fishing trip, selected vessels must complete the trip expense section located at the bottom of the trip report form. These variable cost questions include expenses for bait, ice, groceries, and IFQ leasing (paying to use another vessel’s shares); the amount of fuel used and the cost per gallon of fuel; whether or not the ves- sel owner was present on the trip, and whether or not payment for the catch was determined. If payment was determined, then gross trip revenue and payment to hired crew and hired cap- tain are collected (Fig 2). PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 6 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries Fig 2. Trip-level economic survey instrument (part of logbook report). https://doi.org/10.1371/journal.pone.0287250.g002 Early in the following year, selected vessels are mailed an annual expense survey (S1 Fig). This survey elicits annual, vessel-holistic economic data. The primary purpose of the annual expense survey is to collect fixed costs. These expenses include the costs for maintaining and repairing the vessel and gear, insurance, loan payments, and overhead (such as mooring, utili- ties, office staff, professional services, etc.). Because vessels often engage in (non-federal) fish- eries not in the logbook system, or engage in for-hire fishing, the survey also asks for annual cumulative trip-level expenses such as fuel, supplies, and hired crew payments for a complete picture of the annual, vessel-level expenses. To allow for comparison to the trip-level reporting and to help assign shares of fixed costs, the survey also collects the number of days at sea and total revenue for commercial and for-hire fishing. Finally, it collects an estimate of the vessel’s market value as proxy for the capital invested. The surveys involve three rounds of mail-outs, reminder calls, many call-backs, and send- backs when call-backs fail. Response rates are generally high by fishery standards. In 2016, the raw response rate at the trip-level was 98% and 99% in the SAT SG and GOM RF fisheries, respectively. The annual, vessel-level raw response rate was 77% and 88%, respectively. Many call-backs to respondents are necessary for getting missing values, clarifications, or validating outlier or odd numbers. Only records that are complete in all the financial fields can be used for generating the financial statements, i.e., item non-response cannot be tolerated for most questions. As a result, when counting only observations used in the analysis, the effective response rates drop to 94% and 94% at the trip-level and 71% and 82% at the annual, vessel- level for the SAT SG and GOM RF fisheries, respectively. Missing trip-level revenue or hired crew costs (these questions cannot always be answered by all fishers at the time the logbook is completed) are replaced with the estimated revenue from the logbook and a regression-based estimate of crew costs, respectively. Once the economic data is complete, a careful accounting exercise begins. Two economically relevant values are not collected on the surveys and are instead estimated for each trip or vessel. At the trip-level, the opportunity cost of the owner-operator’s time as captain is estimated based on hired crew compensation and profitability (due to share sys- tems). At the vessel-level, vessel depreciation is simply calculated at 5% of the vessel’s current market value. The resulting number is a rough estimate, identical to that used in the federal Gulf of Mexico shrimp fishery where depreciation is based on surveys and the fact that the Internal Revenue Service requires non-fishing vessels to be depreciated over 23 years. This paper’s comparative analysis of the economics of the SAT SG and GOM RF fisheries builds on earlier technical memoranda [30, 31]. Each report provides economic results for a specific subsets of the overall logbook and survey data, such as for the SAT SG or GOM RF fishery. Participation in a fishery is defined as catching at least one pound of the applicable spe- cies on a trip in the applicable waters, e.g., in the SAT or GOM. As sampling is at the vessel- level, prior to the fishing year, post-stratification is used to statistically estimate appropriate population means for the elements in the fishery financial statements (separately at the trip- PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 7 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries level and at the annual, vessel-level, as they are effectively two different data streams). All vessel and logbook trip data utilized in this report were pulled from the various databases on May 4, 2018. All dollar values are in nominal 2016 USD. Results Aggregating the census-level logbook data specific to each fishery reveals some important dis- tinctions between the patterns of the fishing in the two regions (Table 1). The first is the rela- tive size of the each fishery as the total landings in gutted-weight pounds in the GOM RF fishery are nearly three times those in the SAT SG one. The Gulf of Mexico is a larger area in terms of geography and habitat as the reef species are caught on the continental shelf and pref- erably in shallower waters. To account for the difference in scale of these two fisheries, we cal- culate the ratio of measures in the SAT SG fishery over the GOM RF fishery and then adjust for the difference in pounds (divide by 0.35), i.e., we prorate appropriately for a more legiti- mate comparison. For instance, revenues are higher in the GOM RF fishery and not just due to higher land- ings. The data shows that on average fishers in the GOM RF fishery received a higher price per pound (+$0.74). This implies that SAT SG fishers receive just 82% of revenue per pound of fish landed compared to the GOM RF fishers. Despite its much smaller catch, the SAT SG fleet is approximately the same size as the GOM RF one. When adjusted for pounds caught, the SAT SG fleet contains almost three times as many vessels as the GOM RF fleet. Even starker is the divergence in the number of trips taken in each fishery. For a given amount of landings, the SAT SG fleet takes almost five times more trips than the GOM RF one. It is also worth not- ing that the SAT SG fishery uses 29% more labor per unit of landings than the GOM RF fishery. Looking at 2016 fishing trips (Table 2), we see that GOM RF trips are much longer; averag- ing 4.4 days at sea compared to 1.7 days of SAT SG trips. On these longer trips, with a some- what larger crew, Gulf fishers on average land more than four and a half times as much, 2,262 pounds vs. 499 pounds, as fishers in the SAT SG fishery. Trip limits in the SAT SG fishery are leading to much shorter trips than would otherwise be taken. Regulations for most of the major commercial fishing species in the SAT SG fishery are intended to extend fishing oppor- tunities, i.e., preventing very short seasons, by imposing trip limits. Fig 3 shows two scatter plots of trips landing vermilion snapper, a major species, especially in the SAT SG fishery. The top and bottom plots are for the GOM RF and SAT SG fisheries, Table 1. Census-level aggregate data for the South Atlantic snapper-grouper (SAT SG) and Gulf of Mexico reef fish (GOM RF) fisheries (annual averages for the period 2014–2016). SAT SG GOM RF SG/RF Ratio SG/RF Ratio (per lb basis) SG or RF Fisheries Landings Price Revenue Trips Vessels Other landings All Landing on SG or RF Trips Landings Revenue Crew days https://doi.org/10.1371/journal.pone.0287250.t001 5,341,587 3.29 17,559,439 11,521 518 +11% 5,908,616 19,303,965 40,565 15,176,791 4.03 61,199,156 6,751 522 +5% 15,944,854 62,494,512 89,035 0.35 0.82 0.29 1.71 0.99 0.37 0.31 0.46 1.00 0.82 4.85 2.82 1.05 0.88 1.29 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 8 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries Table 2. Average fishing trip in 2016 in the South Atlantic snapper-grouper (SAT SG) and Gulf of Mexico reef fish (GOM RF) fisheries (census-level data). SAT SG (N = 11,521) GOM RF (N = 6,751) Days at sea Crew size Landings (lbs) % of Landings in Fishery https://doi.org/10.1371/journal.pone.0287250.t002 1.7 2.0 499 90% 4.4 2.8 2,262 96% respectively. Each trip is plotted relative to the scale of vermilion snapper revenue (X-axis; in $) and the specialization on vermilion snapper of the trip (Y-axis; in % of vermilion snapper revenue of total trip revenue). When comparing the two plots, we note that the one for the GOM RF fishery lacks any major discontinuities, but the one for the SAT SG fishery has two areas where trips group up along a seeming vertical line; at approximately $2000 and $3750 in vermillion snapper revenue. This strong behavioral response caused by the step-down trip lim- its implemented by the SAFMC to extend the fishing year. Specifically, as the vermilion snap- per quota gets closer to being reached during the year, the per-vessel trip limit for vermillion snapper steps down from 1000 lbs per trip to 500 lbs per trip (until the fleet quota is met and the season is closed). Somewhat less obvious but equally important to note in Fig 3, is that under the GFMC’s management system relatively few trips specialize, or solely target, vermillion snapper. For Fig 3. Scale and specialization of vermilion snapper trips. Distribution of trips across vermillion snapper revenue (in $) and share of revenue (in %) for the GOM RF (top) and SAT SN fisheries (bottom) in 2016. https://doi.org/10.1371/journal.pone.0287250.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 9 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries Fig 4. Monthly share of total revenue in the GOM RF (top panel) and SAT SG (bottom) fisheries in 2016. https://doi.org/10.1371/journal.pone.0287250.g004 most trips in the GOM RF fishery, vermillion snapper makes up less than a quarter of the reve- nue for the trip (>70% of trips), with relatively few exceeding half (~15% of trips). In contrast, vermillion snapper trips in the SAT SG fishery derive a much bigger share of their revenue from vermilion snapper. The vermilion snapper revenue exceeds 50% on about 55% of the trips. As will be explained more below, some of the higher specialization in the SAT vs. GOM fishery is likely due to SAT SG season closures of other species. Fig 4 shows the monthly share of total revenue generated by the two fisheries. The overall harvest level in the GOM RF fishery is much more stable month-to-month than in the SAT SG fishery. In the SAT SG in 2016 months late in the year generate less than a third of the revenue of top producing months, i.e., January and May. In contrast, in the Gulf the lowest month’s revenue is still at about two-thirds of the highest month’s revenue. In the SAT SG, a pattern of the fleet running out of quota happens consistently enough across the species that constitute the majority of fishing revenue such that fishers catch notice- ably less fish as the winter holiday season approaches. In Fig 4, the last three months of the year, i.e., 25% of the year, constitute only 10% of the year’s revenue. When we disaggregate Fig 4 by species for the SAT SG fishery the impact of seasons, clo- sures, and derby fishing behavior becomes more apparent. Fig 5 shows the monthly share of revenue for three species/species groups. For vermilion snapper (top panel), the quota is split into two seasons, starting in January and July. As each half-year season goes into effect, nearly half of the year’s quota is caught in those two first-months after opening. With continued high landings in the following two months quota is quickly drawn down. As a result, the fishery is all but closed in April through June and again October through December. The small spike in landings at the end of 2016 is due to NMFS’s oversight of the fishery. After estimating that some quota still remained following the October closure (and data lags), the agency reopened the fishery for a few days before the end of the 2016. The middle and bottom panel of Fig 5 show the equivalent results for the deepwater species complex and the shallow-water complex of the SAT SG fishery, respectively. The deepwater season opens in January, and two-thirds of landing occur in the first quarter of the year. For PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 10 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries Fig 5. Monthly share of landings revenue for vermilion snapper (top panel), the deepwater complex (middle), and shallow water complex (bottom) in 2016 in the South Atlantic snapper grouper fleet. https://doi.org/10.1371/journal.pone.0287250.g005 the shallow-water species, a spawning season closure is in effect for the first four months of every year. The expectation of closures due to exhaustion of the quotas induces some element of derby behavior, i.e., racing to fish behavior, in the SAT SG fishery. The use of some gear types in the SAT SG fishery correlates fairly strongly with certain spe- cies or species groups. For instance, longlines are used to catch half the deepwater complex landings, almost entirely during January through March. Diving equipment and traps are used May through October. As a result, much of this gear is idle for at least half a year. While some idle and redundant gear serves economic purposes, too much is a drain on productivity. It is also likely that the derbies and season closures are detrimental to the average ex-vessel prices fishers receive in the SAT SG fishery. Due to regulation, and unlikely to match market demand, a perishable product floods the market at times, only to disappear entirely at other times of the year (Fig 5). At least some of the 18% lower average price in the SAT SG is likely due to these fluctuations. Cost data also reveals substantial differences between the SAT SG and GOM RF fisheries. Based on sample trip-level data for 2016, Table 3 shows that the average GOM RF trip was far more lucrative than those taken in the SAT SG fishery; with gross revenues nearly five times as large. After accounting for trip-level economic costs, i.e., variable costs including the opportu- nity costs of the owner-operators’ time as captain, the GOM RF trips generate over eight times the surplus cash flow (here called trip net revenue) over SAT SG trips. While GOM RF trips PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 11 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries Table 3. Average trip-level economics in 2016 in the South Atlantic snapper-grouper (SAT SG) and Gulf of Mexico reef fish (GOM RF) fisheries (sample data). SAT SG (n = 2,612) GOM RF (n = 1,948) Owner-operated Days at sea Crew size Fuel used (gallons) Landings (lbs) Revenue ($) Costs ($) Fuel Bait Ice Groceries Miscellaneous Hired crew Opportunity cost—Owner-captain Trip net revenue ($) https://doi.org/10.1371/journal.pone.0287250.t003 Mean 82% 1.8 2.0 74 503 1,761 165 126 38 62 49 517 299 505 SE 3.1 0.2 0.1 7 57 206 15 23 5 10 16 84 35 74 Mean 68% 4.4 2.7 179 2,043 8,406 365 303 143 262 250 2,277 630 4,176 SE 3.5 0.2 0.1 11 169 757 21 31 14 21 34 247 101 442 are nearly three times the length of SAT SG trips and catch four times the landings, these scale measures are insufficient to explain the eight-times disparity in trip net revenues. We can compare two proxies for the technical productivity of these trips by calculating the average landings pounds per gallon of fuel used and average landings per crew-day of labor employed. The GOM RF trips generate 68% more landings per gallon of fuel use (11.4/6.8) and 20% more landings per crew-day of labor employed (169/141). This indicates that the SAT SG trips are very inefficient in fuel use vs. the GOM RF trips. At the same time, while labor is employed less efficiently than in the Gulf, the difference is much less than the disparity in fuel consumption. This finding is consistent with the idea that SAT SG trip limit regulation force fishers to cut short many trips, burning more fuel as they frequently return to port to unload. The two fleets have some differences in average trip-level production functions. While the SAT SG trips spend about twice as much (relative to revenue) on fuel and bait compared to GOM RF trips, they spend a very similar portion of revenue on groceries and miscellaneous expenditures. Ice expenditures and overall labor cost (hired crew and owner opportunity costs) as a share of revenue are about 30% higher in the SAT SG fishery. Dividing trip net reve- nue by total revenue generates the gross margin generated by the trip, i.e., the share of revenue available, after subtracting trip-level variable costs, to pay for fixed costs, for financing costs, for compensating the owner and invested equity, and pure profit such as resource rent. The trip-level margins are 49.7% and 28.7% for the GOM RF and SAT SG trips, respectively. A 21 percentage point difference on a margin or return in similar industries demands further explanation. The economic trip-level results- - -while based on a large sample size and being indicative of the economic situation- - -do not provide a full or holistic view of the economic situation in each fishery. Annual surveys reveal that the average vessel in each fishery also engages in some for-hire fishing work and commercial fishing for species beyond the SAT SG and GOM RF fisheries (Table 4). At the annual level, overall annual fishing days at sea and for-hire days at sea and revenue are similar, as is the vessel value. Costs related to the vessel, repair and mainte- nance, insurance, depreciation, and even overhead are also of similar magnitude. The biggest PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 12 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries Table 4. Average annual, vessel-level economics in 2016 in the South Atlantic snapper-grouper (SAT SG) and Gulf of Mexico reef fish (GOM RF) fisheries (sample data). SAT SG (n = 94) GOM RF (n = 121) Owner-operated Days-Commercial fishing Days-For-hire fishing Vessel value ($) Has insurance Total revenue ($) Commercial fishing For-hire fishing Costs($) Fuel Other supplies Hired crew Vessel repair & maintenance Insurance Overhead Opportunity cost—Owner Depreciation Net revenue from operations($) https://doi.org/10.1371/journal.pone.0287250.t004 Mean 89% 80 10 93,685 45% 69,373 57,489 11,883 7,037 10,015 19,274 10,503 1,478 7,100 9,052 4,684 230 SE 3.4 6.3 2.9 10,395 5.3 9,014 7,194 5,442 717 1,277 2,853 1,766 265 974 984 520 4,328 Mean 78% 74 10 85,688 38% 132,167 120,155 12,012 8,907 14,263 32,336 11,271 1,347 6,800 8,825 4,284 44,133 SE 3.4 3.8 3.0 6,327 4 16,043 15,483 3,625 832 1,152 3,942 1,066 200 749 1,100 316 11,310 differences are the revenue from commercial fishing (more than double in the Gulf), followed by hired crew and other supply costs. There are different ways to look at profits. As our objective is a societal, economic perspec- tive (vs. a individual business, financial perspective), we use what we call net revenue from operations. Net revenue from operations starts with operating revenue (i.e., excluding extraor- dinary, i.e., non-fishing, income) and subtracts all real, tangible costs of production. Beyond material inputs (fuel, repairs, etc.), the in-kind contributions to the production process must be accounted for as well. In our case, this includes the opportunity cost of owner-operator’s time spent as captain of the vessel and the vessel’s depreciation accounting for the degradation of the vessel with use and over time. Financial cost that do not represent an actual input to the production function, e.g., loan payments, IFQ purchases, or income taxes, are not counted. From a societal perspective many of these represent transfers of value generated by the fishery to others, i.e., are distributional in nature (and not our focus here). To be more representative of the fisheries than a single year, we collapse the cost categories, express them as percent of revenue, and then average them across three years (Table 5). We thereby generate an aggregate cost structure in percent-of-revenue terms and the economic profit margin implicit in the annual, vessel-level net revenue from operations. We will use these measures for the rest of the paper. In Table 6, we use the 3-year average cost structure and margins from Table 5, multiplied by the 3-year average annual fishery revenue (from Table 1) to estimate the total annual fishery expenses in each cost category and fishery-wide total profit (net revenue from operations). Note that the SAT SG and GOM RF landings that generate the revenue displayed are never landed in isolation from other species. Similarly, fixed costs components are never specific just to these fisheries. Hence the table’s results represent an abstraction (through standardization and prorating) of complex and messy fisheries data down to the a hypothetical concept of a pure SAT SG or GOM RF fishery, respectively. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 13 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries Table 5. Three-year average of economic costs and net revenue as percentage of revenue for the South Atlantic snapper-grouper (SAT SG) and Gulf of Mexico reef fish (GOM RF) fisheries (for the period 2014–2016). Revenue Costs (% of Revenue) Fuel & Supplies Labor—Hired & Owner Vessel R&M, Insure, Overhead Depreciation Net Revenue from Operations SAT Snapper-Grouper GOM Reef Fish 2014 (n = 75) 100% 27.1% 39.0% 23.0% 5.3% 5.6% 2015 (n = 101) 100% 24.6% 36.6% 25.7% 5.3% 7.7% 2016 (n = 94) 100% 24.6% 40.8% 27.5% 6.8% 0.3% Mean 100% 25.4% 38.8% 25.4% 5.8% 4.5% 2014 (n = 84) 100% 2015 2016 Mean (n = 105) (n = 121) 100% 100% 100% 18% 32% 14% 3.7% 31% 15% 31% 13% 3.0% 38% 18% 31% 15% 3.2% 33% 17.1% 31.5% 14.1% 3.3% 34.0% https://doi.org/10.1371/journal.pone.0287250.t005 Conceptually, to derive the resource rent from the net revenue from operations, it is neces- sary to subtract the opportunity costs of capital. The opportunity cost of capital accounts for “fair” compensation for the financial capital invested in the fishing vessel and business. For private businesses investment decisions, a large element accounts for the investment risk involved. Past studies in fisheries have assumed an opportunity cost of capital which is equal to the rate of return on a BAA rated bond, which is considered a somewhat risky bond [32]. Dur- ing the 2014–2016 time period, the rate for such bonds averaged 4.85%. For evaluating a publi- cally-owned natural resource at the aggregate industry level, and not to penalize the more capital intensive SAT SG fishery further, we use a more conservative opportunity cost of capi- tal of 3.5%. We apply this rate to the total market value of the effective vessels in each fishery. To calcu- late the number of effective vessels, we first use the days at sea (from the annual, vessel-level surveys) to prorate the total number of vessels between commercial and for-hire fisheries. In a second step, we prorate the commercial fishery effective vessels between the fishery of interest (SAT SG or GOM RF) and any other fisheries by using the revenues from the logbooks. In this manner, we calculate that the 518 vessels (partially) active in the SAT SG, are “equivalently engaged” as 272 hypothetical vessels that are fishing—solely—for SAT SG species. We then multiply this number by the 3-year average vessel value ($82,793) to estimate the value of the capital stock ($22.5 million invested capital specific to the SAT SG fishery). 3.5% of this capital stock corresponds to the $0.8 million under opportunity cost of capital in Table 6. The equiva- lent is done for the Gulf. Table 6. Estimated total annual economic costs, net revenue, and resource rent for the South Atlantic snapper-grouper (SAT SG) and Gulf of Mexico reef fish (GOM RF) fisheries (for the period 2014–2016). SAT SG GOM RF as % of Rev. in $ million as % of Rev. in $ million Revenue Costs Fuel & Supplies Labor—Hired & Owner Vessel R&M, Insure, Overhead Depreciation Net Revenue from Operations Opportunity Cost—Capital Resource Rent (approximate) https://doi.org/10.1371/journal.pone.0287250.t006 100.0% 25.4% 38.8% 25.4% 5.8% 4.5% 4.5% 0.1% 17.6 4.5 6.8 4.5 1.0 0.8 0.8 0.0 100.0% 17.1% 31.5% 14.1% 3.3% 34.0% 2.4% 31.6% 61.2 10.5 19.3 8.6 2.0 20.8 1.5 19.4 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 14 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries Our calculated resource rent is hence an approximation. First, we acknowledge that there may be some intramarginal rents (IMRs) being generated due to fleet heterogeneity [33]. Any such IMR in the SAT SG fishery would imply a negative resource rent. In the Gulf, large resource rents (approx) have been generated since the introduction of the ITQs [9, 34]. In 2006, pre-ITQ, similar to the SAT SG fishery today, the substantial rent was non-existent. As the ITQ years coincided with a consolidation of vessels, it is unlikely that IMR increased dur- ing this time. Importantly, most of the different regulations in each fishery apply uniformly to all fishers in each fishery, i.e., they should not be a source of heterogeneity within the fishery. We estimate the estimated resource rent in the GOM RF fishery between 2014–2016 in the broad range of $20 million per year or, in percent of revenue terms, over 30% of total revenue. In stark contrast, the SAT SG fishery seems to generate little or no resource rent. All the resource rents have dissipated due to the combination of an inability to limit costs or increase revenue when faced with the regulations. The difference between the two fisheries is 30% of revenue, i.e., a large fraction of total revenue. As we argued throughout the paper, the most likely culprit for this divergence in economic outcomes is the choice management regime. A sensitivity analysis on our assumptions would not change the central results due to their extreme divergence. Also, expanding the data to five years (2014–2018) makes no difference. In summary, in two fisheries with similar geography, biology, technology, and embedded in the same economic environment, the choice of management regime leads to very different economic outcomes. Specifically, the ITQ managed fishery generates substantial resource rent for society; on the order of a large fraction of revenue; while the traditionally managed fishery fails to capture most or all of the potential rents. Discussion Rather than provide analysis of any specific management action, we have attempted to calcu- late the cumulative economic effects of the differing management regimes used in two other- wise very similar southeast U.S. reef fisheries. We considered attempting to estimate the loss of revenue in the SAT SG resulting from the volatility of landings (due to species-specific sea- sons) but decided it was beyond the scope of this research as it would need to be conducted on a species-by-species basis; with sometimes thin price and market data. Nonetheless, the gluts generated by the race-to-fish when seasons open and the frequent shutdowns undoubtedly contribute negatively to the market bargaining position of the SAT SG fishers. Research on the demand for the SAT SG species, especially on how the ex-vessel prices relate to locally landed species, could be used to better predict possible gains if the SAT SG fishery was to bring prod- uct to the market in a more rational manner. Beyond the revenue loss, the SA fishery’s resource rent dissipation is largely due to the larger-than-necessary fleet. A fleet that is larger than necessary incurs additional fixed cost and opportunity cost of capital. The burning of hundreds of thousands of gallons of additional fuel, necessitated by extra travel, dissipates millions of dollars in additional value while adding to US carbon emissions. Despite the warnings in much of the literature about the effects of ITQs on crew labor, the ratio of the number of crew days between the SAT SG and GOM RF fisher- ies is only 1.29 when standardized by poundage (Table 1) and much of that additional labor is a product of constantly traveling back and forth to port. The aggregate expenditure numbers also hide the fact that the GOM RF fishery uses less labor, but pays it more. A known, but unquantified problem in the SAT SG fishery is regulatory discarding due to season closures and size limits. At its worst, fishers might be producing additional valuable product only to discard it. Any management changes that would allow fishers to keep even some of such catch—even if the quota is fixed—would have a disproportionally big economic PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 15 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries effect, as would any efforts by the fleet to work cooperatively to reduce bycatch [35]. The extent of these issues in the Gulf are likewise unknown, although ITQs are hardly free of discarding and high grading problems [36, 37]. Beyond its detrimental effects on economic profitability, there are additional problems with the SAFMC’s consistent use of trip limits as a management tool. The effects of regulatory trip limits are also difficult to predict, as tightening or loosening trip limits causes changes in effort induced by the changes in trip efficiency [38]. Hence NOAA has to take a precautionary approach to fishing seasons, closing them prematurely sometimes lest quotas be exceeded and result in shorter seasons the following year. Following a presentation of our conclusions to the SAFMC, we were also told that owner-operators sometimes spend additional funds keeping crews on payroll during closed seasons, further subtracting from commercial profits. Such payments would not be accounted for in our trip-level analysis, and it is unclear if they would be reported on the annual, vessel-level survey. Finally, there are potential spillover effects from closed fisheries, putting potential strain on other fisheries in the region. We have not discussed the recreational fishing fleet in this paper, but the GFMC has also experimented with a separate quota for the for-hire portion of that fishing sector in an effort to increase economic benefits [39, 40]. The SAFMC, in contrast, has not yet limited entry into for-hire fishing, let alone separated out management of it from the general recreational angler. We have also avoided discussing safety at sea issues, although there is evidence that the Gulf’s management has resulted in improvements [41], or compliance rates [42]. We have also not discussed some of the potential negatives of the GMFC ITQ programs, leaving that for other authors to explore. This paper focuses on economic efficiency and rent returns as a measure of management success. The GFMC introduced ITQs into the Reef Fish fishery primarily to reduce overcapacity and eliminate derbies [9]. The SAFMC has other goals for the Snapper Grouper Fishery Management Plan, including allowing consistent access across all sectors and maximizing social and economic opportunities [43]. Researchers have raised issues with ITQs leading to a loss of local level community [44], increasing fishermen’s dependence on particular fisheries [24], and causing employment loss [45]. Most troublesome, ITQs can cause wealth dissipation for non ITQ owners [46] and promote so-called “armchair fishing” [47], an accusation we have heard in about Gulf fisheries. These negative social effects may be present in the Gulf ITQ fisheries [48]. The older South Atlantic wreckfish ITQ may offer a less problematic alternative, shareholders can only lease quota to other shareholders and the fishery hence many of the same equity issues that may plague the Gulf [27]. Conclusion In summary, the SAFMC and GFMC take very different approaches to commercial fishing management. The GFMC has expanded ITQ management to most of the valuable commercial species in the GOM RF fishery, while the SAFMC continues to rely on traditional manage- ment. We find that the choice of fishery management regime shows up as a regime shift in the productivity and profitability of the fisheries. While many of our specific definitions or assumptions used in the derivation of these results could be adjusted or refined (according to each researcher’s judgement and research question focus), such changes will not eliminate the huge advantage in terms of economic performance of ITQs over traditional management. The decision to switch to ITQs or other catch shares is a politically charged one, and many of the criticisms—enrichment of the initial shareholders of quota at the expense of future ones (and of the public, who ultimately own the resource itself) and “armchair fishing” by share- holders who lease but do not fish—are serious and proven externalities of a different sort. Our analysis here shows that the decision to utilize a traditional management approach comes with PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 16 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries substantial economic consequences of its own. Trip limits, short seasons, and the resulting der- bies—and the rational response of capital stuffing [49]—have resulted in a renewable resource being utilized in a way that does not capture its potential economic value. Supporting information S1 Fig. Annual, vessel-level economic survey instrument (mail survey). (TIF) Author Contributions Conceptualization: Christopher Liese. Data curation: Christopher Liese. Investigation: Scott Crosson. Methodology: Christopher Liese. Project administration: Christopher Liese. Supervision: Christopher Liese. Validation: Christopher Liese. Visualization: Scott Crosson. Writing – original draft: Scott Crosson. Writing – review & editing: Scott Crosson. References 1. Waters JR. Restricted access vs. open access methods of management: Toward more effective regula- tion of fishing effort. Marine Fisheries Review. 1991; 53(3):1–10. 2. Arnason R. Minimum information management in fisheries. Canadian Journal of economics. 1990 Aug 1:630–53. 3. Birkenbach AM, Smith MD, Stefanski S. Feature—taking stock of catch shares: lessons from the past and directions for the future. Review of Environmental Economics and Policy. 2019 Feb 1. 4. Birkenbach AM, Kaczan DJ, Smith MD. Catch shares slow the race to fish. Nature. 2017 Apr 13; 544 (7649):223–6. https://doi.org/10.1038/nature21728 PMID: 28379945 5. Squires D, Campbell H, Cunningham S, Dewees C, Grafton RQ, Herrick SF Jr, et al. Individual transfer- able quotas in multispecies fisheries. Marine Policy. 1998 Mar 1; 22(2):135–59. 6. Costello C, Deacon R. The efficiency gains from fully delineating rights in an ITQ fishery. Marine Resource Economics. 2007 Jan 1; 22(4):347–61. 7. Herrmann M, Criddle K. An econometric market model for the Pacific halibut fishery. Marine Resource Economics. 2006 Jan 1; 21(2):129–58. 8. Toft JE, Punt AE, Little LR. Modelling the economic and ecological impacts of the transition to individual transferable quotas in the multispecies US west coast groundfish trawl fleet. ICES Journal of Marine Science. 2011 Jul 1; 68(7):1566–79. 9. Agar JJ, Stephen JA, Strelcheck A, Diagne A. The Gulf of Mexico red snapper IFQ program: The first five years. Marine Resource Economics. 2014 Jun 1; 29(2):177–98. 10. Gunnlaugsson SB, Saevaldsson H, Kristofersson DM, Agnarsson S. Resource rent and its distribution in Iceland’s fisheries. Marine Resource Economics. 2020 Apr 1; 35(2):113–35. 11. Arnason R, Bjørndal T, Gordon DV, Bezabih M. Measuring potential rents in the North Sea herring fish- ery. American Journal of Agricultural Economics. 2018 Apr; 100(3):889–905. 12. Flaaten O, Heen K, Matthı´asson T. Profit and resource rent in fisheries. Marine Resource Economics. 2017 Jul 1; 32(3):311–28. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 17 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries 13. Asche F, Bjørndal T, Gordon DV. Resource rent in individual quota fisheries. Land Economics. 2009 May 1; 85(2):279–91. 14. Gordon HS. The economic theory of a common-property resource: the fishery. Journal of political econ- omy. 1954 Apr 1; 62(2):124–42. 15. Scott A. The fishery: the objectives of sole ownership. Journal of political Economy. 1955 Apr 1; 63 (2):116–24. 16. DuPont DP. Rent dissipation in restricted access fisheries. Journal of Environmental economics and Management. 1990 Jul 1; 19(1):26–44. 17. Homans FR, Wilen JE. A model of regulated open access resource use. Journal of Environmental Eco- nomics and Management. 1997 Jan 1; 32(1):1–21. 18. Scheld AM, Anderson CM, Uchida H. The economic effects of catch share management: The Rhode Island fluke sector pilot program. Marine Resource Economics. 2012 Sep; 27(3):203–28. 19. Hsueh L. Quasi-experimental evidence on the “rights to fish”: the effects of catch shares on fishermen’s days at sea. Journal of the Association of Environmental and Resource Economists. 2017 Jun 1; 4 (2):407–45. 20. Costello C, Gaines SD, Lynham J. Can catch shares prevent fisheries collapse?. Science. 2008 Sep 19; 321(5896):1678–81. https://doi.org/10.1126/science.1159478 PMID: 18801999 21. McCay BJ. ITQs and community: An essay on environmental governance. Agricultural and Resource Economics Review. 2004 Oct; 33(2):162–70. 22. Pfeiffer L, Gratz T. The effect of rights-based fisheries management on risk taking and fishing safety. Proceedings of the National Academy of Sciences. 2016 Mar 8; 113(10):2615–20. https://doi.org/10. 1073/pnas.1509456113 PMID: 26884188 23. Sumaila UR. A cautionary note on individual transferable quotas. Ecology and Society. 2010 Sep 1; 15 (3). 24. Holland DS, Speir C, Agar J, Crosson S, DePiper G, Kasperski S, et al. Impact of catch shares on diver- sification of fishers’ income and risk. Proceedings of the National Academy of Sciences. 2017 Aug 29; 114(35):9302–7. https://doi.org/10.1073/pnas.1702382114 PMID: 28808006 25. Crosson S. The impact of empowering scientific advisory committees to constrain catch limits in US fisheries. Science and Public Policy. 2013 Apr 1; 40(2):261–73. 26. National Marine Fisheries Service. Fisheries Economics of the United States 2016. 2018. U.S. NOAA Tech. Memo. NMFS-F/SPO-187a, 243 p. 27. Yandle T, Crosson S. Whatever happened to the wreckfish fishery? An evaluation of the oldest finfish ITQ program in the United States. Marine Resource Economics. 2015 Apr 1; 30(2):193–217. 28. Holland DS, Thunberg E, Agar J, Crosson S, Demarest C, Kasperski S, et al. US catch share markets: a review of data availability and impediments to transparent markets. Marine Policy. 2015 Jul 1; 57:103–10. 29. Brinson AA, Thunberg EM. Performance of federally managed catch share fisheries in the United States. Fisheries Research. 2016 Jul 1; 179:213–23. 30. Overstreet E, Perruso L, Liese C. Economics of the U.S. South Atlantic Snapper-Grouper Fishery— 2016. 2018. NOAA Tech. Memo. NMFS-SEFSC-730. 104 p. 31. Overstreet E, Liese C. Economics of the Gulf of Mexico Reef Fish Fishery—2016. 2018. NOAA Tech. Memo. NMFS-SEFSC-725. 116 p. 32. Squires D. Productivity measurement in common property resource industries: an application to the Pacific coast trawl fishery. RAND Journal of Economics. 1992 Jul 1; 23(2):221–36. 33. Coglan L, Pascoe S. Separating resource rents from intra-marginal rents in fisheries’ economic survey data. Agricultural and Resource Economics Review. 1999 Oct; 28(2):219–28. 34. Gulf of Mexico Fishery Management Council. Grouper-Tilefish Individual Fishing Quota Program 5- Year Review. 2018 https://gulfcouncil.org/wp-content/uploads/Grouper-Tilefish-IFQ-Review-Final- version.pdf 35. Abbott JK, Haynie AC, Reimer MN. Hidden flexibility: institutions, incentives, and the margins of selec- tivity in fishing. Land Economics. 2015 Feb 1; 91(1):169–95. 36. Batsleer J, Hamon KG, van Overzee HM, Rijnsdorp AD, Poos JJ. High-grading and over-quota discard- ing in mixed fisheries. Reviews in Fish Biology and Fisheries. 2015 Dec; 25:715–36. 37. Bremner G, Johnstone P, Bateson T, Clarke P. Unreported bycatch in the New Zealand west coast South Island hoki fishery. Marine Policy. 2009 May 1; 33(3):504–12. 38. Pulver JR, Stephen JA, Larkin MF, Gray DiLeone AM. Retrospective Analyses of Commercial Trip Limit Efficacy in the Southeastern USA. Marine and Coastal Fisheries. 2019 Dec; 11(6):414–22. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 18 / 19 PLOS ONEQuantifying the economic effects of different fishery management regimes in two otherwise similar fisheries 39. Gulf of Mexico Fishery Management Council. Revision of the Red Snapper Recreational Sector Separa- tion Sunset Provision. 2016. https://gulfcouncil.org/fishery-management/implemented-plans/reef-fish/ 40. Abbott JK, Lloyd-Smith P, Willard D, Adamowicz W. Status-quo management of marine recreational fisheries undermines angler welfare. Proceedings of the National Academy of Sciences. 2018 Sep 4; 115(36):8948–53. https://doi.org/10.1073/pnas.1809549115 PMID: 30127021 41. Dakhlia S, Marvasti A. Regulatory Change, Market Structure, and Fatalities: The Case of the Gulf of Mexico Reef Fish Fishery. Review of Industrial Organization. 2020 Aug; 57:1–26. 42. Porter RD, Jylkka Z, Swanson G. Enforcement and compliance trends under IFQ management in the Gulf of Mexico commercial reef fish fishery. Marine Policy. 2013 Mar 1; 38:45–53. 43. South Atlantic Fishery Management Council. Amendment 49 to the Fishery Management Plan for the Snapper Grouper Fishery of the South Atlantic Region. 2023. https://safmc.net/documents/snapper- grouper-amendment-49-final/ 44. Carothers C. Fisheries privatization, social transitions, and well-being in Kodiak, Alaska. Marine Policy. 2015 Nov 1; 61:313–22. 45. Olson J. Understanding and contextualizing social impacts from the privatization of fisheries: An over- view. Ocean & Coastal Management. 2011 May 1; 54(5):353–63. 46. Pinkerton E, Edwards DN. The elephant in the room: the hidden costs of leasing individual transferable fishing quotas. Marine Policy. 2009 Jul 1; 33(4):707–13. 47. Soliman A. Using individual transferable quotas (ITQs) to achieve social policy objectives: A proposed intervention. Marine Policy. 2014 Mar 1; 45:76–81. 48. National Academies of Sciences, Engineering, and Medicine. The Use of Limited Access Privilege Pro- grams in Mixed-Use Fisheries. 2021. Washington, DC: The National Academies Press. https://doi.org/ 10.17226/26186. 49. Townsend RE. On "capital-stuffing" in regulated fisheries. Land Economics. 1985 May 1; 61(2):195–7. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287250 June 20, 2023 19 / 19 PLOS ONE
10.1371_journal.pone.0284899	RESEARCH ARTICLE Factors related to changes in visual symptoms after successful photodynamic therapy in central serous chorioretinopathy Geun Woo Lee1, Yun Young Kim1, Kyung Jun Choi2, Se Woong KangID 2* 1 Department of Ophthalmology, Daegu Catholic University School of Medicine, Daegu, Korea, 2 Department of Ophthalmology, Samsung Medical Center, Sungkyunkwan University School of Medicine, Seoul, Korea * swkang@skku.edu Abstract To investigate biomarkers related to visual symptom and best corrected visual acuity (BCVA) improvement after photodynamic therapy (PDT) for central serous chorioretinopa- thy. This retrospective cross-sectional study involved 42 consecutive eyes, from 42 patients who underwent successful PDT, divided into two groups according to improvement in sub- jective visual complaints: complete (20 eyes) and incomplete recovery (22 eyes). The clini- cal characteristics of each group, including central foveal thickness (CFT), foveal avascular zone (FAZ) area, and degree of change in signal voiding of the choriocapillaris on optical coherence tomography angiography, were compared. Correlations between best-corrected visual acuity (BCVA) and baseline clinical features were investigated. At baseline, CFT and FAZ areas showed significant differences between the two groups (all p < 0.05). Multiple binary logistic regression analysis revealed that greater CFT predicted complete recovery from visual complaints (p = 0.002). Reduction or disappearance of signal voiding in the chor- iocapillaris 6 months post-PDT occurred more frequently in the complete recovery group (p < 0.05). FAZ area before PDT correlated with BCVA before and 6 months after PDT and BCVA improvement during the study period (all p < 0.05). CFT and FAZ area before PDT correlated with completeness of visual symptom recovery after PDT. Smaller FAZ area before PDT correlated with better BCVA before and after treatment. Introduction Central serous chorioretinopathy (CSC) is characterized by a leakage point at the retinal pig- ment epithelium (RPE) level, accompanied by serous detachment of the neurosensory retina at the posterior pole [1]. It is related to choroidal vasculature abnormalities, such as congestion of the choroidal vessels and choroidal hyperpermeability [2–4]. Patients generally complain of symptoms, such as decreased visual acuity, central scotoma, metamorphopsia, and hyperopia due to subretinal fluid (SRF) at the fovea [5]. CSC occurs preferentially as an acute disease and often improves on its own. If SRF remains for 4–6 months or longer, leading to a chronic a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Lee GW, Kim YY, Choi KJ, Kang SW (2023) Factors related to changes in visual symptoms after successful photodynamic therapy in central serous chorioretinopathy. PLoS ONE 18(4): e0284899. https://doi.org/10.1371/journal. pone.0284899 Editor: Daisuke Nagasato, Tsukazaki Hospital, JAPAN Received: January 2, 2022 Accepted: April 1, 2023 Published: April 21, 2023 Copyright: © 2023 Lee et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information file. Funding: The author(s) received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284899 April 21, 2023 1 / 13 PLOS ONEVisual symptom change after PDT in CSC course, symptoms like decreased visual acuity and scotoma may persist permanently [6]. Pho- todynamic therapy (PDT) is an effective treatment for chronic CSC, and intravitreal injection of anti-vascular endothelial growth factor is ineffective [7,8]. If PDT is successful, the SRF dis- appears and the retinal sensitivity increases on microperimetry [9,10]. However, even after successful PDT, the degree of recovery from subjective visual symptoms (e.g., central scotoma, metamorphopsia, decreased visual acuity) may be insufficient [11,12]. Optical coherence tomography angiography (OCTA) is a noninvasive method that makes it possible to easily identify the microvasculature of the retina and choroid [13]. Several studies have used optical coherence tomography (OCT) and OCTA to study the retinal and choroidal changes that appear after PDT [11,14–17]. However, there are few studies on the clinical char- acteristics related to the changes in subjective visual symptoms after PDT. Despite the presence of SRF in CSC, best corrected visual acuity (BCVA) could be maintained occasionally [18]. Even though some patients with CSC complain of visual symptoms, they may have normal or near-normal visual acuity when the change in refraction caused by SRF is corrected. In light of this, it has been noted that BCVA may not be an accurate representative of the changes in visual symptoms reported by patients. The purpose of this study was to investigate the clinical characteristics that influence changes in visual symptoms and BCVA after PDT in CSC patients, using OCT and OCTA. Methods Subject selection and design This retrospective cross-sectional study was performed at a single center and approved by the Institutional Review Board of Samsung Seoul Hospital (IRB no. 2018-07-036). The require- ment for written informed consent was waived because of the retrospective design of the study, which was conducted in accordance with the tenets of the Declaration of Helsinki. PDT was performed by a single retinal specialist. We enrolled patients who underwent PDT due to CSC, had symptoms for more than 3 months, and had been followed up without recurrence for at least 6 months. The subjects were recruited retrospectively from January 2017 to April 2019. Successful PDT was defined as complete resolution of SRF on OCT at 1 or 3 months after PDT and, at the same time, no evidence of recurrence (SRF) on OCT 6 months after PDT. Before conducting the PDT, all subjects underwent a comprehensive ophthalmic examina- tion, including measurements of best-corrected visual acuity (BCVA), manifest refraction, anterior segment examination using a slit lamp, dilated fundus examination, spectral-domain optical coherence tomography (SD-OCT, Spectralis HRA-OCT; Heidelberg Engineering, Hei- delberg, Germany), and swept-source optical coherence tomography angiography (SS-OCTA, DRI OCT Triton; Topcon Corporation, Tokyo, Japan). Dilated fundus examination, BCVA, SD-OCT, and SS-OCTA were performed 1 and 6 months after PDT. Exclusion criteria included a history of PDT or laser photocoagulation due to previous CSC, history of anti-vas- cular endothelial growth factor injection treatment in either eye within 3 months, history of other retinal or choroidal diseases other than CSC, reduced OCTA image quality due to media opacity such as cataract (Topcon image quality index <50), and significant myopia (spherical equivalent <−8 D or axial length �26 mm). Moreover, cases of choroidal neovascularization (CNV) diagnosed based on OCTA findings and patients with a history of taking eplerenone were excluded. Pre-PDT (baseline), subjects were also examined for chief complaints (subjective visual symptoms), symptom duration (period from the onset of subjective symptoms to PDT), and systemic diseases. The degree of subjective visual symptoms after PDT, in comparison with PLOS ONE \| https://doi.org/10.1371/journal.pone.0284899 April 21, 2023 2 / 13 PLOS ONEVisual symptom change after PDT in CSC those at baseline, was investigated at every visit. Subjective visual symptoms mentioned by the subject at baseline were classified into four categories: decreased visual acuity, scotoma, meta- morphopsia, and decreased visual acuity with scotoma. In general, CSC is known to complain of various and complex symptoms. Representatives include decreased visual acuity, blurred vision, relative central scotoma, metamorphosis, moderate dyschromatopsia, micropsia, and reduced contrast sensitivity [1]. The patients included in this study also complained about complex symptoms. After explaining the possible symptoms, the main symptom was chosen by themselves. Interestingly, all chosen the main symptoms were included in the four catego- ries mentioned above. According to the 6 months post-PDT results, these were divided into two groups, “complete recovery” if the symptoms subsided completely and “incomplete recov- ery” if the symptoms were relieved incompletely or did not change. Several OCT and OCTA variables at before and 1, 3, and 6 months after PDT were mea- sured by two examiners (G.W.L. and K.J.C.) who were blinded to all medical information. Measurement of the image of the optical coherence tomography An image of a raster scan with a size of 20˚ × 20˚ was obtained using OCT in the enhanced depth imaging mode. The central foveal thickness (CFT) was taken as the average of the dis- tance between the vitreoretinal interface and photoreceptor outer segment, which passed per- pendicular to the fovea on the horizontal and vertical scans. SRF height was defined as the average value of the distance between the inner border of the RPE and outer border of the interdigitation zone, passing perpendicular to the fovea on a horizontal and vertical scan. If hyperreflective material was present, it was used as the boundary of the SRF space. Subfoveal choroidal thickness (SFCT) was set as the average value of the distance between the outer bor- der of the RPE and the sclerochoroidal junction, passing perpendicular to the fovea on hori- zontal and vertical scans. Before PDT, the presence of RPE detachment, hypertrophic outer retinal changes, RPE undulation, and disruption of the ellipsoid zone were identified. The CFT, SFCT, presence of RPE undulation, and disruption of the ellipsoid zone were also identi- fied at 1, 3, and 6 months after PDT. The process of analysis of optical coherence tomography angiography SS-OCTA images that automatically segmented the 3 × 3 mm2 macula area were identified using a viewing software. The en-face images of the superficial capillary plexus (SCP), deep capillary plexus (DCP), and choriocapillaris’ profiles were exported as quality-preserving JPEG files. Vessel density and the foveal avascular zone (FAZ) in the SCP and DCP were measured using an open-source software (ImageJ version 1.52a; National Institutes of Health, Bethesda, MD, USA; http://imagej.nih.gob/ij/). The FAZ was measured from the outline obtained using the polygon selection tool. The vessel density was calculated using methods similar to those reported previously (Fig 1D and 1E) [16,19]. Briefly, the area was converted to a binarized 8-bit image and measured to include the maximum number of vessels using threshold and binarization functions. The density was then obtained by dividing the area excluding the FAZ area by the total area of 9 mm2. In addition, two independent observers intuitively judged the signal void of choriocapillaris and divided them into two categories: “increasing or stationary” and “decreasing or resolved,” according to the comparison of the signal void between the time when the SRF had completely subsided (1 or 3 months post-PDT) and at 6 months post-PDT (Fig 1C and 1F). In cases with different results, a consensus was reached through discussion. The repeatability of the degree of change in the signal void was assessed using Cohen’s kappa coefficient. Cohen’s kappa coefficient was 0.788 (p < 0.001), which corresponds to substantial agreement. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284899 April 21, 2023 3 / 13 PLOS ONEVisual symptom change after PDT in CSC Fig 1. The analysis process of the en-face OCTA image. The 3 mm × 3 mm-sized en-face image of the SCP (A) and DCP (B) was converted to a binarized image (D and E) using ImageJ software. From the binarized image, the FAZ area (black dotted grayish area) was measured at the center, and the black colored vessel area was measured. Vessel density was obtained by dividing the vessel area by the area where the FAZ area was removed by 9 mm2. The signal void (F) of the choriocapillaris at 6 months post-PDT is smaller than the signal void (C) of the choriocapillaris after the subretinal fluid was completely absorbed 1 month post-PDT. https://doi.org/10.1371/journal.pone.0284899.g001 Photodynamic therapy procedure All subjects included in this study underwent half-fluence PDT with verteporfin (Visudyne; Novartis, Basel, Switzerland). We used a standard dose of verteporfin (6 mg/m2), 689 nm laser, a light intensity of 600 mW/cm2, and a shortened irradiation time of approximately 40 s. The delivered radiation covered the area of choroidal hyperpermeability, which had engendered subfoveal SRF, in the mid or late phase of indocyanine green angiography. Statistical analysis BCVA was converted to the logMAR scale, and manifest refraction was converted to a spheri- cal equivalent for analyses. All continuous variables are reported as median (interquartile range). The Mann–Whitney U test was used to compare continuous variables between the two groups. To compare categorical variables between the two groups, cross tabulation analyses (chi-squared test) were used. For repeated measures, a generalized estimation equation was used to analyze the differences between groups and visits. In univariate analysis, when the p- value was less than 0.05, it was used as an explanatory variable for multivariate analysis, and multiple binary logistic regression was used for multivariate analysis. In addition, the cut-off PLOS ONE \| https://doi.org/10.1371/journal.pone.0284899 April 21, 2023 4 / 13 PLOS ONEVisual symptom change after PDT in CSC value in the continuous variable showing a significant difference between the two groups was also checked through the receiver operating characteristic curve. And, spearman correlation analysis was performed to identify factors that correlated with BCVA or changes in BCVA. Results Of the 69 eyes (69 CSC patients) examined retrospectively, 10 eyes met the exclusion criteria and 12 eyes were lost to follow-up. Among 47 eyes (47 CSC patients), five eyes that did not experience subsiding SRF up to 3 months after PDT and one eye with recurrence at 6 months were excluded from the study. Eventually, 42 eyes (42 patients with CSC) were analyzed (ratio of successful PDT = 89.4%). Demographics, OCT and OCTA measurements, and subjective visual symptoms before PDT (baseline) are presented in Table 1. The mean age of all subjects was 49.0(13) years, and included 32 men (76%) and 10 women (24%). At baseline, the mean logMAR BCVA was 0.10(0.16), and the mean CFT was 150(34)μm. The mean area of the SCP Table 1. Comparison of baseline demographics and clinical characteristics between the two groups. Variable Age, years Sex, M:F Symptom duration, months Systemic disease Diabetes mellitus, n (%) Hypertension, n (%) BCVA, logMAR Spherical equivalent OCT measurements CFT, μm SRF height, μm SFCT, μm Presence of PED, n (%) RPE undulation, n (%) Hypertrophic outer retinal change, n (%) OCTA measurements SCP FAZ, mm2 SCP VD, % DCP FAZ, mm2 DCP VD, % Visual symptoms at diagnosis Decreased visual acuity, n (%) Scotoma, n (%) Decreased VA & scotoma, n (%) Metamorphopsia, n (%) Total (n = 42) Complete recovery group (n = 20) Incomplete recovery group (n = 22) p-value 49.0 (13) 32:10 5 (3) 4(9.5) 2(10.0) 0.10 (0.16) 0 (6) 150 (34) 123 (72) 401 (81.5) 9(21.4) 18(42.9) 27(64.3) 0.29 (0.06) 35.9(2.2) 0.31(0.07) 36.9(1.6) 12 (28.6) 14 (33.3) 14 (33.3) 2 (4.8) 47.0 (16) 17:3 3.5(3) 3(15.0) 0(0) 0.05(0.14) 0 (1) 159 (46) 113 (91) 402.5 (60) 5(25.0) 8(40.0) 4(80.0) 0.25(0.07) 35.7(2.3) 0.28(0.09) 36.5(1.9) 9 (40.0) 7 (35.0) 5 (25.0) 0 (0) 50.5 (8) 15:7 5(3) 1(4.5) 2(4.8) 0.10(0.11) 0 (1.5) 131(48) 147 (31) 385.5(88) 4(18.2) 10(45.5) 11(50.0) 0.31(0.04) 35.9(2.1) 0.34(0.06) 37.1(1.7) 4 (18.2) 7 (31.8) 9 (40.9) 2 (9.1) 0.371 0.284 0.058 0.333 0.221 0.139 0.654 <0.001 0.202 0.155 0.714 0.764 0.082 <0.001 0.860 0.005 0.623 0.175 1.000 0.338 0.489 p-values were obtained by mann–whitney U test and pearson’s chi-squared test. Symptom duration refers to the period from occurrence of subjective symptom to PDT. All continuous variables were presented by median(interquartile range). Abbreviations: CSC, central serous chorioretinopathy; BCVA, best corrected visual acuity; logMAR, logarithm of the minimum angle resolution; OCT, optical coherence tomography; OCTA, optical coherence tomography angiography; CFT, central foveal thickness; SRF, subretinal fluid; SFCT, subfoveal choroidal thickness; PED, pigment epithelial detachment; RPE, retinal pigment epithelium; SCP, superficial capillary plexus; DCP, deep capillary plexus; FAZ, foveal avascular zone; VD, vessel density. https://doi.org/10.1371/journal.pone.0284899.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0284899 April 21, 2023 5 / 13 PLOS ONEVisual symptom change after PDT in CSC Fig 2. Horizontal section of OCT image and OCTA image of SCP of a 67-year-old man in the incomplete recovery group (A, B, C) and 63-year-old man in the complete recovery group (D, E, F). Compared with the 67-year-old man (incomplete recovery group), the 63-year-old man (complete recovery group) had better BCVA at baseline and 6 months after PDT (logMAR BCVA, 0.05 vs. 0.22 and 0.05 vs. 0.10, respectively), and shorter symptom duration (4 months vs. 7 months, respectively). Moreover, the central foveal thickness before (A, D) and after (C, F) PDT was thicker (182 μm vs. 145 μm and 201 μm vs. 135 μm, respectively), and the SCP FAZ area (B, E) before PDT was narrower (0.22 mm2 vs. 0.32 mm2, respectively). https://doi.org/10.1371/journal.pone.0284899.g002 FAZ and DCP FAZ was 0.29(0.06)mm2 and 0.31(0.07)mm2, respectively, and mean vessel den- sity of the SCP and DCP was 35.9(2.2)%, and 36.9(1.6)%, respectively. Regarding the frequency of the four subjective visual symptoms, 12 eyes (28.6%) had decreased visual acuity, 14 (33.3%) had scotoma, 14 (33.3%) had decreased visual acuity and scotoma, and 2 (4.8%) had metamor- phopsia. When comparing the subjective visual symptoms at baseline and 6 months after PDT, patients in whom symptoms completely disappeared were included in the complete recovery group (20 eyes). Patients in whom symptoms did not disappear were included in the incom- plete recovery group (22 eyes). The complete recovery group before PDT had greater CFT than that of the incomplete recovery group (p < 0.01). In addition, in the complete recovery group, SCP FAZ and DCP FAZ were smaller (all p < 0.01). Fig 2 shows representative cases for the two groups. A generalized estimation equation has been used to investigate the change in repeated mea- sures according to visits and groups (Table 2). BCVA, CFT, SFCT, SCP FAZ, and DCP FAZ were analyzed. Further, the differences according to visits and groups, separately and simulta- neously, and after adjusting for the age, sex, and symptom duration were comprehensively analyzed. Although the groups did not differ in BCVA or choroidal thickness, the differences with visits were significant (all p < 0.05). For CFT, SCP FAZ, and DCP FAZ, differences among visits were insignificant but those among groups were significant (all p < 0.01). When the signal void of the choriocapillaris at 6 months after PDT was compared with the signal void immediately after the resolution of SRF, the ratio of subjects whose signal void was reduced or disappeared was greater in the complete recovery group (p = 0.012).Multiple binary logistic regression analyses were performed to identify the independent factors associated with the recovery of subjective visual symptoms. Variables with a p-value < 0.05 in the univariate PLOS ONE \| https://doi.org/10.1371/journal.pone.0284899 April 21, 2023 6 / 13 PLOS ONETable 2. Comparison of sequential characteristics by groups and by the visit. Variable Visit Complete recovery group Incomplete recovery group p-value(Crude) p-value(Adjusted) † Visual symptom change after PDT in CSC BCVA, logMAR CFT, μm SFCT, μm SCP FAZ, mm2 DCP FAZ, mm2 Baseline 1 month 3 months 6 months Baseline 1 month 3 months 6 months Baseline 1 month 3 months 6 months Baseline 1 month 3 months 6 months Baseline 1 month 3 months 6 months 0.05(0–0.13) 0.13(0.05–0.22) < .001 0.139 V G VG 0.433 V < .001 G 0.11 VG 0.427 0(0–0.07) 0(0–0) 0(0–0.05) 159(152.5–174.5) 158.5(147.5–175) 166.5(150–178.5) 164.5(154–180) 391(389–438) 377(339.5–425) 379(318.5–460.5) 368(339.5–425) 0.25(0.23–0.30) 0.25(0.24–0.31) 0.25(0.25–0.32) 0.26(0.24–0.30) 0.28(0.27–0.36) 0.29(0.28–0.32) 0.30(0.28–0.33) 0.30(0.28–0.33) 0.05(0–0.1) 0(0–0.1) 0(0–0.05) 130.5(102–150) 142.5(131–146) 142(121–158) 148(112–162) 0.092 < .001 0.063 0.087 < .001 0.067 385.5(341–422) 0.048 0.253 0.059 0.038 0.157 0.095 397(333–433) 346(321–376) 366.5(330–415) 0.31(0.29–0.33) 0.32(0.30–0.33) 0.31(0.30–0.33) 0.31(0.29–0.33) 0.34(0.31–0.36) 0.33(0.31–0.38) 0.33(0.32–0.38) 0.33(0.32–0.38) 0.389 < .001 0.343 0.397 < .001 0.42 0.083 < .001 0.078 0.076 < .001 0.084 Abbreviations: V, visit; G, group. All values were presented by median (interquartile range). p-values were obtained by generalized estimation equation (GEE). †: Adjusted results were obtained with covariate variable as age(years), sex and symptom duration(months). https://doi.org/10.1371/journal.pone.0284899.t002 analysis were used as explanatory variables in the multivariate analysis. The explanatory vari- ables satisfying the condition were CFT, SCP FAZ, and DCP FAZ (all p < 0.01). Because of the high correlation among explanatory variables that were significant in the univariate analysis, forward (condition) was used as the variable selection method to resolve multicollinearity. In multivariate analysis, only 1 variable(CFT) was selected, and when CFT increased by 1 unit, the odds ratio of incomplete recovery was 0.931 times higher than that of complete recovery (95% confidence interval: 0.890 to 0.975, p = 0.002). Among the factors differing significantly between the two groups, the cutoff CFT and SCP FAZ (using the Youden index) were obtained from the receiver operating characteristic curve (Fig 3). Recovery of subjective symptoms was more likely to be incomplete when CFT was < 139.5 μm or SCP FAZ was > 0.265 mm2. Among the measures of CFT, SCP FAZ, and DCP FAZ at baseline, the factors related to BCVA were investigated (Fig 4). Baseline BCVA, 6 months post-PDT BCVA, and the amount of BCVA changes over 6 months showed a significant relationship with SCP FAZ and DCP FAZ at baseline. In other words, as the FAZ area was wider, baseline BCVA and 6 months post-PDT BCVA were worse; however, the magnitude of BCVA improvement was greater. Discussion PDT is a treatment involving irradiation of lesions visible on angiography with a laser after verteporfin injection, known to be effective in chronic CSC. It is presumed that free radicals generated under the damaged RPE site damage the vascular endothelium, resulting in PLOS ONE \| https://doi.org/10.1371/journal.pone.0284899 April 21, 2023 7 / 13 PLOS ONEVisual symptom change after PDT in CSC Fig 3. Receiver operating characteristic curves of baseline CFT and SCP-FAZ. The area under the curve was 0.858 for CFT and 0.816 for SCP FAZ. The cutoff value (asterisk) using the Youden index was 139.5 μm for CFT and 0.265 mm2 for SCP FAZ. CFT, central foveal thickness; SCP, superficial capillary plexus; FAZ, foveal avascular zone. https://doi.org/10.1371/journal.pone.0284899.g003 hypoperfusion and remodeling of the choriocapillaris [20]. In this regard, side effects, such as decreased visual acuity and visual field impairment, have been reported after standard full-set- ting PDT. Recently, to lower the risk of side effects, reduced dose and reduced time PDT have been mainly used and reported to be effective treatments [21–23]. In several previous studies, metamorphopsia, and not visual acuity, was reported to be a fac- tor significantly associated with vision-related quality of life in diseases involving the macula (e.g., epiretinal membrane, macular hole, macular-off retinal detachment) [24–26]. In other words, there are visual symptoms that cause discomfort to patients’ and interfere with their daily life much more than decreased visual acuity. We focused on these points to investigate the changes in subjective visual symptoms. We divided the patients into two groups to deter- mine the factors that influenced the change in subjective visual symptoms. Those with com- plete disappearance of subjective visual symptoms six months after successful PDT were classified into the complete response group. In this study, this included 20 (47.6%) out of total 42 eyes. After PDT, the all subjects showed a significant increase in BCVA, but there was no significant difference in BCVA between the two groups. This means that visual acuity does not sufficiently represent changes in visual symptoms, as reported in previous studies. OCT and OCTA have also been used to examine changes that occur after PDT in CSC, in previous studies. Demircan et al. reported that hypoperfusion of the choriocapillaris on OCTA was clearly observed on the 3rd day after PDT, but returned to normal after 1 month [14]. Demirel et al. reported that choriocapillaris flow increased and total choroidal area decreased at 1 month after PDT, but FAZ area and vessel density did not show any significant difference from those at baseline [17]. Similar to this study, Rochepeau et al. reported that SRF could be a false-positive source, so the degree of signal void was compared with the opposite eye after the SRF subsided, and that the signal void was significantly larger than the opposite eye [27]. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284899 April 21, 2023 8 / 13 PLOS ONEVisual symptom change after PDT in CSC Fig 4. Graphs showing the correlation between variables before PDT (CFT, SCP FAZ, and DCP FAZ) and baseline BCVA, 6 months post-PDT BCVA, and BCVA changes over 6 months. BCVA-related variables show a significant correlation with both SCP FAZ and DCP FAZ (all p < 0.01). CFT, central foveal thickness; SCP, superficial capillary plexus; DCP, deep capillary plexus; FAZ, foveal avascular zone; R, Spearman’s rho coefficient; BCVA, best corrected visual acuity; logMAR, logarithm of the minimum angle of resolution; PDT, photodynamic therapy. https://doi.org/10.1371/journal.pone.0284899.g004 Furthermore, Rabiolo et al. also reported that choroidal thickness decreased at 3 months after PDT, but did not show any significant change in the FAZ area [16]. Similar to previous studies, after PDT, the choroid thickness decreased significantly and the FAZ area did not show any significant difference. When compared with the choriocapillaris at the time when SRF disap- peared and the choriocapillaris at 6 months after PDT, the signal void was reduced in 31 out of 42 eyes (73.8%). Gawecki et al reported that retinal thinning, visual impairment, and choroidal flow defects were observed in resolved chronic CSC [28]. Similarly, in our study, the wider the FAZ area, and the lower the CFT, the higher was the likelihood of having limited recovery from subjec- tive visual symptoms. The larger the FAZ area before PDT, the lower was the BCVA before PDT and 6 months after PDT. In other words, as CSC becomes more chronic, subjective visual symptoms may remain and BCVA may be lower [12]. Haga et al. reported that younger age and better baseline BCVA are more likely to lead to successful treatment without recurrence [29]. In addition, Fujita et al. reported that BCVA was significantly increased after PDT, but PDT was less effective when BCVA was low before PDT [30]. Similar to previous studies, early PDT may be effective in the recovery of BCVA and subjective visual symptoms. In addition to PDT, which is a representative treatment option for CSC, eplerenone may be an effective treatment [31]. A study on overactivation of the mineralocorticoid receptor PLOS ONE \| https://doi.org/10.1371/journal.pone.0284899 April 21, 2023 9 / 13 PLOS ONEVisual symptom change after PDT in CSC inducing a choroidal pathology close to that of the pachychoroid showed that the oral drug eplerenone (mineralocorticoid receptor antagonist) decreases the choroidal blood flow and volume. The choroidal thickness varies with the axial length and examination time, while the choroidal vascular index is constant [31–34]. Toto et al. reported that oral eplerenone effec- tively lowered the choroidal vascular index in chronic CSC [31]. In this study, the subfoveal choroidal thickness and degree of CC signal void change in the choroid were used. The subfo- veal choroidal thickness showed a tendency to decrease with PDT treatment, but the difference between the two groups was insignificant. However, a limitation of the study was the lack of consideration of the measurement time and axial length. Further, the recovery of the signal void 6 months after PDT was better in the complete response group, under the limitation of a lack of measurement results. In the future, an analysis should be performed using the choroidal vascular index in addition to the retinal vessel density. Sulzbacher et al. reported that the visual prognosis of neovascular CSCs with CNV detec- tion on OCTA was poor [35]. In this study, visual symptoms and acuity were also checked, but patients with confirmed CNV were excluded from the study. In a previous long-term study, the development of CNV was higher in the incomplete response group, consistent with our study. Novel conclusions may be drawn if we observe the progression over a longer period. There are two clinically important facts suggested by these results. First, in patients with CSC, those with low CFT or large FAZ of SCP, DCP may not fully recover from their symp- toms after PDT and may have sequelae, patients should be informed of this possibility before treatment. Second, and more importantly, although waiting for natural recovery in CSC patients may be a legitimate form of treatment, it would be better not to postpone PDT until CFT decreases or FAZ increases. Although the change in visual acuity is a major consideration in determining the initiation of PDT treatment, this study suggests that visual complaints may remain even with good visual acuity. In other words, CFT and FAZ should be major consider- ations in treatment decisions. Even if the visual acuity is good, relatively early intervention before the changes in CFT and FAZ become more profound appears preferable; however, fur- ther research is needed in this regard. The limitations of this study are that the sample size was small and the follow-up period was short. Another potential issue is that the groups were defined using subjective symptoms directly reported by the subjects. The severity of symptoms was not considered as it was diffi- cult to measure because of the retrospective nature of the study. However, the analysis of chief complaints would have some degree of reliability. A prospective study on symptom severity should be conducted in the future. Finally, of a total of 69 eyes, 42 (60.8%) were included in the analysis. Since 12 of the 27 excluded eyes had been lost to follow-up, selection bias may have occurred. This was a limitation of the retrospective study design. Nevertheless, this is the first study to investigate the differences in clinical features, including OCT and OCTA, and their relationship with changes in subjective visual symptoms after PDT. In conclusion, before PDT, the thicker the CFT and the smaller the FAZ area, the easier it is to recover from subjective visual symptoms. Before PDT, the wider the FAZ area, the lower the BCVA before and 6 months after PDT. These facts can be meaningful in explaining the degree of recovery from subjective visual symptoms after PDT for patients before PDT. Supporting information S1 Dataset. (XLSX) PLOS ONE \| https://doi.org/10.1371/journal.pone.0284899 April 21, 2023 10 / 13 PLOS ONEVisual symptom change after PDT in CSC Author Contributions Conceptualization: Se Woong Kang. Data curation: Geun Woo Lee, Kyung Jun Choi. Writing – original draft: Geun Woo Lee, Se Woong Kang. Writing – review & editing: Geun Woo Lee, Yun Young Kim, Kyung Jun Choi, Se Woong Kang. References 1. Gass JD. Pathogenesis of disciform detachment of the neuroepithelium. Am J Ophthalmol. 1967; 63: Suppl:1–139. PMID: 6019308. 2. Marmor MF. New hypotheses on the pathogenesis and treatment of serous retinal detachment. Graefes Arch Clin Exp Ophthalmol. 1988; 226:548–52. https://doi.org/10.1007/BF02169203 PMID: 3209082. 3. Spaide RF, Hall L, Haas A, Campeas L, Yannuzzi LA, Fisher YL, et al. Indocyanine green videoangio- graphy of older patients with central serous chorioretinopathy. Retina. 1996; 16:203–13. https://doi.org/ 10.1097/00006982-199616030-00004 PMID: 8789858. 4. 5. Ferrara D, Mohler KJ, Waheed N, Adhi M, Liu JJ, Grulkowski I, et al. En face enhanced-depth swept- source optical coherence tomography features of chronic central serous chorioretinopathy. Ophthalmol- ogy. 2014; 121:719–26. https://doi.org/10.1016/j.ophtha.2013.10.014 PMID: 24289918. Frisen L, Frisen M. Micropsia and visual acuity in macular edema. A study of the neuro-retinal basis of visual acuity. Albrecht Von Graefes Arch Klin Exp Ophthalmol. 1979; 210:69–77. https://doi.org/10. 1007/BF00409993 PMID: 312038. 6. Daruich A, Matet A, Dirani A, Bousquet E, Zhao M, Farman N, et al. Central serous chorioretinopathy: Recent findings and new physiopathology hypothesis. Prog Retin Eye Res. 2015; 48:82–118. https:// doi.org/10.1016/j.preteyeres.2015.05.003 PMID: 26026923. 7. Nicholson B, Noble J, Forooghian F, Meyerle C. Central serous chorioretinopathy: update on patho- physiology and treatment. Surv Ophthalmol. 2013; 58:103–26. https://doi.org/10.1016/j.survophthal. 2012.07.004 PMID: 23410821. 8. Bae SH, Heo JW, Kim C, Kim TW, Lee JY, Song SJ, et al. A randomized pilot study of low-fluence pho- todynamic therapy versus intravitreal ranibizumab for chronic central serous chorioretinopathy. Am J Ophthalmol. 2011; 152:784–92 e2. https://doi.org/10.1016/j.ajo.2011.04.008 PMID: 21742303. 9. 10. Fujita K, Shinoda K, Matsumoto CS, Imamura Y, Tanaka E, Mizutani Y, et al. Microperimetric evaluation of chronic central serous chorioretinopathy after half-dose photodynamic therapy. Clin Ophthalmol. 2012; 6:1681–7. https://doi.org/10.2147/OPTH.S35338 PMID: 23097612. van Dijk EHC, Fauser S, Breukink MB, Blanco-Garavito R, Groenewoud JMM, Keunen JEE, et al. Half- Dose Photodynamic Therapy versus High-Density Subthreshold Micropulse Laser Treatment in Patients with Chronic Central Serous Chorioretinopathy: The PLACE Trial. Ophthalmology. 2018; 125:1547–55. https://doi.org/10.1016/j.ophtha.2018.04.021 PMID: 29776672. 11. Vasconcelos H, Marques I, Santos AR, Melo P, Pires I, Figueira J, et al. Long-term chorioretinal changes after photodynamic therapy for chronic central serous chorioretinopathy. Graefes Arch Clin Exp Ophthalmol. 2013; 251:1697–705. https://doi.org/10.1007/s00417-013-2270-2 PMID: 23389551. 12. 13. van Rijssen TJ, van Dijk EHC, Yzer S, Ohno-Matsui K, Keunen JEE, Schlingemann RO, et al. Central serous chorioretinopathy: Towards an evidence-based treatment guideline. Prog Retin Eye Res. 2019; 73:100770. https://doi.org/10.1016/j.preteyeres.2019.07.003 PMID: 31319157. Jia Y, Tan O, Tokayer J, Potsaid B, Wang Y, Liu JJ, et al. Split-spectrum amplitude-decorrelation angi- ography with optical coherence tomography. Opt Express. 2012; 20:4710–25. https://doi.org/10.1364/ OE.20.004710 PMID: 22418228. 14. Demircan A, Yesilkaya C, Alkin Z. Early choriocapillaris changes after half-fluence photodynamic ther- apy in chronic central serous chorioretinopathy evaluated by optical coherence tomography angiogra- phy: Preliminary results. Photodiagnosis and Photodynamic Therapy. 2018; 21:375–8. WOS:000430644200059. https://doi.org/10.1016/j.pdpdt.2018.01.015 PMID: 29409857 15. Ma DJ, Park UC, Kim ET, Yu HG. Choroidal vascularity changes in idiopathic central serous chorioreti- nopathy after half-fluence photodynamic therapy. PLoS One. 2018; 13:e0202930. https://doi.org/10. 1371/journal.pone.0202930 PMID: 30148858. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284899 April 21, 2023 11 / 13 PLOS ONEVisual symptom change after PDT in CSC 16. Rabiolo A, Zucchiatti I, Marchese A, Baldin G, Sacconi R, Montorio D, et al. Multimodal retinal imaging in central serous chorioretinopathy treated with oral eplerenone or photodynamic therapy. Eye (Lond). 2018; 32:55–66. https://doi.org/10.1038/eye.2017.290 PMID: 29265111. 17. Demirel S, Ozcan G, Yanik O, Batioglu F, Ozmert E. Vascular and structural alterations of the choroid evaluated by optical coherence tomography angiography and optical coherence tomography after half- fluence photodynamic therapy in chronic central serous chorioretinopathy. Graefes Arch Clin Exp Ophthalmol. 2019; 257:905–12. https://doi.org/10.1007/s00417-018-04226-6 PMID: 30617579. 18. Karakus SH, Basarir B, Pinarci EY, Kirandi EU, Demirok A. Long-term results of half-dose photody- namic therapy for chronic central serous chorioretinopathy with contrast sensitivity changes. Eye (Lond). 2013; 27:612–20. https://doi.org/10.1038/eye.2013.24 PMID: 23519277. 19. Battaglia Parodi M, Cicinelli MV, Rabiolo A, Pierro L, Bolognesi G, Bandello F. Vascular abnormalities in patients with Stargardt disease assessed with optical coherence tomography angiography. Br J Ophthalmol. 2017; 101:780–5. https://doi.org/10.1136/bjophthalmol-2016-308869 PMID: 27628426. 20. Chan WM, Lam DS, Lai TY, Tam BS, Liu DT, Chan CK. Choroidal vascular remodelling in central serous chorioretinopathy after indocyanine green guided photodynamic therapy with verteporfin: a novel treatment at the primary disease level. Br J Ophthalmol. 2003; 87:1453–8. https://doi.org/10. 1136/bjo.87.12.1453 PMID: 14660450. 21. Liu HY, Yang CH, Yang CM, Ho TC, Lin CP, Hsieh YT. Half-dose Versus Half-time Photodynamic Ther- apy for Central Serous Chorioretinopathy. Am J Ophthalmol. 2016; 167:57–64. https://doi.org/10.1016/ j.ajo.2016.04.001 PMID: 27084002. 22. Hwang S, Kang SW, Kim SJ, Jang JW, Kim KT. PHOTODYNAMIC THERAPY FOR SYMPTOMATIC SUBFOVEAL RETINAL PIGMENT EPITHELIAL DETACHMENT IN CENTRAL SEROUS CHORIORE- TINOPATHY: Outcomes and Prognostic Factors. Retina. 2019; 39:1117–24. https://doi.org/10.1097/ IAE.0000000000002108 PMID: 29517581. 23. Cheng CK, Chang CK, Peng CH. Comparison of Photodynamic Therapy Using Half-Dose of Verteporfin or Half-Fluence of Laser Light for the Treatment of Chronic Central Serous Chorioretinopathy. Retina. 2017; 37:325–33. https://doi.org/10.1097/IAE.0000000000001138 PMID: 27429374. 24. Fukuda S, Okamoto F, Yuasa M, Kunikata T, Okamoto Y, Hiraoka T, et al. Vision-related quality of life and visual function in patients undergoing vitrectomy, gas tamponade and cataract surgery for macular hole. Br J Ophthalmol. 2009; 93:1595–9. https://doi.org/10.1136/bjo.2008.155440 PMID: 19570766. 25. Okamoto F, Okamoto Y, Hiraoka T, Oshika T. Effect of vitrectomy for epiretinal membrane on visual function and vision-related quality of life. Am J Ophthalmol. 2009; 147:869–74, 74 e1. https://doi.org/ 10.1016/j.ajo.2008.11.018 PMID: 19200531. 26. Lina G, Xuemin Q, Qinmei W, Lijun S. Vision-related quality of life, metamorphopsia, and stereopsis after successful surgery for rhegmatogenous retinal detachment. Eye (Lond). 2016; 30:40–5. https:// doi.org/10.1038/eye.2015.171 PMID: 26427988. 27. Rochepeau C, Kodjikian L, Garcia MA, Coulon C, Burillon C, Denis P, et al. Optical Coherence Tomog- raphy Angiography Quantitative Assessment of Choriocapillaris Blood Flow in Central Serous Choriore- tinopathy. Am J Ophthalmol. 2018; 194:26–34. https://doi.org/10.1016/j.ajo.2018.07.004 PMID: 30053475. 28. Gawęcki M, Jaszczuk-Maciejewska A, Jurska-Jaśko A, Kneba M, Grzybowski A. Impairment of visual acuity and retinal morphology following resolved chronic central serous chorioretinopathy. BMC Oph- thalmology. 2019; 19 https://doi.org/10.1186/s12886-019-1171-5 PMID: 31345183 29. Haga F, Maruko R, Sato C, Kataoka K, Ito Y, Terasaki H. Long-term prognostic factors of chronic central serous chorioretinopathy after half-dose photodynamic therapy: A 3-year follow-up study. PLoS One. 2017; 12:e0181479. https://doi.org/10.1371/journal.pone.0181479 PMID: 28742138. 30. 31. 32. Fujita K, Imamura Y, Shinoda K, Matsumoto CS, Mizutani Y, Hashizume K, et al. One-year outcomes with half-dose verteporfin photodynamic therapy for chronic central serous chorioretinopathy. Ophthal- mology. 2015; 122:555–61. https://doi.org/10.1016/j.ophtha.2014.09.034 PMID: 25444637. Toto L, Ruggeri ML, Evangelista F, Viggiano P, D’Aloisio R, De Nicola C, et al. Choroidal modifications assessed by means of choroidal vascularity index after oral eplerenone treatment in chronic central serous chorioretinopathy. Eye (Lond). 2022 https://doi.org/10.1038/s41433-022-02091-6 PMID: 35590104. Imamura Y, Fujiwara T, Margolis R, Spaide RF. Enhanced depth imaging optical coherence tomogra- phy of the choroid in central serous chorioretinopathy. Retina. 2009; 29:1469–73. https://doi.org/10. 1097/IAE.0b013e3181be0a83 PMID: 19898183. 33. Agrawal R, Chhablani J, Tan KA, Shah S, Sarvaiya C, Banker A. CHOROIDAL VASCULARITY INDEX IN CENTRAL SEROUS CHORIORETINOPATHY. Retina. 2016; 36:1646–51. https://doi.org/10.1097/ IAE.0000000000001040 PMID: 27124882. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284899 April 21, 2023 12 / 13 PLOS ONEVisual symptom change after PDT in CSC 34. Yang J, Wang E, Yuan M, Chen Y. Three-dimensional choroidal vascularity index in acute central serous chorioretinopathy using swept-source optical coherence tomography. Graefes Arch Clin Exp Ophthalmol. 2020; 258:241–7. https://doi.org/10.1007/s00417-019-04524-7 PMID: 31724090. 35. Sulzbacher F, Schu¨ tze C, Burgmu¨ ller M, Ve´ csei-Marlovits PV, Weingessel B. Clinical evaluation of neo- vascular and non-neovascular chronic central serous chorioretinopathy (CSC) diagnosed by swept source optical coherence tomography angiography (SS OCTA). Graefes Arch Clin Exp Ophthalmol. 2019; 257:1581–90. https://doi.org/10.1007/s00417-019-04297-z PMID: 31037488. PLOS ONE \| https://doi.org/10.1371/journal.pone.0284899 April 21, 2023 13 / 13 PLOS ONE
10.1371_journal.pone.0285166	RESEARCH ARTICLE IL-17 cytokines preferentially act on naïve CD4+ T cells with the IL-17AF heterodimer inducing the greatest functional changes Michael P. Crawford1,2‡, Nicholas Borcherding1‡, Nitin J. KarandikarID 1,2* 1 Department of Pathology, University of Iowa Health Care, Iowa City, Iowa, United States of America, 2 Iowa City Veterans Administration Medical Center, Iowa City, IA, United States of America ‡ MPC and NB contributed equally to this work as co-first authors. * nitin-karandikar@uiowa.edu Abstract CD4+ T-helper 17 (Th17) T cells are a key population in protective immunity during infection and in self-tolerance/autoimmunity. Through the secretion of IL-17, Th17 cells act in promo- tion of inflammation and are thus a major potential therapeutic target in autoimmune disor- ders. Recent reports have brought to light that the IL-17 family cytokines, IL-17A, IL-17F and IL-17AF, can directly act on CD4+ T-cells, both in murine and human systems, inducing functional changes in these cells. Here we show that this action is preferentially targeted toward naïve, but not memory, CD4+ T-cells. Naïve cells showed transcriptome changes as early as 48 hours post-IL-17 exposure, whereas memory cells remained unaffected as late as 7 days. These functional differences occurred despite similar IL-17 receptor expression on these subsets and were maintained in co-culture/transwell systems, with each subset maintaining its functional response to IL-17. Importantly, there were differences in down- stream transcriptional signaling by the three IL-17 cytokines, with the IL-17AF heterodimer conferring both the greatest transcriptional change and most altered functional conse- quences. Detailed transcriptome analysis provides important insights into the genes and pathways that are modulated as a result of IL-17-mediated signaling and may serve as tar- gets of future therapies. Introduction The dysfunction of CD4+ T-helper 17 (Th17) cells has been associated with a number of auto- immune conditions, such as type 1 diabetes, multiple sclerosis, and arthritis [1–7]. As a major source of proinflammatory IL-17 secretion, Th17 cells can act on stromal cells to produce inflammatory cascades and recruit neutrophils [8]. In addition, the differentiation of Th17 cells has a bifurcated relationship with the differentiation of regulatory T-cells (Tregs), a sup- pressive T-cell subset, in the periphery [9]. In the current paradigm, the signaling downstream of TGF-β in a naïve CD4+ T-cell can lead to Th17 or Treg differentiation. However, in the presence of proinflammatory cytokines, the fate is biased toward Th17 differentiation. These a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Crawford MP, Borcherding N, Karandikar NJ (2023) IL-17 cytokines preferentially act on naïve CD4+ T cells with the IL-17AF heterodimer inducing the greatest functional changes. PLoS ONE 18(4): e0285166. https://doi.org/10.1371/ journal.pone.0285166 Editor: Pierre Bobe´, Universite Paris-Saclay, FRANCE Received: February 21, 2023 Accepted: April 17, 2023 Published: April 28, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0285166 Copyright: This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication. Data Availability Statement: All data are included in this article, with the exception of raw RNAseq data, which can be accessed in the NCBI Genome PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 1 / 15 PLOS ONEExpression Omnibus public database through GEO Series accession no. GSE150805. Funding: This work was supported, in part, by grant awards from the NIH and VA to N.J.K. (R01 AI121567, I01 CX002319) and N.B. (F30 CA29655). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. IL-17-mediated signaling in CD4+ T cells observations have led to the evaluation of anti-IL-17 therapies for inflammatory and autoim- mune conditions that have had mixed success. We and others recently found that IL-17 is capable of acting directly on CD4+ T-cells in both humans and mice [10, 11]. We found that in the presence of a Th17-differentiating milieu, CD4+ T-cells become more resistant to CD8-mediated immune suppression [11], partly mediated by their production of IL-17A, IL-17F, and IL-17AF heterodimer. Moreover, these cytokines could also act directly on non-Th17 CD4+ T-cells and make them more resis- tant to suppression. This suppressive resistance could be ameliorated by interfering with IL- 1β, IL-6 or STAT3 signaling [11]. Recently published studies in an autoimmune uveitis model also suggest a complex signaling network in developing Th17 cells, where feedback of IL-17A leads to NF-κb activation and subsequent secretion of IL-24 to suppress the Th17 genetic pro- gram [10]. Interestingly, the removal of IL-17A led to compensatory increases in IL-17F and GM-CSF. However, the knockout of IL-17F did not have the same effect, suggesting both redundant and unique signaling for IL-17 family members [10]. Here we further our investigation of IL-17-mediated changes in CD4+ T-cells, finding naïve, but not memory, CD4+ T-cells as the principal cells responding to IL-17. Characterizing the genetic patterns resulting from IL-17A, IL-17F and IL-17AF actions, we found that IL- 17AF induced the greatest magnitude of suppressive resistance and size of genetic programs. We found general downregulation of cytokine and immune effector molecules and upregula- tion in downstream mediators of interferon signaling. Materials and methods Cell preparation and bead sorting Peripheral blood mononuclear cells (PBMC) from healthy subjects were isolated from de- identified leukocyte reduction system (LRS) cones containing leukocyte-rich whole blood from platelet donors at the University of Iowa, DeGowin Blood Center. PBMC isolation was performed with BD Vacutainer CPT tubes (BD, 362753) density gradient centrifugation. CD8 T-cells were positively selected from freshly prepared PBMC with Manual LS Column MACS sorting with Miltenyi Biotec MACS CD8 Bead sorting microbeads (130-045-201) according to manufacturer specifications. Untouched bulk CD4+CD25- T-cells were obtained using the CD4+ T-cell Isolation Kit (130-096-533) followed by a CD25 microbead depletion (130-092- 983). CD45RO (130-046-001) or CD45RA (130-045-901) positive selection microbeads were used to respectively obtain CD4+CD25-CD45RO+ memory T-cells and “untouched” CD4+CD25-CD45RO- naïve T-cells or CD4+CD25-CD45RA+ naïve T-cells and “untouched” CD4+CD25-CD45RA- memory T-cells. Sort purities of all populations were routinely above 93% by flow cytometric analysis (S1 Fig). Sorted CD8 T-cells and CD4+ T-cell populations were frozen in human serum and dimethyl sulfoxide-containing media on the day of sorting for future use. Th subset differentiation Memory CD4+ T-cells were subjected to Th0, Th1, Th2 and Th17 differentiation cultures, as described previously [11]. Briefly, memory cells were thawed in RPMI 1640 (Corning 10- 040-CV) with DNase at 10KU/ml (Sigma D4513-1vl) and then resuspended at 1×10e6 cells/ml in X-VIVO 15 serum-free media (Lonza, 04-418Q), followed by stimulation in various differ- entiation conditions (Media Alone/Th0, Th1, Th2, Th17). Conditions included: (1) Media Alone/Th0: no cytokines/antibodies added; (2) Th1: anti-IL-4 7μg/ml BD554481, IL-2 10ng/ ml BD554603, IL-12 10ng/ml BD554613; (3) Th2: anti-IFNγ 7μg/ml BD554698, IL-2 10ng/ml, IL-4 10ng/ml BD554605; (4) Th17: anti-IL-4 7μg/ml, anti-IFNγ 7μg/ml, TGFβ1 10ng/ml PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 2 / 15 PLOS ONEIL-17-mediated signaling in CD4+ T cells eBiosciences 14-8348-62, IL-1β 10ng/ml BD554602, IL-6 50ng/ml BD550071. Cultures were activated with 1 μg/ml each of fixed anti-CD3 (eBiosciences, 16-0037-85) and anti-CD28 (eBiosciences, 16-0289-85) and incubated for 7 days at 37˚C, followed by suppression assay cultures. n = 7 IL-17 receptor staining IL17RA APC (Miltenyi Biotec #130-104-722), REA-APC isotype control (Miltenyi Biotec #130-104-614), IL17RC PE (Miltenyi Biotec #130-109-150) and the REA-PE isotype control (Miltenyi Biotec #130-104-612) were used to stain ex vivo cells and T-helper subsets. Flow cytometric data was acquired on the Cytek Aurora flow cytometer and analyzed with FlowJo software v10. Transwell exogenous IL-17 addition suppression assays The 24 multiwell tissue culture plates (Costar Ref 3413, tissue culture treated polystyrene with 6.55 mm insert and 0.4 micrometer polycarbonate membranes) were plated with 1 μg/ml each of anti-CD3 and anti-CD28, as previously described [12]. Anti-CD3/anti-CD28 1ug/ml was fixed per manufacturer protocol to Sperotech polystyrene protein G particles (cat PGP-60-05, 0.5 w/v, 6.8 um) and were used within the well insert for stimulation. Ex vivo bead-sorted memory and naïve CD4+ T-cell populations were cultured in X-VIVO 15 media. IL-17A (eBioscience, 34-8179-82), IL-17F (R&D Systems, 1335-IL-025/CF), IL-17A+IL-17F and IL- 17AF (R&D Systems, 5194-IL-025/CF) were added at 10 ng/mL to the indicated cultures (Fig 3) and cultured for 7 days. At 7 days, the cells were removed from the transwell, washed and used as responder cells in standardized suppression assays. Flow cytometric suppression assays Either ex vivo- or 7-day-cultured CD4+ T-cells were placed in flow cytometric suppression assays, as described previously [11, 13, 14]. Briefly, responder CD4+ T-cells were stained with CFSE, followed by culture 1 μg/ml of fixed anti-CD3 (eBiosciences, 16-0037-85) and 1μg/ml of fixed anti-CD28 (eBiosciences, 16-0289-85) in the presence or absence of ex vivo sorted autolo- gous bulk CD8+ T-cells at 1:0, 1:1 and 1:0.5 CD4:CD8 cell concentrations. On day 7 of culture, cells were stained for anti-CD4 PE-Cy7 (BD, 557852), anti-CD3 AlexaFluor700 (BD, 557943), anti-CD8 Pacific Blue (Biolegend, 344718) and flow cytometrically assessed for CD4 proliferat- ing fraction (CFSE dilution). Percent proliferation and percent suppression were calculated as described previously [13]. Other antibodies utilized include: CD45RO Pacific Blue, Biolegend 304216; CD45RO PE, Biolegend 304244; CD45RA Fitc, BD 555488; CD8 PE, BD 340046; CD8 BV786, BD 563823; CD4 BV786, BD 563877; CD4 APC, BD 561841; CD4 PE, BD 555347; CD3 APC, Biolegend 300412; CD25 PE, BD 555432; CD25 APC, BD 555434; CD25 PacBlue, Biolegend 356130. Cells were analyzed on a BD LSRII or Cytek Aurora. Data were analyzed with FlowJo software v10. RNA sequencing/transcriptome analysis Ex vivo-purified bulk CD4+CD25- T-cells and memory CD4+CD25-CD45RO+ T-cells were activated in vitro for 48 hours or 7 days in the presence of media alone (controls), or 10 ng/ml of IL-17A (eBioscience, 34-8179-82), IL-17F (R&DSystems, 1335-IL-025/CF) or IL-17AF (R&D Systems, 5194-IL-025/CF). Cells were incubated with the respective cytokines 90 min- utes prior to activation and culture. Supernatants were separated and frozen for later analysis. Cell pellet samples were snap frozen and were submitted to the University of Chicago PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 3 / 15 PLOS ONEIL-17-mediated signaling in CD4+ T cells Genomics facility for RNA extraction, quality assessment and sequencing. Single-end 50 bp sequencing was performed on the Illumina HiSeq 2000 (Illumina, San Diego, CA). Pseudoa- lignment was performed using kallisto with the GRCh38 human genome build [15]. Pseu- doalignments were processed using sleuth (v0.30) R package using gene-level quantifications [15]. Differential gene expression analysis was performed in the sleuth R package with the Wald test. Differential genes were defined as log2-fold change >1 or <-1 and false discovery rate < 0.05. The significant genes were used for Ingenuity Pathway Analysis (Qiagen) using the same cut points for significance as inputs. Overlap coefficients were calculated using the size of the intersection of the conditions divided by the total size of the smallest condition. ELISAs ELISAs were performed per manufacturer protocol on culture supernatant aliquots of original RNAsequenced cells. Human ELISA Kits for IL6 (Invitrogen, BMS213-2), IL-1β (Invitrogen, KAC1211), IL-10 (Invitrogen, BMS215/2), CXCL1 (Invitrogen BMS2122), CCL1/I-309 (Invi- trogen, EHCCL1), IL-4 (Invitrogen, BMS225-2), IL-5 (R&D D5000B). ELISA data were acquired on a BioTek Synergy H1 Hybrid Reader. Gen5 v2.09 was used for software analysis. Statistics For non-RNAseq data, Graphpad Prism v7.03 was used for statistical graphics using paired t tests for significance. Study approval All experiments were performed on PBMC obtained from de-identified leukocyte reduction system (LRS) cones from healthy platelet donors at the University of Iowa DeGowin Blood Center, as approved by the University of Iowa Institutional Review Board. Results IL-17 cytokines affect naïve and bulk CD4+ T-cells but show no effect on memory T-cells For these studies, we used Miltenyi magnetic bead cell sorting to yield total (bulk) CD4+ CD25- T-cells from healthy donors and further isolated CD45RO+ memory T-cells (Fig 1A), with a sort purity/enrichment of 93.62% +/- 1.13. The bulk ex vivo CD4+25- T cell populations were made up of ~57% memory and ~42% naïve T-cells (S2 Fig). Cells were first incubated with the indicated cytokines and then activated with anti-CD3 and anti-CD28. We expected to see a large change in memory CD4+ T-cells due to their inherently lower threshold for activa- tion. However, we found significantly greater differential gene expression in the bulk CD4+ T- cells at 2 days of activation (Fig 1B). Therefore, we activated memory CD4+ T-cells for 7 days and yet did not observe virtually any changes in the number of differentially expressed genes (Fig 1B). This indicated that the naïve T-cell component of the bulk population may be respon- sible for IL-17-induced resistance to suppression. To test this functionally, we isolated CD4+CD25- bulk, CD4+CD25-CD45RA+RO- naïve, and CD4+CD25-CD45RA-RO+ memory T-cells. To ascertain that the functional effects were not an artifact of magnetic microbeads stuck to the positively sorted cells, we performed the sorting using CD45RA or CD45RO microbeads, resulting in positive or negative selection of each subset, leaving the other subset “untouched.” These cells were activated in media or IL-17AF for 7 days and then assessed for their suppressive resistance, as described previously [11]. Regardless of the beads used PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 4 / 15 PLOS ONEIL-17-mediated signaling in CD4+ T cells Fig 1. IL-17 cytokines affect naïve and bulk CD4+ T-cells but show no effect on memory T-cells. A. Schematic summary of the ex vivo treatment of bead-sorted bulk or memory CD4+ CD25- T-cells activated in media alone (control/baseline), IL-17A, IL-17F, or IL-17AF heterodimer. After activation, RNA was isolated and underwent mRNA sequencing. B. The number of significant genes upregulated and downregulated with IL-17A (yellow), IL-17F (orange) and IL-17AF (blue) within the indicated populations. Significance was defined as q- value < 0.05 and log2-fold change > 1 or < -1. C. Negatively selected CD4+25- T-cells were further sorted using either CD45RO or CD45RA magnetic beads resulting in bulk, naïve and memory T-cell populations, which were αCD3/αCD28-activated for 7 days in media alone (control) or in the presence of IL-17AF and then used as responder cells in flow cytometric suppression assays with autologous CD8+ T-cells as suppressors. p<0.005, * p<0.001, n.s. = not significant by Paired Student T-test. https://doi.org/10.1371/journal.pone.0285166.g001 (CD45RO or CD45RA) or the positive/negative isolation (bead touched/untouched) status, we found that bulk and naïve CD4+ T-cells attained significant suppressive resistance following IL-17 exposure (Fig 1C). The memory cells were unaffected by this exposure, functionally cor- roborating the transcriptome data. Thus, IL-17 cytokines consistently induced functional changes in the naïve CD4+ T-cell subset but did not affect memory CD4+ T-cells. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 5 / 15 PLOS ONEIL-17-mediated signaling in CD4+ T cells Memory and naïve CD4+ T-cells have similar levels of IL-17 receptor expression, are similarly suppressible and have similar plasticity and resistance patterns when exposed to Th0/Th1/Th17 differentiation conditions Next, we investigated the possibility that the IL-17 receptors were differentially expressed on memory versus naïve CD4+ T-cells. Fig 2A shows that IL17RA and IL17RC receptor expres- sion patterns were not significantly different between these populations, with >80% of cells expressing IL17RA and <20% of cells expressing IL17RC. Similarly, previously activated and differentiated Th0 and Th17 cells also showed similar proportions of IL17RA- and IL17RC- expressing cells (Fig 2B and 2C). Moreover, CD8 T-cell-mediated suppression of ex vivo mem- ory CD4+ T-cells was similar to ex vivo naïve CD4+ T-cells (Fig 2D). Notably, we did not observe any differences in % proliferation between ex vivo naïve vs memory in the 1:0, 1:0.5 or 1:1 culture conditions (Fig 2D; p = n.s.). We have shown previously that naïve CD4+ T-cells cultured under different cytokine milieus attain differential resistance to suppression [11], with greatest resistance by Th17 cells. Interestingly, in these types of differentiation conditions, memory cells also showed functional plasticity in that their suppressive resistance was modulated in a manner similar to that of naïve CD4+ T-cells, including suppressive resistance following exposure to Th17 conditions Fig 2. IL-17RA/RC receptor expression does not account for the lack of memory cell response to IL-17. A. IL17RA and IL17RC expression on ex vivo microbead-sorted CD4+ Naïve (CD45RA+ and CD45RO-) and CD4+ Memory (CD45RO+ and CD45RA-) T-cells (Mean+/-SEM; n = 9). B-C. IL17RA and IL17RC expression after 7-day differentiation cultures of Naïve CD4+ T cells into Th0 and Th17 conditions (Mean+/-SEM; n = 16). D. Ex vivo-purified memory (CD4+CD25-CD45RO+) and naïve (CD4+CD25-CD45RO-) T-cells from healthy donors were stained with CFSE, followed by a 7-day culture with autologous irradiated APCs, fixed αCD3 antibody, and +/- increasing numbers of CD8+ T-cells (at the indicated ratios). Column bars depict %proliferation (+/-SEM) of indicated CD4+ T-cells at the indicated CD4:CD8 ratios, showing similar proliferation at 1:0 and similar suppression at 1:0.5 and 1:1. p<0.005, * p<0.001, ***p<0.0001 by paired student t test; n = 6. E-F. Ex vivo-purified memory CD4+CD25-CD45RO+ T-cells from healthy donors were polarized under indicated differentiation conditions for 7 days and then placed in a 7-day CD8+ T-cell mediated suppression assay. (E) Column bars depict %suppression (+/-SEM) of indicated Th-subsets by CD8+ T-cells at the 1:0.5 CD4:CD8 ratio, showing greater sensitivity of “Th1 milieu”-exposed cells and greater resistance by “Th17 milieu”-exposed cells. (F) Paired %suppression data from the Th0 and Th17 suppression cultures. p<0.05, ** p<0.005. https://doi.org/10.1371/journal.pone.0285166.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 6 / 15 PLOS ONEIL-17-mediated signaling in CD4+ T cells (Fig 2E and 2F). Thus, other cytokines could functionally modulate the suppressive resistance of memory CD4+ T-cells. However, unlike these other differentiation cytokines, IL-17 seems to preferentially act on naïve T-cells while uniquely ignoring ex vivo memory cells. Cytokines from IL-17-responsive naïve T-cells do not affect the IL-17-non- responsiveness of memory T-cells Our initial observations (Fig 1) were made on bulk vs. memory CD4+ T-cells. We wondered whether the interactions of naïve and memory cells from bulk cultures might convert memory T-cells into becoming IL-17-responsive. For this, we used transwell assay systems where mem- ory cells would get exposed to cytokines secreted by naïve T-cells in the presence of IL-17 (Fig 3). Thus, ex vivo CD4+25- T-cells were sorted into memory and naïve subsets and acti- vated with anti-CD3 and anti-CD28 for 7 days in the presence or absence of IL-17A, IL-17F, IL-17AF or IL-17A+IL-17F either across from each other through a transwell (Fig 3A) or memory alone (Fig 3B) and naïve alone (Fig 3C). The subsets were then resuspended and placed in a CD8 T-cell mediated suppression assay. Naïve cells retained the IL-17-induced resistance to suppression whether cultured alone or in the presence of memory cells. In con- trast, memory T-cells remained unresponsive to IL-17 in either condition, suggesting that they were not affected by secreted molecules from neighboring naïve cells. There were no signifi- cant differences between the suppression of memory cells that were cultured with naïve cells (through the transwell) versus those cultured alone (p = n.s.). Naïve cells cultured in transwells versus alone also showed largely similar degrees of suppressive resistance in most of the condi- tions, except for the IL-17AF condition, which showed significantly greater resistance when naïve cells were cultured alone, likely suggesting undiluted effect of the heterodimer. Overall, these results emphasize the distinct responsiveness of memory vs. naïve cells to IL-17 cytokines. Overlapping as well as unique transcriptome changes in IL-17-treated CD4+ T-cells In order to better characterize the IL-17-mediated actions on CD4+ T-cells, we next examined the underlying genetic patterns from the three IL-17 conditions. We found that IL-17A and IL-17F treatment had a notable overlap in upregulated (overlap coefficient = 0.435) and down- regulated (overlap coefficient = 0.536) genes compared with the IL-17AF heterodimer (Fig 4). Within upregulated genes shared in at least two conditions, we observed increases in inter- feron response factors (IRFs) and interferon-induced factors (IFITs) (Fig 4A and 4B). In con- trast, commonly downregulated genes across the conditions included chemokines (CCL2, CCL3, and CCL8), cytokines (IL1A, IL1B, IL6, IL17F, and IFNG), and receptors (CCR1 and TLR8) associated with T effector functions (Fig 4C and 4D). Of note, while the specific cyto- kine genes, IL1B and IL6, were downregulated in these conditions at the 48-hour snapshot, the alteration in these cytokines was the common component of the majority of differential path- ways identified [11]. Moreover, interfering with the specific cytokines at a later timepoint dur- ing suppression assays led to the reversal of the suppressive resistance [11]. This shows that each of the potentially targetable genes or pathways would need to be dissected and studied longitudinally for specific functional effects. Ranking the significant genes by log2-fold change, we found no clear difference in the dis- tribution of the unique genes by incubation condition (Fig 4E). In addition, examining the top unique upregulated conditions, we noticed a general absence of immune-related genes for each condition (Fig 4F), with the exception of IL21 in IL-17A and IL36A in IL-17AF. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 7 / 15 PLOS ONEIL-17-mediated signaling in CD4+ T cells Fig 3. Memory T-cell non-responsiveness to IL-17 is not affected by local auto/paracrine action of naïve T-cell cytokines. Transwell cultures were set up in 24-well plates with transwell membrane inserts separating autologous presorted Memory and Naïve CD4+ T cells together (A) or individually (B and C), with indicated cytokine conditions (media alone, IL-17A, IL-17F, IL-17AF and IL-17A+IL-17F, panels D-H, respectively) stimulated with anti-CD3/anti-CD28-coated beads in culture for 7 days. On day 7, the cells were harvested, washed, stained with CFSE and subjected to CD8-mediated T-cell suppression assays with fixed anti-CD3 and anti-CD28 stimulation. Mean suppression +/- SEM for the 5 conditions is shown; n = 5 each; p<0.05, p<0.01, p<0.001, *p<0.0001. https://doi.org/10.1371/journal.pone.0285166.g003 Using Ingenuity Pathway Analysis, we next examined the differential pathways for each condition compared to the baseline Th0. Across each condition, we noted a general trend towards increased enrichment of pathways and a high degree of overlap between IL-17A and IL-17F with an overlap coefficient of 0.60 (Fig 5A). Within common differentially altered path- ways, we found increases in metabolic pathways and pathways downstream of cytokine signal- ing (Fig 5B), despite a general reduction in cytokine expression. Broad trends in common PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 8 / 15 PLOS ONEIL-17-mediated signaling in CD4+ T cells Fig 4. Overlapping and unique expression patterns in IL-17-treated conditions. Overlapping and unique upregulated genes in bulk CD4+CD25- T-cells in IL-17A, IL-17F, and IL-17AF conditions compared to media-only cells. B. Heatmap of upregulated genes in at least two of three conditions; clustering based on Euclidean distance with immune-related genes labeled. C. Overlapping and unique downregulated genes in bulk CD4+CD25- T-cells in IL-17A, IL-17F, and IL-17AF conditions compared to media-only cells. D. Heatmap of downregulated genes in at least two of three conditions; clustering based on Euclidean distance with immune-related genes labeled. E. Dispersion of differentially-regulated genes by condition, grey genes are common between at least two conditions and colored genes are unique to the indicated condition. F. Top 10 unique upregulated genes in each condition ordered by log2-fold change. https://doi.org/10.1371/journal.pone.0285166.g004 pathways with decreased enrichment were not observed, but IL-17F in Allergic Inflammatory Airway Disease, Toll-Like Receptor Signaling, and PPAR Signaling were seen (Fig 5B). Unique pathway enrichment by condition showed increased enrichment of acyl-CoA hydrolase and Cytotoxic T lymphocyte-mediated Apoptosis of Target Cells in IL-17A (Fig 5C). In IL-17F, top pathways enrichment, including PI3K signaling, Phospholipase C Signaling, and calcium- induced T lymphocyte Apoptosis (Fig 5C). The IL-17AF-treated cells had high levels of enrich- ment in EIF2 Signaling, a translation initiation factor, CREB Signaling and the Pyridoxal 5’phosphate Salvage pathway (Fig 5C), among a number of other pathways. Detailed analysis of significant IL-17AF-mediated changes in cytokines and chemokines With IL-17AF leading to the greatest changes in gene expression, we wanted to further charac- terize this condition. We compared IL-17AF to the no cytokine baseline. We found a general decrease in chemokines (CXCL1, CXCL3, CXCL5, and CXCL8), activation markers (LAG3 and TNFRSF8), effector molecules (PRF1), transcriptional regulators (RORC) (Fig 6). Taking a closer look at cytokines and chemokines (Fig 6B), we found an increase in a select number subsequent to IL-17AF treatment. These include interleukins (IL4, IL5, and IL16) and tumor PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 9 / 15 PLOS ONEIL-17-mediated signaling in CD4+ T cells Fig 5. Pathway induction following IL-17 exposure. A. Overlapping and unique upregulated canonical pathway enrichment in bulk CD4+CD25- T-cells in IL-17A, IL-17F, and IL-17AF conditions compared to media-only cells. B. Bar chart of significantly (p < 0.05) altered in at least two of three conditions. C. Z-score of significantly enriched canonical pathways with non-unique pathways colored in grey. Top increased unique pathways labeled. https://doi.org/10.1371/journal.pone.0285166.g005 Fig 6. Detailed analysis of significant IL-17AF mediated changes in cytokines and chemokines. Volcano plot of differential genes identified comparing IL-17AF to media-only cells. Highlighted cells have q-value < 0.05 and log2-fold change > 1 or < -1. Selected genes labeled were in the top 20 genes by q-value. B. Filtered cytokine and chemokine genes with q-value < 0.05. https://doi.org/10.1371/journal.pone.0285166.g006 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 10 / 15 PLOS ONEIL-17-mediated signaling in CD4+ T cells necrosis factor family of ligands (TNFSF4, TNFSF9, TNFSF11, and TNFSF13B), while the base- line condition had the highest levels of the IL1B, IL6 and IL10. To validate the RNAseq data, we selected some of the most up- and down-regulated cyto- kines/chemokines that were measurable by commercially available standardized ELISA assays. We performed these assays on supernatants from all four conditions (media, IL-17A, IL-17F and IL-17AF) on bulk cells at 48 hours, memory cells at 48 hours and memory cells at 7 days of culture. The production of IL-6, IL-1β, IL-10 and CXCL1 was indeed robust in the media- only condition for bulk CD4+ T-cells and significantly downregulated in the presence of IL-17, particularly in the IL-17AF condition (Fig 7). Conversely, IL-4, IL-5 and CCL1 were absent in the media condition and induced by IL-17AF, matching with the RNAseq analysis. Addition- ally, GM-CSF, IFNγ and sIL2r were measured as controls and not found to be different in these conditions (NCBI Genome Expression Omnibus public database Series accession no. GSE150805). Importantly, in contrast to bulk cells, the sorted memory cells did not show dif- ferences between media vs. IL-17 conditions at either 48h or 7d, corroborating the transcrip- tome data (Fig 1). Interestingly, memory cells did not produce appreciable quantities of IL-6, IL-1β or CXCL1 at baseline. Overall, these data validate the RNAseq analysis, demonstrating a strong CD4-intrinsic effect of IL-17, which is selective for naïve T-cells. Discussion Our studies validate the recent observation that IL-17 cytokines can directly act on CD4+ T- cells and further demonstrate the unexpected finding that they act preferentially on naïve CD4+ T-cells while inducing no changes in memory CD4+ T-cells. One theoretical explanation of this differential effect on naïve versus memory CD4+ T-cells might be differences in IL-17 receptor expression. However, when we assessed two known receptors (IL-17RA and IL- 17RC), we did not find significant differences in the ex vivo expression of these receptors between naïve vs. memory cells, either through flow cytometric staining (Fig 2) or by quantifi- cation of message within the RNAseq data (NCBI Genome Expression Omnibus public data- base Series accession no. GSE150805). Similarly, we also did not see differences in previously activated/differentiated cells. While the proportion of cells expressing these receptors remained similar, it may still be possible that subtle differences in expression levels during the activation process may explain these biological effects, or there may be an unknown receptor (s)/co-receptor(s) that is required for the action of these cytokines, or some critical down- stream molecule that may be different between naïve and memory cells. Future studies will be important in dissecting this biology. In the same vein, there are interesting biologic implications of this preferential action on naïve CD4+ T cells. It is tempting to speculate that IL-17 cytokines may have an important role in microenvironments where T-cell priming is ongoing. For example, in the host’s response to infection, the presence of IL-17 in these milieus (presumably coming from either pre-differen- tiated T-cells or other cell types responding to the same antigenic stimuli) would protect naïve T-cells from suppression and allow them to get primed in response to the ongoing infection. One can envision an opposite and detrimental scenario where autoantigen-specific T-cells are afforded similar immune resistance in the context of ongoing IL-17 production. These hypoth- eses can be addressed in future studies using appropriate modeling. Amongst the cytokines tested, the IL-17AF heterodimer has the greatest effect. The combi- nation of IL-17A and IL-17F can mimic the strong functional changes induced by IL-17AF [11]. However, it remains to be seen whether that combination also replicates the genetic pro- gram induced by the heterodimer. Several lines of inquiry further support the observation of partially overlapping functions. Mutations in IL-17RA and IL-17F have been linked to chronic PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 11 / 15 PLOS ONEIL-17-mediated signaling in CD4+ T cells PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 12 / 15 PLOS ONEIL-17-mediated signaling in CD4+ T cells Fig 7. ELISA measurements of selected cytokines and chemokines validate RNAseq analysis with significant differences between CD4+CD25- bulk T-cells at 48 hours but not between CD4+CD25-45RO+ memory T-cells at 48 hours or 7 days. ELISAs were performed on supernatants frozen at -80 from cellular cultures of 48 hour bulk CD4+25- T-cell, 48-hour and 7-day memory CD4+25-45RO+ T-cells that were sent for RNA sequencing. N = 3. Data are +/- SEM. Statistical analysis is by paired Student t test p<0.05, ** p<0.005. https://doi.org/10.1371/journal.pone.0285166.g007 mucocutaneous candidiasis disease and predisposition to Staphylococcus aureus infection [16]. Interestingly, the cases identified showed a difference with inheritance patterns, with IL-17RA mutations being autosomal recessive and IL-17F as autosomal dominant [16]. Differences in expression patterns have been noted in disease processes as well. For example, psoriatic lesions appear to have upregulation of both IL-17A and IL-17F [17], while upregulation of IL-17A, but not IL-17F have been noted in multiple sclerosis brain biopsy [6]. Conversely, IL-17F pro- tein and mRNA seem more closely linked to the severity of asthma than IL-17A [18]. The com- mon and divergent signaling produced by IL-17 ligands is unclear, especially with new observations of functions on CD4+ T-cells themselves. Th17 cells and their cytokines are of immense interest in various clinical settings. For exam- ple, secukinumab, an agent targeting IL-17A, is an FDA-approved therapy for psoriasis, psori- atic arthritis, and ankylosing spondylitis [19–21]. However, this agent has not shown promising results in other autoimmune disease settings [22–24]. These discrepancies may be partially explained by differential functions of IL-17A vs IL-17F vs IL-17AF. Our data suggest that agents that interfere with the IL-17AF heterodimer or a combination of IL-17A and IL- 17F may hold greater promise in some of these disease settings. Alternatively, a common downstream effector molecule may prove to be a better target for intervention. Thus, our stud- ies provide a roadmap related to IL-17-mediated consequences on CD4+ T-cells that can be used to formulate immunotherapeutic intervention strategies and to understand the effects of anti-IL-17 therapy in specific diseases. In summary, our findings provide novel insights into IL-17 function that may have implica- tions in understanding the fundamental biology of these cytokines and inform the design of future interventional strategies. Supporting information S1 Fig. Magnetic bead separation sort purity in both positively and negatively sorted frac- tions. Ex vivo CD4+25- bulk cells were magnetically sorted with either CD45RO+ or CD45RA+ Miltenyi positive selection beads. Mean purity of the sorts is 93.62% +/- 1.13; N = 5. (TIFF) S2 Fig. Distribution of memory and naïve components of ex vivo CD4+CD25- T-cell sorts. The bulk ex vivo CD4+25- T cell populations were made up of ~57–58% memory and ~42– 43% naïve T-cells; N = 5. (TIFF) Acknowledgments The authors would like to thank Drs. Sushmita Sinha, Pranav Renavikar, Alexander Boyden, and Ashutosh Mangalam for helpful discussions and advice. Author Contributions Conceptualization: Michael P. Crawford, Nicholas Borcherding, Nitin J. Karandikar. Data curation: Nicholas Borcherding. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 13 / 15 PLOS ONEIL-17-mediated signaling in CD4+ T cells Formal analysis: Michael P. Crawford. Funding acquisition: Nitin J. Karandikar. Investigation: Nicholas Borcherding. Project administration: Nitin J. Karandikar. Supervision: Nitin J. Karandikar. Writing – original draft: Michael P. Crawford, Nicholas Borcherding. Writing – review & editing: Michael P. Crawford, Nicholas Borcherding, Nitin J. Karandikar. References 1. Noack M, Miossec P. Th17 and regulatory T cell balance in autoimmune and inflammatory diseases. Autoimmun Rev. 2014; 13: 668–677. https://doi.org/10.1016/j.autrev.2013.12.004 PMID: 24418308 2. 3. Tesmer LA, Lundy SK, Sarkar S, Fox DA. Th17 cells in human disease. Immunol Rev. 2008; 223: 87– 113. https://doi.org/10.1111/j.1600-065X.2008.00628.x PMID: 18613831 Luger D, Silver PB, Tang J, Cua D, Chen Z, Iwakura Y, et al. Either a Th17 or a Th1 effector response can drive autoimmunity: Conditions of disease induction affect dominant effector category. Journal of Experimental Medicine. 2008; 205: 799–810. https://doi.org/10.1084/jem.20071258 PMID: 18391061 4. Patel DD, Kuchroo VK. Th17 Cell Pathway in Human Immunity: Lessons from Genetics and Therapeu- tic Interventions. Immunity. 2015; 43: 1040–1051. https://doi.org/10.1016/j.immuni.2015.12.003 PMID: 26682981 5. Ye P, Rodriguez FH, Kanaly S, Stocking KL, Schurr J, Schwarzenberger P, et al. Requirement of inter- leukin 17 receptor signaling for lung CXC chemokine and granulocyte colony-stimulating factor expres- sion, neutrophil recruitment, and host defense. Journal of Experimental Medicine. 2001; 194: 519–528. https://doi.org/10.1084/jem.194.4.519 PMID: 11514607 6. Lock C, Hermans G, Pedotti R, Brendolan A, Schadt E, Garren H, et al. Gene-microarray analysis of multiple sclerosis lesions yields new targets validated in autoimmune encephalomyelitis. Nat Med. 2002; 8: 500–508. https://doi.org/10.1038/nm0502-500 PMID: 11984595 7. Cua DJ, Sherlock J, Chen Y, Murphy CA, Joyce B, Seymour B, et al. Interleukin-23 rather than interleu- kin-12 is the critical cytokine for autoimmune inflammation of the brain. Nature. 2003; 421: 744–748. https://doi.org/10.1038/nature01355 PMID: 12610626 8. Miossec P, Kolls JK. Targeting IL-17 and T H 17 cells in chronic inflammation. Nat Rev Drug Discov. 2012; 11: 763–776. https://doi.org/10.1038/nrd3794 PMID: 23023676 9. Korn T, Bettelli E, Oukka M, Kuchroo VK. IL-17 and Th17 cells. Annu Rev Immunol. 2009; 27: 485–517. https://doi.org/10.1146/annurev.immunol.021908.132710 PMID: 19132915 10. Chong WP, Mattapallil M, Raychaudhuri K, Bing SJ, Wu S, Zhong Y, et al. The Cytokine IL-17A Limits Th17 Pathogenicity via a Negative Feedback Loop Driven by Autocrine Induction of IL-24. Immunity. 2020; 53: 384–397. https://doi.org/10.1016/j.immuni.2020.06.022 PMID: 32673565 11. Crawford MP, Sinha S, Renavikar PS, Borcherding N, Karandikar NJ. CD4 T cell-intrinsic role for the T helper 17 signature cytokine IL-17: Effector resistance to immune suppression. Proceedings of the National Academy of Sciences. 2020;117: 19408–19414. 12. Mohiuddin IH, Pillai V, Baughman EJ, Greenberg BM, Frohman EM, Crawford MP, et al. Induction of regulatory T-cells from memory T-cells is perturbed during acute exacerbation of multiple sclerosis. Clinical Immunology. 2016; 166: 12–18. https://doi.org/10.1016/j.clim.2016.05.001 PMID: 27154631 13. Baughman EJ, Mendoza JP, Ortega SB, Ayers CL, Greenberg BM, Frohman EM, et al. Neuroantigen- specific CD8+ regulatory T-cell function is deficient during acute exacerbation of multiple sclerosis. J Autoimmun. 2011; 36: 115–124. https://doi.org/10.1016/j.jaut.2010.12.003 PMID: 21257291 14. Cunnusamy K, Baughman EJ, Franco J, Ortega SB, Sinha S, Chaudhary P, et al. Disease exacerbation of multiple sclerosis is characterized by loss of terminally differentiated autoregulatory CD8+ T cells. Clinical immunology. 2014; 152: 115–126. https://doi.org/10.1016/j.clim.2014.03.005 PMID: 24657764 15. Bray NL, Pimentel H, Melsted P, Pachter L. Near-optimal probabilistic RNA-seq quantification. Nat Bio- technol. 2016; 34: 525–527. https://doi.org/10.1038/nbt.3519 PMID: 27043002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 14 / 15 PLOS ONEIL-17-mediated signaling in CD4+ T cells 16. Puel A, Cypowyj S, Bustamante J, Wright JF, Liu L, Lim HK, et al. Chronic mucocutaneous candidiasis in humans with inborn errors of interleukin-17 immunity. Science (1979). 2011; 332: 65–68. https://doi. org/10.1126/science.1200439 PMID: 21350122 17. Johansen C, Usher PA, Kjellerup RB, Lundsgaard D, Iversen L, Kragballe K. Characterization of the interleukin-17 isoforms and receptors in lesional psoriatic skin. British Journal of Dermatology. 2009; 160: 319–324. https://doi.org/10.1111/j.1365-2133.2008.08902.x PMID: 19016708 18. Al-Ramli W, Pre´fontaine D, Chouiali F, Martin JG, Olivenstein R, Lemière C, et al. TH17-associated cytokines (IL-17A and IL-17F) in severe asthma. Journal of Allergy and Clinical Immunology. 2009; 123: 1185–1187. https://doi.org/10.1016/j.jaci.2009.02.024 PMID: 19361847 19. Baeten D, Sieper J, Braun J, Baraliakos X, Dougados M, Emery P, et al. Secukinumab, an interleukin- 17A inhibitor, in ankylosing spondylitis. New England Journal of Medicine. 2015; 373: 2534–2548. https://doi.org/10.1056/NEJMoa1505066 PMID: 26699169 20. Mease PJ, McInnes IB, Kirkham B, Kavanaugh A, Rahman P, Van Der Heijde D, et al. Secukinumab inhibition of interleukin-17A in patients with psoriatic arthritis. New England Journal of Medicine. 2015; 373: 1329–1339. https://doi.org/10.1056/NEJMoa1412679 PMID: 26422723 21. Thac¸i D, Blauvelt A, Reich K, Tsai TF, Vanaclocha F, Kingo K, et al. Secukinumab is superior to usteki- numab in clearing skin of subjects with moderate to severe plaque psoriasis: CLEAR, a randomized controlled trial. J Am Acad Dermatol. 2015; 73: 400–409. https://doi.org/10.1016/j.jaad.2015.05.013 PMID: 26092291 22. Hueber W, Sands BE, Lewitzky S, Vandemeulebroecke M, Reinisch W, Higgins PDR, et al. Secukinu- mab, a human anti-IL-17A monoclonal antibody, for moderate to severe Crohn’s disease: Unexpected results of a randomised, double-blindplacebo- controlled trial. Gut. 2012; 61: 1693–1700. https://doi. org/10.1136/gutjnl-2011-301668 PMID: 22595313 23. Dick AD, Tugal-Tutkun I, Foster S, Zierhut M, Melissa Liew SH, Bezlyak V, et al. Secukinumab in the treatment of noninfectious uveitis: Results of three randomized, controlled clinical trials. Ophthalmology. 2013; 120: 777–787. https://doi.org/10.1016/j.ophtha.2012.09.040 PMID: 23290985 24. Genovese MC, Durez P, Richards HB, Supronik J, Dokoupilova E, Mazurov V, et al. Efficacy and safety of secukinumab in patients with rheumatoid arthritis: A phase II, dose-finding, double-blind, randomised, placebo controlled study. Ann Rheum Dis. 2013; 72: 863–869. https://doi.org/10.1136/annrheumdis- 2012-201601 PMID: 22730366 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285166 April 28, 2023 15 / 15 PLOS ONE
10.1371_journal.pone.0285211	RESEARCH ARTICLE An improved golden jackal optimization for multilevel thresholding image segmentation Zihao Wang1☯, Yuanbin MoID 2☯, Mingyue Cui1‡, Jufeng Hu1‡, Yucheng Lyu1‡ 1 School of Artificial Intelligence, Guangxi Minzu University, Nanning, China, 2 Guangxi Key Laboratory of Hybrid Computation and IC Design Analysis, Guangxi Minzu University, Nanning, China ☯ These authors contributed equally to this work. ‡ These authors also contributed equally to this work. moyuanbin2020@gxmzu.edu.cn Abstract Aerial photography is a long-range, non-contact method of target detection technology that enables qualitative or quantitative analysis of the target. However, aerial photography images generally have certain chromatic aberration and color distortion. Therefore, effective segmentation of aerial images can further enhance the feature information and reduce the computational difficulty for subsequent image processing. In this paper, we propose an improved version of Golden Jackal Optimization, which is dubbed Helper Mechanism Based Golden Jackal Optimization (HGJO), to apply multilevel threshold segmentation to aerial images. The proposed method uses opposition-based learning to boost population diversity. And a new approach to calculate the prey escape energy is proposed to improve the conver- gence speed of the algorithm. In addition, the Cauchy distribution is introduced to adjust the original update scheme to enhance the exploration capability of the algorithm. Finally, a novel “helper mechanism” is designed to improve the performance for escape the local optima. To demonstrate the effectiveness of the proposed algorithm, we use the CEC2022 benchmark function test suite to perform comparison experiments. the HGJO is compared with the original GJO and five classical meta-heuristics. The experimental results show that HGJO is able to achieve competitive results in the benchmark test set. Finally, all of the algorithms are applied to the experiments of variable threshold segmentation of aerial images, and the results show that the aerial photography images segmented by HGJO beat the others. Noteworthy, the source code of HGJO is publicly available at https://github.com/ Vang-z/HGJO. 1. Introduction Aerial imagery is an important component of modern photography and scientific research. Through aerial photography technology, we can obtain high-resolution images of natural land- scapes and urban architecture with previously unparalleled accuracy and detail. These images can be used in various fields such as map drawing [1], urban planning [2], land use planning [3], environmental monitoring [4], and agricultural and forestry resource management [5]. However, these high-resolution images may suffer from distortion due to external a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Wang Z, Mo Y, Cui M, Hu J, Lyu Y (2023) An improved golden jackal optimization for multilevel thresholding image segmentation. PLoS ONE 18(5): e0285211. https://doi.org/10.1371/ journal.pone.0285211 Editor: Diego Oliva, Universidad de Guadalajara, MEXICO Received: March 2, 2023 Accepted: April 17, 2023 Published: May 5, 2023 Copyright: © 2023 Wang et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All of the source code of this paper is publicly available at https://github. com/Vang-z/HGJO. Funding: This study was supported in the form of funding by the National Natural Science Foundation of China (Grant No. 21466008) awarded to Dr. Yuanbin Mo, the Natural Science Foundation of Guangxi Province (Grant No. 2019GXNSFAA185017) awarded to Dr. Yuanbin Mo, the Guangxi Minzu University Scientific Foundation (Grant No. 2021MDKJ004) awarded to Dr. Yuanbin Mo, and the Innovation Project of PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 1 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Guangxi Graduate Education (Grant No. YCSW2022255) awarded to Mr. Yucheng Lyu. environmental factors, which can make subsequent work difficult. Therefore, preprocessing aerial images is particularly important. Competing interests: The authors have declared that no competing interests exist. By segmenting, denoising, enhancing, and other operations on aerial images, we can effec- tively improve the quality and accuracy of the images, providing more reliable data support for subsequent applications. On the other hand, segmentation technology can also reduce the complexity of high-resolution images, making subsequent processing more efficient and accu- rate. Image segmentation is one of the most critical processes in computer vision. Preprocess- ing through segmentation technology can effectively reduce the complexity of subsequent processes. Segmentation technology is the process of dividing images into several categories based on pixels, with threshold segmentation being the most effective and commonly used method [6]. Due to its simplicity and stable performance, threshold segmentation has always been the pre- ferred segmentation technology [7]. In addition, there are clustering-based [8], edge-based [9], and region-based [10] segmentation methods that have also attracted the attention of research- ers. However, these methods are currently only effective for binary segmentation problems. As the number of segmentation targets increases beyond two, their processing capabilities decrease exponentially. This phenomenon is due to the exponential growth in complexity of multi-threshold image segmentation problems compared to binary segmentation, making tra- ditional segmentation methods unable to calculate feasible solutions within a limited time frame. Therefore, researchers have focused their attention on metaheuristics for these types of problems. In recent years, some of the most popular meta-heuristics used to solve engineering problems include Genetic Algorithms (GA) [11], Particle Swarm Optimization (PSO) [12], Differential Evolution (DE) [13], Ant Colony Optimization (ACO) [14], Whale Optimization Algorithm (WOA) [15], and Grey Wolf Optimization (GWO) [16]. In addition, some recently proposed metaheuristics have also been widely recognized by researchers, such as Arithmetic Optimization Algorithm [17], Aquila Optimizer (AO) [18], Ebola Optimization Search Algo- rithm (EOSA) [19], Dwarf Mongoose Optimization Algorithm (DMO) [20], Reptile Search Algorithm (RSA) [21], Prairie Dog Optimization Algorithm (PDO) [22], and Gazelle Optimi- zation Algorithm (GOA) [23]. These algorithms have achieved good results in dealing with real word engineering problems, providing new ideas for the development of multi-threshold image segmentation. The following is a summary of the meta-heuristics used by scholars in various fields to solve multilevel segmentation problems in recent years. Yin [24] proposed a recursive method to optimize the minimum cross entropy thresholding (MCET), which is combined with particle swarm optimization to search for the near-optimal MCET. The experimental results demonstrate that the MCET obtained by the proposed algo- rithm is very close to the optimal MCET. Osuna-Enciso, Cuevas and Sossa [25] presented a mixture of Gaussian functions to approximate the 1D histogram of a gray level image, which is combined with Particle Swarm Optimization, Artificial Bee Colony Optimization, and Differ- ential Evolution. The experimental outcomes indicate that each algorithm has its own advan- tages and disadvantages, and all of them could achieve more satisfactory results. Gao, Kwong, Yang and Cao [26] suggested a novel Particle Swarm Optimization with an intermediate per- turbation search strategy, which is called IDPSO. The proposed algorithm is compared with 10 different improved PSO algorithms which achieved outstanding performance on multilevel threshold segmentation problems. Jiang, Yeh, Hao and Yang [27] introduced a new hybrid algorithm which is based on honey bee mating algorithms and cooperative learning, further- more a new initialization population strategy is employed. Upon extensive experimental analy- sis, the algorithm is demonstrated to be able to be applied to complex image processing problems. Han, Yang, Zhou and Gui [28] used the State Transfer Algorithm (STA) to fit the optimal parameters of a linear combined normal distribution functions. The competitive PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 2 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation capability of the STA in thresholding segmentation is illustrated by comparison with the classi- cal meta-heuristics. He and Huang [29] proposed an effective improved multilevel color image thresholding firefly algorithm using Kapur’s entropy, minimum cross-entropy and inter-class variance method as the objective function. The experimental outcomes indicate that the pro- posed algorithm is superior to other classical metaheuristic algorithms in all aspects. Ishak [30] developed a two-dimensional multilevel thresholding technique based on Re´nyi and Tsallis entropies, which combines Quantum Genetic Algorithm and Differential Evolutionary to solve the segmentation problem of multimodal noisy images. A multi-level threshold image segmentation method using the Fruit Fly Optimization Algo- rithm was developed by Ding, Dong and Zou [31]. Extensive experimental indicate that the proposed algorithm could significantly reduce the time cost and also achieve satisfactory computational accuracy. To effectively segment coastal video images, a multilevel thresholding method based on Cuckoo Search Algorithm was designed by Widyantara et al [32]. This method was successful in overcoming a series of problems caused by nonlinear variations in image quality and opaque areas. Bohat and Arya [33] propose a novel threshold heuristic algo- rithm for the multilevel thresholding problem, which embeds the proposed algorithm into the Whale Optimization Algorithm, Gray Wolf Optimizer and Particle Swarm Optimization. Experiments illustrate that this work reduced the computation time of all the embedded algo- rithms. Singh, Mittal and Singh [34] presented an efficient multilevel thresholding image seg- mentation method based on Learning enthusiasm-based Teaching-Learning-Based Optimization. This method conquered the problem with increasing the level of redundant thresholds makes the computational complexity grow exponentially. Qualitative experimental outcomes demonstrate that the proposed algorithm is efficient in the field of image segmenta- tion. Xing [35] proposed a novel color image segmentation method based on Emperor Pen- guin Optimization for Berkeley images, Satellite images, and Plant canopy images. The experimental shows that the method has superior segmentation accuracy. In addition, the computational complexity does not increase exponentially due to the increase of thresholds. Upadhyay and Chhabra [36] suggested a Kapur’s entropy based Crow Search Algorithm to solve the optimal solution for multilevel thresholding image segmentation. By comparing with the classical meta-heuristics, the proposed algorithm achieves satisfactory performance with respect to both quality and consistency. Mousavirad and Ebrahimpour-Komleh [37] proposed a multilevel thresholding for image segmentation using Human Mental Search. This method combines Kapur’s entropy and Otsu method to achieve significant advantages in multi-thresh- old segmentation problems. Zhao et al. [38] proposed an improved Slime Mould Algorithm, which introduced a diffu- sion mechanism to increase the diversity of population. In addition, this methodology was suc- cessfully applied to the CT image segmentation of chronic obstructive pulmonary disease, which could help physicians to analyze the lesion tissues qualitatively and quantitatively, more- over improving the accuracy of diagnosis. Swain et al. [39] developed a multilevel thresholding image segmentation method based on differential exponential entropy. The method was com- bined with Equilibrium-Cuckoo Search Optimizer to achieve satisfactory performance in satel- lite image segmentation. Furthermore, the method is suggested in the paper for segmentation of the brain MR images. Houssein et al. [40] proposed a fresh approach based on the black widow optimizer to overcome the problem of high computational cost of multilevel threshold- ing image segmentation. This method has been compared with six well-known meta-heuris- tics. The comparison reveals that this method is the most potential alternative. Ma and Yue [41] proposed an improved multilevel thresholding image segmentation method based on the Whale Optimization Algorithm. The proposed method obtains satisfactory results for image segmentation in both grayscale and color images, respectively. Emam et al. [42] proposed an PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 3 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation enhanced RSA algorithm for global optimization and image segmentation, which overcomes the tendency of RSA to get stuck in local optima by combining it with the RUNge Kutta Opti- mizer (RUN) [43] and applies it to brain MRI image segmentation. The results showed that it outperformed other advanced meta-heuristics in terms of segmentation accuracy and compu- tational efficiency. To further advance research on COVID-19, Houssein et al. [44] used the opposition-based learning mechanism to improve the Manta Ray Foraging Optimization. Experimental results showed that the proposed method has higher robustness compared to existing meta-heuristics. Additionally, in [45], the Equilibrium Optimizer was further improved to advance research on COVID-19, and experimental results showed that the pro- posed method can be an effective tool for image segmentation. These two works further advance the research on image segmentation for COVID-19, making effective contributions to prevent the spread of COVID-19. In addition, Houssein et al. [46] introduced the opposition- based learning strategy into the Marine Predator Algorithm (MPA) [47] to accelerate the con- vergence speed of MPA. Finally, Otsu and Kapur entropy were used as objective functions to perform segmentation experiments on benchmark images. The experimental results showed that the proposed algorithm outperformed other methods. In [48], the authors used an improved Golden Jackal Optimization (GJO) [49] to segment skin cancer images, enhancing the original GJO algorithm using the opposition-based learning and comparing it with seven different meta-heuristics. The experimental results showed that the proposed method outperformed other alternative algorithms and effectively solved the seg- mentation problem. However, the improvement of this work is limited for the GJO algorithm, as the time complexity of thresholding increases exponentially with increasing image resolu- tion, and the single opposition-based learning mechanism cannot achieve satisfactory results on high-resolution images. Although meta-heuristics have been widely used in image segmen- tation, there are still shortcomings in multi-level threshold image segmentation for complex images. In other words, researchers are currently working on developing a method that can maintain consistent results when dealing with complex problems. The GJO algorithm is a novel and highly scalable swarm intelligence optimization algorithm proposed in 2022, which has been widely used by scholars [50–52]. Therefore, in this paper, we propose an efficient image segmentation method based on the GJO algorithm to further advance multi-level threshold segmentation work at high resolution and apply it to aerial image segmentation. In this study, we make further improvements to the original GJO to enhance its potential in multi-level thresholding image segmentation. In order to evaluate the effectiveness of the improve- ments, the proposed algorithm is compared with numerous classical and novel algorithms on CEC2022 benchmark functions. Moreover, we used Peak Signal to Noise Ratio (PSNR) [53], Struc- tural Similarity Index (SSIM) [54], and Feature Similarity Index (FSIM) [55] to determine the per- formance of image segmentation. The main contributions of this study are as follows: 1. The Opposition-Based Learning (OBL) strategy is integrated into the initialization of the GJO. The OBL strategy could dramatically improve the quality of the candidate solution to escape from the local optimal solution. 2. The Cauchy distribution is introduced to enhance the raw single Le´vy flight, strengthening the distributivity of the population to improve the capability of the algorithm to search the global optimum. Furthermore, a new approach to calculate the prey escape energy is pro- posed. It is a more reasonable nonlinear method of calculation, which leads to a better bal- ance of exploration and exploitation. 3. “Helpers” are introduced to improve the performance of GJO for the first time. They are some special individuals of the golden jackal population. The overall disturbance of the PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 4 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation population by the “Helpers” before the end of each iteration can effectively prevent the algo- rithm trapped into local optimum. 4. The proposed HGJO was compared with numerous classical and novel algorithms on CEC2022 benchmark functions, and a lot of the segmentation of aerial images. The out- comes demonstrate that the proposed HGJO has remarkably superior performance and enables to challenge the current existing algorithms. The rest of this paper is structured as follows: Section 2 encompasses a review of the multi- level threshold segmentation and the original GJO algorithm. The Improved Golden Jackal optimization is proposed in Section 3. Section 4 introduces, discusses, and analyzes the results of CEC2022 benchmark functions. Section 5 investigates the performance of the segmentation of aerial images. Finally, Section 6 concludes by summarizing the research and making recom- mendations for future work. 2. Literature review 2.1.Multilevel thresholding image segmentation Threshold segmentation, as the name implies, is the division of an image into two parts based on the pixel values, with a given threshold. However, for the current needs of computer vision tasks, it is often not enough to simply segment the image into two parts. Therefore, depending upon the current requirements, most scholars are investigating multilevel thresholding. In general, multilevel thresholding is the addition of more thresholds to binary thresholding to segment the image into more units. At the current stage, the most commonly used threshold- ing method for segmentation is the Otsu method. The Otsu method involves histogram of the image as the input, where the generated class information is employed to calculate the optimal threshold for segmenting the image. The Otsu method was firstly proposed by Otsu in 1979 [56] to segment the grey scale image by maximizing the variance between classes. The approach considers L to represent the different gray levels in an image which has the size of MN. n ¼ n0 þ n1 þ . . . þ nL(cid:0) 1 pi ¼ ni n ; XL(cid:0) 1 i¼0 pi ¼ 1 ð1Þ ð2Þ where n means the total number of pixels in the image, ni denotes the number of pixels for gray level i, and pi indicates the probability distribution of gray levels. Suppose there is a threshold k, in which 0<k<L−1, then the current input image will be seg- mented into two classes, namely C1 and C2, where C1 and C2 contain all pixels with the gray- scale in [0, k] and [k+1, L−1], respectively. P1ðkÞ ¼ Xk i¼0 pi; P2ðkÞ ¼ XL(cid:0) 1 i¼kþ1 pi ¼ 1 (cid:0) P1ðkÞ where P1(k) and P2(k) represent the probability of a pixel has been classified into C1 or C2, respectively. m1ðkÞ ¼ Xk i¼0 iPðijC1Þ ¼ Xk i¼0 iPðC1jiÞPðiÞ PðC1Þ ¼ 1 P1ðkÞ Xk i¼0 ipi PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 ð3Þ ð4Þ 5 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation m2ðkÞ ¼ XL(cid:0) 1 i¼kþ1 iPðijC2Þ ¼ XL(cid:0) 1 i¼kþ1 iPðC2jiÞPðiÞ PðC2Þ ¼ 1 P2ðkÞ XL(cid:0) 1 i¼kþ1 ipi mk ¼ Xk i¼0 ipi; mG ¼ XL(cid:0) 1 i¼0 ipi ð5Þ ð6Þ where m1(k) and m2(k) indicate the average gray value of the pixels in C1 and C2, respectively. mk denotes the average grayscale from 0 to k. mG represents the average grayscale of the whole image. Hence, we can derive Eq (7) without ambiguity. Then the between class variance can be expressed as Eq (8). P1ðkÞ � m1ðkÞ þ P2ðkÞ � m2ðkÞ ¼ mG P1ðkÞ þ P2ðkÞ ¼ 1 s2 B ¼ ðmGP1ðkÞ (cid:0) mkÞ2 P1ðkÞð1 (cid:0) P1ðkÞÞ ¼ P1ðkÞðm1ðkÞ (cid:0) mGÞ2 þ P2ðkÞðm2ðkÞ (cid:0) mGÞ2 ¼ P1ðkÞP2ðkÞðm1ðkÞ (cid:0) m2ðkÞÞ2 s2 Bðk∗Þ ¼ max 0�k�L(cid:0) 1 s2 BðkÞ ð7Þ ð8Þ ð9Þ As shown by Eq (8), we are able to determine a k to make the maximum of s2 B, which is denoted as Eq (9). Therefore, the Otsu method can be used as an objective function of an opti- mization problem to solve the optimal threshold value for segmented images. 2.2.Golden jackal optimization GJO is a novel metaheuristic algorithm proposed by Chopra and Ansari in 2022 [49]. GJO simulates the behavior of the golden jackal in natural environments for hunting. The search agents of this algorithm follow male and female jackals to seek, encircle and attack the prey, while the male jackal is considered as the global optimal solution of the problem. The entire description of the GJO is given below: 1. Initialization. As mentioned above, GJO is a population-based meta-heuristic. There- fore, the initialization of GJO is consistent with most meta-heuristics. The process of the ini- tialization is described in detail in Eq (10). �! Xk * ¼ LB * X ¼ ½X1 * þ r * ; X2 * � ðUB * ; . . .; Xn �T * (cid:0) LB Þ; k ¼ 1; 2; . . .; n ð10Þ indicates the position of the prey, n represents the size of is a random * stand for the lower and upper boundary, respectively. r * where X denotes the prey matrix, Xk * * population, LB and UB vector between 0 and 1. Algorithm 1 Pseudo-code of the GJO Inputs: The Population Size N and the Max iterations T. Outputs: The best solution. 1. Initialization the population X. 2. while (t<T) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 6 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Update the Evasion Energy (E) according Eq (13). if (\|E\|�1) Update the population according Eqs (11) and (12). else Update the population according Eqs (14) and (12). ðtÞ ¼ bestðXðtÞÞ ðtÞ ¼ second bestðXðtÞÞ 3. Calculating the fitness of the population. * 4. XM * 5. XFM 6. foreach (X(t)) 7. 8. 9. 10. 11. 12. 13. 14. t = t+1 15. end while * 16. return the best solution (XM end if end foreach ) 1. Exploration. In the exploration phase, the algorithm will search for as many potential solutions as possible over the search space. The position update is performed by a male jackal leading a female jackal, which is shown in the following mathematical model: * X1 * X2 * ¼ Xm * ¼ Xfm * ðtÞ (cid:0) E * ðtÞ (cid:0) E * � jXm * � jXfm * ðtÞ (cid:0) RL * ðtÞ (cid:0) RL * � Xk * � Xk ðtÞj ðtÞj * Xk ðt þ 1Þ ¼ * X1 * þ X2 2 ð11Þ ð12Þ ðt þ 1Þ stand for the position after it has been updated, * ðtÞ means the current position. XM * where t is the current iteration, Xk * ðtÞ denote the current positions of the male Xk * and female jackals, respectively, which are best and second-best fitness of the population. RL * a random vector which is based on Le´vy motion. E is the Evasion Energy of prey which is cal- culated as follows: * ðtÞ and XFM is * E * ¼ E1 � E0 ; E1 ¼ 1:5 � ð1 (cid:0) * t=TÞ; E0 * ¼ 2 � r (cid:0) 1 ð13Þ * where E1 is the decreasing energy of the prey, E0 the maximum iterations. is the initial energy of the prey. T stands for 1. Exploitation. As most of the meta-heuristics, the exploitation phase is based on the exploration phase. Through the exploitation of the candidate solutions which have been searched, the global optimal solution is approximated as closely as possible. The mathematical is shown as follows: * X1 * X2 * ¼ Xm * ¼ Xfm * ðtÞ (cid:0) E * ðtÞ (cid:0) E * � jRL * � jRL * � Xm * � Xfm * ðtÞ (cid:0) Xk * ðtÞ (cid:0) Xk ðtÞj ðtÞj ð14Þ It is worth noting that all of the variables are given in 2.2.2. Therefore, we do not repeat them. In addition, the pseudo-code of the GJO is shown in Algorithm 1. 3. Improved golden jackal optimization 3.1.Opposition-based learning strategy Since the generation for initial solution of GJO is consistent with most meta-heuristics, this approach may cause the initial population to be unevenly distributed and converge sluggishly. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 7 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation However, the performance of the algorithm is strongly influenced by the initial population. A high-quality initial population not only improves the convergence speed of the algorithm but even has the potential to determine the final outcomes. Enhancing population diversity can effectively avoid the algorithm from maturing prematurely and falling into a local optimum. Therefore, the Opposition-Based Learning(OBL) [57] strategy is used to assist with the genera- tion of the initial population in this study. Each individual in the population was given an Opposition solution to select the better individual as the initial solution, which could improve the convergence performance of GJO. The mathematical model of the OBL is shown as fol- lows: * Xk ~¼LB * þ UB * (cid:0) Xk ; k ¼ 1; 2; . . .n ð15Þ * where Xi~is the opposing individual of Xk * Xk , then Xk~will be retained as the initial individual. in the search space. If the fitness of Xk~is better than 3.2.Cauchy distribution and dynamic balance strategy 3.2.1.Cauchy distribution. The Cauchy distribution is a continuous probability distribu- tion without mathematical expectation. Better outcomes tended to be achieved when the motion state of the population was portrayed by the Cauchy distribution [58]. Fig 1 is a com- parison of the motion trajectory employing the Cauchy distribution and the Le´vy flight, which can be visualized that the Cauchy distribution is able to perform a comprehensive search in a given search space. Therefore, it is a more sensible choice to adopt the Cauchy distribution in the exploration stages. The probability density function of the Cauchy distribution is shown Fig 1. The movement of Cauchy distribution and Le´vy flight. https://doi.org/10.1371/journal.pone.0285211.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 8 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation below: " � f ðx; x0; gÞ ¼ 1=pg 1 þ x (cid:0) x0 g # �2 " # ¼ 1 p g ðx (cid:0) x0Þ2 þ g2 ð16Þ where x0 represents the position parameter, specifying the position of the peak of the distri- bution. γ denotes the scale parameter, which specifies the half-width at half-maximum. The Cauchy distribution which obeys X~C(0, 0.5) is utilized in this work. Where γ is fixed at 0.5 was determined by experimental analysis. Table 1 shows the effect of the algorithm using dif- ferent γ on the test results of IEEE CEC 2022. In this experiment, the population size was set to 60, the number of iterations was 10000, and 31 independent experiments were conducted. Finally, the results of 31 times were validated to Friedman mean rank test. Through observa- tion of the data in Table 1, we can realize that although when the value of γ is set to 0.5 does not achieve the optimal results on all test functions, it is still a best choice in general. In addi- tion, the results given by the Friedman calibration, γ set to 0.5 is also the best choice. Therefore, in this paper, the Cauchy distribution which obeys X~C(0, 0.5) is used. 3.2.2.Dynamic balance strategy. The balance of exploration and exploitation essentially * determines the performance of an algorithm [59]. It is not hard to see the escape energy (E * the GJO determines the exploration and exploitation. However, the calculation of E is defined ) in Table 1. The experimental of HGJO with different γ. Problem F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 Mean Std Mean Std Mean Std Mean Std Mean Std Mean Std Mean Std Mean Std Mean Std Mean Std Mean Std Mean Std Friedman mean rank Rank https://doi.org/10.1371/journal.pone.0285211.t001 γ = 0.25 300 3.91E-13 403.94923 2.13E+00 600.00001 9.11E-06 809.56445 3.52E+00 900 4.00E-13 1800.1992 1.12E-01 2010.8465 9.43E+00 2201.6878 1.50E+00 2529.2844 0.00E+00 2500.2164 4.38E-02 2600 5.87E-11 2860.0375 1.21E+00 2.7500 2 γ = 0.5 300 7.190E-14 400 4.896E-10 600.00001 1.446E-05 804.65384 5.963E-01 900 1.164E-12 1800.2297 8.529E-02 2002.9588 4.874E+00 2201.8592 1.340E+00 2529.2844 0.000E+00 2500.2173 5.886E-02 2600 1.422E-10 2859.0842 4.359E-01 2.3333 1 γ = 0.75 300 1.651E-13 403.47218 1.359E+00 600.00001 5.991E-06 809.53234 3.102E+00 900 5.142E-12 1800.2195 1.190E-01 2010.7526 9.548E+00 2202.8334 4.659E+00 2529.2844 0.000E+00 2500.2308 3.934E-02 2600 2.445E-10 2860.0648 1.138E+00 3.5417 5 γ = 1 300 3.226E-13 404.04741 1.636E+00 600.00001 1.747E-05 808.28063 2.509E+00 900 1.415E-12 1800.2166 1.087E-01 2008.7908 9.465E+00 2201.4274 1.191E+00 2529.2844 0.000E+00 2503.8003 1.990E+01 2600 4.723E-10 2859.8318 1.199E+00 3.4583 4 γ = 1.5 300 2.044E-13 403.60078 1.198E+00 600.00001 2.060E-05 808.89044 2.436E+00 900 1.925E-13 1800.2062 9.270E-02 2005.678 7.783E+00 2201.1876 9.500E-01 2529.2844 0.000E+00 2500.2212 4.141E-02 2600 1.364E-10 2860.2065 1.103E+00 2.9167 3 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 9 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Fig 2. The variation curves of the two escape energies. https://doi.org/10.1371/journal.pone.0285211.g002 by E1, which is a value that varies linearly according to the iteration. Thus, GJO might cause exploration and exploitation to be insufficiently balanced during the iteration. In order to * overcome this drawback, we propose a novel formula to calculate E (17). Fig 2 compares the variation curves of the two escape energies. , which is shown in Eq * E * ¼ E1 � E0 ; E1 ¼ 2 � ð1 (cid:0) * t=TÞpt=T; E0 * ¼ 2 � r (cid:0) 1 ð17Þ All variables have the same implications as those mentioned in the previous section. There- fore, they are not explained here. Note that how the proposed new escape energy will affect the exploration and exploitation of the algorithm is described in detail in Section 3.3. 3.3.The new update strategy Combined with the improvements proposed in the previous two sections, the position update strategy is also modified in this section. In the raw algorithm, exploration and exploitation are divided into two opposite parts, the algorithm will proceed to exploration when the escape energy is greater than 1, and vice versa. However, in this study, we split the iteration into three parts. First, there is the stage where the escape energy is greater than 1. In this stage, all PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 10 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation individuals will explore. Then, there is a phase with escape energy greater than 0.5, in which a part of the individuals in the population explores and the rest exploits. Finally, the stage with escape energy less than 0.5, in which all individuals were exploiting. The detailed mathematical model is shown below: * 1. In the first stage, when meanðabsðE ÞÞ > 1, the population will explore. However, their exploration is divided into two parts, part with reference to the current position of the indi- vidual, and the other part with reference to the center position of the whole population. 8 >>>>>>>>< >>>>>>>>: * X1 * X2 * X1 * X2 * ¼ Xk * ¼ Xk * ¼ Xk * ¼ Xk * ðtÞ (cid:0) E * ðtÞ (cid:0) E * ðtÞ (cid:0) E * ðtÞ (cid:0) E * � jXm * � jXfm * � jXm * � jXfm ðtÞj * � Xk * � Xk * ðtÞ (cid:0) RC * ðtÞ (cid:0) RC * ðtÞ (cid:0) RC � meanðXÞj * ðtÞ (cid:0) RC � meanðXÞj ; otherwise ; rand > 0:5 ðtÞj ð18Þ * where RC means a random vector generated by the Cauchy distribution, which obeys X~C(0, 0.5). rand denotes a random number between 0 and 1. mean(X) represents the central position of the current population. The rest of the variables have the same meaning as before. * 2. In the second stage, when 0:5 < meanðabsðE ÞÞ < 1, the population will transition from exploration to exploitation. In this process, a portion of the individuals will maintain the exploration, while the rest will transform to exploitation. 8 >>>>>>>>< >>>>>>>>: * ðtÞ (cid:0) E * ðtÞ (cid:0) RC * X1 * ¼ Xk * ¼ Xk * � jXm * � jXfm * X2 * X1 * X2 * ðtÞ (cid:0) E * ¼ Xm * ¼ Xfm * ðtÞ (cid:0) E * ðtÞ (cid:0) E * ðtÞ (cid:0) RC * * � Xm � jRC * � Xfm * � jRC ðtÞj ðtÞj * � Xk * � Xk * ðtÞ (cid:0) Xk * ðtÞ (cid:0) Xk ðtÞj ; otherwise ðtÞj ; rand > 0:5 ð19Þ All the variables have the same meaning as before. * 3. In the last stage, when meanðabsðE ÞÞ < 0:5, the population will enter the exploitation stage. In this stage, a part of the individuals will be exploitation depending on themselves, while the rest of the individuals will be exploitation depending on the center position of the population. 8 >>>>>>>>< >>>>>>>>: * X1 * X2 * X1 * X2 * ¼ Xm * ¼ Xfm * ¼ Xm * ¼ Xfm * ðtÞ (cid:0) E * � jRL * ðtÞ (cid:0) E * ðtÞ (cid:0) E * � jRL * � jRL * ðtÞ (cid:0) E * � jRL * � Xm * � Xfm * � Xm * � Xfm * ðtÞ (cid:0) Xk * ðtÞ (cid:0) Xk ðtÞj ðtÞj ðtÞ (cid:0) meanðXÞj ; otherwise ðtÞ (cid:0) meanðXÞj ; rand > 0:5 ð20Þ As well, all variables have the same meaning as mentioned before. Therefore, there is no need to go into too much detail. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 11 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation 3.4.The helper mechanism In general, there are individuals in the golden jackal group which are called “helpers” [49]. These helpers are the previous offspring of the golden jackal. Golden jackal populations are strengthened by helpers. This study focuses on “helpers” to enhance the global searchability of the algorithm with the introduction of “helpers”, which can effectively prevent the algorithm trapped into local optimal. 1. The first part of the “helpers” is to support the growth of the golden jackal pups. The mathe- matical model of this part is shown below: ( * Xk ðt þ 1Þ ¼ * Xk * ðt þ 1Þ; fitnessðXk * ðt þ 1ÞÞ < fitnessðXoffspring * Xoffspring * ðtÞ ¼ Xhelper1 * ðtÞ þ E * Xoffspring * � ðXhelper2 ðtÞ; otherwise * ðtÞ (cid:0) Xhelper3 ðtÞÞ ðtÞÞ ð21Þ * ; Xhelper2 * where Xhelper1 obtained offspring has a better fitness than the current updated individual, the individual posi- tion is updated to the position of the offspring. represent three random individuals, respectively. If the * and Xhelper3 2. The second part of the “helpers” is to take care of the pups while the golden jackal parents are out hunting. When other foragers are present, or something happens which could be harmful to the safety of the pups, the “helpers” will assist the pups in avoiding the danger. This part of the mechanism can be shown by how the algorithm escapes from the local opti- mal solution, which the mathematical model is as follows. ( * Xk * ðt þ 1Þ; fitnessðXk * ðt þ 1ÞÞ < fitnessðXhelper ðtÞÞ * Xk ðt þ 1Þ ¼ * Xhelper * ðtÞ ¼ Xi * Xhelper * ðtÞ þ rand � ðXrand1 ðtÞ; otherwise * ðtÞ (cid:0) Xrand2 ðtÞÞ ð22Þ * ; Xrand2 * represent two random individuals, respectively. rand denotes a random where Xrand1 number between 0 and 1. At this point, the improvement of IGJO is almost complete. The pseudo-code of HGJO is given in Algorithm 2. Algorithm 2 Pseudo-code of the HGJO Inputs: The Population Size N and the Max iterations T. Outputs: The best solution. 1. Initialization the population X according Eq (15). 2. while (t<T) 3. Calculating the fitness of the population. * 4. XM * 5. XFM 6. foreach (X(t)) ðtÞ ¼ bestðXðtÞÞ ðtÞ ¼ second bestðXðtÞÞ 7. 8. 9. 10. 11. * Update the Evasion Energy (E * if (meanðjE jÞ > 1) ) according Eq (17). Update the population according Eqs (12) and (18). * elseif (meanðjE jÞ > 0:5) Update the population according Eqs (12) and (19). PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 12 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation else end if Calculating, comparing and updating the fitness of offspring Update the population according Eqs (12) and (20). 12. 13. 14. 15. and current individual according Eq (21). 16. end foreach 17. Global perturbation with helpers according Eq (22). 18. t = t+1 19. end while * 20. return the best solution (XM ) 3.5.Computational complexity 3.5.1. Time complexity. Based on the pseudo-code in Algorithm 2, we could derive the time complexity of HGJO without difficulty. In which, the initial population spends O(NM) time to generated, where N denotes the size of the population, M represents the dimensions of the deci- sion space. Then, the fitness of the population needs O(TNOf) time to calculate, where T indi- cates the maximum iterations, Of is the cost of object function. In addition, the population needs O(TNM)time to be updated. Therefore, the total time complexity is O(N(T(Of+M)+M)). 3.5.2. Space complexity. Since no additional memory space is used in the computation, the space complexity of HGJO is limited only by the population size. Hence, the space com- plexity of HGJO is O(MN). 4. Experimental results and analysis In this section, we will evaluate the performance of the proposed algorithm. We will compare HGJO with six existing meta-heuristics on the CEC2022 test suite. These six meta-heuristics include the original GJO algorithm, the first variant of GJO algorithm called IGJO which uses OBL for improvement, two recently proposed widely used meta-heuristics, the RUN algorithm and the Archimedes Optimization Algorithm (AOA) [60], and the two most classical and stable algorithms, DE and PSO. Additionally, to further ensure that combining the OBL operator with the GJO algo- rithm is the most feasible option, we also included the OBL operator in the DE and PSO algorithms for comparison in the experiments. There are 12 different test functions in the CEC2022 test suite, which can cover a majority of the real-world problems. Therefore, the contents of this section enable us to make a preliminary understanding of the performance for HGJO. The details of the CEC2022 test suite are given in Table 2 and the runtime environment is also shown in Table 3. All Table 2. The benchmark functions of CEC2022. Problem No. F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 Problem name Zakharov Function Rosenbrock’s Function Schaffer’s F7 Rastrigin’s Function Levy Function Hybrid Function 1 Hybrid Function 2 Hybrid Function 3 Composition Function 1 Composition Function 2 Composition Function 3 Composition Function 4 Dim 10 10 10 10 10 10 10 10 10 10 10 10 https://doi.org/10.1371/journal.pone.0285211.t002 Range [–100, 100] [–100, 100] [–100, 100] [–100, 100] [–100, 100] [–100, 100] [–100, 100] [–100, 100] [–100, 100] [–100, 100] [–100, 100] [–100, 100] F_min 300 400 600 800 900 1800 2000 2200 2300 2400 2600 2700 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 13 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Table 3. Runtime environment. Configurations Hardware CPU GPU RAM Software OS Interpreter Language Intel(R) Core(TM) i9-10980HK CPU @ 2.40GHz NVIDIA GeForce RTX 3080 Laptop GPU 64.0 GB Microsoft Windows [Version 10.0.19043.1826] IntelliJ IDEA 2021.3.2 (Ultimate Edition) MATLAB R2021a (9.10.0.1602886) https://doi.org/10.1371/journal.pone.0285211.t003 algorithms are iterated with a population size of 60 and a maximum iteration of 1000. Further- more, in consideration of the suggestion by Arcuri et al. [61], all algorithm parameters are kept at their default values which are derived from their raw papers to ensure they are in a relatively opti- mal state, and these parameters are provided in Table 4. Moreover, the source code of the CEC2022 test set is available at: https://github.com/P-N-Suganthan/2022-SO-BO. 4.1.Statistical results on CEC2022 As mentioned above, the CEC2022 test suite is used to measure the performance of each algo- rithm, which includes both quantitative and qualitative metrics. The quantitative metrics include the mean, median, and standard deviation obtained by all algorithms. Qualitative met- rics are illustrated by convergence curves, which reflect the evolution of the optimal solution throughout the iterations of the algorithm. To ensure the fairness of the experiments, all algo- rithms were run 31 times independently on the CEC2022 benchmark test function. Table 5 provides the average time spent by all algorithms over the 31 runs, and lists the median, mean, and standard deviation of the best values obtained by all algorithms, the best results (minimum value) was highlighted in bold. The Friedman mean rank [62] was used to determine the over- all rank of each algorithm. According to the data in Table 5, we can see that the proposed method is optimal for most problems in terms of mean and median, except for F9 where AOA obtains the most accurate value. Similarly, the proposed algorithm also achieves satisfactory results for standard deviation in most problems. Therefore, we can consider that the proposed algorithm has excellent solving performance in CEC2022. However, in terms of running time, the AOA algorithm is the shortest in all problems, while the HGJO algorithm has some short- comings compared to it. Considering that the introduction of OBL and Cauchy operator will affect the efficiency of the algorithm to some extent, Fig 3 further analyzes the running time of all algorithms in CEC2022. Fig 3 is the average time slot of all algorithms running in CEC2022, Table 4. The parameters setting of all algorithms. Algorithm Parameters HGJO GJO IGJO RUN AOA DE(OBL) PSO(OBL) C1 = 2 C1 = 1.5 C1 = 1.5 a = 20, b = 12 C1 = 2, C2 = 6, C3 = 2, C4 = 0.5 PCr = 0.8, F = 0.85 C1 = 1.49445, C2 = 1.49445 https://doi.org/10.1371/journal.pone.0285211.t004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 14 / 37 PLOS ONETable 5. The results of CEC2022 for all algorithm. An improved golden jackal optimization for multilevel thresholding image segmentation Problem F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 Mean Median Std Time Mean Median Std Time Mean Median Std Time Mean Median Std Time Mean Median Std Time Mean Median Std Time Mean Median Std Time Mean Median Std Time Mean Median Std Time Mean Median Std Time Mean Median Std Time HGJO 300.000 300.000 GJO 682.748 439.932 IGJO 699.541 456.433 RUN 300.000 300.000 AOA 300.001 300.000 DE 1215.047 1153.934 PSO 1220.769 1217.032 1.228E-13 7.593E+02 7.395E+02 6.385E-05 1.627E-03 3.405E+02 6.384E+01 1.072 403.472 403.987 0.362 425.274 411.882 0.395 422.150 409.887 1.973 403.587 400.013 0.268 405.245 400.257 0.524 405.504 404.124 0.329 434.336 427.549 1.359E+00 2.210E+01 1.993E+01 4.363E+00 1.139E+01 2.219E+00 1.857E+01 1.149 600.000 600.000 0.362 605.511 604.463 0.389 605.925 605.740 1.976 611.779 608.887 0.271 600.245 600.097 0.533 600.064 600.061 0.327 615.090 612.536 1.419E-05 4.645E+00 4.768E+00 6.793E+00 2.906E-01 2.314E-02 5.356E+00 1.460 808.762 808.955 0.457 819.404 818.711 0.572 822.509 821.114 2.241 823.847 823.879 0.371 817.139 817.909 0.639 861.592 861.478 0.437 845.546 845.007 2.819E+00 5.913E+00 7.876E+00 6.765E+00 5.615E+00 1.967E+00 2.538E+00 1.246 900.000 900.000 0.394 959.189 951.961 0.448 967.710 942.420 2.041 977.192 968.285 0.303 908.414 900.454 0.568 900.041 900.014 0.363 977.753 977.414 3.771E-13 6.716E+01 8.875E+01 4.827E+01 3.302E+01 6.664E-02 6.523E+00 1.275 1800.226 1800.211 9.605E-02 1.202 2006.963 2001.372 0.399 7151.939 8154.988 0.457 7110.417 8182.686 2.066 2921.693 2533.530 0.307 2532.632 2171.087 0.542 1814.253 1814.067 2.193E+03 1.907E+03 1.012E+03 9.174E+02 1.567E+00 0.371 2032.196 2031.836 0.408 2037.376 2034.197 1.988 2037.731 2037.757 0.280 2021.994 2021.643 0.540 2019.993 2021.894 0.369 4354616.518 4264276.228 7.624E+05 0.343 2064.246 2062.244 8.907E+00 1.136E+01 1.121E+01 9.988E+00 9.576E+00 6.431E+00 5.565E+00 1.674 2201.458 2201.065 0.510 2224.956 2225.255 0.689 2225.259 2225.280 2.138 2223.042 2223.540 0.398 2219.205 2221.317 0.654 2210.591 2207.073 0.457 2247.440 2235.872 1.409E+00 4.580E+00 2.978E+00 1.564E+00 6.726E+00 8.650E+00 3.649E+01 1.667 2529.284 2529.284 0.548 2549.393 2531.291 0.761 2551.107 2541.380 2.302 2529.285 2529.284 0.000E+00 3.477E+01 2.645E+01 2.205E-04 1.436 2500.230 2500.225 4.586E-02 1.350 2600.000 2600.000 1.947E-10 1.711 0.470 2570.959 2614.641 0.604 2543.721 2500.606 2.127 2530.174 2500.631 6.120E+01 5.936E+01 5.094E+01 5.767E+01 0.460 2800.106 2735.136 0.579 2725.587 2731.824 2.026 2622.627 2600.018 0.352 2644.051 2600.001 1.735E+02 9.684E+01 7.946E+01 1.051E+02 0.531 0.729 2.175 0.431 0.482 2493.923 2493.923 8.775E-05 0.375 2551.838 2500.519 0.747 2529.284 2529.284 0.528 2545.972 2536.194 0.000E+00 3.849E+01 0.608 2500.876 2500.874 6.567E-02 0.580 2600.001 2600.001 4.481E-04 0.695 0.426 2579.107 2648.197 7.631E+01 0.413 2808.208 2761.214 1.260E+02 0.513 (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 15 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Table 5. (Continued) Problem F12 Mean Median Std Time Friedman mean rank Rank HGJO 2859.847 2859.369 GJO 2865.104 2864.532 IGJO 2865.137 2864.505 RUN 2863.978 2863.883 AOA 2878.827 2867.589 1.183E+00 3.054E+00 4.156E+00 1.668E+00 1.965E+01 1.699 1.1667 1 0.554 4.8333 5 0.768 5.0833 6 2.202 3.9583 4 0.443 3.3333 3 DE 2861.082 2861.405 9.283E-01 0.699 2.9583 2 PSO 2866.534 2866.482 7.997E-01 0.513 6.6667 7 https://doi.org/10.1371/journal.pone.0285211.t005 which can reflect the percentage of time each algorithm consumes when processing the same problem. The horizontal coordinate is the percentage of time consumed, and the vertical coor- dinate is the test function. Through Fig 3, we can clearly observe that compared with the origi- nal GJO, the time complexity of HGJO has increased significantly, but it still has some advantages compared with RUN. In addition, through the observation of 12 test functions, we can see that the running time of HGJO has not fluctuated significantly. Although OBL and Cauchy operator will increase certain time costs, the proposed algorithm shows the most satis- factory results in terms of overall performance. In addition, the Friedman mean rank is also used to comprehensively rank the algorithms, and the top three algorithms are HGJO, DE (OBL), and AOA, respectively. Table 6 uses the Wilcoxon rank-sum test [63] to further evaluate the running results of each algorithm. The Wilcoxon rank-sum test is used to verify whether there is a significant differ- ence between algorithms. When the p-value is less than 0.05, it can be considered that there is a significant difference between algorithms. On the contrary, it means that the performance of Fig 3. The average time slots for all algorithms. https://doi.org/10.1371/journal.pone.0285211.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 16 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Table 6. The Wilcoxon signed-rank test for CEC2022. Problem HGJO vs. GJO HGJO vs. IGJO HGJO vs. RUN HGJO vs. AOA HGJO vs. DE HGJO vs. PSO F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 7.329E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 1.192E-06 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.642E-07 ++ 6.018E-07 ++ 1.588E-06 ++ 6.018E-07 ++ 6.642E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 4.457E-02 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.642E-07 ++ 6.018E-07 ++ 6.642E-07 ++ 6.018E-07 ++ 1.192E-06 ++ 6.018E-07 ++ 3.001E-01 — 6.018E-07 ++ 1.588E-06 ++ 6.018E-07 ++ 6.018E-07 ++ 6.252E-05 ++ 1.312E-06 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 5.057E-04 ++ 6.018E-07 ++ 6.352E-06 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 2.308E-02 ++ 2.906E-03 ++ NaN — 6.642E-07 ++ 6.018E-07 ++ 1.451E-02 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ 6.018E-07 ++ https://doi.org/10.1371/journal.pone.0285211.t006 the proposed algorithm is similar to that of the compared algorithm. To better represent the analysis of values, we use the symbols "++" and "—" to indicate the cases where the p-value is less than 0.05 and greater than 0.05, respectively. From Table 6, we can clearly see that the pro- posed algorithm has significant differences compared with the original GJO, IGJO, RUN, and PSO(OBL). Combined with the data in Table 5, we can consider that it has significant improvements compared with the above algorithms. For AOA and DE(OBL), only similar per- formance was shown in F2 and F9, respectively. It is worth noting that we can see from the data in Table 5 that AOA achieved the best result in F9, which is also reflected in the Wilcoxon rank-sum test (significant difference between HGJO and AOA). Overall, according to the results of the Wilcoxon rank-sum test, we can consider that the proposed HGJO algorithm has higher performance on the CEC2022 test set. Fig 3 further shows the average time slots achieved by each algorithm on the IEEE CEC2022 benchmark test function. The figure displays the percentage of time consumed by each algorithm in processing the same test function. The horizontal coordinate is the percent- age of time consumed, and the vertical coordinate is the test function. Through this figure, we can observe that the time consumption of the HGJO algorithm increases to some extent, but there is no significant fluctuation for the overall. Therefore, this phenomenon verifies that the performance of the proposed algorithm is not limited to a particular problem. This makes the algorithm more extensible and can be more widely ap-plied to other optimization problems. 4.2.Convergence behavior analysis This subsection further presents the results of the convergence analysis of HGJO compared to other meta-heuristics. Fig 4 shows the convergence curves of all the algorithms for the CEC2022 benchmark test function. It is worth noting that all the curves in Fig 4 are calculated as the difference between the optimal values of their benchmark functions. It does not alter the exact meaning of the convergence curves but additionally enhances the observability of the images on the logarithmic axis. In short, the value closer to 0 indicates that the optimal solu- tion obtained by the algorithm is closer to the true optimal solution. With these convergence plots, we can see that the proposed algorithm reaches a stable point in all benchmark test func- tions, which can further indicate that the proposed algorithm is convergent. In addition, for most test functions, HGJO achieved the optimal solution with the least number of iterations, except for F2 and F9 where it was surpassed by RUN and AOA. Therefore, by analyzing the PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 17 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Fig 4. The convergence curves for all algorithms on CEC2022 benchmark functions. https://doi.org/10.1371/journal.pone.0285211.g004 convergence of the proposed algorithm and other competing algorithms, the superiority of HGJO is further validated, making it possible for HGJO to replace existing algorithms to solve complex problems. 4.3.Boxplot behavior analysis Due to the many local optima of the CEC2022 benchmark test functions, solving these prob- lems can easily fall into local optima. In order to analyze the algorithmic results more intui- tively, in this section we use boxplots to analyze HGJO and other metaheuristics. Fig 5 shows boxplots for all algorithms on the CEC2022 benchmark test function. The boxplots provide a visual representation of the distribution characteristics of the data, with the maximum and minimum values of the data corresponding to the highest and lowest points of the image, respectively. Therefore, the narrower the image reflects, the more stable the data. For most benchmark test functions, the proposed method has the narrowest and lowest boxplots. In fact, HGJO outperforms other metaheuristic algorithms in most test functions, except for F9. Combining the performance analysis of HGJO mentioned above, we can reasonably speculate that the proposed algorithm has the ability to solve complex engineering problems in the real world, providing a new candidate solution for scientific researchers to choose from. In the next section, we will use HGJO to handle the optical aerial image segmentation problem. 5. The optical aerial image segmentation In this section, HGJO is applied to a real-world scenario of multi-level thresholding segmenta- tion. The HGJO and all the comparison algorithms in the previous section are used to perform PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 18 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Fig 5. The box plots for all algorithms on CEC2022 benchmark functions. https://doi.org/10.1371/journal.pone.0285211.g005 threshold segmentation on the optical aerial image. The objective function used is the Otsu method introduced in Section 2.1. Finally, all algorithms are evaluated by PSNR, SSIM, and FSIM obtained after processing. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 19 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Fig 6. The selected images for segmentation. https://doi.org/10.1371/journal.pone.0285211.g006 5.1.Dataset and runtime environment The optical aerial image used for image segmentation experiments are from the MASATI dataset [64], where each image has a size of 512×512. In this study, 16 images are selected for segmenta- tion, which were named C0080, C0088, C0132, C0135, C0180, C0536, C1088, L0032, L0064, L0135, L0158, L0226, L0699, L0879, L1074, and X0017, as shown in Fig 6. All of these images are characterized by their own features which cover the vast majority of optical aerial image types. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 20 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Fig 7. The RGB histogram of C0080 and C0088. https://doi.org/10.1371/journal.pone.0285211.g007 Furthermore, Figs 7–14 show the RGB histogram of all the selected images. The threshold values of each image are set to 8, 16, 24, 32, respectively. To ensure the fairness of the experiments, the population size was fixed to 60, the maximum number of iterations was set to 1000 and all the algorithms were run independently 31 times with the same configured environment. 5.2.The evaluation metrics As mentioned above, in this study PSNR, SSIM, and FSIM will be used as the performance metrics for image segmentation. This subsection will describe the effects of these three metrics in detail. First is PSNR, peak signal-to-noise ratio, which is used to measure the difference between two images, which has a minimum value of 0 to represent the maximum difference. The Fig 8. The RGB histogram of C0132 and C0135. https://doi.org/10.1371/journal.pone.0285211.g008 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 21 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Fig 9. The RGB histogram of C0180 and C0536. https://doi.org/10.1371/journal.pone.0285211.g009 mathematical formula of PSNR can be described as follows: PSNR ¼ 20 log10 255 RMSE ; RMSE ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi XM XN ðIinði; jÞ (cid:0) Ioutði; jÞÞ2 v u u u u t i¼1 j¼1 M � N ð23Þ where RMES is the root mean square error, Iin and Iout are the original image and the seg- mented image, which have the size of M×N, respectively. Next is SSIM, Structural Similarity, which is used to evaluate the similarity between two images. The value of SSIM is between 0 and 1, and if two images are completely identical, then Fig 10. The RGB histogram of C1088 and L0032. https://doi.org/10.1371/journal.pone.0285211.g010 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 22 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Fig 11. The RGB histogram of L0064 and L0135. https://doi.org/10.1371/journal.pone.0285211.g011 the SSIM is equal to 1. Its mathematical model can be described as follows: SSIMðIin; IoutÞ ¼ ð2mIinmIout þ c1Þð2dIin;Iout þ c2Þ Iin þ m2 Iout þ c1Þðd2 Iin þ d2 Iout þ c2Þ ðm2 ð24Þ where μIin and μIout are the mean intensities of the original image and segmented image, respectively, σIin and σIout are the standard deviations of the original image and segmented image, respectively, σIin,Iout is the covariance of the original and segmented images, and C1 and C2 are two constants. Finally, is FSIM, Feature Similarity, which is calculated by Phase Consistency (PC) and Gra- dient Magnitude (GM). It reflects the difference in features between the two images. The FSIM has the same value as the SSIM, which is also between 0 and 1, and the closer to 1 indicates that Fig 12. The RGB histogram of L0158 and L0226. https://doi.org/10.1371/journal.pone.0285211.g012 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 23 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Fig 13. The RGB histogram of L0699 and L0879. https://doi.org/10.1371/journal.pone.0285211.g013 the feature information of the two images is more similar. It can be described as follows: X FSIM ¼ SLðxÞPCmðxÞ X x2O PCmðxÞ x2O SLðxÞ ¼ ½SPCðxÞ�a½SGðxÞ�b SPCðxÞ ¼ 2PCIinðxÞPCIoutðxÞ þ T1 IoutðxÞ þ T1 PC2 IinðxÞ þ PC2 SGðxÞ ¼ 2GIinðxÞGIoutðxÞ þ T2 IoutðxÞ þ T2 IinðxÞ þ G2 G2 ð25Þ ð26Þ ð27Þ ð28Þ Fig 14. The RGB histogram of L1074 and X0017. https://doi.org/10.1371/journal.pone.0285211.g014 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 24 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation where PCIin and PCIout are the PC of the original image and segmented image, respectively, and T1 is a positive constant used to increase the stability of SPC. GIin and GIout represent the gradients of the original and segmented images, respectively, while T2 is also a positive con- stant that controls the range of GM. α and β are two constants, respectively. In short, for all three metrics, PSNR, SSIM, and FSIM, the bigger the better. 5.3.Experimental results and analysis In this section, the experimental outcomes of multilevel thresholding segmentation are ana- lyzed by combining images and tables. To verify the adaptivity of HGJO in handling the seg- mentation task, the selected optical aerial images were segmented at threshold levels of 8, 16, 24, and 32, respectively. The algorithms involved in the segmentation comparison are consis- tent with the previous section, and the experimental results are evaluated by three metrics: PSNR, SSIM, and FSIM. In addition, the comprehensive performance of all algorithms is ranked using the Friedman mean rank test, and the Wilcoxon rank-sum test is used to analyze the fitness of all algorithms on the Otsu method. Tables 7–9 show the mean and std value of the PSNR, SSIM, and FSIM, respectively. It is noteworthy that the maximum mean and minimum std are highlighted in the tables. In addi- tion, it should be mentioned that the segmentation results of the same algorithm on different images may behave differently. This is because each image corresponds to a different problem being processed. When compared to other algorithms, HGJO performs exceptionally well at the specified threshold level. Additionally, as the threshold level increases, this performance does not degrade. According to the analysis of the results of PSNR recorded in Table 7, the proposed HGJO algorithm obtained the best experimental data in terms of accuracy and stability in the seg- mentation experiments of C0080, C0088, C0132, C0135, C0180, C1088, L0226, L0699 and X0017 for a total of 9 images. The other compared algorithms did not obtain the best results in the segmentation experiments for any of the images. The top three algorithms ranked by the Friedman mean rank test are as follows: the first ranked is the proposed HGJO algorithm, the AOA algorithm ranks second, and the RUN ranks third. In Table 8, which records the results about SSIM, the proposed algorithm achieves optimal results in the segmentation experiments of C0080, C0536, and X0017 for a total of 3 images. The rest of the algorithms also did not achieve optimal results on any of the images. According to the Friedman mean rank test, the top three algorithms were, HGJO, RUN, and AOA. Table 9 shows the experimental results of FSIM, which are similar to Tables 7 and 8. By evaluating the FSIM metrics, the proposed HGJO algorithm achieves optimal results in a total of 7 images, C0080, C0088, C0180, C0536, C1088, L1074, and X0017. The performance far exceeds that of other comparable algorithms. According to the Friedman average ranking test, the top three algorithms are HGJO, AOA, and RUN. In addition, Table 10 shows the results of the Wilcoxon rank-sum test. Consistent with the evaluation criteria in the previous section, if P>0.05, the null hypothesis is true; otherwise, the alternative hypothesis is true. By analyzing the results of the Wilcoxon rank-sum test, we can see that the proposed method has significant differences with other compared algorithms in all experiments, except for some experiments where it performs similarly to DE(OBL). Therefore, we can conclude that the performance of HGJO is significantly different from other algorithms. Despite the fact that the algorithms used for comparison have some competitiveness on some images. In all, however, the proposed algorithm is still outstanding in optical aerial image segmentation. These results demonstrate that the HGJO algorithm can achieve better PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 25 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Table 7. The PSNR results for all algorithms. Image Name Level HGJO GJO IGJO RUN AOA DE PSO C0080 C0088 C0132 C0135 C0180 C0536 C1088 L0032 L0064 L0135 L0158 Mean STD Mean STD Mean STD Mean STD Mean STD Mean STD Mean STD 24.49 4.05E-02 22.23 7.44E-01 23.29 4.29E-01 22.48 6.72E-01 23.35 2.77E-01 24.38 1.37E-02 18.67 1.23E+00 29.64 1.23E-01 27.32 9.44E-01 27.37 9.60E-01 28.06 6.36E-01 28.32 5.83E-01 27.62 3.88E-01 23.61 9.81E-01 34.82 5.66E-01 30.57 8.25E-01 29.67 1.13E+00 31.79 5.29E-01 31.60 6.71E-01 29.30 1.08E+00 26.08 1.35E+00 37.86 2.54E-01 31.39 1.12E+00 31.66 9.20E-01 33.13 7.09E-01 34.18 6.69E-01 31.09 1.16E+00 29.66 1.01E+00 24.43 3.82E-03 23.80 1.87E-01 24.14 9.64E-02 24.19 5.86E-02 24.32 3.60E-02 24.42 6.67E-03 19.90 9.88E-01 30.31 3.29E-02 27.69 5.86E-01 28.11 5.34E-01 28.63 3.54E-01 29.67 1.49E-01 28.87 1.31E-01 25.37 7.67E-01 33.88 7.34E-02 30.29 7.56E-01 30.99 4.90E-01 31.65 3.20E-01 32.76 1.47E-01 31.43 2.36E-01 26.86 8.28E-01 36.45 1.20E-01 32.80 6.64E-01 32.77 5.70E-01 33.59 5.92E-01 35.21 1.44E-01 33.11 3.22E-01 29.08 8.26E-01 25.06 9.41E-03 24.04 3.70E-01 24.00 3.24E-01 24.50 1.32E-01 24.47 1.84E-01 25.06 2.55E-03 21.80 7.87E-01 30.21 1.12E-01 28.45 4.60E-01 28.44 5.82E-01 28.83 3.68E-01 28.56 3.12E-01 27.15 7.10E-01 24.53 1.03E+00 33.67 1.32E-01 30.53 8.49E-01 30.76 5.64E-01 31.96 4.37E-01 32.31 2.78E-01 30.18 6.07E-01 27.45 8.25E-01 36.54 2.06E-01 31.11 1.11E+00 33.01 7.12E-01 34.66 4.48E-01 34.42 3.38E-01 31.44 7.07E-01 29.51 9.41E-01 24.88 1.49E-02 22.72 6.23E-01 23.80 2.50E-01 24.62 7.07E-02 24.66 4.91E-02 24.85 1.29E-02 20.93 6.83E-01 30.86 2.71E-02 27.30 7.52E-01 28.88 4.79E-01 29.04 3.22E-01 29.84 1.88E-01 28.69 3.60E-01 22.41 1.42E+00 34.36 5.83E-02 31.40 4.87E-01 30.84 7.32E-01 31.60 4.42E-01 32.33 3.44E-01 31.46 2.91E-01 28.01 8.52E-01 36.99 1.55E-01 33.25 7.34E-01 32.73 8.07E-01 33.97 5.09E-01 34.77 2.39E-01 33.18 3.62E-01 28.91 1.12E+00 25.93 7.28E-15 24.13 7.54E-01 24.14 4.32E-01 25.79 4.54E-02 24.48 6.67E-01 25.92 1.60E-02 19.93 1.34E+00 31.50 1.29E-01 28.27 8.29E-01 29.09 7.22E-01 29.28 4.50E-01 29.58 6.41E-01 28.28 5.83E-01 26.59 1.08E+00 35.56 1.79E-01 30.85 9.47E-01 31.25 9.23E-01 31.90 6.97E-01 31.40 8.09E-01 31.20 7.19E-01 26.83 1.45E+00 38.02 3.85E-01 33.10 7.69E-01 34.51 5.80E-01 34.81 7.18E-01 34.78 6.03E-01 32.67 8.31E-01 29.27 1.46E+00 23.96 1.20E-02 23.10 5.06E-01 23.15 3.92E-01 23.96 1.72E-01 23.92 1.30E-01 23.93 7.28E-15 19.21 9.90E-01 30.47 2.06E-01 27.36 8.19E-01 28.59 5.59E-01 29.66 3.74E-01 28.43 4.67E-01 28.44 5.46E-01 25.25 1.10E+00 34.82 2.39E-01 30.63 8.74E-01 31.74 6.40E-01 32.40 3.95E-01 32.08 3.16E-01 30.74 6.31E-01 26.26 1.42E+00 37.09 2.12E-01 32.63 1.08E+00 33.44 6.58E-01 34.26 5.18E-01 34.63 3.94E-01 33.00 5.55E-01 30.15 1.10E+00 22.72 3.61E-02 21.41 6.58E-01 20.96 6.48E-01 21.90 5.84E-01 22.55 1.57E-01 22.64 1.70E-02 14.25 2.08E+00 30.20 1.52E-01 25.01 1.14E+00 26.89 8.42E-01 27.74 5.48E-01 28.49 6.76E-01 26.10 9.36E-01 20.20 1.72E+00 34.42 2.71E-01 28.28 1.25E+00 28.86 8.17E-01 29.57 8.20E-01 31.94 7.32E-01 30.19 8.93E-01 25.24 1.76E+00 37.08 2.96E-01 29.26 1.13E+00 30.97 1.08E+00 30.14 1.23E+00 33.39 8.25E-01 32.09 7.79E-01 25.89 1.83E+00 19.41 8.40E-02 18.27 3.89E-01 18.27 3.90E-01 18.27 1.36E+00 18.77 3.27E-01 19.13 5.91E-04 17.15 1.24E+00 23.11 2.20E-01 18.27 3.27E+00 19.61 3.33E+00 20.23 3.08E+00 20.01 2.88E+00 21.07 9.06E-01 19.21 1.87E+00 26.59 5.89E-01 22.93 2.89E+00 21.06 3.70E+00 28.12 1.30E+00 22.29 2.55E+00 20.99 1.83E+00 19.33 3.11E+00 33.67 1.51E+00 24.71 2.36E+00 26.27 2.36E+00 23.23 3.22E+00 24.63 2.52E+00 21.87 2.73E+00 21.09 3.25E+00 22.97 9.47E-02 21.97 4.48E-01 22.70 2.81E-01 21.11 7.57E-01 22.70 2.40E-01 22.70 8.62E-02 20.32 7.46E-01 29.35 2.97E-01 26.34 1.38E+00 28.05 9.55E-01 29.90 3.12E-01 27.30 1.18E+00 25.09 9.36E-01 22.62 1.72E+00 33.15 3.16E-01 27.95 1.45E+00 30.39 8.34E-01 32.42 5.99E-01 31.42 8.71E-01 28.40 9.43E-01 23.92 1.75E+00 37.55 6.83E-01 32.91 8.16E-01 32.90 1.06E+00 34.96 6.14E-01 33.69 6.63E-01 29.35 1.70E+00 24.30 2.22E+00 22.65 2.40E-03 21.18 5.20E-01 22.15 8.36E-01 21.20 9.25E-01 22.64 2.73E-01 22.64 2.51E-03 18.91 1.23E+00 26.98 2.31E-01 24.73 1.78E+00 23.84 2.53E+00 27.72 1.25E+00 25.12 2.11E+00 24.58 1.10E+00 22.73 1.42E+00 34.26 1.30E+00 27.25 2.22E+00 27.33 2.12E+00 32.20 1.07E+00 27.91 2.58E+00 25.72 1.54E+00 24.53 2.32E+00 37.40 2.29E+00 31.09 1.32E+00 28.63 2.04E+00 31.15 1.81E+00 28.48 2.80E+00 26.62 1.69E+00 26.11 2.46E+00 22.18 3.73E-02 20.09 8.82E-01 20.95 4.25E-01 22.15 7.18E-01 21.80 3.12E-01 22.01 7.28E-15 16.66 1.40E+00 28.77 3.08E-01 25.99 1.28E+00 25.58 1.38E+00 28.95 5.23E-01 28.96 5.72E-01 25.42 9.80E-01 21.43 2.13E+00 33.98 5.43E-01 29.91 9.75E-01 28.55 1.47E+00 32.27 4.67E-01 32.12 6.11E-01 25.62 1.27E+00 23.21 2.32E+00 36.91 6.09E-01 32.29 8.77E-01 30.86 1.50E+00 31.23 1.23E+00 33.77 8.02E-01 29.66 1.42E+00 25.87 1.86E+00 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 26 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Table 7. (Continued) Image Name Level HGJO GJO IGJO RUN AOA DE PSO L0226 L0699 L0879 L1074 X0017 Mean STD Mean STD Mean STD Mean STD Mean STD Mean STD Mean STD 21.42 7.28E-15 19.79 9.06E-01 21.40 2.31E-01 21.29 5.09E-01 21.02 2.70E-01 21.42 1.58E-03 19.04 1.48E+00 29.47 4.40E-01 26.65 1.48E+00 26.62 1.75E+00 28.15 1.14E+00 25.90 1.91E+00 24.03 1.62E+00 21.29 2.26E+00 36.22 6.15E-01 32.38 8.56E-01 32.12 9.04E-01 34.04 4.84E-01 32.57 9.25E-01 25.81 2.38E+00 25.16 2.00E+00 39.28 2.34E-01 33.20 9.84E-01 34.61 8.55E-01 36.43 4.99E-01 35.06 6.74E-01 28.01 2.30E+00 25.39 2.28E+00 21.47 3.64E-15 18.85 1.04E+00 20.49 5.91E-01 20.50 1.39E+00 20.50 8.32E-01 21.47 3.64E-15 17.97 1.52E+00 30.10 5.00E-01 24.33 2.06E+00 27.08 1.34E+00 25.85 1.87E+00 27.55 1.56E+00 24.52 1.57E+00 20.41 2.70E+00 36.47 6.57E-01 28.99 1.53E+00 29.70 1.47E+00 27.37 1.95E+00 32.20 9.90E-01 28.58 1.94E+00 23.38 3.08E+00 39.63 3.95E-01 31.96 1.50E+00 31.48 1.66E+00 31.51 1.21E+00 35.91 6.84E-01 29.68 2.19E+00 24.74 2.34E+00 21.24 2.73E-04 20.50 5.17E-01 19.65 9.02E-01 20.50 1.29E+00 21.24 1.96E-01 21.24 3.64E-15 19.04 1.04E+00 27.92 4.45E-01 25.16 2.11E+00 24.97 2.05E+00 29.20 9.69E-01 24.41 2.17E+00 23.62 1.07E+00 21.55 1.50E+00 33.62 8.29E-01 29.70 1.29E+00 24.72 2.66E+00 31.98 1.05E+00 26.94 2.52E+00 25.51 1.89E+00 20.90 2.84E+00 38.05 1.48E+00 29.03 1.84E+00 30.59 1.70E+00 30.69 1.98E+00 33.78 1.39E+00 26.49 2.54E+00 24.02 2.54E+00 24.33 8.48E-03 23.13 6.21E-01 23.10 3.77E-01 24.49 2.97E-01 24.30 2.08E-01 24.30 7.28E-15 20.66 1.03E+00 31.32 1.25E-01 28.94 5.60E-01 29.48 4.96E-01 30.08 3.87E-01 29.82 4.15E-01 28.31 5.99E-01 25.21 9.50E-01 35.34 1.26E-01 31.97 5.45E-01 32.37 6.48E-01 32.90 4.09E-01 33.08 2.61E-01 31.26 4.99E-01 25.63 1.33E+00 38.08 1.89E-01 33.39 6.92E-01 34.05 8.71E-01 35.33 4.20E-01 35.10 3.77E-01 32.96 5.24E-01 29.60 8.36E-01 23.45 8.22E-03 22.19 4.50E-01 22.10 4.04E-01 22.72 1.84E-01 23.25 5.07E-02 23.39 1.47E-02 20.65 6.03E-01 29.26 7.51E-02 26.58 7.20E-01 27.56 4.75E-01 27.48 3.74E-01 28.68 1.21E-01 27.58 2.35E-01 24.42 9.23E-01 32.99 6.53E-02 30.39 4.81E-01 29.71 7.04E-01 30.36 4.55E-01 32.18 1.29E-01 30.40 2.57E-01 27.84 5.01E-01 35.60 9.65E-02 32.26 6.18E-01 32.25 7.39E-01 32.36 5.06E-01 34.54 1.36E-01 32.14 3.50E-01 28.82 7.64E-01 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 Friedman mean rank 1.1328 Rank 1 4.9844 6 4.5781 5 3.0156 3 2.8828 2 4.4219 4 6.98441 7 https://doi.org/10.1371/journal.pone.0285211.t007 Table 8. The SSIM results for all algorithms. Image Name Level HGJO GJO IGJO RUN AOA DE PSO C0080 C0088 C0132 C0135 C0180 Mean STD Mean STD Mean STD Mean STD Mean STD Mean STD Mean STD 0.892 8.99E-04 0.833 1.74E-02 0.868 7.74E-03 0.844 1.39E-02 0.869 5.55E-03 0.890 3.69E-04 0.660 4.64E-02 0.957 1.16E-03 0.937 9.46E-03 0.940 9.50E-03 0.945 6.07E-03 0.945 5.63E-03 0.932 5.13E-03 0.840 2.12E-02 0.987 3.13E-03 0.966 4.53E-03 0.963 6.18E-03 0.977 2.34E-03 0.974 3.20E-03 0.951 8.67E-03 0.900 1.83E-02 0.993 5.42E-04 0.973 4.29E-03 0.970 4.15E-03 0.983 1.95E-03 0.985 1.84E-03 0.963 6.76E-03 0.948 8.88E-03 0.938 4.63E-04 0.933 1.56E-03 0.937 7.18E-04 0.940 6.52E-04 0.940 8.79E-04 0.937 2.69E-04 0.841 2.42E-02 0.983 1.40E-04 0.970 2.84E-03 0.974 2.33E-03 0.978 1.41E-03 0.980 8.15E-04 0.976 1.17E-03 0.951 6.26E-03 0.993 1.33E-04 0.984 1.82E-03 0.986 1.25E-03 0.989 6.29E-04 0.990 4.02E-04 0.985 1.03E-03 0.961 5.25E-03 0.996 1.14E-04 0.991 1.11E-03 0.991 7.98E-04 0.993 6.94E-04 0.994 2.35E-04 0.989 8.64E-04 0.974 3.34E-03 0.859 2.48E-04 0.835 1.06E-02 0.809 1.59E-02 0.838 5.96E-03 0.837 6.73E-03 0.859 6.78E-04 0.615 6.47E-02 0.940 1.50E-03 0.917 1.31E-02 0.937 1.45E-02 0.944 3.81E-03 0.911 8.40E-03 0.872 1.77E-02 0.795 3.49E-02 0.977 1.68E-03 0.951 8.95E-03 0.945 8.01E-03 0.974 2.32E-03 0.965 3.12E-03 0.938 9.27E-03 0.859 2.28E-02 0.989 1.13E-03 0.939 1.05E-02 0.980 5.62E-03 0.986 1.57E-03 0.978 2.33E-03 0.941 8.56E-03 0.927 1.57E-02 0.907 7.95E-04 0.858 1.58E-02 0.882 5.95E-03 0.914 2.17E-03 0.910 1.88E-03 0.905 7.59E-04 0.759 3.13E-02 0.974 3.54E-04 0.946 6.34E-03 0.960 4.39E-03 0.968 1.96E-03 0.964 2.18E-03 0.951 4.17E-03 0.775 3.83E-02 0.989 2.38E-04 0.979 2.56E-03 0.977 3.05E-03 0.981 1.45E-03 0.980 1.58E-03 0.976 2.09E-03 0.939 9.22E-03 0.993 3.01E-04 0.985 1.74E-03 0.984 2.36E-03 0.989 8.99E-04 0.988 7.17E-04 0.981 1.69E-03 0.948 9.13E-03 0.888 1.14E-16 0.845 2.11E-02 0.843 1.32E-02 0.890 1.75E-03 0.849 2.03E-02 0.888 1.32E-04 0.689 5.44E-02 0.962 1.93E-03 0.928 1.05E-02 0.939 8.61E-03 0.945 5.41E-03 0.939 8.48E-03 0.908 1.18E-02 0.885 2.21E-02 0.986 8.68E-04 0.957 6.91E-03 0.958 6.82E-03 0.970 3.44E-03 0.956 6.58E-03 0.949 7.61E-03 0.900 2.12E-02 0.992 1.11E-03 0.977 4.08E-03 0.980 2.45E-03 0.984 2.33E-03 0.980 2.55E-03 0.960 6.16E-03 0.926 1.60E-02 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 (Continued ) 27 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Table 8. (Continued) Image Name Level HGJO GJO IGJO RUN AOA DE PSO C0536 C1088 L0032 L0064 L0135 L0158 L0226 L0699 L0879 L1074 X0017 Mean STD Mean STD Mean STD Mean STD Mean STD Mean STD Mean STD 0.832 5.78E-04 0.812 1.58E-02 0.814 1.36E-02 0.832 8.51E-03 0.830 6.65E-03 0.831 3.41E-16 0.626 5.13E-02 0.959 4.78E-03 0.932 1.09E-02 0.925 9.92E-03 0.951 5.21E-03 0.925 9.26E-03 0.923 1.14E-02 0.862 2.94E-02 0.987 1.46E-03 0.963 6.30E-03 0.970 4.19E-03 0.979 2.26E-03 0.969 3.01E-03 0.951 7.51E-03 0.868 2.90E-02 0.992 6.71E-04 0.974 5.01E-03 0.984 2.64E-03 0.986 1.56E-03 0.984 1.69E-03 0.968 5.25E-03 0.961 1.08E-02 0.847 1.02E-03 0.812 1.97E-02 0.795 2.03E-02 0.824 1.71E-02 0.842 4.92E-03 0.845 4.82E-04 0.240 1.59E-01 0.965 2.46E-03 0.898 1.87E-02 0.926 1.08E-02 0.954 4.77E-03 0.941 1.15E-02 0.907 1.40E-02 0.655 5.95E-02 0.988 1.26E-03 0.938 9.89E-03 0.960 5.27E-03 0.968 4.17E-03 0.983 2.66E-03 0.952 8.91E-03 0.886 2.28E-02 0.993 4.01E-04 0.966 5.92E-03 0.977 4.26E-03 0.974 3.72E-03 0.986 1.80E-03 0.969 4.01E-03 0.884 2.23E-02 0.582 6.35E-03 0.496 2.95E-02 0.496 2.92E-02 0.496 7.65E-02 0.534 2.39E-02 0.561 1.41E-05 0.533 9.14E-02 0.778 8.88E-03 0.496 1.24E-01 0.595 1.08E-01 0.639 7.73E-02 0.620 8.59E-02 0.684 3.84E-02 0.576 9.30E-02 0.869 1.39E-02 0.773 5.81E-02 0.686 9.87E-02 0.916 1.50E-02 0.742 5.78E-02 0.521 8.35E-02 0.579 1.19E-01 0.963 2.15E-02 0.829 3.49E-02 0.865 3.35E-02 0.797 5.47E-02 0.880 3.06E-02 0.799 4.73E-02 0.702 8.43E-02 0.812 3.68E-03 0.781 1.39E-02 0.802 1.10E-02 0.751 2.72E-02 0.802 9.50E-03 0.802 3.33E-03 0.702 3.21E-02 0.941 5.31E-03 0.894 2.56E-02 0.919 1.84E-02 0.964 4.50E-03 0.907 2.09E-02 0.853 2.07E-02 0.813 4.32E-02 0.972 3.10E-03 0.921 1.53E-02 0.947 8.76E-03 0.984 1.84E-03 0.959 8.91E-03 0.916 1.35E-02 0.819 3.57E-02 0.992 3.66E-03 0.980 2.92E-03 0.972 6.22E-03 0.988 1.33E-03 0.978 3.10E-03 0.927 1.84E-02 0.827 3.80E-02 0.793 1.37E-04 0.740 1.85E-02 0.778 3.00E-02 0.740 3.41E-02 0.793 1.05E-02 0.793 1.48E-04 0.582 5.97E-02 0.904 4.98E-03 0.857 3.03E-02 0.834 4.23E-02 0.923 1.39E-02 0.863 3.45E-02 0.847 2.41E-02 0.790 3.71E-02 0.982 1.37E-02 0.910 2.34E-02 0.909 2.27E-02 0.979 3.20E-03 0.918 2.40E-02 0.875 2.30E-02 0.846 3.80E-02 0.990 1.30E-02 0.961 6.98E-03 0.929 1.42E-02 0.971 5.55E-03 0.924 2.05E-02 0.890 2.18E-02 0.884 3.26E-02 0.865 1.03E-03 0.809 2.35E-02 0.834 1.07E-02 0.868 1.58E-02 0.855 7.76E-03 0.861 4.55E-16 0.670 4.68E-02 0.960 3.19E-03 0.931 1.34E-02 0.924 1.58E-02 0.970 3.64E-03 0.963 5.19E-03 0.918 1.31E-02 0.825 3.69E-02 0.988 2.69E-03 0.968 5.32E-03 0.957 8.87E-03 0.987 1.13E-03 0.983 2.22E-03 0.915 1.35E-02 0.880 2.94E-02 0.994 1.82E-03 0.983 2.24E-03 0.972 5.72E-03 0.979 3.31E-03 0.987 2.18E-03 0.962 8.50E-03 0.921 1.78E-02 0.680 2.28E-16 0.614 4.48E-02 0.679 1.04E-02 0.684 2.44E-02 0.661 1.23E-02 0.680 7.45E-05 0.561 6.85E-02 0.910 1.07E-02 0.866 3.50E-02 0.840 4.26E-02 0.912 2.30E-02 0.822 4.62E-02 0.760 4.64E-02 0.664 8.24E-02 0.987 5.91E-03 0.951 1.01E-02 0.973 9.77E-03 0.984 1.89E-03 0.953 1.11E-02 0.809 4.83E-02 0.798 4.95E-02 0.994 1.03E-03 0.953 8.19E-03 0.984 3.13E-03 0.990 1.09E-03 0.971 4.35E-03 0.859 3.39E-02 0.791 4.62E-02 0.760 2.28E-16 0.670 3.33E-02 0.727 1.83E-02 0.728 3.77E-02 0.728 2.37E-02 0.760 2.28E-16 0.601 5.31E-02 0.934 7.78E-03 0.832 3.56E-02 0.889 2.27E-02 0.865 2.85E-02 0.896 2.45E-02 0.833 2.84E-02 0.662 7.43E-02 0.988 4.64E-03 0.922 1.65E-02 0.925 1.43E-02 0.897 2.04E-02 0.953 8.70E-03 0.905 2.32E-02 0.808 7.42E-02 0.994 1.22E-03 0.960 6.83E-03 0.960 1.03E-02 0.963 6.47E-03 0.984 2.45E-03 0.920 1.91E-02 0.788 4.26E-02 0.789 9.89E-06 0.765 1.71E-02 0.733 3.13E-02 0.765 4.03E-02 0.789 6.74E-03 0.789 2.28E-16 0.705 4.08E-02 0.936 7.58E-03 0.891 2.92E-02 0.884 2.79E-02 0.963 6.92E-03 0.873 2.91E-02 0.853 2.23E-02 0.792 4.52E-02 0.981 5.95E-03 0.959 8.38E-03 0.877 2.76E-02 0.983 2.66E-03 0.919 1.91E-02 0.893 2.59E-02 0.760 5.82E-02 0.994 5.51E-03 0.945 1.09E-02 0.961 7.91E-03 0.969 6.45E-03 0.981 4.26E-03 0.908 2.60E-02 0.857 3.41E-02 0.929 6.34E-05 0.909 1.08E-02 0.909 7.15E-03 0.932 6.41E-03 0.929 4.24E-03 0.929 0.00E+00 0.833 2.63E-02 0.986 8.00E-04 0.976 2.79E-03 0.976 2.51E-03 0.983 1.33E-03 0.978 2.20E-03 0.965 5.00E-03 0.944 1.01E-02 0.995 2.31E-04 0.989 9.63E-04 0.989 1.60E-03 0.991 6.29E-04 0.990 5.82E-04 0.981 2.24E-03 0.923 1.32E-02 0.997 1.56E-04 0.992 9.63E-04 0.993 1.08E-03 0.995 3.91E-04 0.994 4.66E-04 0.986 1.69E-03 0.969 4.78E-03 0.921 2.38E-04 0.898 8.22E-03 0.896 7.49E-03 0.908 3.50E-03 0.918 1.06E-03 0.921 3.04E-04 0.869 1.15E-02 0.978 3.25E-04 0.959 4.91E-03 0.966 2.80E-03 0.966 2.54E-03 0.975 6.07E-04 0.966 1.63E-03 0.924 1.02E-02 0.989 3.28E-04 0.981 1.50E-03 0.979 2.33E-03 0.981 1.50E-03 0.988 3.13E-04 0.980 1.48E-03 0.964 3.61E-03 0.994 1.62E-04 0.988 1.50E-03 0.987 1.67E-03 0.988 1.12E-03 0.993 2.16E-04 0.986 1.16E-03 0.969 3.93E-03 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 Friedman mean rank 1.3203 Rank 1 4.7969 5 4.4219 4 2.5469 2 3.1172 3 4.8750 6 6.9219 7 https://doi.org/10.1371/journal.pone.0285211.t008 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 28 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Table 9. The FSIM results for all algorithms. Image Name Level HGJO GJO IGJO RUN AOA DE PSO C0080 C0088 C0132 C0135 C0180 C0536 C1088 L0032 L0064 L0135 L0158 Mean STD Mean STD Mean STD Mean STD Mean STD Mean STD Mean STD 0.975 1.33E-04 0.962 4.17E-03 0.970 2.03E-03 0.964 3.40E-03 0.971 1.18E-03 0.975 3.94E-05 0.885 2.12E-02 0.992 2.11E-04 0.984 2.12E-03 0.987 2.02E-03 0.988 1.49E-03 0.990 9.56E-04 0.986 1.52E-03 0.939 1.04E-02 0.997 5.07E-04 0.993 1.23E-03 0.993 1.33E-03 0.996 4.77E-04 0.996 6.36E-04 0.990 1.48E-03 0.977 3.62E-03 0.998 2.77E-04 0.995 9.76E-04 0.993 1.17E-03 0.997 3.93E-04 0.998 4.65E-04 0.992 1.43E-03 0.992 1.91E-03 0.988 1.61E-05 0.986 6.69E-04 0.987 3.89E-04 0.987 3.53E-04 0.987 3.03E-04 0.988 3.22E-05 0.957 7.57E-03 0.997 1.62E-04 0.995 6.08E-04 0.995 5.93E-04 0.995 3.75E-04 0.997 9.38E-05 0.995 6.02E-04 0.989 2.20E-03 0.999 5.39E-05 0.997 5.21E-04 0.998 3.82E-04 0.998 2.77E-04 0.999 5.21E-05 0.998 3.63E-04 0.991 1.30E-03 0.999 3.48E-05 0.998 3.44E-04 0.997 3.99E-04 0.998 1.92E-04 0.999 3.25E-05 0.998 1.97E-04 0.996 8.69E-04 0.972 3.60E-04 0.963 5.11E-03 0.974 1.56E-03 0.976 5.06E-04 0.976 2.05E-03 0.972 3.03E-04 0.951 9.26E-03 0.993 2.17E-04 0.988 1.28E-03 0.989 1.69E-03 0.990 9.38E-04 0.991 9.38E-04 0.989 2.02E-03 0.977 6.35E-03 0.996 2.06E-04 0.993 1.42E-03 0.995 1.09E-03 0.995 6.34E-04 0.996 4.78E-04 0.993 1.20E-03 0.983 3.29E-03 0.998 2.79E-04 0.992 1.47E-03 0.996 9.72E-04 0.998 3.16E-04 0.998 2.39E-04 0.997 5.51E-04 0.990 2.11E-03 0.977 8.04E-06 0.968 2.00E-03 0.976 4.70E-04 0.976 4.25E-04 0.978 2.87E-04 0.977 7.67E-06 0.944 8.20E-03 0.995 1.73E-04 0.987 1.87E-03 0.992 1.10E-03 0.990 9.83E-04 0.994 4.07E-04 0.989 1.27E-03 0.951 8.64E-03 0.998 1.60E-04 0.994 9.36E-04 0.994 1.19E-03 0.996 4.94E-04 0.997 1.81E-04 0.995 6.62E-04 0.984 2.68E-03 0.999 9.61E-05 0.997 7.34E-04 0.997 5.38E-04 0.997 3.68E-04 0.999 1.57E-04 0.996 6.31E-04 0.995 1.33E-03 0.956 0.00E+00 0.947 4.18E-03 0.948 2.06E-03 0.951 1.22E-03 0.955 3.55E-03 0.955 3.45E-04 0.917 2.50E-02 0.986 6.60E-04 0.974 3.50E-03 0.978 3.18E-03 0.979 2.06E-03 0.980 1.59E-03 0.979 2.74E-03 0.961 8.50E-03 0.995 2.80E-04 0.982 3.12E-03 0.985 2.72E-03 0.988 1.51E-03 0.987 1.59E-03 0.987 2.30E-03 0.954 1.14E-02 0.998 4.18E-04 0.991 2.08E-03 0.992 1.16E-03 0.995 9.35E-04 0.992 1.19E-03 0.991 2.81E-03 0.967 6.64E-03 0.947 1.43E-04 0.941 7.41E-03 0.942 5.99E-03 0.947 6.61E-03 0.947 2.75E-03 0.947 4.55E-16 0.934 1.08E-02 0.988 1.77E-03 0.981 2.34E-03 0.983 2.10E-03 0.985 1.29E-03 0.983 1.02E-03 0.981 4.16E-03 0.967 8.89E-03 0.996 7.85E-04 0.990 2.24E-03 0.990 1.86E-03 0.992 1.19E-03 0.989 8.61E-04 0.986 2.58E-03 0.972 9.37E-03 0.997 2.70E-04 0.990 2.03E-03 0.995 1.17E-03 0.995 6.55E-04 0.995 7.87E-04 0.991 1.78E-03 0.979 3.89E-03 0.951 8.77E-04 0.928 8.46E-03 0.921 8.23E-03 0.937 5.06E-03 0.947 1.83E-03 0.949 4.13E-04 0.799 4.52E-02 0.990 6.26E-04 0.963 6.40E-03 0.975 4.30E-03 0.985 1.73E-03 0.983 2.61E-03 0.975 4.34E-03 0.878 2.48E-02 0.996 4.42E-04 0.982 3.54E-03 0.989 1.61E-03 0.992 1.46E-03 0.994 1.00E-03 0.994 2.96E-03 0.979 1.06E-02 0.998 2.02E-04 0.990 1.57E-03 0.993 1.68E-03 0.993 1.31E-03 0.996 7.79E-04 0.994 1.83E-03 0.971 7.71E-03 0.921 6.75E-04 0.917 1.76E-03 0.917 1.57E-03 0.917 1.25E-02 0.919 1.96E-03 0.919 1.44E-04 0.884 3.37E-02 0.943 1.93E-03 0.917 2.62E-02 0.923 2.59E-02 0.927 1.88E-02 0.924 2.33E-02 0.924 9.88E-03 0.926 2.44E-02 0.980 8.13E-03 0.944 1.74E-02 0.932 2.61E-02 0.994 2.82E-03 0.900 2.24E-02 0.972 2.24E-02 0.918 3.44E-02 0.997 5.54E-03 0.955 9.36E-03 0.974 7.09E-03 0.947 1.52E-02 0.958 1.34E-02 0.829 3.63E-02 0.939 2.12E-02 0.969 5.92E-04 0.965 1.73E-03 0.968 1.98E-03 0.958 4.17E-03 0.967 1.31E-03 0.967 5.74E-04 0.944 1.07E-02 0.993 6.84E-04 0.985 3.19E-03 0.991 1.23E-03 0.995 4.15E-04 0.990 2.22E-03 0.983 4.46E-03 0.980 7.06E-03 0.997 3.60E-04 0.991 1.65E-03 0.994 1.10E-03 0.997 4.47E-04 0.997 9.12E-04 0.996 1.80E-03 0.975 7.58E-03 0.999 2.20E-04 0.998 6.91E-04 0.997 6.62E-04 0.999 3.46E-04 0.998 2.71E-04 0.996 2.42E-03 0.990 5.83E-03 0.967 4.77E-04 0.950 4.98E-03 0.963 3.52E-03 0.950 5.04E-03 0.967 5.39E-04 0.967 4.28E-04 0.865 3.01E-02 0.992 1.15E-03 0.978 3.88E-03 0.972 5.53E-03 0.993 1.86E-03 0.982 4.50E-03 0.979 4.13E-03 0.952 9.37E-03 0.998 6.89E-04 0.992 1.43E-03 0.991 1.36E-03 0.995 9.63E-04 0.995 8.49E-04 0.990 2.44E-03 0.981 7.88E-03 0.999 3.56E-04 0.996 7.33E-04 0.996 7.45E-04 0.994 1.70E-03 0.995 8.54E-04 0.987 2.41E-03 0.984 6.54E-03 0.964 5.09E-04 0.948 7.99E-03 0.954 3.67E-03 0.967 5.80E-03 0.960 3.09E-03 0.962 0.00E+00 0.921 1.65E-02 0.992 1.14E-03 0.983 3.95E-03 0.981 5.72E-03 0.996 7.57E-04 0.994 1.24E-03 0.988 6.61E-03 0.959 1.48E-02 0.998 7.64E-04 0.994 1.22E-03 0.991 2.42E-03 0.998 2.76E-04 0.998 3.94E-04 0.990 2.85E-03 0.977 1.17E-02 0.999 3.70E-04 0.998 3.13E-04 0.994 1.37E-03 0.997 4.39E-04 0.998 3.72E-04 0.992 1.80E-03 0.984 3.77E-03 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 29 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Table 9. (Continued) Image Name Level HGJO GJO IGJO RUN AOA DE PSO L0226 L0699 L0879 L1074 X0017 Mean STD Mean STD Mean STD Mean STD Mean STD Mean STD Mean STD 0.960 0.00E+00 0.941 5.84E-03 0.963 1.44E-03 0.953 2.15E-03 0.961 1.00E-03 0.961 2.56E-04 0.876 2.49E-02 0.990 8.55E-04 0.986 2.51E-03 0.986 2.03E-03 0.987 1.25E-03 0.984 3.05E-03 0.989 3.25E-03 0.972 8.56E-03 0.998 4.03E-04 0.994 1.01E-03 0.991 1.49E-03 0.995 7.68E-04 0.996 6.68E-04 0.988 2.81E-03 0.981 7.68E-03 0.999 1.09E-04 0.995 1.05E-03 0.996 6.03E-04 0.997 4.02E-04 0.997 3.57E-04 0.995 1.75E-03 0.991 3.28E-03 0.943 0.00E+00 0.918 9.03E-03 0.937 3.67E-03 0.936 9.32E-03 0.937 6.95E-03 0.943 0.00E+00 0.948 2.85E-02 0.986 2.34E-03 0.960 8.36E-03 0.973 5.58E-03 0.968 6.71E-03 0.974 5.98E-03 0.958 8.86E-03 0.923 2.14E-02 0.997 9.79E-04 0.987 3.22E-03 0.985 3.11E-03 0.981 4.08E-03 0.990 2.14E-03 0.979 6.18E-03 0.935 1.77E-02 0.999 2.49E-04 0.994 1.88E-03 0.993 2.09E-03 0.995 8.85E-04 0.999 7.23E-04 0.982 5.05E-03 0.953 1.18E-02 0.962 7.77E-05 0.955 4.93E-03 0.944 8.41E-03 0.955 8.43E-03 0.962 1.80E-03 0.962 2.28E-16 0.958 2.30E-02 0.993 1.05E-03 0.987 2.93E-03 0.985 3.04E-03 0.994 8.56E-04 0.984 3.28E-03 0.987 3.62E-03 0.975 1.08E-02 0.998 5.37E-04 0.995 9.02E-04 0.983 3.30E-03 0.997 4.97E-04 0.992 1.82E-03 0.985 2.70E-03 0.958 9.37E-03 0.999 3.64E-04 0.994 1.08E-03 0.997 5.13E-04 0.998 1.49E-03 0.998 2.75E-04 0.995 1.86E-03 0.990 6.74E-03 0.983 1.08E-04 0.978 2.66E-03 0.978 1.74E-03 0.983 7.57E-04 0.983 8.91E-04 0.983 0.00E+00 0.982 8.03E-03 0.996 2.65E-04 0.994 8.05E-04 0.993 1.17E-03 0.996 5.68E-04 0.996 6.06E-04 0.994 1.36E-03 0.989 3.70E-03 0.999 1.15E-04 0.998 6.61E-04 0.998 6.14E-04 0.997 3.67E-04 0.999 9.92E-05 0.996 9.59E-04 0.995 1.89E-03 0.999 7.77E-05 0.999 3.86E-04 0.998 5.77E-04 0.999 1.52E-04 0.999 4.11E-05 0.998 3.90E-04 0.994 1.60E-03 0.984 4.13E-05 0.979 1.94E-03 0.979 1.70E-03 0.980 1.13E-03 0.984 2.07E-04 0.984 9.19E-05 0.968 4.20E-03 0.996 1.18E-04 0.990 1.52E-03 0.993 8.46E-04 0.993 8.67E-04 0.995 1.69E-04 0.991 9.11E-04 0.981 3.24E-03 0.998 1.31E-04 0.996 3.31E-04 0.996 5.48E-04 0.996 3.13E-04 0.998 1.13E-04 0.995 6.34E-04 0.990 1.91E-03 0.999 7.92E-05 0.997 3.82E-04 0.997 4.43E-04 0.998 2.82E-04 0.999 5.66E-05 0.997 3.27E-04 0.990 1.43E-03 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 Friedman mean rank 1.5547 Rank 1 5.0938 6 4.4531 5 3.2031 3 2.6484 2 4.3125 4 6.7344 7 https://doi.org/10.1371/journal.pone.0285211.t009 results in processing optical aerial images. Combined with the analysis of the results for CEC2022 test suite in the previous section, it is reasonable to presume that the current results can be maintained when HGJO is applied to a wider range of optical aerial image segmentation in the future. In addition, an interesting phenomenon worth our attention is that the improved algorithm has a certain loss in the performance of land images which is named at the begin- ning of “L”. By comparing with the original image, we can observe that these images all have a common feature, including many small objects, which greatly increases the difficulty of seg- mentation. Therefore, in the future work, we can further optimize the segmentation problem of such images. Fig 15 displays the segmentation results of each algorithm on the test image C0180 with a threshold level of 8. They are the original image, HGJO segmentation result, GJO segmenta- tion result, IGJO segmentation result, RUN segmentation result, AOA segmentation result, DE(OBL) segmentation result and PSO(OBL) segmentation result in order. It is worth noting that only the image results with a threshold level of 8 are shown here, because it is impossible to intuitively feel the quality of the segmentation results in the high threshold segmentation results through human eye observation. Therefore, image results with a threshold level above 8 are not listed separately. With this figure, we can intuitively feel that the island in the optical aerial image segmented by HGJO have more distinct contours. In addition, by comparing the seabed distribution features in the segmentation results, we can observe that the competition algorithms ignore these details. However, HGJO preserves almost all the features of ocean dis- tribution. Therefore, it can be concluded that HGJO can effectively segment complex ocean distribution optical aerial images with high quality. In summary, the proposed HGJO can PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 30 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Table 10. Comparison of the Wilcoxon signed-rank test for Otsu method. Image Name Level HGJO vs. GJO HGJO vs. IGJO HGJO vs. RUN HGJO vs. AOA HGJO vs. DE HGJO vs. PSO C0080 C0088 C0132 C0135 C0180 C0536 C1088 L0032 L0064 L0135 L0158 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.578E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 2.643E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.906E-03 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 1.734E-04 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 6.104E-04 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 1.090E-04 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 9.766E-02 — 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 8.411E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 8.731E-03 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 1.289E-01 — 3.090E-05 ++ 3.090E-05 ++ 3.051E-05 ++ 6.168E-02 — 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 8.356E-04 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 1.367E-02 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 2.875E-05 ++ 2.489E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ (Continued ) PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 31 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Table 10. (Continued) Image Name Level HGJO vs. GJO HGJO vs. IGJO HGJO vs. RUN HGJO vs. AOA HGJO vs. DE HGJO vs. PSO L0226 L0699 L0879 L1074 X0017 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 8 16 24 32 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ NaN — 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 4.648E-01 – 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 9.766E-04 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 6.250E-02 — 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 5.662E-04 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ 3.090E-05 ++ https://doi.org/10.1371/journal.pone.0285211.t010 Fig 15. The segmentation results for C0180. https://doi.org/10.1371/journal.pone.0285211.g015 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 32 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation effectively handle optical aerial image segmentation and provide effective help for the subse- quent data processing and data acquisition. Furthermore, the segmentation histogram for each channel is also shown in Appendix A–G in S1 Appendix. By comparing these images, we can see that, the results of HGJO segmenta- tion on the red, green, and blue channels can retain more details. This shows that for optical aerial images, the proposed algorithm has more advantages and can retain more details after segmentation than other algorithms, which is helpful for subsequent image processing. 6. Conclusions Aerial photography images can provide a wealth of information for scientific researchers. Multi- level threshold segmentation of aerial images can effectively reduce the complexity of subsequent image processing while preserving the original features. This paper introduces a new optimization algorithm, the Golden Jackal Optimization (GJO) algorithm, which suffers from convergence issues and the tendency to get stuck in local optima. Therefore, an improved version of GJO, which is named HGJO, is proposed in this paper to enhance the search capabilities and avoid get- ting stuck in local optima, to optimize the process of multilevel thresholding segmentation. The performance of the proposed method is compared with six different meta-heuristics, including GJO, IGJO, RUN, AOA, DE(OBL), and PSO(OBL), on the IEEE CEC2022 benchmark test func- tion. Based on the experimental results, the proposed algorithm outperforms all other algorithms in terms of convergence accuracy and stability. In addition, the Otsu method is used as an objec- tive function to perform multi-level threshold segmentation on a set of aerial images. PSNR, SSIM, and FSIM are used as evaluation metrics to assess the quality of the segmented images pro- duced by each algorithm. Moreover, the Friedman mean rank test and the Wilcoxon rank-sum test are used to verify the segmentation results. The experimental results show that HGJO outper- forms other algorithms in terms of overall performance. The proposed algorithm can effectively reduce the image complexity while preserving the original features, thereby improving the effi- ciency of the subsequent image processing. In general, the results of this study are satisfactory, but there are certain shortcomings. Firstly, the introduction of OBL and Cauchy operators increases the computation time of the original algorithm, making the proposed method less efficient than most algorithms (only better than RUN). Secondly, there is a performance loss in complex image segmentation, such as test images L0032, L0064, and L0135, thus further work is needed to improve the algorithm’s performance in these types of image segmentation. In future work, we will further validate and improve the proposed algorithm through more extensive problems, such as medical image segmentation. In addition, we will try to apply HGJO to other more complex problems, such as neural networks, remote sensing data pro- cessing, and UAV path analysis. Furthermore, improving the computational efficiency of HGJO would be a significant contribution. Supporting information S1 Appendix. (DOCX) S1 File. (ZIP) Acknowledgments The authors would like to thank two anonymous reviewers for providing their valuable insights that could improve the quality of this study. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 33 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation Author Contributions Conceptualization: Zihao Wang. Methodology: Zihao Wang. Software: Zihao Wang. Supervision: Yuanbin Mo. Validation: Zihao Wang. Writing – original draft: Zihao Wang. Writing – review & editing: Yuanbin Mo, Mingyue Cui, Jufeng Hu, Yucheng Lyu. References 1. Fu Y, Liang S, Chen D, Chen Z. Translation of Aerial Image Into Digital Map via Discriminative Segmen- tation and Creative Generation. IEEE Transactions on Geoscience and Remote Sensing. 2022; 60: 1– 15. https://doi.org/10.1109/TGRS.2021.3110894 2. Vanegas CA, Aliaga DG, Benes B, Waddell P. Visualization of Simulated Urban Spaces: Inferring Parameterized Generation of Streets, Parcels, and Aerial Imagery. IEEE Transactions on Visualization and Computer Graphics. 2009; 15: 424–435. https://doi.org/10.1109/TVCG.2008.193 PMID: 19282549 3. Saing Z, Djainal H, Deni S. Land use balance determination using satellite imagery and geographic information system: case study in South Sulawesi Province, Indonesia. Geodesy and Geodynamics. 2021; 12: 133–147. https://doi.org/10.1016/j.geog.2020.11.006 4. Manfreda S, McCabe MF, Miller PE, Lucas R, Pajuelo Madrigal V, Mallinis G, et al. On the Use of Unmanned Aerial Systems for Environmental Monitoring. Remote Sensing. 2018; 10: 641. https://doi. org/10.3390/rs10040641 5. Shahbazi M, The´ au J, Me´ nard P. Recent applications of unmanned aerial imagery in natural resource management. GIScience & Remote Sensing. 2014; 51: 339–365. https://doi.org/10.1080/15481603. 2014.926650 6. Elaziz MA, Ewees AA, Yousri D, Alwerfali HSN, Awad QA, Lu S, et al. An Improved Marine Predators Algorithm With Fuzzy Entropy for Multi-Level Thresholding: Real World Example of COVID-19 CT Image Segmentation. IEEE Access. 2020; 8: 125306–125330. https://doi.org/10.1109/ACCESS.2020. 3007928 PMID: 34192114 7. Al-amri SS, Kalyankar NV, D KS. Image Segmentation by Using Threshold Techniques. arXiv; 2010. https://doi.org/10.48550/arXiv.1005.4020 8. Siva Raja PM, rani AV. Brain tumor classification using a hybrid deep autoencoder with Bayesian fuzzy clustering-based segmentation approach. Biocybernetics and Biomedical Engineering. 2020; 40: 440– 453. https://doi.org/10.1016/j.bbe.2020.01.006 9. Esnet: Edge-Based Segmentation Network for Real-Time Semantic Segmentation in Traffic Scenes. [cited 11 Jul 2022]. Available: https://ieeexplore.ieee.org/abstract/document/8803132/. 10. Lewis JJ O’Callaghan RJ, Nikolov SG, Bull DR, Canagarajah N. Pixel- and region-based image fusion with complex wavelets. Information Fusion. 2007; 8: 119–130. https://doi.org/10.1016/j.inffus.2005.09. 006 11. Mirjalili S. Genetic Algorithm. In: Mirjalili S, editor. Evolutionary Algorithms and Neural Networks: Theory and Applications. Cham: Springer International Publishing; 2019. pp. 43–55. https://doi.org/10.1007/ 978-3-319-93025-1_4 12. Bansal JC. Particle Swarm Optimization. Evolutionary and Swarm Intelligence Algorithms. 2019; 11– 23. https://doi.org/10.1007/978-3-319-91341-4_2 13. Das S, Suganthan PN. Differential Evolution: A Survey of the State-of-the-Art. IEEE Transactions on Evolutionary Computation. 2011; 15: 4–31. https://doi.org/10.1109/TEVC.2010.2059031 14. Dorigo M, Birattari M, Stutzle T. Ant colony optimization. IEEE Computational Intelligence Magazine. 2006; 1: 28–39. https://doi.org/10.1109/MCI.2006.329691 15. Mirjalili S, Lewis A. The Whale Optimization Algorithm. Advances in Engineering Software. 2016; 95: 51–67. https://doi.org/10.1016/j.advengsoft.2016.01.008 16. Mirjalili S, Mirjalili SM, Lewis A. Grey Wolf Optimizer. Advances in Engineering Software. 2014; 69: 46– 61. https://doi.org/10.1016/j.advengsoft.2013.12.007 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 34 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation 17. Abualigah L, Diabat A, Mirjalili S, Abd Elaziz M, Gandomi AH. The Arithmetic Optimization Algorithm. Computer Methods in Applied Mechanics and Engineering. 2021; 376: 113609. https://doi.org/10.1016/ j.cma.2020.113609 18. Abualigah L, Yousri D, Abd Elaziz M, Ewees AA, Al-qaness MAA, Gandomi AH. Aquila Optimizer: A novel meta-heuristic optimization algorithm. Computers & Industrial Engineering. 2021; 157: 107250. https://doi.org/10.1016/j.cie.2021.107250 19. Oyelade ON, Ezugwu AE-S, Mohamed TIA, Abualigah L. Ebola Optimization Search Algorithm: A New Nature-Inspired Metaheuristic Optimization Algorithm. IEEE Access. 2022; 10: 16150–16177. https:// doi.org/10.1109/ACCESS.2022.3147821 20. Agushaka JO, Ezugwu AE, Abualigah L. Dwarf Mongoose Optimization Algorithm. Computer Methods in Applied Mechanics and Engineering. 2022; 391: 114570. https://doi.org/10.1016/j.cma.2022.114570 21. Abualigah L, Elaziz MA, Sumari P, Geem ZW, Gandomi AH. Reptile Search Algorithm (RSA): A nature- inspired meta-heuristic optimizer. Expert Systems with Applications. 2022; 191: 116158. https://doi.org/ 10.1016/j.eswa.2021.116158 22. Ezugwu AE, Agushaka JO, Abualigah L, Mirjalili S, Gandomi AH. Prairie Dog Optimization Algorithm. Neural Comput & Applic. 2022; 34: 20017–20065. https://doi.org/10.1007/s00521-022-07530-9 23. Agushaka JO, Ezugwu AE, Abualigah L. Gazelle optimization algorithm: a novel nature-inspired meta- heuristic optimizer. Neural Comput & Applic. 2023; 35: 4099–4131. https://doi.org/10.1007/s00521- 022-07854-6 24. Yin P-Y. Multilevel minimum cross entropy threshold selection based on particle swarm optimization. Applied Mathematics and Computation. 2007; 184: 503–513. https://doi.org/10.1016/j.amc.2006.06. 057 25. Osuna-Enciso V, Cuevas E, Sossa H. A comparison of nature inspired algorithms for multi-threshold image segmentation. Expert Systems with Applications. 2013; 40: 1213–1219. https://doi.org/10.1016/ j.eswa.2012.08.017 26. Gao H, Kwong S, Yang J, Cao J. Particle swarm optimization based on intermediate disturbance strat- egy algorithm and its application in multi-threshold image segmentation. Information Sciences. 2013; 250: 82–112. https://doi.org/10.1016/j.ins.2013.07.005 27. Jiang Y, Yeh W-C, Hao Z, Yang Z. A cooperative honey bee mating algorithm and its application in multi-threshold image segmentation. Information Sciences. 2016; 369: 171–183. https://doi.org/10. 1016/j.ins.2016.06.020 28. Han J, Yang C, Zhou X, Gui W. A new multi-threshold image segmentation approach using state transi- tion algorithm. Applied Mathematical Modelling. 2017; 44: 588–601. https://doi.org/10.1016/j.apm. 2017.02.015 29. He L, Huang S. Modified firefly algorithm based multilevel thresholding for color image segmentation. Neurocomputing. 2017; 240: 152–174. https://doi.org/10.1016/j.neucom.2017.02.040 30. Ben Ishak A. A two-dimensional multilevel thresholding method for image segmentation. Applied Soft Computing. 2017; 52: 306–322. https://doi.org/10.1016/j.asoc.2016.10.034 31. Ding G, Dong F, Zou H. Fruit fly optimization algorithm based on a hybrid adaptive-cooperative learning and its application in multilevel image thresholding. Applied Soft Computing. 2019; 84: 105704. https:// doi.org/10.1016/j.asoc.2019.105704 32. Widyantara IMO, Pramaita N, Asana IMDP, Adnyana IBP, Pawana IGNA. Multilevel Thresholding for Coastal Video Image Segmentation Based on Cuckoo Search Algorithm. Proceedings of the 2019 5th International Conference on Computing and Artificial Intelligence. New York, NY, USA: Association for Computing Machinery; 2019. pp. 143–149. https://doi.org/10.1145/3330482.3330515 33. Bohat VK, Arya KV. A new heuristic for multilevel thresholding of images. Expert Systems with Applica- tions. 2019; 117: 176–203. https://doi.org/10.1016/j.eswa.2018.08.045 34. Singh S, Mittal N, Singh H. A multilevel thresholding algorithm using LebTLBO for image segmentation. Neural Comput & Applic. 2020; 32: 16681–16706. https://doi.org/10.1007/s00521-020-04989-2 35. Xing Z. An improved emperor penguin optimization based multilevel thresholding for color image seg- mentation. Knowledge-Based Systems. 2020; 194: 105570. https://doi.org/10.1016/j.knosys.2020. 105570 36. Upadhyay P, Chhabra JK. Kapur’s entropy based optimal multilevel image segmentation using Crow Search Algorithm. Applied Soft Computing. 2020; 97: 105522. https://doi.org/10.1016/j.asoc.2019. 105522 37. Mousavirad SJ, Ebrahimpour-Komleh H. Human mental search-based multilevel thresholding for image segmentation. Applied Soft Computing. 2020; 97: 105427. https://doi.org/10.1016/j.asoc.2019.04.002 38. Zhao S, Wang P, Heidari AA, Chen H, Turabieh H, Mafarja M, et al. Multilevel threshold image segmen- tation with diffusion association slime mould algorithm and Renyi’s entropy for chronic obstructive PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 35 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation pulmonary disease. Computers in Biology and Medicine. 2021; 134: 104427. https://doi.org/10.1016/j. compbiomed.2021.104427 PMID: 34020128 39. Swain M, Tripathy TT, Panda R, Agrawal S, Abraham A. Differential exponential entropy-based multi- level threshold selection methodology for colour satellite images using equilibrium-cuckoo search opti- mizer. Engineering Applications of Artificial Intelligence. 2022; 109: 104599. https://doi.org/10.1016/j. engappai.2021.104599 40. Houssein EH, Helmy BE, Oliva D, Elngar AA, Shaban H. A novel Black Widow Optimization algorithm for multilevel thresholding image segmentation. Expert Systems with Applications. 2021; 167: 114159. https://doi.org/10.1016/j.eswa.2020.114159 41. Ma G, Yue X. An improved whale optimization algorithm based on multilevel threshold image segmen- tation using the Otsu method. Engineering Applications of Artificial Intelligence. 2022; 113: 104960. https://doi.org/10.1016/j.engappai.2022.104960 42. Emam MM, Houssein EH, Ghoniem RM. A modified reptile search algorithm for global optimization and image segmentation: Case study brain MRI images. Computers in Biology and Medicine. 2023; 152: 106404. https://doi.org/10.1016/j.compbiomed.2022.106404 PMID: 36521356 43. Ahmadianfar I, Heidari AA, Gandomi AH, Chu X, Chen H. RUN beyond the metaphor: An efficient opti- mization algorithm based on Runge Kutta method. Expert Systems with Applications. 2021; 181: 115079. https://doi.org/10.1016/j.eswa.2021.115079 44. Houssein EH, Emam MM, Ali AA. Improved manta ray foraging optimization for multi-level thresholding using COVID-19 CT images. Neural Comput & Applic. 2021; 33: 16899–16919. https://doi.org/10.1007/ s00521-021-06273-3 PMID: 34248291 45. Houssein EH, Helmy BE, Oliva D, Jangir P, Premkumar M, Elngar AA, et al. An efficient multi-threshold- ing based COVID-19 CT images segmentation approach using an improved equilibrium optimizer. Bio- medical Signal Processing and Control. 2022; 73: 103401. https://doi.org/10.1016/j.bspc.2021.103401 46. Houssein EH, Hussain K, Abualigah L, Elaziz MA, Alomoush W, Dhiman G, et al. An improved opposi- tion-based marine predators algorithm for global optimization and multilevel thresholding image seg- mentation. Knowledge-Based Systems. 2021; 229: 107348. https://doi.org/10.1016/j.knosys.2021. 107348 47. Faramarzi A, Heidarinejad M, Mirjalili S, Gandomi AH. Marine Predators Algorithm: A nature-inspired metaheuristic. Expert Systems with Applications. 2020; 152: 113377. https://doi.org/10.1016/j.eswa. 2020.113377 48. Houssein EH, Abdelkareem DA, Emam MM, Hameed MA, Younan M. An efficient image segmentation method for skin cancer imaging using improved golden jackal optimization algorithm. Computers in Biol- ogy and Medicine. 2022; 149: 106075. https://doi.org/10.1016/j.compbiomed.2022.106075 PMID: 36115303 49. Chopra N, Mohsin Ansari M. Golden jackal optimization: A novel nature-inspired optimizer for engineer- ing applications. Expert Systems with Applications. 2022; 198: 116924. https://doi.org/10.1016/j.eswa. 2022.116924 50. Rezaie M, karamnejadi azar K, kardan sani A, Akbari E, Ghadimi N, Razmjooy N, et al. Model parame- ters estimation of the proton exchange membrane fuel cell by a Modified Golden Jackal Optimization. Sustainable Energy Technologies and Assessments. 2022; 53: 102657. https://doi.org/10.1016/j.seta. 2022.102657 51. Zhang J, Zhang G, Kong M, Zhang T. Adaptive infinite impulse response system identification using an enhanced golden jackal optimization. J Supercomput. 2023 [cited 12 Mar 2023]. https://doi.org/10. 1007/s11227-023-05086-6 52. Mahdy A, Hasanien HM, Turky RA, Abdel Aleem SHE. Modeling and optimal operation of hybrid wave energy and PV system feeding supercharging stations based on golden jackal optimal control strategy. Energy. 2023; 263: 125932. https://doi.org/10.1016/j.energy.2022.125932 53. Huynh-Thu Q, Ghanbari M. Scope of validity of PSNR in image/video quality assessment. Electronics Letters. 2008; 44: 800–801. 54. Wang Z, Bovik AC, Sheikh HR, Simoncelli EP. Image quality assessment: from error visibility to struc- tural similarity. IEEE Transactions on Image Processing. 2004; 13: 600–612. https://doi.org/10.1109/ tip.2003.819861 PMID: 15376593 55. Zhang L, Zhang L, Mou X, Zhang D. FSIM: A Feature Similarity Index for Image Quality Assessment. IEEE Transactions on Image Processing. 2011; 20: 2378–2386. https://doi.org/10.1109/TIP.2011. 2109730 PMID: 21292594 56. Otsu N. A threshold selection method from gray-level histograms. IEEE transactions on systems. 1979; 9: 62–66. PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 36 / 37 PLOS ONEAn improved golden jackal optimization for multilevel thresholding image segmentation 57. Tizhoosh HR. Opposition-Based Learning: A New Scheme for Machine Intelligence. International Con- ference on Computational Intelligence for Modelling, Control and Automation and International Confer- ence on Intelligent Agents, Web Technologies and Internet Commerce (CIMCA-IAWTIC’06). 2005. pp. 695–701. https://doi.org/10.1109/CIMCA.2005.1631345 58. Ali M, Pant M. Improving the performance of differential evolution algorithm using Cauchy mutation. Soft Comput. 2011; 15: 991–1007. https://doi.org/10.1007/s00500-010-0655-2 59. Hashim FA, Hussien AG. Snake Optimizer: A novel meta-heuristic optimization algorithm. Knowledge- Based Systems. 2022; 242: 108320. https://doi.org/10.1016/j.knosys.2022.108320 60. Hashim FA, Hussain K, Houssein EH, Mabrouk MS, Al-Atabany W. Archimedes optimization algorithm: a new metaheuristic algorithm for solving optimization problems. Appl Intell. 2021; 51: 1531–1551. https://doi.org/10.1007/s10489-020-01893-z 61. Arcuri A, Fraser G. Parameter tuning or default values? An empirical investigation in search-based soft- ware engineering. Empir Software Eng. 2013; 18: 594–623. https://doi.org/10.1007/s10664-013-9249-9 62. 63. Friedman M. The Use of Ranks to Avoid the Assumption of Normality Implicit in the Analysis of Vari- ance. Journal of the American Statistical Association. 1937; 32: 675–701. https://doi.org/10.1080/ 01621459.1937.10503522 Liao C, Li S, Luo Z. Gene Selection Using Wilcoxon Rank Sum Test and Support Vector Machine for Cancer Classification. In: Wang Y, Cheung Y, Liu H, editors. Computational Intelligence and Security. Berlin, Heidelberg: Springer; 2007. pp. 57–66. https://doi.org/10.1007/978-3-540-74377-4_7 64. Gallego A-J, Pertusa A, Gil P. Automatic Ship Classification from Optical Aerial Images with Convolu- tional Neural Networks. Remote Sensing. 2018; 10: 511. https://doi.org/10.3390/rs10040511 PLOS ONE \| https://doi.org/10.1371/journal.pone.0285211 May 5, 2023 37 / 37 PLOS ONE
10.1371_journal.pone.0287011	RESEARCH ARTICLE Time series and power law analysis of crop yield in some east African countries Idika E. Okorie1, Emmanuel Afuecheta2,3, Saralees NadarajahID 4* 1 Department of Mathematics, Khalifa University, Abu Dhabi, UAE, 2 Department of Mathematics and Statistics, King Fahd University of Petroleum & Minerals, Dhahran, Saudi Arabia, 3 Interdisciplinary Research Center for Finance and Digital Economy, KFUPM, Dhahran, Saudi Arabia, 4 Department of Mathematics, University of Manchester, Manchester, United Kingdom * mbbsssn2@manchester.ac.uk Abstract a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Okorie IE, Afuecheta E, Nadarajah S (2023) Time series and power law analysis of crop yield in some east African countries. PLoS ONE 18(6): e0287011. https://doi.org/10.1371/journal. pone.0287011 Editor: Steven Arthur Loiselle, University of Siena, ITALY Received: July 7, 2022 Accepted: May 27, 2023 Published: June 13, 2023 Copyright: © 2023 Okorie et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting information files. Funding: The authors received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. We carry out a time series analysis on the yearly crop yield data in six east African countries (Burundi, Kenya, Somalia, Tanzania, Uganda and Rwanda) using the autoregressive inte- grated moving average (ARIMA) model. We describe the upper tail of the yearly crop yield data in those countries using the power law, lognormal, Fre´ chet and stretched exponential distributions. The forecast of the fitted ARIMA models suggests that the majority of the crops in different countries will experience neither an increase nor a decrease in yield from 2019 to 2028. A few exceptional cases correspond to significant increase in the yield of sor- ghum and coffee in Burundi and Rwanda, respectively, and significant decrease in the yield of beans in Burundi, Kenya and Rwanda. Based on Vuong’s similarity test p–value, we find that the power law distribution captured the upper tails of yield distribution better than other distributions with just one exceptional case in Uganda, suggesting that these crops have the tendency for producing high yield. We find that only sugar cane in Somalia and sweet potato in Tanzania have the potential of producing extremely high yield. We describe the yield behaviour of these two crops as black swan, where the “rich getting richer” or the “preferen- tial attachment” could be the underlying generating process. Other crops in Burundi, Kenya, Somalia, Tanzania, Uganda and Rwanda can only produce high but not extremely high yields. Various climate adaptation/smart strategies (use of short-duration pigeon pea varie- ties, use of cassava mosaic disease resistant cassava varieties, use of improved maize vari- eties, intensive manuring with a combination of green and poultry manure, early planting, etc) that could be adapted to increase yields in east Africa are suggested. The paper could be useful for future agricultural planning and rates calibration in crop risk insurance. 1 Introduction Africa is the poorest continent. It is struggling to feed its people. Hence, enhancement of crop production is important. Furthermore, farmers are more interested in investing in crops that are capable of produc- ing high yields not crops that can produce extremely low yield. They want to maximize the profit on their investment. Crops that have the potential for high yield are likely to attract low premium in crop yield insurance. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 1 / 36 PLOS ONECrop yield in some east African countries There have been several papers on high crop yield in African countries. While discussing nutrients in the west African Sudano-Sahelian zone, [1] noted that “shrubs and trees with their alternating periods of nutrient storing and recycling in leaves and wood, micro-depressions, termite mounts and ant nests become localised points of nutrient concentration and high crop productivity”. While investigating the importance of liming acid soils, [2] demonstrated that “severely acidified soils of the western highlands of Cameroon should be limed at moderate rates to sustain crop productivity”. While examining the seed supply system for maize produc- tion in southwestern Nigeria, [3] observed that “about 39% of farmers used improved varieties for high crop yields, 24% for disease resistance and 22% for market preferences, whereas local varieties were cultivated by 37% of farmers because of market preferences and availability, 16% because of low cost and 12% because of disease resistance”. [4] demonstrated that continuous- flow drip irrigation in Bauchi state of Nigeria delivers “high crop yields especially if the crops are grown under appropriate agronomic practices that enable protraction of the growth sea- son”. [5] demonstrated that high maize yields on sandy soils in Zimbabwe can be achieved by using mineral fertilizers. According to [6], among many oilseed crops (for example, sunflower, soybeans, rapeseed/mustard, sesame, groundnuts, etc) grown in Kenya, oilseed rape is pre- ferred because of its high yields (1.5 tons—4.0 tons / hectare) with high oil content of 42–46%. While comparing three fertigation strategies of grapes in the Berg River Valley region of South Africa, [7] found that “less berry crack contributed to a higher yield and higher export percent- age of grapes”. While analysing the benefits of soil conservation in the Kondoa eroded area of Tanzania by conducting a household survey of 240 households, [8] observed that 56% of the respondents gained high crop yields. [9] investigated limited nitrogen content, a major chal- lenge to sustainable and high crop production, for agricultural soils of lower eastern Kenya. While evaluating small holder farmers’ preferences for climate smart agricultural practices in Tehuledere district, northeastern Ethiopia, [10] found that “high and moderate climate resil- ience and high crop yield agricultural practices had a positive utility”. [11] demonstrated that phosphorus treatment for rice fields in lowlands in the central highlands of Madagascar signif- icantly and consistently accelerated initial production with high crop growth rate and short- ened days to heading. According to [12], “rain fed agriculture has a high crop yield potential if rainfall and soil nutrient input resources are utilized effectively”. But none of these papers discuss the distribution of crop yield or forecasts. The distribu- tions of crop yields is very useful in agribusiness. These distributions can help to tackle food shortages and insecurity by understanding how natural resources and farmers attitude towards crops selection and cultivation can control agricultural productivity, in agricultural policy assessment and to calibrate rates and premiums in crop insurance. Similarly, understanding the trend of crop yield and the insights gained from crop yield predictions can go a long way in helping to address the current global issue of increase in food prices and demand as well as to understand the associated risk of food production by helping farmers to make informed decisions especially on what and where to grow. We are also not aware of any previous research that has focused on predicting crop yield in east Africa let alone doing so in such an almost holistic manner as we have done in this paper; so, to bridge this research gap, we follow [13] to provide some crop yield forecast in some east African countries. We believe that the results herein will be of extreme importance to east Afri- can regional farmers. The aim of this paper is two folded. First, to forecast the crop yield and secondly to identify cash crops that are capable of producing extremely high yield in some east African countries by modelling the tail region of crop yield data. The remainder of this paper contains data in Section 2, methods in Section 3, results and discussion in Section 4 and conclusions in Section 5. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 2 / 36 PLOS ONECrop yield in some east African countries We use two methods for analyzing the data: time series analysis and fit of heavy tailed distri- butions. Time series analysis and forecasting is a branch of statistics. Time series forecasting uses models to predict future outcomes based on past observations. With time series visualiza- tions, trends and seasonal patterns could be identified. We could then seek to gain deeper insight as regards to the reason behind these trends. Several time series models have been devel- oped, studied and widely applied in many fields. Box-Jenkins’ auto-regressive integrated mov- ing average (ARIMA) model [14] arguably stands out among others as the most widely used perhaps due to its simplistic application appeal and high precision in modelling. For instance, [15] used the ARIMA model to forecast rice production, consumption, importation, exporta- tion and self-sufficiency in the Benin Republic. [16] used the ARIMA model to forecast the consumption of some livestock products such as eggs, milk, chicken and cow meat to see if the forecast of consumption was on the increase. [17] highlighted that the past century has wit- nessed significant rise and fall of cocoa production in Nigeria due to diverse institutional and climate changes. They used the ARIMA model to predict cocoa production in Nigeria between 2018 and 2025. Their forecast showed a decreasing trend where cocoa production is expected to fall by more than 20% in 2025 against the 2017 value. [18] used the ARIMA model to forecast maize production in India from 2018 to 2022. The model predicted about 13.76% increase in maize production in India. [19] used the ARIMA model to forecast soybean yield in Zambia. The forecast suggested 23430.3 hectogram / hectare yield increase in 2020 compared to the 2016 figure of 19624 hectogram / hectare. [20] used the ARIMA model to forecast Kharif rice production in West Bengal, India which contributes about 15% of the total paddy in India. [21] used the ARIMA model to forecast sorghum production in South Africa from 2017 to 2020. Their forecast depicted an increasing trend. [22] used the ARIMA model to forecast sugar cane production in Pakistan from 2019 to 2030. Their forecast indicated a significant increase. Quantifying the tail of the crop yield distribution is vital for managing agricultural produc- tion risk and rating crop insurance [23]. The simplest and the most widely used distribution for modelling rare outcomes occurring in the tail region is the power law distribution. Many processes follow the power law over large magnitude of values. Recent examples are the distri- bution of stock returns [24], income [25, 26], wealth of world billionaires [27], persisters-anti- biotic-tolerant cells [28], duration size of unhealthy air pollution events [29], tourism recommendations [30], cumulative coal production [31], agricultural land size [32], rates of wetland loss [18], union size [33], strike size [34] and growth rate of CO2 [35]. Popular alterna- tives to the power law distribution are the lognormal, stretched exponential, and Fre´chet distributions. 2 Data Yearly data from 1961 to 2018 on the yield of cash crops like banana, plantain, beans, cassava, coffee, sorghum, potato, sweet potato, maize, rice, sugar cane, wheat, millet and cotton seed from six countries in east Africa (namely, Burundi, Kenya, Somalia, Tanzania, Uganda and Rwanda) were obtained from Food and Agriculture Organization of United Nations-FAO, see http://www.fao.org/faostat/en/#home. The data obtained were yields aggregated at national levels. The time plots of the crops in different countries are shown in Figs 1 and 2. Some sudden changes, particularly big drops and falls could be seen at different times indicating periods of high and low yields. These changes could be as a result of the global economic outlook, envi- ronmental/climate changes or even changes in farming practices. Some descriptive statistics of the data for crops are presented in Table 1. The statistics include the mean, median, standard deviation, minimum, maximum, skewness and kurtosis. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 3 / 36 PLOS ONECrop yield in some east African countries Fig 1. Time series plots for crop yield in different countries. https://doi.org/10.1371/journal.pone.0287011.g001 The discrepancy between the mean and the median values appears not to be large for almost all the crops across the countries. The mean is larger than the standard deviation for all the crops across the countries. This suggests that the data are underdispersed. Note that underdis- persion could be as a result of serial correlation which is typical of time series data. We can remove serial correlation by random variable transformation. But, this may lead to (a) loss of data information and (b) limits us to specific class of models to use. The data exhibit varying degrees of skewness and kurtosis across crops and countries. The lowest (highest) positive PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 4 / 36 PLOS ONECrop yield in some east African countries Fig 2. Time series plots for crop yield in different countries. https://doi.org/10.1371/journal.pone.0287011.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 5 / 36 PLOS ONECrop yield in some east African countries Table 1. Descriptive measures for the crop yield data sets. Country Crop Min. 1st qu. Median Mean 3rd qu. Max. Std. dev. Skewness Kurtosis Burundi Kenya Somalia Tanzania Uganda Rwanda Banana Beans Cassava Coffee Sorghum Sweet potato Beans Coffee Maize Rice 46915 6044 41867 2687 5890 59048 3127 9212 10713 13076 Sugar Cane 297552 Wheat Banana Maize Sorghum 9212 88430 4149 2040 54433 9067 85880 6961 9837 63302 4783 14918 12957 34802 689809 14918 169641 8173 3320 56947 10036 89890 8251 10000 63837 5556 16839 15813 39621 808548 16839 170374 9758 3522 60700 9599 85078 8100 10426 68215 5413 18293 15507 39988 774305 18293 185025 10065 3992 62984 10380 90894 9379 11881 65848 6122 21078 17266 46510 885805 21078 198616 11765 4280 127352 13184 112378 11598 14042 133015 8382 31991 20712 61813 1211845 31991 317500 17901 9824 12204.970 1375.755 12870.850 1769.862 1886.384 13063.090 1034.896 5003.894 2605.723 10166.130 217545.200 5003.894 46006.370 3029.413 1306.699 Sugar Cane 272727 350000 407143 604201 887500 1455975 346839.200 Maize Millet Rice Cotton Seed Sorghum Sweet potato Banana Cassava Coffee Millet Plantain Sweet potato Beans Cassava Coffee Potato Sorghum Sweet potato 4808 4522 7143 2328 4423 10448 23298 32973 3839 8092 42971 24009 5606 11778 2678 22821 6850 34388 9170 7010 12826 4356 6554 18029 39412 44771 5443 11486 52141 35504 6980 55212 5267 64313 10014 53682 12722 8308 16286 5136 9151 29252 42070 66988 6131 14017 56585 41558 8020 91644 5994 68656 11000 62550 12734 8697 16396 5117 8832 34621 40891 71711 6402 13341 59014 40660 7858 82820 6051 71994 11109 63978 14414 9982 19172 5783 10100 49412 44927 89971 7150 15986 60867 44017 8522 116873 6776 82871 12086 75039 31359 19507 27382 7936 17963 72759 48333 144083 10283 16751 84235 62075 10258 164000 11019 130600 15084 96163 4893.517 2401.514 4933.184 1245.132 2717.268 19284.670 5636.359 32354.340 1408.237 2637.066 11223.650 6398.938 1037.677 37331.23 1473.642 20435.600 1754.313 13269.530 3.0704 -0.8491 -2.0220 -0.5861 -0.3327 3.1150 0.0585 0.7453 0.0261 -0.0934 -0.4537 0.7453 0.7969 0.5619 2.2982 1.0554 1.3077 1.5969 0.2870 0.2257 0.4293 0.5541 -1.3614 0.7664 0.6347 -0.6120 0.8573 0.3221 -0.0566 -0.2683 0.6369 0.2417 0.2655 0.0977 https://doi.org/10.1371/journal.pone.0287011.t001 13.0916 0.9814 4.2142 0.1126 -0.1633 10.0345 0.3111 0.2387 -1.0478 -0.0593 -0.0124 0.2387 0.7965 -0.1600 6.9370 -0.2674 2.9604 5.1457 -0.7743 -0.4752 0.5903 -1.0485 1.613 -0.5467 -0.0043 -0.945 -0.0373 1.5048 -0.8005 -1.0465 1.6200 0.7926 0.0410 -0.4860 skewness of 0.0261 (3.1150) corresponds to maize (sweet potato) in Kenya (Burundi). The low- est (highest) negative skewness of -0.0934 (-2.0220) corresponds to rice (cassava) in Kenya (Burundi). The lowest (highest) positive kurtosis of 0.0410 (13.0916) corresponds to sorghum (banana) in Rwanda (Burundi). The lowest (highest) negative kurtosis of -0.0043 (-1.0478) corresponds to coffee (maize) in Uganda (Kenya). Crop yield skewness has been used to char- acterize crop yield tendencies. [36] reported that crop yield is positively skewed in the presence of independent, identical and uniform resource availability distribution. Crop yield is nega- tively skewed whenever the distributions are Gaussian, i.e. skewness depends on asymmetries in resource availabilities, meaning that a negatively skewed yield occurs whenever production is tightly controlled so that the left tails of some resources availabilities distributions are thin [36]. However, in addition to the observable similarities between the mean and the median PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 6 / 36 PLOS ONECrop yield in some east African countries crop yield values, we notice that for majority of the cases, the skewness and kurtosis values are close to zero, suggesting possible symmetry and mesokurtosis. Figs 3 and 4 show boxplots to support the descriptive statistics in Table 1 and to compare the yield performance of some of the crops that are produced in more than one east African country. We see that Somalia recorded the highest banana and sugar cane yields. Burundi recorded the highest beans, coffee and sweet potato yields. Rwanda recorded the highest cas- sava yield. Kenya recorded the highest rice yield. Tanzania recorded the highest sorghum, maize and millet yields. Also, evident enough in Figs 3 and 4 are the presence of extreme (high Fig 3. Box plots for crop yield in different countries. https://doi.org/10.1371/journal.pone.0287011.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 7 / 36 PLOS ONECrop yield in some east African countries Fig 4. Box plots for crop yield in different countries. https://doi.org/10.1371/journal.pone.0287011.g004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 8 / 36 PLOS ONECrop yield in some east African countries and low) yields for some of the crops which are indicated by observations lying outside of the whiskers in the box plots. The power law distribution discussed later is especially useful for modelling unusually high yields. We tested heavytailedness of the each data set using [37]’s test based on Kolmogorov-Smir- nov statistic corrected for correlation [38]. The p–values of this test for banana, beans, cassava, coffee, sorghum and sweet potato in Burundi were 0.182, 0.0664, 0.151, 0.102, 0.156 and 0.115, respectively. The p–values for the crops in Kenya were 0.162, 0.171, 0.059, 0.120, 0.166 and 0.145. The p–values for the crops in Somalia were 0.167, 0.157, 0.098 and 0.115. The p–values for the crops in Tanzania were 0.168, 0.112, 0.095, 0.068, 0.096 and 0.125. The p–values for the crops in Uganda were 0.177, 0.114, 0.087, 0.171, 0.105 and 0.077. The p–values for the crops in Rwanda were 0.075, 0.169, 0.068, 0.098, 0.061 and 0.158. The p–values reported show that there is no significant evidence against the fact that each data has a heavy tail. Hence, unusually high yields can be modeled by heavy tailed distributions as done in Section 4. 3 Methods 3.1 Time series analysis of crop yields One possible technique for time series analysis is to assume that the overall mean is either con- stantly increasing or constantly decreasing with respect to time. In this case, the fit of a sloping line might be appropriate for the time series. This type of line is typically referred to as a linear trend model or a trend-line model and it is a special case of a simple linear regression model with time index t as the only predictor variable, i.e. t = 1, 2, 3, . . .. The estimated trend line is the line that minimizes the sum of the squared vertical deviations from the data. Trend lines serve as important visual aids. However, they often perform poorly in forecasting beyond the historical data. In practice, majority of the time series data that arise in different areas cannot be described by some straight lines because their trends often undergo evolution. Given the past observations, the trend-line model attempts to find the intercept and slope that give the best average fit to the data. Unfortunately, the deviation of the linear trend model from the data is usually greatest at the end of the time series where the forecasting starts. Therefore, in time series analysis and forecasting, the important question ‘what is the appropriate model?’ can first be addressed by visually inspecting the time series data for any constantly changing trend or randomly changing trend. Based on Figs 1 and 2, we see that assuming a steady upward or downward linear trend for any of the crop yield data is apparently illogical and out of place because a randomly changing trend is overwhelmingly evident for all the time series data. To model the nonlinear trend in all the time series, we may need to regress the time series on second or higher order terms of t and this may require some trial and errors which may possibly lead to some overestimated or underestimated models. To circumvent the issue of model selection, we consider the most reliable models for nonlinear trends in time series and they are referred to as stochastic time-series models. Examples of such models are the one pro- posed by [14] which involve straightforward laid down iterative procedures for model fitting unlike the nonlinear regression method mentioned earlier. In this section, we carry out a time series analysis to study the yield pattern of crops over a specified period of time. We need to isolate first the impact of trends (the overall pattern in the series) and second the impact of random disturbances (the vigorous wiggles in the series). The impact of trends could be due to planting strategies and techniques, advanced mechanized farming, farm management, irrigation, the use of fertilizers and genetically improved seed- lings/crops. The impact of random disturbances could be due to pandemics, crop disease out- breaks, wars, recessions, environmental degradations (for example, erosion) and extreme weather conditions such as droughts and floods. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 9 / 36 PLOS ONECrop yield in some east African countries Let xt denote the observed yield of a crop at time t. Suppose we denote all the observed information up to time t by I t. We are interested in forecasting xt. We can specify the forecast as xtjI t or more specifically as ^xtþhjt. The forecast of xt+h given all previous observations up to time t (x1, x2, . . ., xt) is known as the h–step forecast. The h–step forecasting method can be easily implemented through the famous Box-Jenkins autoregressive integrated moving average (ARIMA) modelling framework. ARIMA models are used for trend analysis and forecasting. The ARIMA (p, d, q) model is defined by � 1 (cid:0) � Xp �iBi i¼1 � ð1 (cid:0) BÞdxt ¼ c þ 1 þ Xq yjBj � xt; j¼1 where ϕ’s are the autoregressive (AR) parts of the model, θ’s are the moving average (MA) parts of the model, d is the order of difference, B is known as the backshift operator, c is a con- stant which is equal to μ(1 − ϕ1 − � � � − ϕp), μ is the mean of the dth differenced series (1 − B)dxt and ξt is white noise. ξt are generally assumed to be independent, identically distributed variables sampled from a normal distribution with zero mean. In ARIMA modelling, we make the following assumptions about the time series: there are no seasonality or cyclical trends, there are no outliers, and that the variation about the mean is consistent. After fitting the ARIMA model, we can check the model adequacy viz-a-viz a popular portmanteau test called Ljung-Box test by simply testing whether the residuals from the fitted model are white noise. For Ljung–Box test, we test the hypothesis H0: ρk = 0 versus H1: ρk 6¼ 0. The test statistic of Ljung-Box test is Q? ¼ nðn þ 2Þ Xh k¼1 ^r2 k ðn (cid:0) kÞ ; where n is the sample size, ^rk is the sample autocorrelation at lag k, and h is the number of lags being tested. Under H0, the statistic Q? is asymptotically chi-square distributed with h degrees of freedom. At α significance level, the critical region for rejecting the hypothesis of random- ness is Q? > w2 1(cid:0) a;h, where w2 1(cid:0) a;h denotes the (1 − α)th quantile of the chi-squared distribution with h degrees of freedom. A detailed discussion of Box-Jenkins ARIMA (p, d, q) model could be read from [39] and [40]. In Figs 1 and 2, we find some evidence of changing variance in some of the series. Each series appears clearly non-stationary as the series wanders up and down. Before proceeding with the data analysis, we ensured that the variance for each series is stabilized by the Box-Cox transformation [41]. The Box Cox transformation involves an exponent, λ 2 [−5, 5]. In this paper, all values of λ are considered but the optimal value for each data is applied. The optimal value of λ is the one that gives the best approximation of the Gaussian distribution. The transformation of xt has the form: 8 >>>>>>>>< >>>>>>>>: xtðlÞ ¼ xl t (cid:0) 1 l ; if l 6¼ 0; ln ðxtÞ; if l ¼ 0: ð1Þ The formula in (1) is not as simple as it appears because testing for all possible values one by one is unnecessarily time consuming. However, most software packages include an option for PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 10 / 36 PLOS ONECrop yield in some east African countries a Box-Cox transformation. In this paper, we used the 0auto:arima0 function in the 0forecast0 package in the R (R Core Team, 2022) software to fit the ARIMA (p, d, q) models. Setting the 0lambda0 argument to 0auto0 allows a transformation to be automatically selected and imple- mented using the Box-Cox method. The routinely transformed data are then coerced into sta- tionarity by implementing first or second order differences whenever there is any need to do so before estimating the appropriate model. Each coerced series was tested for stationarity using [42]’s test. The null hypothesis was that the series is stationary. The p–values for banana, beans, cassava, coffee, sorghum and sweet potato in Burundi were 0.085, 0.085, 0.089, 0.095, 0.075 add 0.083, respectively. The p–values for the crops in Kenya were 0.057, 0.081, 0.051, 0.069, 0.086 and 0.056. The p–values for the crops in Somalia were 0.098, 0.078, 0.089 and 0.083. The p–values for the crops in Tanzania were 0.067, 0.078, 0.082, 0.083, 0.081 and 0.052. The p–values for the crops in Uganda were 0.094, 0.086, 0.095, 0.051, 0.092 and 0.090. The p–values for the crops in Rwanda were 0.098, 0.099, 0.053, 0.073, 0.090 and 0.080. 3.2 Analysis of the maximum crop yields Suppose we denote the crop yield random variable by X with realizations xi, i = 1, 2, . . ., n, where n represents the number of observations. For the convenience of fitting distributions to the available data, we assume that the xi are random. The assumption of independence is not technically correct as the data are actually serially correlated. But ignoring dependence in a data set and treating the data as being independent has no effect on parameter estimates, it only affects standard errors (see, for example, [43]). Hence, the results presented later on the fit of heavy tailed distributions are correct as accuracy of estimation is not taken into account. The probability density functions (PDFs) of the fitted heavy tailed distributions are 1. The power law distribution also known as Pareto distribution of type I [44] specified by the PDF f ðxÞ ¼ �(cid:0) a a (cid:0) 1 xmin � x xmin for x � xmin > 0, where xmin is the lower bound and α > 0 is the exponent. At or above xmin, the distribution exhibits properties of a power law distribution. 2. The lognormal distribution specified by the PDF f ðxÞ ¼ � 1 p exp (cid:0) ffiffiffiffiffiffi 2p bx � ðln x (cid:0) aÞ2 2b2 for x > 0, where −1 < a < 1 and b > 0 are the location and scale parameters, respectively. 3. The stretched exponential distribution specified by the PDF f ðxÞ ¼ � �b(cid:0) 1 x a b a � exp (cid:0) � � �b x a for x > 0, where a > 0 is the scale parameter and b > 0 is the shape parameter. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 11 / 36 PLOS ONECrop yield in some east African countries 4. Fre´chet distribution [45] specified by the PDF � f ðxÞ ¼ babx(cid:0) 1(cid:0) b exp (cid:0) � � �(cid:0) b x a for x > 0, where a > 0 is the scale parameter and b > 0 is the shape parameter. We estimated the parameters of all the distributions by the method of maximum likelihood through the optim routine in R [46]. We estimated xmin in the power law distribution by fol- lowing the method in [47]. That is, we chose xmin that minimized KS ¼ max x�xmin jFnðxÞ (cid:0) ^FðxÞj; where Fn(x) and ^FðxÞ denote, respectively, the empirical and fitted power law distribution functions for x � xmin. We have used the method of maximum likelihood because of its popularity. There are other methods for estimation; in particular, for estimating α of the power law distribution. Some of these estimators include the rank estimator due to [48], [49]’s estimator and the median esti- mator due to [50]. Note that each of the four distributions has two free parameters. So, no one distribution is more flexible than the others in terms of the number of parameters. Unlike the power law dis- tribution, the lognormal, Fre´chet and stretched exponential distributions model the entire data. We can compare their fits by the following goodness-of-fit measures: 1. Bayes information criterion (BIC) due to [51] defined by BIC ¼ (cid:0) 2 ^L þ k ln ðnÞ; 2. Akaike information criterion with a correction (AICc) due to [52] defined by AICc ¼ AIC þ 2kðk þ 1Þ n (cid:0) k (cid:0) 1 ; where ^L and k denote, respectively, the maximized log likelihood value and the number of unknown parameters. We can also compare all of the fitted distributions through the Kolmogorov–Smirnov test. Its statistic is given by KS ¼ max x2Data jFnðxÞ (cid:0) ^FðxÞj; which was corrected as in [38] to account for correlation in the data. The larger the value of the corresponding KS p–value the better the fitted distribution. We require the p–value of the Kolmogorov–Smirnov test to be greater than 0.05 to conclude that the distribution is a reason- able model for the data. A p–value less than 0.05 suggests an absolute rejection of the distribu- tion as a candidate for the data. However, one major drawback of the Kolmogorov–Smirnov p–value is that it depends on fixed parameters, hence it does not reflect sampling variability. We can calculate more conservative p–values by a bootstrapping method in [47]. We imple- mented this method by using 5000 bootstrap replications to obtain the final p–value for the Kolmogorov–Smirnov test. In this paper, we shall use the non-bootstrapped KS p–value to PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 12 / 36 PLOS ONECrop yield in some east African countries verify the plausibility of each distribution as a candidate model for data. We use the boot- strapped KS p–value to discriminate among competing distributions and to generalize our findings. Vuong test [53] can be used to discriminate between two non-nested models by testing the ln Pðxj ^Y2Þ. The test statistic for Voung’s test is L ¼ null hypothesis that the models provide indistinguishable fits for the same data. Suppose we denote the probabilities for models 1 and 2 by Pðxj ^Y1 Þ and Pðxj ^Y2 Þ, respectively, where ^Y1 and ^Y2 denote the parameter estimates for models 1 and 2, respectively. Let d ¼ ln Pðxj ^Y1Þ (cid:0) denote the mean and standard deviation of d, respectively. A large, positive test statistic value provides evidence that model 1 is superior to model 2. A large, negative test statistic value gives evidence that model 2 is superior to model 1. Under the null hypothesis that the models are inseparable, the test statistic Λ is asymptotically standard normal distributed. Two finite sample corrections of Vuong’s test are sometimes considered based on the AIC and BIC pen- alty terms, depending on the complexity of the two models. However, these corrections some- times generate conflicting conclusions. d� , where d� and sd ffiffi n sd p 4 Results and discussion Ljung–Box p–values in Table 2 are > 0.05 suggesting that the residuals of the fitted ARIMA models are not statistically significant from white noise at 0.05 significance level for all the crops except for plantain in Uganda which is not statistically significant from white noise at 0.01 significance level. All of the fitted models are suitable for prediction based on the residual analysis. From the 10 years (2019–2028) point forecast (solid blue lines) of the fitted ARIMA models in Figs 5 to 10, we observe the following for Burundi: an initial sharp drop in 2019 fol- lowed by an upward swing of yield for banana; a sharp increase in 2019 followed by increasing oscillations of yield for sweet potato; the yield for sorghum shows a quick increase from 2019 to 2028; the yield for beans shows an immediate decline from 2019 to 2028; neither cassava nor coffee indicate any increasing or decreasing pattern from 2019 to 2028. In Kenya, we observe the following: the yield for beans shows a continuous decline from 2019 to 2028; nei- ther upward nor downward yield trend is evident for coffee, rice, wheat and sugar cane from 2019 to 2028; the yield of maize shows a sharp drop in 2019 followed by an increase and then a stable trend. In Somalia, we observe the following: the yield for maize or sugar cane does not indicate any pattern; the yield for banana shows an initial moderate increase in 2019 followed by a period of no trend up to 2028; the yield for sorghum first experienced a sharp drop in 2019 followed by a stable period of no trend up to 2028. In Tanzania, we observe the following: no significant trend could be identified for maize, rice, sweet potato and cotton seed for the entire forecast period; millet is characterized by a slight yield decrease in 2019 followed by a period of no significant trend up to 2028. In Uganda, we observe the following: the forecast for banana, cassava, millet, plantain and sweet potato did not show any significant trend from 2019 to 2028; the yield for coffee shows a slight increase in 2019 followed by a period of neither increase nor decrease. In Rwanda, we observe the following: the yield for beans shows a persis- tent decline from 2019 to 2028; the yield for sweet potato shows initial jump followed by a slow decline; coffee indicated an upward trend tendency from 2019 to 2028; cassava, potato and sor- ghum did not indicate any significant trend. The changes observed in Figs 5 to 10 are consistent with findings in the literature. [54] established that both intra- and interseasonal changes in temperature and precipitation influ- ence cereal yields in Tanzania. [55] reported that climate change will reduce mean yields in Africa by 17% for wheat, 5% for maize, 15% for sorghum and 10% for millet. No mean change in yield for rice was detected. Using data from the northern Tanzanian highlands, [56] PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 13 / 36 PLOS ONETable 2. Ljung–Box test statistic (Q?), its degree of freedom and its p–value for the fitted ARIMA models at lag 10 (i.e. h = 10). Crop yield in some east African countries Country Burundi Kenya Somalia Tanzania Uganda Rwanda Crop Banana Beans Cassava Coffee Sorghum Sweet potato Beans Coffee Maize Rice Sugar Cane Wheat Banana Maize Sorghum Sugar Cane Maize Millet Rice Cotton Seed Sorghum Sweet potato Banana Cassava Coffee Millet Plantain Sweet potato Beans Cassava Coffee Potato Sorghum Sweet potato Q? Fitted ARIMA (p, d, q) a ARIMA(2,1,2) ARIMA(0,1,1) with drift ARIMA(0,1,0) ARIMA(0,1,1) ARIMA(1,0,0) with non-zero mean ARIMA(2,1,1) ARIMA(1,0,0) ARIMA(0,1,1) ARIMA(0,1,3) ARIMA(0,1,0) ARIMA(0,1,0) ARIMA(0,1,1) ARIMA(0,0,4) with non-zero mean ARIMA(0,1,0) ARIMA(1,1,1) ARIMA(0,1,0) ARIMA(1,1,1) ARIMA(1,0,0) with non-zero mean ARIMA(0,1,1) ARIMA(0,1,1) ARIMA(2,1,0) ARIMA(0,1,1) ARIMA(0,1,2) ARIMA(0,1,0) ARIMA(1,0,0) with non-zero mean ARIMA(0,1,1) ARIMA(0,1,0) ARIMA(0,1,0) ARIMA(1,0,0) with non-zero mean ARIMA(1,1,0) ARIMA(1,0,0) with non-zero mean ARIMA(0,1,2) ARIMA(1,0,0) with non-zero mean ARIMA(2,0,2) with non-zero mean 1.5746 5.9108 12.0820 8.3159 8.9597 7.8161 12.0780 13.2300 3.4473 10.4230 17.9800 13.9030 4.1658 15.1300 3.8376 13.8200 11.8720 6.2974 7.7560 18.8250 4.2521 12.8150 3.8909 12.1600 15.3660 6.1006 19.0490 17.7340 7.9098 2.3680 3.5770 4.4913 4.4304 2.4531 df 6 8 10 9 8 7 8 9 7 10 10 9 5 10 8 10 8 8 9 9 8 9 8 10 8 9 10 10 8 9 8 8 8 5 p–value 0.9544 0.6572 0.2796 0.5027 0.3457 0.3491 0.1478 0.1525 0.8408 0.4042 0.0553 0.1258 0.5258 0.1274 0.8715 0.1814 0.1570 0.6140 0.5589 0.0267 0.8337 0.1711 0.8668 0.2745 0.0524 0.7298 0.0397 0.0596 0.4423 0.9842 0.8931 0.8103 0.8164 0.7835 aDue to space constraints, we omit the coefficients of the fitted ARIMA models; interested readers can obtain them from the authors upon reasonable request. https://doi.org/10.1371/journal.pone.0287011.t002 demonstrated that increasing night time temperature is the most significant climatic variable responsible for diminishing coffea arabica yields between 1961 and 2012. According to [57], annual food crops in the Kilimanjaro region of Tanzania were particularly sensitive to the drought and maize and beans yields were lower than perennial crops during the years of drought. Through a simulation study, [58] predicted climate change in east Africa and found its negative impact on crop production in that region. They projected that the crop output decrease will lie between 1.2% and 4.5%. [59] identified soil erosion by water as one of the major causes of land degradation and dwindling agricultural produce in Africa resulting in an estimated yearly crop yield loss of about 280 million tons. [60] provided evidence to suggest PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 14 / 36 PLOS ONECrop yield in some east African countries Fig 5. Time series plots and 10 years yield forecast with fitted ARIMA models showing 80% and 95% confidence bands for crops in Burundi. https://doi.org/10.1371/journal.pone.0287011.g005 that climate change severely impacted rice production in Rwanda. [61] produced evidence to suggest that temperature increases lead to decline in maize and cassava crops for Tanzania, Malawi, Zambia and South Africa. [62] observed that the yields for maize, sorghum or millet fluctuated at a decreasing trend in the Kongwa district of Tanzania. According to [63], increased temperatures in Kenya due to climate change have a general tendency to reduce rice yields. [64] showed that the impacts of projected changes in climate on maize production areas are the reduction in the suitability of the crop, especially around central and western Tanzania, mid-northern and western Uganda, and parts of western Kenya by 20–40%, and patches of east Africa will experience a reduction as high as 40–60%, especially in northern Uganda, and western Kenya. According to [65], maize production in southern highlands of Tanzania has decreased during the past two decades, since the year 2000. According to [66], climate change has induced a devastating effect on agricultural production in Somalia leading to crop yield to decline including sorghum. Tables 3 and 4 give the BIC, AICc and the KS p–values of the fitted distributions. The BIC and AICc values for the power law distribution are smaller than those for the remaining distri- butions. The KS p–value > 0.05 in all the cases except for Millet in Uganda indicating that the PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 15 / 36 PLOS ONECrop yield in some east African countries Fig 6. Time series plots along with 10 years yield forecast for the fitted ARIMA models showing 80% and 95% prediction confidence bands for crops in Kenya. https://doi.org/10.1371/journal.pone.0287011.g006 power law is not a plausible distribution in this case. We cannot compare the values of the goodness-of-fit measures of the power law distribution with those of the other distributions because the power law distribution fits only the tails whereas the lognormal, Fre´chet, and the stretched exponential distributions fit the entire data. Thus, we can only compare the BIC and AICc values of the lognormal, Fre´chet, and the stretched exponential distributions. Based on the KS p–value, we can observe that the lognormal distribution could be a plausible distribu- tion for banana and coffee in Burundi, all the crops in Kenya except for sugar cane, maize and sorghum in Somalia, all the crops in Tanzania, all but banana in Uganda and all but cassava in Rwanda. Fre´chet distribution appears to be a plausible distribution for banana in Burundi, maize and wheat in Kenya, maize and sorghum in Somalia, all except for maize and sorghum in Tanzania and cassava, coffee and plantain in Uganda and all except for cassava and potato in Rwanda. The stretched exponential distribution appears to be a plausible distribution for beans and coffee in Burundi, all the crops in Kenya, maize in Somalia, all the crops in Tanza- nia, all except for plantain in Uganda and all the crops in Rwanda. Based on the AICc and BIC values in Tables 3 and 4, we can see that none of the three dis- tributions that model the entire data (i.e. the lognormal, Fre´chet and the stretched exponential PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 16 / 36 PLOS ONECrop yield in some east African countries Fig 7. Time series plots along with 10 years yield forecast for the fitted ARIMA models showing 80% and 95% prediction confidence bands for crops in Somalia. https://doi.org/10.1371/journal.pone.0287011.g007 distributions) consistently provide the best fit. None of them consistently gave the smallest AICc or smallest BIC values across the countries. The bootstrapped KS p–values in Table 5 indicate that the power law distribution is a plausible model for all the crop yield data. In gen- eral, the distribution with the smallest AICc and smallest BIC values corresponds to the distri- bution with the largest bootstrapped KS p–values. Fitting of such distributions to the tail of the data can be compared with that of the power law distribution by using Vuong’s test. The results of this comparison are presented in Table 6. We can observe that the stretched expo- nential distribution emerges as the best model for millet in Uganda and the power law distri- bution emerges as the best model for the rest of the crops except for a few cases where the winner is undecided. For instance, for sorghum in Burundi (power law and lognormal), sweet potato in Burundi (power law and Fre´chet) and for banana in Somalia (power law and lognor- mal). The log-log plots of the fitted distributions superimposed with the empirical distribu- tions are displayed in Figs 11 to 16. We can see that the power law distribution fits all the crop yield data well across the countries. Since the power law model appears to be a plausible distribution for virtually all the crops across countries, we present the estimate for the parameters of the distribution in Table 7. We see that the power law mechanism may occur at varying degrees depending on the type of crop and country. See the ntail values for crops in Table 7, where ntail denotes the total number of observations equal to or above the threshold value xmin, i.e. the total number of data points fol- lowing the power law distribution. The occurrence of such extremely high crop yield definitely has positive impact on farmers and food security. In this case, farmers can make huge profits. Crop yield risk insurance policies for such crops can attract relatively lower premium rates compared to crops with lower yields. The α value of the fitted power law model describes the heaviness of the tail distribution corresponding to extremely high crop yield events with yield > xmin. According to Table 7, the estimates of α are all > 2 indicating that the data in the right PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 17 / 36 PLOS ONECrop yield in some east African countries Fig 8. Time series plots along with 10 years yield forecast for the fitted ARIMA models showing 80% and 95% prediction confidence bands for crops in Tanzania. https://doi.org/10.1371/journal.pone.0287011.g008 tail of the distribution show significant high inequality (i.e. large crop yield). However, there are two special cases satisfying 2 < α � 3 specifically in Somalia for sugar cane and Tanzania for sweet potato. In these cases, the variance and higher-order moments for the crop yields are infinite regardless of whether their mean yield exists or not. Hence, the classical central limit theorem does not hold for these yield data. The consequence of the infinite variance and higher order moments is that empirical estimates of the means converge very slowly due to the regular occurrence of extremely large crop yield values. These characteristics suggest that crop harvest with extremely large yield could sometimes occur for sugar cane in Somalia and sweet potato in Tanzania. Such events could often be of great importance to the farmers and other investors in agribusiness. This behavior is referred to as the black swan mechanism (see [67]). The black swan mechanism describes events coming as a surprise. It has a major effect (posi- tive or negative) and is often inappropriately rationalized. Farmers can have the tendency to break even and even enjoy lower crop yield risk insurance policies in Somalia and Tanzania if they invest in sugar cane and sweet potato, respectively, due to their potential for extremely high yield. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 18 / 36 PLOS ONECrop yield in some east African countries Fig 9. Time series plots along with 10 years yield forecast for the fitted ARIMA models showing 80% and 95% prediction confidence bands for crops in Uganda. https://doi.org/10.1371/journal.pone.0287011.g009 All the estimated α values for the power law distribution in Table 7 are > 3 except for sugar cane in Somalia and sweet potato in Tanzania. This indicates that the sample means for these crops are Gaussian distributed and that their variances are finite. Hence, the standard central limit theorem applies for these crop yield data. The finite mean and variance and the observed evidence of underdispersion in Table 1 suggest that east African regional food security does not seem to be extremely volatile as regular crop yields for these crops tend to cluster around the mean crop yield. Ignoring the impacts of climate and environment, soil structures and compositions/nutri- ents, crop species, mechanization and technology, etc on crop yields, the observed black swan behaviour for the yields of sugar cane in Somalia and sweet potato in Tanzania could be explained by the so called “rich getting richer” principle or the “preferential attachment” princi- ple. Based on these principles, these two crops have potentials for extremely high yield perhaps because of either high demand (so every farmer tends to make them their choice crops for cul- tivation) or common practice such as irrigation adopted by all the farmers being capable of increasing crop yield [36]. So, speaking of crop harvest, yield could follow the pattern of the PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 19 / 36 PLOS ONECrop yield in some east African countries Fig 10. Time series plots along with 10 years yield forecast for the fitted ARIMA models showing 80% and 95% prediction confidence bands for crops in Rwanda. https://doi.org/10.1371/journal.pone.0287011.g010 rich getting richer or the preferential attachment principle. The extremely high yields for sugar cane in Somalia and sweet potato in Tanzania are not just a little bit higher than the normal yield for the same or different crops in the same or other countries. Instead they are so much higher that they cause their distributions to skew significantly. 5 Conclusions We have analyzed the trend and tail of some yearly crop yield data such as banana, plantain, beans, cassava, coffee, sorghum, potato, sweet potato, maize, rice, sugar cane, wheat, millet and cotton seed from 1961 to 2018 in six east African countries: Burundi, Kenya, Somalia, Tanza- nia, Uganda and Rwanda. An exploratory analysis of the crop yield data reveals three structural patterns in each of the series. They are: increasing, decreasing and stagnant trends. Ten years (2019–2028) time series point forecast based on the fitted ARIMA models shows that majority of the crops will experience stagnant yield in different countries with only sorghum and coffee showing the tendency for significant and persistent upward trend in Burundi and Rwanda, PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 20 / 36 PLOS ONECrop yield in some east African countries Table 3. AIC, BIC, AICc and KS p–values (as defined in Section 3) for the fitted distributions. Country—Crop Power law BIC AICc Burundi—Banana 723.717 Beans 519.823 Cassava 585.320 Coffee 307.680 Sorghum 78.290 Sweet potato 134.892 Kenya—Beans 493.325 Coffee 308.382 Maize 352.530 Rice 785.164 Sugar Cane 771.239 Wheat 418.665 Somalia—Banana 306.400 Maize 577.876 Sorghum 299.456 719.814 515.920 581.417 303.777 74.387 130.989 489.422 304.479 348.627 781.262 767.336 414.762 302.498 573.973 295.554 Sugar Cane 1561.743 1557.841 Tanzania—Maize 602.620 598.718 Millet 357.070 Rice 645.884 Cotton Seed 480.611 Sorghum 264.040 Sweet potato 823.410 Uganda—Banana 269.292 Cassava 688.846 Coffee 614.601 Millet 700.174 Plantain 884.580 Sweet potato 618.508 353.168 641.981 476.708 260.138 819.507 265.389 684.944 610.698 696.271 880.678 614.605 p–value 0.954 0.532 0.132 0.931 0.930 0.901 0.504 0.721 0.756 0.303 0.944 0.835 0.889 0.811 0.954 0.252 0.632 0.922 0.532 0.332 0.834 0.200 0.909 0.518 0.833 0.003 0.667 0.654 https://doi.org/10.1371/journal.pone.0287011.t003 Lognormal BIC AICc 1239.032 1235.129 1019.932 1016.029 1293.669 1289.766 1054.181 1050.278 1053.391 1049.489 1243.862 1239.959 980.215 976.313 1043.274 1039.371 1084.795 1080.893 1251.779 1247.876 1613.395 1609.492 1153.186 1149.283 1412.801 1408.899 1097.639 1093.737 978.390 974.487 1621.627 1617.724 1146.010 1142.107 1061.506 1057.604 1157.416 1153.513 999.445 995.543 1088.711 1084.809 1303.455 1299.553 1187.916 1184.013 1361.801 1357.898 1007.740 1003.837 1094.174 1090.271 1245.936 1242.033 1189.226 1185.324 p–value 0.051 0.004 0.000 0.112 0.001 0.000 0.223 0.109 0.332 0.121 0.007 0.821 0.000 0.943 0.131 0.009 0.100 0.966 0.802 0.854 0.211 0.661 0.005 0.535 0.907 0.074 0.138 0.087 Fre´chet BIC AICc 1214.761 1210.859 1048.091 1044.188 1337.395 1333.492 1091.141 1087.238 1076.397 1072.495 1184.906 1181.003 999.376 995.474 1390.577 1386.674 1093.394 1089.491 1282.416 1278.513 1639.809 1635.906 1160.712 1156.810 1430.688 1426.786 1109.941 1106.038 976.162 972.259 1608.998 1605.095 1157.454 1153.551 1066.259 1062.356 1169.649 1165.746 1016.260 1012.358 1100.409 1096.506 1307.969 1304.066 1224.341 1220.438 1361.721 1357.818 1013.23 1009.328 1113.196 1109.294 1242.476 1238.573 1209.564 1205.662 p–value 0.629 0.001 0.000 0.001 0.000 0.000 0.010 0.000 0.112 0.012 0.000 0.712 0.000 0.423 0.203 0.019 0.005 0.533 0.242 0.166 0.023 0.183 0.000 0.515 0.516 0.008 0.384 0.012 Stretched exponential BIC AICc 1282.202 1278.299 1005.110 1001.208 1250.241 1246.338 1036.031 1032.129 1045.201 1041.298 1290.785 1286.882 979.803 975.900 1038.009 1034.106 1085.083 1081.180 1242.221 1238.318 1595.751 1591.848 1162.411 1158.508 1421.742 1417.839 1102.289 1098.386 1006.621 1002.718 1636.254 1632.351 1155.097 1151.194 1080.704 1076.801 1157.795 1153.893 999.811 995.908 1089.489 1085.586 1305.698 1301.795 1158.594 1154.691 1369.707 1365.804 1017.461 1013.558 1079.048 1075.145 1261.018 1257.115 1196.161 1192.258 p–value 0.000 0.092 0.000 0.645 0.000 0.000 0.432 0.323 0.532 0.231 0.321 0.402 0.000 0.434 0.015 0.003 0.103 0.121 0.499 0.664 0.512 0.402 0.443 0.323 0.420 0.187 0.004 0.065 respectively, while beans indicates significant and persistent yield decrease in Burundi, Kenya and Rwanda. We used the power law, lognormal, Fre´chet and stretched exponential distributions to describe high yields in all the crops across the countries. Based on Vuong’s test, we observed that the stretched exponential distribution gave the best fit for millet in Uganda while the Table 4. Continuation of Table 3. Country—Crop Power law BIC AICc Rwanda—Beans 171.782 167.879 Cassava 152.243 148.340 Coffee 578.925 575.022 Potato 945.602 941.699 Sorghum 617.552 613.650 Sweet potato 278.219 274.317 Lognormal BIC AICc 978.580 974.677 1410.743 1406.840 1017.965 1014.062 1332.154 1328.251 1038.192 1034.289 1274.881 1270.979 p–value 0.524 0.019 0.332 0.051 0.543 0.675 Fre´chet BIC AICc 989.914 986.011 1433.857 1429.954 1040.145 1036.243 1363.027 1359.124 1054.371 1050.468 1292.205 1288.303 p–value 0.165 0.004 0.054 0.001 0.223 0.263 Stretched exponential BIC AICc 978.688 974.785 1392.647 1388.744 1022.827 1018.925 1324.835 1320.932 1043.405 1039.502 1274.463 1270.560 p–value 0.642 0.176 0.105 0.142 0.091 0.243 p–value 0.935 0.909 0.903 0.810 0.732 0.983 https://doi.org/10.1371/journal.pone.0287011.t004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 21 / 36 PLOS ONETable 5. Bootstrap KS-test p–values (as defined in Section 3) for the fitted distributions. Crop yield in some east African countries Country Burundi Kenya Somalia Tanzania Uganda Rwanda Crop Banana Beans Cassava Coffee Sorghum Sweet potato Beans Coffee Maize Rice Sugar Cane Wheat Banana Maize Sorghum Sugar Cane Maize Millet Rice Cotton Seed Sorghum Sweet potato Banana Cassava Coffee Millet Plantain Sweet potato Beans Cassava Coffee Potato Sorghum Sweet potato https://doi.org/10.1371/journal.pone.0287011.t005 Power law Lognormal Fre´chet Stretched exponential 0.966 0.902 0.832 0.943 0.952 0.903 0.913 0.928 0.929 0.805 0.945 0.905 0.923 0.933 0.955 0.903 0.956 0.931 0.902 0.933 0.939 0.804 0.965 0.901 0.955 0.201 0.932 0.909 0.958 0.987 0.943 0.924 0.932 0.949 0.732 0.424 0 0.831 0.305 0 0.902 0.821 0.931 0.821 0.611 0.962 0.016 0.942 0.813 0.523 0.787 0.962 0.951 0.971 0.827 0.965 0.514 0.961 0.980 0.612 0.831 0.810 0.951 0.612 0.902 0.732 0.933 0.910 0.925 0.112 0 0.322 0.101 0.221 0.591 0 0.820 0.461 0.132 0.901 0.012 0.912 0.828 0.631 0.511 0.936 0.921 0.822 0.609 0.919 0.055 0.919 0.911 0.423 0.925 0.424 0.901 0.301 0.613 0.321 0.841 0.906 0.094 0.732 0.021 0.941 0.324 0 0.944 0.922 0.954 0.844 0.940 0.922 0 0 0 0 0.723 0.902 0.952 0.949 0.906 0.915 0.931 0.924 0.923 0.919 0.506 0.734 0.911 0.812 0.832 0.815 0.822 0.931 power law distribution gave the best fit for the other crops except for a few undecided cases. The log-log plots were used to visually inspect the performance of the fitted distributions. The power law distribution appeared to fit the upper tail of all the crop yield data better than the other distributions in all the countries. Based on the estimated α value of the fitted power law model, we found potential for extremely high yield in sugar cane in Somalia and sweet potato in Tanzania indicating the inappropriateness of the Gaussian distribution for describing these crop yields. Other crops in Burundi, Kenya, Somalia, Tanzania, Uganda and Rwanda can pro- duce only high but not extremely high yields. Though the time series point forecasts for major- ity of the crops show yield stagnancy with a few exceptions, the evidence from the power law analysis indicates the potential for high yield for all the crops and provides specific calibrations for the yield of all the crops in terms of what quantity of yield is considered high. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 22 / 36 PLOS ONETable 6. Vuong test statistic (Λ) and its p–value for comparing the upper tail (i.e. x > xmin) of the fitted power law distribution and the best among the rest of the competing distributions. Crop yield in some east African countries Country Burundi Kenya Somalia Tanzania Uganda Rwanda Crop Banana Beans Cassava Coffee Sorghum Sweet potato Beans Coffee Maize Rice Sugar Cane Wheat Banana Maize Sorghum Sugar Cane Maize Millet Rice Cotton Seed Sorghum Sweet potato Banana Cassava Coffee Millet Plantain Sweet potato Beans Cassava Coffee Potato Sorghum Sweet potato Contest Statistic (Λ) Power law vs Fre´chet Power law vs Stretched exponential Power law vs Stretched exponential Power law vs Stretched exponential Power law vs Lognormal Power law vs Fre´chet Power law vs Stretched exponential Power law vs Stretched exponential Power law vs Stretched exponential Power law vs Stretched exponential Power law vs Stretched exponential Power law vs Lognormal Power law vs Lognormal Power law vs Lognormal Power law vs Fre´chet Power law vs Fre´chet Power law vs Lognormal Power law vs Lognormal Power law vs Stretched exponential Power law vs Lognormal Power law vs Stretched exponential Power law vs Lognormal Power law vs stretched exponential Power law vs Lognormal Power law vs Lognormal Power law vs Stretched exponential Power law vs Fre´chet Power law vs Lognormal Power law vs Stretched exponential Power law vs Stretched exponential Power law vs Lognormal Power law vs Stretched exponential Power law vs Lognormal Power law vs Lognormal 2.6671 6.7366 10.4399 4.1021 1.2454 0.7442 3.9727 3.5192 2.8513 4.4303 4.3048 2.5905 1.7339 2.6642 1.9648 4.5875 5.4006 3.8995 2.8193 2.8774 2.8893 2.3193 2.9194 2.4766 2.7274 −2.8155 3.1373 3.9022 3.5970 2.9573 4.3420 5.4570 4.1122 2.6648 p–value 0.0077 0 0 0 0.2130 0.4568 1.0×10−4 4.0×10−4 0.0044 0 0 0.0096 0.0829 0.0077 0.0494 0 0 1.0×10−4 0.0048 0.0040 0.0039 0.0204 0.0035 0.0133 0.0064 0.0049 0.0017 1.0×10−4 3.0×10−4 0.0031 0 0 0 0.0077 Winner Power law Power law Power law Power law Undecided Undecided Power law Power law Power law Power law Power law Power law Undecided Power law Power law Power law Power law Power law Power law Power law Power law Power law Power law Power law Power law Stretched exponential Power law Power law Power law Power law Power law Power law Power law Power law https://doi.org/10.1371/journal.pone.0287011.t006 We characterize the evidence for extremely high yield for sugar cane and sweet potato in Somalia and Tanzania, respectively, as black swan where the “rich getting richer” or the “prefer- ential attachment” could be the underlying generating process, meaning that either the two crops are increasingly at lower risk of climate change and environmental challenges such as being drought resistant or farmers are constantly doing many things right (such as adopting favorable planting strategies, large crop areas, etc) as far as the cultivation of the two crops are concerned in the two countries. ARIMA(0,1,1) was used to model and predict coffee in Burundi and Kenya; beans in Burundi; wheat in Kenya; rice, cotton seed, and sweet potato in Tanzania; millet in Uganda. ARIMA(2,1,0) was used to model and predict sorghum in Tanzania. ARIMA(2,1,2) was used PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 23 / 36 PLOS ONECrop yield in some east African countries Fig 11. Log-log plots for crops yield in Burundi where the red line corresponds to the value of xmin in Table 7. https://doi.org/10.1371/journal.pone.0287011.g011 to model and predict banana in Burundi. ARIMA(0,1,2) was used to model and predict banana in Uganda and potato in Rwanda. ARIMA(0,1,0) was used to model and predict cassava in Burundi and Uganda; sugarcane in Kenya and Somalia; rice in Kenya; maize in Somalia; plan- tain and sweet potato in Uganda. ARIMA(1,0,0) was used to model and predict Sorghum in Burundi and Rwanda; beans in Kenya and Rwanda; millet in Tanzania; coffee in Uganda. ARIMA(2,1,1) was used to model and predict sweet potato in Burundi. ARIMA(0,1,3) was used to model and predict maize in Kenya. ARIMA(0,0,4) was used to model and predict PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 24 / 36 PLOS ONECrop yield in some east African countries Fig 12. Log-log plots for crops yield in Kenya where the red line corresponds to the value of xmin in Table 7. https://doi.org/10.1371/journal.pone.0287011.g012 banana in Somalia. ARIMA(1,1,1) was used to model and predict sorghum in Somalia; maize in Tanzania. ARIMA(2,0,2) was used to model and predict sweet potato in Rwanda. The yield forecast in Burundi shows an initial quick decline in 2019 followed by an increase for banana; a sharp increase in 2019 followed by an increase for sweet potato; sorghum shows a quick increase from 2019 to 2028; beans shows a sharp decrease from 2019 to 2028; neither cassava nor coffee show any tendency to increase or decrease from 2019 to 2028. The forecast of the crop yield in Kenya indicates continuous decline of beans yield from 2019 to 2028; no PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 25 / 36 PLOS ONECrop yield in some east African countries Fig 13. Log-log plots for crops yield in Somalia where the red line corresponds to the value of xmin in Table 7. https://doi.org/10.1371/journal.pone.0287011.g013 decrease or increase pattern in yield is evident for coffee, rice, wheat and sugar cane from 2019 to 2028; maize shows a sharp decline in 2019 with an immediate increase followed by a stable trend. In Somalia, the yield forecast for maize and sugar cane does not indicate any pattern; banana shows an initial moderate increase in 2019 followed by the lack of pattern until 2028; sorghum experienced a sharp drop in 2019 followed by a period of no trend up to 2028. The yield forecast in Tanzania indicates no significant trend for maize, rice, sweet potato and cot- ton seed for the whole forecast period; millet is slightly decreased in 2019 and remained stag- nant until 2028. The yield forecast in Uganda indicates that banana, cassava, millet, plantain and sweet potato did not show any significant pattern from 2019 to 2028; coffee shows a slight increase in 2019 followed by a period of no change in yield. The yield forecast in Rwanda indi- cates that beans persistently decreased from 2019 to 2028; sweet potato shows initial increase followed by a slow decrease; coffee indicated an upward trend from 2019 to 2028; cassava, potato and sorghum did not show any significant pattern. In our discussion in Section 4, we saw how the literature points in the direction of climate change as the major cause of the observed yield stagnancy and decline. On this backdrop, we PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 26 / 36 PLOS ONECrop yield in some east African countries Fig 14. Log-log plots for crops yield in Tanzania where the red line corresponds to the value of xmin in Table 7. https://doi.org/10.1371/journal.pone.0287011.g014 suggest that a promising future in favour of high crop yield could await east Africa if urgent changes or improvements on the cropping systems and infrastructures that currently exist in east Africa could be made in order to meet up with the inevitable future demand of agricultural produce due to the increasing population and the challenge of negative impacts of climate change. Science and technology could be useful in showing how agricultural production can be significantly improved in east Africa. For instance, the construction of irrigation systems and rainwater harvesting structures could help cushion the impact of climate change. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 27 / 36 PLOS ONECrop yield in some east African countries Fig 15. Log-log plots for crops yield in Uganda where the red line corresponds to the value of xmin in Table 7. https://doi.org/10.1371/journal.pone.0287011.g015 Further, various climate adaptation/smart strategies could be adapted to increase yields in east Africa. According to [68], short-duration pigeon pea varieties developed by the Interna- tional Crops Research Institute for Semi-Arid Tropics and the Kenya Agricultural Research Institute can give high yields and escape drought, but require non-traditional management practices (for example, sole-cropping, spraying against insect pests). According to [69], NER- ICA, a new rice for Africa, has shown high potential to revolutionize rice farming, producing high yield with minimum inputs in stress-afflicted ecologies. [70] observed that cassava mosaic PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 28 / 36 PLOS ONECrop yield in some east African countries Fig 16. Log-log plots for crops yield in Rwanda where the red line corresponds to the value of xmin in Table 7. https://doi.org/10.1371/journal.pone.0287011.g016 disease (CMD) resistant cassava varieties released in western Kenya and Uganda yielded up to three times more than local varieties. [71] demonstrated that high yields of maize were recorded from certain varieties (Pwani Hybrid 4-PH4, Coast Composite Maize-CCM and the local check-Mdzihana) but they usually required relatively high rainfall amounts in order for them to produce better yields. [72] showed that increased knowledge of varieties, environment and management factors can double total yield of maize, sorghum, millet and groundnut from 1.67 to 3.29 tons per hectare from the average 5.1 hectares that farmers usually crop in south PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 29 / 36 PLOS ONETable 7. Parameter estimates for the power law distribution for all the crop yield data sets (xmin and α are parameters of the power distribution; αse is the standard error corresponding to α; ntail is the number of data exceeding xmin). Crop yield in some east African countries Country Burundi Kenya Somalia Tanzania Uganda Rwanda Crop Banana Beans Cassava Coffee Sorghum Sweet potato Beans Coffee Maize Rice Sugar Cane Wheat Banana Maize Sorghum Sugar Cane Maize Millet Rice Cotton Seed Sorghum Sweet potato Banana Cassava Coffee Millet Plantain Sweet potato Beans Cassava Coffee Potato Sorghum Sweet potato n ntail 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 58 35 35 35 35 5 6 32 20 21 39 30 22 13 32 18 57 33 21 34 30 15 37 16 30 37 39 43 33 12 7 36 44 37 14 xmin 55455 9888 89591 9180 12964 89127 5429 6669 16922 37568 806694 19359 222222 9544 4143 300000 12427 9530 15583 5132 10133 25323 44748 66988 5873 12496 53474 40960 8914 120265 5817 64088 10688 75333 α αse 7.4026 19.9503 52.1287 15.9219 33.5816 8.1898 9.4680 11.9168 14.5952 6.7330 8.5033 6.5818 8.2251 5.4249 5.3367 2.7445 6.0445 8.3193 5.4751 7.2276 7.3606 2.9081 36.7026 4.0035 6.3153 6.6790 7.3752 12.6778 28.4564 13.0159 7.6296 6.0225 9.6451 15.2064 1.0822 3.2032 8.7685 3.3366 14.5710 2.9352 1.4969 2.4411 2.9667 0.9180 1.3699 1.1901 2.0039 0.7822 1.0222 0.2311 0.8781 1.5972 0.7675 1.1370 1.6423 0.3137 8.9257 0.5484 0.8738 0.9094 0.9722 2.0328 7.9260 4.5416 1.1049 0.7572 1.4212 3.7968 https://doi.org/10.1371/journal.pone.0287011.t007 east Zimbabwe. [73] showed that improved maize varieties outyielded the traditional control variety by 26–46% across sites and season in central Mozambique. [74] showed that the use of organic soil management practices such as reduced tillage, mulching and leguminous crops in the northern part of Tanzania increased the production of food crops from an average of 0.5 ton per hectare to 1.5 ton per hectare; subsequently, maize yields increased from 12,000 kilo- gram to 20,000 kilogram per 4.8 hectares. [75] suggested that relaxing liquidity constraints could help to encourage farmers’ adaptation through the implementation of soil, water and land management strategies; thereby, positioning east Africa for food sufficiency in the face of the current global food crisis. [76] noted that intensive manuring with a combination of green and poultry manure produced high yields of maize in central Uganda that were comparable to PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 30 / 36 PLOS ONECrop yield in some east African countries those with mineral fertilizers. [77] demonstrated that households in Kenya adapting to climate change and climate variability through uptake of technologies such as early planting, use of improved crop varieties, and crop diversification produced 4877 kilograms of maize yield equivalent / hectare per year against 3238 kilograms of maize yield equivalent / hectare per year for households that did not adapt (a 33.6% difference between the two groups). [78] found that fertilizer application in the intercropping system is eastern and southern Africa improved cereal yields by 71–282% and pigeon pea yields by 32–449%, increased benefit-cost ratios by 10–40%, and reduced variability in cereal yields by 40–56% and pigeon pea yields by 5–52% compared with unfertilized intercrops. [79] showed that drought resistant climate- smart maize hybrids in Kenya increased yields 33 to 54% relative to conventional hybrids. According to [80], climate adaptation strategies in the central highlands of Kenya included the use of fertilizer and manure in combination (71%), terracing (66%), and crop rotation (60%). [81] showed that climate-smart adaptation practices significantly enhanced wheat yield by 34.35% in southern Ethiopia. [82] showed that use of mulching and permanent planting basin dimensions on maize in western Uganda relatively increased yield by 11–66% and water use efficiency by 33–94% compared to conventional practices. The findings in this paper underscore the importance of using climate-smart agricultural alternatives to improve resilience farming system and the livelihood of subsistence farmers due to the impact of climate change in east Africa. Currently, crop yield for majority of the crops in different countries has been confirmed to neither increase nor decrease with only few crops experiencing all time increase or decrease in yield. Urgent attention should be paid to beans production in the affected countries in order to reverse the persistent down- ward trend of its yield. This paper brings good news of hope for crop yield increase in east Africa if adaptive farming methods and strategies are adequately harnessed in the region in the face of climate and environmental challenges and rising global demand for agricultural produce. The data from 1961 to 2018 consist of only 58 observations. Hence, the results and forecasts in this paper should be treated conservatively. A future work is to see if more frequent and more up-to-date data are available. Another is to consider multivariate modelling of yield by considering country and crop. The disadvantage of the length of the observed series can be interpolated by explaining the common factor for each country and crop. Supporting information S1 Data. (CSV) S2 Data. (CSV) S3 Data. (CSV) S4 Data. (CSV) S5 Data. (CSV) S6 Data. (CSV) PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 31 / 36 PLOS ONECrop yield in some east African countries S7 Data. (CSV) S8 Data. (CSV) S1 File. (TXT) Acknowledgments The authors would like to thank the Editor and the two referees for careful reading and com- ments which greatly improved the paper. Author Contributions Formal analysis: Idika E. Okorie, Emmanuel Afuecheta, Saralees Nadarajah. Methodology: Idika E. Okorie, Emmanuel Afuecheta, Saralees Nadarajah. Resources: Emmanuel Afuecheta. Software: Idika E. Okorie, Saralees Nadarajah. References 1. Buerkert A, Hiernaux P. Nutrients in the west African Sudano-Sahelian zone: Losses, transfers and role of external inputs. Zeitschrift fuer Pflanzenernaehrung und Bodenkunde. 1988; 161: 365–383. https:// doi.org/10.1002/jpln.1998.3581610405 2. Nair KPP, Ngachie V, Nzetchoung F. The importance of liming acid soils of the Sahel regions of Africa to sustain high crop productivity—A case-study from the republic of Cameroon. Experimental Agricul- ture. 1992; 28: 473–476. https://doi.org/10.1017/S0014479700020184 3. Daniel IO, Adetumbi JA. Maize seed supply systems and implications for seed sector development in southwestern Nigeria. Journal of Sustainable Agriculture. 2006; 28: 25–40. https://doi.org/10.1300/ J064v28n02_04 4. Mofoke ALE, Adewumi JK, Babatunde FE, Mudiare OJ, Ramalan AA. Yield of tomato grown under con- tinuous-flow drip irrigation in Bauchi state of Nigeria. Agricultural Water Management. 2006; 84: 166– 172. https://doi.org/10.1016/j.agwat.2006.02.001 5. Mtambanengwe F, Mapfumo P. Effects of organic resource quality on soil profile N dynamics and maize yields on sandy soils in Zimbabwe. Plant and Soil. 2006; 281: 173–191. https://doi.org/10.1007/ s11104-005-4182-3 6. Mahasi MJ, Kamundia JW. Cluster analysis in rapeseed (Brassica napus L.). African Journal of Agricul- tural Research. 2007; 9: 409–411. 7. Howell CL, Myburgh PA, Conradie WJ. Comparison of three different fertigation strategies for drip irri- gated table grapes—Part III: Growth, yield and quality. South African Journal of Enology and Viticulture. 2013; 34: 21–29. 8. Shrestha RP, Ligonja PJ. Social perception of soil conservation benefits in Kondoa eroded area of Tan- zania. International Soil and Water Conservation Research. 2015; 3: 183–195. https://doi.org/10.1016/ j.iswcr.2015.08.001 9. Ondieki DK, Nyaboga EN, Wagacha JM, Mwaura FB. Morphological and genetic diversity of Rhizobia Nodulating Cowpea (Vigna unguiculata L.) from agricultural soils of lower eastern Kenya. International Journal of Microbiology. 2017; article number 8684921. https://doi.org/10.1155/2017/8684921 PMID: 29463983 10. Wassie A, Pauline N. Evaluating small holder farmers’ preferences for climate smart agricultural prac- tices in Tehuledere district, northeastern Ethiopia. Singapore Journal of Tropical Geography. 2018; 39: 300–316. https://doi.org/10.1111/sjtg.12240 11. Rakotoarisoa NM, Tsujimoto Y, Oo AZ. Dipping rice seedlings in p-enriched slurry increases grain yield and shortens days to heading on p-deficient lowlands in the central highlands of Madagascar. Field Crops Research. 2020; 254: article number 107806. https://doi.org/10.1016/j.fcr.2020.107806 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 32 / 36 PLOS ONECrop yield in some east African countries 12. Kiboi MN, Ngetich FK, Mucheru-Muna MW, Diels J, Mugendi DN. Soil nutrients and crop yield response to conservation-effective management practices in the sub-humid highlands agro-ecologies of Kenya. Heliyon. 2021; 7: article number 07156. https://doi.org/10.1016/j.heliyon.2021.e07156 PMID: 34141923 13. Ray DK, Ramankutty N, Mueller ND, West PC, Foley JA. Recent patterns of crop yield growth and stag- nation. Nature Communications. 2012; 3: 1–7. https://doi.org/10.1038/ncomms2296 PMID: 23250423 14. Box GE, Jenkins GM. Time series analysis, forecasting and control. San Francisco: Holden-Day; 1976. 15. Akouegnonhou O, Demirbas N. Forecasting of rice self-sufficiency in the Benin Republic using ARIMA model. Selcuk Journal of Agriculture and Food Sciences. 2019; 33: 204–214. 16. Mgaya JF. Application of ARIMA models in forecasting livestock products consumption in Tanzania. Cogent Food and Agriculture. 2019; 5: article number 1607430. https://doi.org/10.1080/23311932. 2019.1607430 17. Ajetomobi JO, Olaleye AO. Auto-regressive integrated moving average (ARIMA) modelling of cocoa production in Nigeria: 1900-2025. Journal of Crop Improvement. 2019; 33: 445–455. https://doi.org/10. 1080/15427528.2019.1610534 18. Sharma PK, Dwivedi S, Ali L, Arora RK. Forecasting maize production in India using ARIMA model. Agro Economist-An International Journal. 2018; 5: 1–6. 19. Sikalubya M, Xu S, Yu W, Moonga P. Study on forecasting soybean production: An application of ARIMA model. In: Proceedings of the 2019 International Conference on Intelligent Computing, Automa- tion and Systems (ICICAS). 2019. pp. 447-452. 20. Biswas R. Study on ARIMA modelling to forecast area and production of Kharif rice in West Bengal. Cut- ting-edge Research in Agricultural Sciences. 2021; 12: 142–151. https://doi.org/10.9734/bpi/cras/v12/ 10700D 21. Shoko RR, Belete A. Efficient planning of sorghum production in South Africa–Application of the Box- Jenkin’s method. Journal of Agribusiness and Rural Development. 2017; 4: 835–841. https://doi.org/ 10.17306/J.JARD.2017.00352 22. Mehmood Q, Sial M, Riaz M, Shaheen N. Forecasting the production of sugarcane crop of Pakistan for the year 2018–2030, using Box-Jenkin’s methodology. Journal of Animal and Plant Sciences. 2019; 5: 1396–1401. 23. Chen SL, Miranda MJ. Modelling multivariate crop yield densities with frequent extreme events (No. 377-2016-20672). In: Proceedings of the Agricultural and Applied Economics Association Conference. 2004. 24. Warusawitharana M. Time-varying volatility and the power law distribution of stock returns. Journal of Empirical Finance. 2018; 49: 123–141. https://doi.org/10.1016/j.jempfin.2018.09.004 25. Masseran N, Yee LH, Safari MAM, Ibrahim K. Power law behavior and tail modelling on low income dis- tribution. Mathematics and Statistics. 2019; 7: 70–77. https://doi.org/10.13189/ms.2019.070303 26. Safari MAM, Masseran N, Ibrahim K, Al-Dhurafi NA. The power-law distribution for the income of poor households. Physica A: Statistical Mechanics and Its Applications. 2020; 557: article number 124893. https://doi.org/10.1016/j.physa.2020.124893 27. Asif M, Hussain Z, Asghar Z, Hussain MI, Raftab M, Shah SF, et al. A statistical evidence of power law distribution in the upper tail of world billionaires’ data 2010–20. Physica A: Statistical Mechanics and Its Applications. 2021; 581: article number 126198. https://doi.org/10.1016/j.physa.2021.126198 28. Simsek E, Kim M. Power-law tail in lag time distribution underlies bacterial persistence. Proceedings of the National Academy of Sciences. 2019; 116: 17635–17640. https://doi.org/10.1073/pnas. 1903836116 PMID: 31427535 29. Masseran N. Power-law behaviors of the duration size of unhealthy air pollution events. Stochastic Environmental Research and Risk Assessment. 2021; 35: 1499–1508. https://doi.org/10.1007/s00477- 021-01978-2 30. Chen X, Pan Y, Luo B. Research on power-law distribution of long-tail data and its application to tourism recommendation. Industrial Management and Data Systems. 2021; 121: 1268–1286. https://doi.org/ 10.1108/IMDS-10-2019-0584 31. Balthrop A, Quan S. The power-law distribution of cumulative coal production. Physica A: Statistical Mechanics and Its Applications. 2019; 530: article number 121573. https://doi.org/10.1016/j.physa. 2019.121573 32. Akhundjanov SB, Chamberlain L. The power-law distribution of agricultural land size. In: Proceedings of the 2019 Annual Meeting, Agricultural and Applied Economics Association. 2019. 33. Campolieti M. The distribution of union size: Canada, 1913–2014. Physica A: Statistical Mechanics and Its Applications. 2020; 558: article number 125007. https://doi.org/10.1016/j.physa.2020.125007 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 33 / 36 PLOS ONECrop yield in some east African countries 34. Campolieti M, Ramos A. The distribution of strike size: Empirical evidence from Europe and North America in the 19th and 20th centuries. Physica A: Statistical Mechanics and Its Applications. 2021; 563: article number 125424. https://doi.org/10.1016/j.physa.2020.125424 35. Pena G, Puente-Ajovı´n M, Ramos A, Sanz-Gracia F. Log-growth rates of CO2: An empirical analysis. Physica A: Statistical Mechanics and Its Applications. 2022; 588: article number 126550. https://doi. org/10.1016/j.physa.2021.126550 36. Hennessy DA. Crop yield skewness under law of the minimum technology. American Journal of Agricul- tural Economics. 2009; 91: 197–208. https://doi.org/10.1111/j.1467-8276.2008.01181.x 37. Koning AJ, Peng L. Goodness-of-fit tests for a heavy tailed distribution. Journal of Statistical Planning and Inference. 2008; 138: 3960–3981. https://doi.org/10.1016/j.jspi.2008.02.013 38. Zhang H, Wu Z. The general goodness-of-fit tests for correlated data. Computational Statistics and Data Analysis. 2022; 167: article number 107379. https://doi.org/10.1016/j.csda.2021.107379 39. Pankratz A. Forecasting with univariate Box-Jenkins models: Concepts and cases. New York: John Wiley and Sons; 2009. 40. Box GE, Jenkins GM, Reinsel GC, Ljung GM. Time series analysis: Forecasting and control. New York: John Wiley and Sons; 2015. 41. Box GE, Cox DR. An analysis of transformations. Journal of the Royal Statistical Society, B. 1964; 26: 211–243. 42. Kwiatkowski D, Phillips PCB, Schmidt P, Shin Y. Testing the null hypothesis of stationarity against the alternative of a unit root. Journal of Econometrics. 1992; 54: 159–178. https://doi.org/10.1016/0304- 4076(92)90104-Y 43. Cox DR, Hinkley DV. (1979). Theoretical statistics. London: Chapman and Hall; 1979. 44. Pareto V. La legge della domanda. Giornale Degli Economisti. 1895; 10: 59–68. 45. Fre´ chet M. Sur la loi de probabilite de l’ecart maximum. Annales de la Societe Polonaise de Mathema- tique. 1927; 6: 93. 46. R Core Team R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Sta- tistical Computing; 2022. https://www.R-project.org/. 47. Clauset A, Shalizi CR, Newman ME. Power law distributions in empirical data. SIAM Review. 2009; 51: 661–703. https://doi.org/10.1137/070710111 48. Gabaix X, Ibragimov R. Rank—1 / 2: A simple way to improve the OLS estimation of tail exponents. Journal of Business and Economic Statistics. 2011; 29: 24–39. https://doi.org/10.1198/jbes.2009. 06157 49. Hill BM. A simple general approach to inference about the tail of a distribution. Annals of Statistics. 1975; 3: 1163–1174. https://doi.org/10.1214/aos/1176343247 50. Theil H. A rank-invariant method of linear and polynomial regression analysis, I. Proceedings of the Koninklijke Nederlandse Akademie van Wetenschappen A. 1950; 53: 386–392. 51. Schwarz GE. Estimating the dimension of a model. Annals of Statistics. 1978; 6: 461–464. https://doi. org/10.1214/aos/1176344136 52. Hurvich CM, Tsai CL. Regression and time series model selection in small samples. Biometrika. 1989; 76: 297–307. https://doi.org/10.1093/biomet/76.2.297 53. Vuong QH. Likelihood ratio tests for model selection and non-nested hypotheses. Econometrica: Jour- nal of the Econometric Society. 1989; 57: 307–333. https://doi.org/10.2307/1912557 54. Rowhani P, Lobell DB, Linderman M, Ramankutty N. Climate variability and crop production in Tanza- nia. Agricultural and Forest Meteorology. 2011; 151: 449–460. https://doi.org/10.1016/j.agrformet. 2010.12.002 55. Knox J, Hess T, Daccache A, Wheeler T. Climate change impacts on crop productivity in Africa and south Asia. Environmental Research Letters. 2012; 7: article number 034032. https://doi.org/10.1088/ 1748-9326/7/3/034032 56. Craparo ACW, Van Asten PJA, Laederach P, Jassogne LTP, Grab SW. Coffea arabica yields decline in Tanzania due to climate change: Global implications. Agricultural and Forest Meteorology. 2015; 207: 1–10. https://doi.org/10.1016/j.agrformet.2015.03.005 57. Munishi LK, Lema AA, Ndakidemi PA. Decline in maize and beans production in the face of climate change at Hai district in Kilimanjaro region, Tanzania. International Journal of Climate Change Strate- gies and Management. 2015; 7: 17–26. https://doi.org/10.1108/IJCCSM-07-2013-0094 58. Kahsay GA, Hansen LG. The effect of climate change and adaptation policy on agricultural production in eastern Africa. Ecological Economics. 2016; 121: 54–64. https://doi.org/10.1016/j.ecolecon.2015.11. 016 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 34 / 36 PLOS ONECrop yield in some east African countries 59. Wolka K, Mulder J, Biazin B. Effects of soil and water conservation techniques on crop yield, runoff and soil loss in sub-Saharan Africa: A review. Agricultural Water Management. 2018; 207: 67–79. https:// doi.org/10.1016/j.agwat.2018.05.016 60. Mukamuhirwa A, Hovmalm HP, Bolinsson H, Ortiz R, Nyamangyoku O, Johansson E. Concurrent drought and temperature stress in rice. A possible result of the predicted climate change: Effects on yield attributes, eating characteristics, and health promoting compounds. International Journal of Envi- ronmental Research and Public Health. 2019; 16: article number 1043. https://doi.org/10.3390/ ijerph16061043 PMID: 30909476 61. Chapman S, Birch CE, Pope E, Sallu S, Bradshaw C, Davie J, et al. Impact of climate change on crop suitability in sub-Saharan Africa in parameterized and convection-permitting regional climate models. Environmental Research Letters. 2020; 15: article number 094086. https://doi.org/10.1088/1748-9326/ ab9daf 62. Mkonda MY, He XH. The influence of soil organic carbon and climate variability on crop yields in Kongwa district, Tanzania. Environmental Management. 2022 PMID: 34994818 63. Narita D, Sato I, Ogawada D, Matsumura A. Integrating economic measures of adaptation effectiveness into climate change interventions: A case study of irrigation development in Mwea, Kenya. Plos ONE. 2020; 15: article number e0243779. https://doi.org/10.1371/journal.pone.0243779 64. Ojara MA, Lou YS, Ongoma V, Mumo L, Akodi D, Ayugi B, et al. Projected changes in east African cli- mate and its impacts on climatic suitability of maize production areas by the mid-twenty-first century. Environmental Monitoring and Assessment. 2021; 193: article number 831. https://doi.org/10.1007/ s10661-021-09547-4 PMID: 34797418 65. Shabani Y, Pauline NM. Perceived effective adaptation strategies against climate change impacts: Per- spectives of maize growers in the southern highlands of Tanzania. Environmental Management. 2022 PMID: 34997302 66. Warsame AA, Sheik-Ali IA, Jama OM, Hassan AA, Barre GM. Assessing the effects of climate change and political instability on sorghum production: Empirical evidence from Somalia. Journal of Cleaner Production. 2022; 360: article number 131893. https://doi.org/10.1016/j.jclepro.2022.131893 67. Newman ME. Power laws, Pareto distributions and Zipf’s law. Contemporary Physics. 2005; 46: 323– 351. https://doi.org/10.1080/00107510500052444 68. Muli JM, Omanga PA, Jones RB. Farmer-to-farmer seed supply: Case study of pigeon pea seed distri- bution in Kenya. In: Rohrbach DD, Bishaw Z, Vangastel AJG, editors. Alternative strategies for small- holder seed supply. 1997. pp. 188–191. 69. Atera EA, Onyango JC, Azuma T, Asanuma S, Itoh K. Field evaluation of selected NERICA rice culti- vars in western Kenya. African Journal of Agricultural Research. 2011; 6: 60–66. 70. Fermont AM, Obiero HM, van Asten PJA, Bagurna Y, Okwuosa E. Improved cassava varieties increase the risk of soil nutrient mining: An ex-ante analysis for western Kenya and Uganda. In: Bationo A, Waswa, B, Kihara J, Kimetu J, editors, Advances in integrated soil fertility management in sub-Saharan Africa—Challenges and opportunities. 2007. pp. 511-519. 71. Pole FN, Saha HM, Mangale N, Mzingirwa AM, Munyambu P. On-Station evaluation of maize geno- types for nutrient and water use efficiency in the semi arid lands of coastal Kenya. In: Filho WL, Esilaba AO, Rao KPC, Sridhar G, editors, Adapting African agriculture to climate change—Transforming rural livelihoods. 2015. pp. 199-206. 72. Murungweni C, Van Wijk MT, Smaling EMA, Giller KE. Climate-smart crop production in semi-arid areas through increased knowledge of varieties, environment and management factors. Nutrient Cycling in Agroecosystems. 2016; 105: 183–197. https://doi.org/10.1007/s10705-015-9695-4 73. Thierfelder C, Rusinamhodzi L, Setimela P, Walker F, Eash NS. Conservation agriculture and drought- tolerant germplasm: Reaping the benefits of climate-smart agriculture technologies in central Mozam- bique. Renewable Agriculture and Food Systems. 2016; 31: 414–428. https://doi.org/10.1017/ S1742170515000332 74. Mkonda MY, He XH. Conservation agriculture in Tanzania. Sustainable Agriculture Reviews. 2017; 25: 309–324. https://doi.org/10.1007/978-3-319-48006-0_10 75. Shikuku KM, Winowiecki L, Twyman J, Eitzinger A, Perez JG, Mwongera C, et al. Small holder farmers’ attitudes and determinants of adaptation to climate risks in east Africa. Climate Risk Management. 2017; 16: 234–245. https://doi.org/10.1016/j.crm.2017.03.001 76. Alibu S, Neuhoff D, Senthilkumar K, Becker M, Kopke U. Potential of cultivating dry season maize along a hydrological gradient of an inland valley in Uganda. Agronomy-Basel. 2019; 9: article number 606. 77. Kabubo-Mariara J, Mulwa R. Adaptation to climate change and climate variability and its implications for household food security in Kenya. Food Security. 2019; 11: 1289–1304. https://doi.org/10.1007/ s12571-019-00965-4 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 35 / 36 PLOS ONECrop yield in some east African countries 78. Kiwia A, Kimani D, Harawa R, Jama B, Sileshi GW. Sustainable intensification with cereal-legume inter- cropping in eastern and southern Africa. Sustainability. 2019; 11: article number 2891. https://doi.org/ 10.3390/su11102891 PMID: 33552560 79. Obunyali CO, Karanja J, Oikeh SO, Omanya GO, Mugo S, Beyene Y, et al. On-farm performance and farmers’ perceptions of DroughtTEGO climate-smart maize hybrids in Kenya. Agronomy Journal. 2019; 111: 2754–2768. https://doi.org/10.2134/agronj2019.08.0600 80. Mairura FS, Musafiri CM, Kiboi MN, Macharia JM, Ng’etich OK, Shisanya CA, et al. Determinants of farmers’ perceptions of climate variability, mitigation, and adaptation strategies in the central highlands of Kenya. Weather and Climate Extremes. 2021; 34: article number 100374. https://doi.org/10.1016/j. wace.2021.100374 81. Sedebo DA, Li GC, Etea BG, Abebe KA, Ahiakpa JK, Arega Y, et al. Impact of small holder farmers’ cli- mate-smart adaptation practices on wheat yield in southern Ethiopia. Climate and Development. 2022; 14: 282–296. https://doi.org/10.1080/17565529.2021.2014777 82. Zizinga A, Mwanjalolo JGM, Tietjen B, Bedadi B, Gabiri G, Luswata KC. Effect of mulching and perma- nent planting basin dimensions on maize (Zea mays L.) production in a sub-humid climate. Water. 2022; 14: article number 79. https://doi.org/10.3390/w14010079 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287011 June 13, 2023 36 / 36 PLOS ONE
10.1371_journal.ppat.1011304	RESEARCH ARTICLE EphA2 is a functional entry receptor for HCMV infection of glioblastoma cells Xiao-Dong Dong1☯, Yan Li1,2☯, Ying Li3, Cong Sun1, Shang-Xin Liu1, Hao Duan1, Run Cui1, 1, Min-Hua Luo4, Qian Zhong1, Yong-Gao Mou1, Le Wen4,5, Bo Yang4, Mu-Sheng ZengID Hua Zhang1,3* 1 State Key Laboratory of Oncology in South China, Collaborative Innovation Center for Cancer Medicine, Guangdong Key Laboratory of Nasopharyngeal Carcinoma Diagnosis and Therapy, Sun Yat-sen University Cancer Center, Guangzhou, China, 2 Department of Pathology, Sun Yat-sen University Cancer Center, Guangzhou, China, 3 MOE Key Laboratory of Tropical Disease Control, Shenzhen Centre for Infection and Immunity Studies (CIIS), School of Medicine, Shenzhen Campus of Sun Yat-sen University, Shenzhen, Guangdong, China, 4 State Key Laboratory of Virology, CAS Center for Excellence in Brain Science and Intelligence Technology, Wuhan Institute of Virology, Chinese Academy of Sciences, Wuhan, China, 5 The Joint Center of Translational Precision Medicine, Guangzhou Institute of Pediatrics, Guangzhou Women and Children Medical Center; Wuhan Institute of Virology, Chinese Academy of Sciences, China ☯ These authors contributed equally to this work. * zengmsh@sysucc.org.cn (M-SZ); luomh@wh.iov.cn (M-HL); zhangh255@mail.sysu.edu.cn (HZ) Abstract Human cytomegalovirus (HCMV) infection is associated with human glioblastoma, the most common and aggressive primary brain tumor, but the underlying infection mechanism has not been fully demonstrated. Here, we show that EphA2 was upregulated in glioblastoma and correlated with the poor prognosis of the patients. EphA2 silencing inhibits, whereas overexpression promotes HCMV infection, establishing EphA2 as a crucial cell factor for HCMV infection of glioblastoma cells. Mechanistically, EphA2 binds to HCMV gH/gL com- plex to mediate membrane fusion. Importantly, the HCMV infection was inhibited by the treatment of inhibitor or antibody targeting EphA2 in glioblastoma cells. Furthermore, HCMV infection was also impaired in optimal glioblastoma organoids by EphA2 inhibitor. Taken together, we propose EphA2 as a crucial cell factor for HCMV infection in glioblastoma cells and a potential target for intervention. Author summary Human cytomegalovirus (HCMV) belonged to β-human herpesvirus is a ubiquitous path- ogen causing congenital infection and relating to morbidity and mortality in immuno- compromised transplant patients. Moreover, HCMV has been demonstrated to promote the progression of glioblastoma, the most common and aggressive primary brain tumor. Many studies highlighted the inevitable relationship between HCMV and glioblastoma. But the underlying mechanism of HCMV infection of glioblastoma cells has not been fully demonstrated. Here we found that EphA2, a member of receptor tyrosine kinases (RTKs) family, played a crucial role in HCMV infection of glioblastoma cells. Our research suggests that EphA2 mediates the infection of HCMV by interacting with a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Dong X-D, Li Y, Li Y, Sun C, Liu S-X, Duan H, et al. (2023) EphA2 is a functional entry receptor for HCMV infection of glioblastoma cells. PLoS Pathog 19(5): e1011304. https://doi.org/ 10.1371/journal.ppat.1011304 Editor: Wolfram Brune, Leibniz Institute of Virology (LIV), GERMANY Received: September 23, 2022 Accepted: March 20, 2023 Published: May 5, 2023 Copyright: © 2023 Dong et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting information files. Primary datasets have been generated and deposited in the Research Data Deposit (RDD) bank (http://www.researchdata.org.cn), with the approval RDD number RDDB2023226947. Funding: This work was supported by the National Natural Science Foundation of China (81830090) to M.S.Z; the National Natural Science Foundation of China (81802729) to Y.L; the National Natural Science Foundation of China (81872224), Guangdong Basic and Applied Basic Research PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 1 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma Foundation (2021A1515010734) to H.Z; the National Natural Science Foundation of China (82002128) to L.W. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. glycoproteins gH/gL/gO displaying on the surface of HCMV virion. We find that anti- EphA2 antibody and 2,5-dimethylpyrrolyl benzoic acid derivatives can block HCMV infection in glioblastoma cells, while 2,5-dimethylpyrrolyl benzoic acid derivatives can block HCMV infection in GBOs in a dose-dependent manner, indicating that the strategy of developing an anti-HCMV drug targeting EphA2 is feasible. Our findings may assist future studies striving for a better understanding of how HCMV infects glioblastoma cells and for potential new targets of innovative antiviral strategies. Introduction Glioblastoma, as the most common and aggressive primary brain tumor, exhibits an approxi- mately 15 months median longevity following medical treatment [1]. Many studies highlighted the inevitable relationship between human cytomegalovirus (HCMV) and glioblastoma. The viral proteins immediate-early 1 (IE1), pp65, or HCMV nucleic acids, have been detected in glioblastoma specimens [2–6], and IE1 is negatively associated with patient survival and median longevity [7]. HCMV can promote the progression of glioblastoma through various mechanisms, including affecting the cell cycle, invasion, and metastasis of tumor cells [8–10]. The antiviral drug ganciclovir can restore temozolomide sensitivity, indicating a potential mechanism underlying the positive effects observed in glioblastoma patients treated with anti- viral therapy [11]. But the role of HCMV in the pathogenesis of glioblastoma still remains con- troversial [12]. As a ubiquitous β-human herpesvirus, HCMV is an important pathogen in immunocom- promised individuals and tumor patients, causes congenital infection which targets the neural progenitor/stem cells in fetal brain [13]. In vivo HCMV infects and replicates in a wide range of cells, such as epithelial cells, smooth muscle cells, fibroblasts, macrophages, dendritic cells, liver cells, and vascular endothelial cells [14,15]. The widespread cellular tropism promotes the systematic spread of the virus in the human body. The glycoproteins on the surface of the envelope are key molecules that mediate virus infection. There are at least 25 glycoproteins dis- playing on the surface of HCMV virion, including gB, gM, gO, gH/gL, and UL128-131 [16– 18]. Infection of all cell types thus far tested seems to require gH/gL/gO, whereas gH/gL/ UL128-131 is an additional requirement some cell types, like epithelial and endothelial cells [19–21]. However, the mechanism of HCMV infection in glioblastoma has remained poorly clear. Eph receptors are the largest family of receptor tyrosine kinases (RTKs), consisting of 10 EphA receptor members and 5 EphB receptor members. These receptors play important roles in various developmental, physiological, and pathological processes [22]. Specifically, more and more evidence has shown Eph receptors’ involvement in tumorigenesis. The increased expression of ephrin receptor A2 (EphA2) and ephrin receptor A3 (EphA3) correlated with poor prognosis in patients with glioblastomas [23,24]. Additionally, the Eph receptor and its Ephrin ligand were reported to mediate herpesvirus infection. For example, EphA2 and Ephrin receptor A4 (EphA4) were required for Kaposi’s sarcoma-associated herpesvirus (KSHV) infection [25–27]. We and others identified EphA2, which is bound to gH/gL, to serve as a receptor for EBV infection of epithelial cells [28,29]. To explore whether Eph recep- tors implicate in HCMV infection of glioblastoma, we used siRNA screening to target the members of the Eph receptor family and found that EphA2 played a crucial role in HCMV infection. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 2 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma Results Eph receptor family siRNA screening identifies EphA2 as a host factor in HCMV infection of glioblastoma cells The Eph receptors mRNA levels were analyzed using the online database Gene Expression Profiling Interactive Analysis (GEPIA) to determine Eph receptors expression in glioblastoma and normal brain tissues [30]. The results revealed that EphA2, EphB2, EphB3, and EphB4 were significantly upregulated, while EphA4 and EphB6 were significantly downregulated in glioblastoma than normal brain tissues, respectively (S1 and S2 Figs). The mRNA expression of other Eph receptors showed no significant difference between glioblastoma and normal brain tissues (S2 Fig). Furthermore, the prognosis potential of Eph receptors in glioblastoma was determined using GEPIA. Higher EphA2 and EphA8 expression were associated with poorer, while higher EphA3 expression was associated with better, prognosis in patients with glioblastoma, respectively (S1 and S3 Figs). There was no obvious correlation between the expression of other Eph receptors and prognosis (S3 Fig). To explore the potential cell factors associated with HCMV infection in glioblastoma, we performed a siRNA screening in glioblastoma cells U138 targeting the Eph receptors family with a pool of 4 siRNAs targeting each gene. Each siRNA pool was transfected into U138 indi- vidually for 48 h, then infected with a reconstructed HCMV Towne strain (ATCC-VR977) expressing GFP [31,32]. The RT-qPCR assay showed that each of these siRNA pools efficiently reduced the expression of its targeted genes (S4 Fig). HCMV-positive cells were determined by flow cytometry, which revealed that knockdown EphA2 and EphB6 reduced the percentage of HCMV infected cells compared to the control siRNA transfected cells (Fig 1A). Notably, Fig 1. Identifying EphA2 as a potential cellular factor to mediate HCMV infection in glioblastoma cells. A, U138 cells were transfected with siRNA pools targeting the Eph receptors family for 36 h. Then cells were infected with HCMV for 3 days, and HCMV-positive cells were quantified by flow cytometry. Bars represent the percentage of infection determined by flow cytometry, with infection of control siRNA duplex (siCtrl) transfected cells normalized to 100%. Data are mean ± s.e.m. (n = 3 biological replicates) and represent three independent experiments. One-way ANOVA was carried out with Dunnett’s correction for multiple comparisons. NS, not significant; P<0.01; P < 0.001. https://doi.org/10.1371/journal.ppat.1011304.g001 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 3 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma EphA2 knockdown resulted a more than 50 percent reduction of HCMV infection (Fig 1A). Therefore, we selected EphA2 for further investigation. EphA2 plays a crucial role in HCMV infection of glioblastoma cells We further explored the role of EphA2 in HCMV infection in glioblastoma cells. Firstly, we performed EphA2 knockdown in U138 cells by three independent siRNAs. These siRNAs decreased EphA2 protein expression remarkably compared with a siRNA control (Fig 2A). Simultaneously, HCMV infection efficiency was reduced by >70% based on flow cytometry analysis for GFP (Fig 2B). Next, we performed similar experiment on another glioblastoma cell line, U251. Consistently, EphA2 siRNAs also significantly decreased the EphA2 protein expression (Fig 2C) and HCMV infection (Fig 2D). To exclude the potential siRNA off-target effects, we knockout EphA2 using CRISPR/Cas9. EphA2 protein and HCMV infection were remarkably decreased in U138 cells transduced with either sgA2-1# or sgA2-2# sgRNA (Fig 2E and 2F). Altogether, EphA2 is crucial for HCMV infection of glioblastoma cells. As for the reconstructed HCMV Towne strain also can infect fibroblast cells, we also performed the EphA2 knockdown assay in MRC-5 cells. Two independent siRNAs both can reduce the EphA2 expression as well as the HCMV infection efficiency (S5 Fig). Fig 2. EphA2 plays a key role in HCMV infection of glioblastoma cells. A to D, The U138 cells (A) and U251 cells (C) were transfected with EphA2 siRNAs (siA2- 1#, siA2-2#, siA2-3#) or siCtrl for 36 h. Part of the cells was harvested, and their EphA2 protein level was analyzed by WB, using α-tubulin as a loading control (representative of 3 independent experiments). The remaining cells were infected with HCMV and HCMV-positive cells and then were analyzed by flow cytometry (B and D). Bars represent the percentage of infection determined by flow cytometry, with infection of siCtrl transfected cells normalized to 100%. Data are mean ± s.e.m. (n = 3 biological replicates) and represent 3 independent experiments. One-way ANOVA was carried out with Dunnett’s correction for multiple comparisons. P < 0.001. E, F, sgRNAs targeting EphA2 (sgA2-1#, sgA2-2#) were delivered into U138 cells by the lentivirus package system. Cells were selected by puromycin for 3 days. The empty vector was used as control (sgVector). Part of the cells was harvested, and their EphA2 protein level was analyzed by WB (E), using α-tubulin as a loading control (representative of three independent experiments). The remaining cells were infected with HCMV and HCMV-positive cells and then were analyzed by flow cytometry (F). Bars represent the percentage of infection determined by flow cytometry, with infection of sgVector transduced cells normalized to 100%. Data are mean ± s.e.m. (n = 3 biological replicates) and represent 3 independent experiments. One-way ANOVA was carried out with Dunnett’s correction for multiple comparisons. P < 0.001. G, H, The plasmid expressing Flag-tagged EphA2, EphA2Δcyto or empty vector was stably transduced into T98G cells. Cells were harvested, and the EphA2 expression was analyzed by WB (G). α-tubulin was used as a loading control. Data are representative of 3 independent experiments. HCMV was used to infect these cells, and the HCMV infection efficiency was analyzed by flow cytometry (H). Bars represent the percentage of infection, with infection of empty vector- transfected cells normalized to 1. Data are mean ± s.e.m. (n = 3 biological replicates) and represent 3 independent experiments. One-way ANOVA was carried out with Dunnett’s correction for multiple comparisons. P < 0.001. https://doi.org/10.1371/journal.ppat.1011304.g002 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 4 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma Previous studies showed that deletion of EphA2 intracellular domain (EphA2Δcyto) inhib- ited KSHV infection of both epithelial and endothelial cells [26]. We stably overexpressed the wild-type EphA2 (EphA2wt)or EphA2Δcyto in glioblastoma T98G cells (Fig 2G), which were then infected with HCMV. Both EphA2wt and EphA2Δcyto remarkably increased the HCMV infection (Fig 2H). These results demonstrated that the extracellular domain was more essen- tial for HCMV infection of glioblastoma cells, which is consistent with the role of EphA2 in EBV infection of epithelial cells [29]. EphA2 phisically interacts with HCMV glycoprotein gH/gL The broad tropism of HCMV infection suggested that it interacted with multiple receptors via combination with different glycoproteins to enter different type of host cells [33]. HCMV gH/ gL and gB glycoprotein complex was essential for enter host cells [34,35]. Thus, we performed co-immunoprecipitation assays to test if HCMV gH/gL and gB could interact with EphA2 in HEK-293FT cells co-transfected with Myc-tagged EphA2 and Flag-tagged gH/gL or Flag- tagged gB. The results showed that EphA2 was co-immunoprecipitated by Flag-gH/gL pull- down but not Flag-gB (Fig 3A). The interaction of EphA2 with gH/gL was verified by Myc- EphA2 pulldown (Fig 3B). To further validate whether there was a direct interaction between EphA2 and gH/gL, recombinant His-gH/gL and EphA2 extracellular domains were purified (Fig 3C). Biolayer interferometry (BLI) was used to confirm the direct interaction between EphA2 and gH/gL by demonstrating high-affinity binding of EphA2 extracellular domain to His-gH/gL (KD = 3.64E-08 M) (Fig 3D). Altogether, these results demonstrated a strong, specific, and direct interaction between EphA2 and gH/gL. The EphA2 ectodomain contains three parts, the Ephrin-binding domain (EBD), the cyste- ine-rich domain (CRD), and the Fibronectin III type repeats (FNR). We next investigated which EphA2 ectodomain was involved in interaction with gH/gL. So we performed co-immu- noprecipitation assays between three EphA2 ectodomains and gH/gL. The result showed that gH/gL interacted with the EphA2 EBD domain and FNR domain but not the CRD domain (Fig 3E). HCMV gH/gL can form two kinds of complexes Trimer and Pentamer. The HCMV strain we used in this study has defect in genetic integrity and unable to form the Pentamer. We per- formed the co-IP assay to explore the interaction between EphA2 and the Trimer. The sequen- tial co-IP revealed that both EphA2 and gO were pulled down by anti-Flag immunoprecipitation, followed by anti-Myc immunoprecipitation in cells co-expressing Myc- EphA2, Flag-gH/gL and HA-gO (Fig 3F). These results suggest that EphA2 could also interact with gH/gL/gO trimer. EphA2 inhibitor and antibody block HCMV infection of U138 cells in a dose-dependent manner A 2,5-dimethylpyrrolyl benzoic acid derivatives are potent antagonists of EphA2. To evaluate its effects on HCMV infection, U138 cells were first treated with 2,5-dimethylpyrrolyl benzoic acid derivatives and then co-incubated with HCMV. The cytoxic assay showed that 2,5-dimethylpyrrolyl benzoic acid derivatives had no influence on U138 cells proliferation (S6A Fig). The result showed that 2,5-dimethylpyrrolyl benzoic acid derivatives treatment remarkably inhibited HCMV infection of glioblastoma cells in a dose-dependent manner (Fig 4A). Furthermore, to determine if antibodies against EphA2 ectodomain could block HCMV infection, we generated a rabbit polyclonal antibody against EphA2. Antibody against EphA2 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 5 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma Fig 3. EphA2 interacts with HCMV gH/gL. A, Co-immunoprecipitation assay of Myc tagged EphA2 (Myc-EphA2) with Flag-tagged gH (Flag-gH)/gL (Flag-gL)or Flag-tagged gB (Flag-gB), followed by western blot analysis with indicated antibodies. IB, immunoblotting. Data are representative of 2 independent experiments. B, Co-immunoprecipitation assay of Flag-gH/gL and Myc-EphA2, followed by western blot analysis with indicated antibodies. IB, immunoblotting. Data are representative of 2 independent experiments. C, Purified gH/gL, and EphA2 ectodomain proteins were presented by coomassie blue staining. Data are representative of 2 independent experiments. D, BLI analysis for the binding of EphA2 to gH/gL. gH/gL proteins were captured onto SA biosensors and assayed for binding to EphA2 ectodomain proteins at the indicated concentrations. Kinetic values calculated from the fit model for binding curves are shown in the table. E, Co-immunoprecipitation assay of Flag-gH/gL with EphA2 EBD domain (Myc-EphA2EBD), EphA2 CRD domain (Myc-EphA2CRD) or EphA2 FNR domain (Myc-EphA2FNR), followed by western blot analysis with indicated antibodies. IB, immunoblotting. Data are representative of 2 independent experiments. F, HEK-293T cells were transfected with Myc-EphA2, HA-gO together with FLAG-gH/gL or vector, lysed, and immunoprecipitated with antibody against FLAG as indicated IPx1:FLAG. The immunoprecipitated proteins were eluted by FLAG peptide and re-immunoprecipitated with antibody against Myc as indicated IPx2:Myc, followed by WB analysis with indicated antibodies. Data are representative of two independent experiments. https://doi.org/10.1371/journal.ppat.1011304.g003 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 6 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma Fig 4. EphA2 inhibitor and anti-EphA2 antibody inhibit HCMV infection in a dose-dependent manner. A, U138 cells were infected with HCMV in the presence of an indicated concentration of 2,5-dimethylpyrrolyl benzoic acid derivative. HCMV infection efficiency was quantified by flow cytometry after 3 days. Bars represent the percentage of infection determined by flow cytometry, with infection of no 2,5-dimethylpyrrolyl benzoic acid derivative treated cells normalized to 100%. Data are mean ± s.e.m. (n = 3 biological replicates) and represent 3 independent experiments. One-way ANOVA was carried out with Dunnett’s correction for multiple comparisons. P < 0.001. B, U138 cells were infected with HCMV in the presence of an indicated concentration of EphA2 antibody or rabbit IgG control. HCMV infection efficiency was quantified by flow cytometry after 3 days. Bars represent the percentage of infection determined by flow cytometry, with infection of rabbit IgG control-treated cells normalized to 100%. Data are mean ± s.e.m. (n = 3 biological replicates) and represent 3 independent experiments. One-way ANOVA was carried out with Dunnett’s correction for multiple comparisons. P < 0.001. C, Glioblastoma organoids were stained with hematoxylin-eosin staining (HE) and EphA2 antibody. Images of insets were magnified 3 times. Scale bars: 100 μm. Representative images from the samples were detected. D, E, Glioblastoma organoids were infected with HCMV in the presence of an indicated concentration of 2,5-dimethylpyrrolyl benzoic acid derivative. HCMV genome DNA was extracted from HCMV-infected organoids, and the copy number of HCMV was measured using qPCR. The GAPDH DNA was used for cell counting estimation (D). Bars represent relative HCMV DNA copy number determined by qPCR, with the copy number of no 2,5-dimethylpyrrolyl benzoic acid derivative treated cells normalized to 100%. Data are mean ± s.e.m. (n = 3 biological replicates) and represent 2 independent experiments. One-way ANOVA was carried out with Dunnett’s correction for multiple comparisons. P < 0.001. HCMV-positive cells were shown green in representative fluorescence images after 3 days (E). https://doi.org/10.1371/journal.ppat.1011304.g004 or rabbit IgG control was pre-incubated with U138 cells prior to HCMV infection, and infec- tion efficiency was measured by flow cytometry. Treatment with anti-EphA2 antibody signifi- cantly decreased HCMV infection of glioblastoma cells in a dose-dependent manner (Fig 4B) and showed no cytotoxicity (S6B Fig). Recently, organoids have been applied to model various kinds of tumors. We generated optimal glioblastoma organoids (GBOs) from tissue along the tumor margin with minimal necrosis and little surrounding brain tissue. Immunohistochemistry (IHC) analyses showed that the GBOs resembled their corresponding parental tumors with expression of GFAP and olig-2, high ki-67 proliferation index, and SMA-positive vessel (S7 Fig). Simultaneously, EphA2 could be detected in the GBOs by IHC (Fig 4C). PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 7 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma We then treated the GBOs with 2,5-dimethylpyrrolyl benzoic acid derivatives and then co- incubated with HCMV. Three days post-infection, the qPCR analysis showed that HCMV copy number was reduced in GBOs pretreated with EphA2 inhibitors in a dose-dependent manner (Fig 4D). We also observed under the fluorescence microscope that GBOs were per- missive for HCMV infection in vitro, and 2,5-dimethylpyrrolyl benzoic acid derivatives could inhibit HCMV infection of GBOs (Fig 4E). Altogether, these results suggested that blocking EphA2 impairs HCMV infection of glioblastoma cells and highlighted EphA2 as a potential therapeutic target to halt HCMV infection in glioblastoma cells. EphA2 mediates HCMV entry and fusion in glioblastoma cells HCMV glycoprotein gB and gM/gN glycoprotein complex were involved in the initial attach- ment of HCMV particles to host cells [36,37]. The gH/gL complex in concert with gB was believed to mediate the fusion of the virus with cellular membranes [38]. We performed entry assays, which revealed that overexpression of EphA2 in T98G cells promoted HCMV entry (Fig 5A). However, EphA2 knockout inhibited HCMV entry in U138 cells (Fig 5B). We further explored whether EphA2 was required for fusion. Using cell-based HCMV fusion assay, we found that gH/gL or gB alone slightly increased HCMV fusion, while co-expression of gH/gL and gB or gH/gL/gO and gB remarkably enhanced HCMV fusion (Fig 5C). Compared to the vector control, overexpression of EphA2 increased HCMV fusion (Fig 5D). We used HEK- 293T cells expressing gH/gL/gO/gB as effector cells and HEK-293T cells as target cells. HEK- 293T cells were treated with EphA2 antibody or IgG. EphA2 antibody also can block cell based fusion in a dose-dependent manner (Fig 5E). Altogether, these results indicated that EphA2 promoted the fusion of HCMV and cellular membranes, which was largely depended on the coexistence of gH/gL/gO and gB. Discussion Eph receptor family has been reported to mediate infection of several important human patho- gens [25–29]. By siRNA screen of Eph receptors family members, combined with a series of loss-of-function, gain-of-function, and protein-protein interaction assays, we identified EphA2 as an entry co-receptor for HCMV infection of glioblastoma. The relationship between HCMV and glioblastoma has been debated for a long time [2,5,39,40]. The expression of HCMV proteins in glioblastoma tissues was first reported in 2002. Then, more and more studies have detected HCMV proteins or DNA in human glioblas- toma tissue samples [4,41,42]. Moreover, HCMV has been demonstrated to promote the pro- gression of glioblastoma, negatively associate with patient’s survival, and the antiviral drug can restore sensitivity of glioblastoma chemotherapeutic drug [7–11]. These studies supply evi- dences for a role of HCMV infection in the pathogenesis of glioblastoma and emerging thera- peutic interventions directed against HCMV. Several molecules, such as EGFR, Integrin αvβ3, CD90, CD147, Neuropilin-2, and platelet- derived growth factor receptor alpha (PDGFRα), have been identified as receptors or co-recep- tors for HCMV infection in fibroblasts, epithelial, or endothelial cells [15,43–46]. However, the receptor(s) that mediates HCMV infection of human glioblastoma cells is unclear. Eph receptors and their Ephrin ligands are strongly related to herpes virus infection. We identified EphA2 as a host factor for HCMV infection of human glioblastoma cells by siRNA screen. EGFR, Integrin αvβ3, Neuropilin-2, and PDGFRα, but not CD90 and CD147, were signifi- cantly upregulated in glioblastoma than normal brain tissues by GEPIA analysis (S8 Fig). Whether these genes play some role in the HCMV infection of human glioblastoma cells needs further investigation. It has been reported that HCMV glycoprotein complex gH/gL/gB is PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 8 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma Fig 5. EphA2 mediates HCMV entry and fusion. A, The T98G cells stably expressing EphA2 were incubated with HCMV on ice for 2 h and then 30 minutes at 37˚C. Cells were washed with Hanks solution three times. The surface-bound virus was removed by proteinase K digestion. Then HCMV DNA copy number per cell was measured by qPCR. Bars represent the percentage of HCMV entry, with the entry of empty vector-transfected cells normalized to 100%. Data are mean ± s.e. m. (n = 3 biological replicates) and represent 2 independent experiments, two-tailed unpaired Student’s t-test. P < 0.001. B, EphA2 knockout cells were incubated with HCMV on ice for 2 h and then 30 minutes at 37˚C. Cells were washed with Hanks solution three times. The surface-bound virus was removed by proteinase K digestion. Then HCMV DNA copy number per cell was measured by qPCR. Bars represent the percentage of HCMV entry, with the entry of PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 9 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma sgVector transfected cells normalized to 100%. Data are mean ± s.e.m. (n = 3 biological replicates) and represent 2 independent experiments. One-way ANOVA was carried out with Dunnett’scorrection for multiple comparisons. P < 0.001. C, Effector HEK-293T cells were transfected with pT7-EMCLuc, pRL-SV40 and gB or gH/gL or gO or gH/ gL/gB or gH/gL/gO/gB. Cells then were co-cultured with HEK-293T cells transfected with a plasmid for T7 polymerase. The relative fusion activity was calculated as the ratio of firefly to Renilla luciferase activity after 24 h. Data are mean ± s.e. m. (n = 4 biological replicates) and represent 3 independent experiments. One-way ANOVA was carried out with Dunnett’s correction for multiple comparisons. NS P>0.05, P < 0.001. D, Cell-based HCMV fusion assay by co- culturing HEK-293T cells transfected with plasmids expressing EphA2 or vector and HEK-293T transfected with gB or gH/ gL or gO or gH/gL/gB or gH/gL/gO/gB. Bars represent the percentage of fusion, with the fusion of vector-transfected cells normalized to 100%. Data are mean ± s.e.m. (n = 4 biological replicates) and represent 2 independent experiments, One- way ANOVA was carried out with Dunnett’s correction for multiple comparisons. NS P>0.05, P < 0.01, *P < 0.001. E, Effector HEK-293T cells were transfected with pT7-EMCLuc, pRL-SV40 and gH/gL/gO/gB. Cells then were co-cultured with HEK-293T cells transfected with a plasmid for T7 polymerase in the presence of an indicated concentration of EphA2 antibody or rabbit IgG control. The relative fusion activity was calculated as the ratio of firefly to Renilla luciferase activity after 24 h. Data are mean ± s.e.m. (n = 4 biological replicates) and represent 3 independent experiments. One-way ANOVA was carried out with Dunnett’s correction for multiple comparisons. P < 0.001. https://doi.org/10.1371/journal.ppat.1011304.g005 important for HCMV fusion [38]. EphA2 also has been found to interact with KSHV gH/gL, EBV gH/gL and gB [25,26,28,29,47]. The binding site of EphA2 on EBV gH/gL (gL NGSN69- 72) is different from the binding site of EphA2 on KSHV gH/gL (gH ELEFN50-54) [48]. The three subunits of HCMV Trimer interact in a linear order, where the gL subunits bridge the gH and gO subunits in the center of the complex [49]. In our study, we found that EphA2 interacts with gH N-terminal, and the major binding regions are EBD and FNR domains. It is more like EBV gH/gL but Unlike KSHV gH/gL, for which the EphA2 EBD is the major bind- ing region. The specific binding site of EphA2 and HCMV Trimer need further investigation. We have identified that EphA2 could interact with HCMV gH/gL complex, but not gB to pro- mote cell-cell fusion, which may support the development of gH/gL vaccines or drugs target- ing gH/gL to block the HCMV infection. Two gH/gL-containing complexes, gH/gL/gO (trimer) and gH/gL/pUL128-131A (pentamer), regulate viral tropism [15]. Whether EphA2 could interact with gH/gL-containing pentamer need further investigation. Although many studies have detected HCMV DNA or protein in human glioblastoma tis- sues, it remains undetermined whether HCMV could directly infect the glioblastoma tissues. Organoids have been applied to model various kinds of tumors and drug screening. We dem- onstrate that the GBOs model is susceptibility to HCMV infection, suggesting GBOs model could be a potential tool to for studying the role HCMV in glioblastoma as well as for screening anti-HCMV drugs. Furthermore, EphA2 has been found to drive tumorigenicity in glioblas- toma [50]. We find that anti-EphA2 antibody and 2,5-dimethylpyrrolyl benzoic acid deriva- tives can block HCMV infection in glioblastoma cells, while 2,5-dimethylpyrrolyl benzoic acid derivative can block HCMV infection in GBOs in a dose-dependent manner, indicating that the strategy of developing an anti-HCMV drug targeting EphA2 is feasible. In conclusion, we have identified EphA2 as an important host factor mediating entry and fusion of HCMV with glioblastoma cells, which may assist future studies striving for a better understanding of how HCMV infects glioblastoma cells and for potential new targets of inno- vative antiviral strategies. Materials and methods Cell lines U138, U251, T98G, HFF, MRC-5 and HEK-293T cells (Thermo Scientific, R70007) were grown in DMEM (C11995500BT, GIBCO, California) supplemented with 10% (vol/vol) FBS (10099-141C, GIBCO, Australia). Cells were cultured in humidified 5% CO2 incubators at 37˚C. HEK-293T was purchased from ATCC; U251, HFF, MRC-5 were gifted from Professor PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 10 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma Luo Minhua (WuHan institute of virology, CAS, China); U138 and T98G were gifted from Professor Chen Zhongping (Sun Yat-sen University, Guangzhou, China). Reagents The reagents used were as follows: antibodies against EphA2 (6997, CST, Massachusetts), α -tubulin (sc-8035, Santa Cruz, Texas), β -actin (66009-1-Ig, Proteintech, Chicago), Myc-tag (16286-1-Ig, Proteintech, Chicago), Flag-tag (66008-1-Ig, Proteintech, Chicago), Flag-tag (F4049, Sigma-Aldrich, Germany), normal rabbit IgG (AB-105-C, R&D, Minnesota), the horseradish peroxidase (HRP)-conjugated goat-anti-mouse/rabbit secondary antibodies (#31460, #61–6520, Invitrogen, California), IgG (AB-105-c, R&D, Minnesota) and 2,5-dimethylpyrrolyl benzoic acid derivative (sc-314230, Santa Cruz, Texas). All other reagents were obtained from Sigma-Aldrich unless otherwise indicated. Gene silencing assays The siRNAs pools against Eph family genes and non-targeting siRNA duplexes denoted as siCtrl were synthesized from RIBOBIO (China). The three single siRNA duplexes against EphA2 (RIBOBIO, China) were listed as follows: EphA2 si1#: 5ʹ-GCAGCAAGGTGCAC- GAATT-3ʹ; EphA2 si2#: 5ʹ-TGACCAACGACGACATCAA-3ʹ; EphA2 si3#: 5ʹ-GCA- GACTGTGAACTTGACT-3ʹ. Under the instructions, all the siRNAs were delivered by RNAi MAX (13778150; Invitrogen, California). RT-qPCR Total RNA was extracted using TRIzol reagent (T9424; Sigma-Aldrich, Germany). 1 μg of RNA was reversely transcribed using the RNA Reverse Transcription System (A5001, Pro- mega, Wisconsin) to analyze gene expression. The mRNA level was calculated by RT-qPCR using the LightCycler 480 SYBR Green I Master (04887352001, Roche, Switzerland) and ana- lyzed on a Roche LightCycler 480. All the gene expressions were normalized to the housekeep- ing gene actin beta (ACTB). Plasmids For HCMV infection of T98G assays, cDNA of EphA2 and EphA2Δcyto were integrated into pHAGE vector; for co-IP assays, cDNA of EphA2, EphA2ΔEBD, EphA2ΔCRD, EphA2ΔFNR, gB, gH, gL were integrated into the pCDNA6-Myc vector, Flag-tagged gH, gL, were integrated into pCDNA3.1 vector; for cell-based fusion assay, expression plasmids for pCAG-T7, pT7EMC-Luc were gifted from Professor R. Longnecker (Northwestern University), and Wolfgang Hammerschmidt (Helmholtz Zentrum Mu¨nchen), cDNA of gH, gL, gB were inte- grated into Phage vector; for purification assay, cDNA of EphA2 (27–534) was integrated into pCDNA3.1 vector with N-terminal Kozak sequence and CD5 signal peptide and C-ter- minal 6Histidine tag, pLko.-gH-gL was gifted from Professor Qian Zhikang (Fudan University). Plasmid transfection Indicated plasmids were delivered by Lipofectamine 3000 followed the instructions. The pHA- GE-EphA2 or pHAGE- EphA2Δcyto was delivered into T98G cells by a lentivirus package system. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 11 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma CRISPR-Cas9-mediated EphA2 gene knockout The EphA2 U138 knockout cell line was generated by CRISPR-Cas9 gene-editing technology. The two guide RNA (sgRNA) sequences are gRNA-1 5ʹ-GAAGCGCGGCATGGAGCTCC-3ʹ and gRNA-2 5ʹ-CGAAGCAGGCGCGGGCTGCC-3ʹ. Oligonucleotides corresponding to the sgRNA were synthesized and cloned into Cas9-expressing plasmid lenti-CRISPR-V2. EphA2 gRNA-encoding plasmids were packaged into lentivirus and infected U138 cells to establish stable cells. Cells were screened by puro (1 μg mL-1) for 3 days. HCMV preparation and infection of cells Recombinant HCMV virus (rHCMV), derived from Towne strain by inserting a GFP gene driven by an SV40 promoter into the viral genome [51]. To provide flanking DNA for homolo- gous recombination HCMV BAC, two fragments of HCMV DNA were PCR amplified from cosmid clone CM1052, which contains the HindIII K/Q, X, V, and Wfragments of AD169 HCMV. This final construct, pUSF-3, contains the prokaryotic genetic elements necessary to confer maintenance as a BAC in E. coli, HCMV DNA sequences to direct homologous recom- bination to the unique short (US) region of the viral genome, and the GFP marker to facilitate identification and purification of recombinant HCMV in eukaryotic cells. The flanking DNA deletes 8.9 kb of DNA within the US region of HCMV that has been defined as dispensable for HCMV replication in cell culture, truncating IRS1 after amino acid 719 and removing reading frames US1 to US11 plus the carboxyterminal third of US12. Recombinant virus having pUSF- 3 substituted for US1-12 was enriched by plaque purification using the GFP marker. The virus was propagated and titrated in HFFs. The indicated cells were infected with with HCMV and incubated for 3 h at 37˚C to allow virus enter into cells, then unbound virus was discarded by washing with PBS three times. Then, cells were cultured in a fresh medium for 72 h, then ana- lyzed GFP positive cells with flow cytometry (cytoFLEX; Beckman). HCMV entry assay 5 × 104 cells were seeded in 24-well plates overnight. Cells were incubated with HCMV for 2 h at 4˚C to allow virus attach to cells, then moved into incubators at 37˚C for 30 min. To remove the unbound HCMV Hanks solution was used to wash cells for three times. Then 200 μl tryp- sin with EDTA and proteinase K were added for 15 min at 4˚C and stopped by a trypsin inhib- itor. HCMV genome DNA was extracted from HCMV-infected cells using Omega tissue DNA Mini Kit (D3396, Omega) followed by the manufacturer’s instruction. The copy number of internalized HCMV was measured using TaqMan qPCR. qPCR for the GAPDH DNA was used for cell counting estimation. Primers included 5’-GACTAGTGTGATGCTGGCCAAG-3’ and 5’-GCTACAATAGCCTCTTCCTCATCTG-3’ for HCMV, and 5’-CCCCACACACATG CACTTACC-3’ and 5’- CCTAGTCCCAGGGCTTTGATT-3’ for GAPDH. Cell-based fusion assay Effector HEK-293T cells were transiently transfected with plasmid pT7EMCLuc and pRL-SV40; expressing the luciferase gene driven by the T7 polymerase, Renilla luciferase as internal control. Effector cells were also transiently transfected gH/gL, gO, gB Separately or simultaneously. Target cells (HEK-293T) were transfected with expression plasmid for pCAGT7 (expression T7 DNA polymerase) together with EphA2 or empty vector. At 24 h post transfection, 2.5× 105 effector HEK-293T cells were co-cultured with 2.5×105 target HEK- 293T cells in 24-well plates for 24 h. The luciferase activity was measured according to a dual- luciferase reporter assay system (E2920, Promega, Wisconsin) by the GloMax-96 Microplate PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 12 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma Luminometer. The ratio of firefly luciferase activity to Renilla luciferase activity was used as the relative fusion activity. Blocking assay For the antibody blocking assays, U138 cells were preincubated with rabbit polyclonal anti- EphA2 antibody costumed from Abmart (China), which were validated by ELISA and western blot, by diluting to the concentrations of 100 and 200 μg ml–1 in FBS-free DMEM at 4˚C for 1 h. Rabbit IgG at the indicated concentration was used as the negative control. Preincubated cells were exposed to HCMV in the presence of indicated antibodies at 37˚C for 3 h. The per- centage of HCMV infected cells were determined by flow cytometry at 72 h post-infection. For the EphA2 inhibitor 2,5-dimethylpyrrolyl benzoic acid derivative blocking assays, U138 cells were preincubated with 2,5-dimethylpyrrolyl benzoic acid derivative at the concentrations of 25 and 50 μg ml–1 in FBS-free DMEM at room temperature for 1 h. DMSO was used as a con- trol. The pretreated cells were infected with HCMV, and HCMV infection efficiency was determined by flow cytometry at 72 h post-infection. Immunoprecipitation The transfected cells were lysed in lysis buffer: 50 mM Tris-HCl, pH 7.4; 250 mM NaCl; 5 mM EDTA, pH 8.0; 0.5% Nonidet P40 (NP40) containing 1 mM phenylmethylsulfonyl fluoride and Roche Complete protease inhibitor cocktail (04693159001, Roche, Switzerland). After centrifug- ing at 15,000g for 20 min at 4˚C, the supernatant was incubated with 25 μl Anti-c-Myc Agarose Affinity Gel (A7470, Sigma-Aldrich, Germany) or ANTI-FLAG M2 Affinity Gel (A2220, Sigma- Aldrich, Germany) for 2 h at 4˚C. After washing three times with lysis buffer to remove unbound proteins, the sample was suspended in 2x SDS-sample buffer and boiled for 10min at 98˚C. The complex was then analyzed by western blotting with the indicated antibodies. 3D organoid culture The tumor tissues were obtained from patients medically required in Sun Yat-sen University Cancer Center. Tissues were minced into approximately 0.5 to 1 mm diameter pieces and dis- tributed in ultra-low attachment 6-well culture plates with 4 mL of GBO medium containing 50% DMEM:F12 (Thermo Fisher Scientific, Massachusetts), 50% Neurobasal (Thermo Fisher Scientific, Massachusetts), 1X GlutaMax (Thermo Fisher Scientific, Massachusetts), 1X NEAAs (Thermo Fisher Scientific, Massachusetts), 1X PenStrep (Thermo Fisher Scientific, Massachusetts), 1X N2 supplement (Thermo Fisher Scientific, Massachusetts), 1XB27 w/o vitamin A supplement (Thermo Fisher Scientific, Massachusetts), 1X 2-mercaptoethanol (Thermo Fisher Scientific, Massachusetts), and 2.5 μg/ml human insulin (Sigma-Aldrich, Ger- many) per well and placed on an orbital shaker rotating at 120 rpm within a 37˚C, 5% CO2, and 90% humidity sterile incubator. Roughly 75% of the medium was changed every 48 h by tilting the plates at a 45˚ angle and aspirating the medium above the sunken organoids. Immunohistochemistry staining To detect EphA2, CD68, GFAP, SMA, ki-67 and olig-2 in GBOs, antibodies against EphA2 (ab5386, Abcam, UK), CD68(ZM-0060, ZSGB-Bio, China), GFAP (ZA-0529, ZSGB-Bio, China), SMA (ZM-0003, ZSGB-Bio, China), ki-67(TA800648, ZSGB-Bio, China) and olig-2 (ZA-0561, ZSGB-Bio, China) were used as primary antibodies overnight at 4˚C. After washing three times in PBST, the tissue sections were incubated with anti-rabbit secondary antibody (1:1000, Zymed, California) or anti-mouse secondary antibody (1:1000, Zymed, California), PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 13 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma and then treated with 3-diaminobenzidine tetrahydrochloride for 10 seconds, finally stained with 10% Mayer’s hematoxylin (ZSGB-Bio, China). Protein purification Eukaryotic expression system was used to purify human EphA2 and HCMV gH/gL. Briefly, expression plasmid was mixed with Polyethyleneimine Linear (PEI) MW40000 (40816ES, Yea- sen, China) at a molar ratio of 1:3 in 293F medium (UP1000, Union, China), and the mixture was transfected to Expi293F (A14527, ThermoFisher, Massachusetts). Then the cells were cul- tured at 37˚C with 5% CO2 and 120 rpm shaking for 6 days, and the supernatant was harvested by centrifuge, filtered with 0.45 um filtering membrane, and applied to gravity column loaded with Ni Sepharose Excel (17371202, Cytiva, Washington, DC). The column was washed by PBS with 30 mM imidazole and eluted by PBS with 500 mM imidazole. The elution was con- centrated by ultracentrifuge and further purified by size exclusion chromatography (SEC) using Superdex 200 increase 10/300GL (28990944, Cytiva, Washington, DC) on an AKTA Pure25M (GE healthcare, Massachusetts). Biolayer interferometry The kinetic assay was performed on Octet Red96 (18–1127, ForteBio, California) to determine the affinity of EphA2 with HCMV gH/gL. Briefly, the HCMV gH/gL was first biotinylated using a sulfo-LC-LC-biotin kit (21335, ThermoFisher, Massachusetts) following the manufac- turer’s recommended procedures. During the kinetic assay, PBS with 0.1% Tween20 was used as a kinetic buffer (KB) to dilute protein or equilibration biosensors. To detect the affinity, HCMV gH/gL was loaded on SA biosensors (Fortebio, California) at a concentration of 5 ug/ mL. After the baseline process, a series of diluted EphA2 proteins were associated with the bio- sensors, and the dissociation process followed. During the data processing in Octet data analy- sis software, the curve (0 mM EphA2) was used as blank to recalibrate the other raw curves, and a global fitting model was used to calculate the general kinetic parameters. Statistical analyses Results are expressed as mean ± s.e.m. from three independent experiments. The unpaired parametric two-sided Student’s t-test was used for statistical analysis involving two-group comparisons, and One-way ANOVA was carried out with Dunnett’s correction for multiple comparisons involving more than two groups (P< 0.05, P< 0.01, P< 0.001). Statistical analyses were performed with Graphpad Prism (GraphPad Software, San Diego, CA, USA). Supporting information S1 Fig. The correlation of EphA2 with glioblastoma patients’ survival. A, Analysis of EphA2 expression in glioblastoma tissues and normal brain tissues using the samples in the GEPIA database. B, Kaplan Meier survival analysis of overall survival rates between high EphA2 expression group and low expression group in the GEPIA database. (TIF) S2 Fig. The expression of Eph receptors in glioblastoma compared to normal brain tissues. Analysis of EphA1, EphA3, EphA4, EphA5, EphA6, EphA7, EphA8, EphB1, EphB2, EphB3, EphB4 or EphB6 expression levels in glioblastoma tissues and normal brain tissues using GEPIA database. (TIF) PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 14 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma S3 Fig. The correlation of Eph receptors with glioblastoma patients’ survival. Kaplan Meier survival analysis of overall survival rates between high EphA1, EphA3, EphA4, EphA5, EphA6, EphA7, EphA8, EphB1, EphB2, EphB3, EphB4 or EphB6 expression group and low expression group using GEPIA database. (TIF) S4 Fig. The knockdown efficiency of siRNA pools. The U138 cells were transfected with siRNA pools targeting the indicated genes or control siRNA (siCtrl) for 36 h. RT-qPCR was used to quantify the mRNA level of the respective targeted gene. Results were quantified rela- tive to the housekeeping gene beta-actin (ACTB) expression and shown as fold-change of mRNA abundance normalized to siCtrl, which was normalized to 100%. Data are mean ± s.e. m. (n = 3 biological replicates) and represent 2 independent experiments, two-tailed unpaired Student’s t-test. (TIF) S5 Fig. EphA2 plays a key role in HCMV infection of fibroblast cells. A, The MRC-5 cells were transfected with EphA2 siRNAs (siA2-1#, siA2-2#) or siCtrl for 36 h. Part of the cells was harvested, and their EphA2 protein level was analyzed by WB, using α-tubulin as a loading control (representative of 3 independent experiments). The remaining cells were infected with HCMV and HCMV-positive cells were analyzed by flow cytometry (B). Bars represent the per- centage of infection determined by flow cytometry, with infection of siCtrl transfected cells normalized to 100%. Data are mean ± s.e.m. (n = 3 biological replicates) and represent 3 inde- pendent experiments. One-way ANOVA was carried out with Dunnett’s correction for multi- ple comparisons. *P < 0.001. (TIF) S6 Fig. Cytotoxic assays of 2,5-dimethylpyrrolyl benzoic acid derivatives and EphA2 anti- body. A, B MTT assay of U138 cell line treated with benzoic acid at concentration of 0, 25, 50 μg /mL (A) or IgG at concentration of 200 μg /mL or EphA2 antibody at concentration of 100, 200 μg /mL. n = 4 biological replicates. (TIF) S7 Fig. Verification of the glioblastoma organoids by IHC. Glioblastoma organoids were stained with GFAP, olig-2, ki-67, and SMA antibodies. Images of insets were magnified 3 times. Scale bars: 100 μm. Representative images from the samples were detected. (TIF) S8 Fig. The expression of identified HCMV receptors or co-receptors in glioblastoma com- pared to normal brain tissues. EGFR, Integrin αvβ3, Neuropilin-2, PDGFRα, CD90 or CD147 expression levels were analyzed in glioblastoma tissues and normal brain tissues using the GEPIA database. (TIF) Author Contributions Conceptualization: Xiao-Dong Dong, Yan Li, Mu-Sheng Zeng, Min-Hua Luo, Hua Zhang. Data curation: Xiao-Dong Dong, Yan Li, Ying Li, Cong Sun, Hao Duan, Run Cui. Formal analysis: Xiao-Dong Dong, Yan Li, Shang-Xin Liu. Funding acquisition: Yan Li, Le Wen, Mu-Sheng Zeng, Hua Zhang. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 15 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma Investigation: Xiao-Dong Dong, Yan Li, Ying Li, Cong Sun, Shang-Xin Liu, Hao Duan, Run Cui, Qian Zhong, Yong-Gao Mou, Le Wen, Bo Yang, Mu-Sheng Zeng, Min-Hua Luo, Hua Zhang. Methodology: Xiao-Dong Dong, Yan Li, Cong Sun, Hao Duan, Run Cui, Bo Yang, Mu-Sheng Zeng, Min-Hua Luo, Hua Zhang. Project administration: Xiao-Dong Dong, Yan Li, Qian Zhong, Yong-Gao Mou, Hua Zhang. Resources: Qian Zhong, Yong-Gao Mou, Mu-Sheng Zeng, Min-Hua Luo, Hua Zhang. Software: Xiao-Dong Dong, Yan Li, Ying Li, Cong Sun, Shang-Xin Liu. Supervision: Qian Zhong, Mu-Sheng Zeng, Hua Zhang. Validation: Xiao-Dong Dong, Yan Li, Ying Li, Mu-Sheng Zeng, Hua Zhang. Visualization: Xiao-Dong Dong, Yan Li. Writing – original draft: Xiao-Dong Dong, Yan Li. Writing – review & editing: Xiao-Dong Dong, Yan Li, Ying Li, Mu-Sheng Zeng, Hua Zhang. References 1. Pu W, Nassar ZD, Khabbazi S, Xie N, McMahon KA, Parton RG, Riggins GJ, Harris JM, Parat MO: Cor- relation of the invasive potential of glioblastoma and expression of caveola-forming proteins caveolin-1 and CAVIN1. Journal of neuro-oncology 2019, 143(2):207–220. https://doi.org/10.1007/s11060-019- 03161-8 PMID: 30949900 2. Cobbs CS, Harkins L, Samanta M, Gillespie GY, Bharara S, King PH, Nabors LB, Cobbs CG, Britt WJ: Human cytomegalovirus infection and expression in human malignant glioma. Cancer Res 2002, 62 (12):3347–3350. PMID: 12067971 3. Lucas KG, Bao L, Bruggeman R, Dunham K, Specht C: The detection of CMV pp65 and IE1 in glioblas- toma multiforme. Journal of neuro-oncology 2011, 103(2):231–238. https://doi.org/10.1007/s11060- 010-0383-6 PMID: 20820869 4. Scheurer ME, Bondy ML, Aldape KD, Albrecht T, El-Zein R: Detection of human cytomegalovirus in dif- ferent histological types of gliomas. Acta Neuropathol 2008, 116(1):79–86. https://doi.org/10.1007/ s00401-008-0359-1 PMID: 18351367 5. dos Santos CJ, Stangherlin LM, Figueiredo EG, Correa C, Teixeira MJ, da Silva MC: High prevalence of HCMV and viral load in tumor tissues and peripheral blood of glioblastoma multiforme patients. J Med Virol 2014, 86(11):1953–1961. https://doi.org/10.1002/jmv.23820 PMID: 24173908 6. Ranganathan P, Clark PA, Kuo JS, Salamat MS, Kalejta RF: Significant association of multiple human cytomegalovirus genomic Loci with glioblastoma multiforme samples. Journal of virology 2012, 86 (2):854–864. https://doi.org/10.1128/JVI.06097-11 PMID: 22090104 7. Wen L, Zhao F, Qiu Y, Cheng S, Sun JY, Fang W, Rayner S, McVoy MA, Jiang XJ, Tang Q et al: Human cytomegalovirus DNA and immediate early protein 1/2 are highly associated with glioma and prognosis. Protein Cell 2020, 11(7):525–533. https://doi.org/10.1007/s13238-020-00696-9 PMID: 32189197 8. Zhu X, Hu B, Hu M, Qian D, Wang B: Human cytomegalovirus infection enhances invasiveness and migration of glioblastoma cells by epithelial-to-mesenchymal transition. International journal of clinical and experimental pathology 2020, 13(10):2637–2647. PMID: 33165445 9. Soroceanu L, Matlaf L, Khan S, Akhavan A, Singer E, Bezrookove V, Decker S, Ghanny S, Hadaczek P, Bengtsson H et al: Cytomegalovirus Immediate-Early Proteins Promote Stemness Properties in Glio- blastoma. Cancer research 2015, 75(15):3065–3076. https://doi.org/10.1158/0008-5472.CAN-14- 3307 PMID: 26239477 10. 11. Fornara O, Bartek J Jr., Rahbar A, Odeberg J, Khan Z, Peredo I, Hamerlik P, Bartek J, Stragliotto G, Landazuri N et al: Cytomegalovirus infection induces a stem cell phenotype in human primary glioblas- toma cells: prognostic significance and biological impact. Cell Death Differ 2016, 23(2):261–269. https://doi.org/10.1038/cdd.2015.91 PMID: 26138445 Liu C, Clark PA, Kuo JS, Kalejta RF: Human Cytomegalovirus-Infected Glioblastoma Cells Display Stem Cell-Like Phenotypes. mSphere 2017, 2(3). https://doi.org/10.1128/mSphere.00137-17 PMID: 28656174 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 16 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma 12. Herbein G: Tumors and Cytomegalovirus: An Intimate Interplay. Viruses 2022, 14(4). https://doi.org/ 10.3390/v14040812 PMID: 35458542 13. Zhou YP, Mei MJ, Wang XZ, Huang SN, Chen L, Zhang M, Li XY, Qin HB, Dong X, Cheng S et al: A congenital CMV infection model for follow-up studies of neurodevelopmental disorders, neuroimaging abnormalities, and treatment. JCI Insight 2022, 7(1). https://doi.org/10.1172/jci.insight.152551 PMID: 35014624 14. Li G, Kamil JP: Viral Regulation of Cell Tropism in Human Cytomegalovirus. Journal of virology 2016, 90(2):626–629. https://doi.org/10.1128/JVI.01500-15 PMID: 26559829 15. Martinez-Martin N, Marcandalli J, Huang CS, Arthur CP, Perotti M, Foglierini M, Ho H, Dosey AM, Shri- ver S, Payandeh J et al: An Unbiased Screen for Human Cytomegalovirus Identifies Neuropilin-2 as a Central Viral Receptor. Cell 2018, 174(5):1158–1171.e1119. https://doi.org/10.1016/j.cell.2018.06.028 PMID: 30057110 16. Varnum SM, Streblow DN, Monroe ME, Smith P, Auberry KJ, Pasa-Tolic L, Wang D, Camp DG 2nd, Rodland K, Wiley S et al: Identification of proteins in human cytomegalovirus (HCMV) particles: the HCMV proteome. J Virol 2004, 78(20):10960–10966. https://doi.org/10.1128/JVI.78.20.10960-10966. 2004 PMID: 15452216 17. Wang D, Shenk T: Human cytomegalovirus virion protein complex required for epithelial and endothelial cell tropism. Proc Natl Acad Sci U S A 2005, 102(50):18153–18158. https://doi.org/10.1073/pnas. 0509201102 PMID: 16319222 18. Adler B, Scrivano L, Ruzcics Z, Rupp B, Sinzger C, Koszinowski U: Role of human cytomegalovirus UL131A in cell type-specific virus entry and release. J Gen Virol 2006, 87(Pt 9):2451–2460. https://doi. org/10.1099/vir.0.81921-0 PMID: 16894182 19. Wille PT, Knoche AJ, Nelson JA, Jarvis MA, Johnson DC: A human cytomegalovirus gO-null mutant fails to incorporate gH/gL into the virion envelope and is unable to enter fibroblasts and epithelial and endothelial cells. J Virol 2010, 84(5):2585–2596. https://doi.org/10.1128/JVI.02249-09 PMID: 20032184 20. Zhou M, Lanchy JM, Ryckman BJ: Human Cytomegalovirus gH/gL/gO Promotes the Fusion Step of Entry into All Cell Types, whereas gH/gL/UL128-131 Broadens Virus Tropism through a Distinct Mechanism. Journal of virology 2015, 89(17):8999–9009. https://doi.org/10.1128/JVI.01325-15 PMID: 26085146 21. Stegmann C, Abdellatif MEA, Laib Sampaio K, Walther P, Sinzger C: Importance of Highly Conserved Peptide Sites of Human Cytomegalovirus gO for Formation of the gH/gL/gO Complex. J Virol 2017, 91 (1). https://doi.org/10.1128/JVI.01339-16 PMID: 27795411 22. Sharma P, Roberts C, Herpai D, Fokt ID, Priebe W, Debinski W: Drug Conjugates for Targeting Eph Receptors in Glioblastoma. Pharmaceuticals (Basel) 2020, 13(4). https://doi.org/10.3390/ph13040077 PMID: 32340173 23. Liu F, Park PJ, Lai W, Maher E, Chakravarti A, Durso L, Jiang X, Yu Y, Brosius A, Thomas M et al: A genome-wide screen reveals functional gene clusters in the cancer genome and identifies EphA2 as a mitogen in glioblastoma. Cancer research 2006, 66(22):10815–10823. https://doi.org/10.1158/0008- 5472.CAN-06-1408 PMID: 17090523 24. Debinski W, Priebe W, Tatter SB: Maximizing Local Access to Therapeutic Deliveries in Glioblastoma. Part I: Targeted Cytotoxic Therapy. In: Glioblastoma. Edited by De Vleeschouwer S. Brisbane (AU); 2017. 25. Chakraborty S, Veettil MV, Bottero V, Chandran B: Kaposi’s sarcoma-associated herpesvirus interacts with EphrinA2 receptor to amplify signaling essential for productive infection. Proceedings of the National Academy of Sciences of the United States of America 2012, 109(19):E1163–1172. https://doi. org/10.1073/pnas.1119592109 PMID: 22509030 26. Hahn AS, Kaufmann JK, Wies E, Naschberger E, Panteleev-Ivlev J, Schmidt K, Holzer A, Schmidt M, Chen J, Konig S et al: The ephrin receptor tyrosine kinase A2 is a cellular receptor for Kaposi’s sar- coma-associated herpesvirus. Nature medicine 2012, 18(6):961–966. 27. Chen J, Zhang X, Schaller S, Jardetzky TS, Longnecker R: Ephrin Receptor A4 is a New Kaposi’s Sar- coma-Associated Herpesvirus Virus Entry Receptor. mBio 2019, 10(1). https://doi.org/10.1128/mBio. 02892-18 PMID: 30782663 28. Chen J, Sathiyamoorthy K, Zhang X, Schaller S, Perez White BE, Jardetzky TS, Longnecker R: Ephrin receptor A2 is a functional entry receptor for Epstein-Barr virus. Nat Microbiol 2018, 3(2):172–180. https://doi.org/10.1038/s41564-017-0081-7 PMID: 29292384 29. 30. Zhang H, Li Y, Wang HB, Zhang A, Chen ML, Fang ZX, Dong XD, Li SB, Du Y, Xiong D et al: Ephrin receptor A2 is an epithelial cell receptor for Epstein-Barr virus entry. Nat Microbiol 2018, 3(2):1–8. https://doi.org/10.1038/s41564-017-0080-8 PMID: 29292383 Tang Z, Li C, Kang B, Gao G, Li C, Zhang Z: GEPIA: a web server for cancer and normal gene expres- sion profiling and interactive analyses. Nucleic acids research 2017, 45(W1):W98–W102. https://doi. org/10.1093/nar/gkx247 PMID: 28407145 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 17 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma 31. Cheng S, Jiang X, Yang B, Wen L, Zhao F, Zeng WB, Liu XJ, Dong X, Sun JY, Ming YZ et al: Infected T98G glioblastoma cells support human cytomegalovirus reactivation from latency. Virology 2017, 510:205–215. https://doi.org/10.1016/j.virol.2017.07.023 PMID: 28750324 32. Cheng S, Zhao F, Wen L, Yang B, Wang XZ, Huang SN, Jiang X, Zeng WB, Sun JY, Zhang FK et al: iTRAQ-Based Proteomics Analysis of Human Cytomegalovirus Latency and Reactivation in T98G Cells. J Virol 2022, 96(2):e0147621. https://doi.org/10.1128/JVI.01476-21 PMID: 34730396 33. Vanarsdall AL, Johnson DC: Human cytomegalovirus entry into cells. Current opinion in virology 2012, 2(1):37–42. https://doi.org/10.1016/j.coviro.2012.01.001 PMID: 22440964 34. Eisenberg RJ, Atanasiu D, Cairns TM, Gallagher JR, Krummenacher C, Cohen GH: Herpes virus fusion and entry: a story with many characters. Viruses 2012, 4(5):800–832. https://doi.org/10.3390/ v4050800 PMID: 22754650 35. Connolly SA, Jardetzky TS, Longnecker R: The structural basis of herpesvirus entry. Nature reviews Microbiology 2021, 19(2):110–121. https://doi.org/10.1038/s41579-020-00448-w PMID: 33087881 36. Boyle KA, Compton T: Receptor-binding properties of a soluble form of human cytomegalovirus glyco- protein B. Journal of virology 1998, 72(3):1826–1833. https://doi.org/10.1128/JVI.72.3.1826-1833. 1998 PMID: 9499033 37. Kari B, Gehrz R: A human cytomegalovirus glycoprotein complex designated gC-II is a major heparin- binding component of the envelope. Journal of virology 1992, 66(3):1761–1764. https://doi.org/10. 1128/JVI.66.3.1761-1764.1992 PMID: 1310777 38. Vanarsdall AL, Ryckman BJ, Chase MC, Johnson DC: Human cytomegalovirus glycoproteins gB and gH/gL mediate epithelial cell-cell fusion when expressed either in cis or in trans. Journal of virology 2008, 82(23):11837–11850. https://doi.org/10.1128/JVI.01623-08 PMID: 18815310 39. Holdhoff M, Guner G, Rodriguez FJ, Hicks JL, Zheng Q, Forman MS, Ye X, Grossman SA, Meeker AK, Heaphy CM et al: Absence of Cytomegalovirus in Glioblastoma and Other High-grade Gliomas by Real- time PCR, Immunohistochemistry, and In Situ Hybridization. Clinical cancer research: an official journal of the American Association for Cancer Research 2017, 23(12):3150–3157. https://doi.org/10.1158/ 1078-0432.CCR-16-1490 PMID: 28034905 40. Baumgarten P, Michaelis M, Rothweiler F, Starzetz T, Rabenau HF, Berger A, Jennewein L, Braczynski AK, Franz K, Seifert V et al: Human cytomegalovirus infection in tumor cells of the nervous system is not detectable with standardized pathologico-virological diagnostics. Neuro-oncology 2014, 16 (11):1469–1477. https://doi.org/10.1093/neuonc/nou167 PMID: 25155358 41. Mitchell DA, Xie W, Schmittling R, Learn C, Friedman A, McLendon RE, Sampson JH: Sensitive detec- tion of human cytomegalovirus in tumors and peripheral blood of patients diagnosed with glioblastoma. Neuro Oncol 2008, 10(1):10–18. https://doi.org/10.1215/15228517-2007-035 PMID: 17951512 42. Dziurzynski K, Wei J, Qiao W, Hatiboglu MA, Kong LY, Wu A, Wang Y, Cahill D, Levine N, Prabhu S et al: Glioma-associated cytomegalovirus mediates subversion of the monocyte lineage to a tumor propagating phenotype. Clin Cancer Res 2011, 17(14):4642–4649. https://doi.org/10.1158/1078-0432. CCR-11-0414 PMID: 21490182 43. Wang X, Huong SM, Chiu ML, Raab-Traub N, Huang ES: Epidermal growth factor receptor is a cellular receptor for human cytomegalovirus. Nature 2003, 424(6947):456–461. https://doi.org/10.1038/ nature01818 PMID: 12879076 44. Wang X, Huang DY, Huong SM, Huang ES: Integrin alphavbeta3 is a coreceptor for human cytomega- lovirus. Nature medicine 2005, 11(5):515–521. https://doi.org/10.1038/nm1236 PMID: 15834425 45. Li Q, Fischer E, Cohen JI: Cell Surface THY-1 Contributes to Human Cytomegalovirus Entry via a Macropinocytosis-Like Process. J Virol 2016, 90(21):9766–9781. 46. Kabanova A, Marcandalli J, Zhou T, Bianchi S, Baxa U, Tsybovsky Y, Lilleri D, Silacci-Fregni C, Foglier- ini M, Fernandez-Rodriguez BM et al: Platelet-derived growth factor-alpha receptor is the cellular recep- tor for human cytomegalovirus gHgLgO trimer. Nature microbiology 2016, 1(8):16082. 47. Hahn AS, Desrosiers RC: Binding of the Kaposi’s sarcoma-associated herpesvirus to the ephrin binding surface of the EphA2 receptor and its inhibition by a small molecule. Journal of virology 2014, 88 (16):8724–8734. https://doi.org/10.1128/JVI.01392-14 PMID: 24899181 48. Chen J, Schaller S, Jardetzky TS, Longnecker R: Epstein-Barr Virus gH/gL and Kaposi’s Sarcoma- Associated Herpesvirus gH/gL Bind to Different Sites on EphA2 To Trigger Fusion. J Virol 2020, 94 (21). https://doi.org/10.1128/JVI.01454-20 PMID: 32847853 49. Kschonsak M, Rouge L, Arthur CP, Hoangdung H, Patel N, Kim I, Johnson MC, Kraft E, Rohou AL, Gill A et al: Structures of HCMV Trimer reveal the basis for receptor recognition and cell entry. Cell 2021, 184(5):1232–1244.e1216. https://doi.org/10.1016/j.cell.2021.01.036 PMID: 33626330 50. Binda E, Visioli A, Giani F, Lamorte G, Copetti M, Pitter KL, Huse JT, Cajola L, Zanetti N, DiMeco F et al: The EphA2 receptor drives self-renewal and tumorigenicity in stem-like tumor-propagating cells PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 18 / 19 PLOS PATHOGENSEphA2 is a receptor for HCMV infection of glioblastoma from human glioblastomas. Cancer cell 2012, 22(6):765–780. https://doi.org/10.1016/j.ccr.2012.11. 005 PMID: 23238013 51. Marchini A, Liu H, Zhu H: Human cytomegalovirus with IE-2 (UL122) deleted fails to express early lytic genes. J Virol 2001, 75(4):1870–1878. https://doi.org/10.1128/JVI.75.4.1870-1878.2001 PMID: 11160686 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1011304 May 5, 2023 19 / 19 PLOS PATHOGENS
10.1371_journal.pone.0288068	RESEARCH ARTICLE Complete genome sequencing and characterization of a potential new genotype of Citrus tristeza virus in Iran Abozar GhorbaniID Keramatollah Izadpanah4 1, Mohammad Mehdi Faghihi2, Faezeh Falaki3, 1 Nuclear Science and Technology Research Institute, Nuclear Agriculture Research School, Karaj, Iran, 2 Plant Protection Research Department, Fars Agricultural and Natural Resources Research and Education Centre, AREEO, Zarghan, Iran, 3 Department of Plant Protection, College of Agriculture Sciences and Food Industries, Science and Research Branch, Islamic Azad University, Tehran, Iran, 4 Plant Virology Research Centre, College of Agriculture, Shiraz University, Shiraz, Iran abghorbany@aeoi.org.ir Abstract Citrus tristeza virus (CTV) is one of the economically destructive viruses affecting citrus trees worldwide, causing significant losses in fruit production. Comparative genomic studies have shown genetic diversity in various regions of the genome of CTV isolates, which has classified the virus into several genotypes. In recent years, some orange citrumelo-tolerant rootstocks showed yellowing, decline, and vein clearing in northern Iran (Mazandaran prov- ince, Sari). We confirmed the presence of CTV in the symptomatic trees by reverse tran- scription PCR (RT-PCR). The complete genome of a Sari isolate of CTV (Sari isolate) was sequenced using next-generation sequencing (NGS) technology. In addition, phylogenetic analysis, differential gene expression of the virus and identification of its variants in a popu- lation were studied. We obtained the final contigs of the virus (nt) and annotated all genomes to viral ORFs, untranslated regions (UTRs), intergenic regions, and 5’ and 3’ ends of the genome. Phylogenetic analysis of the Sari isolate and other genotypes of CTV showed that the Sari isolates were placed in a distinct cluster without a sister group. Based on the num- ber of specific transcripts (TPM) in CTV RNA -Seq, P13 was the most highly expressed gene related to the host range of the virus and its systemic infection. The ORFs of the poly- protein, P33, and P18 showed variation in a single population of the sari isolate. The CTV has a potential for variation in a population in a host, and these variations may contribute to the best fit of the CTV in different situations. In Iran, whole genome sequencing of the CTV was performed for the first time, and we gained new insights into CTV variation in a population. Introduction Citrus tristeza is an economically destructive disease affecting citrus trees throughout the world, causing significant losses in fruit production [1]. The causal agent of the disease, Citrus tristeza virus (CTV), is an RNA plant virus of the genus Closterovirus (family Closteroviridae) a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Ghorbani A, Faghihi MM, Falaki F, Izadpanah K (2023) Complete genome sequencing and characterization of a potential new genotype of Citrus tristeza virus in Iran. PLoS ONE 18(6): e0288068. https://doi.org/10.1371/journal. pone.0288068 Editor: Shirin Farzadfar, Iranian Research Institute of Plant Protection, ISLAMIC REPUBLIC OF IRAN Received: February 17, 2023 Accepted: June 17, 2023 Published: June 29, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0288068 Copyright: © 2023 Ghorbani et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. PLOS ONE \| https://doi.org/10.1371/journal.pone.0288068 June 29, 2023 1 / 12 PLOS ONEFunding: The authors received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. A potential new genotype of Citrus tristeza virus in Iran composed of a variety of strains. The virus host range is limited to Citrus and related species [2]. The symptomology and severity of CTV infection widely vary depending on the type of viral strain, host cultivar, and root-stock/scion combination. However, classic disease symp- toms include stem-pitting, vein clearing, leaf cupping, and yellowing [3]. The CTV is distrib- uted in the world as mild or severe strains. The main mode of transmission of CTV is through multiple aphid vectors including Aphis gossypii, Toxoptera citricida, A. citricola, T. aurantii, A. craccivora, and Myzus persicae [4–6]. Other spread pathways include graft transmission and plant material infested with aphid vectors [7]. CTV is a single-stranded positive-sense RNA virus and the genome is encapsulated by the major and minor coat proteins (CP and CPm, respectively) [8]. The complete genomes have been sequenced from the biologically and geographically distinct strains and reveal that the CTV has the largest plant viral genome. The sequences range from 19,226 to 19,306 nucleo- tides and are organized into 12 open reading frames (ORFs) which potentially encode for 17 proteins [9–11]. The most comprehensively studied part of the CTV genome is the CP gene which encodes for the major coat protein (25 KDa). The CP gene is commonly used for molec- ular typing of CTV isolates around the world [9,12–14]. ORF1 encodes a replication-associated protein (p349) directly translated from the genome (gRNA) and proteins p33, p6, p65, p61, p27, p25, p18, p13, p20 and p23 (5’ to 3’) expressed via 30-terminal subgenomic RNAs (sgRNAs). Based on genome-wide sequence diversity, CTV iso- lates are classified as strains or genotypes. The recognized genotypes of CTV are constantly expanding and to date include T36, VT, T3, RB, T68, T30, HA16-5, S1, AT -1, and Taiwan- Pum. CTV strains are known to recombine frequently and the Maximum likelihood (ML) phylogenetic method has been used for classification [15]. Factors that may influence genetic diversity include geographic differences and mode of transmission [16]. Studying the whole genome of diverse isolates from around the world pro- vides a deeper understanding of molecular epidemiology. In particular, the use of next-genera- tion sequencing (NGS) tools such as RNA sequencing (RNA-Seq) is now a major focus in the field of plant virology for accurate and specific whole-genome analyses and rapid diagnoses [17,18]. In this study, high-throughput sequencing (HTS) was used to determine the genetic diversity and phylogenetic relationships for the Sari isolate of CTV. Materials and methods Ethics statement This research was carried out in the laboratories of the Shiraz University and Nuclear Science and Technology Research Institute. No other permits were required to conduct this research. We also confirm that no endangered or protected species were involved in the studies. Sample collection Samples of leaves and young shoot tips of sweet orange trees on citrumelo rootstocks showing various virus-like symptoms, including stem-pitting, vein clearing, leaf cupping, and yellow- ing, were collected in Mazandaran Province, Sari, Iran, in 2021–22. A total of five orchards were visited, and samples were stored individually in microfuge tubes at −80˚C until process- ing. Samples were used for RT-PCR and RNA-Seq tests. RAN extraction and initial RT-PCR screening of CTV Total RNA was extracted from 100 mg of leaves using the TRIzol1 reagent (USA) following the manufacturer’s instructions. RNA was quantified by Nanodrop1 spectrophotometer PLOS ONE \| https://doi.org/10.1371/journal.pone.0288068 June 29, 2023 2 / 12 PLOS ONEA potential new genotype of Citrus tristeza virus in Iran (Thermo Fisher Scientific, USA). Total RNA was treated by DNase (Thermo Scientific). For initial RT-PCR screening of CTV, cDNA was synthesized with Random hexamer primers and an M-MuLV Reverse Transcriptase cDNA synthesis kit (Thermo Scientific). RT-PCR reaction was carried out in a total volume of 20 μL of reaction mixture containing 1 μL of cDNA as tem- plate, Taq DNA polymerase (Takara Bio, Otsu, Shiga, Japan), (1.25 U/50 μL) and capsid pro- tein gene primers (forward: 50 ATGGACGACGAAACAAAGAA 30), (reverse: 50 TCAACGTGTG TTGAATTTCC 30) [19]. PCR reactions were performed using the Ampliqon Taq DNA Poly- merase 2x Master Mix (Denmark) in a total volume of 12.5 μL containing 20 ng cDNA and 10 μM of each primer. The PCR condition comprised one initial denaturation cycle at 95˚C for 10 min, followed by 35 cycles of 95˚C for 30 s, 55˚C for 30 s, and 75˚C for 30 s, with a final extension step at 75˚C for 10 min. Aliquots of PCR-amplified fragments were loaded on 1% agarose gels in Tris-borate (TBE) buffer (0.09 M Tris base, 0.09 M boric acid, 0.002 M EDTA, pH 8.0). A 100-bp DNA Ladder (Promega, Madison, WI) was used as a nucleic acid marker. After electrophoresis, the gels were stained with ethidium bromide at 0.5 μg/ml and then ana- lyzed using a BIO imaging system (Syngene, Frederick, MD). RNA library prep and CTV whole genome sequencing Total RNA from three leaf samples of a tree was used for RNA-Seq. The rRNA was removed from the total RNA using a Ribo-zero rRNA Removal Kit (Epicentre, WI, USA). RNA-Seq libraries using TruSeq Stranded Total RNA for Illumina were prepared according to the manu- facturer’s instructions and sequenced on the Illumina HiSeq 2000 platform (Novogene, China), generating paired-end reads of 150 bp. Bioinformatics analysis Data analysis of FASTQ files was performed using CLC Genomics Workbench (version 20, QIAGEN, Venlo, The Netherlands). Sequencing adaptors and low-quality sequences with ambiguous nucleotides were trimmed to obtain sequences of approximate size (using default parameters: Reads less than 15 nt were trimmed, ambiguous nucleotides maximum 2). FASTQ files of paired-end sequences from the library were assembled into transcriptomes de novo using CLC Genomic Workbench (word size 15, minimum length of contigs 150 nt). To distinguish and retrieve viral sequences from the entire transcriptome, the assembled tran- scriptome contigs were mapped to the chromosol and non-chromosol reference genome of Citrus (Citrus sinensis, GCF_000317415). Subsequently, the unmapped contigs were com- pared with sequences available in the NCBI viral reference database (https://www.ncbi.nlm. nih.gov/genome/viruses/) using the CLC BLAST tool, which is more reliable than other sequence similarity programs for virus identification (the E-value cut-off was 1e-5). Following the initial analysis, putative virus-associated contigs were compared with sequences in the NCBI NR (non-redundant proteins) database. Subsequently, endogenous virus-like sequences were removed from the data set, and virus-related contigs were retained for further analysis. Viral sequence mapping and genome assembly To assemble the whole viral genomes, the reads transcriptome sequences were aligned with a reference whole viral genome sequence (NC_001661). Sequences associated with the viral genome were mapped to the viral reference genome using Geneious Prime 2019, and consen- sus sequences were generated with a threshold of 95% identity. The viral-associated sequences were then analyzed using Geneious version R10 (Biomatters, New Zealand) for sequence trimming, nucleotide analysis, viral ORF determination, gene annotation, and phylogenetic analysis. In addition, the phylogenetic maximum likelihood tree PLOS ONE \| https://doi.org/10.1371/journal.pone.0288068 June 29, 2023 3 / 12 PLOS ONEA potential new genotype of Citrus tristeza virus in Iran was constructed using a 50% bootstrap threshold with 1000 bootstrap repeats and a score defined using the Kimura 2-parameter model (MEGA 11). Genomic diversity of CTV at interpopulation and intrapopulation levels To evaluate the genetic diversity of the selected CTVs at the ORF level, new reference sequences (accession number: OP900953) were generated from the RNA-Seq data examined using CLC Genomics Workbench. The clean reads were mapped to the viral contig genome. The minimum coverage, minimum variant frequency, and maximum variant P values were set to 2, 0.01, and 10−6, respectively. Single-nucleotide variants (SNVs) were filtered out as synon- ymous SNVs. The frequency and distribution of polymorphisms in ORFs of CTV (Sari isolate) for which transcriptome profiling was performed were assessed and values were compared. The Protein Data Bank (PDB) was downloaded from RCSB PDB (https://www.rcsb.org) to visualize SNVs in tertiary protein structures using CLC Genomics Workbench. Profiling the gene expression using RNA-Seq data Reads trimmed to RNA-Seq data were subjected to expression analysis by mapping all sequences to the synthesized master sequences. Transcript per million (TPM) was calculated for the detected ORFs using CLC Genomics Workbench, with the default parameters of the software and the length fraction and similarity fraction of 0.8. Phylogenetic analysis Phylogenetic trees for sequences deposited in the NCBI nucleotide database were constructed based on whole-genome and phylogenetic analysis. Briefly, the tree construction was based on the ClustalW alignment of the concatenating sequence of CTV using Geneious Prime 2019 and applying the maximum likelihood method in the MEGA11 program (Tamura et al., 2021). The substitution models used for each phylogenetic tree were selected via the best-fit model tool of MEGA software. Carrot yellow leaf virus (NC013007) was used as an out-group to root the tree, and bootstrapping (threshold: 60) was carried out using 1000 replicates. Results and discussion Symptomatology and conventional RT-PCR During the 2021–22 field surveys in different districts of Mazandaran Province, Sari, Iran, sweet orange trees (on citrumelo rootstocks) showed the typical tristeza symptoms including chlorosis, yellow leaves, leaf cupping, vein clearing, vein flecking, stem pitting and grooving, poor growth and decline condition, (Fig 1). Samples that showed diverse virus-like symptoms were screened by RT-PCR for the presence of CTV. The virus was detected in all symptomatic samples (5 samples) as indicated by the amplification of specific fragments with approximately 670 bp (Fig 2). No amplification was observed with the RNA templates of healthy citrus plants. Genome assembly of CTV After preprocessing the raw data (42,133,736 reads, 150 nt), clean reads (39,652,356) were obtained from the whole transcriptome sequencing sample. Subsequently, citrus-related reads (80%) were removed by comparing transcriptome contigs with the citrus sequences available in NCBI using MEGABLAST. Assembly of the remaining viral-associated reads produced 0.6% CTV-related contigs of 300 to 18,397 nt. No additional contigs with high sequence simi- larity to other viruses or viroids were detected from the transcriptomic reads. After assembly, we obtained the final contigs of the virus (19,300 nt), and we annotated all genomes at viral PLOS ONE \| https://doi.org/10.1371/journal.pone.0288068 June 29, 2023 4 / 12 PLOS ONEA potential new genotype of Citrus tristeza virus in Iran Fig 1. Citrus tristeza virus symptoms on sweet orange trees on citrumelo rootstocks. Yellowing of leaves (A) andstem pitting and grooving at graft union (B). https://doi.org/10.1371/journal.pone.0288068.g001 ORFs, untranslated regions (UTRs), intergenic regions, and 5’ and 3’ ends of the genome (Fig 3 and Table 1). Phylogenetic analysis of full-length CTV sequences Phylogenetic analysis of the Sari isolate and other genotypes of CTV that were downloaded from NCBI was performed. The maximum likelihood method and whole genome sequence of Fig 2. Electrophoresis pattern of RT-PCR products using total RNA extracts. Lane 1: A healthy orange seedling, Lane 2: Infected trees, Lane 3: CTV positive control, Lane 4: DNA ladder (Thermo Scientific, UK). https://doi.org/10.1371/journal.pone.0288068.g002 PLOS ONE \| https://doi.org/10.1371/journal.pone.0288068 June 29, 2023 5 / 12 PLOS ONEA potential new genotype of Citrus tristeza virus in Iran Fig 3. Schematic diagram of the CTV genome organization. The open boxes represent ORFs and their translation products. https://doi.org/10.1371/journal.pone.0288068.g003 CTV genotypes were used for genotype clustering. Phylogenetic analysis of our data revealed that the Sari isolate (accession number OP900953) was placed in a distinct cluster from other genotypes and not in a sister group with a known genotype (Fig 4). On the other hand, identi- fying the similarity matrix of the Sari isolate with others showed at least 80% identity with other genotypes and the highest similarity (92%) with the T3 genotype (S1 File). Differential expression of CTV ORFs Raw transcriptomic data were used to profile the differential gene expression of CTV in citrus transcriptomic data. Differentially expressed genes (DEGs) were characterized for RNA-Seq data (adjusted-value p < 0.01) (Fig 5). Based on the number of specific transcripts (TPM) identified in CTV RNA-Seq. P13 which is positioned toward the 30 termini of the genome near the CP gene, was the most highly expressed gene than the 50 terminal gene encoding polypro- tein and P6 protein which is related to the host range (Fig 5). P13 has the same function as P18 Table 1. Genome features and ORFs ofCTV sequence of Sari isolate. Type 3’UTR CDS CDS CDS CDS CDS CDS CDS CDS CDS CDS mat_peptide mat_peptide mat_peptide CDS 5’UTR Length (nt) Product 273 630 549 360 504 672 723 1611 1785 156 912 7818 1475 1454 10750 107 23-kDa protein 20-kDa protein 13-kDa protein 18-kDa protein 25-kDa coat protein 27-kDa protein 61-kDa protein 65-kDa protein 6-kDa protein 33-kDa protein replicas papain-like protease papain-like protease 401-kDa viral polyprotein CDS: Protein Coding Sequence. mat_peptide: Mature peptide. https://doi.org/10.1371/journal.pone.0288068.t001 Locus_tag CTVgp11 CTVgp10 CTVgp09 CTVgp08 CTVgp07 CTVgp06 CTVgp05 CTVgp04 CTVgp03 CTVgp02 CTVgp01 CTVgp01 CTVgp01 CTVgp01 PLOS ONE \| https://doi.org/10.1371/journal.pone.0288068 June 29, 2023 6 / 12 PLOS ONEA potential new genotype of Citrus tristeza virus in Iran Fig 4. Phylogenetic tree based on full-length genomes of the Sari isolate (Sari, shown with the red circle) of CTV, and the isolates from GenBank shown as accession number and country of origin. The tree was constructed by the maximum likelihood method using MEGA 11 software. Numbers on branches are bootstrap values of 1,000 replicates. Beet yellows virus was used as an outgroup : A18, : L1, : Outgroup. : This study, : HA-65, : HA-65, : T36, : T68, : T30, : M1, : VT, : RB, : T3, : S1, https://doi.org/10.1371/journal.pone.0288068.g004 PLOS ONE \| https://doi.org/10.1371/journal.pone.0288068 June 29, 2023 7 / 12 PLOS ONEA potential new genotype of Citrus tristeza virus in Iran Fig 5. Expression levels (transcript per million, TPM) of Citrus tristeza virus (CTV) Gene IDs in RNA-seq data of citrus infected with CTV. https://doi.org/10.1371/journal.pone.0288068.g005 and P23. A quantitative comparison of TPM reads showed that the gene P13 is the most abun- dantly expressed transcript, followed by genes P20 and CPm. The P6 ORF showed lower TPM in this study. The P13 protein resulted in systemic infection of calamondin was required for infection of a broader range of citrus varieties [20]. It is also reported to interact with the P20, P23 and P33 [21]. However, other functions of the P13 have not been elucidated yet and the highest expres- sion of this gene warrants further study of its function. The P20 protein has a key role in the suppression of RNA silencing [22] and might block the loading of CTV sRNAs into the RNA silencing complex or interfere with it through alter- native mechanisms [23]. The CPm protein was the third highest-expressed gene with functions in virus movement and assembly. These results show that CTV needs more copies of P13, P20, and CPm as they are related to the virus infection, replication and movement. SNV profiling on CTV draft genome Using RNA-Seq data, this study was able to determine the SNVs on an unprecedented scale for the CTV population. For better coverage and access to all micro and macro variants of CTV in the citrus tree, we mapped the raw reads to the CTV sequence obtained from this study. Identification of these SNVs is critical to understanding the potential of CTV variation. SNV was determined by annotating ORFs for the virus population. At least 15 sites displayed substantial differences with frequency ranging from 33% to 100% across the mapped reads when applied with a threshold of 1% for SNV detection. We filtered synonymous SNVs that did not change amino acids. The positions of SNVs in the coding regions of CTV annotated ORFs were in the polyprotein ORF (13 SNVs), P33 ORF (1 SNV) and P18 ORF (1 SNV) (Table 2). The CP genes were the most conserved and represented the same SNVs reported by PLOS ONE \| https://doi.org/10.1371/journal.pone.0288068 June 29, 2023 8 / 12 PLOS ONEA potential new genotype of Citrus tristeza virus in Iran Table 2. SNVs in the full-length genome of CTV, Sari isolate. Name Polyprotein Glu236Asp (Papain-like protease) Polyprotein Ala242Val (Papain-like protease) Polyprotein Ile246Thr (Papain-like protease) Polyprotein Arg248Pro (Papain-like protease) Polyprotein Ile252Leu (Papain-like protease) Polyprotein Leu255Arg (Papain-like protease) Polyprotein Ala313Val (Papain-like protease) Polyprotein Tyr323His (Papain-like protease) Polyprotein Arg327Trp (Papain-like protease) Polyprotein Thr2748Ala (Replicase) Polyprotein Asp2919Glu (Replicase) Polyprotein Arg2931Pro (Replicase) Polyprotein Val2934Ala (Replicase) P33 Val111Leu P18 Ala166Val https://doi.org/10.1371/journal.pone.0288068.t002 Type SNV SNV SNV SNV MNV MNV SNV SNV SNV SNV MNV SNV MNV SNV SNV Reference Allele Count Coverage Frequency A C T G GA TC C T A A TT G TA G C T T C C CC GA T C T G AC C CG T T 3 3 3 3 3 3 3 3 4 3 3 3 3 2 2 3 3 3 3 3 3 3 3 4 3 3 3 3 3 6 100 100 100 100 100 100 100 100 100 100 100 100 100 66.66 33.33 previous studies [24]. Most SNVs in polyprotein were on papain-like proteins and four SNVs were observed in replication protein. Fig 6 shows the exact location of SNVs in the tertiary structure of the polyprotein which may suggest potential areas for further research into the mechanisms underlying virus-host interactions and understanding the molecular mechanisms underlying the effects of specific SNVs on the virus. Polyproteins are the biggest protein in CTV and have several functions. The P33 protein has multiple functions. It is a unique non-conserved movement protein that also interacts with plant immunity [25]. In addition, the P33 has an important role in the ability of a CTV variant to protect the host from superinfection by a second closely related Fig 6. The tertiary structure of polyprotein and location of SNVs (red color). A and B are figures of polyproteins from different views. https://doi.org/10.1371/journal.pone.0288068.g006 PLOS ONE \| https://doi.org/10.1371/journal.pone.0288068 June 29, 2023 9 / 12 PLOS ONEA potential new genotype of Citrus tristeza virus in Iran virus variant [26]. While the CP protein of the virus is involved in virion assembly and virus translocation, P33 controls the asymmetrical accumulation of the positive- and negative- stranded RNAs during viral replication [27]. A recent study demonstrated that p33 participates in many different viral processes, and interacts with multiple protein partners such as CP, p20, and p23 [21]. The P33, P18, and P13 have been shown to expand the virus host range. For instance, several citrus genotypes can be infected with virus mutants containing deletions in those three genes [20]. The differences between strains in the different regions showed the recombination in the parental CTV genome that changed over a long time in different hosts [28]. It has been sug- gested that if a mixture of severe and mild strains of CTV exist in infected plant cells at differ- ent levels, the restriction of disease development can happen by a larger amount of mild viral genome, even though the other strain of the virus remains at low ratios in the viral populations [29]. Mixed viral infections of citrus trees are usually expected as they may be visited by virulif- erous aphids multiple times leading to the transmission of different strains of CTV to the same tree. Consequently, it is supposed that in the numerous infected citrus trees, there are co-infec- tions of different strains of the virus [29]. The distribution of the mild strains of CTV in a region implies good adaptation of the virus to its host [29,30]. This study does not provide direct evidence of mixed infections of different strains or variants in the sample or sequences. However, our findings do support the idea that mixed infections are possible in citrus trees. Previous studies have shown that the CTV genotypes can be changed after passage through different hosts [31]. Studies demonstrated the existence of genotype population modification in the genotype of a single isolate after passage through two different hosts, demonstrating that the presence and dominance of population genotypes were modified by virus transmission from sweet orange to Mexican lime [31,32]. So, CTV has a variation potential in a population in a host which our results confirm previous studies and these variations can support CTV for best fit in a different situation. Conclusion We sequenced and analyzed the complete genome of a CTV isolate for the first time from Iran. Phylogenetic analysis of the whole genome shows that this isolate is distinct from other isolates in the GenBank. Additionally, our study of the differential gene expression of the virus revealed that the P13 ORF was highly expressed in infected plants. Previous research has sug- gested that P13 plays a key role in the systemic infection and host range of the virus. Therefore, the high expression of P13 in infected plants suggests that it may be important for the virus to successfully infect and replicate within its host. The potential implications of this finding include the possibility of targeting P13 for the development of new control strategies for CTV. Our study of the potential variation of the virus in citrus trees revealed that the Sari isolate of CTV had mild variation, which could contribute to the virus’s best fit in different environmen- tal conditions. This potential variation has also been reported in previous studies [31,32], which suggests that it could be a fitness tool for the virus. Our findings add to this body of research and provide new insights into the potential for CTV to adapt to different host environments. Supporting information S1 File. Similarity matrix of full-length genomes of the Sari isolate of CTV and the isolates from GenBank. (CSV) PLOS ONE \| https://doi.org/10.1371/journal.pone.0288068 June 29, 2023 10 / 12 PLOS ONEA potential new genotype of Citrus tristeza virus in Iran S1 Raw images. (TIF) Author Contributions Conceptualization: Abozar Ghorbani. Supervision: Keramatollah Izadpanah. Validation: Mohammad Mehdi Faghihi, Faezeh Falaki. Visualization: Faezeh Falaki. Writing – original draft: Abozar Ghorbani. Writing – review & editing: Mohammad Mehdi Faghihi, Keramatollah Izadpanah. References 1. Bar-Joseph M, Marcus R, Lee RF. The continuous challenge of citrus tristeza virus control. Annual Review of Phytopathology. 1989; 27(1):291–316. 2. Dawson WO, Garnsey SM, Tatineni S, Folimonova SY, Harper SJ, Gowda S. Citrus tristeza virus-host interactions. Frontiers in Microbiology. 2013; 4:88. https://doi.org/10.3389/fmicb.2013.00088 PMID: 23717303 3. Moreno P, Ambro´s S, Albiach-Martı´ MR, Guerri J, Pena L. Citrus tristeza virus: a pathogen that changed the course of the citrus industry. Molecular Plant Pathology. 2008; 9(2):251–68. https://doi.org/10.1111/ j.1364-3703.2007.00455.x PMID: 18705856 4. Yokomi R, Polek M, Gumpf D. Transmission and spread of Citrus tristeza virus in Central California. Cit- rus Tristeza Virus Complex and Tristeza Diseases American Phytopathological Society, St Paul, MN. 2010:151–65. 5. Delkhosh H, Tousi N, D’Onghia K. Historical review of Citrus tristeza virus and its vectors in Iran. Citrus tristeza virus and Toxoptera citricidus: a serious threat to the Mediterranean citrus industry. 2009:135– 40. 6. Benı´tez-Galeano MJ, Vallet T, Carrau L, Herna´ndez-Rodrı´guez L, Bertalmı´o A, Rivas F, et al. Complete genome sequence of a novel recombinant citrus tristeza virus, a resistance-breaking isolate from Uru- guay. Genome Announcements. 2018; 6(22):e00442–18. https://doi.org/10.1128/genomeA.00442-18 PMID: 29853501 7. Hajeri S, Killiny N, El-Mohtar C, Dawson WO, Gowda S. Citrus tristeza virus-based RNAi in citrus plants induces gene silencing in Diaphorina citri, a phloem-sap sucking insect vector of citrus greening disease (Huanglongbing). Journal of Biotechnology. 2014; 176:42–9. https://doi.org/10.1016/j.jbiotec.2014.02. 010 PMID: 24572372 8. Febres V, Ashoulin L, Mawassi M, Frank A, Bar-Joseph M, Manjunath K, et al. The p27 protein is pres- ent at one end of citrus tristeza virus particles. Phytopathology. 1996; 86(12):1331–5. 9. Pappu H, Pappu S, Niblett C, Lee R, Civerolo E. Comparative sequence analysis of the coat proteins of biologically distinct citrus tristeza closterovirus isolates. Virus Genes. 1993; 7:255–64. https://doi.org/ 10.1007/BF01702586 PMID: 8279124 10. Karasev A, Boyko V, Gowda S, Nikolaeva O, Hilf M, Koonin E, et al. Complete sequence of the citrus tristeza virus RNA genome. Virology. 1995; 208(2):511–20. https://doi.org/10.1006/viro.1995.1182 PMID: 7747424 11. Harper S, Dawson T, Pearson M. Isolates of Citrus tristeza virus that overcome Poncirus trifoliata resis- tance comprise a novel strain. Archives of Virology. 2010; 155:471–80. 12. Roy A, Ramachandran P, Brlansky R. Grouping and comparison of Indian citrus tristeza virus isolates based on coat protein gene sequences and restriction analysis patterns. Archives of Virology. 2003; 148:707–22. https://doi.org/10.1007/s00705-002-0947-7 PMID: 12664295 13. Biswas KK, Tarafdar A, Sharma SK. Complete genome sequence of mandarin decline Citrus tristeza virus of the Northeastern Himalayan hill region of India: comparative analyses determine recombinant. Archives of Virology. 2012; 157:579–83. https://doi.org/10.1007/s00705-011-1165-y PMID: 22160128 14. Wu G-W, Tang M, Wang G-P, Jin F-Y, Yang Z-K, Cheng L-J, et al. Genetic diversity and evolution of two capsid protein genes of citrus tristeza virus isolates from China. Archives of Virology. 2015; 160:787–94. https://doi.org/10.1007/s00705-014-2281-2 PMID: 25387862 PLOS ONE \| https://doi.org/10.1371/journal.pone.0288068 June 29, 2023 11 / 12 PLOS ONEA potential new genotype of Citrus tristeza virus in Iran 15. Wang J, Zhou T, Shen P, Zhang S, Cao M, Zhou Y, et al. Complete genome sequences of two novel genotypes of Citrus tristeza virus infecting Poncirus trifoliata in China. Journal of Plant Pathology. 2020; 102:903–7. 16. Korkmaz S, Karanfil A, Satar S, Uslu T, Koc¸ NK, C¸ evik B. Effects of graft and aphid transmission on the genetic diversity and population structure of Turkish citrus tristeza virus isolates. European Journal of Plant Pathology. 2022; 162(2):369–88. 17. Jones S, Baizan-Edge A, MacFarlane S, Torrance L. Viral diagnostics in plants using next generation sequencing: computational analysis in practice. Frontiers in Plant Science. 2017; 8:1770. https://doi. org/10.3389/fpls.2017.01770 PMID: 29123534 18. Ghorbani A, Izadpanah K, Dietzgen RG. Gene expression and population polymorphism of maize Ira- nian mosaic virus in Zea mays, and intracellular localization and interactions of viral N, P, and M pro- teins in Nicotiana benthamiana. Virus Genes. 2018; 54:290–6. 19. Huang Z, Rundell PA, Guan X, Powell CA. Detection and isolate differentiation of Citrus tristeza virus in infected field trees based on reverse transcription-polymerase chain reaction. Plant Disease. 2004; 88 (6):625–9. https://doi.org/10.1094/PDIS.2004.88.6.625 PMID: 30812582 20. Tatineni S, Robertson CJ, Garnsey SM, Dawson WO. A plant virus evolved by acquiring multiple non- conserved genes to extend its host range. Proceedings of the National Academy of Sciences. 2011; 108(42):17366–71. https://doi.org/10.1073/pnas.1113227108 PMID: 21987809 21. Dao TNM, Kang S-H, Bak A, Folimonova SY. A non-conserved p33 protein of Citrus tristeza virus inter- acts with multiple viral partners. Molecular Plant-Microbe Interactions. 2020; 33(6):859–70. https://doi. org/10.1094/MPMI-11-19-0328-FI PMID: 32141354 22. Chen AY, Peng JH, Polek M, Tian T, Ludman M, Fa´tyol K, et al. Comparative analysis identifies amino acids critical for citrus tristeza virus (T36CA) encoded proteins involved in suppression of RNA silencing and differential systemic infection in two plant species. Molecular Plant Pathology. 2021; 22(1):64–76. https://doi.org/10.1111/mpp.13008 PMID: 33118689 23. Ruiz-Ruiz S, Navarro B, Peña L, Navarro L, Moreno P, Di Serio F, et al. Citrus tristeza virus: Host RNA silencing and virus counteraction. Citrus Tristeza Virus: Methods and Protocols. 2019;195–207. 24. Erkiş-Gu¨ngo¨ r G, C¸ evik B. Genetic Diversity and phylogenetic analysis of Citrus tristeza virus isolates from Turkey. Advances in Virology. 2019;2019. https://doi.org/10.1155/2019/7163747 PMID: 30906322 25. Sun YD, Folimonova SY. The p33 protein of Citrus tristeza virus affects viral pathogenicity by modulat- ing a host immune response. New Phytologist. 2019; 221(4):2039–53. https://doi.org/10.1111/nph. 15482 PMID: 30220089 26. Folimonova SY. Superinfection exclusion is an active virus-controlled function that requires a specific viral protein. Journal of Virology. 2012; 86(10):5554–61. https://doi.org/10.1128/JVI.00310-12 PMID: 22398285 27. Satyanarayana T, Gowda S, Ayllo´ n MaA, Albiach-Martı´ MaR, Rabindran S, Dawson WO. The p23 pro- tein of Citrus tristeza virus controls asymmetrical RNA accumulation. Journal of Virology. 2002; 76 (2):473–83. 28. Biswas KK, Palchoudhury S, Sharma SK, Saha B, Godara S, Ghosh DK, et al. Analyses of 3’half genome of citrus tristeza virus reveal existence of distinct virus genotypes in citrus growing regions of India. VirusDisease. 2018; 29:308–15. https://doi.org/10.1007/s13337-018-0456-2 PMID: 30159365 29. Herrera-Isidro´ n L, Ochoa-Sa´nchez JC, Rivera-Bustamante R, Martı´nez-Soriano JP. Sequence diversity on four ORFs of Citrus tristeza virus correlates with pathogenicity. Virology Journal. 2009; 6:1–10. 30. Harper SJ. Citrus tristeza virus: evolution of complex and varied genotypic groups. Frontiers in Microbi- ology. 2013; 4:93. https://doi.org/10.3389/fmicb.2013.00093 PMID: 23630519 31. Camps R, Fiore N, Riquelme N, Barros-Parada W, Besoain X. Genotype variation of citrus tristeza virus after passage on different hosts, and changes in the virus genotype populations by the vector Aphis gossypii. Phytopathologia Mediterranea. 2022; 61(1):55–63. 32. Harper S, Cowell S, Dawson W. Finding balance: Virus populations reach equilibrium during the infec- tion process. Virology. 2015; 485:205–12. https://doi.org/10.1016/j.virol.2015.07.017 PMID: 26291064 PLOS ONE \| https://doi.org/10.1371/journal.pone.0288068 June 29, 2023 12 / 12 PLOS ONE
10.2196_44326	JOURNAL OF MEDICAL INTERNET RESEARCH Martins et al Original Paper Consensus on the Terms and Procedures for Planning and Reporting a Usability Evaluation of Health-Related Digital Solutions: Delphi Study and a Resulting Checklist Ana Isabel Martins1, PhD; Gonçalo Santinha2, PhD; Ana Margarida Almeida3, PhD; Óscar Ribeiro4, PhD; Telmo Silva3, PhD; Nelson Rocha5, PhD; Anabela G Silva6, PhD 1Center for Health Technology and Services Research, University of Aveiro, Aveiro, Portugal 2Governance, Competitiveness and Public Policies, Department of Social, Political and Territorial Sciences, University of Aveiro, Aveiro, Portugal 3Digital Media and Interaction Research Centre, Department of Communication and Art, University of Aveiro, Aveiro, Portugal 4Center for Health Technology and Services Research, Department of Education and Psychology, University of Aveiro, Aveiro, Portugal 5Institute of Electronics and Informatics Engineering of Aveiro, Department of Medical Sciences, University of Aveiro, Aveiro, Portugal 6Center for Health Technology and Services Research, School of Health Sciences, University of Aveiro, Aveiro, Portugal Corresponding Author: Anabela G Silva, PhD Center for Health Technology and Services Research School of Health Sciences University of Aveiro Campus Universitário de Santiago Aveiro, 3810-193 Portugal Phone: 351 234370200 Email: asilva@ua.pt Abstract Background: Usability evaluation both by experts and target users is an integral part of the process of developing and assessing digital solutions. Usability evaluation improves the probability of having digital solutions that are easier, safer, more efficient, and more pleasant to use. However, despite the widespread recognition of the importance of usability evaluation, there is a lack of research and consensus on related concepts and reporting standards. Objective: The aim of the study is to generate consensus on terms and procedures that should be considered when planning and reporting a study on a usability evaluation of health-related digital solutions both by users and experts and provide a checklist that can easily be used by researchers when conducting their usability studies. Methods: A Delphi study with 2 rounds was conducted with a panel of international participants experienced in usability evaluation. In the first round, they were asked to comment on definitions, rate the importance of preidentified methodological procedures using a 9-item Likert scale, and suggest additional procedures. In the second round, experienced participants were asked to reappraise the relevance of each procedure informed by round 1 results. Consensus on the relevance of each item was defined a priori when at least 70% or more experienced participants scored an item 7 to 9 and less than 15% of participants scored the same item 1 to 3. Results: A total of 30 participants (n=20 females) from 11 different countries entered the Delphi study with a mean age of 37.2 (SD 7.7) years. Agreement was achieved on the definitions for all usability evaluation–related terms proposed (usability assessment moderator, participant, usability evaluation method, usability evaluation technique, tasks, usability evaluation environment, usability evaluator, and domain evaluator). A total of 38 procedures related to usability evaluation planning and reporting were identified across rounds (28 were related to usability evaluation involving users and 10 related to usability evaluation involving experts). Consensus on the relevance was achieved for 23 (82%) of the procedures related to usability evaluation involving users and for 7 (70%) of the usability evaluation procedures involving experts. A checklist was proposed that can guide authors when designing and reporting usability studies. Conclusions: This study proposes a set of terms and respective definitions as well as a checklist to guide the planning and reporting of usability evaluation studies, constituting an important step toward a more standardized approach in the field of https://www.jmir.org/2023/1/e44326 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e44326 \| p. 1 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Martins et al usability evaluation that may contribute to enhancing the quality of planning and reporting usability studies. Future studies can contribute to further validating this study work by refining the definitions, assessing the practical applicability of the checklist, or assessing whether using this checklist results in higher-quality digital solutions. (J Med Internet Res 2023;25:e44326) doi: 10.2196/44326 KEYWORDS usability evaluation; Delphi study; user-centered design; design; usability; evaluation; process; development; user; digital; efficient; reporting; quality; applicability Introduction Background Usability evaluation is essential to ensure the adaptation of digital solutions to their users [1,2]. Usability evaluation is defined as the evaluation of the extent to which a product can be used by specified users to achieve specified goals with effectiveness, efficiency, and satisfaction in a specific context of use [3]. When usability evaluation is part of the development process of digital solutions, these are more likely to allow an interaction that is intuitive, efficient, memorable, effective, and pleasant, which is key for user acceptance and digital solutions dissemination [4]. Poor usability impacts the quality of digital solutions and undermines the objective of ensuring that it is suitable for their users [5]. Usability evaluation methods can be based on expert analysis (inspection methods) or on real user data (test and inquiry methods). Within each method, there are numerous evaluation techniques that must be selected according to the characteristics of users and the stage of development of the digital solution to be evaluated [6,7]. In most cases, usability evaluation involves a combination of several techniques [8]. Usability assumes amplified importance when referring to health digital solutions that intend to help preventing, diagnosing, treating, monitoring, or alleviating a disease or injury in human beings [9,10]. The errors and problems related to digital solutions that occur during the process of interaction between users and digital solutions in the real context of use [11,12] could be avoided, at least partially, if a comprehensive usability evaluation performed continuously throughout the design, development, and implementation process of the health-related digital solution had been implemented [13,14]. Usability testing creates opportunities to make digital health solutions easier, safer, more efficient, and pleasant to use [15,16]. These improved interactive qualities benefit not only the user (patient or caregiver) but also the manufacturer and society. Pressing the wrong button, misreading a number, misplacing a component, skipping a step, or overlooking a warning message are examples of potentially catastrophic actions that can be minimized with proper usability evaluation [15]. Another reason to conduct usability tests of health digital solutions is to meet the device regulators’ expectations [15]. There are already a series of regulations, standards, and guides that standardize the evaluation of usability [14,17]. Human factors engineering has been added to regulatory requirements to reduce the number of errors in the use of medical devices, develop intuitive devices, and reduce training costs for both manufacturers and end users [11]. In addition, usability assessment is now integrated into https://www.jmir.org/2023/1/e44326 XSL•FO RenderX the design and development of health software, and the need to consider the usability of health information technologies is widely accepted [18]. However, despite the widespread recognition of the importance of usability evaluation, there is a lack of research on consensus on related concepts and reporting of usability assessment [19]. Previous studies have highlighted the need to improve the quality of health-related digital solutions. For example, a review of the quality and content of mobile apps to support lifestyle modifications following a stroke has concluded that overall quality was low [20]. A similar conclusion was reported in a study on the quality of smoking cessation apps [21]. Another systematic review on the methodological quality of mobile apps for pain management and assessment concluded that studies fail to report on several important methodological aspects regarding the evaluation of usability, including the use of valid and reliable measurement instruments or the previous experience of the investigator who conducted the evaluation [22]. Similar findings were reported in a recent scoping review aiming to synthesize the characteristics and procedures reported in the existing literature on the usability evaluation of digital solutions relevant to older adults [19]. A few attempts have been made to provide guidance for good evaluation practice in health informatics [23]. However, these guidelines are not specific to usability [23], not for research purposes [24], and lack detail and specification, which might explain why its use is not widespread. An objective and simple tool that clearly identifies what should be considered and how it should be considered when planning and reporting a usability study are needed. Objective A lack of standardized terms and procedures and good practices across the studies on usability evaluation has been identified, such as failure to report on the characteristics of study evaluators and participants, detail the tasks used for usability evaluation, or triangulate methods and techniques in the evaluation of usability. It is unclear whether the lack of information provided in the manuscripts results only from poor reporting as poor reporting may also reflect insufficient planning. Overall, these findings suggest that there is a need for consensus on the planning and reporting of studies on usability evaluation. Furthermore, a systematic review of criteria to assess mobile health apps and respective definitions found great diversity across the literature, further reinforcing the need for consensus, which may help improve the existing tools [25]. Therefore, this study aims to generate consensus on the terms and procedures that should be considered when planning and reporting a study on usability evaluation both by users and experts and provide a checklist that can easily be used by researchers when J Med Internet Res 2023 \| vol. 25 \| e44326 \| p. 2 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Martins et al conducting their usability studies. Consensus on the planning and reporting of usability studies is likely to improve the quality and comparability of usability results across studies and facilitate further research on the impact of usability on the acceptance and use of digital solutions. Conceivably, it might also contribute to increasing the probability that usability issues are detected and corrected before final solutions reach the market. This level of standardization already exists in other health areas. A concrete example of standardization is the CONSORT (Consolidated Standards of Reporting Trials) Statement [26], which is an evidence-based that presents a minimum set of recommendations for reporting randomized trials and offers a standard way for authors to prepare reports of trial findings. tool Methods Ethical Considerations This study was conducted according to the ethical principles that have their origin in the Declaration of Helsinki and was approved by the Data Protection Office of the University of Aveiro (nº27). All participants read the participant information sheet outlining the study objectives and procedures and gave informed consent before entering the study. Delphi Method Overview The Delphi method is a structured process that includes several phases and relies on experts to reach a consensus on a specific topic. In a Delphi study, the experts group participates in several rounds and the results of previous rounds are supplied with each new round, so that the experts are able to reconsider their judgments, revising them when appropriate [27,28]. This method was chosen because it is recommended when the aim is to determine consensus for a predefined problem on which there is little information available, and one must rely on the opinion of experts [28]. Furthermore, previous publications from this study authors [19,29] have shown that usability evaluation is a field where there is high heterogeneity in terms of reporting and meaning of concepts that would benefit from contrasting and congregating opinions from experienced individuals and by allowing the possibility of analyzing their answers in light of other experienced individuals’ answers as facilitated by the Delphi method. Delphi Survey Preparation Two previously conducted scoping reviews on procedures of usability evaluation for digital solutions contributed to the identification of the terms and definitions, as well as a list of items regarding procedures of usability evaluation both with users and experts [19,29] that were sent to participants in round 1 of the Delphi. The terms identified as being used with inconsistent meanings across the studies were usability assessment moderator, participant, usability evaluation method, usability evaluation technique, tasks, usability evaluation environment, usability evaluator, and domain evaluator. Based on the findings of the same review, a definition was proposed for each term. In addition, the reviews allowed the identification of a list of items regarding the procedures of usability evaluation, including 29 items: 6 related to the usability assessment https://www.jmir.org/2023/1/e44326 XSL•FO RenderX moderator, 6 to the participants, 5 to usability evaluation methods and usability evaluation techniques, 2 related to tasks, 2 related to the usability evaluation environment, 5 related to the usability evaluator and the domain evaluator, and 3 related with the inspection method. Innovation program Recruitment of Participants for the Delphi Invitations to enter the Delphi were sent to the Coordinators of the European Projects of the Health and Care Cluster, Horizon (ACTIVAGE 2020 Research and [Activating Innovative IoT Smart Living Environments for Ageing Well], ADLIFE [Integrated Personalized Care for Patients With Advanced Chronic Diseases to Improve Health and Quality of Life] Project, FAITH [Federated Artificial Intelligence Solution for Monitoring Mental Health Status After Cancer Treatment], Gatekeeper, InteropEHRate [Interoperable Electronic Health Records at user edge], Pharaon, SMART BEAR, and Smart4Health) who were asked to send the invitation to all partners in the project or to individual participants (Smart and Health Ageing through People Engaging in Supportive Systems [SHAPES]). To enter the Delphi and be considered experienced on usability, at least one of the following criteria had to be met: (1) have 2 publications on usability evaluation, (2) have participated in the evaluation of usability for at least 2 projects, or (3) have designed at least 2 studies on usability. A sample size of at least 20 participants has been suggested as appropriate [27,28,30]. Data Collection and Analysis Overview This Delphi study was organized in 2 rounds and took place between December 2020 and March 2021 and was held on the internet. In both rounds, individual participants (for the SHAPES project as authors are partners in this project) or the coordinators of the European projects (for the remaining projects) were sent an email explaining the study objectives with a link to a survey, which also included the participant information sheet and the informed consent. Participants remained anonymous during the whole study. The anonymity of participants was kept across the 2 rounds. Round 1 The survey for the first round of the Delphi was divided into 2 parts. The first consisted of a list of terms and respective definitions. Experienced participants were asked whether they agreed or disagreed with the terms and definitions, and then to provide a comment or propose alternative definitions. The second part of the survey was on the procedures of usability evaluation. The study participants were asked to rate the importance of each of the procedures for the planning and reporting of a study on usability evaluation using a 9-item Likert scale (1—“item not at all important” to 9— “item very important”) and add any other procedure that, in their opinion, was important and was not already included in the list of procedures provided. Participants were also asked to provide basic demographic information (age, sex, and country of origin) and professional background. Once the first round of the Delphi was completed, the results were collated. For the terms and definitions, the suggestions J Med Internet Res 2023 \| vol. 25 \| e44326 \| p. 3 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Martins et al given by the experienced participants were aggregated by definition and then analyzed by a panel of 3 members of our team (AIM—a gerontologist and an expert on usability with more than 10 years of experience conducting studies on usability assessment, AGS—a physiotherapist involved in usability for more than 10 years, and NR—an engineer with more than 30 years of experience conducting and leading research on usability) against the following two criteria: (1) the number of participants giving similar suggestions and (2) the level of consensus already achieved on the definition. Based on these criteria and the overall analysis of the suggested changes and commentaries, this panel decided on the final terms and respective definitions. The consensus was defined a priori as having at least 80% agreement [26] for each term and definition. The agreed definitions were then included in round 2. The same panel analyzed the commentaries and suggestions of new items for the list of procedures of usability evaluation using the same 2 criteria and an additional 1 regarding whether suggestions were specific to the planning and reporting of a study on usability evaluation (rather than general suggestions that would apply to any study). The new items that resulted from this analysis were added to the list of procedures and included in the next round. For items already included in round 1, an additional analysis was made; it consisted of the calculation of Figure 1. Delphi method diagram. the number and percentage of ratings attributed to each item grouped from 1 to 3, 4 to 6, and 7 to 9. Round 2 In this round, experienced participants were asked to reappraise the relevance of each procedure. The graphical representation of round 1 results informed the participants and served as a basis for their decision-making on the degree of importance they wanted to assign to each specific item. Care was taken in writing neutral instructions to minimize influencing participants’ responses. After round 2, the results were analyzed. Consensus on the relevance of each item was defined a priori when at least 70% or more experienced participants scored an item 7 to 9 and less than 15% of participants scored the same item 1 to 3. Consensus on the irrelevance of an item was considered when 70% or more of participants scored the item 1 to 3 and less than 15% of participants scored the item 7 to 9 on the Likert scale [31]. In both rounds, experienced participants were given 3 weeks to complete the survey and were sent 1 to 2 reminders. Participants’ identity was not disclosed at any time. The diagram represented in Figure 1 summarizes the method implemented for the Delphi study. Results Participants’ Characteristics A total of 30 different participants entered the Delphi study: 29 in the first round and 27 in the second round. Participants were from 11 different countries (Table 1), had a mean age of 37.2 (SD 7.7) years, and had experience in designing usability evaluation studies, evaluating usability, or publishing on usability evaluation. Most of them were females (n=20, 67%), with a background related to communication and technology sciences (n=8, 27%) or computer and biomedical engineering (n=12, 40%). https://www.jmir.org/2023/1/e44326 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e44326 \| p. 4 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Martins et al Table 1. Participants’ characteristics (N=30). Characteristics Gender Female Male Prefer not to disclose Background Value, n (%) 20 (67) 9 (30) 1 (3) Health and related areas (eg, eHealth researcher, rehabilitation specialist, medical device specialist, and gerontologist) 5 (17) Social sciences (eg, psychologist, accessibility specialist, and learning technologies specialist) Communication and technology sciences (eg, usability researcher, technology manager, and assistive technology developer) Computer and biomedical engineering (eg, computer science researcher, software developer, bioengineer, and robotics engineer) Country Portugal Spain Italy Greece Germany France Belgium Switzerland England Netherlands Norway 5 (17) 8 (27) 12 (40) 10 (33) 5(17) 4 (13) 3 (10) 2 (7) 1 (3) 1 (3) 1 (3) 1 (3) 1 (3) 1 (3) Experience with usability evaluation At least designed 2 studies + participated in the evaluation of usability for at least 2 projects and had 2 usability evaluation publications 8 (27) At least designed 2 studies and participated in the evaluation of usability for at least 2 projects At least 2 usability evaluation publications At least participated in the evaluation of usability for at least 2 projects 6 (20) 3 (10) 13 (43) Delphi Rounds Round 1—Agreement on Terms and Definitions The agreement on the terms and respective definitions varied between a minimum of 82.8% (n=24) for the definition of “usability assessment moderator” and 100% (n=29) for the definition of “tasks.” As the predefined minimum agreement rate of 80% was achieved, definitions were not included in round 2 of the Delphi. Nevertheless, 4 definitions were amended following experienced participants’ comments and suggestions (Table 2). https://www.jmir.org/2023/1/e44326 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e44326 \| p. 5 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Martins et al Table 2. Results of the agreement and changes made to definitions following round 1 of the Delphi (N=29). Term Proposed definition Agreement, n (%) Final definition Usability assessment moderator The person who conducts the usability evaluation inter- acts with the participant and guides the session. 24 (83) The same as the initially proposed Participant The person who is asked to evaluate the usability of a product or service and who completes the tasks, follow- ing the indications of the evaluator. Usability evaluation method A set of techniques used to perform usability evaluation at different stages of the product or service development (eg, inquirya or testb methods). 27 (93) The same as the initially proposed 25 (86) The same as the initially proposed Usability evaluation technique A set of procedures used to perform the usability evalu- ation and collect data of a certain type (eg, brainstorm- ing, questionnaire, and think aloud). 25 (86) Tasks The activities that participants are asked to perform when evaluating the usability of a product or service. 29 (100) A set of procedures used to perform the us- ability evaluation and collect either qualita- tive or quantitative data (eg, focus group, survey, think aloud, and performance) Self-contained or independent activities that participants are asked to perform when eval- uating the usability of a product or service within a limited period Usability evaluation environment Usability evaluator The environment where the evaluation of usability takes place: (1) laboratory or controlled conditions and (2) in a real context, that is, the usability evaluation is carried out in the same context and circumstances where the end product is expected to be used. A person with knowledge and experience on HCIc, us- ability, and user experience who conducts the usability inspection (eg, an HCI specialist assigned to evaluate the interface of technology for patients with diabetes). 27 (93) The same as the initially proposed 29 (100) The same as the initially proposed Domain evaluator A person with knowledge on the application area of the technology under development (eg, a physician involved in treating patients with diabetes). 28 (97) A person with knowledge of the application area of the technology under development that provides feedback about functionalities of the technological product or service (eg, a physician involved in treating patients with diabetes who provides feedback about tech- nology for patients with diabetes) aInquiry methods involve collecting qualitative data from users. bTesting methods involve observing the user while interacting with the product or service and consist of collecting mostly quantitative data. cHCI: human-computer interaction. Round 2—Procedures for Usability Evaluation In round 1, experienced participants suggested a total of 24 additional items for usability evaluation involving users. Of these, 6 were considered to be repeated items (eg, “have experience with usability methods and techniques,” or “know the methods well and be comfortable in applying the techniques”), 6 were not specific to usability evaluation (eg, “follow ethical guidelines and comply with data protection recommendations” or “create consent the participants”), and another 5 were out of scope (eg, “stress the relevance of the stakeholders and decision makers standpoints” or “detect and match the participants with appropriate user personas”) and were not considered. Therefore, 7 new items were added to the initial list of items and submitted for round 2. In addition, 2 of the items included in round 1 were amended (Table 3). A total of 28 items on procedures of usability evaluation with users were included in round 2. forms for https://www.jmir.org/2023/1/e44326 XSL•FO RenderX For procedures of usability evaluation regarding experts, 5 new items resulted from the participants’ suggestions in round 1. Of these, 1 was repeated (“Always use a combination of experts from different domains”) and 2 were not specific for usability planning or reporting (eg, the session should be structured in such a way that avoids bias), hence were not considered, resulting in the inclusion of 2 items in round 2 (Table 3). In addition, the wording of 1 item from round 1 was rephrased for clarification. A total of 10 items on usability evaluation with experts were included in round 2. Consensus on relevance was reached for 23 (82%) of the 28 items on procedures of usability evaluation with users (Table 4). For the remaining 5 items, consensus was not achieved neither on its relevance nor on its irrelevance.. Consensus on relevance was reached for 7 (70%) of the 10 items on procedures of usability evaluation with experts (Table 5). For the remaining 3 items consensus was not achieved neither on its relevance nor on its irrelevance. J Med Internet Res 2023 \| vol. 25 \| e44326 \| p. 6 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Martins et al Table 3. Results from experienced participants’ suggestions on the procedures for usability evaluation involving users. Usability evaluation Items that were rephrased, n New items proposed by ex- perienced participants, n Items excluded and reason, n New items includ- ed in round 2, n Usability evaluation involving users Usability assessment moderator Participants Usability evaluation methods and usability evaluation techniques Tasks Usability evaluation environment Usability evaluation involving experts Usability evaluator and domain evaluator Inspection method 1 0 1 0 0 0 1 4 8 3 7 2 3 2 aItems excluded as they were repeated. bItems excluded as they were not specific for usability planning and reporting. cItems excluded as they were out of scope. • • • • • • • • • • • • • 1a 2b 1a 3b 3c 1a 1b 2a 2c 1a 1a 1b 1b 1 1 1 3 1 1 1 https://www.jmir.org/2023/1/e44326 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e44326 \| p. 7 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Martins et al Table 4. Agreement on the items for usability evaluation involving users (N=27). Planning and reporting procedures for usability evaluation with users Usability assessment moderator Determine the number of usability assessment moderators. Provide the rationale used to establish the number of usability assessment moder- ators. Specify as inclusion criteria having previous experience with usability evaluation with users or consider adequate training and provide details of the training plan. Detail inclusion and exclusion criteria other than previous experience (eg, academic background or age). Specify whether the usability assessment moderators are external to the service or product development team. Specify if observers are included, define their responsibilities, and collect their characteristics (eg, gender, academic background, and previous experience in usability evaluation).b Detail the usability assessment moderators’ characteristics that should be collected (eg, gender, academic background, and previous experience conducting usability evaluation).c Items grouped from 1 to 3, n (%) Items grouped from 4 to 6, n (%) Items grouped from 7 to 9, n (%) 1 (4) 2 (7) 0 (0) 0 (0) 0 (0) 0 (0) 13 (48) 10 (37) 2 (7) 9 (33) 4 (15) 7 (26) 13 (48) 15 (56) 25 (93) a 18 (67) 13 (85) 20 (74) 0 (0) 9 (33) 18 (67) Participants Determine sample size (ie, the total number of participants involved in the evalu- ation). Provide a rationale to establish the sample size. Provide clear inclusion and exclusion criteria (eg, profile definition including age, gender, educational level, digital literacy, and previous experience using the product or service being evaluated). Provide sampling methods (eg, random, systematic, cluster, convenience, and snowball). Indicate the setting of participants’ recruitment (eg, community and hospital). Detail clinical conditions (if relevant for the study) (eg, asymptomatic or with a specific clinical condition or from a specific group—occupational group, the severity of the clinical condition, disabilities, cognitive impairment). Detail the participant’s characteristics that should be collected (such as age, gender, educational level, and digital literacy).c Usability evaluation method and usability evaluation technique Specify whether a combination of usability evaluation methods was used (eg, using both inquiry and test methods). Specify whether a combination of usability evaluation techniques was used (eg, for the inquiry method, combine the questionnaire and interview techniques). Provide the rationale for the choice of usability evaluation methods and techniques. Describe the usability evaluation methods and techniques used and how they are implemented.b When using measuring instruments such as scales or questionnaires, give indicators of their validity and reliability. Describe the data analysis plan for both quantitative and qualitative data.c Tasks Provide a detailed description of tasks or present the session script. Indicate the total number of tasks. Detail the task-related outcomes and how they are measured (eg, task completion and duration and number of errors).c 0 0 0 0 0 0 0 2 (7) 1 (4) 0 0 0 0 0 0 0 3 (11) 6 (22) 1 (4) 8 (30) 4 (15) 2 (7) 24 (89) 21 (78) 26 (96) 19 (70) 23 (85) 25 (93) 0 27 (100) 4 (15) 4 (15) 5 (19) 2 (7) 2 (7) 4 (15) 1 (4) 6 (22) 5 (19) 21 (78) 22 (82) 22 (82) 25 (93) 25 (93) 23 (85) 26 (96) 21 (78) 22 (82) https://www.jmir.org/2023/1/e44326 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e44326 \| p. 8 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Martins et al Planning and reporting procedures for usability evaluation with users Items grouped from 1 to 3, n (%) Items grouped from 4 to 6, n (%) Items grouped from 7 to 9, n (%) Detail the conditions for carrying out the tasks (eg, with or without supervision, individually or in the group, with or without a period for familiarization with the digital product or service).c Detail the instructions to participants and the way they are presented (eg, verbally, written, and both) and registered (eg, audio, video, screen recorder, and notes from an observer).c Usability evaluation environment Justify the choice of the usability evaluation environment (eg, lab or field test and remote or face-to-face test). Specify usability evaluation environment requirements (eg, recording equipment or observer room availability). Detail the procedures to make the usability evaluation environment safe and comfortable for the participants.c 0 0 0 0 0 aValues in italics denote the items that reached consensus on inclusion. bItems that were rephrased. cNew items that emerged from round 1. 0 27 (100) 2 (7) 27 (100) 8 (30) 5 (19) 9 (33) 19 (70) 22 (82) 18 (67) Table 5. Agreement on items for usability evaluation involving experts (N=27). Planning and reporting procedures for usability evaluation with experts Items grouped from 1 to 3, n (%) Items grouped from 4 to 6, n (%) Items grouped from 7 to 9, n (%) Usability evaluator and the domain evaluator Determine the number of evaluators involved in the evaluation. Provide the rationale to establish the number of evaluators. Define as inclusion criteria having previous experience in inspection usability evaluation or consider adequate training and provide details of training. State whether the evaluators are external to the product or service development team. Specify whether a combination of evaluators from different domains was used (eg, for a health-related digital service, use both usability and health domain evaluators). Provide clear inclusion and exclusion criteria.b Inspection method Detail the protocol to conduct the inspection (including the techniques used and how they are implemented).c State whether a combination of techniques was used (eg, heuristic evaluation and cognitive walkthrough). Provide the rationale for the choice of the techniques. Detail the criteria to prioritize the resolution of problems identified (eg, ac- cording to the severity criteria, problems with higher impact on users are solved first).a 0 (0) 1 (4) 0 (0) 1 (4) 0 (0) 1 (4) 0 (0) 2 (7) 1 (4) 0 (0) aValues in italics denote the items that reached consensus on inclusion. bNew items that emerged from round 1. cItems that were rephrased. 10 (37) 14 (52) 5 (19) 17 (63) 12 (44) 22 (82) a 11 (41) 15 (56) 3 (11) 24 (89) 1 (4) 1 (4) 6 (22) 5 (19) 4 (15) 25 (93) 26 (96) 19 (70) 21 (78) 23 (85) https://www.jmir.org/2023/1/e44326 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e44326 \| p. 9 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Martins et al Discussion Principal Results and Comparison With Previous Work The results of this Delphi study are a set of agreed definitions of common terms used in the field of usability evaluation and a checklist of procedures that should guide the planning and reporting of usability evaluation studies. This consensus was achieved with a panel of international-experienced participants in usability evaluation, and these findings provide an important step toward a more standardized approach in the field of usability evaluation of digital health solutions for which the intended user is a layperson. We believe that the results of this study are particularly relevant for the evaluation of digital health solutions as they include terms and items that are specific to this field, such as the domain evaluator or details of the clinical conditions of the participants involved in the usability evaluation. Nevertheless, the general nature of most items suggests that the checklist is also relevant to inform usability evaluation in other fields. Of the 38 items on procedures included in the Delphi, consensus on their relevance for the planning and reporting of usability studies was reached for 30. For the remaining 8, it was not possible to reach a consensus neither on their relevance nor on their irrelevance. These items were scored by more than 90% of participants with a rating of 4 or higher suggesting that they were considered moderately to highly relevant. Therefore, these items were also included in a checklist developed to facilitate the planning and reporting of usability studies (Multimedia Appendix 1). Nevertheless, results might suggest a division of the 38 items into 30 items that should be considered for all studies and 8 items that are important but not essential. Interestingly, the items for which consensus was not reached report mainly to aspects of the usability evaluation moderator (for evaluation involving users) and usability evaluator and domain evaluator (for evaluation involving experts), including the rationale for characteristics and sample size, inclusion criteria, and personal characteristics, which are seldom reported in individual characteristics of the person who conducts the usability evaluation and interacts with the participant guiding the session might impact the results as shown in previous studies [33-35]. The so-called “evaluator effect” is well known and has a great influence on the evaluation results as usability evaluation involves direct contact of the evaluator with the participant and a certain amount of subjective interpretation [33,34] that might impact results. For example, body language and tone of voice might influence how the user evaluates the digital solution [35]. Both the lack of reporting on existing literature and the percentage of participants classifying these as less relevant items might suggest a lack of awareness of the implications of the evaluator impact on the process of evaluation and, consequently, on its results. [19,32]. Conceivably, literature the the A clear and distinctive characteristic of our checklist is its simplicity of use, objectivity, and high level of specification. For example, previous guidelines on evaluation in health informatics refer that users’ characteristics should be clearly https://www.jmir.org/2023/1/e44326 XSL•FO RenderX identified and defined but do not identify a minimum set of user characteristics that need to be considered across studies allowing for different authors to report on different characteristics and making comparisons across studies difficult. Contrary, our checklist, exemplifies what information should be provided (eg, age, gender, educational level, digital literacy, and existing clinical conditions). We hope that this checklist can be a contribution toward a more standardized approach and high-quality planning and reporting of usability studies. This is likely to result in more robust studies and more transparent reports, which increase the interpretability and transferability of the results. In addition, it is likely to (1) increase comparability across studies and consequently, the aggregation of higher amounts of data into meta-analysis, (2) higher percentage of errors being detected during evaluations, (3) overarching studies comparing the level of sensitivity of different usability methods and techniques, to name a few potential gains for the field of usability. We suggest that the list of terms and respective definitions should be used with the checklist to guarantee a common understanding. Furthermore, we acknowledge that the items of the checklist differ in the level of complexity and specificity. For example, while some refer to objectively stating what was done in the study (eg, “state whether the usability assessment moderators are external to service or product development team”), others report on the rationale of the decisions (eg, “provide a rationale to establish the sample size”), but we believe that this translates what is expected in the Methods section of a manuscript, where one needs to report both on methodological procedures and their justification. The checklist includes items that tend to be specific to usability studies, and important aspects such as data analysis, which is transversal to all studies, are not included. In these cases, authors should refer to existing guidelines and checklists, such as the Consolidated Criteria for Reporting Qualitative Research (COREQ) [36]. Given the lack of consensus for usability reporting, the process adopted in this Delphi method, which was conducted on the internet, was appropriate to assemble the views of an international panel of experienced participants on usability evaluation. In addition to identifying areas of consensus, this study was able to highlight areas where there is less certainty in the usability field, potentially requiring further research. One of the advantages of this method is that it allows participants to suggest new items that were not initially foreseen. Although the items sent to participants of the Delphi in the first round resulted from an extensive literature review [19,29], 9 new items were added to the initial list in the first round. Of the new items proposed by participants in the first round, the majority (7 out of 9) reached a consensus with only 1 round, which shows the adequacy and value of the Delphi method. Strengths and Limitations The strength of the proposed checklist is that it was developed based on the perspective of an international panel of participants following a detailed analysis of existing evidence [27,28]. Being simultaneously a checklist to inform planning and reporting, it helps ensure that the important methodological aspects that need to be reported are also considered during the study planning. J Med Internet Res 2023 \| vol. 25 \| e44326 \| p. 10 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Martins et al Although 30 participants are a number considered to be reasonable [27,28,37], it cannot be guaranteed that the views of the included participants are representative of the views of the broader community. In addition, Delphi participants were all from Europe, and most of them were from a small number of countries as half of the sample came from 2 countries (Portugal and Spain) potentially limiting the generalizability of the findings, particularly to those outside Europe. The participation rate cannot be calculated, as the invitation to participate in the Delphi study was spread across several European Projects of the Health and Care Cluster, Horizon 2020 Research and Innovation program, and there are no data on how many potential participants had access to the invitation to enter the Delphi. Furthermore, the inclusion criteria to be considered an experienced participant on usability were broad. Although they were selected among participants of European projects with a strong focus on usability and clarifications were made that participants had to be experienced on usability assessment, our inclusion criteria might not have been robust against the inclusion of individuals without an in-depth knowledge or experience of usability evaluation. Conclusions This study proposes a set of terms and respective definitions and a checklist to guide the planning and reporting of usability evaluation studies both in the health area as well as for usability in general. These can be used both to guide the planning and reporting of usability evaluation studies as well as to inform quality assessment for these studies. Future studies can contribute to further validating this study work by refining the definitions, assessing the practical applicability of the current checklist for specific digital health solutions, or assessing whether using this checklist results in higher-quality digital health solutions. Acknowledgments The research reported in this publication was supported by the SHAPES (Smart and Health Ageing through People Engaging in Supportive Systems), which is funded by the Horizon 2020 Framework Programme of the European Union for Research Innovation (grant 857159-SHAPES-H2020-SC1-FA-DTS-2018-2020). Data Availability The data sets generated and analyzed during this study are available from the corresponding author on request. Conflicts of Interest None declared. Multimedia Appendix 1 Glossary and checklists of procedures for planning and reporting procedures for usability evaluation with users and experts. [PDF File (Adobe PDF File), 152 KB-Multimedia Appendix 1] References 2. 3. 1. Martins AI, Queirós A, Rocha NP. Validation of a usability assessment instrument according to the evaluators’ perspective about the users’ performance. Univ Access Inf Soc 2019;19(3):515-525 [FREE Full text] [doi: 10.1007/s10209-019-00659-w] Nielsen J. Usability Engineering. London: Academic Press; 1993. Ergonomics of human-system interaction — part 210: human-centred design for interactive systems (ISO 9241-210:2019). International Organization for Standardization. 2019. URL: https://www.iso.org/standard/77520.html [accessed 2023-05-02] Dix A, Finlay J, Abowd GD, Beale R. Human-Computer Interaction, 3rd ed. Hoboken, NJ: Prentice Hall; 2004. 4. 5. Middleton B, Bloomrosen M, Dente MA, Hashmat B, Koppel R, Overhage JM, et al. Enhancing patient safety and quality of care by improving the usability of electronic health record systems: recommendations from AMIA. J Am Med Inform Assoc 2013;20(e1):e2-e8 [FREE Full text] [doi: 10.1136/amiajnl-2012-001458] [Medline: 23355463] 6. Martins AI, Queirós A, Silva AG, Rocha NP. Usability evaluation methods: a systematic review. In: Bajwa IS, Saeed S, Mahmood Z, editors. Human Factors in Software Development and Design. Hershey, PA: IGI Global; 2015:250-273 7. Maguire M. Methods to support human-centred design. Int J Hum Comput Stud 2001;55(4):587-634 [FREE Full text] [doi: 8. 10.1006/ijhc.2001.0503] Petrie H, Bevan N. The evaluation of accessibility, usability, and user experience. In: Stephanidis C, editor. The Universal Access Handbook. Boca Raton, FL: CRC Press; 2009:1-16 Dispositivos médicos. Infarmed. 2013. URL: http://www.webcitation.org/6OqjwM9NN [accessed 2023-05-02] 9. 10. European Union. Council directive of 20 June 1990 on the approximation of the laws of the member states relating to active implantable medical devices (90/385/EEC). Off J Eur Comm 1990;189:17-35 [FREE Full text] 11. Human factors engineering for medical devices. Emergo UL. 2014. URL: https://www.emergobyul.com/resources/human- factors-engineering-medical-devices [accessed 2023-05-02] 12. Kaye RD, North RA, Peterson MK. UPCARE: an analysis, description, and educational tool for medical device use problems. 2003 Presented at: Eighth Annual International Conference of Industrial Engineering Theory, Applications and Practice; https://www.jmir.org/2023/1/e44326 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e44326 \| p. 11 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Martins et al 2003; Las Vegas, NV p. 27 URL: https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=308668cf7912ca5ec10 cefeabc0772f1feb3bcf8 13. Woods D, Cook RI. The new look at error, safety, and failure: a primer for health care. Chicago National Patient Safety Foundation. 1999. URL: http://csel.eng.ohio-state.edu/productions/pexis/readings/submod1/NewLookPrimer.pdf [accessed 2023-04-05] Shaver E. Human Factors and Medical Device Resources. 2013. URL: https://www.ericshaver.com/?s=Human+Factors+and +Medical+Device+Resources [accessed 2023-05-02] 14. 15. Wiklund ME, Kendler J, Strochlic AY. Usability Testing of Medical Devices. 1st ed. Boca Raton, FL: CRC Press; 2011:412 16. Bastien JMC. Usability testing: a review of some methodological and technical aspects of the method. Int J Med Inform 2010;79(4):e18-e23 [FREE Full text] [doi: 10.1016/j.ijmedinf.2008.12.004] [Medline: 19345139] 17. Association for the Advancement of Medical Instrumentation. Medical devices—application of usability engineering to medical devices ANSI/AAMI/IEC 62366:2007 (R2013). ANSI Webstore. 2013. URL: https://webstore.ansi.org/standards/ aami/ansiaamiiec623662007r2013a1 [accessed 2023-05-02] 18. Christ-Neumann M, Escrich A, Anguita A, Stenzhorn H, Taylor M, Ramay H, et al. Usability on the p-medicine infrastructure: an extended usability concept. Ecancermedicalscience 2014;8:399 [FREE Full text] [doi: 10.3332/ecancer.2014.399] [Medline: 24567756] Silva AG, Caravau H, Martins A, Almeida AMP, Silva T, Ribeiro O, et al. Procedures of user-centered usability assessment for digital solutions: scoping review of reviews reporting on digital solutions relevant for older adults. JMIR Hum Factors 2021;8(1):e22774 [FREE Full text] [doi: 10.2196/22774] [Medline: 33439128] 19. 20. O'Connor SR, Kee F, Thompson DR, Cupples ME, Donnelly M, Heron N. A review of the quality and content of mobile apps to support lifestyle modifications following a transient ischaemic attack or 'minor' stroke. Digit Health 2021;7:1-9 [FREE Full text] [doi: 10.1177/20552076211065271] [Medline: 34950500] 21. Regmi D, Tobutt C, Shaban S. Quality and use of free smoking cessation apps for smartphones. Int J Technol Assess Health Care 2018;34(5):476-480 [FREE Full text] [doi: 10.1017/S0266462318000521] [Medline: 30226123] 22. Almeida AF, Rocha NP, Silva AG. Methodological quality of manuscripts reporting on the usability of mobile applications for pain assessment and management: a systematic review. Int J Environ Res Public Health 2020;17(3):785 [FREE Full text] [doi: 10.3390/ijerph17030785] [Medline: 32012674] 23. Nykänen P, Brender J, Talmon J, de Keizer N, Rigby M, Beuscart-Zephir M, et al. Guideline for good evaluation practice in health informatics (GEP-HI). Int J Med Inform 2011;80(12):815-827 [FREE Full text] [doi: 10.1016/j.ijmedinf.2011.08.004] [Medline: 21920809] 24. Bevan N. Common industry format usability tests. In: Proceedings of Usability Professionals Association. 1999 Presented at: UPA'98; June 29-July 2, 1999; Scottsdale, AZ URL: https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=c 321434c988f73477b12d03f6bebe5dece26e5cc 25. Nouri R, R Niakan Kalhori S, Ghazisaeedi M, Marchand G, Yasini M. Criteria for assessing the quality of mhealth apps: a systematic review. J Am Med Inform Assoc 2018;25(8):1089-1098 [FREE Full text] [doi: 10.1093/jamia/ocy050] [Medline: 29788283] 26. CONSORT Group. CONSORT—Consolidated Standards of Reporting Trials. URL: http://www.consort-statement.org/ [accessed 2021-03-30] 27. Hasson F, Keeney S, McKenna H. Research guidelines for the Delphi survey technique. J Adv Nurs 2000;32(4):1008-1015 [FREE Full text] [Medline: 11095242] 28. Hsu CC, Sandford BA. The Delphi technique: making sense of consensus. Pract Assessment Res Eval 2007;12(1):10 [FREE 29. Full text] [doi: 10.7275/pdz9-th90] Silva A, Isabel Martins A, Caravau H, Martins Almeida A, Silva T, Ribeiro O, et al. Experts evaluation of usability for digital solutions directed at older adults: a scoping review of review. New York, NY: Association for Computing Machinery; 2020 Presented at: 9th International Conference on Software Development and Technologies for Enhancing Accessibility and Fighting Info-exclusion; December 2-4, 2020; Virtual Event, Portugal p. 174-181 URL: https://dl.acm.org/doi/abs/ 10.1145/3439231.3439238 [doi: 10.1145/3439231.3439238] 30. Veugelers R, Gaakeer MI, Patka P, Huijsman R. Improving design choices in Delphi studies in medicine: the case of an exemplary physician multi-round panel study with 100% response. BMC Med Res Methodol 2020;20(1):156 [FREE Full text] [doi: 10.1186/s12874-020-01029-4] [Medline: 32539717] 31. Thorn JC, Brookes ST, Ridyard C, Riley R, Hughes DA, Wordsworth S, et al. Core items for a standardized resource use measure: expert Delphi consensus survey. Value Health 2018;21(6):640-649 [FREE Full text] [doi: 10.1016/j.jval.2017.06.011] [Medline: 29909868] 32. Ellsworth MA, Dziadzko M, O'Horo JC, Farrell AM, Zhang J, Herasevich V. An appraisal of published usability evaluations of electronic health records via systematic review. J Am Med Inform Assoc 2017;24(1):218-226 [FREE Full text] [doi: 10.1093/jamia/ocw046] [Medline: 27107451] 33. Hertzum M, Jacobsen NE. The evaluator effect: a chilling fact about usability evaluation methods. Int J Hum-Comput Interact 2003;15(1):183-204 [FREE Full text] [doi: 10.1207/s15327590ijhc1501_14] https://www.jmir.org/2023/1/e44326 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e44326 \| p. 12 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Martins et al 34. Hertzum M, Molich R, Jacobsen NE. What you get is what you see: revisiting the evaluator effect in usability tests. Behav Inf Technol 2014;33(2):144-162 [FREE Full text] [doi: 10.1080/0144929x.2013.783114] 35. Riihiaho S. Experiences with usability testing: effects of thinking aloud and moderator presence. In: Kirakowski J, Norman K, editors. The Wiley Handbook of Human Computer Interaction. Vol 1. Chichester, West Sussex: John Wiley & Sons Ltd; 2018:258-275 36. Tong A, Sainsbury P, Craig J. Consolidated criteria for reporting qualitative research (COREQ): a 32-item checklist for interviews and focus groups. Int J Qual Health Care 2007;19(6):349-357 [FREE Full text] [doi: 10.1093/intqhc/mzm042] [Medline: 17872937] 37. Murphy MK, Black NA, Lamping DL, McKee CM, Sanderson CF, Askham J, et al. Consensus development methods, and their use in clinical guideline development. Health Technol Assess 1998;2(3):i-iv, 1-i-iv,88 [Medline: 9561895] Abbreviations ACTIVAGE: Activating Innovative IoT Smart Living Environments for Ageing Well ADLIFE: Integrated Personalized Care for Patients With Advanced Chronic Diseases to Improve Health and Quality of Life CONSORT: Consolidated Standards of Reporting Trials COREQ: Consolidated Criteria for Reporting Qualitative Research FAITH: Federated Artificial Intelligence Solution for Monitoring Mental Health Status After Cancer Treatment InteropEHRate: Interoperable Electronic Health Records at user edge SHAPES: Smart and Health Ageing through People Engaging in Supportive Systems Edited by T Leung; submitted 15.11.22; peer-reviewed by B Chaudhry, P Worthy, J Kaipio; comments to author 29.01.23; revised version received 16.02.23; accepted 10.03.23; published 06.06.23 Please cite as: Martins AI, Santinha G, Almeida AM, Ribeiro Ó, Silva T, Rocha N, Silva AG Consensus on the Terms and Procedures for Planning and Reporting a Usability Evaluation of Health-Related Digital Solutions: Delphi Study and a Resulting Checklist J Med Internet Res 2023;25:e44326 URL: https://www.jmir.org/2023/1/e44326 doi: 10.2196/44326 PMID: 37279047 ©Ana Isabel Martins, Gonçalo Santinha, Ana Margarida Almeida, Óscar Ribeiro, Telmo Silva, Nelson Rocha, Anabela G Silva. Originally published in the Journal of Medical Internet Research (https://www.jmir.org), 06.06.2023. This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included. https://www.jmir.org/2023/1/e44326 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e44326 \| p. 13 (page number not for citation purposes)
10.1371_journal.ppat.1009995	RESEARCH ARTICLE Acquisition of yersinia murine toxin enabled Yersinia pestis to expand the range of mammalian hosts that sustain flea-borne plague David M. BlandID, Ade´ laïde MiarinjaraID Joseph Hinnebusch ¤a, Christopher F. Bosio, Jeanette CalarcoID ¤b, B. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 Laboratory of Bacteriology, Rocky Mountain Laboratories, National Institute of Allergy and Infectious Diseases, NIH, Hamilton, Montana, United State of America ¤a Current address: Department of Environmental Sciences, Emory University, Atlanta, Georgia, United State of America ¤b Current address: Department of Integrative Biology, University of South Florida, Tampa, Florida, United State of America david.bland@nih.gov OPEN ACCESS Citation: Bland DM, Miarinjara A, Bosio CF, Calarco J, Hinnebusch BJ (2021) Acquisition of yersinia murine toxin enabled Yersinia pestis to expand the range of mammalian hosts that sustain flea-borne plague. PLoS Pathog 17(10): e1009995. https:// doi.org/10.1371/journal.ppat.1009995 Editor: Deborah M. Anderson, University of Missouri, UNITED STATES Received: July 14, 2021 Accepted: September 30, 2021 Published: October 14, 2021 Copyright: This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. Funding: This research was funded by the Intramural Research Program of the NIH, NIAID. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Abstract Yersinia murine toxin (Ymt) is a phospholipase D encoded on a plasmid acquired by Yersi- nia pestis after its recent divergence from a Yersinia pseudotuberculosis progenitor. Despite its name, Ymt is not required for virulence but acts to enhance bacterial survival in the flea digestive tract. Certain Y. pestis strains circulating in the Bronze Age lacked Ymt, suggest- ing that they were not transmitted by fleas. However, we show that the importance of Ymt varies with host blood source. In accordance with the original description, Ymt greatly enhanced Y. pestis survival in fleas infected with bacteremic mouse, human, or black rat blood. In contrast, Ymt was much less important when fleas were infected using brown rat blood. A Y. pestis Ymt− mutant infected fleas nearly as well as the Ymt+ parent strain after feeding on bacteremic brown rat blood, and the mutant was transmitted efficiently by flea bite during the first weeks after infection. The protective function of Ymt correlated with red blood cell digestion kinetics in the flea gut. Thus, early Y. pestis strains that lacked Ymt could have been maintained in flea-brown rat transmission cycles, and perhaps in other hosts with similar blood characteristics. Acquisition of Ymt, however, served to greatly expand the range of hosts that could support flea-borne plague. Author summary The bacterium Yersinia pestis causes highly lethal bubonic plague in a wide variety of mammals and is transmitted primarily by the bites of infected fleas. During its recent evo- lutionary divergence from Yersinia pseudotuberculosis, a mild pathogen incapable of flea- borne transmission, Y. pestis acquired a new gene that encodes a phospholipase enzyme called Yersinia murine toxin (Ymt). This was a critical step in the transition to an insect- PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 1 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood borne life cycle as it was reported that Ymt activity greatly enhances bacterial survival in the flea gut. Recent genomic sequencing of ancient Y. pestis strains revealed that some lacked Ymt, leading to the conclusion that these strains were not transmitted by flea bite. Here, we report that the importance of Ymt for survival in the flea is greatly dependent on host blood source. Ymt is required if fleas take up Y. pestis in mouse, human, or black rat blood, but is not required if brown rat blood is used. We conclude that ancient Y. pestis strains lacking Ymt could have circulated in certain flea-rodent transmission cycles. Acquisition of Ymt, however, enabled Y. pestis to greatly expand its host range to an eco- logically broad range of mammals and their fleas. Introduction Yersinia pestis evolved from the closely related Yersinia pseudotuberculosis, a food-borne path- ogen that generally causes self-limiting enteric disease, within the last 6,000 years [1,2]. Making only 5 specific genetic changes to Y. pseudotuberculosis results in a strain able to produce a transmissible infection in the flea [3]. One key gene acquired during transition to the flea- borne life cycle encodes Yersinia murine toxin (Ymt), a phospholipase D enzyme that has an important role in the ability of Y. pestis to colonize the flea midgut [4]. Ymt is encoded on the Y. pestis-specific pMT1 plasmid, which was acquired through horizontal gene transfer [5]. Ymt was once believed to be an important virulence factor in the mammalian host, as Ymt- enriched protein fractions are highly lethal to mice and rats [6,7]. However, Ymt is not required for typical plague disease progression and virulence and the LD50 of a Ymt-negative strain in mice is equivalent to that of wild-type Y. pestis [8]. Murine toxicity of Ymt is likely related to its ability to act as a β-andrenergic-blocking agonist in mice and rats [9,10], but tox- icity is not observed in other mammals such as guinea pigs, rabbits, dogs, and primates [11]. Application of molecular Koch’s postulates to a standardized flea model of Y. pestis infection revealed that Ymt’s true biological function is to enhance bacterial survival in the flea midgut, significantly improving the ability of the plague bacillus to stably infect and be transmitted by its vector [4,12]. In the original characterization, a Ymt mutant was rapidly eliminated from ~90% of Xenop- sylla cheopis fleas, and those few fleas with chronic infections had reduced bacterial burdens in which only the proventricular valve in the foregut (and not the midgut) was colonized. The incidence of transmission-enhancing proventricular blockage due to Y. pestis biofilm accumu- lation was correspondingly rare, indicating low potential for Ymt− strains to be vectored by fleas [4]. The Y. pestis Ymt mutant was eliminated from fleas within the first 24h following uptake in a blood meal, preceded by conversion of the bacilli to an atypical spheroplast mor- phology in the midgut [4]. Bacterial spheroplast formation usually indicates damage to, or loss of, the bacterial outer membrane and a reduction in osmotolerance. Addition of recombinant Ymt protein to the infectious blood meal did not protect mutant bacilli from clearance, and in fleas coinfected with Ymt− and Ymt+ Y. pestis, Ymt− bacteria persisted in the midgut only if they were embedded within a biofilm of Ymt+ bacilli. Immunohistochemistry and immunoas- says of culture supernatants indicate that Ymt is not secreted and is released only upon cell lysis [8]. Collectively, current data indicate that the Ymt phospholipase exerts its protective function intracellularly and that Ymt mutant bacteria are better able to survive in the flea gut if protected from the surrounding digestive and/or immunological milieu of the midgut [4]. In seeming contradiction to the rapid clearance phenotype observed for Ymt mutant bacte- ria [4], a separate study showed that Ymt− Y. pestis could survive in and be transmitted by fleas PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 2 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood up to 3 days after infection nearly as efficiently as the parental strain [13]. Notably, the study indicating that Ymt was dispensable for this early-phase transmission used brown rat (Rattus norvegicus) blood for the infectious blood meal [13], whereas the study demonstrating rapid clearance of the Ymt mutant from fleas used mouse blood [4]. Recently, we have shown that the source of infectious host blood alters the nature of the Y. pestis infection in the flea foregut [14]. Specifically, the slow digestion rate of brown rat blood and the relative insolubility of its hemoglobin promotes more rapid and extensive foregut infection (proventriculus and esopha- gus) in which partially digested blood meal contents mixed with Y. pestis are refluxed from the midgut into the esophagus; a phenomenon we have termed post-infection esophageal reflux (PIER) [14]. PIER-inducing blood sources reduce the time required for some rodent fleas to become infectious; increasing the number of bacilli transmitted during the first few days fol- lowing an infectious blood meal. Because brown rat blood promotes infection of the esophagus and the bactericidal agent of Ymt− strains is believed to be generated during blood digestion in the midgut, we thought Ymt− Y. pestis might be better able to survive in fleas if PIER-inducing blood sources were used for the infection. To test this hypothesis and evaluate the permissive- ness of different host blood sources to flea colonization, we infected rodent fleas with either wild-type or Ymt− Y. pestis suspended in blood collected from mice, brown rats, black “roof” rats (Rattus rattus), or humans. In resolution of the seemingly contradictory results, we found that Ymt mutant Y. pestis can chronically infect and be transmitted by the rodent fleas X. cheopis and Oropsylla montana at much higher levels if brown rat blood is used for the infectious blood meal than if mouse, human, or black rat blood is used. Our results suggest that ancestral Y. pestis strains lacking Ymt could have been maintained in flea-borne transmission cycles involving brown rats and perhaps other mammals with similarly permissive blood biochemistry. Acquisition of Ymt, however, fortified that ability and allowed Y. pestis to greatly expand its host range to involve many other mammals and their fleas, resulting in strong positive selective pressure for the Ymt+ lineage. Results The Y. pestis Ymt mutant induces PIER in fleas following an infectious brown rat blood meal When fleas ingest Y. pestis suspended in blood that is digested relatively slowly and is charac- terized by a poorly soluble hemoglobin molecule, many of them exhibit post-infection esoph- ageal reflux (PIER) [14]. The foregut of these fleas contains a mixture of partially digested blood components and Y. pestis aggregates that extends from the proventriculus forward into the esophagus within 24 h after an infectious blood meal. This phenomenon is seen following infections using brown rat and guinea pig blood, but not when mouse or gerbil blood is used [14]. Because digestive enzymes are likely not present at high concentration in the foregut, we hypothesized that bacteria aggregated there would be protected from bactericidal agents gener- ated in the midgut, and that the foregut thus might provide a niche for Ymt− Y. pestis to tem- porarily colonize if PIER-inducing blood is used for the infection. To determine if PIER induction occurs and could provide protection to Ymt− strains, we infected X. cheopis fleas using one of four blood sources (brown rat, black rat, mouse, and human) and screened them 24 h later for PIER (Fig 1). Consistent with our previous study, PIER was evident in ~20% of fleas infected with wild-type Y. pestis KIM6+ using brown rat blood, but not when mouse blood was used. PIER was also induced in fleas infected using black rat blood, but at lower incidence (~5%) than for brown rat blood. Fleas infected using human blood did not develop PIER (Fig 1A). Notably, PIER was also observed in fleas PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 3 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood Fig 1. A Y. pestis Ymt mutant induces PIER in X. cheopis fleas when brown rat blood is used for the infectious blood meal. A) Incidence of post-infection esophageal reflux (PIER) in groups of 25 to 220 X. cheopis fleas 24 h after feeding on brown rat (Rn), black rat (Rr), mouse (M), or human blood (H) containing 1.5 x 108–1.1 x 109 CFU/ml KIM6+ or KIM6+ymtH188N Y. pestis. Bars show the mean and standard error of 3 independent experiments (n = 164–438 mixed sex fleas). �p < 0.005 by chi-square test. B) Female X. cheopis with PIER 24 h after feeding on black rat blood containing GFP-positive Y. pestis KIM6+; blue arrow indicates where blood and Y. pestis has been refluxed from the proventriculus and/or midgut into the esophagus. C) light and D) fluorescence microscopy images of the digestive tract dissected from this flea showing the presence of partially digested blood components and bacteria in the proventriculus (PV) and esophagus (E). Scale bar = 50 μm. https://doi.org/10.1371/journal.ppat.1009995.g001 following infection with Ymt− Y. pestis in brown rat blood (but not black rat blood), but only about half as often as in fleas infected with the parental KIM6+ strain (Fig 1A). As in our previous study, PIER correlated with the presence of hemoglobin crystals, par- tially digested red blood cell stroma, and Y. pestis in the proventriculus and esophagus of fleas (Fig 1B–1D) [14]. Hemoglobin crystals were commonly observed in the midgut of infected fleas when black rat blood was used for the infectious blood meal but appeared to be more sol- uble than brown rat hemoglobin crystals. Black rat hemoglobin crystals typically had a long rod-like shape and rapidly dissolved in the PBS we used to prepare wet mounts of infected flea digestive tracts, making them difficult to image and possibly causing us to underestimate their prevalence. In addition, unlike brown rat blood [14], hemolysis of black rat red blood cells in water did not result in hemoglobin crystallization. Hemoglobin crystals were not observed in the gut of fleas infected using mouse blood [14], and rarely observed in fleas infected using human blood. Collectively, these data suggest that Ymt− Y. pestis can colonize the flea foregut, induce PIER, and potentially be protected from elimination when brown rat blood is used for the infectious blood meal. Blood source affects colonization of rodent fleas by Ymt-deficient Y. pestis To determine whether blood source and PIER affect the overall ability of the Ymt mutant to colonize the flea, X. cheopis were fed mouse, human, black rat, or brown rat blood containing ~ 5x108 CFU/ml Y. pestis KIM6+, KIM6+ymtH188N, or KIM6+ymtH118N (pYmt). Infected fleas subsequently received two sterile maintenance blood meals over the course of 1 week to evaluate their potential to become blocked. Replicating previously published results [4], 80– 90% of female fleas infected with the Ymt mutant in mouse blood cleared the infection within 24 h, whereas strains that produce the functional Ymt enzyme were rarely cleared by fleas dur- ing the first week (Fig 2A). The foregut of the few fleas that remained infected with the Ymt PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 4 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood Fig 2. Ymt− and Ymt+ Y. pestis colonize female X. cheopis similarly when brown rat blood is used for the infectious blood meal, but not if mouse, human, or black rat blood are used. Groups of female X. cheopis fleas that fed on mouse (blue), black rat (black), human (orange), or brown rat (red) blood containing 1.5x108–1.1x109 CFU/ml Y. pestis KIM6+, KIM6+ymtH188N, or KIM6+ymtH188N (pYmt) were scored for 1 week for A) the percentage of fleas that remained infected; B) the percentage that developed obstruction of the foregut (partial or complete blockage) that interfered with normal blood-feeding; and C) bacterial burden. Data are cumulative from 3 (KIM6+ and KIM6 +ymtH188N groups) or 1 (KIM6+ymtH188N(pYmt) groups) independent experiments. Samples consisted of 7–20 female (A and C) or 25–220 fleas (roughly equal numbers of males and females; B) per experiment. The mean and standard error (A, B) or median (C) are indicated. �p < 0.05 by chi-square (A, B) or by Kruskal-Wallis test with Dunn’s post-test (mouse, human, and brown rat groups) or Mann-Whitney test (black rat group) (C). Dotted lines indicate the limit of detection (40 CFU). KIM6+ymtH188N(pYmt) was not used for black rat blood infections due to the limited availability of this blood. https://doi.org/10.1371/journal.ppat.1009995.g002 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 5 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood mutant in mouse blood rarely became obstructed by a bacterial mass (partially or fully blocked) during the first week of the infection and had reduced bacterial burdens (Fig 2B and 2C). Comparable results were observed for fleas infected with the Ymt mutant in human or black rat blood (Fig 2). In contrast, the average infection rate (73%) and median bacterial bur- den (1.2 x106 CFU) after 1 week for fleas infected with the Ymt mutant in brown rat blood were only modestly lower than the infection rate (mean 95%) and bacterial burden (median 1.5 x106 CFU) of fleas infected with the wild-type parent strain. Furthermore, fleas infected with the Ymt mutant in brown rat blood developed proventricular blockage at a rate similar to that of fleas infected with the parent strain at the first feeding following infection (12% vs 13%) and at a slightly reduced rate (6% vs 8%) after the second feeding (Fig 2B). To verify that these results were not unique to X. cheopis rat fleas, we replicated the experiments using mouse and brown rat blood with Oropsylla montana, a North American ground squirrel flea. The results mirrored those seen for X. cheopis: the Ymt mutant was rapidly cleared from O. montana fleas infected using mouse blood, but those infected using brown rat blood had infection rates, bac- terial burdens, and proventricular obstruction rates that were equivalent to or only slightly reduced from wild-type levels (S1 Fig). These results were surprising, because although a minority of the fleas infected using brown rat blood developed PIER (10–20%), much higher proportions (30–100%) remained infected for up to 1 week. Thus, it seems unlikely that PIER alone accounted for the high rates of flea colonization observed for the Ymt mutant-brown rat blood infections. However, the data pro- vide insight into a previous report that the Ymt mutant can be as efficiently transmitted as its wild-type parent during the early phase when brown rat blood is used for the infectious blood meal [13]. In sum, our results show that the previously reported lability of Ymt− Y. pestis in the flea gut varies depending on the infectious blood source. This mutant fares poorly after infec- tious mouse, black rat, or human blood meals, but survives much better after brown rat infec- tious blood meals. This effect is conserved in two rodent flea species from distinct taxonomic families. The protective role of Ymt is more pronounced in female fleas than in male fleas In the original characterization of the Ymt mutant strain in fleas, infection rates were deter- mined only for female X. cheopis fleas infected using mouse blood [4], and the rates in Fig 2 were also based on female fleas. Because the metabolism and physiology of insects is not identi- cal between sexes, we evaluated infection rates separately for male and female fleas infected with Ymt mutant Y. pestis using either mouse or brown rat blood. Unexpectedly, when mouse blood was used for the flea infection, 61% of male fleas remained infected after 24 h, whereas only a single female (4%) had evidence of GFP+ bacteria in the digestive tract (Fig 3A and Table 1). In contrast, when brown rat blood was used for the flea infections, male and female fleas had equivalently high rates of Y. pestis colonization and 25% (including examples of both sexes) had more severe bacterial infections in the proventriculus (Table 1). These data show an enhanced capacity for Ymt− bacteria to survive in the male flea midgut. The infection status and bacterial load of fleas 1 and 7 days after infection was determined to assess whether the Ymt mutant persisted in male fleas infected using mouse blood. Signifi- cantly more males (40–70%) than females (0–25%; Fig 3B) remained infected for up to 1 week with Ymt mutant Y. pestis. The mean bacterial load of fleas infected with Ymt− Y. pestis was higher for males at both 1 and 7 days after infection, but the difference was not statistically sig- nificant (Fig 3C). In contrast, infection rates were identical between sexes when infected with the wild-type parent strain (Fig 3A and 3B). Regardless of sex, the few fleas that remained PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 6 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood Fig 3. The Y. pestis Ymt mutant colonizes male fleas more efficiently than females following infection using mouse blood. Infection rates for groups of female or male X. cheopis infected using mouse blood (blue symbols) or brown rat blood (red symbols) containing 1x108–5.7x108 CFU/ml GFP-positive KIM6+ or KIM6+ymtH188N Y. pestis were determined 1 day after infection by fluorescence microscopy of dissected flea digestive tracts A); or 0, 1, and 7 days after infection by CFU counts from individual triturated fleas (B, C). For A, each symbol represents the percentage of fleas containing GFP+ bacteria in their digestive tract. n = 4–10 fleas of each sex in 3 independent experiments (Table 1). For B and C, the mean and standard error (B) or median (C) of pooled data from 3 independent experiments for groups of 5–20 fleas infected using mouse blood are shown. �p < 0.05 by chi-square test (B) or two-way ANOVA with Tukey’s post-test (C). D) Examples of the foregut infection in female or male X. cheopis 1 day after ingesting KIM6+ymtH188N Y. pestis suspended in mouse blood (Left) or brown rat blood (Right). Scale bar = 50 μm. https://doi.org/10.1371/journal.ppat.1009995.g003 Table 1. Flea Dissection Summary. Blood Source /Experiment % Fleas Infected (KIM6 +ymtH188N) Bacteria Present In: PV Infection Severity X. cheopis Mouse Blood #1 Mouse Blood #2 Mouse Blood #3 Total/Average Rat Blood #1 Rat Blood #2 Rat Blood #3 Male 67% (9) 100% (4) 40% (10) 61% (14/23) 70% (10) 83% (6) 100% (10) Female 10% (10) 0% (5) 0% (9) 4% (1/24) 100% (10) 100% (6) 90% (10) PV+ MG PV only MG only 66% 100% 50% 0% 0% 0% 33% 0% 50% Light 67% 100% 50% Moderate Heavy 0% 0% 0% 0% 0% 0% 67% (10/15) 0% (0/15) 33% (5/15) 67% (10/15) 0% (0/15) 0% (0/15) 82% 100% 100% 6% 0% 0% 12% 0% 0% 47% 73% 100% 35% 27% 11% 6% 0% 0% Total/Average 85% (22/26) 96% (25/26) 94% (44/47) 2% (1/47) 4% (2/47) 70% (33/47) 23% (11/47) 2% (1/47) PV = proventriculus, MG = midgut. Numbers in parentheses indicate flea sample sizes. https://doi.org/10.1371/journal.ppat.1009995.t001 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 7 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood infected had a lightly colonized proventriculus (� 25% coverage of the proventricular spines by a bacterial mass; (Fig 3D and Table 1). Clearance of Ymt− Y. pestis correlates with the rate of RBC digestion To confirm that the differential survival of the Ymt mutant in the flea was not due to inhibitory components in certain blood sources, in vitro growth of wild-type and mutant Y. pestis in defi- brinated mouse blood, rat blood, or BHI-hemin broth was monitored during 24 h of incuba- tion at 21˚ or 37˚C. As expected, the Ymt mutant and the parental strain grew equally well in all three substrates (S2A and S2B Fig). Additionally, exposure to hemolyzed mouse RBCs or defibrinated mouse plasma did not affect bacterial viability (S2C Fig). These data indicate that the Ymt mutant phenotype observed in the flea gut is unrelated to differential growth charac- teristics in mouse blood and that the clearance of the Ymt mutant may require processing of the blood meal by flea digestive enzymes [4]. To determine the fraction of blood responsible for clearance of the Ymt mutant in the flea gut, we infected X. cheopis using reconstituted, plasma-swapped mouse or brown rat blood (rat plasma mixed with mouse RBCs or vice versa) containing KIM6+ymtH188N Y. pestis. One day after infection, only 7% of fleas infected using brown rat plasma with mouse RBCs remained infected compared to 80% of those infected using mouse plasma and brown rat RBCs (Fig 4A). In sum, addition of brown rat plasma to mouse RBCs did not rescue the Ymt mutant in the flea gut, and addition of mouse plasma to rat RBCs did not result in impaired bacterial infectivity. These results indicate that the bactericidal agent is primarily produced as a consequence of digestion of RBCs, such as those from a mouse, and that the contribution of plasma to the Ymt− strain phenotype is likely modest or inconsequential. Given that Ymt appeared to protect against a bactericidal product of RBC digestion, we decided to test whether female and male X. cheopis digested mouse and rat RBCs at equivalent rates. First, to get a better understanding of flea digestion kinetics between flea sexes, we deter- mined that female fleas ingest, on average, roughly twice as much blood as male fleas (Fig 4B). Next, we found that the RBC concentration in the flea gut was similar, regardless of blood source or flea sex, within the first 30 minutes after the bloodmeal (Fig 4C). However, by two hours after feeding, female fleas that ingested mouse blood had the largest reduction in red cell counts. The majority of mouse RBCs had lysed by 2 h in ~1/3rd of female fleas, whereas all other flea sex-blood source combinations showed a lower RBC digestion rate (Fig 4C). Beyond 2 h, both mouse and rat RBCs frequently aggregated in large clusters in the flea digestive tract, ren- dering hemocytometer counts unfeasible. To address this, we imaged digestive tracts excised from fleas every 2 h for the first 8 h after an uninfected blood meal. We found that most female fleas completely digest and liquify mouse RBCs within 4–6 h (Figs 4D and S3). At 6 h, the gut of 70% of the female fleas (9 of 13) contained only a moderately viscous pink fluid, devoid of cellu- lar material. Male fleas took longer to digest mouse blood; after 6–8 h of digestion, only ~30% had completely liquified their blood meal (Figs 4D and S3). In contrast, brown rat blood took considerably longer for both male and female fleas to digest. By 6 h after feeding, the digestive tract always contained a thick, viscid, brownish-red slurry of aggregated RBC stroma in various stages of breakdown, distinctly more viscous than what was present in fleas fed mouse blood (S3 Fig). By 8 h, no female fleas (0 of 11) and only 7% of males (1 of 15) had completely liquified their brown rat blood meal (Fig 4D). The relative amount of solid material in the fleas fed brown rat blood remained fairly constant over the first 8 h of digestion, indicating that both male and female fleas typically require more than 8 h to liquify brown rat blood. The identical temporal patterns of RBC digestion were also observed in fleas infected with Ymt− Y. pestis. By 24 h after infection, fleas infected using brown rat blood routinely contained PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 8 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood Fig 4. Survival of the Ymt mutant in the flea correlates with slower RBC digestion. A) Bacterial titers and infection rates for groups of female X. cheopis infected using reconstituted, plasma-swapped mouse blood (brown rat plasma mixed with mouse RBCs; blue) or brown rat blood (mouse plasma mixed with rat RBCs; red) containing 1.3x108– 2.8x108 CFU/ml KIM6+ymtH188N. Data are the pooled results from 3 independent experiments (n = 10); bars represent the median. �p < 0.0001 by Mann-Whitney test. B) Blood meal volumes of individual female or male X. cheopis allowed to feed for 1 h on a neonatal mouse. Mean blood meal volumes are indicated, �p <0.0001 by Student’s t-test. C) The RBC concentration in individual X. cheopis female or male digestive tracts 0.5 or 2 h after ingestion of sterile mouse or rat blood. Bars represent the mean of 3 independent assays using n = 3–6 (0.5 h) or n = 6–10 (2 h) fleas. �p <0.05 by two-way ANOVA with Tukey’s post-test. D) The mean proportion and range of male or female X. cheopis that completely liquified sterile mouse or brown rat blood during the first 8 h of digestion. Data are from groups of 3–6 digestive tracts excised from fleas at each timepoint and condition from 3 independent experiments; n = 9–15. �p <0.05 by Fisher’s exact test compared to rat blood group. A representative image series of these data is shown in S3 Fig. https://doi.org/10.1371/journal.ppat.1009995.g004 significant quantities of undigested midgut material (Fig 3D). In contrast, fleas infected using mouse blood typically contained only the viscous pink or red liquid, with little to no solid material (Fig 3D). Collectively, these data indicate a correlation between the rate of RBC diges- tion and the clearance of Ymt− Y. pestis from the flea gut. Ymt− Y. pestis can be transmitted beyond the early phase when fleas are infected using brown rat blood To assess transmission of the mutant strain, groups of O. montana or X. cheopis fleas were infected with the Ymt mutant or the parent strain in either brown rat or mouse blood and PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 9 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood Fig 5. Rodent fleas can transmit Ymt mutant Y. pestis for at least 3 weeks when infected using brown rat blood. Y. pestis transmission dynamics were monitored for 3 to 4 weeks for groups of 150–267 X. cheopis (A) or O. montana fleas infected using 3.4 x 108−1.9 x109 CFU/ml KIM6+ or KIM6+ymtH188N Y. pestis (B) in either mouse (blue) or brown rat (red) blood and subsequently maintained on sterile blood of the same type. Numbers in parentheses indicate the total number of fleas that fed followed by the number of fleas with evidence of foregut obstruction (partially or fully blocked). Roughly equivalent numbers of male and female fleas were used for transmission assays. Infection rate was determined for groups of 10–20 female C) X. cheopis or D) O. montana at various times following infection. https://doi.org/10.1371/journal.ppat.1009995.g005 were fed periodically on sterile blood of the same source. After each maintenance feed, the blood was collected from the feeding device and plated to determine the number of CFUs transmitted. Early-phase transmission (3 days post-infection) of Ymt mutant Y. pestis was detected for O. montana infected using brown rat blood, but not if mouse blood was used (Fig 5A and 5B). Furthermore, X. cheopis and O. montana infected using brown rat blood transmit- ted moderate to high levels of the Ymt mutant for at least 2 or 3 weeks, respectively, during the biofilm-dependent phase of transmission. Transmission of the Ymt mutant and the parental Ymt+ Y. pestis strains by X. cheopis infected using brown rat blood was roughly comparable (Fig 5A). In contrast, fleas infected using mouse blood rarely became blocked and only a single instance of transmission was observed (X. cheopis, day 17), in which very few CFU were trans- mitted (Fig 5A). Infection rates of fleas used for transmission tests were similar to those shown in Figs 2 and S1 (Fig 5C and 5D). Reduced transmission by fleas infected using brown rat blood during the later weeks of infection may be partially attributable to the higher mortality rate of these fleas. Greater than 80% of both flea species had died by 3 weeks after infection. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 10 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood The elevated mortality may be attributable to the more severe Y. pestis infection in the foregut of fleas infected using brown rat blood (Figs 1 and 5) [14]. The overall incidence of foregut obstruction (fully and partially blocked fleas) was 39–45% for rodent fleas infected and main- tained on brown rat blood but only 0–2.5% for those infected using mouse blood (Fig 5). Con- sistent with our finding that male X. cheopis are more susceptible to infection by Ymt− Y. pestis in mouse blood (Fig 3), all 5 fleas in this group that became blocked were males (Fig 5A). Discussion Gene gain and gene loss were both major drivers of the recent evolutionary emergence of Y. pestis, and acquisition of Ymt was critical for the transition to a flea-borne life cycle because it greatly enhanced survival of Y. pestis in the flea gut [15]. The original report of the protective effect of Ymt hypothesized that the Phospholipase D activity of Ymt directly or indirectly pro- tects Y. pestis against a bactericidal byproduct of blood digestion [4], and was based mainly on female fleas infected using mouse blood. We extend that original characterization here, show- ing that the protective function of Ymt is much less important when fleas feed on bacteremic brown rat blood than on bacteremic mouse, human, or black rat blood. With brown rat blood, flea infection and proventricular blockage rates were not significantly different for the first few days of infection and only slightly reduced after 1 week for Ymt− compared to Ymt+ Y. pestis, whereas these rates were greatly reduced in the first 24 h of infection for the Ymt mutant using the other blood sources. Correspondingly, both X. cheopis and O. montana fleas infected using brown rat blood transmitted Ymt− Y. pestis relatively efficiently, whereas as predicted by the previous study [4] fleas infected using mouse blood rarely transmitted, and few CFU were transmitted. Other major findings of this study are that Ymt likely protects Y. pestis from a product of RBC digestion, and not a plasma digestion product as hypothesized previously [4]; and that the importance of Ymt correlates with RBC digestion kinetics. Previous studies of RBC diges- tion by female X. cheopis demonstrated that fleas digest their blood meals more rapidly than many other blood-feeding arthropods, such as mosquitos and ticks, and that the digestive tract expresses a number of trypsin-like transcripts within the first hours following feeding [16,17]. Electron microscopic analysis of the X. cheopis midgut epithelium indicates that secretory vesi- cles, likely containing digestive enzymes, are produced in advance of feeding and are released more or less immediately following ingestion of blood [18]. However, we found that the diges- tion rate can vary depending on the host blood source. Brown rat RBCs were digested more slowly and incompletely than mouse RBCs. Fleas infected using brown rat blood routinely had large quantities of undigested material in their gut 24 h after infection, whereas fleas infected using mouse blood were essentially devoid or had greatly reduced amounts of solid blood material. In addition, our results show that female X. cheopis digest mouse blood more rapidly than males do, despite ingesting roughly twice as much blood. Digestion kinetics often differ between insects of the opposite sex, as oviposition and egg maturation are physiologically asso- ciated with digestion, and females typically have greater energetic demands due to these bio- logical imperatives [19,20]. The digestion patterns we observed for female fleas and mouse blood, in which most erythrocytes are digested during the first few hours, are consistent with previous estimates [16]. Other microscopic analyses of flea gut contents also have indicated that X. cheopis females digest mouse blood more rapidly than males [21]. Overall, the surviv- ability of Ymt− Y. pestis after being ingested by a flea correlated well with RBC digestion rate: good, nearly normal survival in both sexes with rat blood infections; and intermediate survival in male fleas but poor survival in female fleas with mouse blood infections. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 11 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood In addition to digestive enzymes, RNA-Seq analysis of the X. cheopis digestive tract tran- scriptome revealed that antimicrobial peptides are rapidly produced in response to ingestion of Y. pestis [17]. However, Ymt does not appear to have a role in protection against the flea immune response. Drosophila, an insect that has been developed as a surrogate Y. pestis infec- tion model, produces a diverse array of antimicrobials in response to Gram-negative bacteria, yet Ymt mutant strains show no defect in fruit fly colonization or bacterial burden [22]. In this model, Drosophila larvae ingest Y. pestis-laden cornmeal agar, rather than blood, to initiate infection [23], suggesting that the Ymt mutant colonization defect in the flea is uniquely related to blood digestion rather than to insect innate immunity. Furthermore, the Ymt mutant shows no enhanced susceptibility to common antimicrobials that target the outer membrane (polymyxin B, SDS, lysozyme, etc.) or to other potentially bacteriolytic enzymes and environmental stressors that would be encountered in the arthropod gut environment (proteases, lipases, osmotic and oxidative stress) [24]. It was previously hypothesized that the Phospholipase D activity of Ymt provides protection against a bacteriolytic byproduct of blood digestion by either modifying the bacterial envelope to make the bacteria resistant to lysis (pro- phylaxis model), or by direct or indirect neutralization of the lytic agent (antidote model) [24]. While we have not resolved the mechanism by which Ymt provides protection to the bacteria, our data further indicate that it is a byproduct of RBC digestion that induces the abnormal spheroplast morphology indicative of cell envelope damage to the Ymt− mutant in the flea. Notably, Ymt− Y. pestis grows normally in all blood sources during in vitro growth assays, and can produce septicemic plague in mice [8]. Based on these findings, we propose a nuanced model for the role of Ymt in flea-borne transmission. When fleas are infected from a host (e.g. mouse) whose RBCs are digested rap- idly, the bactericidal byproduct generated reaches cytotoxic levels within the first few hours and eradicates Ymt− Y. pestis from the midgut. This is more pronounced in female fleas, which take larger blood meals and digest them more rapidly than males. A significantly higher per- centage of male X. cheopis become infected after feeding on mouse blood containing Ymt− Y. pestis, and the infection can involve the midgut and the proventriculus. In females, only those few fleas in which the mutant localizes to the foregut, sequestered from the digestive milieu of the midgut, remain colonized [4]. In contrast, when fleas are infected from a host (e.g. brown rat) whose RBCs are digested slowly, we hypothesize that the bactericidal byproduct does not reach lethal levels before the bacteria have time to coalesce into large dense aggregates. These aggregates develop in the midgut and proventriculus within a few hours after ingestion and appear to be surrounded by a viscous matrix [25,26], suggesting that they may be protected from exposure to bactericidal factors in the midgut. Supporting this idea is the observation that providing fleas infected using brown rat blood with two maintenance mouse blood meals 2–7 days later did not significantly reduce their high infection rates, which remained compara- ble to those seen for fleas provided brown rat blood maintenance meals. This model could also account for the disparity in infection rates between male and female fleas infected with Ymt− Y. pestis using mouse blood. Our results suggest that the bactericidal agent is produced in the flea gut regardless of the host blood ingested, but the digestion kinetics of the various blood sources dictate the frequency and rate at which an absolute lethal concentration is achieved rel- ative to the time it takes for Y. pestis to coalesce into large dense masses. Although the model emphasizes RBC digestion kinetics, it is also possible that biochemical differences between mouse and rat RBCs contribute to the much greater sensitivity of the Ymt− mutant to mouse RBC digestion. These results suggest a revision to the evolutionary history of Y. pestis (Fig 6). Ancestral strains that had not yet acquired ymt, such as those circulating in the Neolithic and Bronze Age, have been thought to be fully virulent for mammals but incompetent for flea-borne PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 12 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood Fig 6. Model of the adaptive function of Yersinia murine toxin (Ymt) during the evolution of the flea-borne life cycle of Y. pestis. In this model, ancestral Y. pestis strains lacking the ymt gene (Ymt−; left) could cycle between fleas and certain species of rodent with flea-colonization-permissive host blood, such as brown rats (Rattus norvegicus), but not those with non-permissive blood, such as mice (Mus spp). Following acquisition of ymt on the pMT1 plasmid (Ymt+ strains; right), the progenitor of modern, extant strains of Y. pestis was able to stably colonize fleas that fed on bacteremic hosts with a blood chemistry that is not permissive for Ymt-negative strains. Thus, acquisition of ymt effectively greatly expanded the range of mammalian hosts that could support a flea-mammal transmission cycle. Although male fleas become infected at a moderate rate with Ymt− Y. pestis infected with non-permissive mouse blood (Fig 3), their potential to transmit is likely not sufficient to maintain a stable transmission cycle (Fig 5). https://doi.org/10.1371/journal.ppat.1009995.g006 transmission [27–30]. However, our data indicate that they could have been maintained in stable flea-borne transmission cycles among brown rats and other hosts with similarly permissive blood characteristics. In addition to early-phase transmission, which does not require Ymt following rat blood infection [13], transmission by the later, proventricular blockage mechanism would be robust within these host populations. Acquisition of ymt, however, would have been adaptive for two reasons. First, it modestly augments flea infectivity even when the bacteremic host blood has a largely permissive biochemical profile, such as the blood of the brown rat. More consequen- tially, Ymt enzymatic activity greatly enhances the percentage of fleas that develop a chronic, transmissible infection when Y. pestis is acquired from a host with blood biochemistry that was (originally) poorly permissive for flea infection (e.g., mice, humans, black rats) (Fig 6). The ances- tral Y. pestis lineages that lacked ymt are extinct, suggesting that host restriction and reduced flea transmissibility of these strains contributed to reduced Darwinian fitness and their eventual dis- appearance [31]. It’s tempting to hypothesize that rodents involved in ancestral plague transmis- sion cycles have blood biochemistry similar to that of the brown rat. For example, Tarbagan marmots (Marmota siberica) have been proposed as host to the original Y. pestis clone, if so, their blood would likely support chronic flea infection by Ymt− Y. pestis strains [32]. Collectively, our results indicate that acquisition of Ymt did not allow Y. pestis to colonize fleas per se, but significantly improved Y. pestis survival in the flea gut in the context of RBC digestion and processing kinetics of blood meals from different mammals. In effect, acquisi- tion of Ymt greatly expanded the range of hosts that could support a stable mammal-flea trans- mission cycle (Fig 6). The antibacterial product of RBC digestion and the mechanism of Ymt- mediated resistance to it remain to be determined. However, this study provides an important PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 13 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood update and revision to Ymt’s adaptive function during the recent evolutionary transition of Y. pestis to a flea-borne pathogen involving the ecologically broad range of mammals that charac- terizes modern strains. Methods Ethics statement Experiments involving animals were approved by the Rocky Mountain Laboratories, National Institute of Allergy and Infectious Diseases, National Institutes of Health Animal Care and Use Committee (Animal Protocol #2019–011E) and were conducted in accordance with all National Institutes of Health guidelines. Bacterial strains and plasmids Y. pestis strains, plasmids, and primers used in this study are listed in Table 2. Y. pestis KIM6+ ymtH188N expresses a non-functional Ymt with a point mutation in one of the two HKD cata- lytic domains that typify this class of phospholipase D enzymes [33] and is referred to here as the Ymt− or Ymt mutant strain. pCH16 (referred to hereafter as pYmt) contains the wild type ymt gene expressed by its native promoter and was used for complementation of the ymtH188N mutant [8]. All Y. pestis strains were transformed with pAcGFP1 (Clontech/Takara Bio) or pGFP-Kmr (this study), respectively. pGFP-Kmr was used for strains complemented with pCH16 to maintain selection for both plasmids prior to use in flea infections. Flea infection and host blood Prior to infection, X. cheopis or O. montana fleas were randomly pulled from colonies estab- lished at the Rocky Mountain Laboratories and starved for three days. Y. pestis strains were Table 2. Strain and Plasmid List. Strain/Plasmid Y. pestis strains KIM6+ pCD1-, pMT1+, pPCP1+, Pgm+ Key Properties KIM6+ymtH188N KIM6+ modified to express a non-functional Ymt with a point mutation in one of the two HKD catalytic Plasmids pAcGFP1 pGFP-Kmr pCH16 (pYmt) Primers pGFP-Kmr (Inverse PCR) pGFP-Kmr (Kmr Casette) domains. Apr, constitutively expresses GFP pAcGFP1 was amplified by inverse PCR to selectively exclude the bla gene and replace it with a SacI site. A kanamycin resistance cassette with terminal SacI sites was inserted into the linear inverse PCR product of pAcGFP1, which was then religated. Apr, expresses ymt from its native promoter. F: CGTCGAGCTCTTCGTTCCACTGAGCGTCA R: CGTAGAGCTCGTACAATCTGCTCTGATGCCG F: CGTAGAGCTCTCCAGCCAGAAAGTGAGGGAG R: GCATGAGCTCGGGAAAGCCACGTTGTGTCTC (Amplified from pKD4; [36]) pCD1 virulence plasmid, encodes type 3 secretion system pMT1 plasmid, encodes the phospholipase D Yersinia murine toxin (Ymt) and capsule antigen (F1) pPCP1 plasmid, encodes plasminogen activator/protease (Pla), the bacteriocin pesticin (Pst), and pesticin immunity protein (Pim) Pgm pigmentation locus and pathogenicity island, encodes the hemin storage locus (hmsHFRS operon) and iron acquisition genes Apr ampicillin resistance; Kmr kanamycin resistance https://doi.org/10.1371/journal.ppat.1009995.t002 Reference [34,35] [33] Clontech/Takara Bio (Mountain View, CA) This Study [8] This Study This Study PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 14 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood grown in brain-heart infusion (BHI) broth with appropriate antibiotic selection as described previously [37]. Briefly, 100 ml Y. pestis cultures, grown at 37˚C overnight, were centrifuged and the bacterial pellet resuspended in 1 ml sterile PBS. Bacterial suspensions were added to a final concentration of ~5x108 CFU/ml to 5 ml of heparinized Swiss-Webster mouse blood, defi- brinated Sprague-Dawley brown rat (R. norvegicus) or human blood (both from BioIVT, New York), or to heparinized wild black rat (R. rattus) blood collected and shipped overnight by Ala- meda County, CA, Vector Control personnel. Prior to use in flea infections, black rat blood was treated with carbenicillin (100 μg/ml) and plated on 5% sheep blood agar to ensure sterility. The blood and bacterial mixture was added to a membrane feeding apparatus and groups of fleas were allowed to feed for 1 h [4]. Fleas (approximately equal numbers of males and females) that took an infectious blood meal were collected and kept at 21˚C in a humidified chamber (75% RH). These fleas were provided maintenance feedings on neonatal mice 2 to 3 days after infec- tion and again 6 to 7 days after infection. Following each maintenance feed, fleas were screened for the presence of fresh red blood in the esophagus, a condition of fleas with partial or complete blockage or PIER. At 0, 1, and 7 days following infection, 10 to 20 infected fleas were frozen at -80˚C for later determination of infection status and bacterial load per flea by plating individual triturated fleas in BHI soft agar overlays as previously described [38]. For flea infections using plasma-swapped blood, the plasma fraction was separated from mouse or rat red blood cells (RBC) following centrifugation at 3000 rpm, the RBCs were washed 3 times with an equivalent volume of sterile PBS, and whole blood was reconstituted with heterologous plasma from the other rodent. Dissection and imaging of flea digestive tracts Fleas infected with KIM6+ymtH188N suspended in mouse or brown rat blood were dissected one day after infection to determine the localization of bacteria in the digestive tract and their phenotype. The severity of proventricular infection was scored as light, moderate, or heavy as described previously [38]. Images of flea digestive tracts and bacterial biofilms were taken with a Nikon Eclipse E800 microscope equipped with a DP72 Olympus camera (Center Valley, PA) and a G-2E/C (540/25 EX) fluorescent filter (Nikon), and were processed using Olympus cell- Sens software. Blood meal volume and red blood cell digestion rate For blood meal volume determination, individual adult X. cheopis not fed for 5 days prior, were weighed using a Sartorius SC 2 Microbalance (Goettingen, Germany) before and imme- diately after feeding on a neonatal mouse. Fleas were anesthetized with CO2 and placed in a microcentrifuge tube prior to each weighing. Bloodmeal weight was determined by subtracting the pre-feed weight from the post-feed weight and then converted to volume based on the spe- cific gravity of mouse blood [39]. To assess digestion kinetics, X. cheopis were allowed to feed on sterile Swiss Webster mouse or Sprague-Dawley rat blood for 30 minutes. Digestive tracts were dissected from groups of male or female fleas immediately (0.5 h) following feeding or 2 h after feeding. Excised diges- tive tracts were placed in 20 μl sterile PBS on a microscope slide and expressed with forceps to release midgut contents. Expelled gut contents were diluted 1:5 in PBS containing 0.4% (w/v) trypan blue, mixed, and the number of RBCs determined using a hemocytometer. For the digestion image series, fleas were fed as described above and digestive tracts were imaged using a Nikon SMZ1500 dissection microscope with a DP72 Olympus camera. Diges- tive tracts were visually scored for the presence or absence of cellular material to determine if they had completely liquified the blood meal. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 15 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood Transmission assays Transmission by O. montana or X. cheopis fleas was assayed as described previously with minor modifications [40]. Fleas infected as described above were refed using the artificial feed- ing system every 3 days following infection on sterile host blood of the same source they were initially infected with. After each feed (except for the X. cheopis-mouse blood experiment in which transmission was assessed every 6 days following the initial assay on day 3) the entirety of the blood from the feeding reservoir was collected and distributively spread onto blood agar-carbenicillin plates. In addition, the feeding reservoir was washed 6 times with 5 ml of sterile PBS; these washes were combined, centrifuged, and the resulting pellet resuspended in 2 to 3 ml of PBS and plated. Blood agar plates were incubated for 48 h at 28˚C and GFP positive colonies counted to determine the number of CFUs transmitted. Bacterial in vitro growth and susceptibility assays Y. pestis strains were grown in BHI containing 10 μg/ml hemin. After 18 h incubation at 28˚C without shaking, cultures were diluted to an OD600 of 0.1, centrifuged at 6000 rpm for 10 min, and the bacterial pellets resuspended in an equal volume of sterile PBS. Y. pestis was then added to 10 ml of BHI-hemin, defibrinated brown rat blood, or defibrinated mouse blood to a final concentration of ~1x106 CFU/ml and cultures were incubated in 50 ml conical tubes at either 21˚ or 37˚C without shaking. After 0, 2, 4, 6, 8, and 24 h incubation the cultures were mixed well and a 100 μl sample was removed, serially diluted, and plated on blood agar for CFU determination. For susceptibility assays, ~1x106 Y. pestis, prepared as above, were added to 1 ml of defibrin- ated mouse plasma, a suspension of lysed mouse RBCs, or BHI broth in an 8-well culture dish. Lysed mouse RBCs were prepared by mixing washed cells 1:1 with sterile PBS followed by three freeze-thaw cycles. After 1 h at 25˚C, 10-fold serial dilutions of the suspensions were plated on blood agar. The percentage of CFU recovered from each medium relative to the BHI control was calculated to assess antibacterial activity. Supporting information S1 Fig. The Y. pestis Ymt mutant phenotype in female Oropsylla montana fleas is identical to that in X. cheopis. Groups of O. montana fleas that fed on mouse or brown rat blood con- taining 2.8 x 108–7.1 x 108 CFU/ml Y. pestis KIM6+, KIM6+ymtH188N, or KIM6+ymtH188N (pYmt) were screened for 1 week for A) the percentage of fleas that remained infected; B) development of a foregut obstruction that interfered with normal blood-feeding; and C) bacte- rial burden. Data are the results from 3 (KIM6+ymtH188N groups) or 1–2 (KIM6+ and KIM6 +ymtH188N(pYmt) groups) independent experiments. Samples consisted of 9–20 female fleas (A, C) or 40 to 112 fleas (approximately equal numbers of males and females; B) per experi- ment. The mean and standard error (A, B) or median (C) are indicated. �p <0.05 by chi- square (A, B) or by Kruskal-Wallis test with Dunn’s post-test (C). (TIF) S2 Fig. The Ymt mutant has no growth or survival defects under in vitro conditions. Growth kinetics of Y. pestis KIM6+ and KIM6+ymtH188N grown in mouse blood, brown rat blood, or BHI broth supplemented with hemin and incubated for 24 h at A) 21˚C or B) 37˚C. The mean and standard error of 3 independent experiments are shown. C) Bacterial survival assay in which 1x106 CFU KIM6+ymtH188N were added to BHI broth, defibrinated mouse plasma, lysed mouse red blood cells and incubated for 1 h at 25˚C. Dilutions of each medium were then plated to determine CFU concentrations. The mean and standard error of 3 PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 16 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood independent experiments are shown and expressed as the percent CFU recovered relative to the BHI control. (TIF) S3 Fig. X. cheopis blood source- and sex-related differences in red blood cell digestion rates. Representative image series of the data shown in Fig 4D. Digestive tract preparations were scored for the presence or absence of particulates that exuded from the flea midgut into the surrounding saline. Mouse blood meals were completely liquified by most female fleas in 4–6 h (far left) but partially digested RBC stroma were still present in most males for 6–8 h (middle left). With rare exception, fleas that ingested sterile rat blood, regardless of sex, con- tained a fairly stable amount of partially digested RBCs for at least 8 h following feeding (right). Scale bar = 100 μm. (TIF) S1 Data. File containing numerical data used for Figs 1–5, S1 and S2. (XLSX) Acknowledgments We thank David K. James and colleagues at Alameda County Vector Control (Alameda, CA) for generously collecting black rat blood; Ryan Kissinger for assistance with graphic design; and Clayton Jarrett, Jeff Shannon, and Phil Stewart for critical review of the manuscript. Author Contributions Conceptualization: David M. Bland, B. Joseph Hinnebusch. Data curation: David M. Bland. Formal analysis: David M. Bland, Ade´laïde Miarinjara, B. Joseph Hinnebusch. Funding acquisition: B. Joseph Hinnebusch. Investigation: David M. Bland, Ade´laïde Miarinjara, Christopher F. Bosio, Jeanette Calarco, B. Joseph Hinnebusch. Methodology: David M. Bland, Ade´laïde Miarinjara, Christopher F. Bosio, B. Joseph Hinnebusch. Project administration: B. Joseph Hinnebusch. Resources: B. Joseph Hinnebusch. Supervision: B. Joseph Hinnebusch. Validation: David M. Bland, B. Joseph Hinnebusch. Visualization: David M. Bland, B. Joseph Hinnebusch. Writing – original draft: David M. Bland. Writing – review & editing: David M. Bland, Ade´laïde Miarinjara, Christopher F. Bosio, Jean- ette Calarco, B. Joseph Hinnebusch. References 1. Achtman M, Zurth K, Morelli G, Torrea G, Guiyoule A, Carniel E. Yersinia pestis, the cause of plague, is a recently emerged clone of Yersinia pseudotuberculosis. Proc. Natl. Acad. Sci. USA. 1999; 96 (24):14043–8. https://doi.org/10.1073/pnas.96.24.14043 PMID: 10570195. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 17 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood 2. Cui Y, Yu C, Yan Y, Li D, Li Y, Jombart T, et al. Historical variations in mutation rate in an epidemic path- ogen, Yersinia pestis. Proc. Natl. Acad. Sci. USA. 2013; 110(2):577–82. https://doi.org/10.1073/pnas. 1205750110 PMID: 23271803. 3. Sun YC, Jarrett CO, Bosio CF, Hinnebusch BJ. Retracing the evolutionary path that led to flea-borne transmission of Yersinia pestis. Cell Host Microbe. 2014; 15(5):578–86. https://doi.org/10.1016/j.chom. 2014.04.003 PMID: 24832452. 4. Hinnebusch BJ, Rudolph AE, Cherepanov P, Dixon JE, Schwan TG, Forsberg A. Role of Yersinia murine toxin in survival of Yersinia pestis in the midgut of the flea vector. Science. 2002; 296 (5568):733–5. https://doi.org/10.1126/science.1069972 PMID: 11976454. 5. Lindler LE, Plano GV, Burland V, Mayhew GF, Blattner FR. Complete DNA sequence and detailed anal- ysis of the Yersinia pestis KIM5 plasmid encoding murine toxin and capsular antigen. Infect. Immun. 1998; 66(12):5731–42. https://doi.org/10.1128/IAI.66.12.5731-5742.1998 PMID: 9826348. 6. Hildebrand GJ, Ng J, Von Metz EK, Eisler DM. Studies on the mechanism of circulatory failure induced in rats by Pasteurella pestis murine toxin. J. Infect. Dis. 1966; 116(5):615–29. https://doi.org/10.1093/ infdis/116.5.615 PMID: 5957268. 7. Ajl SJ, Reedal JS, Durrum EL, Warren J. Studies on plague. I. Purification and properties of the toxin of Pasteurella pestis. J. Bacteriol. 1955; 70(2):158–69. https://doi.org/10.1128/jb.70.2.158-169.1955 PMID: 13251981 8. Hinnebusch BJ, Cherepanov P, Du Y, Rudolph A, Dixon JD, Schwan T, et al. Murine toxin of Yersinia pestis shows phospholipase D activity but is not required for virulence in mice. Int. J. Med. Microbiol. 2000; 290(4–5):483–7. https://doi.org/10.1016/S1438-4221(00)80070-3 PMID: 11111930. 9. Brown SD, Montie TC. Beta-adrenergic blocking activity of Yersinia pestis murine toxin. Infect. Immun. 1977; 18(1):85–93. https://doi.org/10.1128/iai.18.1.85-93.1977 PMID: 198377. 10. Fan Y, Zhou Y, Feng N, Wang Q, Tian G, Wu X, et al. Recombinant murine toxin from Yersinia pestis shows high toxicity and beta-adrenergic blocking activity in mice. Microbes Infect. 2016; 18(5):329–35. https://doi.org/10.1016/j.micinf.2016.01.001 PMID: 26774329. 11. Montie TC, Ajl SJ. Microbial Toxins. Montie TC, Kadis S, Ajl SJ, editors. New York: Academic Press; 1970. 1–37 p. 12. 13. Falkow S. Molecular Koch’s postulates applied to microbial pathogenicity. Rev. Infect. Dis. 1988; 10 Suppl 2:S274–6. https://doi.org/10.1093/cid/10.supplement_2.s274 PMID: 3055197. Johnson TL, Hinnebusch BJ, Boegler KA, Graham CB, MacMillan K, Montenieri JA, et al. Yersinia murine toxin is not required for early-phase transmission of Yersinia pestis by Oropsylla montana (Siphonaptera: Ceratophyllidae) or Xenopsylla cheopis (Siphonaptera: Pulicidae). Microbiology. 2014; 160(Pt 11):2517–25. https://doi.org/10.1099/mic.0.082123-0 PMID: 25187626. 14. Bland DM, Jarrett CO, Bosio CF, Hinnebusch BJ. Infectious blood source alters early foregut infection and regurgitative transmission of Yersinia pestis by rodent fleas. PLoS Path. 2018; 14(1):e1006859. https://doi.org/10.1371/journal.ppat.1006859 PMID: 29357385. 15. Hinnebusch BJ, Chouikha I, Sun YC. Ecological opportunity, evolution, and the emergence of flea- borne plague. Infect. Immun. 2016; 84(7):1932–40. https://doi.org/10.1128/IAI.00188-16 PMID: 27160296. 16. Vaughan JA, Azad AF. Patterns of erythrocyte digestion by bloodsucking insects: constraints on vector competence. J. Med. Entomol. 1993; 30(1):214–6. https://doi.org/10.1093/jmedent/30.1.214 PMID: 8094460. 17. Bland DM, Martens CA, Virtaneva K, Kanakabandi K, Long D, Rosenke R, et al. Transcriptomic profiling of the digestive tract of the rat flea, Xenopsylla cheopis, following blood feeding and infection with Yersi- nia pestis. PLoS Negl. Trop. Dis. 2020; 14(9):e0008688. https://doi.org/10.1371/journal.pntd.0008688 PMID: 32946437. 18. Reinhardt CA. Ultrastructural comparison of the midgut epithelia of fleas with different feeding behavior patterns (Xenopsylla cheopis, Echidnophaga gallinacea, Tunga penetrans, Siphonaptera, Pulicidae). Act. Trop. 1976; 33(2):105–32. 8972. 19. Sarfati M, Krasnov BR, Ghazaryan L, Khokhlova IS, Fielden LJ, Degen AA. Energy costs of blood diges- tion in a host-specific haematophagous parasite. J. Exp. Biol. 2005; 208(Pt 13):2489–96. https://doi. org/10.1242/jeb.01676 PMID: 15961735. 20. Lehane MJ. The Biology of Blood Sucking Insects. 2nd ed. Cambridge: Cambridge University Press; 2005. 21. Vashchenok VS, Solina LT, Zhirnov AE. Digestion of blood of different animals by fleas Xenopsylla cheopis [In Russian]. Parazitologiya. 1976; 10(6):544–9. PMID: 1023147. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 18 / 19 PLOS PATHOGENSProtective effect of Yersinia murine toxin in the flea gut is dependent on source of host blood 22. Earl SC, Rogers MT, Keen J, Bland DM, Houppert AS, Miller C, et al. Resistance to innate immunity contributes to colonization of the insect gut by Yersinia pestis. PloS One. 2015; 10(7):e0133318. https:// doi.org/10.1371/journal.pone.0133318 PMID: 26177454. 23. Ludlow AB, Pauling CD, Marketon MM, Anderson DM. Drosophila as a model for understanding the insect host of Yersinia pestis. Methods Mol. Biol. 2019; 2010:167–78. https://doi.org/10.1007/978-1- 4939-9541-7_12 PMID: 31177438. 24. Hinnebusch BJ. The evolution of flea-borne transmission in Yersinia pestis. Curr. Issues Mol. Biol. 2005; 7(2):197–212. PMID: 16053250. 25. Hinnebusch BJ, Jarrett CO, Bland DM. "Fleaing" the plague: adaptations of Yersinia pestis to its insect vector that lead to transmission. Annu. Rev. Microbiol. 2017; 71:215–32. https://doi.org/10.1146/ annurev-micro-090816-093521 PMID: 28886687. 26. Dewitte A, Bouvenot T, Pierre F, Ricard I, Pradel E, Barois N, et al. A refined model of how Yersinia pes- tis produces a transmissible infection in its flea vector. PLoS Path. 2020; 16(4):e1008440. https://doi. org/10.1371/journal.ppat.1008440 PMID: 32294143. 27. Rasmussen S, Allentoft ME, Nielsen K, Orlando L, Sikora M, Sjogren KG, et al. Early divergent strains of Yersinia pestis in Eurasia 5,000 years ago. Cell. 2015; 163(3):571–82. https://doi.org/10.1016/j.cell. 2015.10.009 PMID: 26496604. 28. Rascovan N, Sjogren KG, Kristiansen K, Nielsen R, Willerslev E, Desnues C, et al. Emergence and spread of basal lineages of Yersinia pestis during the Neolithic decline. Cell. 2019; 176(1–2):295–305 e10. https://doi.org/10.1016/j.cell.2018.11.005 PMID: 30528431. 29. Spyrou MA, Tukhbatova RI, Wang CC, Valtuena AA, Lankapalli AK, Kondrashin VV, et al. Analysis of 3800-year-old Yersinia pestis genomes suggests Bronze Age origin for bubonic plague. Nat. Commun. 2018; 9(1):2234. https://doi.org/10.1038/s41467-018-04550-9 PMID: 29884871. 30. Andrades Valtueña A, Mittnik A, Key FM, Haak W, Allmae R, Belinskij A, et al. The Stone Age plague and its persistence in Eurasia. Curr. Biol. 2017; 27(23):3683–91 e8. https://doi.org/10.1016/j.cub.2017. 10.025 PMID: 29174893. 31. Demeure C, Dussurget O, Fiol GM, Le Guern AS, Savin C, Pizarro-Cerda J. Yersinia pestis and plague: an updated view on evolution, virulence determinants, immune subversion, vaccination and diagnos- tics. Microbes. Infect. 2019; 21(5–6):202–12. https://doi.org/10.1016/j.micinf.2019.06.007 PMID: 31252217. 32. Suntsov VV. Sympatric speciation of the plague microbe Yersinia pestis: monohostal specialization in the host-parasite marmot-flea (Marmota sibirica-Oropsylla silantiewi) system. Izv. Akad. Nauk. Ser. Biol. 2016;(2):117–27. PMID: 27396172. 33. Rudolph AE, Stuckey JA, Zhao Y, Matthews HR, Patton WA, Moss J, et al. Expression, characteriza- tion, and mutagenesis of the Yersinia pestis murine toxin, a phospholipase D superfamily member. J. Biol. Chem. 1999; 274(17):11824–31. https://doi.org/10.1074/jbc.274.17.11824 PMID: 10207000. 34. Perry RD, Fetherston JD. Yersinia pestis—etiologic agent of plague. Clin. Microbiol. Rev. 1997; 10 (1):35–66. https://doi.org/10.1128/CMR.10.1.35 PMID: 8993858. 35. Sikkema DJ, Brubaker RR. Resistance to pesticin, storage of iron, and invasion of HeLa cells by Yersi- niae. Infect. Immun. 1987; 55(3):572–8. https://doi.org/10.1128/iai.55.3.572-578.1987 PMID: 3818085. 36. Datsenko KA, Wanner BL. One-step inactivation of chromosomal genes in Escherichia coli K-12 using PCR products. Proc. Natl. Acad. Sci. USA. 2000; 97(12):6640–5. https://doi.org/10.1073/pnas. 120163297 PMID: 10829079. 37. Bland DM, Brown LD, Jarrett CO, Hinnebusch BJ, Macaluso KR. Methods in Flea Research. 2017. https://www.beiresources.org/Catalog/VectorResources.aspx 38. Bland DM, Hinnebusch BJ. Feeding behavior modulates biofilm-mediated transmission of Yersinia pes- tis by the cat flea, Ctenocephalides felis. PLoS Negl. Trop. Dis. 2016; 10(2):e0004413. https://doi.org/ 10.1371/journal.pntd.0004413 PMID: 26829486. 39. Altman PL. Blood and Other Body Fluids. 1 ed. Washington, D.C.: Federation of American Societies for Experimental Biology; 1961. 15 p. 40. Hinnebusch BJ, Bland DM, Bosio CF, Jarrett CO. Comparative ability of Oropsylla montana and Xenop- sylla cheopis fleas to transmit Yersinia pestis by two different mechanisms. PLoS Negl. Trop. Dis. 2017; 11(1):e0005276. https://doi.org/10.1371/journal.pntd.0005276 PMID: 28081130. PLOS Pathogens \| https://doi.org/10.1371/journal.ppat.1009995 October 14, 2021 19 / 19 PLOS PATHOGENS
10.2196_43965	JOURNAL OF MEDICAL INTERNET RESEARCH D'Amico et al Original Paper The Evolution of Primary Care Telehealth Disparities During COVID-19: Retrospective Cohort Study Rachel D'Amico1, MD; Patrick M Schnell2, PhD; Randi Foraker3, PhD; J Nwando Olayiwola4, MPH, MD; Daniel E Jonas1, MPH, MD; Seuli Bose Brill1, MD 1Department of Internal Medicine, The Ohio State University College of Medicine, Columbus, OH, United States 2Division of Biostatistics, College of Public Health, The Ohio State University, Columbus, OH, United States 3Division of General Medical Sciences, Washington University School of Medicine, St Louis, MO, United States 4Humana, Inc, Columbus, OH, United States Corresponding Author: Rachel D'Amico, MD Department of Internal Medicine The Ohio State University College of Medicine 395 W 12th Ave Columbus, OH, 43210 United States Phone: 1 6142938000 Email: rachel.d'amico@osumc.edu Abstract Background: Telehealth has become widely used as a novel way to provide outpatient care during the COVID-19 pandemic, but data about telehealth use in primary care remain limited. Studies in other specialties raise concerns that telehealth may be widening existing health care disparities, requiring further scrutiny of trends in telehealth use. Objective: Our study aims to further characterize sociodemographic differences in primary care via telehealth compared to in-person office visits before and during the COVID-19 pandemic and determine if these disparities changed throughout 2020. Methods: We conducted a retrospective cohort study in a large US academic center with 46 primary care practices from April-December 2019 to April-December 2020. Data were subdivided into calendar quarters and compared to determine evolving disparities throughout the year. We queried and compared billed outpatient encounters in General Internal Medicine and Family Medicine via binary logic mixed effects regression model and estimated odds ratios (ORs) with 95% CIs. We used sex, race, and ethnicity of the patient attending each encounter as fixed effects. We analyzed socioeconomic status of patients in the institution’s primary county based on the patient’s residence zip code. Results: A total of 81,822 encounters in the pre–COVID-19 time frame and 47,994 encounters in the intra–COVID-19 time frame were analyzed; in the intra–COVID-19 time frame, a total of 5322 (11.1%) of encounters were telehealth encounters. Patients living in zip code areas with high utilization rate of supplemental nutrition assistance were less likely to use primary care in the intra–COVID-19 time frame (OR 0.94, 95% CI 0.90-0.98; P=.006). Encounters with the following patients were less likely to be via telehealth compared to in-person office visits: patients who self-identified as Asian (OR 0.74, 95% CI 0.63-0.86) and Nepali (OR 0.37, 95% CI 0.19-0.72), patients insured by Medicare (OR 0.77, 95% CI 0.68-0.88), and patients living in zip code areas with high utilization rate of supplemental nutrition assistance (OR 0.84, 95% CI 0.71-0.99). Many of these disparities persisted throughout the year. Although there was no statistically significant difference in telehealth use for patients insured by Medicaid throughout the whole year, subanalysis of quarter 4 found encounters with patients insured by Medicaid were less likely to be via telehealth (OR 0.73, 95% CI 0.55-0.97; P=.03). Conclusions: Telehealth was not used equally by all patients within primary care throughout the first year of the COVID-19 pandemic, specifically by patients who self-identified as Asian and Nepali, insured by Medicare, and living in zip code areas with low socioeconomic status. As the COVID-19 pandemic and telehealth infrastructure change, it is critical we continue to reassess the use of telehealth. Institutions should continue to monitor disparities in telehealth access and advocate for policy changes that may improve equity. (J Med Internet Res 2023;25:e43965) doi: 10.2196/43965 https://www.jmir.org/2023/1/e43965 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e43965 \| p. 1 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH D'Amico et al KEYWORDS primary care; health care disparities; COVID-19; office visits; telemedicine; telehealth; health inequality; patient care; telehealth access; health care access Introduction The impact of the SARS-CoV-2 strain of coronavirus (also known as COVID-19) pandemic was felt throughout the world in 2020 and will influence the health care landscape for years to come. The COVID-19 pandemic has highlighted existing health care disparities; even when comorbidities are controlled for, Black patients have 2.7 times increased odds of hospitalization from COVID-19 compared to non-Hispanic White patients [1]. Asian Americans have 2.1 times higher percentage of deaths attributed to COVID-19 compared to non-Hispanic White Americans, with 1 in 7 Asian American deaths in 2020 attributable to COVID-19 [2]. The pandemic has also highlighted the importance of reliable primary care use. Primary care physicians or clinicians manage chronic conditions, like hypertension and diabetes, which are linked with increased mortality secondary to COVID-19 [3]. Primary care physicians are essential in decreasing health care disparities. Increased availability of primary care has been associated with reduced effects of income inequality on self-reported health [4] and all-cause mortality [5]. Despite this, before the COVID-19 pandemic, there were already significant racial and ethnic disparities in primary care access. Urban areas with a high proportion of Black patients were up to 28 times more likely to have limited access to primary care providers [6]. Asian Americans are more likely to be uninsured than non-Hispanic White Americans even after the Affordable Care Act [7]. Given the importance of primary care in patient outcomes and mitigation of health care disparities, it is essential that new models of providing primary care be analyzed critically for equity. Telehealth is defined by the Centers for Medicare and Medicaid Services as “exchange of medical information from one site to another through electronic communication [8].” At the beginning of the COVID-19 pandemic, telehealth infrastructure and reimbursement developed rapidly. Prior to March 2020, telehealth was reimbursed by Medicare in limited capacities only for patients in designated rural areas. In March 2020, Centers for Medicare and Medicaid Services broadened telehealth access to include all Medicare beneficiaries [8]. The US Department of Health and Human Services also waived restrictions on technology use not compliant with the Health Insurance Portability and Accountability Act to increase options for telehealth platforms [9]. Congress provided US $200 million in April 2020 to help US providers expand telehealth through the COVID-19 Telehealth Program [10]. These policy changes resulted in exponential growth of telehealth. Previous research shows that weekly telehealth visits increased 50-fold for one insurer [11]. Despite the swift increase in telehealth use, studies during the COVID-19 pandemic have demonstrated that Black [12] and Hispanic [13] patients, patients insured by Medicare and Medicaid [14], and those in lower-income areas [12] were less likely to use telehealth within subspecialty care. The data for https://www.jmir.org/2023/1/e43965 XSL•FO RenderX disparities in adult primary care telehealth use is conflicting but overall concerning for racial inequity [15,16]. In pediatric populations, Black and publicly insured individuals were less likely to use telehealth in primary care [17,18]. However, in patients 65 years of age and older, Black patients used telehealth more frequently than White patients [19]. Little is known about the chronological trends of these disparities. One previous study suggested that the disparities in telehealth lessened throughout 2020 [20]. Our study aimed to further characterize telehealth disparities in adult primary care and to assess how the increase in telehealth use impacted existing health care disparities in primary care, specifically examining differences by race or ethnicity, insurance status, and geographic location in who is using in-person office visits versus telehealth intra–COVID-19. We wanted to determine if any existing disparities were significant only at the beginning of the COVID-19 pandemic related to inequity with initial use or if these disparities continued despite increased clinician and institutional comfort with telehealth use. Methods Design We acquired retrospective data from an informational database for the electronic health record of The Ohio State University Wexner Medical Center, which allows researchers to access deidentified clinical data. All billed outpatient encounters in the Division of General Internal Medicine and Department of Family and Community Medicine were examined in 2 time and periods: pre–COVID-19 intra–COVID-19 (April-December 2020). These ranges were picked to have comparative time points for the pre- and intra–COVID-19 time frames during the initial peak of the pandemic. Our study aimed to determine how health care disparities may have changed throughout the first year of the COVID-19 pandemic, and data were subdivided into calendar quarters (eg, April-June). (April-December 2019) Participants Included variables were type of encounter, age, self-identified race, ethnicity, zip code, insurance type, and visit date. Exclusion criteria were ages outside the range of 18-99 years and encounters in departments other than Family Medicine and General Internal Medicine to focus on primary care usage only. Insurance types were separated into Medicaid, Medicare, private, marketplace (including exchange and marketplace policies), worker’s compensation, self-pay, and uninsured. Our analysis included zip codes for Franklin County, the primary catchment area for patients at The Ohio State University. Franklin County has a population of 1.3 million, of which around 13.5% live below the federal poverty line; 66.8% of the population is White and 23.8% is Black. There are multiple refugee populations located in Franklin County, particularly J Med Internet Res 2023 \| vol. 25 \| e43965 \| p. 2 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH D'Amico et al Nepali (around 23,000 people) and Somali (around 45,000) [12]. Ethical Considerations Data were deidentified prior to being given to the research team. The deidentified data request was approved by the Honest Broker Committee (HBOC Study ID #1273), which oversees all clinical data sets within The Ohio State University. Given that the data were deidentified and there were no interventions, the study was designated exempt from institutional review board approval. Statistical Analysis To determine how demographics changed between the pre–COVID-19 and intra–COVID-19 time frames and how they were associated with encounter type (ie, telehealth versus in-person) during the intra–COVID-19 time frame, we applied a binary logistic mixed effects regression model and estimated odds ratios (ORs) and 95% CIs. Time period and encounter type were used as the outcome variables for the respective analyses. Sex, race, and ethnicity of the patient attending each encounter were used as fixed effects. The model was fixed to account for within-office correlation. Because a patient could have multiple encounters, we attempted to fit a mixed effects model in which a random effect for patient was used to account for correlation of encounters by the same patient. However, the mixed effects model indicated no detectable within-patient correlation, so the random effects were dropped from the model. Zip Code Analysis Zip code data and insurance provider were analyzed via separate models. US Census Bureau data for the 48 zip codes within Franklin County was used [21] to determine the effect of geographic socioeconomic status on telehealth use. The percentage of people living below the federal poverty line (FPL) and the percentage of people receiving Supplemental Nutrition Assistance Program (SNAP; eg, “food stamps”) within the zip code were used as indicators of socioeconomic status. Two measurements were used to attempt to fully capture the nuances of socioeconomic disparities. The 10 zip codes with the highest percentage of people living below the FPL or receiving SNAP were classified as “high,” and the 5 zip codes with the lowest percentage of people below FPL and SNAP use were classified as “low,” with all other zip codes in the county categorized as “moderate.” High poverty level was thus identified as >30% of households under FPL, moderate poverty level was considered as 10%-30% of households under FPL, and low poverty level was when <10% of households were under FPL; high SNAP use was >25% of households receiving SNAP, moderate SNAP use was 5%-25% of households receiving SNAP, and low SNAP use was considered as <5% of households receiving SNAP. State and national levels of income were in the moderate categories; 12.6% of the population of Ohio and 11.9% of the US population lived under FPL in 2020 [22], and 13% of the population of Ohio and of the US used SNAP in 2021 [23]. For zip code data, FPL and SNAP were analyzed separately due to multicollinearity (FPL and SNAP categories were highly correlated). Zip codes associated with the university were excluded from analysis to avoid confounding of students artificially lowering income data (those zip codes were classified as a high percentage below FPL and moderate percentage receiving SNAP). Results A total of 81,822 encounters in the pre–COVID-19 time frame and 47,994 encounters in the intra–COVID-19 time frame were analyzed (Table 1). The sample’s racial or ethnic demographics were found to be similar to those of Franklin County, the institution’s primary county [24]. Table 1. Comparison of pre–COVID-19 and intra–COVID-19 encounter types. Pre–COVID-19 (n=81,822), n (%) Intra–COVID-19 (n=47,994), n (%) Overall (n=12,9816), n (%) Characteristics Encounter type Telehealth Office visit Department 9 (0) 81,813 (100) Family medicine 48,821 (59.7) General or internal medicine 33,001 (40.3) In Franklin County Yes No 61,246 (74.9) 20,576 (25.1) 5322 (11.1) 42,672 (88.9) 29,475 (61.4) 18,519 (38.6) 35,271 (73) 12,723 (26.5) 5331 (4.1) 124,485 (95.9) 78,296 (60.3) 51,520 (39.7) 96,517 (74.3) 33,299 (25.7) Comparison of Pre–COVID-19 and Intra–COVID-19 Time Frames There was a substantial decrease in overall primary care encounters during the intra–COVID-19 time frame compared to pre–COVID-19 time frame: between April-December 2019 and April-December 2020, the volume of overall primary care encounters decreased by over 40% (Table 1). Although significant decreases in primary care have been seen nationally, this seems to be a more significant decrease than national trends [25]. There were no significant differences by race or ethnicity between the patients with encounters in the pre–COVID-19 time frame compared to intra–COVID-19 time frame (Tables S1 and S2 in Multimedia Appendix 1). Compared to pre–COVID-19 time frame, there was a higher proportion of patients with marketplace insurance versus private insurance in https://www.jmir.org/2023/1/e43965 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e43965 \| p. 3 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH D'Amico et al the intra–COVID-19 time frame (OR 1.11, 95% CI 1.03-1.2; P=.006). The proportions of encounters with patients covered by private insurance, Medicare, and Medicaid were similar between pre- and intra–COVID-19 time frames. Encounters with patients living in Franklin County data were then separately analyzed (Table S1 in Multimedia Appendix 1). Patients living in a zip code area with high utilization rate of SNAP were less likely to use primary care in the intra–COVID-19 time frame compared to those living in a zip code area with low SNAP use (OR 0.94, 95% CI 0.90-0.98; P=.006). Telehealth Use Disparities The intra–COVID-19 data (April-December 2020) were then analyzed separately (Table 2). A total of 5322 telehealth and 42,672 office visits were identified. Of note, there were only 9 telehealth encounters in the pre–COVID-19 time frame. Table 2. Racial or ethnic and gender demographics of intra–COVID-19 encounters separated by encounter type. Characteristics Telehealth (n=5322), n (%) Office visit (n=42,672), n (%) Overall (n=47,994), n (%) Sex Female Male Unknown Race or ethnicity White African or Black American Indian Asian Middle Eastern Multiple Nepali Somali Pacific Islander Other Unknown Missing Ethnicity 3282 (61.7) 2040 (38.3) 0 (0) 3704 (69.6) 1056 (19.8) 13 (0.2) 213 (4) 33 (0.6) 58 (1.1) 15 (0.3) 12 (0.2) 2 (0) 181 (3.4) 34 (0.6) 1 (0) 24,511 (57.4) 18,160 (42.6) 1 (0) 28,190 (66.1) 9312 (21.8) 78 (0.2) 2189 (5.1) 216 (0.5) 403 (0.9) 300 (0.7) 118 (0.3) 31 (0.1) 1527 (3.6) 305 (0.7) 43 (0) 27,793 (57.9) 20,200 (42.1) 1 (0) 31,894 (66.5) 10,368 (21.6) 91 (0.2) 2402 (5) 249 (0.5) 461 (1) 315 (0.7) 130 (0.3) 33 (0.1) 1708 (3.6) 339 (0.7) 4 (0) Not Hispanic or Latinx 5146 (96.7) 41,215 (96.6) 46,361 (96.6) Hispanic or Latinx Ashkenazi Jew Unknown 140 (2.6) 2 (0) 34 (0.6) 1115 (2.6) 20 (0) 322 (0.8) 1255 (2.6) 22 (0) 356 (0.7) Racial or Ethnic Disparities Compared to patients who identified as White, encounters during 2020 with patients of the following racial or ethnic groups were significantly less likely to be via telehealth: Asian (OR 0.74, 95% CI 0.63-0.86; P<.001) and Nepali (OR 0.37, 95% CI 0.19-0.72; P=.003). In quarterly subanalyses (Table S3 in Multimedia Appendix 1), there was a significant disparity in telehealth use within the same quarter for encounters with Asian patients in quarters 2 and 3 and for Nepali patients in quarters 3 and 4, compared to White patients. Insurance Disparities Compared to private insurance, overall encounters in 2020 with patients covered by Medicare (OR 0.77, 95% CI 0.68-0.88; P<.001) were less likely to be via telehealth (Table 3). When analyzed by quarters, encounters with patients insured by Medicare were less likely to be via telehealth in quarters 2 and 3. Although there was no statistically significant difference in telehealth use for patients insured by Medicaid throughout the whole year, subanalysis of quarter 4 found encounters with patients insured by Medicaid were less likely to be via telehealth (OR 0.73, 95% CI 0.55-0.97; P=.03). https://www.jmir.org/2023/1/e43965 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e43965 \| p. 4 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH D'Amico et al Table 3. Primary insurance coverage of intra–COVID-19 encounters separated by encounter type. Insurance type Telehealth (n=5322), n (%) Office visit (n=42,672), n (%) Overall (n=47,994), n (%) Private Marketplace Medicaid Medicare Other Self-pay Uninsured Veterans Affairs Worker’s compensation Missing 2968 (55.8) 140 (2.6) 747 (14) 1315 (24.7) 129 (2.4) 1 (0) 14 (0) 2 (0) 0 (0) 6 (0) 21,710 (50.9) 1061 (2.5) 6345 (14.9) 12,422 (29.1) 1048 (2.5) 4 (0) 2 (0) 14 (0) 5 (0) 61 (0.1) 24,678 (51.4) 1201 (2.5) 7092 (14.8) 13,737 (28.6) 1177 (2.5) 5 (0) 16 (0) 16 (0) 5 (0) 67 (0.1) Geographic Disparities Encounters with patients living in zip code areas with high SNAP use compared to those living in areas with low SNAP use (OR 0.84, 95% CI 0.71 to 0.99; P=.04) throughout 2020 were less likely to be via telehealth compared to in-person office visits (Table 4). When comparing quarterly data, there were statistically significant differences in only quarter 4. Table 4. Subanalysis of Franklin County zip code data by encounter type. Characteristics Telehealth (n=3937), n (%) Office visit (n=31,334), n (%) Overall (n=35,271), n (%) FPLa Low Moderate High SNAP useb Low Moderate High 469 (8.8) 3042 (57.2) 426 (8) 650 (12.2) 2653 (49.8) 634 (11.9) 3837 (9) 23,434 (54.9) 4063 (9.5) 4887 (11.5) 20,761 (48.7) 5686 (13.3) 4306 (9) 26,476 (55.2) 4489 (9.4) 5537 (11.5) 23,414 (48.8) 6320 (13.2) aFPL: federal poverty line; high poverty level was identified as >30% of households under FPL; moderate poverty level was considered as 10%-30% of households under FPL, and low poverty level was considered as <10% of households under FPL. bSNAP: Supplemental Nutrition Assistance Program; high SNAP use: >25% of households receiving SNAP; moderate SNAP use: 5%-25% of households receiving SNAP; and low SNAP use: <5% of households receiving SNAP. Discussion The results of our retrospective cohort study demonstrate that certain patient populations who are more likely to experience health care disparities, including Asian and Nepali patients, those that are insured with Medicare, and patients in areas with high rates of poverty, are not using telehealth at equal rates in primary care settings (Table 5). We found that these disparities persisted throughout 2020. Previous research has shown that disparities in telehealth use in medically underserved areas lessened during 2020 [20]; however, our data contradict this. In addition, patients insured with Medicaid were found to have a new decrease in telehealth use at the end of 2020, suggesting continuing development in disparities that should be explored. There are several possible explanations for these disparities. Inequitable patient access to telehealth infrastructure is one of the largest concerns. The recent Infrastructure Investment and Jobs Act will allocate US $65 billion for digital-inclusion https://www.jmir.org/2023/1/e43965 XSL•FO RenderX initiatives, including increasing broadband accessibility [26], in the hopes to curb worsening access disparities. Given the high proportion of refugee populations in Franklin County, including Nepali population, language barriers may make telehealth more difficult to access. As we found disparities in patients insured by Medicare, older patients may lack the education or comfort to adequately use telehealth. Although provider capabilities certainly play a role in telehealth use, one of the strengths of our single center design is that all clinics included in this study had similar telehealth infrastructure and support. With the public health emergency declaration coming to an end in 2022, government officials and insurance payors must decide to what extent telehealth will continue to be reimbursed [27]. If telehealth becomes an increasingly central part of ambulatory care, we may continue to see widening disparities in primary care engagement if telehealth comes at the expense of in-person office visits. Patients must also be included in these conversations to determine what barriers, as well as preferences, affect how they access primary care. J Med Internet Res 2023 \| vol. 25 \| e43965 \| p. 5 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH D'Amico et al There are some potential opportunities to improve equity within telehealth; quality metric use may improve disparities by allowing providers to offer care options that work best for their patients. Quality metric use was recently shown to decrease racial disparities in postpartum follow-up [28]. Some Medicare plans already incorporate Accountable Care Organizations that have quality metrics [29] that could be modified to be telehealth appropriate. Patient education on electronic portals and telehealth applications is critical to ensuring those with less familiarity with broadband can use telehealth. Interventions as easy as previsit telephone calls have been shown to increase completion of telehealth in underserved populations [30]. Our research does have limitations. As a single-center study, our results may be difficult to generalize to other centers. Additionally, we did not examine the type of care provided during these visits (eg, acute, chronic, and preventative) or if telehealth visits were done by video or telephone, which provides important context. We did not assess patient’s primary language, which may be an additional barrier to telehealth [31]. Our study combined both General Internal Medicine and Family Medicine encounters to develop a comprehensive understanding of primary care telehealth at our institution; however, this may make our data difficult to analyze in the setting of one department. Nationally, telehealth accounted for 14% of commercially insured ambulatory encounters in 2020 [32], and these rates were maintained into 2021 [33], suggesting that telehealth has entered the mainstream of ambulatory care. The persistent disparities described by our study suggest that ongoing policy discussions about telehealth should focus on health equity to prevent further widening of health disparities in primary care. Our study was also able to examine the changing of telehealth use throughout the year of 2020. By monitoring how the pandemic has continued to evolve, we can determine if changes in telehealth infrastructure are inadvertently affecting specific populations. Continuing to monitor changes in telehealth access is critical to evaluating equity interventions. Table 5. Summary of significant findings that were statistically significant for 2020 and the quarters in which these disparities were found. Population Odds ratio (95% CI) P value Quarters this was statistically significant Less likely to access care in intra–COVID-19 Zip code area with high SNAPa use 0.94 (0.9-0.98) .006 2 Less likely to use telehealth Male gender Asian Nepali Medicare beneficiaries 0.83 (0.73-0.94) 0.74 (0.63-0.86) 0.37 (0.19-0.72) 0.77 (0.68-0.88) Zip code area with high SNAP use 0.84 (0.71-0.99) aSNAP: Supplemental Nutrition Assistance Program. .003 <.001 .003 <.001 .04 3 and 4 2 and 3 3 and 4 2 and 3 4 Acknowledgments We would like to thank The Ohio State University Information Warehouse for their support in data housing and extraction. Data Availability The data sets generated and analyzed during this study are available from the corresponding author on reasonable request. Interested parties will be required to complete an institutional Data Use Agreement, and data will be made available via Secure Data transfer. Conflicts of Interest None declared. Multimedia Appendix 1 Supplementary material. [DOCX File , 31 KB-Multimedia Appendix 1] References 1. Azar KMJ, Shen Z, Romanelli RJ, Lockhart SH, Smits K, Robinson S, et al. Disparities in outcomes among COVID-19 patients in a large health care system in California. Health Aff (Millwood) 2020 Jul 01;39(7):1253-1262. [doi: 10.1377/hlthaff.2020.00598] [Medline: 32437224] https://www.jmir.org/2023/1/e43965 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e43965 \| p. 6 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH D'Amico et al 2. 3. 4. 5. 6. 7. Yan BW, Hwang AL, Ng F, Chu JN, Tsoh JY, Nguyen TT. Death toll of COVID-19 on Asian Americans: disparities revealed. J Gen Intern Med 2021 Nov 04;36(11):3545-3549 [FREE Full text] [doi: 10.1007/s11606-021-07003-0] [Medline: 34347256] CDC COVID-19 Response Team. Preliminary estimates of the prevalence of selected underlying health conditions among patients with coronavirus disease 2019 - United States, February 12-March 28, 2020. MMWR Morb Mortal Wkly Rep 2020 Apr 03;69(13):382-386 [FREE Full text] [doi: 10.15585/mmwr.mm6913e2] [Medline: 32240123] Shi L, Starfield B. Primary care, income inequality, and self-rated health in the United States: a mixed-level analysis. Int J Health Serv 2016 Jun 22;30(3):541-555. [doi: 10.2190/n4m8-303m-72ua-p1k1] Shi L, Macinko J, Starfield B, Politzer R, Wulu J, Xu J. Primary care, social inequalities, and all-cause, heart disease, and cancer mortality in us counties, 1990. Am J Public Health 2005 Apr;95(4):674-680. [doi: 10.2105/ajph.2003.031716] Brown EJ, Polsky D, Barbu CM, Seymour JW, Grande D. Racial disparities in geographic access to primary care in Philadelphia. Health Aff (Millwood) 2016 Aug 01;35(8):1374-1381. [doi: 10.1377/hlthaff.2015.1612] [Medline: 27503960] Park S. The effects of the affordable care act on health care access and utilization among Asian American subgroups. Med Care 2019;57(11):861-868. [doi: 10.1097/mlr.0000000000001202] 8. Medicare telemedicine health care provider fact sheet. Centers for Medicare & Medicaid Services. URL: https://www. 9. 10. 11. cms.gov/newsroom/fact-sheets/medicare-telemedicine-health-care-provider-fact-sheet [accessed 2022-01-18] OCR announces notification of enforcement discretion for telehealth remote communications during the COVID-19 nationwide public health emergency. US Department of Health & Human Services. URL: https://public3.pagefreezer.com/ content/HHS.gov/31-12-2020T08:51/https:/www.hhs.gov/about/news/2020/03/17/ ocr-announces-notification-of-enforcement-discretion-for-telehealth-remote-communications-during-the-covid-19.html [accessed 2022-01-18] Federa Communications Commission. URL: https://www.fcc.gov/covid-19-telehealth-program-invoices-reimbursements [accessed 2022-01-18] Friedman AB, Gervasi S, Song H, Bond AM, Chen AT, Bergman A, et al. Telemedicine catches on: changes in the utilization of telemedicine services during the COVID-19 pandemic. Am J Manag Care 2022 Jan 01;28(1):e1-e6 [FREE Full text] [doi: 10.37765/ajmc.2022.88771] [Medline: 35049260] 12. Quick Facts: Franklin County, Ohio. United States Census Bureau. URL: https://www.census.gov/quickfacts/ 13. 14. franklincountyohio [accessed 2022-05-30] Schmidt AL, Bakouny Z, Bhalla S, Steinharter JA, Tremblay DA, Awad MM, et al. Cancer care disparities during the COVID-19 pandemic: COVID-19 and cancer outcomes study. Cancer Cell 2020 Dec 14;38(6):769-770 [FREE Full text] [doi: 10.1016/j.ccell.2020.10.023] [Medline: 33176161] Schifeling CH, Shanbhag P, Johnson A, Atwater RC, Koljack C, Parnes BL, et al. Disparities in video and telephone visits among older adults during the COVID-19 pandemic: cross-sectional analysis. JMIR Aging 2020 Nov 10;3(2):e23176 [FREE Full text] [doi: 10.2196/23176] [Medline: 33048821] 15. Eberly LA, Kallan MJ, Julien HM, Haynes N, Khatana SAM, Nathan AS, et al. Patient characteristics associated with telemedicine access for primary and specialty ambulatory care during the COVID-19 pandemic. JAMA Netw Open 2020 Dec 01;3(12):e2031640 [FREE Full text] [doi: 10.1001/jamanetworkopen.2020.31640] [Medline: 33372974] 16. Rodriguez JA, Betancourt JR, Sequist TD, Ganguli I. Differences in the use of telephone and video telemedicine visits during the COVID-19 pandemic. Am J Manag Care 2021 Jan 14;27(1):21-26 [FREE Full text] [doi: 10.37765/ajmc.2021.88573] [Medline: 33471458] 17. Walters J, Johnson T, DeBlasio D, Klein M, Sikora K, Reilly K, et al. Integration and impact of telemedicine in underserved pediatric primary care. Clin Pediatr (Phila) 2021 Oct 12;60(11-12):452-458. [doi: 10.1177/00099228211039621] [Medline: 34382880] Schenker RB, Laguna MC, Odisho AY, Okumura MJ, Burnett H. Are we reaching everyone? A cross-sectional study of telehealth inequity in the COVID-19 pandemic in an urban academic pediatric primary care clinic. Clin Pediatr (Phila) 2022 Jan 11;61(1):26-33 [FREE Full text] [doi: 10.1177/00099228211045809] [Medline: 34514898] 18. 19. Ryskina KL, Shultz K, Zhou Y, Lautenbach G, Brown RT. Older adults' access to primary care: Gender, racial, and ethnic disparities in telemedicine. J Am Geriatr Soc 2021 Oct 08;69(10):2732-2740 [FREE Full text] [doi: 10.1111/jgs.17354] [Medline: 34224577] 20. Adepoju O, Liaw W, Chae M, Ojinnaka C, Britton E, Reves S, et al. COVID-19 and telehealth operations in Texas primary care clinics: disparities in medically underserved area clinics. J Health Care Poor Underserved 2021;32(2):948-957. [doi: 10.1353/hpu.2021.0073] [Medline: 34120986] Franklin County HealthMap 2019. The Ohio State University. URL: https://tinyurl.com/yknawceh [accessed 2021-02-02] Small Area Income and Poverty Estimates (SAIPE). US Census Bureau. URL: https://www.census.gov/data-tools/demo/ saipe/#/?s_state=39&s_geography=county&s_county= [accessed 2022-04-18] 21. 22. 23. Hall L, Nchako C. A closer look at who benefits from SNAP: state-by-state fact sheets. Center on Budget and Policy Priorities. 2023. URL: https://tinyurl.com/7fahrc5y [accessed 2022-04-18] 24. Quick facts Franklin County, Ohio. US Census Bureau. URL: https://www.census.gov/quickfacts/franklincountyohio [accessed 2021-05-21] https://www.jmir.org/2023/1/e43965 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e43965 \| p. 7 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH D'Amico et al 25. Mehrotra A. The impact of COVID-19 on outpatient visits in 2020: visits remained stable, despite a late surge in cases. The Commonwealth Fund. 2021. URL: https://tinyurl.com/3vdypfa2 [accessed 2022-04-18] 26. Rodriguez JA, Shachar C, Bates DW. Digital inclusion as health care — supporting health care equity with 27. digital-infrastructure initiatives. N Engl J Med 2022 Mar 24;386(12):1101-1103. [doi: 10.1056/nejmp2115646] Jercich K. House extends virtual care flexibilities beyond public health emergency. HealthcareITnews. 2022. URL: https:/ /www.healthcareitnews.com/news/house-extends-virtual-care-flexibilities-beyond-public-health-emergency [accessed 2022-04-30] 28. Khosla K, Suresh S, Mueller A, Perdigao JL, Stewart K, Duncan C, et al. Elimination of racial disparities in postpartum hypertension follow-up after incorporation of telehealth into a quality bundle. Am J Obstet Gynecol MFM 2022 May;4(3):100580. [doi: 10.1016/j.ajogmf.2022.100580] [Medline: 35121193] Philip S. Telehealth under alternative payment models. Milliman. 2017. URL: https://www.milliman.com/en/insight/ telehealth-under-alternative-payment-models [accessed 2022-04-22] 29. 30. Gusdorf RE, Shah KP, Triana AJ, McCoy AB, Pabla B, Scoville E, et al. A patient education intervention improved rates of successful video visits during rapid implementation of telehealth. J Telemed Telecare 2021 May 11:1357633X2110087. [doi: 10.1177/1357633x211008786] 31. Thomason J, Bays A, Mantilla B, Huang I, Najjar R, Singh N, et al. Non-English language preference associated with decreased rheumatology telehealth use during the COVID-19 pandemic. ACR Open Rheumatol 2022 May 27;4(5):385-394 [FREE Full text] [doi: 10.1002/acr2.11407] [Medline: 35084116] 32. Hatef E, Lans D, Bandeian S, Lasser EC, Goldsack J, Weiner JP. Outcomes of in-person and telehealth ambulatory encounters during COVID-19 within a large commercially insured cohort. JAMA Netw Open 2022 Apr 01;5(4):e228954 [FREE Full text] [doi: 10.1001/jamanetworkopen.2022.8954] [Medline: 35471570] 33. Bestsennyy O, Gilbert G, Harris A, Rost J. Telehealth: a quarter-trillion-dollar post-COVID-19 reality? McKinsey & Company. 2021. URL: https://tinyurl.com/3mv8yxxt [accessed 2022-04-30] Abbreviations FPL: federal poverty line OR: odds ratio SNAP: Supplemental Nutrition Assistance Program Edited by A Mavragani; submitted 31.10.22; peer-reviewed by JF Echelard, F Petrazzuoli, C Gibson-Gill; comments to author 06.01.23; revised version received 25.01.23; accepted 26.01.23; published 17.05.23 Please cite as: D'Amico R, Schnell PM, Foraker R, Olayiwola JN, Jonas DE, Brill SB The Evolution of Primary Care Telehealth Disparities During COVID-19: Retrospective Cohort Study J Med Internet Res 2023;25:e43965 URL: https://www.jmir.org/2023/1/e43965 doi: 10.2196/43965 PMID: 37146176 ©Rachel D'Amico, Patrick M Schnell, Randi Foraker, J Nwando Olayiwola, Daniel E Jonas, Seuli Bose Brill. Originally published in the Journal of Medical Internet Research (https://www.jmir.org), 17.05.2023. This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included. https://www.jmir.org/2023/1/e43965 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e43965 \| p. 8 (page number not for citation purposes)
10.1371_journal.pone.0287584	RESEARCH ARTICLE Effect of speech-stimulus degradation on phoneme-related potential Min-Jae Jeon1, Jihwan WooID 1,2* 1 Department of Electrical, Electronic and Computer Engineering, University of Ulsan, Ulsan, Republic of Korea, 2 Department of Biomedical Engineering, University of Ulsan, Ulsan, Republic of Korea * jhwoo@ulsan.ac.kr Abstract a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Jeon M-J, Woo J (2023) Effect of speech- stimulus degradation on phoneme-related potential. PLoS ONE 18(6): e0287584. https://doi. org/10.1371/journal.pone.0287584 Editor: Ru¨diger Land, Hannover Medical School: Medizinische Hochschule Hannover, GERMANY Received: September 26, 2022 Accepted: June 8, 2023 Published: June 23, 2023 Copyright: © 2023 Jeon, Woo. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All data files are available from the OSF database (https://doi.org/ 10.17605/OSF.IO/PQUNJ). Funding: This work was supported by the 2020 Research Fund of the University of Ulsan. There was no additional external funding received for this study. Competing interests: The authors have declared that no competing interests exist. Auditory evoked potential (AEP) has been used to evaluate the degree of hearing and speech cognition. Because AEP generates a very small voltage relative to ambient noise, a repetitive presentation of a stimulus, such as a tone, word, or short sentence, should be employed to generate ensemble averages over trials. However, the stimulation of repetitive short words and sentences may present an unnatural situation to a subject. Phoneme- related potentials (PRPs), which are evoked-responses to typical phonemic stimuli, can be extracted from electroencephalography (EEG) data in response to a continuous storybook. In this study, we investigated the effects of spectrally degraded speech stimuli on PRPs. The EEG data in response to the spectrally degraded and natural storybooks were recorded from normal listeners, and the PRP components for 10 vowels and 12 consonants were extracted. The PRP responses to a vocoded (spectrally-degraded) storybook showed a sta- tistically significant lower peak amplitude and were prolonged compared with those of a nat- ural storybook. The findings in this study suggest that PRPs can be considered a potential tool to evaluate hearing and speech cognition as other AEPs. Moreover, PRPs can provide the details of phonological processing and phonemic awareness to understand poor speech intelligibility. Further investigation with the hearing impaired is required prior to clinical application. 1. Introduction Speech perception, which is the ability to hear and process speech information, is widely evalu- ated using either behavioral or electrophysiological tests in clinical settings. Behavioral tests are simple and straightforward tasks that use sentences, words, and phonemes [1, 2]. Behav- ioral tests require the behavioral response of a subject, and hence, the outcomes may be affected by the subject’s condition [1–3]. Electroencephalography (EEG) is a popular method for measuring the electrical activity of the auditory nerve or brain in response to acoustic sti- muli, which is typically known as auditory evoked potential (AEP). AEP has been reliably used in clinic settings and research. The speech perception abilities of hearing aids and cochlear implant users have been sys- tematically accessed by cortical event responses [4, 5]. Cortical AEP can be effectively used to understand speech discrimination and intelligibility of people for whom it is difficult to PLOS ONE \| https://doi.org/10.1371/journal.pone.0287584 June 23, 2023 1 / 12 PLOS ONEEffects of degraded speech stimuli on phoneme-related potentials (PRPs) behaviorally test auditory function. However, because the measurement of AEP requires the use of electrodes placed on the scalp, which can result in a low signal-to-noise ratio, AEP requires repetitive presentation of a stimulus to obtain an ensemble average. An approach that uses repetitive stimuli may present an uncomfortable and unnatural situation for a subject [3, 6, 7]. Recently, continuous sentence stimuli, instead of a word or short-duration tone, have been used to access speech processing because they provide a more ecological measure of speech perception. The potential techniques, cortical tracking of speech envelope, temporal response function, and decoding speech from EEG can allow access to higher-level speech pro- cessing [3, 8, 9]. However, the AEP has a limitation in understanding phonemic processing at the cortical level and phonemic awareness in detail, which may be important for reading and spelling. It is also important to access the improvement of phonemic processing for users of hearing aids and cochlear implants to emphasize the benefit of auditory prosthesis [10, 11]. Recently, it has been investigated that the phoneme-related potential (PRP) can be obtained from time-locked responses to phoneme instances from a continuous storybook rather than repetitive stimulations of words and short sentences [12]. The findings of the above study pro- vide evidence that phonetic information propagates along the auditory pathway and is subse- quently encoded in the brain. The PRP showed morphological similarity to the P1-N1-P2 complex of auditory middle latency response. However, as the PRP in hearing-impaired and in response to degraded speech has not been investigated yet, it cannot be considered a possi- ble tool to evaluate hearing and phonemic awareness. Each peak component of AEP is related to neural information processing, and an increase in the amplitude and decrease in the latency of each component generally denotes more informative processing in the human brain [13]. Therefore, we hypothesize that PRP, similar to other types of AEPs, can cause changes in peak amplitude and latency dependent on phonological processing. In this study, we investigated the effect of spectrally degraded speech on the amplitude and latency of PRP. A continuous storybook evoked EEG data from normal listeners were recorded to extract the PRP compo- nents for 10 vowels and 12 consonants. The results showed that the PRP responses to a vocoded (spectrally-degraded) storybook resulted in a statistically significant lower peak amplitude and were prolonged compared with those of a natural storybook. Finally, this study provided evidence on the feasibility of using PRP as an objective test for speech perception and as a useful tool for understanding phonemic awareness. 2. Methods 2.1. Participants Twenty subjects (21.4±1.7 years old, 10 males and 10 females) participated in this study. All participants had no speech or hearing impairment and were native Korean speakers. The experimental procedures were reviewed and approved by the Institutional Review Board of the University of Ulsan and all participants signed an informed consent. 2.2. Natural and vocoded continuous storybook stimuli A female speaker recorded the stimulus storybook. The storybook consisted of noise-vocoded speech with degraded spectral detail and natural speech to provide different intelligible speech conditions [14]. The natural storybook was spectrally degraded using a noise vocoder. It was filtered using eight bandpass filters whose cutoff frequencies were logarithmically spaced between 200 Hz and 5000 Hz. After modulation with white Gaussian noise, the filtered data were synthesized for the vocoded storybook. The noise-vocoded speech story consisted of 398 sentences and the natural speech story consisted of 458 sentences. The duration of each story- book was 30 min. The storybook was presented at a comfortable hearing level of 60–70 dBA PLOS ONE \| https://doi.org/10.1371/journal.pone.0287584 June 23, 2023 2 / 12 PLOS ONEEffects of degraded speech stimuli on phoneme-related potentials (PRPs) Fig 1. Comparison of questionnaire scores between the vocoded and natural cases using the Wilcoxon signed rank test (n = 20). https://doi.org/10.1371/journal.pone.0287584.g001 using a loudspeaker placed 1 m from the participant in a soundproof room. The participants were asked to listen attentively to the story and watch the cross on the monitor. The experi- ment consisted of six sessions (10 min per session) with a 5-min rest between sessions. A ques- tionnaire was provided during each rest to confirm the attentiveness of the participants. The questionnaire consisted of nine questions about the story, and each question was designed to evaluate whether the story was comprehensible. The scores of each subject are detailed in Fig 1. The score for the vocoded story (mean: 53.3) is typically lower compared that of the natural story (mean: 81.5) (p < 0.001, Wilcoxon Signed Rank Test), indicating that there is a difference in understanding between the two cases. 2.3. Electroencephalography Brain activity in response to storybook stimulation was measured using a 64-channel EEG sys- tem at a rate of 2048 Hz (Biosemi Co., Netherlands). The EEG data were preprocessed using common referencing and 2–57 Hz bandpass filtering for baseline correction. The data were PLOS ONE \| https://doi.org/10.1371/journal.pone.0287584 June 23, 2023 3 / 12 PLOS ONEEffects of degraded speech stimuli on phoneme-related potentials (PRPs) Table 1. Number of utterances for each phoneme in the original and vocoded storybooks. The phonemes with incidences of over 100 were used to compute the pho- neme-related potential (PRP). The PRP of ‘ㅢ /ui/’ (marked ‘’) was particularly computed to compare with the vocoded PRP. The number in parentheses denotes the number of averaging single-trials PRP. Natural storybook Vowel Phoneme Incidence Phoneme ㅏ /a/ ㅣ /i/ ㅡ /eu/ ㅓ /eo/ ㅗ /o/ ㅜ /u/ ㅐ /ae/ ㅔ /e/ ㅕ /yeo/ ㅢ /ui/ 1494(1453) 812(798) 736(721) 559(548) 568(493) 326(321) 279(274) 251(239) 244(232) 70(69) ㄴ /n/ ㄱ /g/ ㄷ /d/ ㄹ /l/ ㅁ /m/ ㅅ /s/ ㅈ /dz/ ㄹ /r/ ㅎ /h/ ㅂ /b/ ㅇ /ng/ ㅊ /ch/ https://doi.org/10.1371/journal.pone.0287584.t001 Consonant Incidence 1296(1268) 874(857) 800(762) 595(565) 486(476) 452(444) 407(393) 342(339) 368(344) 270(265) 251(229) 87(86)* Vocoded storybook Phoneme ㅏ /a/ ㅣ /i/ ㅡ /eu/ ㅓ /eo/ ㅗ /o/ ㅜ /u/ ㅐ /ae/ ㅔ /e/ ㅕ /yeo/ ㅢ /ui/ Vowel Incidence 1541(1453) 920(798) 857(721) 679(548) 509(493) 367(321) 311(274) 250(239) 242(232) 106(69) Consonant Phoneme Incidence ㄴ /n/ ㄱ /g/ ㄷ /d/ ㄹ /l/ ㅁ /m/ ㅅ /s/ ㅈ /dz/ ㄹ /r/ ㅎ /h/ ㅂ /b/ ㅇ /ng/ ㅊ /ch/ 1471(1268) 931(857) 793(762) 590(565) 532(476) 530(444) 413(393) 423(339) 358(344) 326(265) 235(229) 119(86) resampled to 256 Hz to enhance computational efficiency. Eye blink-related noise was removed using independent component analysis [15]. 2.4. Phoneme-related potential There are 20 noun-related phonemes and 20 vowel-related phonemes in the Korean story- book, as listed in Table 1. The phoneme onset was extracted from the storybook using Praat software (Paul Boersma and David Weenink, Phonetic Sciences, University of Amsterdam, Netherlands), which is an open-source software program for speech phonetics [16, 17]. The utterance rates of the phonemes are summarized in Table 1. In this study, phonemes with a sufficient number of utterances (n > 100) were used to reliably compute the ensemble neural activity. There was no statistical difference (p = 0.782) in the number of phoneme utterances of natural and noise-vocoded speech stories. To compute the PRP of a neural activity in response to a specific phoneme [12], we segmented the EEG signals before/after phoneme onset (0 ms) of the utterance time to have each interval of 100–600 ms, as shown in Fig 2. Fig 2 shows an example of the PRP of a typical phoneme /a/ obtained by averaging 100 segmented EEG signals. As spectral dominance of PRPs was observed in the range of 4–9 Hz, the PRPs were post-processed using a bandpass (2–15 Hz) filter [12]. Although the frequency of occurrence of each phoneme exceeds 100, the frequencies of occur- rence of a phoneme differ between the natural and vocoded stories. The number of averaging sin- gle-trials PRP was matched to the smaller of the two, as noted in parentheses in Table 1. 2.5. Statistical analysis The PRPs were calculated by averaging single trials of EEG signals in response to a phonemic stimulus. The amplitude and latency of the PRP component of P1, N1, and P2 were deter- mined and subsequently compared between two experimental conditions (natural and vocoded story) within a subject. As this study was designed with only two conditions, the paired t-test was used to investigate if a statistically significant difference existed. Furthermore, the statistically significant difference over time in PRP waveforms between two conditions was determined by computing the t-value using a paired t-test at each time point. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287584 June 23, 2023 4 / 12 PLOS ONEEffects of degraded speech stimuli on phoneme-related potentials (PRPs) Fig 2. Example of the computation of the phoneme related potential (PRP) in response to the typical phoneme of ‘ㅏ /a/’. The stimulus spectrograms of part of the storybook are shown in the two upper panels, and the following panel shows a plot of details of electroencephalography (EEG) data in response to storybook stimuli. The EEG data were segmented based on the onset time of phoneme ‘ㅏ /a/’ and averaged to obtain the PRP of phoneme ‘ㅏ /a/’. https://doi.org/10.1371/journal.pone.0287584.g002 3. Results 3.1. Comparison of the grand-averaged PRPs of natural and vocoded speech stories Fig 3A shows the average PRPs in response to all natural (upper panel) and vocoded phonemes (middle panel). The gray thin line in each panel denotes the averaged PRP across all channels and all phonemes for each participant. The grand-averaged PRPs are indicated by a thick line, and the PRPs from each subject were plotted using a thin gray line. The lower panel in Fig 3A shows a comparison of the grand-averaged PRPs to natural (red) and vocoded (blue) story- books. Differences on the P1, N1, and P3 components in the grand-averaged PRPs were clearly observed. Higher and earlier peaks in each component in natural PRP are prominent. Here, the t-values at each time point over time were computed by a paired t-test between natural and vocoded grand averaged PRPs. The intervals of R1 (50–80 ms), R2 (110–140 ms), R3 (170–260 ms), R4 (350–410 ms), and R5 (450–470 ms) were set based on the statistical differences (p < 0.05) between two plots as seen in the bottom panel of Fig 3A. To understand the neural processing in the brain at each interval, a topographical PRP map was obtained by averaging the PRPs within each interval at a channel (Fig 3B). The p-value topographic map (a paired t- test) shows the statistical differences between two topographic PRPs, as shown in the third row of Fig 3B. Stronger responses in natural PRP than in vocoded PRP are shown in the PLOS ONE \| https://doi.org/10.1371/journal.pone.0287584 June 23, 2023 5 / 12 PLOS ONEEffects of degraded speech stimuli on phoneme-related potentials (PRPs) Fig 3. (A) Grand-averaged phoneme-related potential (PRP) computed using the mean PRP of all participants, all phonemes, and all channels. The grand- averaged PRPs in response to natural storybook (red) and vocoded storybook (blue) are plotted in each panel. The PRPs from each subject are plotted in gray. (B) Average topographic map of z-score across participants in the intervals R1 (50–80 ms), R2 (110–140 ms), R3 (170–260 ms), R4 (350–410 ms), and R5 (450– 470 ms) in response to natural storybook (first row) and vocoded storybook (second row). The p-values of the t-test (third row) are colored in red (< 0.01) to yellow (< 0.05). https://doi.org/10.1371/journal.pone.0287584.g003 frontocentral area at early R1 (positive) and R2 (negative), whereas the vocoded PRP was stronger in the late R4 interval. Fig 4 describes the effect of speech degradation on amplitude and latency of the PRP com- ponent. In Fig 4, each dot represents the z-score and latency of the grand-averaged PRP across all phoneme stimuli and 64 channels in each subject. A paired t-test was conducted to compare natural and vocoded cases. The statistical analysis indicates that the P1, N1, and P2 peaks are significantly larger in natural PRP than in vocoded PRP (*: p < 0.001, paired t-test) as seen in Fig 4A. Although prolonged mean latencies were observed in the P1, N1, and P2 peaks in vocoded PRP compared with those in natural PRP, a significantly longer latency was only observed in the N1 peak, as shown in the lower panel in Fig 4B (: p < 0.001, paired t-test) and Table 2. 3.2. Comparison of individual phoneme-related potentials Fig 5 shows the natural and vocoded PRPs for each 10 vowels and 12 consonants. Each PRP at FCz was averaged across the participants. The color bar at each PRP component provides intu- itive z-score information about the positive (red) and negative (blue) amplitudes. The manner of articulation clearly affects the amplitude and latency of the PRP waveform. There is a signifi- cant decrease in the amplitude of vocoded PRP compared with that of natural PRP. The comparisons of the natural and vocoded PRPs in response to vowels and plosive-, frica- tive-, and nasal-consonants are shown in Fig 6. Each panel shows plots of the averaged PRPs at FCz across each phoneme group of 10 vowels, 3 plosives, 4 fricatives, and 5 nasals. Although the incidence of each phoneme varied, as shown in Table 1, the averaged PRP for each pho- neme stimulation could be calculated sufficiently. The averaged PRPs from the subjects were employed to investigate the statistical differences depending on manner and place of PLOS ONE \| https://doi.org/10.1371/journal.pone.0287584 June 23, 2023 6 / 12 PLOS ONEEffects of degraded speech stimuli on phoneme-related potentials (PRPs) Fig 4. Comparison of the P1-N1-P2 complex of phoneme-related potentials (PRPs) in response to natural and vocoded storybooks. (A) The z-score of the amplitude of each peak P1, N1, and P2 of the PRPs evoked by vocoded and natural storybooks. The gray-circle denotes individual (each subject) data. (B) Peak latency of P1, N1, and P2 of PRP. The latencies were measured after each phoneme onset. The statistical analysis yielded the significant difference (**: p < 0.001) between the vocoded and natural cases. https://doi.org/10.1371/journal.pone.0287584.g004 articulation. The gray-bar in each panel represents the statistical difference over time between the averaged PRPs of the natural and vocoded cases (p < 0.01, paired t-test). It was noted that significant differences were observed in the R1 (50–80 ms), R2 (110–140 ms), R3 (170–260 ms), and R4 (350–410 ms) intervals of the grand-averaged PRP following both natural and vocoded storybook stimulation, as illustrated in Fig 3. The results were consistent across all four cases examined in this section, categorized by phonetic groups based on articulation. Sig- nificant differences were found between natural and vocoded PRPs in peak amplitudes during early R1 (50–80 ms) and R2 (110–140 ms) intervals, with higher amplitudes in natural PRPs. Conversely, during late R3 (170–260 ms) and R4 (350–410 ms) intervals, higher amplitudes were observed in vocoded PRPs for nasal, plosive, and vowel sounds. The late latency compo- nent of the vocoded PRPs was observed for the phonemes of vowels, nasals, and plosives. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287584 June 23, 2023 7 / 12 PLOS ONEEffects of degraded speech stimuli on phoneme-related potentials (PRPs) Table 2. Summary of PRP peak amplitude, latency, and statistical comparison between natural and vocoded cases. P1 N1 P2 P1 N1 P2 z-score (amplitude) natural vocoded natural vocoded natural vocoded Latency (ms) natural vocoded natural vocoded natural vocoded Mean 0.035 0.016 -0.022 -0.043 0.022 0.002 Mean 68.6 71.6 127.0 146.9 209.2 211.1 SD 0.014 0.011 0.010 0.017 0.016 0.009 SD 17.7 17.9 17.0 18.8 18.9 23.5 p-value p < 0.001 p < 0.001 p < 0.001 p-value - p < 0.001 - SD: standard deviation; p-value: between the paired (natural and vocoded cases in a subject) samples https://doi.org/10.1371/journal.pone.0287584.t002 4. Discussion In this study, we evaluated the effect of speech degradation on PRP in response to a continuous storybook. The latency and amplitude of PRP in natural continuous storybooks were signifi- cantly shorter and higher, respectively, than those in vocoded storybooks. 4.1. Similarity to late AEP P1-N1-P2 The latter AEPs are generally known to be involved in stimulus recognition and information pro- cessing. PRP has the latter components of P1-N1-P2 that are similar to those of general AEP that occur during central auditory processing (see Fig 4). P1 (first positive peak) of PRP was observed between 40 ms and 75 ms, N1 (first negative peak) between 90 ms and 200 ms, and P2 (second positive peak) between 100 ms and 250 ms after stimulus onset. It has been reported that the amplitude and latency of P1-N1-P2 are useful to objectively evaluate auditory functions [18, 19]. The current study revealed that the amplitude and latency of the PRP peak varied depending on speech intelligibility, which is consistent with the findings of a previous AEP study [20]. 4.2. Phoneme related potential according to phoneme class Phonemes are generally classified according to the place of articulation during their utterances [21]. Kova´cs et al. (2017) reported that different AEPs occurred in response to syllabic non- sense words based on the phoneme classes (fricative, plosive, nasal, affricate, and liquid) [22]. It has been demonstrated that the human auditory system is effectively sensitive to sudden changes in spectrotemporal information [23]. The results of this study highlight the consis- tency in the effect of speech-stimulus degradation in each phoneme class on PRP, as seen in Figs 5 and 6. However, the similarity of PRP waveforms within each phoneme class was not observed. It is necessary to evaluate the similarity of waveforms classified based on stimulus feature such as fundamental frequency rather than utterances. 4.3. Clinical relevance for the prediction of speech intelligibility Behavioral speech tests have been used to evaluate speech recognition in clinical settings. An objective approach using continuous speech-evoked EEG has also been recently proposed [3, PLOS ONE \| https://doi.org/10.1371/journal.pone.0287584 June 23, 2023 8 / 12 PLOS ONEEffects of degraded speech stimuli on phoneme-related potentials (PRPs) Fig 5. Comparison of phoneme-related potential (PRP) waveforms in response to natural and vocoded storybooks. The PRPs were plotted in articulate (vowel, plosive, fricative, and nasal) and alphabetical order. The color in each PRP indicates positive (red) and negative (blue) z-scores. https://doi.org/10.1371/journal.pone.0287584.g005 24]. The repetitive presentation of words or short sentences to compute the synchronized AEP deteriorates the task performance efficiency [6, 25]. Because running speech stimuli with a story, which is not repetitive, can arouse the subject’s interest in maintaining attention, PRP has the advantage of evaluating speech intelligibility compared with conventional AEP tests as well as understanding phonemic awareness. 4.4. Limitations on phoneme related potential The phonemic awareness, which represents the ability of separating and identifying individual phonemes in spoken words, can be assessed by behavioral tasks or objective ERPs [26, 27] This study showed that PRP can be a valuable tool to test the ability of hearing individual sounds in words. Coarticulation sensitivity, the ability to perceive the overlap phonemes, also helps indi- viduals to better understand spoken language, even in challenging listening conditions. Several studies have used the mismatch negative component of ERP to allow the processing of vowel- consonant or vowel-vowel coarticulation [28, 29]. While the use of PRP has provided insights into the phonemic processing, limitations exist in our understanding of the underlying coarti- culation. Therefore, further research is needed to fully explore and comprehend the cognitive and neural processes involved in perceiving individual phonemes. 5. Conclusions This study demonstrated that significant differences occurred in the grand-averaged PRP as well as the PRP in response to each phoneme between the natural and vocoded cases. These PLOS ONE \| https://doi.org/10.1371/journal.pone.0287584 June 23, 2023 9 / 12 PLOS ONEEffects of degraded speech stimuli on phoneme-related potentials (PRPs) Fig 6. Grand-averaged phoneme-related potentials in an articulate group (nasal, fricative, plosive, and vowel) in response to natural (red) and vocoded (blue) storybooks. The transparent area indicates the range of individual data. The gray bars denote the statistically (p < 0.01, t-test) different intervals. https://doi.org/10.1371/journal.pone.0287584.g006 findings indicate that PRP can be used as an objective measure to evaluate speech intelligibility in clinical settings. However, some issues need to be addressed prior to clinical implementa- tion. The EEG data were acquired from only normal hearing subjects using natural and vocoded story stimuli. Therefore, additional validations for subjects with hearing impairment should be performed. Moreover, as the phonemes in the storybook used in this study were asymmetrically distributed, typical phonemes were limited to achieve reliable PRPs with a high signal-to-noise ratio. A storybook that thoroughly covers a sufficient number of phonemes should be further examined. Author Contributions Conceptualization: Jihwan Woo. Data curation: Min-Jae Jeon. Investigation: Min-Jae Jeon, Jihwan Woo. Methodology: Min-Jae Jeon, Jihwan Woo. Writing – original draft: Min-Jae Jeon. Writing – review & editing: Jihwan Woo. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287584 June 23, 2023 10 / 12 PLOS ONEEffects of degraded speech stimuli on phoneme-related potentials (PRPs) References 1. Miller N. Measuring up to speech intelligibility. International Journal of Language & Communication Dis- orders. 2013; 48(6):601–12. https://doi.org/10.1111/1460-6984.12061 PMID: 24119170 2. Lee J. Standardization of Korean speech audiometry. Audiology and Speech Research. 2016; 12:S7– S9. 3. Vanthornhout J, Decruy L, Wouters J, Simon JZ, Francart T. Speech intelligibility predicted from neural entrainment of the speech envelope. Journal of the Association for Research in Otolaryngology. 2018; 19(2):181–91. https://doi.org/10.1007/s10162-018-0654-z PMID: 29464412 4. He S, Grose JH, Teagle HF, Woodard J, Park LR, Hatch DR, et al. Gap detection measured with electri- cally-evoked auditory event-related potentials and speech perception abilities in children with auditory neuropathy spectrum disorder. Ear and hearing. 2013; 34(6). https://doi.org/10.1097/AUD. 0b013e3182944bb5 PMID: 23722354 5. Korczak PA, Kurtzberg D, Stapells DR. Effects of sensorineural hearing loss and personal hearing aids on cortical event-related potential and behavioral measures of speech-sound processing. Ear and hear- ing. 2005; 26(2):165–85. https://doi.org/10.1097/00003446-200504000-00005 PMID: 15809543 6. Iotzov I, Parra LC. EEG can predict speech intelligibility. Journal of Neural Engineering. 2019; 16 (3):036008. https://doi.org/10.1088/1741-2552/ab07fe PMID: 30776785 7. Ding N, Simon JZ. Cortical entrainment to continuous speech: functional roles and interpretations. Fron- tiers in human neuroscience. 2014; 8:311. https://doi.org/10.3389/fnhum.2014.00311 PMID: 24904354 8. 9. Fiedler L, Wo¨stmann M, Herbst SK, Obleser J. Late cortical tracking of ignored speech facilitates neural selectivity in acoustically challenging conditions. Neuroimage. 2019; 186:33–42. https://doi.org/10. 1016/j.neuroimage.2018.10.057 PMID: 30367953 Fuglsang SA, Dau T, Hjortkjær J. Noise-robust cortical tracking of attended speech in real-world acous- tic scenes. Neuroimage. 2017; 156:435–44. https://doi.org/10.1016/j.neuroimage.2017.04.026 PMID: 28412441 10. Schmitt N, Winkler A, Boretzki M, Holube I. A phoneme perception test method for high-frequency hear- ing aid fitting. Journal of the American Academy of Audiology. 2016; 27(05):367–79. https://doi.org/10. 3766/jaaa.15037 PMID: 27179256 11. Zeng F-G, Galvin JJ III. Amplitude mapping and phoneme recognition in cochlear implant listeners. Ear and hearing. 1999; 20(1):60–74. https://doi.org/10.1097/00003446-199902000-00006 PMID: 10037066 12. Khalighinejad B, da Silva GC, Mesgarani N. Dynamic encoding of acoustic features in neural responses to continuous speech. Journal of Neuroscience. 2017; 37(8):2176–85. https://doi.org/10.1523/ JNEUROSCI.2383-16.2017 PMID: 28119400 13. Picton TW, Stuss DT, Champagne SC, Nelson RF. The effects of age on human event-related poten- tials. Psychophysiology. 1984; 21(3):312–26. https://doi.org/10.1111/j.1469-8986.1984.tb02941.x PMID: 6739673 14. van Schijndel NH, Houtgast T, Festen JM. Effects of degradation of intensity, time, or frequency content on speech intelligibility for normal-hearing and hearing-impaired listeners. The Journal of the Acoustical Society of America. 2001; 110(1):529–42. https://doi.org/10.1121/1.1378345 PMID: 11508977 15. Delorme A, Makeig S. EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. Journal of neuroscience methods. 2004; 134(1):9–21. https://doi.org/10.1016/j.jneumeth.2003.10.009 PMID: 15102499 16. Boersma P. Praat, a system for doing phonetics by computer. Glot Int. 2001; 5(9):341–5. 17. Yoon T-J, Kang Y. Monophthong analysis on a large-scale speech corpus of read-style Korean. Phonet- ics and Speech Sciences. 2014; 6(3):139–45. 18. Wilkinson RT, Morlock HC. Auditory evoked response and reaction time. Electroencephalography and Clinical Neurophysiology. 1967; 23(1):50–6. https://doi.org/10.1016/0013-4694(67)90140-x PMID: 4165560 19. Boston JR. Spectra of auditory brainstem responses and spontaneous EEG. IEEE Transactions on Bio- medical Engineering. 1981;(4):334–41. https://doi.org/10.1109/TBME.1981.324801 PMID: 7287033 20. Finke M, Bu¨ chner A, Ruigendijk E, Meyer M, Sandmann P. On the relationship between auditory cogni- tion and speech intelligibility in cochlear implant users: An ERP study. Neuropsychologia. 2016; 87:169–81. https://doi.org/10.1016/j.neuropsychologia.2016.05.019 PMID: 27212057 21. Vogt H. Phoneme classes and phoneme classification. Word. 1954; 10(1):28–34. 22. Kova´ cs A, Winkler I, Vicsi K, editors. EEG correlates of speech: Examination of event related potentials elicited by phoneme classes. 2017 8th IEEE International Conference on Cognitive Infocommunications (CogInfoCom); 2017: IEEE. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287584 June 23, 2023 11 / 12 PLOS ONEEffects of degraded speech stimuli on phoneme-related potentials (PRPs) 23. Babel M, Bulatov D. The role of fundamental frequency in phonetic accommodation. Language and speech. 2012; 55(2):231–48. https://doi.org/10.1177/0023830911417695 PMID: 22783633 24. Nogueira W, Dolhopiatenko H. Predicting speech intelligibility from a selective attention decoding para- digm in cochlear implant users. Journal of Neural Engineering. 2022; 19(2):026037. https://doi.org/10. 1088/1741-2552/ac599f PMID: 35234663 25. Boksem MA, Meijman TF, Lorist MM. Effects of mental fatigue on attention: an ERP study. Cognitive brain research. 2005; 25(1):107–16. https://doi.org/10.1016/j.cogbrainres.2005.04.011 PMID: 15913965 26. Yopp HK. The validity and reliability of phonemic awareness tests. Reading research quarterly. 1988:159–77. 27. Harwood V, Kleinman D, Puggioni G, Baron A. The P300 event related potential predicts phonological working memory skills in school-aged children. Frontiers in Psychology. 2022; 13. https://doi.org/10. 3389/fpsyg.2022.918046 PMID: 36312112 28. Grosvald M, Corina D. Perception of long-distance coarticulation: An event-related potential and behav- ioral study. Applied Psycholinguistics. 2012; 33(1):55–82. 29. Na¨a¨ta¨nen R, Paavilainen P, Rinne T, Alho K. The mismatch negativity (MMN) in basic research of cen- tral auditory processing: a review. Clinical neurophysiology. 2007; 118(12):2544–90. https://doi.org/10. 1016/j.clinph.2007.04.026 PMID: 17931964 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287584 June 23, 2023 12 / 12 PLOS ONE
10.1371_journal.pone.0287750	RESEARCH ARTICLE A procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed Lin WangID 2 1, Qiang Zu1, Qiang ZhangID 1 State Key Laboratory of Eco-hydraulics in Northwest Arid Region of China, Xi’an University of Technology, Xi’an, China, 2 Research Institute of Geotechnical Engineering, China Institute of Water Resources and Hydropower Research, Beijing, China a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 linwang@xaut.edu.cn Abstract OPEN ACCESS Citation: Wang L, Zu Q, Zhang Q (2023) A procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed. PLoS ONE 18(6): e0287750. https://doi.org/ 10.1371/journal.pone.0287750 Editor: Pramod K. Pandey, University of California Davis, UNITED STATES Received: June 1, 2022 Accepted: June 13, 2023 Published: June 27, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0287750 Copyright: © 2023 Wang et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files. the GitHub in .csv format, and its URL is: https://github.com/XAUT-WangLin/ Summary-table-of-evaluation-index. Flood-based hydrodynamic damage to check dam systems on the Loess Plateau of China occurs frequently, and there is a strong desire to carry out risk assessments of such check dam systems. This study proposes a weighting method that combines the analytic hierarchy process, entropy method, and TOPSIS to assess the risk of check dam systems. The com- bined weight-TOPSIS model avoids weight calculation only considers the influence of sub- jective or objective preference and the bias of the single weighting method. The proposed method is capable of multi-objective risk ranking. It is applied to the Wangmaogou check dam system located in a small watershed on the Loess Plateau. The result of risk ranking matches the reality of the situation. The gray correlation theory model is utilized to rank the risks in the same research area and compared with the results of the combined weight-TOP- SIS model. The combined weight-TOPSIS model is more favorable to risk assessment than the gray correlation theory model. The resolution level and decisive judgment of the com- bined weight-TOPSIS model are more advantageous. These results are in line with the actual conditions. It proves that the combined weight-TOPSIS model can provide a technical reference for the risk assessment of check dam systems in small watersheds. Introduction The Chinese Loess Plateau is located in the middle and upper reaches of the Yellow River. This area is known for its severe soil erosion and high sediment yield [1]. Check dams are an effec- tive long-term engineering measure created by residents in the northwest to control soil ero- sion on the Loess Plateau [2]. As of the end of 2020, there were 58,776 check dams on the Plateau [3]. Check dams have provided many ecological benefits with regard to silting up land, reducing debris flows, blocking mud and improving the ecological environment [4–10], as shown in Fig 1. The basic structure of the check dam body has a similar fundamental design to the earth- rock dam. The height of the dam is up to 30 meters. Most check dams use shaft or culvert drainage, and flood discharge capacity is insufficient [11]. Due to the lack of spillways and PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 1 / 20 PLOS ONEFunding: The research work is supported by the National Natural Sciences Foundation of China Fund Project (Grant No.51909214), National Key R&D Program Project (2017YFC1501103) provided us with all the funds for the site investigation of the check dam. Shaanxi Water Conservancy Science and Technology Plan Project (2021slkj-9) participated in the data collection and analysis of this study. Competing interests: The authors have declared that no competing interests exist. A procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed limited flood control standards, there is a large gap between earth-rock dams and check dams [12]. Many check dams are full of silt or aging in disrepair. In recent years, flood-based hydrodynamic damage caused by frequent rainstorms on the Loess Plateau has often caused check dam to breach. Given the unique material composition and geological conditions, once a flood occurs, it will cause unpredictable disasters to the lives and properties of the local people [13–15]. Continuous heavy rainfall in the Yanhe River Basin in Shaanxi induced damage to 516 check dams in July 2013 [16]. In August 2016, an extremely heavy rainfall event occurred in Dalate Banner, Inner Mongolia, and 19 check dams were bro- ken [17]. In July 2017, the disastrous flood on the Wuding River in Shaanxi destroyed 337 check dams in Suide County (Fig 2) [18]. The check dam system is an effective method to control soil erosion and prevent floods on the Loess Plateau [19–22], but once the check dam breaks, it will cause a huge loss of life and property to local residents [23]. A comprehensive risk assessment should be carried out to real- ize the risk ranking of the check dam system in small basins, which can reduce to the greatest extent the risk caused by the burst of the check dam [24]. The analytic hierarchy process (AHP) is an index weight calculation method that considers subjective factors. It has been widely used in check dams [25], earth-rock dam safety evaluation [26], flood assessment [27], dam site selection [28], land restoration [29], etc. The entropy method is used to calculate the index weight that considers objective factors. It has many appli- cations in the fields of dam risk assessment [30], land use [31], environmental science [32], and other fields. By combining the analytic hierarchy process and entropy method to calculate the combined weight, studies can avoid the bias of a single weighting method. After calculating the combined weight of the evaluation index, the risk of the dam system still needs to be evaluated. Fig 1. Check dams with ecological benefits. https://doi.org/10.1371/journal.pone.0287750.g001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 2 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed Fig 2. Breach of check dams in Suide County. https://doi.org/10.1371/journal.pone.0287750.g002 TOPSIS (a technique for order preference by similarity to the ideal solution), developed by Hwang and Yoon in 1981 [33], is a simple sequencing method in conception and application. This method can analyze principles quickly and perform rapid calculations, and the sample demand is not large. In recent years, it has been used in many fields, such as the evaluation of agricultural water-saving development strategies [34], sustainable building safety analyses [35], and site selection of fixed refuge places in cities and towns [36]. The essence of risk evaluation is to carry out a comprehensive and objective risk ranking under multi-objective situation. This study adopts the subjective and objective combination weight empowerment method combining hierarchical analysis method and entropy weight method was put forward, which aims to avoid the shortcomings of considering only the sub- jective or the objective factors unilaterally and improve the accuracy of weight determination; The comprehensive risk evaluation model of the check dam system in the small watershed was constructed based on the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) method, which can realize the risk ranking of check dam system under multiple objectives. It includes three first-level indices of flood, operational and economic risks and ten second-level indices. The combination weight index is established by using the analytic hierar- chy process and entropy weight method to obtain the combination weight of the evaluation index, and then combined with the TOPSIS method to carry out the comprehensive risk assessment with the Wangmaogou watershed check dam system as an example. The study is expected to provide a technical reference for the safety prevention and control of check dam systems in small watersheds on the Loess Plateau. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 3 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed Theory and methodology 1.1 Technical route of the risk assessment model The analytic hierarchy process and the entropy method are used to establish combined weight indicators with the TOPSIS method [37,38] to carry out risk assessments of check dam systems in small watersheds (Fig 3). 1.2 Evaluation index weight 1.2.1 Analytic hierarchy process. The analytic hierarchy process is a method of “mea- surement through pairwise comparisons and relies on the judgments of experts to derive pri- ority scales”. Therefore, it is widely used in multiple criteria decision-making tools considering subject factors, such as evaluating the risk of a check dam system in this study [39,40]. The aim of the analytic hierarchy process method is to obtain the weight vector ωsj, which not only considers subject risk evaluation but also satisfies the demand of the consistency check. This subjective weight vector is used to the calculate combined weight ωj. The analytic hierarchy process method is specified in 4 steps as follows. Step 1: Establish a hierarchical structure. The hierarchical structure is shown in Fig 4 as the risk evaluation index system of the check dam system. Step 2: Construct the judgment matrix Aij of each level. Fig 3. Technology road map of the combined weight-TOPSIS model. https://doi.org/10.1371/journal.pone.0287750.g003 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 4 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed Fig 4. Risk evaluation index system of check dam systems. https://doi.org/10.1371/journal.pone.0287750.g004 The judgment matrix is composed of the comparison value of the mutual importance of each evaluation index at the same level. Aij ¼ 2 6 6 6 6 4 a11 a21 � � � an1 a12 a22 � � � an2 � � � � � � � � � � � � 3 7 7 7 7 5 a1n a2n � � � amn ð1Þ where aij = the scale value, which is expressed by numbers 1–9 and its reciprocal; we give the principle of determining the scale in Table 1. Step 3: Consistency check. The consistency index CI is calculated as follows: CI ¼ lmax (cid:0) n n (cid:0) 1 ð2Þ Table 1. Determining principles of scales. Scales Meaning 1 3 5 7 9 2, 4, 6, 8 Index i is as important as index j Index i and index j are slightly more important Index i and index j are obviously important Index i and index j are strongly important Index i and index j are extremely important Indicates the intermediate value of the abovementioned adjacent importance Note: When the importance scale of index i over index j is aij, the importance scale of index j over index i is 1/aij. https://doi.org/10.1371/journal.pone.0287750.t001 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 5 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed Table 2. Random consistency index. n RI 1 0 2 0 3 0.58 4 0.90 5 1.12 6 1.24 7 1.32 8 1.41 9 1.45 10 1.49 The consistency ratio CR is calculated as follows: https://doi.org/10.1371/journal.pone.0287750.t002 where λmax = the largest characteristic root of the matrix Aij, and n is the order of the judgment matrix Aij. The corresponding random consistency index RI is determined (Table 2). CR ¼ CI RI ð3Þ If CR<0.1, the consistency of the judgment matrix can be considered acceptable. Step 4: The eigenvector of the judgment matrix is normalized to be the desired weight vec- tor ωsj. 1.2.2 Entropy method. The entropy method can be used to derive criteria weights objec- tively from pertinent decision data in case preferential judgments are either partial or unavail- able [41]. Using entropy, the weight assigned to a decision criterion is directly related to the average intrinsic information generated by a given set of alternative evaluations at that crite- rion, as well as to its subjective assessment. Therefore, the entropy method is capable of evalu- ating the objective weight as a supplement to the analytic hierarchy process method. The aim of the entropy method is to obtain the weight vector ωgj, which is similar to ωsj and is also used in calculating of the combined weight ωj. The use of the entropy method is specified in 4 steps as follows. Step 1: Suppose there are a total of m dam systems, each dam system has n evaluation indi- ces, and a judgment matrix R is constructed. R ¼ ðrijÞmnði ¼ 1; 2; . . .; m; j ¼ 1; 2; . . .; nÞ ð4Þ where rij = the value of the jth index of the ith dam system. Step 2: Normalize the judgment matrix to obtain the normalized judgment matrix D = (dij)mn. Step 3: For the case of m dam systems and n indicators, the entropy Sj of the evaluation indi- cators can be determined as: Xm i¼1 fijlnfij lnm Sj ¼ (cid:0) ði ¼ 1; 2; . . .; m; j ¼ 1; 2; . . .; nÞ ð5Þ where fij ¼ ð1 þ dijÞ= ð1 þ dijÞ. Xm Step 4: The entropy weight vector W of the evaluation index is: i¼1 W ¼ ðogjÞ1�n ¼ 0 B B B B @ 1 (cid:0) Sj Xn n (cid:0) 1 C C C C A Sj j¼1 1�n where ωgj = the entropy weight of the jth evaluation index. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 ð6Þ 6 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed 1.2.3 Combination weight. The subjective weighting ωsj and the objective weighting ωgj are integrated to obtain the combined weight ωj: oj ¼ ðosj � ogjÞ0:5 Xm j¼1 ðosj � ogjÞ0:5 ðj ¼ 1; 2; � � � ; mÞ ð7Þ 1.3 TOPSIS evaluation method The evaluation steps are divided into the following five steps: Step 1: Standardization of the indicator data. For the case of m dam systems and n evaluation indicators, the initial evaluation matrix X = (xij)mn can be obtained by referring to Formula (1). The data in X are standardized. When a lager positive index is better, the positive index Pij calculation formula is as follows: Pij ¼ xij (cid:0) minðxijÞ maxðxijÞ (cid:0) minðxijÞ ð8Þ When a smaller negative index is better, the positive index Pij calculation formula is as fol- lows: Pij ¼ maxðxijÞ (cid:0) xij maxðxijÞ (cid:0) minðxijÞ ð9Þ The final standardized matrix is P = [pij]m×n. Step 2: Establish a weighted decision evaluation matrix. The weighted normalized matrix V after considering the weight of each evaluation index is: V ¼ P � W ¼ ½vij�m�n ð10Þ where W = the evaluation index weight vector. This study adopts subjective and objective com- bination weights, as calculated by Formula (7). Step 3: Determine the positive and negative ideal solutions V+ and V-. V þ ¼ fmaxðvijÞji ¼ 1; 2; . . .; mg ¼ fvþ 1 ; vþ 2 ; . . .; vþ n g V (cid:0) ¼ fminðvijÞji ¼ 1; 2; . . . ; mg ¼ fv1 (cid:0) ; v2 (cid:0) ; . . . ; vn (cid:0) g ð11Þ ð12Þ Step 4: Calculate the distance. The Euclidean distances from the evaluation index to the positive and negative ideal solu- tions V+ and V- are Di -. + and Di v u u t v u u t Dþ i ¼ D(cid:0) i ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Xn ðvþ j (cid:0) vijÞ2 j¼1 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Xn ðvij (cid:0) v(cid:0) j Þ2 j¼1 ði ¼ 1; 2; . . .; mÞ ði ¼ 1; 2; . . .; mÞ ð13Þ ð14Þ where vj + = the positive ideal point of evaluation index j and vj - = the negative ideal point of PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 7 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed evaluation index j. The closer the evaluation indicator is to the positive ideal point, the better the indicator and the lower the risk will be. Step 5: Calculation and sorting of relative closeness. The relative closeness Ti is used to indicate the degree of the evaluation value and the opti- mal value. The larger the value is, the closer it is to the optimal value, i.e., the lower the risk. The calculation formula is as follows: Ti ¼ D(cid:0) i i þ D(cid:0) i Dþ ð i ¼ 1; 2; . . .; m Þ ð15Þ The closeness of each dam system is calculated and sorted to obtain the relative risk of each dam system. Risk evaluation index system 2.1 Construction of the risk evaluation index system for check dam systems First, the risk evaluation index system of check dam systems can be established to evaluate the system risk. The establishment of this system needs to identify the risk factors for check dams. Through the investigation of a large number of water-damaged check dams and dam systems, the common causes of water damage to check dams can be determined [42–44]. These factors include the following: 1. Dams that are suffering from excessive rainstorms and floods exhibit greater risk. 2. The effective storage capacity is full, and the flood detention capacity is insufficient. 3. Unmatched drainage engineering facilities lead to poor drainage. 4. The dams are of poor construction quality. 5. There is a lack of control over backbone projects in the watershed, the dam system layout is unreasonable, and chain dam failures are likely to occur. 6. Later management and protection are relatively weak. Based on the historical water damage, considering that the check dam provides the special function of flood detention and silt retention and silt land reclamation, its water damage will cause submergence losses to downstream facilities and economic crops. The risk assessment of the check dam system is carried out from the three perspectives of external man-made man- agement strategies, protection measures and economic losses. Based on the established system of predecessors [45–47], three first-level indicators that affect the safety of the check dam sys- tem (flood disaster, operation, economy) and a risk evaluation index system with 10 secondary indicators are established (Fig 4 & Table 3). 2.2 Valuation method of the risk evaluation index for check dam systems Based on the survey data, 10 indicators in the risk evaluation index system are quantitatively assigned [48]. To eliminate the dimensional influence, the evaluation index matrix is standard- ized after the assignment. To reasonably reflect the degree of each evaluation index influence on the risk of the dam system, it is necessary to calculate each index weight vector [48]. The weighted evaluation matrix is the decision evaluation matrix. In the flood risk layer, evaluation index C3 is assigned to the water release facility. The value is 1 when the main dam has a com- plete spillway, 0.9 when the spillway has a small amount of damage, 0.5 when horizontal pipes and vertical wells are used for water discharge, and 0.1 when the discharge port is blocked and PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 8 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed Table 3. Connotation of risk evaluation indices for check dam systems. First-level index Secondary indicators Indicator meaning Flood risk (B1) Detention flood (C1) The runoff of the rainstorm within a certain return period within the control area of the check dam minus the amount of discharge from the discharge structure Remaining storage capacity (C2) The capacity of the check dam to withstand floods, the larger the remaining storage capacity, the stronger the ability to withstand floods Drain facility (C3) Including the discharge capacity of vertical shafts, horizontal pipes, and spillways Dam integrity (C4) The extent of damage to the check dam body Dam system layout coefficient (C5) The rationality of the dam system layout, the more reasonable the layout, the larger the coefficient Daily management (C6) Whether there is a corresponding department to manage and maintain the check dam Emergency response (C7) Whether there are emergency measures for accidents Monitoring facility (C8) Whether there are correspondingly complete and normal monitoring facilities Downstream loss risk (C9) Whether there are roads, houses, factories, etc. downstream to determine the indicators of downstream economic risks Crop security risk (C10) Allowable submerged water depth of dam system crops is higher than the depth of stagnant flood Operational risk (B2) Economic risk (B3) Index nature Negative index Positive index Positive index Positive index Positive index Positive index Positive index Positive index Negative index Negative index https://doi.org/10.1371/journal.pone.0287750.t003 cannot be discharged. An evaluation index of integrity degree C4 is assigned to the dam body, and the value is 1 when the main dam body is intact, 0.8 when there are real cracks, and 0.2 when there are small caves. The evaluation index dam system layout coefficient C5 reflects the rationality of the dam system layout, and the value ranges from 0 to 1. A value of more than 0.65 indicates a reasonable layout, a value less than 0.4 indicates an unreasonable layout, and values in the middle indicate a reasonable layout. In the operation risk layer, the evaluation index daily management risk C6 is assigned. When the relevant department maintains the check dam, the value is 1; otherwise, it is 0.1. The evaluation index emergency risk C7 is assigned. When there is an accident, the value is 1 for emergency measures; otherwise, it is 0. A value is assigned to the evaluation index monitoring risk C8. When there are complete and normal monitoring facilities, the value is 1; otherwise, it is 0.1. In the economic risk layer, the evaluation index downstream loss risk of C9 is assigned. When there are important residential buildings downstream, the backbone dam system unit is set to 1, and the branch dam system unit is set to 0.8. The evaluation index crop yield risk C10 is assigned, and its risk value is related to the flood return period corresponding to the stagnant flood depth. Risk assessment of check dam systems—a case study of Wangmaogou watershed 3.1 Study area Wangmaogou is located in Suide County, Yulin City, Shaanxi Province. It is a secondary branch of Jiuyuangou located on the left bank of the Wuding River. The geographical position is 940~1188 m east longitude, the drainage area is 5.97 km2, and the main ditch length is 3.75 km. The slope is generally above 20˚. The amount of rainfall in the basin is small, and it is unevenly distributed. The average annual rainfall is 513 mm, and the rainfall in the flood PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 9 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed Fig 5. Location of the area (a, b); the altitude of the Wangmaogou watershed (c, Drawn by ESRI’s ArcGIS); the layout of check dam types (d). https://doi.org/10.1371/journal.pone.0287750.g005 season accounts for more than 70% of the total annual rainfall [49,50]. In 2012 and 2017, flood-based hydrodynamic damage to the check dam system caused by frequent rainstorms occurred; thus, risk analysis is urgently needed. According to the data from the Suide Soil and Water Conservation Scientific Experiment Station of the Yellow River Conservancy Commis- sion, there are 22 check dams in the Wangmaogou watershed, including 2 backbone dams, 8 medium dams, and 12 small dams [51]. The check dam that breached after flood-based hydro- dynamic damage in the Wangmaogou watershed on July 15, 2012 was used as an example to carry out the research. After excluding the check dams that were breached and full before 2012, 19 check dams were selected for analysis, as shown in Fig 5. This study divides Wang- maogou check dam systems into the Guandigou unit, Wangtagou unit, Wangmaogou Unit 2, Nianyangou unit, Kanghegou unit and Huangbaigou unit. 3.2 Determination of the index weight The risk evaluation index approach of the check dam system is shown in Fig 5. Based on the 3 first-level and 10 second-level indicators, when the Wangmaogou watershed encountered a once-in-a-hundred-year heavy rainfall event, the data used in the risk evaluation index are shown in Table 4. Among the evaluation indicators, detention floods, downstream economic risks, and income-guaranteed risks are negative indicators, and the rest are positive indicators. The matrix P is obtained from the initial evaluation matrix standardized, which is based on the assignment results of the dam system layout coefficient, the downstream economic risk, and PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 10 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed Table 4. Summary table of evaluation index. Detention flood/104 m3 Remaining storage capacity/104 m3 Drain facility Dam integrity Daily management Emergency measures Monitor facility Guandigou Unit Wangtagou Unit Wangmaogou Unit 2 Nianyangou Unit Kanghegou Unit Huangbaigou Unit 6.57 4.26 5.53 4.93 2.28 1.48 43.29 5.23 64.38 32.76 9.67 3.85 https://doi.org/10.1371/journal.pone.0287750.t004 Spillway Tiny cave No Shaft Lying tube intact intact intact Blocked pipe Rills, cracks Shaft intact No No Yes No No No No No Yes No No No No No Yes No No No the income-guaranteed risk: 2 6 6 6 6 6 6 6 6 6 6 4 P ¼ 0:000 0:652 1:000 0:000 0:178 0:000 0:000 0:000 1:000 0:030 0:454 0:023 0:000 1:000 0:089 0:000 0:000 0:000 1:000 0:000 0:204 1:000 0:444 1:000 1:000 1:000 1:000 1:000 0:000 0:020 0:322 0:478 0:444 1:000 0:333 0:000 0:000 0:000 1:000 0:677 0:843 0:096 0:000 0:750 0:000 0:000 0:000 0:000 1:000 0:687 1:000 0:000 0:444 1:000 0:267 0:000 0:000 0:000 1:000 1:000 3 7 7 7 7 7 7 7 7 7 7 5 According to the principle of scale determination, a judgment matrix is constructed which meets the consistency requirement. Then, the eigenvector of the judgment matrix is standard- ized to obtain the subjective weight vector from the analytic hierarchy process. Third, the entropy weight of each evaluation index, i.e., the objective weight, is calculated by Formula (5) and Formula (6). Finally, the combined weight vector is obtained based on the Formula (7) from the subjective weight vector and the objective weight vector. The evaluation index weight vector is shown in Fig 6, which is calculated by the analytic hierarchy process, entropy weight method, and combined weight method. The combination weight value lies between the subjec- tive and objective weight values, indicating the effect of information integration. 3.3 Risk ranking The standardized evaluation matrix P is weighted to obtain the weighted decision evaluation, and matrix V is obtained as shown below. V ¼ P � W 2 0:000 0:057 0:092 0:000 0:012 0:000 0:000 0:000 0:101 0:004 6 6 6 6 6 6 6 6 6 6 4 ¼ 0:044 0:002 0:000 0:058 0:006 0:000 0:000 0:000 0:101 0:000 0:020 0:088 0:041 0:058 0:068 0:116 0:127 0:118 0:000 0:003 0:031 0:042 0:041 0:058 0:023 0:000 0:000 0:000 0:101 0:091 0:082 0:008 0:000 0:043 0:000 0:000 0:000 0:000 0:101 0:092 0:097 0:000 0:041 0:058 0:018 0:000 0:000 0:000 0:101 0:134 3 7 7 7 7 7 7 7 7 7 7 5 where W = the combination weight, and P = the standardized evaluation matrix. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 11 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed Fig 6. The weight of the check dam system risk evaluation index in the Wangmaogou watershed. https://doi.org/10.1371/journal.pone.0287750.g006 The positive and negative ideal solutions V+ and V- are calculated by Formula (11) and For- mula (12): V þ ¼ ½ 0:097 0:088 0:092 0:058 0:068 0:116 0:127 0:118 0:101 0:134 � V (cid:0) ¼ ½ 0 0 0 0 0 0 0 0 0 0 � The distances of D+ and D- from each dam system unit to the positive and negative ideal points are calculated by Formula (13) and Formula (14): Dþ ¼ ½ 0:278 0:290 0:190 0:238 0:255 0:238 � D(cid:0) ¼ ½ 0:149 0:125 0:248 0:164 0:166 0:208 � Based on the combined weight-TOPSIS model to calculate the relative closeness T of each dam system unit, the following can be obtained: T ¼ ½ 0:349 0:301 0:566 0:408 0:393 0:466 � Results and analysis 4.1 Results and analysis for risk assessment of check dam systems At present, there is a lack of comprehensive evaluation index systems and general evaluation methods for check dam systems in small watersheds. Due to the limited data, there are no PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 12 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed Fig 7. Euclidean distance and relative closeness degree of check dam system units in the Wangmaogou watershed. https://doi.org/10.1371/journal.pone.0287750.g007 direct data that can reflect the risk assessment of check dam systems. On-site investigation of the Wangmaogou dam system unit, it can reflect the risk level, and provide a risk assessment of check dam systems. The evaluation results are summarized as follows. (1) The higher the relative closeness is, the lower the risk. The risk ranking of Wangmaogou small watershed dam system units is conducted by using the combined-weight-TOPSIS risk assessment model. The result is Wangmaogou Unit 2<Huangbaigou Unit<Nianyangou Unit<KangHegou unit<Guandigou unit<Wangtagou unit (Fig 7). Fitting the relative closeness in Fig 7, the coefficient of determination R2 after fitting is 0.9439, which indicates that the distribution of evaluation values calculated by the combined weight-TOPSIS model is uniform and reasonable. (2) The Wangtagou unit has the greatest risk. The Wangtagou unit is located in a branch of the Wangmaogou watershed, and the risk is greater than that of the backbone dam Wangmao- gou No. 2 Dam. After conducting field surveys in the Wangmaogou small watershed, it is found that the dam heights of the Wangtagou No. 1 and No. 2 dams of the Wangtagou unit dam system are 9 m and 13 m, respectively. The widths are 3.6 m and 4 m, without any drainage facilities, and the drainage is not smooth. The distances between the bottom and the top are 0.1 m and 4.7 m, respectively. The remaining storage capacity is relatively small, and the risk is higher. Therefore, it is necessary to improve the drainage structure construc- tion of branch check dams. (3) The construction standard of Unit 2 in Wangmaogou, i.e., the backbone dam of the river basin, is the highest, and its risk is the lowest. According to the measured data from the Suide Water Conservation Station, during flood-based hydrodynamic damage caused by the heavy rainfall that hit the Wangmaogou watershed on July 15 in 2012, a total of 8 dams breached, and the Wangtagou No. 1 dam and No. 2 dam in the Wangtagou unit dam PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 13 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed Fig 8. The Breach of Wangmaogou No. 2 backbone dam. https://doi.org/10.1371/journal.pone.0287750.g008 system breached the width of Wangtagou No. 1 dam breach was 1 m, and the width of dam No. 2 was 11.2 m (Fig 8). The Wangmaogou No. 2 dam did not breach. The risk of the No. 2 dam is lower than that of the Wangtagou unit. According to the actual survey after the “7.15” rainstorm [51]. Neither of the two check dams in Unit 2 of Wangmaogou broke during the rainstorm. In the Huangbaigou Unit, the Huangbaigou #2 dam did not break, and the Huangbaigou #2 dam break is circular with a breach of 2m. There are three check dams in the Nianyangou Unit, and two of them have bro- ken. There are also three check dams in the Kanghegou Unit, and all of them have broken. Among them, only the #3 dam of Kanghegou has broken up to 7 m wide, and the breaking is relatively serious. The two check dams in Wangtagou Unit broke seriously, and the maximum breach reached 11.2 m wide.” Through the actual investigation after the rainstorm, the evalua- tion result "Wangmaogou Unit 2<Huangbaigou Unit<Nianyangou Unit<KangHegou uni- t<Guandigou unit<Wangtagou unit" of the siltation dam unit is verified to be correct. Above all, the risk ranking results are consistent with the actual situation. 4.2 Comparison with existing risk assessment models (1) The entropy, AHP and combined weight-TOPSIS model, are used to carry out a compara- tive analysis of check dam systems risk assessment in small watersheds. The ideal points’ relative closeness and the risk ranking results of each dam system unit are shown in Table 5. The Spearman rank correlation coefficient method is used to test the correlation degree [52]. The correlation coefficient between the combined weight-TOPSIS model and PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 14 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed Table 5. Closeness degree and risk ranking result. Dam system unit name Combination weight-TOPSIS AHP-TOPSIS Entropy Method-TOPSIS Closeness Risk ranking Closeness Risk ranking Closeness Risk ranking Guandigou Unit Wangtagou Unit Wangmaogou Unit 2 Nianyangou Unit Kanghegou Unit Huangbaigou Unit 0.349 0.301 0.566 0.408 0.393 0.466 https://doi.org/10.1371/journal.pone.0287750.t005 2 1 6 4 3 5 0.396 0.360 0.493 0.462 0.461 0.530 2 1 5 4 3 6 0.315 0.262 0.632 0.337 0.370 0.412 2 1 6 3 4 5 the subjective weight-TOPSIS model ranking result is 0.943. The correlation coefficient between the combined weight-TOPSIS model and the objective weight-TOPSIS model ranking result is 1.000. The correlation is extremely high, and the evaluation results are highly consistent, indicating that the combined weight-TOPSIS model achieves subjective and objective data information. Considering the subjective and objective weights, the corre- lation coefficient of the TOPSIS model ranking results is 0.943, indicating the difference between subjective judgment and objective information. (2) The combined weight-TOPSIS model and gray relational theory [53] are usually used to carry out a comparative evaluation of dam systemic risks in the same study area. The results are shown in Table 6. Both methods are simple to operate and require quick calculations. The ranking results obtained by the combined weight-TOPSIS model and the gray rela- tional analysis are consistent, indicating that the risk evaluation model proposed is reason- able and reliable. A linear fitting is performed on the ranking value of the posting progress, and it is considered that the slope value can explain the overall resolution level of the rank- ing result [54]. After calculation, the linear fitting functions of the combined weight-TOP- SIS model and the gray relational analysis model’s post schedule value are y = 0.0483x + 0.2445 and y = 0.0263x + 0.0925 (Fig 9), respectively. The slope values are 0.0483 and 0.0263, respectively. Therefore, the combined weight-TOPSIS model is better at the evalua- tion and resolution levels and is more conducive to decision-making and judgment. (3) Subjective weight and objective weight contribute equally to the combined weight. The weight can be further optimized by adjusting the degree of contribution of the subjective weight and objective weight. After subsequent calculations, it is found that when different subjective weights and objective weights are used, such as 0.3: 0.7, 0.4: 0.6, 0.5: 0.5, 0.6: 0.4, and 0.7: 0.3, the risk ranking results are consistent. After a linear fitting is made on the Table 6. Comparison of closeness degree and risk ranking results between the combination weight-TOPSIS and gray relational analysis. Dam system unit name Combination weight-TOPSIS Gray relational analysis Closeness Risk ranking Evaluation value Risk ranking Guandigou Unit Wangtagou Unit Wangmaogou Unit 2 Nianyangou Unit Kanghegou Unit Huangbaigou Unit 0.349 0.301 0.566 0.408 0.393 0.466 https://doi.org/10.1371/journal.pone.0287750.t006 2 1 6 4 3 5 0.154 0.136 0.303 0.170 0.158 0.187 2 1 6 4 3 5 PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 15 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed Fig 9. Comparison of fitting results of correlation closeness between the combined weight-TOPSIS model and gray theory model. https://doi.org/10.1371/journal.pone.0287750.g009 sorted ideal point-posting progress value, the slopes are 0.040, 0.0432, 0.0483, 0.0497, and 0.0529. When the subjective weight and objective weight ratio is 0.3:0.7, the ideal point- posting progress value of each unit is relatively scattered. The resolution level is elevated, which is convenient for decision-making. However, the degree of optimization is not high, to facilitate the calculation. The combination weight is still calculated based on the equal contribution of the subjective weight and objective weight. Conclusions 1. The weighting choice has a considerable impact on the assessment of check dam threats in small watersheds. Three first-level indicators and ten second-level indicators that have an impact on risk assessment are determined using three features of flood catastrophe, opera- tion, and economic risk. The analytical hierarchy process (AHP) and the entropy weight method are used to determine the subjective and objective weights of risk assessment indi- cators, and their combined weights are used to ensure the objectivity and impartiality of the evaluation indicators when computing the weights. 2. For the check dam system in the constrained watershed of the Loess Plateau, a combination weight TOPSIS-based risk assessment model has been developed. By applying the devel- oped model to the check dam system units in Wangmaogou basin in China, the risk rank- ing result from the lowest to the highest risk is as follows: Wangmaogou 2# unit < Huangbaigou unit < Kanghegou unit < Nianyangou unit < Guandigou unit < Wangtagou unit. The risk ranking results are consistent with the actual situation. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 16 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed 3. The correlation between the ranking results generated by the combination weight method, AHP method, and entropy weight technique combined with TOPSIS was tested using the Spearman rank correlation coefficient test method. According to the correlation coefficient, which is not less than 0.943, the combination weight TOPSIS model efficiently integrates subjective experience assessment and objective data information. The idea is understand- able, and the outcome makes sense. 4. In the given case, the outcomes of the combined weight TOPSIS model and the gray corre- lation theory’s risk evaluation are consistent, although their linear slopes are 0.0483 and 0.0270, respectively. Therefore, in terms of evaluation resolution level and choice judgment, the combined weight TOPSIS model is more advantageous. The factors affecting the safety of the check dam are not only the 10 indices selected in this paper, but also the selection of evaluation indices for the check dam connected by different operation states and different dam systems. Therefore, the selection of evaluation indices and the grade standards of indices should be further studied. Supporting information S1 Table. (CSV) S1 File. (XLSX) S2 File. (RAR) Author Contributions Conceptualization: Lin Wang. Data curation: Lin Wang. Formal analysis: Lin Wang. Funding acquisition: Lin Wang. Investigation: Lin Wang. Methodology: Lin Wang. Project administration: Lin Wang. Resources: Lin Wang. Software: Lin Wang, Qiang Zu. Supervision: Lin Wang. Validation: Lin Wang, Qiang Zu. Visualization: Lin Wang. Writing – original draft: Lin Wang, Qiang Zu, Qiang Zhang. Writing – review & editing: Lin Wang, Qiang Zu, Qiang Zhang. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 17 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed References 1. Zhou ZX, Li J. The correlation analysis on the landscape pattern index and hydrological processes in the Yanhe watershed, China. J Hydrol. 2015; 524:417–26. http://doi.org/10.1016/j.jhydrol.2015.02.028. 2. Xu XZ, Zhang HW, Zhang OY. Development of check-dam systems in gullies on the Loess Plateau, China. Environ Sci Policy. 2004; 7(2):79–86. http://doi.org/10.1016/j.envsci.2003.12.002. 3. Liu YL, Jia LL, Zhang YD. Planning Ideas and Layout of check dams in the Loess Plateau in the New Era. China Soil and Water Conservation. 2020(10). 4. Cascini L, Cuomo S, Pastor M, Rendina I. Modelling of debris flows and flash floods propagation: a case study from Italian Alps. Eur J Environ Civ En. 2020:1–24. http://doi.org/10.1080/19648189.2020. 1756418. 5. Huang Y, Zhang B. Challenges and perspectives in designing engineering structures against debris- flow disaster. Eur J Environ Civ En. 2020:1–22. http://doi.org/10.1080/19648189.2020.1854126. 6. Nichols MH, Polyakov VO. The impacts of porous rock check dams on a semiarid alluvial fan. Sci Total Environ. 2019; 664:576–82. https://doi.org/10.1016/j.scitotenv.2019.01.429 PMID: 30763838 7. Previati M, Canone D, Bevilacqua I, Boetto G, Pognant D, Ferraris S. Evaluation of wood degradation for timber check dams using time domain reflectometry water content measurements. Ecol Eng. 2012; 44:259–68. http://doi.org/10.1016/j.ecoleng.2012.03.004. 8. Sun CJ, Zhang WQ, Li XG, Sun JL. Evaluation of ecological effect of gully region of loess plateau based on remote sensing image. Transactions of the Chinese Society of Agricultural Engineering (Transac- tions of the CSAE). 2019; 35(12):165–72. 9. Yazdi J, Sabbaghian Moghaddam M, Saghafian B. Optimal Design of Check Dams in Mountainous Watersheds for Flood Mitigation. Water Resour Manag. 2018; 32(14):4793–811. http://doi.org/10.1007/ s11269-018-2084-4. 10. Yuan SL, Li ZB, Li P, Xu GC, Gao HD, Xiao L, et al. Influence of Check Dams on Flood and Erosion Dynamic Processes of a Small Watershed in the Loss Plateau. Water-Sui. 2019; 11(4):834. http://doi. org/10.3390/w11040834. 11. Guo WZ, Wang WL, Xu Q, Hu JJ, Zhu LL. Distribution, failure risk and reinforcement necessity of check-dams on the Loess Plateau: a review. J Mt Sci-Engl. 2021(2). http://doi.org/10.1007/s11629- 020-6090-7. 12. China MOWR. Technical specification of check dams for farmland forming. In: China MOWR, editor SL/ T 804–2020: China Water Conservancy and Hydropower Press; 2020. p. 57. 13. 14. Li PY, Qian H, Wu JH. Environment: Accelerate research on land creation. Nature. 2014; 510 (7503):29–31. https://doi.org/10.1038/510029a PMID: 24910868 Zhang FY, Yan BB, Feng XM, Lan HX, Kang C, Lin XS, et al. A rapid loess mudflow triggered by the check dam failure in a bulldoze mountain area, Lanzhou, China. Landslides. 2019; 16(10):1981–92. http://doi.org/10.1007/s10346-019-01219-2. 15. Wei X, Li ZB, Wu JH, Li BB, Du Z. A Discussion on Some Ideological Problems in Research of Water Damage Hazards of Check Dam. Research of Soil and Water Conservation. 2007. 16. Wei YH, Wang ZJ, He Z, Yu WJ, Li YJ, Jiao JY. Investigation and Evaluation on Check Dams Damaged Condition Under Continuous Rainstorm in Yanhe River Basin in July 2013. Bulletin of Soil and Water Conservation. 2015; 35(3):250–5. 17. Wang ZL, Zhang BZ, Liu HZ, LI CJ, Lin XZ, Yang JS. Damage causes and sand-blocking effects of warping dams in Dalat Banner in 2016. Advances in Science and Technology of Water Resources. 2019; 39(04):1–6. 18. Dang WQ, Hao LD, Gao JJ, Dang TM, Bai Y. Roles of silt retention dam in rainstorm flood disaster on July. China Water Resources. 2019(08):52–5. 19. Abbasi NA, Xu XZ, Lucas-Borja ME, Dang WQ, Liu B. The use of check dams in watershed manage- ment projects: Examples from around the world. Sci Total Environ. 2019; 676:683–91. https://doi.org/ 10.1016/j.scitotenv.2019.04.249 PMID: 31054413 20. Cucchiaro S, Cavalli M, Vericat D, Crema S, Llena M, Beinat A, et al. Geomorphic effectiveness of check dams in a debris-flow catchment using multi-temporal topographic surveys. Catena (Giessen). 2019; 174:73–83. http://doi.org/10.1016/j.catena.2018.11.004. 21. Pal D, Galelli S, Tang HL, Ran QH. Toward improved design of check dam systems: A case study in the Loess Plateau, China. J Hydrol. 2018;559. http://doi.org/10.1016/j.jhydrol.2018.02.051. 22. Piton G, Carladous S, Recking A, Tacnet JM, Lie´ bault F, Kuss D, et al. Why do we build check dams in Alpine streams? An historical perspective from the French experience. Earth Surf Proc Land. 2017; 42 (1):91–108. http://doi.org/10.1002/esp.3967. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 18 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed 23. 24. Zhang FY, Wang GH, Kamai T, Chen WW, Zhang DX, Yang J. Undrained shear behavior of loess satu- rated with different concentrations of sodium chloride solution. Eng Geol. 2013; 155:69–79. http://doi. org/10.1016/j.enggeo.2012.12.018. Zhang FY, Huang XW. Trend and spatiotemporal distribution of fatal landslides triggered by non-seis- mic effects in China. Landslides. 2018; 15(8):1663–74. http://doi.org/10.1007/s10346-018-1007-z. 25. Wang D, Li ZB, Li P, Gao HD, Zhao BH, Yuan SL. Evaluation of Overall Distribution of Check Dam Sys- tem in the Jiuyuangou Watershed. Research of Soil and Water Conservation. 2016; 23(05):49–55. 26. Peng H, Liu DF, Tian B, Zuo JJ. Study on the Comprehensive Safety Assessment of Earth Fill Dam Based on AHP Methods. Advances in Water Resources and Hydraulic Engineering. 2009. http://doi. org/10.1007/978-3-540-89465-0_318. 27. Stefanidis S, Stathis D. Assessment of flood hazard based on natural and anthropogenic factors using analytic hierarchy process (AHP). Nat Hazards. 2013; 68(2):569–85. http://doi.org/10.1007/s11069- 013-0639-5. 28. Yasser M, Jahangir K, Mohmmad A. Earth dam site selection using the analytic hierarchy process (AHP): a case study in the west of Iran. Arab J Geosci. 2013; 6(9):3417–26. http://doi.org/10.1007/ s12517-012-0602-x. 29. Wang B, Xie HL, Ren HY, Li X, Chen L, Wu BC. Application of AHP, TOPSIS, and TFNs to plant selec- tion for phytoremediation of petroleum-contaminated soils in shale gas and oil fields. J Clean Prod. 2019; 233(OCT.1):13–22. http://doi.org/10.1016/j.jclepro.2019.05.301. 30. 31. Li ZK, Li W, Ge W. Weight analysis of influencing factors of dam break risk consequences. Nat Hazard Earth Sys. 2018; 18(12):3355–62. http://doi.org/10.5194/nhess-18-3355-2018. Li C, Zhang FR, Zhu TF, Feng T, An PL. Evaluation and correlation analysis of land use performance based on entropy-weight TOPSIS method. Transactions of the Chinese Society of Agricultural Engi- neering (Transactions of the CSAE). 2013; 29(05):217–27. 32. Sahoo MM, Patra KC, Swain JB, Khatua KK. Evaluation of water quality with application of Bayes’ rule and entropy weight method. Eur J Environ Civ En. 2017; 21(6):730–52. http://doi.org/10.1080/ 19648189.2016.1150895. 33. Hwang CL, Yoon K. Methods for Multiple Attribute Decision Making.: Multiple Attribute Decision Making; 1981. 34. Cui NB, Zhang ZP, Lou YH, Gong DZ, Liu XZ. Evaluation model of comprehensive agricultural water- saving development level based on TOPSIS in a regional area. Journal of Basic ence and Engineering. 2016. http://doi.org/10.16058/j.issn.1005-0930.2016.05.011. 35. Terracciano G, Di Lorenzo G, Formisano A, Landolfo R. Cold-formed thin-walled steel structures as ver- tical addition and energetic retrofitting systems of existing masonry buildings. Eur J Environ Civ En. 2015; 19(7):850–66. http://doi.org/10.1080/19648189.2014.974832. 36. Chu JY, Ma DX, Su YP. Study on site selection of resident emergency congregate shelters based on combination weighting TOPSIS. China Civil Engineering Journal. 2013. 37. Mathew M, Chakrabortty RK, Ryan MJ. Selection of an Optimal Maintenance Strategy Under Uncertain Conditions: An Interval Type-2 Fuzzy AHP-TOPSIS Method. Ieee T Eng Manage. 2020; PP(99):1–14. http://doi.org/10.1109/TEM.2020.2977141. 38. 39. Liu L, Wan X, Li J, et al. An Improved Entropy-Weighted Topsis Method for Decision-Level Fusion Eval- uation System of Multi-Source Data[J]. Sensors, 2022, 22(17): 6391. de FSM Russo Rosaria, Camanho Roberto. Criteria in AHP: a systematic review of literature. Procedia Computer Science 55 (2015): 1123–1132. 40. Ho W, Ma X. The state-of-the-art integrations and applications of the analytic hierarchy process. Euro- pean Journal of Operational Research, 2018; 267(2): 399–414. 41. 42. Zhao J, Ji G, Tian Y, et al. Environmental vulnerability assessment for mainland China based on entropy method. Ecological Indicators, 2018; 91: 410–422. Li J, Zhang JZ, Wang X. Cause analysis of check dam flood damage in 1970s. China Water Resources. 2003(17):55–6. 43. Wang D, Ha YL, Li ZB, Yu KX, Bu CD, Zhang HW, et al. Evaluation on the operation risk of dam Tsys- tem in the typical watersheds of Ningxia. Science of Soil and Water Conservation. 2017; 15(03):17–25. 44. Wang YS, Wang YS. Silt Arresters destroyed by storm flood in 1994 and investigation on effect of retaining silt in middle Yellow River Region. Soil and Water Conservation in China. 1995(8):23–6. 45. Yang R, Li ZL, Wang D, Yuan SL. Safety evaluation of check dam system in small watershed of loess plateau. Journal of Yan’an University (Natural Science Edition). 2018; 37(01):41–45. (in Chinese) PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 19 / 20 PLOS ONEA procedure for risk assessment of check dam systems: A case study of Wangmaogou watershed 46. Yuan SL, Zhang WH, Wang Z. Study on safety evaluation system of check dam system in small water- shed in loess hilly region. Western Development (land development project research).2018; 3(05):40– 45. 47. Wang D, Ha YL, Li ZB, Yu KX, Bu CD, Zhang HW, et al. Operational risk assessment of check dam sys- tem in Ningxia typical watershed. Science of Soil and Water Conservation in China. 2017; 15(03):17– 25. (in Chinese). 48. Wang D, Li ZB, Li P, Gao HD, Zhao BH, Yuan SL, et al. Evaluation on layout of warping dam system in Jiuyuangou watershed. Research on water and soil conservation. 2016; 23(05):49–55. (in Chinese)46. 49. Chang EH, Li P, Li ZB, Xiao L, Zhao BH, Su YY, et al. Using water isotopes to analyze water uptake dur- ing vegetation succession on abandoned cropland on the Loess Plateau, China. Catena. 2019; 181:104095. http://doi.org/10.1016/j.catena.2019.104095. 50. Zhao BH, Li ZB, Li P, Xu GC, Gao HD, Cheng YT, et al. Spatial distribution of soil organic carbon and its influencing factors under the condition of ecological construction in a hilly-gully watershed of the Loess Plateau, China. Geoderma. 2017; 296:10–7. http://doi.org/10.1016/j.geoderma.2017.02.010. 51. Chen ZY, Huang XP, Yu S, Cao W, Dang WQ, Wang YQ. Risk analysis for clustered check dams due to heavy rainfall. Int J Sediment Res. 2021; 36(2):291–305. http://doi.org/10.1016/j.ijsrc.2020.06.001. 52. Zhang Z, Chai J, Li Z, et al. Reconstruction and effects of a failure of a typical check dam system under an extreme rainstorm on the Loess Plateau, China[J]. Natural Hazards, 2022, 111(2): 1401–1419. 53. Kumar JA, Abirami S. Aspect-based opinion ranking framework for product reviews using a Spearman’s rank correlation coefficient method. Information ences. 2018:23–41. http://doi.org/10.1016/j.ins.2018. 05.003. 54. Zhao P, Guo B, Wang PZ. The Depreciation Method of Construction Machinery Based on Gray Rela- tional Theory. Advanced Materials Research. 2012;374–377:1265–8. http://doi.org/10.4028/www. scientific.net/AMR.374-377.1265. PLOS ONE \| https://doi.org/10.1371/journal.pone.0287750 June 27, 2023 20 / 20 PLOS ONE
10.2196_46793	JOURNAL OF MEDICAL INTERNET RESEARCH Li et al Original Paper Delivery of WeChat-Based HIV Result e-Reports in Social Networks for Recruitment of High-Risk Population: Baseline Data From a Cluster Randomized Controlled Trial Ju-Shuang Li1,2, MSc; Yu-Zhou Gu3, MSc; Feng-Su Hou2,4, DPhil; Yong-Heng Lu5,6, BSc; Xiao-Ru Fan1,2, BSc; Jia-Ling Qiu7, MSc; Qing-Ling Yang1,2, MSc; Jing Gu1,2, DPhil; Jing-Hua Li1,2, DPhil; Dong Roman Xu8,9,10, DPhil; Chun Hao1,2, DPhil 1Department of Medical Statistics, School of Public Health, Sun Yat-Sen University, Guangzhou, China 2Sun Yat-Sen Global Health Institute, Institute of State Governance, Sun Yat-Sen University, Guangzhou, China 3Guangzhou Center for Disease Control and Prevention, Guangzhou, China 4Department of Public Health, Shenzhen Kangning Hospital, Shenzhen, China 5Guangzhou Lingnan Community Support Center, Guangzhou, China 6Kangyuan Community Support Center of Yuexiu District, Guangzhou, China 7Department of Public Health, Guangdong Women and Children Hospital, Guangzhou, China 8Center for World Health Organization Studies, School of Health Management, Southern Medical University, Guangzhou, China 9Department of Health Management, School of Health Management, Southern Medical University, Guangzhou, China 10Acacia Lab for Implementation Research, Southern Medical University Institute for Global Health, Dermatology Hospital, Southern Medical University, Guangzhou, China these authors contributed equally Corresponding Author: Chun Hao, DPhil Department of Medical Statistics School of Public Health Sun Yat-Sen University 74 Zhongshan 2nd Rd Yuexiu District Guangzhou, Guangdong 510080 China Phone: 86 87332517 Email: haochun@mail.sysu.edu.cn Abstract Background: Disclosure of infectious disease status to social network peers can facilitate reaching and early detection among high-risk populations. In this era of social media, globally, HIV/AIDS represents a high burden of infectious disease. Thus, delivery of an HIV result e-report via social media presents a new approach that has the potential to improve contact with and enrollment of the high-risk population in research studies and routine practice. Objective: This study explores the effectiveness and associated factors of a recruitment strategy (ie, WeChat-based HIV e-report delivery in social networks) on the enrollment of men who have sex with men (MSM) for an HIV testing intervention study. Methods: This was an enrollment result analysis of an ongoing cluster randomized controlled trial (RCT) aiming to promote HIV testing among MSM. Recruitment of potential participants was based on the unit of an egocentric social network, which includes 1 core member (an offline tested ego as the recruiter) and several network members (online alters as network associates). Alters’ enrollment and alters’ transformation to ego-recruiters (alter-ego) were measured as outcomes. Recruitment outcomes were compared between the exchangeable and regular e-report groups of the RCT. Associated factors of both outcomes were also investigated, including sociodemographic characteristics, health behaviors, social network characteristics, e-report types, and online delivery information. Binary outcomes were modeled using logistic models, with Firth correction for rare events. Qualitative interviews were conducted to understand facilitators and barriers in detail for alter-ego as the subsequent wave’s recruiter. https://www.jmir.org/2023/1/e46793 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 1 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al Results: The e-report of 1157 egos who tested offline were delivered to 5165 alters in 3 recruitment waves; eventually, 1162 eligible alters enrolled in this RCT (response rate: 22.5%). In the exchangeable e-report group, 544 egos recruited 467 alters, of which 35 alters transformed to alter-egos (7.5%), whereas in the regular e-report group, 613 egos recruited 695 alters, of which 40 alters transformed to alter-egos (5.8%). Alters’ enrollment at first wave was associated with a higher number of e-reports being forwarded by the egos. Alters’ transformation to alter-egos for the subsequent wave was associated with the exchangeable e-report, higher income, being a Guangzhou resident, unprotected anal intercourse, preferring self-testing, and viewing senders’ e-reports frequently. Qualitative interviews revealed that the lack of awareness of e-reports’ function and inadequate access to e-reports at offline testing facilities were major barriers to alters’ transformation to offline ego-recruiters. Conclusions: The delivery of e-report was feasible in MSM social network, and the success and sustainability of online recruitment depended on high levels of familiarity among MSM with the digital tool. The HIV e-report exchange mechanism might promote MSM to test HIV offline to get their own e-report for exchange in the community. The e-report provides an innovative recruitment method with great potential to trace direct contacts for infectious diseases studies. (J Med Internet Res 2023;25:e46793) doi: 10.2196/46793 KEYWORDS social network strategy; HIV result e-report; recruitment; MSM Introduction Controlling the spread of viruses, such as HIV, is essential for preventing and managing infectious diseases [1]. Early testing plays a crucial role in controlling infectious disease outbreaks [2,3]. The Joint United Nations Programme on HIV/AIDS (UNAIDS) has set an ambitious target of ensuring that 95% of people living with HIV are aware of their status by 2030 [4]. Despite these efforts, the coverage of HIV testing, particularly among high-risk populations such as men who have sex with men (MSM), remains insufficient [5]. In 2020, the HIV prevalence rate among MSM in China stood at 6%, but only 62.2% of these individuals were aware of their HIV status [6,7]. To address this challenge, it is imperative that efforts are made to scale-up HIV testing among MSM in China. Enhancing internal demand for HIV testing services and interventions is essential for the successful and sustainable scaling up of HIV testing. To achieve this goal, community engagement plays a critical role. It is widely recognized that fulfilling one’s sexual needs is a fundamental aspect of life, and community-driven efforts aimed at reducing HIV risk through the disclosure of HIV status before engaging in intimate relationships are important [8,9]. In the MSM community, disclosing one’s HIV test results to potential partners is a prerequisite as “show me your safe sex license,” which is vital to increasing HIV testing rates and decreasing HIV infections [10,11]. To support this initiative, we have developed a mechanism for exchanging HIV test results electronically (HIV e-report) as an intervention. In this system, individuals who do not have their own HIV e-report will need to undergo testing in exchange for access to the e-report of others. The efficacy on promoting HIV testing will be evaluated by a randomized controlled trial (RCT) [12]. The delivery of the intervention and recruitment of participants in this study were based on the internet-based social network strategy (SNS). SNS using an initial group of individuals as “seeds” or “egos” to recruit others within their social networks has been shown to effectively recruit a large number of high-risk MSM for HIV testing [13,14]. However, the impact of these networks on recruitment and the factors that influence it are not https://www.jmir.org/2023/1/e46793 XSL•FO RenderX well documented. Moreover, the effectiveness of combining SNS with the HIV e-report disclosure approach is not evaluated before, but it is expected that internet-based social networking will become increasingly prevalent in the future. Therefore, the aim of this study was to investigate the effectiveness and associated factors of a recruitment strategy (WeChat-based HIV e-report delivery in social networks) in enrolling MSM in an HIV testing intervention study. Methods Overview This study analyzed the baseline data of MSM recruited in a cluster RCT in Guangzhou (Guangdong, China) from September 2019 to January 2022 [12]. The RCT aimed to evaluate the efficacy of an HIV e-report exchange mechanism in promoting HIV testing among MSM by comparing HIV testing behavior between a group that adopted the certified HIV e-report exchange mechanism (the exchangeable e-report group) and another that adopted the regular e-report delivery mechanism (the regular e-report group). Participants and Study Recruitment Procedures Recruitment was conducted at the Guangzhou Lingnan Community Support Center (hereinafter, Lingnan), a local MSM-friendly HIV testing clinic supported by the Guangzhou Center for Disease Control and Prevention (CDC) [15]. Lingnan completes approximately 400,000 person-times per year of HIV testing among local MSM [12]. Potential participants were recruited according to the unit of an egocentric social network, which included 1 core member (ego) and several network members (alters), with the tie of HIV e-report delivery. The e-reports were only accessible to MSM who had undergone HIV testing at Lingnan. Ego recruiters were defined as MSM who underwent offline HIV testing at Lingnan and met the inclusion criteria. After agreeing to participate in this study as recruiters and filling in the written consent form, they shared their e-reports via the WeChat (Tencent Holdings Ltd.) mini-program “ChaBei” (an app for online booking of HIV test appointments, online-to-offline referral, offline clinic J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 2 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al to testing registration, and HIV e-report delivery) to their WeChat MSM contacts. Egos were randomly assigned the exchangeable HIV e-report or regular HIV e-report group through the WeChat mini-program “ChaBei.” Alters were the MSM contacts who received and read the shared reports; those who click the link will be candidate alters in our study. An online page that provides study details will be presented to the candidate alters after they read the content of ego’s report. Those interested in this study will be asked to complete the screening questionnaire and eligible individuals will fill in an online consent form and become alters. The specific details of e-reports distributed through “ChaBei” were described in our published protocol [12]. The e-report was convenient and secure on the internet, and its authenticity was guaranteed by the CDC. The recruitment of MSM participants was based on the type of e-report they received: exchangeable e-reports, which could only be viewed after an exchange of reports between the sender and the receiver; and regular e-reports, which could be viewed by the receiver regardless of whether they had their own e-report. Assessments Recruitment Outcomes The first recruitment outcome was whether egos recruited alters completing online enrollment by delivering HIV e-reports. Egos who successfully recruited alters were defined as ego-RA; otherwise they were defined as ego-NRA. For the first outcome, egos were the target population for analysis. The second recruitment outcome was alters’ transformation into alter-egos. Alters were considered to have transformed into offline ego recruiters for the subsequent recruitment wave (hereinafter “alter-egos”) when they had transformed from online to offline egos after they tested for HIV at Lingnan and received an HIV e-report. For the second outcome, alters were the target population for analysis. Alter-egos delivering HIV e-reports to their friends in the MSM community could expand recruitment in subsequent waves. From these data, we created 2 dichotomous dependent variables: (1) whether ego successfully recruited alters; and (2) whether alter transformed into alter-ego. Additionally, when e-report online delivery was identified as a significant influencing factor for recruitment, it was further applied as a dependent variable to explore its influencing factors. Independent Variables Associations Between Egos’ Characteristics and Alters’ Online Enrollment The following egos’ characteristics were included as independent variables to analyze their influence on alters’ online enrollment: demographic characteristics, socioeconomic status, information, and HIV e-report delivery MSM-related information. Demographic characteristics were age, marital status, education level, length of residence in Guangzhou, and registered permanent residence. The socioeconomic status variable was monthly income. MSM features contained sexual orientation, sex roles, main venues to make gay friends, and anal sexual intercourse. HIV e-report online delivery information included type of HIV e-report, the number and duration of viewing own/others’ e-report, the number of times forwarding the e-report, and the number of people/times clicked on e-report. Associations Between Alters’ Characteristics and Alters’ Transformation into Alter-Egos included as The following alters’ characteristics were independent variables to analyze their influence on alters’ transformation into offline ego recruiters (alter-egos) for the second recruitment wave: demographic characteristics, socioeconomic status, MSM-related information, HIV e-report delivery information, HIV/sexually transmitted infection (STI) testing information, high-risk sexual behavior, HIV-related information, and social network characteristics. HIV/STI testing information included HIV testing, preference for HIV testing, STI testing, and self-report STI status. High-risk sexual behavior included having regular/casual partners in the past 3 months, having unprotected anal intercourse (UAI) with regular/casual partners in the past 3 months, and awareness of the HIV status of regular/casual partners. HIV-related information included intervened by any HIV prevention program, HIV risk perceptions, HIV stigma, and HIV testing social norms. The variables included in the egocentric social network analysis were similarity of characteristics, which is defined as the degree of consistency of demographic characteristics between each respondent (ego) and each of his associates (alter); and strength of relational ties, which is measured by the degree of relational intimacy. Demographic characteristics such as age, education, marriage, and income were collected from egos and alters, and similarities in social attributes between them were calculated using an edit distance algorithm [16]. Age similarity was calculated as follows: simage(ego, alter) = 1 – (d/D) where d is the absolute value of the age difference between an ego and his alter, and D is the absolute value of the maximum difference in age attributes between them. Meanwhile, education similarity, simedu(ego, alter); marital status similarity, simmar(ego, alter); and personal income similarity, siminc(ego, alter) were calculated as above. The total similarity between an ego and his alter was calculated as follows: Data Collection Egos’ demographic characteristics, socioeconomic status, and MSM-related information were collected face-to-face through well-trained surveyors in Lingnan using an electronic standardized questionnaire. Alters’ demographic characteristics, socioeconomic status, MSM-related information, high-risk sexual behavior, and HIV-related information were collected online using an electronic standardized questionnaire. information, HIV/STI testing Recruitment outcomes, egos’ and alters’ HIV e-report online delivery information, and alters’ social network data were obtained from participant recruitment logs and data recorded through the WeChat-based data portal. https://www.jmir.org/2023/1/e46793 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 3 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al Data Analysis Descriptive Analysis of Overall Recruitment The social network visualization graph was used to show the overall recruitment situation. Recruitment outcomes were compared between the 2 arms of the RCT. Continuous variables with normal or near-normal distribution are described as the mean (SD) and were compared between groups using the Student t test. Those with skewed distribution are described as the median (Q1-Q3) and were compared between groups using the Mann-Whitney U test. Categorical data are presented as the number of cases (%) and were compared between groups using the chi-square or Fisher exact test. Regression Analysis of Recruitment-Associated Factors Two-part models were fitted to estimate correlates of alters who had successfully enrolled. In the first part, a logistic regression model was used to estimate the probability of alters who successfully enrolled among those to whom egos had forwarded e-reports. In the second part, for egos that successfully recruited alters, the number of alters recruited was estimated using ordinary least squares regression. In addition, factors associated with e-report delivery were analyzed using logistic regression. As a relatively small percentage of alters transformed into alter-egos in the subsequent recruitment waves, Firth’s penalized likelihood [17,18] can be used to minimize the analytical bias caused by small samples and rare events. Therefore, to explore the transformation into alter-egos, logistic regression analysis with Firth correction was used for impact factor analysis. Before that, the multilevel model was used to test for aggregation within clusters. To clearly illustrate factors influencing recruitment, the aforementioned univariate and multivariate regression models were applied to the first-wave egos (NEgo-wave1=1083) and alters (NAlter-wave1=1050). Given the large sample size, mixing subsequent-wave egos and alters was avoided to prevent dilution of the results. In addition, we conducted subsequent waves recruitment and subgroup analysis based on the grouping of the types of e-reports as supplementary results. Qualitative Analysis Qualitative interviews were conducted to understand the facilitators of and barriers to online alters’ transformation into offline egos for subsequent recruitment waves. The qualitative interviews were focused on 2 primary topics: alters’ demand for appointment-based HIV testing services and the utilization of e-reports. All the recorded data were transcribed into text and imported into ATLAS.ti 6.2 (ATLAS.ti GmbH) software for analysis. The textual information was coded according to the research objectives and interview outline, and the comparative method was used for multistage data thematic analysis, data mining by coding and categorizing, and screening meaningful content. All data management and statistical analyses were conducted using RStudio for Windows (version 1.4.1103; RStudio, PBC). A social network visualization graph was obtained using Gephi 0.9.4 for Windows (Gephi Team). All tests were 2-sided, and significance was set at P<.05. https://www.jmir.org/2023/1/e46793 XSL•FO RenderX Ethical Considerations The study was reviewed and approved by the Ethics Committee of Sun Yat-sen University (Institutional Review Board number 054/19; February 28, 2019) and confirmed by Lingnan before the study. Participation in this study was voluntary and all data were made anonymous. We included only those that had provided written or online informed consent. Egos received a compensation of 20 RMB (≈US $3) for each alter recruited, and alters received a compensation of 50 RMB (≈US $7.5) after completing the questionnaire. Compensation was issued through a reward (E-Red Pocket) on WeChat. Results Overall Recruitment Figure 1 presents the study recruitment flowchart. From September 2019 to January 2022, 1157 eligible egos were invited from Lingnan to participate as recruiters, and they generated 5165 online alters through 3 recruitment waves; of these, 1162 eligible alters enrolled in the RCT, giving a 22.5% (1162/5165) response rate. The 3 consecutive waves recruited 1083 egos and 1050 alters (response rate: 1050/4716, 22.3%), 70 egos and 100 alters (100/325, 30.8%), and 5 egos and 12 alters (12/124, 9.7%), respectively. Figure 2 visually depicts the social networks of the study participants. The mean (Q1-Q3) age of egos and alters across all waves was 27.0 (23.0-31.0) years and 26.0 (23.0-30.0) years, respectively. Among all egos, 92.7% (1072/1157) were unmarried, 85.0% (984/1157) had received a high school education or above, and 70.9% (820/1157) were permanent Guangdong residents. Among all alters, 94.2% (1095/1162) were unmarried, 54.9% (638/1162) had received a high school education or above, and 69.6% (809/1162) were permanent Guangdong residents. A total of 613 egos were assigned to the regular HIV e-report group and 695 alters were recruited; 544 egos were assigned to the exchangeable HIV e-report group and 467 alters were recruited. Further background information for both the egos and alters from each wave can be found in Tables S1 and S2 in Multimedia Appendix 1. Table 1 shows egos’ characteristics and the overall recruitment outcome between the exchangeable and regular HIV e-report groups. The mean (Q1-Q3) age of egos was 27.0 (23.0-31.0) years. Most were unmarried (1072/1157, 92.7%), with a monthly income over 5000 RMB (US $771; 698/1157, 60.3%) and education level above high school (984/1157, 85.0%). There were imbalances in viewing others’ e-report as well as in e-reports’ forwarding and click-throughs. Further, the successful recruitment rate of egos was higher in the regular HIV e-report group than in the exchangeable HIV e-report group (294/613, 48.0% vs 223/544, 41.0%; P=.02). All other characteristics of the 2 groups of egos were similar. For alters’ transition to offline ego-recruiters (alter-ego), 70/1050 (6.67%) and 5/100 (5%) alter-egos were transformed from 1050 first-wave alters and 100 second-wave alters, respectively. The 5 alter-egos that participated in all recruitment waves were from 4 social networks (Figure S1 in Multimedia Appendix 1). The median age of alters in the all waves was 26.0 J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 4 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al (IQR 23.0-30.0) years; further, they were mostly unmarried (1095/1162, 94.2%), with monthly income over 5000 RMB (US $771; 681/1162, 58.6%) and education level above high school (638/1162, 54.9%). There were 467 alters in the exchangeable HIV e-report group, of which 35 transformed into alter-ego (transformation rate: 7.5%); and 695 alters in the regular HIV e-report group, of which 40 transformed into alter-ego (transformation rate: 5.8%). Overall, the baseline characteristics of the 2 groups of alters were well balanced. However, there were significant differences between the exchangeable HIV e-report group and the regular HIV e-report group in terms of receiving HIV prevention services (P=.05), stigma (P=.04), viewing others’ e-report (P<.001), and social network similarity (P=.01; Table 2). Figure 1. Flowchart showing recruitment of study participants for the HIV e-report project. *One alter-ego did not complete the ego questionnaire and his basic information was missing. Alter: those contactors who received and read the shared report from ego; alter-ego: when the alter received his friend's e-report, he transformed into the ego after taking an HIV test at the Lingnan Center and forwarded his e-report to his gay friends to expand recruitment; ego: men who have sex with men who tested HIV at the Lingnan Center and shared their HIV e-report via the WeChat mini-program to their WeChat contactors. Figure 2. Social network visualization of participants' recruitment. (A) Overall social network visualization; (B) partial social network visualization. Alter: those contactors who received and read the shared report from ego; alter-ego: when the alter received his friend's e-report, he transformed into an ego after taking an HIV test at the Lingnan Center and forwarded his e-report to his gay friends to expand recruitment; ego: men who have sex with men that tested HIV at the Lingnan Center and shared his HIV e-report via the WeChat mini-program to his WeChat contactors. https://www.jmir.org/2023/1/e46793 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 5 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al Table 1. Characteristics of egos and the recruitment outcome of alters’ online enrollment. Characteristics Sociodemographic characteristics Total (NEgo=1157) Regular HIV e-report group (n1=613) Exchangeable HIV e-report group (n1=544)a P value Age (year), median (IQR) 27.0 (23.0-31.0) 27.0 (23.0-31.0) 27.0 (23.0-30.5) Currently unmarried, n (%) Yes No Monthly incomeb, n (%) ≤5000 RMB >5000 RMB Educational level, n (%) High school or below Above high school Missing Length of residence in Guangzhou, n (%) ≤1 year >1 year Missing Registered permanent residence, n (%) Guangdong province Other provinces MSMc-related information Sex role, n (%) Receptive Insertive Versatile Missing Sexual orientation, n (%) Homosexual Others Main ways to make friends , n (%) Internet Others Ever had anal sexual intercourse , n (%) Yes No HIV e-report deliver process information Has viewed his own e-report , n (%) 1072 (92.7) 85 (7.3) 459 (39.7) 698 (60.3) 148 (12.8) 984 (85.0) 25 (2.2) 178 (15.4) 971 (83.9) 8 (0.7) 820 (70.9) 337 (29.1) 383 (33.1) 468 (40.4) 296 (25.6) 10 (0.9) 899 (77.7) 258 (22.3) 1083 (93.6) 74 (6.4) 1120 (96.8) 37 (3.2) 567 (92.5) 46 (7.5) 242 (39.5) 371 (60.5) 84 (13.7) 517 (84.3) 12 (2.0) 104 (17.0) 505 (82.4) 4 (0.7) 449 (73.2) 164 (26.8) 197 (32.1) 255 (41.6) 153 (25.0) 8 (1.3) 461 (75.2) 152 (24.8) 572 (93.3) 41 (6.7) 590 (96.2) 23 (3.8) Yes No 1134 (98.0) 23 (2.0) 605 (98.7) 8 (1.3) 505 (92.8) 39 (7.2) 217 (39.9) 327 (60.1) 64 (11.8) 467 (85.8) 13 (2.4) 74 (13.6) 466 (85.7) 4 (0.7) 371 (68.2) 173 (31.8) 186 (34.2) 213 (39.2) 143 (26.3) 2 (0.4) 438 (80.5) 106 (19.5) 511 (93.9) 33 (6.1) 530 (97.4) 14 (2.6) 529 (97.2) 15 (2.8) Duration of viewing own e-report (seconds), median (IQR) 11.9 (7.7-18.7) 11.9 (7.8-18.7) 12.1 (7.5-18.7) .56 .83 .89 .56 .28 .06 .27 .03 .67 .26 .08 .62 https://www.jmir.org/2023/1/e46793 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 6 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Characteristics Number of times viewing own e-report (time), median (IQR) Has viewed others’ e-report, n (%) Yes No Duration of viewing others’ e-report (seconds), median (IQR) Number of times viewing others’ e-report (time), median (IQR) Number of times forwarding e-report (time), median (IQR) Number of people clicked on e-report, median (IQR) Number of times clicked on e-report, median (IQR) HIV e-report recruitment outcome Alters online enrolled, n (%) Total (NEgo=1157) Regular HIV e-report group (n1=613) Exchangeable HIV e-report group (n1=544)a 9.0 (4.0-16.0) 9.0 (4.0-17.0) 9.0 (4.0-16.0) 802 (69.3) 355 (30.7) 451 (73.6) 162 (26.4) 351 (64.5) 193 (35.5) 5.9 (0.0-11.0) 6.4 (0.0-12.1) 4.9 (0.0-9.7) 1.0 (0.0-3.0) 2.0 (0.0-4.0) 1.0 (0.0-3.0) 2.0 (1.0-5.0) 2.0 (1.0-6.0) 2.0 (1.0-4.0) 1.0 (0.0-2.0) 2.0 (1.0-4.0) 0.0 (0.0-0.0) 1.0 (0.0-3.0) 3.0 (1.0-8.0) 0.0 (0.0-0.0) Li et al P value .38 .001 d <.001 d <.001 d .003 d <.001 d <.001 d .0 2 d Yes (egos-RA) No (egos-NRA) 517 (44.7) 640 (55.3) 294 (48.0) 319 (52.0) 223 (41.0) 321 (59.0) aIn the exchangeable HIV e-report group, most participants did not complete the exchange process and therefore lacked click-related data of the e-reports. b1 RMB=US $0.15. cMSM: men who have sex with men. dP<.05. https://www.jmir.org/2023/1/e46793 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 7 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al Table 2. Characteristics of altersa and the recruitment outcome of alters’ transition to offline alter-ego. Characteristics Sociodemographic characteristics Age (year), median (IQR) Currently unmarried, n (%) Yes No Monthly incomec, n (%) ≤5000 RMB >5000 RMB Educational level, n (%) High school or below Above high school Length of residence in Guangzhou, n (%) ≤1 year >1 year Registered permanent residence, n (%) Guangdong province Other provinces MSMd-related information Sex role, n (%) Receptive Insertive Versatile Sexual orientation, n (%) Homosexual Others Main ways to make friends, n (%) Internet Others High-risk sexual behavior in the past 3 months Had casual sexual partner, n (%) Yes No UAIe with casual partners, n (%) Yes No Knowing the HIV status of casual partners, n (%) Yes No Had regular sexual partner, n (%) Total (NAlter=1162) Regular HIV e-re- port group (n3=695) Exchangeable HIV e-report group (n3=467)b P value 26.0 (23.0-30.0) 26.0 (23.0-30.0) 26.0 (23.0-30.0) 1095 (94.2) 67 (5.8) 481 (41.4) 681 (58.6) 524 (45.1) 638 (54.9) 507 (43.6) 655 (56.4) 809 (69.6) 353 (30.4) 404 (34.8) 410 (35.3) 348 (29.9) 932 (80.2) 230 (19.8) 1117 (96.1) 45 (3.9) 584 (50.3) 578 (49.7) 141 (12.1) 1021 (87.9) 218 (18.8) 944 (81.2) 657 (94.5) 38 (5.5) 291 (41.9) 404 (58.1) 301 (43.3) 394 (56.7) 306 (44.0) 389 (56.0) 473 (68.1) 222 (31.9) 247 (35.5) 249 (35.8) 199 (28.6) 556 (80.0) 139 (20.0) 664 (95.5) 31 (4.5) 362 (52.1) 333 (47.9) 88 (12.7) 607 (87.3) 119 (17.1) 576 (82.9) 438 (93.8) 29 (6.2) 190 (40.7) 277 (59.3) 223 (47.8) 244 (52.2) 201 (43.0) 266 (57.0) 336 (71.9) 131 (28.1) 157 (33.6) 161 (34.5) 149 (31.9) 376 (80.5) 91 (19.5) 453 (97.0) 14 (3.0) 222 (47.5) 245 (52.5) 53 (11.3) 414 (88.7) 99 (21.2) 368 (78.8) .36 .60 .69 .14 .74 .16 .49 .83 .21 .13 .50 .08 .88 Yes 736 (63.3) 439 (63.2) 297 (63.6) https://www.jmir.org/2023/1/e46793 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 8 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al Characteristics No Total (NAlter=1162) Regular HIV e-re- port group (n3=695) Exchangeable HIV e-report group (n3=467)b P value 426 (36.7) 256 (36.8) 170 (36.4) UAI with regular partners, n (%) Yes No Knowing the HIV status of regular partners, n (%) Yes No HIV testing and awareness information Preference of HIV testing, n (%) Not tested MSM community–based facility Health care facility Self-testing test strips Tested for HIV in the past 3 months, n (%) Yes No Tested for other STIsg in the past 3 months, n (%) Yes No Infection with other STIs, n (%) Yes No Received HIV prevention services, n (%) Yes No 242 (20.8) 920 (79.2) 424 (36.5) 738 (63.5) 184 (15.8) 262 (22.6) 296 (25.5) 420 (36.1) 469 (40.4) 693 (59.6) 147 (12.7) 1015 (87.3) 671 (57.8) 491 (42.2) 909 (78.2) 253 (21.8) Awareness of HIV infection status among gay men in Guangzhou, n (%) 1 in 100 MSM HIV positive 1 in 50 MSM HIV positive Knowing someone with HIV, n (%) Yes No 634 (54.6) 528 (45.4) 366 (31.5) 796 (68.5) 143 (20.6) 552 (79.4) 244 (35.1) 451 (64.9) 109 (15.7) 147 (21.2) 171 (24.6) 268 (38.6) 263 (37.8) 432 (62.2) 81 (11.7) 614 (88.3) 412 (59.3) 283 (40.7) 530 (76.3) 165 (23.7) 331 (47.6) 364 (52.4) 483 (69.5) 212 (30.5) 99 (21.2) 368 (78.8) 180 (38.5) 287 (61.5) 75 (16.1) 115 (24.6) 125 (26.8) 152 (32.5) 206 (44.1) 261 (55.9) 66 (14.1) 401 (85.9) 259 (55.5) 208 (44.5) 379 (81.2) 88 (18.8) 197 (42.2) 270 (57.8) 313 (67.0) 154 (33.0) HIV testing norms (score), median (IQR) 3.0 (2.7-3.0) 3.0 (2.7-3.0) 3.0 (2.7-3.3) HIV stigma (score), median (IQR) 19.0 (17.0-20.0) 19.0 (17.0-20.0) 19.0 (17.0-20.0) HIV e-report deliver process information Has viewed others’ e-report, n (%) Yes No Duration of viewing others’ e-report (seconds), median (IQR) Number of times viewing others’ e-report (time), median (IQR) Social network characteristics, n (%) https://www.jmir.org/2023/1/e46793 XSL•FO RenderX 771 (66.4) 391 (33.6) 695 (100.0) 0 (0.0) 76 (16.3) 391 (83.7) 5.7 (0.0-11.4) 9.2 (5.9-15.8) 0.0 (0.0-0.0) 1.0 (0.0-3.0) 2.0 (1.0-3.0) 0.0 (0.0-0.0) J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 9 (page number not for citation purposes) .80 .23 .19 .03 f .21 .20 .05 .07 .37 .96 .04 f <.001 f <.001 f <.001 f JOURNAL OF MEDICAL INTERNET RESEARCH Li et al Characteristics Total (NAlter=1162) Regular HIV e-re- port group (n3=695) Exchangeable HIV e-report group (n3=467)b P value Similarity to its egoh demographics 0.9 (0.8-1.0) 0.9 (0.8-1.0) 0.9 (0.8-0.9) Age similarity Similarity in educational level Income similarity Similarity in marital status, n (%) 0 0.5 1 Relationship with his ego, n (%) Boyfriends Sex partners Gay friends HIV e-report recruitment outcome Has transformed into alter-egoi, n (%) 0.9 (0.8-1.0) 1.0 (1.0-1.0) 0.7 (0.7-1.0) 0.9 (0.8-1.0) 0.9 (0.8-1.0) 1.0 (1.0-1.0) 1.0 (1.0-1.0) 0.7 (0.7-1.0) 0.7 (0.7-1.0) 77 (6.6) 40 (3.4) 44 (6.3) 17 (2.4) 33 (7.1) 23 (4.9) 1045 (90.0) 634 (91.2) 411 (88.0) 231 (19.9) 127 (10.9) 804 (69.2) 135 (19.4) 76 (10.9) 484 (69.6) 96 (20.6) 51 (10.9) 320 (68.5) .01 f .58 .34 .11 .06 .89 .24 Yes No 75 (6.5) 1087 (93.5) 40 (5.8) 655 (94.2) 35 (7.5) 432 (92.5) aAlter: those contactors who received and read the shared report from ego. bIn the exchangeable HIV e-report group, most alters did not complete the exchange process and therefore lacked data related to the viewing of the e-reports. c1 RMB=US $0.15. dMSM: men who have sex with men. eUAI: unprotected anal intercourse. fP<.05. gSTI: sexually transmitted infection. hEgo: an MSM who tested HIV at the Lingnan Center and shared his HIV e-report via the WeChat mini-program to his WeChat contactors. iAlter-ego: when the alter received his friend’s e-report, he transformed into the ego after taking an HIV test at the Lingnan Center and forwarded his e-report to his gay friends to expand recruitment. Recruitment Outcomes and Associated Factors at First Wave Alters’ Online Enrollment and Associated Factors at First Wave Compared with egos who did not recruit alters (egos-NRA), egos who successfully recruited alters (egos-RA) were more likely to be Guangdong permanent residents (354/473, 74.8% vs 411/610, 67.4%; P=.007) and have had anal sexual intercourse (463/473, 97.9% vs 584/610, 95.7%; P=.05). Egos-RA were also more likely to view their own (11.0 times vs 7.0 times; P<.001) and others’ reports (2.0 times vs 1.0 times; P<.001) and forward reports (5.0 times vs 2.0 times; P<.001) more frequently (Table S3 in Multimedia Appendix 1). The results of the 2-part model indicated that (1) egos with a history of anal sex were (multivariate-adjusted odds ratio [ORm]) 2.53 times (95% CI 1.01-6.32) more likely to successfully recruit alters than their counterparts and (2) the number of times e-reports were forwarded was proportional to the number of alters recruited, with the adjusted ORs (95% CI) being 1.00 (1.00-1.00), 2.23 (1.46-3.40), 5.98 (4.01-8.93), and 41.99 https://www.jmir.org/2023/1/e46793 XSL•FO RenderX (25.33-69.62) when the numbers of forwards stratified by quartile were 1-2, 2-3, 3-6, and ≥6 times, respectively (Table 3). Egos who were permanent Guangdong residents (ORm 1.54, 95% CI 1.13-2.10) and viewed their own e-reports (ORm 2.12-11.96) and others’ e-reports (ORm 1.37-3.74) more frequently were more likely to forward e-reports; in addition, egos with exchangeable e-reports (ORm 0.74, 95% CI 0.56-0.98) were less likely to forward e-reports than egos with regular e-reports (Tables S4 and S5 in Multimedia Appendix 1). Further analysis revealed that egos who viewed others’ e-reports more frequently were more likely to have lived in Guangzhou for more than 1 year and received regular e-reports (Tables S6-S8 in Multimedia Appendix 1). Tables S11-S13 in Multimedia Appendix 1 present the results of the subgroup analysis for egos of the regular e-reports group: egos who frequently viewed their own or others’ e-report were more likely to forward them, and forwarding of e-reports was strongly correlated with the number of people who clicked on e-reports and influenced alters’ recruitment. Similarly, the results J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 10 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al of the exchangeable e-reports group were consistent with those of the regular e-reports group. However, most participants in the exchangeable e-reports group did not complete the exchange process and were missing data on the number of clicks on the e-reports (Tables S9 and S10 in Multimedia Appendix 1). Table 3. Two-part modelsa estimating associated factors of successfully recruiting alters at wave 1 among egos who forwarded the HIV e-report (NEgo-wave1=1083). Variable Part I logistic regression model (NEgo-wave1=1083) Part II ordinary least squares model (NEgo-RA-wave1=473) β coefficient ORm b (95% CI) P value β coefficient P value Intervention groups Regular HIV e-report group Reference 1.00 (1.00-1.00) Reference Reference Reference Exchangeable HIV e-report group –0.13 0.88 (0.65-1.18) .39 0.25 .14 Registered permanent residence Other province Reference 1.00 (1.00-1.00) Reference Reference Reference Guangdong provinces 0.26 1.30 (0.93-1.80) .12 0.02 .90 Ever had anal sexual intercourse No Yes Reference 1.00 (1.00-1.00) Reference Reference Reference 0.93 2.53 (1.01-6.32) .05 –0.85 .15 Number of times viewing own e-report Q1 (0-4) Q2 (5-8) Q3 (9-16) Q4 (17-186) Reference 1.00 (1.00-1.00) Reference Reference Reference 0.31 0.22 –0.32 1.37 (0.89-2.11) 1.25 (0.81-1.92) 0.73 (0.45-1.18) .15 .31 .20 –0.37 –0.22 0.01 .18 .41 .97 Number of times forwarding e -report Q1 (1-1) Q2 (2-2) Q3 (3-5) Q4 (6-58) Reference 1.00 (1.00-1.00) Reference Reference Reference 0.80 1.79 3.74 2.23 (1.46-3.40) 5.98 (4.01-8.93) 41.99 (25.33-69.62) <.001 c <.001 c <.001 c 0.21 0.81 2.93 .53 .004 c <.001 c aThe model includes registered permanent residence, anal sexual intercourse, type of HIV e-report received, and number of times viewing own e-report and forwarding e-report. bORm: multivariate-adjusted odds ratio. cP<.05. Alters’ Transformation into Offline Alter-Egos and Associated Factors at First Wave Compared with online alters who did not transform into offline ego-recruiters, alter-egos were more likely to view others’ reports more frequently (1.0 times vs 3.0 times; P<.001) over longer periods (4.9 seconds vs 8.5 seconds; P<.001). Alter-egos were also more likely to be longer time Guangzhou residents (543/980, 55.4% vs 50/70, 71.4%; P=.009), have had casual sexual partners in the past 3 months (481/980, 49.1% vs 44/70, 62.9%; P=.03), have had UAI with them (112/980, 11.4% vs 15/70, 21.4%; P=.01), and have undergone HIV testing in the MSM-friendly clinic in the past (209/980, 21.3% vs 23/70, 32.9%; P=.04). Within egocentric social networks, alters who did not transform into ego-recruiters and alter-egos had similar proportions of relationships and high sociodemographic similarities with their egos (Table S15 in Multimedia Appendix 1). According to the result of the null model fit, the random effect for cluster level was not statistically significant (P=.17), suggesting that alters’ transformation into egos was not aggregated at the social network level (Table S14 in Multimedia Appendix 1). According to the logistic regression model with Firth correction, alters in the exchangeable e-report group (ORm 1.95, 95% CI 1.15-3.32) had longer time of residency (ORm 2.19, 95% CI 1.26-3.79), UAI with casual partners in the past 3 months (ORm 2.30, 95% CI 1.25-4.25), larger number of times viewing others’ e-report (Q4 vs Q1: ORm 4.45, 95% CI 2.20-8.97), and were more likely to transit to alter-ego for the subsequent wave, thereby expanding recruitment. By contrast, alters with higher monthly incomes (ORm 0.50, 95% CI https://www.jmir.org/2023/1/e46793 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 11 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al 0.30-0.84) and alters who preferred HIV self-testing (ORm 0.55, 95% CI 0.26-1.18; reference: alters who had not tested for HIV) were less likely to transit from online alter to offline ego as the next wave’s potential recruiters (Table 4). Further analysis was conducted to understand which e-reports alters tended to view more frequently. In the multivariable logistic regression model, the following variables showed statistically significant associations with the number of times others’ e-reports were viewed: preferring self-testing (P=.02), having undergone testing for other STIs in the past 3 months (P=.001), receiving regular HIV e-reports (P<.001), being the boyfriend of the ego (P=.01), and greater age similarity (P=.002; Tables S16 and S17 in Multimedia Appendix 1). Tables S19 and S20 in Multimedia Appendix 1 show the results of the subgroup analysis for alters of the regular e-reports group: alters’ transformation was positively associated with the number of times viewing others’ e-report and history of other STIs. In addition, alters who had lived in Guangzhou for a longer period were egos’ boyfriends and those who had a lower age viewed others’ e-reports more frequently. For the exchangeable e-reports group, alters’ conversion was associated with length of residence in Guangzhou and temporary partner’s UAI. The transformation of alters for the exchangeable e-reports group was related to length of residence in Guangzhou and UAI with casual partners in the past 3 months (Table S18 in Multimedia Appendix 1). All of the results mentioned above are summarized in Figure 3. Table 4. Unadjusted and multivariable logistic regression modelsa with Firth correction estimating associated factors of expanded recruitment among alters at wave1 (NAlter-wave1=1050). Variable Intervention groups Regular HIV e-report group Exchangeable HIV e-report group Monthly incomee ≤5000 RMB >5000 RMB Length of residence in Guangzhou ≤1 year >1 year Preference of HIV testing Not tested Men who have sex with men community Health care facility Self-testing test strips N Cases, n (%) ORb (95% CI) P value ORm c (95% CI) P value 609 441 437 613 457 593 168 232 265 385 35 (5.7) 1.00 (1.00-1.00) Reference 1.00 (1.00-1.00) Reference 35 (7.9) 1.41 (0.87-2.29) .16 1.95 (1.15-3.32) .01 d 36 (8.2) 1.00 (1.00-1.00) Reference 1.00 (1.00-1.00) Reference 34 (5.5) 0.65 (0.40-1.06) .09 0.50 (0.30-0.84) .009 d 20 (4.4) 1.00 (1.00-1.00) Reference 1.00 (1.00-1.00) Reference 50 (8.4) 1.98 (1.17-3.37) .01 d 2.19 (1.26-3.79) .005 d 13 (7.7) 1.00 (1.00-1.00) Reference 1.00 (1.00-1.00) Reference 23 (9.9) 1.29 (0.64-2.61) 18 (6.8) 0.86 (0.41-1.79) 16 (4.2) 0.51 (0.24-1.08) .05 .96 .02 d 1.23 (0.59-2.57) 0.95 (0.45-2.00) 0.55 (0.26-1.18) .13 .80 .03 d Unprotected anal intercourse with casual partners in the past 3 months No Yes Number of times viewing others’ e-report Q1 (0-0) Q2 (1-1) Q3 (2-2) Q4 (3-25) 923 127 374 201 179 296 55 (6.0) 1.00 (1.00-1.00) Reference 1.00 (1.00-1.00) Reference 15 (11.8) 2.16 (1.19-3.92) .01 d 2.30 (1.25-4.25) .008 d 10 (2.7) 1.00 (1.00-1.00) Reference 1.00 (1.00-1.00) Reference 11 (5.5) 2.10 (0.89-4.93) 13 (7.3) 2.81 (1.23-6.45) .70 .42 2.16 (0.92-5.05) 2.88 (1.26-6.57) .81 .35 36 (12.2) 4.86 (2.40-9.85) <.001 d 4.45 (2.20-8.97) <.001 d aThe model includes monthly income, length of residence in Guangzhou, pathways to HIV testing, unprotected anal intercourse with casual partners in the past 3 months, and number of times viewing others’ e-report. bOR: odds ratio. cORm: multivariate-adjusted odds ratio. dP<.05. e1 RMB=US $0.15. https://www.jmir.org/2023/1/e46793 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 12 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al Figure 3. Compendium of analysis results. Phase 1: analysis of associated factors with Alters' online enrollment; Phase 2: analysis of associated factors with Alters' transformation to Alter-ego. (A) Compendium of phase 1 analysis results; (B) Compendium of phase 2 analysis results; (C) Compendium of phase 1 analysis results for exchangeable HIV E-report group; (D) Compendium of phase 2 analysis results for exchangeable HIV E-report group; (E) Compendium of phase 1 analysis results for regular HIV E-report group; (F) Compendium of phase 2 analysis results for regular HIV E-report group; (a) HIV E-report delivery information, the data was sourced from the data portal records of WeChatl; (b) Socio-demographic characteristics, the data from questionnaire surveies of Egos and Alters; (c) Grouped according to type of E-report, the grouping information is derived from an electronic random number table generated by the WeChat mini-program; (d) Sexuality and testing behaviour, the data from questionnaire surveies of Egos and Alters; (e) Social networking features, the data from questionnaire surveies of Egos and Alters. Ego: an MSM who tested HIV at the Lingnan Center and shared his HIV E-report via WeChat mini-program to his WeChat contactors; Alter: those contactors who received and read the shared report from ego; UAI: unprotected anal intercourse; STI: sexually transmitted infection. Qualitative Results About Barriers for Alters’ Transformation into Offline Alter-Egos Overview Qualitative interviews were conducted with 8 participants: 5 alters (A1-A5) and 3 alter-egos (B1-B3). The major barriers to transformation from online alters to offline egos (ie, the reasons why online alters did not undergo HIV testing at Lingnan and thus did not transform into alter-egos) were (1) low awareness of the functions of the WeChat-based mini-program “ChaBei” for booking HIV testing services, (2) limited access to HIV e-reports at offline testing facilities, and (3) no demand for offline HIV testing. Awareness of the Functions of the WeChat-Based Mini-Program “ChaBei” “ChaBei” is a WeChat-based tool to promote HIV testing in Guangzhou, China. Some participants believed the authenticity and anonymity of HIV e-reports. B1 stated, “I think in https://www.jmir.org/2023/1/e46793 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 13 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al testing at healthcare facilities will be more accurate than self-testing.” B3 said, “I think Lingnan Center’s e-reports are as reliable as hospitals’ test reports.” Some alters were educated about the functions of “ChaBei” by egos when the egos forwarded them the e-reports. These alters then proceeded to make appointments on “ChaBei” and underwent testing. B1 stated, “My friend said that this place could do the testing for free, so I booked an appointment.” However, some MSM were not familiar with “ChaBei” and expressed that the mini-program was not well advertised and not well known. A1 said, “There are probably few people using this app at the moment, and you need to make advertisements for ChaBei.” Consequently, they were not familiar with the functions of “ChaBei.” A5 said, “I didn’t use this mini-program because I don’t know how to use it.” They were also concerned about privacy problems. A1 said, “I don’t know about the mini-program, so I’m worried about its privacy.” These barriers prevented alters from opening HIV e-reports and using the mini-program to make appointments. Inadequate Access to HIV e-Reports at Offline Testing Service Facilities At the time of this study, “ChaBei” was a recently developed app and initially only allowed appointment booking at Lingnan; therefore, participants who tested at Lingnan were able to obtain HIV e-reports. Despite 60% of MSM in Guangzhou undergoing HIV testing at Lingnan [12], inadequate access to HIV e-reports at offline testing facilities was reported as a barrier to transformation of online alters into offline egos. The location and service time of the center were not convenient. A4 said, “I have not been to Lingnan Center, mainly because the location is not very convenient and it is far from my place.” A3 said, “I have done HIV testing, but usually self-testing using test strips. The service hours of Lingnan are mainly on weekdays, which is not very convenient.” Demand for Offline HIV Testing The alter-egos preferred regular offline HIV testing. B2 stated, “I find it convenient to book tests online.” B3 said, “I am mainly in the habit of regular testing.” Among alters who did not transform, some expressed that they did not have high-risk sexual behavior or regular testing habits. A5 said, “I don’t need to get tested; I haven’t had high-risk sexual behavior.” Some preferred self-testing. A2 stated, “I am most used to using test strips because it is convenient and I can test whenever I want.” Discussion Principal Findings This study assessed the use of HIV e-reports as a tool for recruiting MSM participants for an HIV testing study via social networks. Similar to offline recruitment, MSM egos forwarded their e-reports to more MSM peers (thus, to larger active social networks) and were thus more likely to recruit alters for the study. However, unlike offline recruitment, the success and sustainability of online recruitment were found to depend on high levels of familiarity and awareness among MSM, with the “ChaBei” tool used for delivering e-reports to their network associates. Remarkably, the exchangeable e-report associated with alter’s transformation into offline ego means that it might https://www.jmir.org/2023/1/e46793 XSL•FO RenderX promote MSM to test HIV offline to get their own e-report for exchange with others. This indicates that promoting awareness and increasing recognition of online communicable disease reporting tools are crucial for effectively reaching and recruiting high-risk individuals. This study recruited sufficient eligible alters (n=1162) from the distribution of HIV e-reports in online social networks, although the recruitment was conducted over 2 years during the COVID-19 pandemic. Compared with traditional recruitment methods in previous studies, the recruitment methods in this study showed improved recruitment efficiency. For example, Rhodes et al [19] recruited 304 participants in gay bars and clubs over 2 years, Outlaw et al [20] recruited 188 MSM from the community over 2 years, and Katz et al [21] recruited 230 MSM from sexually transmitted disease clinics over 4 years. Compared with a study that recruited over 400 MSM through multiple social media platforms simultaneously [22], these studies recruited only 100-300 MSM in about a year [23,24]. Although the recruitment strategy in this study showed no advantages over traditional online recruitment methods, it has some notable differences. The newly developed e-reports can be distributed in social networks for the disclosure of infection status based on the demand for safe sex in the MSM community. This recruitment approach has a sustainable mechanism as it involves MSM themselves in the recruitment of other MSM for HIV testing by sex drive. In addition, this e-report delivery strategy can be extended beyond MSM to populations of other high-risk communicable diseases. in facilitate Egos in the regular HIV e-reports group had a higher probability of successful recruitment on alters’ online enrollment, which may be attributed to the increased accessibility of others’ test results in the regular HIV e-report group. After the regular e-reports were forwarded, report receivers could directly see the test results [12]. Direct disclosure of test results promoted understanding between partners [25], which was able to effectively communicate recruitment. A to nonsignificantly higher proportion of alters transformed into the exchangeable HIV e-reports group. alter-egos Nevertheless, in the first wave, alters in the exchangeable HIV e-reports group were more likely to transform into alter-egos for the subsequent wave, thus expanding recruitment. HIV testing at Linnan and obtaining an e-report were the necessary prerequisites for alters to transform into alter-egos. After the exchangeable e-reports were forwarded, report receivers also needed to show their reports and completed the exchange process to see the test results [12]. Therefore, driven by the need for safe sex in the MSM community, alters in the exchangeable HIV e-reports group would perform HIV testing to know the infection status of their partners. The recruitment outcome between the 2 groups on alters that transformed into alter-egos is in the line with our RCT hypothesis, that is, the exchangeable e-report may promote HIV testing behaviors. This study primarily represented the promotion of exchangeable e-reports for facility-based HIV testing behaviors, and additional follow-up results will be expected in recent future. Egos who forwarded e-reports to more MSM recruited more alters and more frequently viewed their own and others’ e-reports. The number of times e-reports were forwarded showed J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 14 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al a strong positive association with the size of the social network, which is consistent with previous studies that used social networks for participant recruitment [26,27]. According to the quantitative and qualitative results, the number of times own and others’ e-reports were viewed reflected the participants’ familiarity with the functions of the “ChaBei” app. Compared with unfamiliar things, people rely more on familiar things and are more likely to share them with others [28]. As we studied the settings for e-reports, after regular e-reports were forwarded, report receivers could directly see the test results. However, when exchangeable e-reports were forwarded, report receivers also needed to show their reports and completed the exchange process before they can see the test results [12]. Therefore, the accessibility of test results also affects egos’ ability to forward e-reports. Egos with Guangzhou-registered residence are more likely to have a stable circle of friends and life circle in the area and are therefore more conducive to forward e-reports [29]. In addition, egos with a “1” role (insertive) and who have had anal sexual intercourse were more likely to recruit alters. MSM with “1” roles are dominant during sexual activity and also have a greater sense of self-protection and competence [30,31]. As a result, they tend to be more active in HIV testing [32]. Lack of awareness of and familiarity with “ChaBei” in the community was a main barrier to MSM recruitment. Hence, it is important to establish the reputation of “ChaBei” within the MSM community and among Guangzhou residents over time. Further, this conclusion was well confirmed in the exchangeable HIV e-report group. Because of the mechanism of e-report delivery, the e-reports sent by egos in the regular HIV e-report group can be directly clicked and viewed by alters for test results. Therefore, in the regular e-report group, actions such as forwarding and viewing of e-reports affected alters’ recruitment by influencing the number of clicks on e-reports. According to the qualitative interviews, the reasons for their reluctance to forward e-reports were lack of demand for HIV disclosure and privacy concerns. To address privacy concerns, the mini-program was developed by developers from the MSM community. The CDC ensures the anonymity and authenticity of e-reports, and only the corresponding MSM are allowed to forward their own e-reports [12]. Our recruitment strategy demonstrated an effective and sustainable recruitment mechanism. e-Reports were forwarded many times in social networks in the first, second, and third waves, creating multiple recruitment waves (Figure 2B). Alters who normally lived in Guangzhou for more than 1 year, had lower incomes, underwent offline testing at Lingnan, and had UAI with casual partners were more likely than others to transform into alter-egos for the subsequent recruitment wave. Alters who have lived in Guangzhou for a longer period indicate that they are more likely to have a stable circle of friends and life circle in the area, and are therefore more conducive to extended recruitment [29]. Previous studies have shown that participants with a higher level of income were less likely to be tested for HIV [33]. Because of their high-risk sexual behavior, such alters may perceive an elevated risk of infection and may therefore be more inclined to take steps to mitigate that risk (ie, undergoing HIV testing). In addition, having knowledge of e-reports and viewing others’ e-reports frequently were facilitators of recruitment in the subsequent waves. This finding is also consistent with those of the qualitative interviews. In the subgroup analysis, the facilitator of participation in subsequent recruitments in the regular HIV e-report group was also primarily familiar with and frequently viewed ego reports, which was influenced by the length of residence in Guangzhou, social network ties, and social network age similarity. Alters who were egos’ boyfriends (ie, their social tie) were more likely to participate in subsequent recruitment. This familiar and stable relationship could increase their sense of identification and recognition of our program and the e-report. By contrast, in the exchangeable HIV e-report group, subsequent recruitment was mainly influenced by the length of residence in Guangzhou and unprotected anal sex because of the influence of the e-report delivery mechanism, which does not allow the content of others’ e-reports to be seen until the exchange process is completed. Limitations This study has several limitations that should be acknowledged. First, the implementation of this study was affected by the COVID-19 pandemic, particularly during the lockdown period (February to May 2020) in China during which HIV testing services were disrupted. Second, this study lacked sufficient resources to investigate alters who failed to be recruited by egos, which is essential to guide future efforts for e-report promotion in such populations. Third, this was a baseline analysis for an RCT targeting alters; as such, available data on egos’ characteristics were insufficient. Therefore, qualitative interviews were conducted to explore the reasons for forwarding and not forwarding e-reports. Conclusions This study recruited sufficient MSM to participate in an HIV testing intervention study through the delivery of HIV e-reports in social networks using a WeChat-based mini-program “ChaBei.” The delivery of e-reports was acceptable in MSM social networks. The HIV e-report exchange mechanism might promote MSM to test HIV offline to get their own e-report for exchange in the community and thus achieve sustainability of recruitment. Further, promoting awareness and increasing recognition of online communicable disease reporting tools are key to ensuring the success and sustainability of online recruitment. This study provides a feasible and sustainable recruitment strategy to trace the direct contacts of individuals with communicable diseases for early testing and participation in infectious disease research. Acknowledgments This study was supported by the National Natural Science Foundation of China (grants 71974212, 72204059), Guangdong Basic and Applied Basic Research Foundation (grant 2020A1515010737), and Science and Technology Program of Guangzhou, China (grant 202201010078). Mr Qi Liu, Mr Gang Meng, and Mr Jie Lu (Guangzhou Lingnan Community Support Center) assisted with WeChat-based mini-program development and ego offline recruitment. Tailing Chen, Sha Chen, and Jinyue Li (all from https://www.jmir.org/2023/1/e46793 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 15 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al School of Public Health); Minghui Zhang (from School of Life Science); and Yijun Guo (from School of Pharmaceutical Science, Sun Yat-sen University) assisted with offline and online recruitment. Professional English language editing support was provided by AsiaEdit. Data Availabilty The data sets generated during or analyzed during this study are available from the corresponding author on reasonable request. Authors' Contributions JSL wrote the manuscript and conducted all statistical analyses. FSH participated in the research design, data collection design, and manuscript revision. YHL is the co-principal investigator responsible for the online part of this project, including research design, WeChat-based mini-program design, management, and supervision. XRF conducted baseline and follow-up data collection, quality control, project implementation, and manuscript revision. JLQ and QLY participated in the research design, data collection design, project launch, beginning of baseline data collection, beginning of baseline quality control, and beginning of baseline project implementation. JG, JHL, and DRX participated in the research design, data collection design, and part of human and funding resources support during the project implementation. YZG is the co-principal investigator responsible for site works, including research design, WeChat-based mini-program design, clinic and telephone data collection, management, and supervision. CH is the principal investigator of this project in the management and leadership of the research project. Conflicts of Interest None declared. Multimedia Appendix 1 Characteristics of egos and alters, results of regression models, results of the null model, and an image showing social networks containing the 3 waves of recruitment. [DOCX File , 216 KB-Multimedia Appendix 1] Multimedia Appendix 2 The CONSORT-eHEALTH checklist (V 1.6.1). [PDF File (Adobe PDF File), 1224 KB-Multimedia Appendix 2] References 1. 2. 3. Rothman KJ. Modern Epidemiology (3rd ed). Philadelphia, PA: Lippincott; 2008. Ortblad KF, Musoke DK, Ngabirano T, Salomon JA, Haberer JE, McConnell M, et al. Is knowledge of HIV status associated with sexual behaviours? A fixed effects analysis of a female sex worker cohort in urban Uganda. J Int AIDS Soc 2019 Jul;22(7):e25336 [FREE Full text] [doi: 10.1002/jia2.25336] [Medline: 31287625] Suthar AB, Ford N, Bachanas PJ, Wong VJ, Rajan JS, Saltzman AK, et al. Towards universal voluntary HIV testing and counselling: a systematic review and meta-analysis of community-based approaches. PLoS Med 2013 Aug;10(8):e1001496 [FREE Full text] [doi: 10.1371/journal.pmed.1001496] [Medline: 23966838] 6. 5. 4. Maheu-Giroux M, Vesga JF, Diabaté S, Alary M, Baral S, Diouf D, et al. Population-level impact of an accelerated HIV response plan to reach the UNAIDS 90-90-90 target in Côte d'Ivoire: Insights from mathematical modeling. PLoS Med 2017 Jun;14(6):e1002321 [FREE Full text] [doi: 10.1371/journal.pmed.1002321] [Medline: 28617810] UNAIDS DATA 2021. UNAIDS. Geneva: Joint United Nations Programme on HIV/AIDS; 2021. URL: https://www. unaids.org/sites/default/files/media_asset/JC3032_AIDS_Data_book_2021_En.pdf Zhou Y, Lu Y, Ni Y, Wu D, He X, Ong JJ, et al. Monetary incentives and peer referral in promoting secondary distribution of HIV self-testing among men who have sex with men in China: A randomized controlled trial. PLoS Med 2022 Mar;19(2):e1003928 [FREE Full text] [doi: 10.1371/journal.pmed.1003928] [Medline: 35157727] Key Populations Atlas. UNAIDS. URL: https://kpatlas.unaids.org/dashboard [accessed 2022-05-18] Granich RM, Gilks CF, Dye C, De Cock KM, Williams BG. Universal voluntary HIV testing with immediate antiretroviral therapy as a strategy for elimination of HIV transmission: a mathematical model. Lancet 2009 Jan 03;373(9657):48-57 [doi: 10.1016/S0140-6736(08)61697-9] [Medline: 19038438] Coates TJ, Kulich M, Celentano DD, Zelaya CE, Chariyalertsak S, Chingono A, NIMH Project Accept (HPTN 043) study team. Effect of community-based voluntary counselling and testing on HIV incidence and social and behavioural outcomes (NIMH Project Accept; HPTN 043): a cluster-randomised trial. Lancet Glob Health 2014 May;2(5):e267-e277 [FREE Full text] [doi: 10.1016/S2214-109X(14)70032-4] [Medline: 25103167] 7. 8. 9. https://www.jmir.org/2023/1/e46793 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 16 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al 10. Konda KA, Castillo R, Leon SR, Silva-Santisteban A, Salazar X, Klausner JD, et al. HIV Status Communication with Sex Partners and Associated Factors Among High-Risk MSM and Transgender Women in Lima, Peru. AIDS Behav 2017 Jan;21(1):152-162 [FREE Full text] [doi: 10.1007/s10461-016-1444-4] [Medline: 27262875] 12. 11. Grov C, Rendina HJ, Moody RL, Ventuneac A, Parsons JT. HIV Serosorting, Status Disclosure, and Strategic Positioning Among Highly Sexually Active Gay and Bisexual Men. AIDS Patient Care STDS 2015 Oct;29(10):559-568 [FREE Full text] [doi: 10.1089/apc.2015.0126] [Medline: 26348322] Shi Y, Qiu J, Yang Q, Chen T, Lu Y, Chen S, et al. Increasing the HIV testing among MSM through HIV test result exchange mechanism: study protocol for a cluster randomized controlled trial. BMC Infect Dis 2021 Aug 06;21(1):764 [FREE Full text] [doi: 10.1186/s12879-021-06484-y] [Medline: 34362323] Stojanovski K, Naja-Riese G, King EJ, Fuchs JD. A Systematic Review of the Social Network Strategy to Optimize HIV Testing in Key Populations to End the Epidemic in the United States. AIDS Behav 2021 Sep;25(9):2680-2698 [doi: 10.1007/s10461-021-03259-z] [Medline: 33871730] 13. 14. Grov C, Westmoreland DA, Carneiro PB, Stief M, MacCrate C, Mirzayi C, et al. Recruiting vulnerable populations to participate in HIV prevention research: findings from the Together 5000 cohort study. Ann Epidemiol 2019 Jul;35:4-11 [FREE Full text] [doi: 10.1016/j.annepidem.2019.05.003] [Medline: 31182379] 15. Cheng W, Cai Y, Tang W, Zhong F, Meng G, Gu J, et al. Providing HIV-related services in China for men who have sex with men. Bull World Health Organ 2016 Mar 01;94(3):222-227 [FREE Full text] [doi: 10.2471/BLT.15.156406] [Medline: 26966334] 16. Ristad E, Yianilos P. Learning string-edit distance. IEEE Trans. Pattern Anal. Machine Intell 1998;20(5):522-532 [doi: 10.1109/34.682181] Firth D. Bias Reduction of Maximum Likelihood Estimates. Biometrika 1993 Mar;80(1):27-38 [doi: 10.2307/2336755] 17. 18. Heinze G, Schemper M. A solution to the problem of separation in logistic regression. Stat Med 2002 Aug 30;21(16):2409-2419 [doi: 10.1002/sim.1047] [Medline: 12210625] 19. Rhodes SD, Alonzo J, Mann L, Song EY, Tanner AE, Arellano JE, et al. Small-Group Randomized Controlled Trial to Increase Condom Use and HIV Testing Among Hispanic/Latino Gay, Bisexual, and Other Men Who Have Sex With Men. Am J Public Health 2017 Jun;107(6):969-976 [doi: 10.2105/AJPH.2017.303814] [Medline: 28426301] 20. Outlaw AY, Naar-King S, Parsons JT, Green-Jones M, Janisse H, Secord E. Using motivational interviewing in HIV field outreach with young African American men who have sex with men: a randomized clinical trial. Am J Public Health 2010 Apr 01;100 Suppl 1(Suppl 1):S146-S151 [doi: 10.2105/AJPH.2009.166991] [Medline: 20147689] 21. Katz DA, Golden MR, Hughes JP, Farquhar C, Stekler JD. HIV Self-Testing Increases HIV Testing Frequency in High-Risk Men Who Have Sex With Men: A Randomized Controlled Trial. J Acquir Immune Defic Syndr 2018 Aug 15;78(5):505-512 [FREE Full text] [doi: 10.1097/QAI.0000000000001709] [Medline: 29697595] 22. Merchant RC, Clark MA, Liu T, Romanoff J, Rosenberger JG, Bauermeister J, et al. Comparison of Home-Based Oral Fluid Rapid HIV Self-Testing Versus Mail-in Blood Sample Collection or Medical/Community HIV Testing By Young Adult Black, Hispanic, and White MSM: Results from a Randomized Trial. AIDS Behav 2018 Jan;22(1):337-346 [FREE Full text] [doi: 10.1007/s10461-017-1802-x] [Medline: 28540562] 23. Horvath KJ, Lammert S, Danh T, Mitchell JW. The Feasibility, Acceptability and Preliminary Impact of Mobile Application to Increase Repeat HIV Testing Among Sexual Minority Men. AIDS Behav 2020 Jun;24(6):1835-1850 [doi: 10.1007/s10461-019-02755-7] [Medline: 31823111] 24. Ybarra ML, Prescott TL, Phillips GL, Bull SS, Parsons JT, Mustanski B. Pilot RCT Results of an mHealth HIV Prevention Program for Sexual Minority Male Adolescents. Pediatrics 2017 Jul;140(1):e20162999 [FREE Full text] [doi: 10.1542/peds.2016-2999] [Medline: 28659456] 25. Walcott MM, Hatcher AM, Kwena Z, Turan JM. Facilitating HIV status disclosure for pregnant women and partners in rural Kenya: a qualitative study. BMC Public Health 2013 Dec 02;13:1115 [FREE Full text] [doi: 10.1186/1471-2458-13-1115] [Medline: 24294994] 26. Cornwell B, Schneider JA. Social venue range and referral chain impact: Implications for the sampling of hidden communities. PLoS One 2017 Aug 23;12(8):e0181494 [FREE Full text] [doi: 10.1371/journal.pone.0181494] [Medline: 28832587] 27. Kuhns LM, Kwon S, Ryan DT, Garofalo R, Phillips G, Mustanski BS. Evaluation of respondent-driven sampling in a study of urban young men who have sex with men. J Urban Health 2015 Feb 16;92(1):151-167 [FREE Full text] [doi: 10.1007/s11524-014-9897-0] [Medline: 25128301] 28. Han G, Liu L, Jiang J, Shu L, Rodrigues JJPC. A Collaborative Secure Localization Algorithm Based on Trust Model in Underwater Wireless Sensor Networks. Sensors (Basel) 2016 Feb 16;16(2):229 [FREE Full text] [doi: 10.3390/s16020229] [Medline: 26891300] 29. Yang K, Liu Y, Yin X, Wu S, Wu Q, Wang L, et al. A survey of social network status and its related factors for older adults with type 2 diabetes in Beijing, China. Nurs Open 2022 Mar;9(2):1005-1014 [FREE Full text] [doi: 10.1002/nop2.1138] [Medline: 34850591] 30. Zeng X, Zhong X, Peng B, Zhang Y, Kong C, Liu X, et al. Prevalence and associated risk characteristics of HIV infection based on anal sexual role among men who have sex with men: a multi-city cross-sectional study in Western China. Int J Infect Dis 2016 Aug;49:111-118 [FREE Full text] [doi: 10.1016/j.ijid.2016.06.009] [Medline: 27312581] https://www.jmir.org/2023/1/e46793 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 17 (page number not for citation purposes) JOURNAL OF MEDICAL INTERNET RESEARCH Li et al 31. Johns MM, Pingel E, Eisenberg A, Santana ML, Bauermeister J. Butch tops and femme bottoms? Sexual positioning, sexual decision making, and gender roles among young gay men. Am J Mens Health 2012 Nov 27;6(6):505-518 [FREE Full text] [doi: 10.1177/1557988312455214] [Medline: 22843811] 32. Dangerfield DT, Gravitt P, Rompalo AM, Tai R, Lim SH. Correlates of anal sex roles among Malay and Chinese MSM in Kuala Lumpur, Malaysia. Int J STD AIDS 2016 Mar 16;27(4):313-320 [doi: 10.1177/0956462415581125] [Medline: 25887064] 33. Gerbi GB, Archie-Booker E, Claridy MD, Miles-Richardson S. Factors Associated with Self-Reported HIV Testing Among Persons 18 Years and Over in the United States - Findings from the 2012 Behavioral Risk Factor Surveillance System (BRFSS). HSOA J Infect Non Infect Dis 2016 Jun;2(1):013 [FREE Full text] [doi: 10.24966/INID-8654/100013] [Medline: 30498787] Abbreviations CDC: Center for Disease Control and Prevention MSM: men who have sex with men ORm: multivariate-adjusted odds ratio RCT: randomized controlled trial SNS: social network strategy STI: sexually transmitted infection UAI: unprotected anal intercourse UNAIDS: Joint United Nations Programme on HIV/AIDS Edited by A Mavragani; submitted 25.02.23; peer-reviewed by Y Lu, Y Zhou; comments to author 05.04.23; revised version received 20.05.23; accepted 23.05.23; published 15.06.23 Please cite as: Li JS, Gu YZ, Hou FS, Lu YH, Fan XR, Qiu JL, Yang QL, Gu J, Li JH, Xu DR, Hao C Delivery of WeChat-Based HIV Result e-Reports in Social Networks for Recruitment of High-Risk Population: Baseline Data From a Cluster Randomized Controlled Trial J Med Internet Res 2023;25:e46793 URL: https://www.jmir.org/2023/1/e46793 doi: 10.2196/46793 PMID: ©Ju-Shuang Li, Yu-Zhou Gu, Feng-Su Hou, Yong-Heng Lu, Xiao-Ru Fan, Jia-Ling Qiu, Qing-Ling Yang, Jing Gu, Jing-Hua Li, Dong Roman Xu, Chun Hao. Originally published in the Journal of Medical Internet Research (https://www.jmir.org), 15.06.2023. This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included. https://www.jmir.org/2023/1/e46793 XSL•FO RenderX J Med Internet Res 2023 \| vol. 25 \| e46793 \| p. 18 (page number not for citation purposes)

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.